| // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _ |
| // expected-no-diagnostics |
| #ifndef HEADER |
| #define HEADER |
| |
| // Test host codegen. |
| // RUN: %clang_cc1 -DCK1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1 |
| // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK1 |
| // RUN: %clang_cc1 -DCK1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK3 |
| // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK3 |
| |
| // RUN: %clang_cc1 -DCK1 -verify -Wno-vla -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK5 |
| // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK5 |
| // RUN: %clang_cc1 -DCK1 -verify -Wno-vla -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK7 |
| // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK7 |
| |
| // RUN: %clang_cc1 -DCK1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK9 |
| // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK9 |
| // RUN: %clang_cc1 -DCK1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK11 |
| // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK11 |
| |
| #ifdef CK1 |
| |
| template <typename T, int X, long long Y> |
| struct SS{ |
| T a[X]; |
| float b; |
| int foo(void) { |
| |
| #pragma omp target teams distribute parallel for simd |
| for(int i = 0; i < X; i++) { |
| a[i] = (T)0; |
| } |
| #pragma omp target teams distribute parallel for simd schedule(static) |
| for(int i = 0; i < X; i++) { |
| a[i] = (T)0; |
| } |
| #pragma omp target teams distribute parallel for simd schedule(static, X/2) |
| for(int i = 0; i < X; i++) { |
| a[i] = (T)0; |
| } |
| |
| #pragma omp target teams distribute parallel for simd schedule(dynamic) |
| for(int i = 0; i < X; i++) { |
| a[i] = (T)0; |
| } |
| |
| #pragma omp target teams distribute parallel for simd schedule(dynamic, X/2) |
| for(int i = 0; i < X; i++) { |
| a[i] = (T)0; |
| } |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| return a[0]; |
| } |
| }; |
| |
| int teams_template_struct(void) { |
| SS<int, 123, 456> V; |
| return V.foo(); |
| |
| } |
| #endif // CK1 |
| |
| // Test host codegen. |
| // RUN: %clang_cc1 -DCK2 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK13 |
| // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK13 |
| // RUN: %clang_cc1 -DCK2 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK15 |
| // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK15 |
| |
| // RUN: %clang_cc1 -DCK2 -verify -Wno-vla -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK17 |
| // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK17 |
| // RUN: %clang_cc1 -DCK2 -verify -Wno-vla -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK19 |
| // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK19 |
| |
| // RUN: %clang_cc1 -DCK2 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK21 |
| // RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK21 |
| // RUN: %clang_cc1 -DCK2 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK23 |
| // RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s |
| // RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK23 |
| #ifdef CK2 |
| |
| template <typename T, int n> |
| int tmain(T argc) { |
| T a[n]; |
| int m = 10; |
| #pragma omp target teams distribute parallel for simd |
| for(int i = 0; i < n; i++) { |
| a[i] = (T)0; |
| } |
| #pragma omp target teams distribute parallel for simd schedule(static) |
| for(int i = 0; i < n; i++) { |
| a[i] = (T)0; |
| } |
| #pragma omp target teams distribute parallel for simd schedule(static, m) |
| for(int i = 0; i < n; i++) { |
| a[i] = (T)0; |
| } |
| #pragma omp target teams distribute parallel for simd schedule(dynamic) |
| for(int i = 0; i < n; i++) { |
| a[i] = (T)0; |
| } |
| #pragma omp target teams distribute parallel for simd schedule(dynamic, m) |
| for(int i = 0; i < n; i++) { |
| a[i] = (T)0; |
| } |
| return 0; |
| } |
| |
| int main (int argc, char **argv) { |
| int n = 100; |
| int a[n]; |
| int m = 10; |
| #pragma omp target teams distribute parallel for simd |
| for(int i = 0; i < n; i++) { |
| a[i] = 0; |
| } |
| #pragma omp target teams distribute parallel for simd dist_schedule(static) |
| for(int i = 0; i < n; i++) { |
| a[i] = 0; |
| } |
| #pragma omp target teams distribute parallel for simd dist_schedule(static, m) |
| for(int i = 0; i < n; i++) { |
| a[i] = 0; |
| } |
| #pragma omp target teams distribute parallel for simd schedule(dynamic) |
| for(int i = 0; i < n; i++) { |
| a[i] = 0; |
| } |
| #pragma omp target teams distribute parallel for simd schedule(dynamic, m) |
| for(int i = 0; i < n; i++) { |
| a[i] = 0; |
| } |
| return tmain<int, 10>(argc); |
| } |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| #endif // CK2 |
| #endif // #ifndef HEADER |
| // CHECK1-LABEL: define {{[^@]+}}@_Z21teams_template_structv |
| // CHECK1-SAME: () #[[ATTR0:[0-9]+]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[V:%.*]] = alloca [[STRUCT_SS:%.*]], align 4 |
| // CHECK1-NEXT: [[CALL:%.*]] = call noundef signext i32 @_ZN2SSIiLi123ELx456EE3fooEv(ptr noundef nonnull align 4 dereferenceable(496) [[V]]) |
| // CHECK1-NEXT: ret i32 [[CALL]] |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@_ZN2SSIiLi123ELx456EE3fooEv |
| // CHECK1-SAME: (ptr noundef nonnull align 4 dereferenceable(496) [[THIS:%.*]]) #[[ATTR0]] comdat { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_BASEPTRS3:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_PTRS4:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_MAPPERS5:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[_TMP6:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_BASEPTRS11:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_PTRS12:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_MAPPERS13:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[_TMP14:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_BASEPTRS19:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_PTRS20:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_MAPPERS21:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[_TMP22:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_BASEPTRS27:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_PTRS28:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[DOTOFFLOAD_MAPPERS29:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK1-NEXT: [[_TMP30:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP0]], align 8 |
| // CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK1-NEXT: store ptr [[A]], ptr [[TMP1]], align 8 |
| // CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP2]], align 8 |
| // CHECK1-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK1-NEXT: store i32 3, ptr [[TMP5]], align 4 |
| // CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK1-NEXT: store i32 1, ptr [[TMP6]], align 4 |
| // CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK1-NEXT: store ptr [[TMP3]], ptr [[TMP7]], align 8 |
| // CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK1-NEXT: store ptr [[TMP4]], ptr [[TMP8]], align 8 |
| // CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP9]], align 8 |
| // CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP10]], align 8 |
| // CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP11]], align 8 |
| // CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8 |
| // CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK1-NEXT: store i64 123, ptr [[TMP13]], align 8 |
| // CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK1-NEXT: store i64 0, ptr [[TMP14]], align 8 |
| // CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4 |
| // CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4 |
| // CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK1-NEXT: store i32 0, ptr [[TMP17]], align 4 |
| // CHECK1-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK1-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0 |
| // CHECK1-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK1: omp_offload.failed: |
| // CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]] |
| // CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK1: omp_offload.cont: |
| // CHECK1-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0 |
| // CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8 |
| // CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0 |
| // CHECK1-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8 |
| // CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8 |
| // CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0 |
| // CHECK1-NEXT: store i32 3, ptr [[TMP25]], align 4 |
| // CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1 |
| // CHECK1-NEXT: store i32 1, ptr [[TMP26]], align 4 |
| // CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2 |
| // CHECK1-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8 |
| // CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3 |
| // CHECK1-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8 |
| // CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4 |
| // CHECK1-NEXT: store ptr @.offload_sizes.1, ptr [[TMP29]], align 8 |
| // CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5 |
| // CHECK1-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP30]], align 8 |
| // CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP31]], align 8 |
| // CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP32]], align 8 |
| // CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8 |
| // CHECK1-NEXT: store i64 123, ptr [[TMP33]], align 8 |
| // CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9 |
| // CHECK1-NEXT: store i64 0, ptr [[TMP34]], align 8 |
| // CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10 |
| // CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4 |
| // CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11 |
| // CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4 |
| // CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12 |
| // CHECK1-NEXT: store i32 0, ptr [[TMP37]], align 4 |
| // CHECK1-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]]) |
| // CHECK1-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0 |
| // CHECK1-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]] |
| // CHECK1: omp_offload.failed8: |
| // CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]] |
| // CHECK1: omp_offload.cont9: |
| // CHECK1-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0 |
| // CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8 |
| // CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0 |
| // CHECK1-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8 |
| // CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP42]], align 8 |
| // CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0 |
| // CHECK1-NEXT: store i32 3, ptr [[TMP45]], align 4 |
| // CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1 |
| // CHECK1-NEXT: store i32 1, ptr [[TMP46]], align 4 |
| // CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2 |
| // CHECK1-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8 |
| // CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3 |
| // CHECK1-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8 |
| // CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4 |
| // CHECK1-NEXT: store ptr @.offload_sizes.3, ptr [[TMP49]], align 8 |
| // CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5 |
| // CHECK1-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP50]], align 8 |
| // CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP51]], align 8 |
| // CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP52]], align 8 |
| // CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8 |
| // CHECK1-NEXT: store i64 123, ptr [[TMP53]], align 8 |
| // CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9 |
| // CHECK1-NEXT: store i64 0, ptr [[TMP54]], align 8 |
| // CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10 |
| // CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4 |
| // CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11 |
| // CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4 |
| // CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12 |
| // CHECK1-NEXT: store i32 0, ptr [[TMP57]], align 4 |
| // CHECK1-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]]) |
| // CHECK1-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0 |
| // CHECK1-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]] |
| // CHECK1: omp_offload.failed16: |
| // CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT17]] |
| // CHECK1: omp_offload.cont17: |
| // CHECK1-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0 |
| // CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 8 |
| // CHECK1-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0 |
| // CHECK1-NEXT: store ptr [[A18]], ptr [[TMP61]], align 8 |
| // CHECK1-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP62]], align 8 |
| // CHECK1-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0 |
| // CHECK1-NEXT: store i32 3, ptr [[TMP65]], align 4 |
| // CHECK1-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1 |
| // CHECK1-NEXT: store i32 1, ptr [[TMP66]], align 4 |
| // CHECK1-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2 |
| // CHECK1-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 8 |
| // CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3 |
| // CHECK1-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 8 |
| // CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4 |
| // CHECK1-NEXT: store ptr @.offload_sizes.5, ptr [[TMP69]], align 8 |
| // CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5 |
| // CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP70]], align 8 |
| // CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP71]], align 8 |
| // CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP72]], align 8 |
| // CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8 |
| // CHECK1-NEXT: store i64 123, ptr [[TMP73]], align 8 |
| // CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9 |
| // CHECK1-NEXT: store i64 0, ptr [[TMP74]], align 8 |
| // CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10 |
| // CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4 |
| // CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11 |
| // CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4 |
| // CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12 |
| // CHECK1-NEXT: store i32 0, ptr [[TMP77]], align 4 |
| // CHECK1-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]]) |
| // CHECK1-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0 |
| // CHECK1-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]] |
| // CHECK1: omp_offload.failed24: |
| // CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT25]] |
| // CHECK1: omp_offload.cont25: |
| // CHECK1-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0 |
| // CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 8 |
| // CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0 |
| // CHECK1-NEXT: store ptr [[A26]], ptr [[TMP81]], align 8 |
| // CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP82]], align 8 |
| // CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0 |
| // CHECK1-NEXT: store i32 3, ptr [[TMP85]], align 4 |
| // CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1 |
| // CHECK1-NEXT: store i32 1, ptr [[TMP86]], align 4 |
| // CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2 |
| // CHECK1-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 8 |
| // CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3 |
| // CHECK1-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 8 |
| // CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4 |
| // CHECK1-NEXT: store ptr @.offload_sizes.7, ptr [[TMP89]], align 8 |
| // CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5 |
| // CHECK1-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP90]], align 8 |
| // CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP91]], align 8 |
| // CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7 |
| // CHECK1-NEXT: store ptr null, ptr [[TMP92]], align 8 |
| // CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8 |
| // CHECK1-NEXT: store i64 123, ptr [[TMP93]], align 8 |
| // CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9 |
| // CHECK1-NEXT: store i64 0, ptr [[TMP94]], align 8 |
| // CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10 |
| // CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4 |
| // CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11 |
| // CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4 |
| // CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12 |
| // CHECK1-NEXT: store i32 0, ptr [[TMP97]], align 4 |
| // CHECK1-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]]) |
| // CHECK1-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0 |
| // CHECK1-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]] |
| // CHECK1: omp_offload.failed32: |
| // CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT33]] |
| // CHECK1: omp_offload.cont33: |
| // CHECK1-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i64 0, i64 0 |
| // CHECK1-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4 |
| // CHECK1-NEXT: ret i32 [[TMP100]] |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36 |
| // CHECK1-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1:[0-9]+]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined, ptr [[TMP0]]) |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined |
| // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK1-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK1-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK1-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK1-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK1: cond.true: |
| // CHECK1-NEXT: br label [[COND_END:%.*]] |
| // CHECK1: cond.false: |
| // CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: br label [[COND_END]] |
| // CHECK1: cond.end: |
| // CHECK1-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK1-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK1: omp.inner.for.cond: |
| // CHECK1-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP8:![0-9]+]] |
| // CHECK1-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK1-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK1-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK1: omp.inner.for.body: |
| // CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK1-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK1-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP8]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK1: omp.inner.for.inc: |
| // CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK1-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]] |
| // CHECK1: omp.inner.for.end: |
| // CHECK1-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK1: omp.loop.exit: |
| // CHECK1-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK1-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK1-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK1: .omp.final.then: |
| // CHECK1-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK1-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK1: .omp.final.done: |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined.omp_outlined |
| // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK1-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK1-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK1-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK1-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK1-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK1-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 122 |
| // CHECK1-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK1: cond.true: |
| // CHECK1-NEXT: br label [[COND_END:%.*]] |
| // CHECK1: cond.false: |
| // CHECK1-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: br label [[COND_END]] |
| // CHECK1: cond.end: |
| // CHECK1-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK1-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK1: omp.inner.for.cond: |
| // CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12:![0-9]+]] |
| // CHECK1-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK1-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK1-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK1: omp.inner.for.body: |
| // CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK1-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK1-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK1-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64 |
| // CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK1-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK1-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK1: omp.body.continue: |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK1: omp.inner.for.inc: |
| // CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK1-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]] |
| // CHECK1: omp.inner.for.end: |
| // CHECK1-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK1: omp.loop.exit: |
| // CHECK1-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK1-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK1: .omp.final.then: |
| // CHECK1-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK1-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK1: .omp.final.done: |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40 |
| // CHECK1-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined, ptr [[TMP0]]) |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined |
| // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK1-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK1-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK1-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK1-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK1: cond.true: |
| // CHECK1-NEXT: br label [[COND_END:%.*]] |
| // CHECK1: cond.false: |
| // CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: br label [[COND_END]] |
| // CHECK1: cond.end: |
| // CHECK1-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK1-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK1: omp.inner.for.cond: |
| // CHECK1-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17:![0-9]+]] |
| // CHECK1-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK1-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK1-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK1: omp.inner.for.body: |
| // CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK1-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK1-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP17]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK1: omp.inner.for.inc: |
| // CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK1-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP18:![0-9]+]] |
| // CHECK1: omp.inner.for.end: |
| // CHECK1-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK1: omp.loop.exit: |
| // CHECK1-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK1-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK1-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK1: .omp.final.then: |
| // CHECK1-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK1-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK1: .omp.final.done: |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined.omp_outlined |
| // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK1-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK1-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK1-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK1-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK1-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK1-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 122 |
| // CHECK1-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK1: cond.true: |
| // CHECK1-NEXT: br label [[COND_END:%.*]] |
| // CHECK1: cond.false: |
| // CHECK1-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: br label [[COND_END]] |
| // CHECK1: cond.end: |
| // CHECK1-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK1-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK1: omp.inner.for.cond: |
| // CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20:![0-9]+]] |
| // CHECK1-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK1-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK1-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK1: omp.inner.for.body: |
| // CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK1-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK1-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK1-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64 |
| // CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK1-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK1-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK1: omp.body.continue: |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK1: omp.inner.for.inc: |
| // CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK1-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP21:![0-9]+]] |
| // CHECK1: omp.inner.for.end: |
| // CHECK1-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK1: omp.loop.exit: |
| // CHECK1-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK1-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK1: .omp.final.then: |
| // CHECK1-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK1-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK1: .omp.final.done: |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44 |
| // CHECK1-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined, ptr [[TMP0]]) |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined |
| // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK1-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK1-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK1-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK1-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK1: cond.true: |
| // CHECK1-NEXT: br label [[COND_END:%.*]] |
| // CHECK1: cond.false: |
| // CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: br label [[COND_END]] |
| // CHECK1: cond.end: |
| // CHECK1-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK1-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK1: omp.inner.for.cond: |
| // CHECK1-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23:![0-9]+]] |
| // CHECK1-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK1-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK1-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK1: omp.inner.for.body: |
| // CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK1-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK1-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP23]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK1: omp.inner.for.inc: |
| // CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK1-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP24:![0-9]+]] |
| // CHECK1: omp.inner.for.end: |
| // CHECK1-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK1: omp.loop.exit: |
| // CHECK1-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK1-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK1-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK1: .omp.final.then: |
| // CHECK1-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK1-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK1: .omp.final.done: |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined.omp_outlined |
| // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK1-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK1-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK1-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK1-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK1-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 61) |
| // CHECK1-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK1: omp.dispatch.cond: |
| // CHECK1-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV2:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], [[CONV2]] |
| // CHECK1-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK1: cond.true: |
| // CHECK1-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV3:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK1-NEXT: br label [[COND_END:%.*]] |
| // CHECK1: cond.false: |
| // CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: br label [[COND_END]] |
| // CHECK1: cond.end: |
| // CHECK1-NEXT: [[COND:%.*]] = phi i32 [ [[CONV3]], [[COND_TRUE]] ], [ [[TMP8]], [[COND_FALSE]] ] |
| // CHECK1-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[TMP9]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4 |
| // CHECK1-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[CMP4:%.*]] = icmp sle i32 [[TMP10]], [[TMP11]] |
| // CHECK1-NEXT: br i1 [[CMP4]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK1: omp.dispatch.body: |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK1: omp.inner.for.cond: |
| // CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26:![0-9]+]] |
| // CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK1-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]] |
| // CHECK1-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK1: omp.inner.for.body: |
| // CHECK1-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK1-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP14]], 1 |
| // CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK1-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP15:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK1-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP15]] to i64 |
| // CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK1-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK1-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK1: omp.body.continue: |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK1: omp.inner.for.inc: |
| // CHECK1-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK1-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP16]], 1 |
| // CHECK1-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP27:![0-9]+]] |
| // CHECK1: omp.inner.for.end: |
| // CHECK1-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK1: omp.dispatch.inc: |
| // CHECK1-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP17]], [[TMP18]] |
| // CHECK1-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: [[ADD8:%.*]] = add nsw i32 [[TMP19]], [[TMP20]] |
| // CHECK1-NEXT: store i32 [[ADD8]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK1: omp.dispatch.end: |
| // CHECK1-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK1-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0 |
| // CHECK1-NEXT: br i1 [[TMP22]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK1: .omp.final.then: |
| // CHECK1-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK1-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK1: .omp.final.done: |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49 |
| // CHECK1-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined, ptr [[TMP0]]) |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined |
| // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK1-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK1-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK1-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK1-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK1: cond.true: |
| // CHECK1-NEXT: br label [[COND_END:%.*]] |
| // CHECK1: cond.false: |
| // CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: br label [[COND_END]] |
| // CHECK1: cond.end: |
| // CHECK1-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK1-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK1: omp.inner.for.cond: |
| // CHECK1-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29:![0-9]+]] |
| // CHECK1-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK1-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK1-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK1: omp.inner.for.body: |
| // CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK1-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK1-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP29]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK1: omp.inner.for.inc: |
| // CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK1-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP30:![0-9]+]] |
| // CHECK1: omp.inner.for.end: |
| // CHECK1-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK1: omp.loop.exit: |
| // CHECK1-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK1-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK1-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK1: .omp.final.then: |
| // CHECK1-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK1-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK1: .omp.final.done: |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined.omp_outlined |
| // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK1-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK1-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK1-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK1-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK1-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 35, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 1) |
| // CHECK1-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK1: omp.dispatch.cond: |
| // CHECK1-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK1-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK1: omp.dispatch.body: |
| // CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK1: omp.inner.for.cond: |
| // CHECK1-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32:![0-9]+]] |
| // CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK1-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK1-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK1: omp.inner.for.body: |
| // CHECK1-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK1-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK1-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK1-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP12]] to i64 |
| // CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK1-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK1-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK1: omp.body.continue: |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK1: omp.inner.for.inc: |
| // CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK1-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK1-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP33:![0-9]+]] |
| // CHECK1: omp.inner.for.end: |
| // CHECK1-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK1: omp.dispatch.inc: |
| // CHECK1-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK1: omp.dispatch.end: |
| // CHECK1-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK1-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK1: .omp.final.then: |
| // CHECK1-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK1-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK1: .omp.final.done: |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54 |
| // CHECK1-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined, ptr [[TMP0]]) |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined |
| // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK1-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK1-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK1-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK1-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK1: cond.true: |
| // CHECK1-NEXT: br label [[COND_END:%.*]] |
| // CHECK1: cond.false: |
| // CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: br label [[COND_END]] |
| // CHECK1: cond.end: |
| // CHECK1-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK1-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK1-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK1: omp.inner.for.cond: |
| // CHECK1-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35:![0-9]+]] |
| // CHECK1-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK1-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK1-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK1: omp.inner.for.body: |
| // CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK1-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK1-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP35]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK1: omp.inner.for.inc: |
| // CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK1-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP36:![0-9]+]] |
| // CHECK1: omp.inner.for.end: |
| // CHECK1-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK1: omp.loop.exit: |
| // CHECK1-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK1-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK1-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK1: .omp.final.then: |
| // CHECK1-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK1-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK1: .omp.final.done: |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined.omp_outlined |
| // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK1-NEXT: entry: |
| // CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK1-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK1-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK1-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK1-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK1-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK1-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK1-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK1-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK1-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK1-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK1-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 35, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 61) |
| // CHECK1-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK1: omp.dispatch.cond: |
| // CHECK1-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK1-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK1-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK1: omp.dispatch.body: |
| // CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK1-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK1: omp.inner.for.cond: |
| // CHECK1-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38:![0-9]+]] |
| // CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK1-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK1-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK1: omp.inner.for.body: |
| // CHECK1-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK1-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK1-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK1-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK1-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP12]] to i64 |
| // CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK1-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK1-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK1: omp.body.continue: |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK1: omp.inner.for.inc: |
| // CHECK1-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK1-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK1-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP39:![0-9]+]] |
| // CHECK1: omp.inner.for.end: |
| // CHECK1-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK1: omp.dispatch.inc: |
| // CHECK1-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK1: omp.dispatch.end: |
| // CHECK1-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK1-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK1-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK1: .omp.final.then: |
| // CHECK1-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK1-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK1: .omp.final.done: |
| // CHECK1-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@_Z21teams_template_structv |
| // CHECK3-SAME: () #[[ATTR0:[0-9]+]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[V:%.*]] = alloca [[STRUCT_SS:%.*]], align 4 |
| // CHECK3-NEXT: [[CALL:%.*]] = call noundef i32 @_ZN2SSIiLi123ELx456EE3fooEv(ptr noundef nonnull align 4 dereferenceable(496) [[V]]) |
| // CHECK3-NEXT: ret i32 [[CALL]] |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@_ZN2SSIiLi123ELx456EE3fooEv |
| // CHECK3-SAME: (ptr noundef nonnull align 4 dereferenceable(496) [[THIS:%.*]]) #[[ATTR0]] comdat align 2 { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK3-NEXT: [[DOTOFFLOAD_BASEPTRS3:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[DOTOFFLOAD_PTRS4:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[DOTOFFLOAD_MAPPERS5:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[_TMP6:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK3-NEXT: [[DOTOFFLOAD_BASEPTRS11:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[DOTOFFLOAD_PTRS12:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[DOTOFFLOAD_MAPPERS13:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[_TMP14:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK3-NEXT: [[DOTOFFLOAD_BASEPTRS19:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[DOTOFFLOAD_PTRS20:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[DOTOFFLOAD_MAPPERS21:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[_TMP22:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK3-NEXT: [[DOTOFFLOAD_BASEPTRS27:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[DOTOFFLOAD_PTRS28:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[DOTOFFLOAD_MAPPERS29:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK3-NEXT: [[_TMP30:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP0]], align 4 |
| // CHECK3-NEXT: [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr [[A]], ptr [[TMP1]], align 4 |
| // CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP2]], align 4 |
| // CHECK3-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK3-NEXT: store i32 3, ptr [[TMP5]], align 4 |
| // CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK3-NEXT: store i32 1, ptr [[TMP6]], align 4 |
| // CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK3-NEXT: store ptr [[TMP3]], ptr [[TMP7]], align 4 |
| // CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK3-NEXT: store ptr [[TMP4]], ptr [[TMP8]], align 4 |
| // CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP9]], align 4 |
| // CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP10]], align 4 |
| // CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP11]], align 4 |
| // CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4 |
| // CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK3-NEXT: store i64 123, ptr [[TMP13]], align 8 |
| // CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK3-NEXT: store i64 0, ptr [[TMP14]], align 8 |
| // CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4 |
| // CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4 |
| // CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK3-NEXT: store i32 0, ptr [[TMP17]], align 4 |
| // CHECK3-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK3-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0 |
| // CHECK3-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK3: omp_offload.failed: |
| // CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]] |
| // CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK3: omp_offload.cont: |
| // CHECK3-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4 |
| // CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4 |
| // CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4 |
| // CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0 |
| // CHECK3-NEXT: store i32 3, ptr [[TMP25]], align 4 |
| // CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1 |
| // CHECK3-NEXT: store i32 1, ptr [[TMP26]], align 4 |
| // CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2 |
| // CHECK3-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4 |
| // CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3 |
| // CHECK3-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4 |
| // CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4 |
| // CHECK3-NEXT: store ptr @.offload_sizes.1, ptr [[TMP29]], align 4 |
| // CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5 |
| // CHECK3-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP30]], align 4 |
| // CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP31]], align 4 |
| // CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP32]], align 4 |
| // CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8 |
| // CHECK3-NEXT: store i64 123, ptr [[TMP33]], align 8 |
| // CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9 |
| // CHECK3-NEXT: store i64 0, ptr [[TMP34]], align 8 |
| // CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10 |
| // CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4 |
| // CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11 |
| // CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4 |
| // CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12 |
| // CHECK3-NEXT: store i32 0, ptr [[TMP37]], align 4 |
| // CHECK3-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]]) |
| // CHECK3-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0 |
| // CHECK3-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]] |
| // CHECK3: omp_offload.failed8: |
| // CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]] |
| // CHECK3: omp_offload.cont9: |
| // CHECK3-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4 |
| // CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4 |
| // CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4 |
| // CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0 |
| // CHECK3-NEXT: store i32 3, ptr [[TMP45]], align 4 |
| // CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1 |
| // CHECK3-NEXT: store i32 1, ptr [[TMP46]], align 4 |
| // CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2 |
| // CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4 |
| // CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3 |
| // CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4 |
| // CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4 |
| // CHECK3-NEXT: store ptr @.offload_sizes.3, ptr [[TMP49]], align 4 |
| // CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5 |
| // CHECK3-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP50]], align 4 |
| // CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4 |
| // CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4 |
| // CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8 |
| // CHECK3-NEXT: store i64 123, ptr [[TMP53]], align 8 |
| // CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9 |
| // CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8 |
| // CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10 |
| // CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4 |
| // CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11 |
| // CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4 |
| // CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12 |
| // CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4 |
| // CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]]) |
| // CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0 |
| // CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]] |
| // CHECK3: omp_offload.failed16: |
| // CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT17]] |
| // CHECK3: omp_offload.cont17: |
| // CHECK3-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 4 |
| // CHECK3-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr [[A18]], ptr [[TMP61]], align 4 |
| // CHECK3-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP62]], align 4 |
| // CHECK3-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0 |
| // CHECK3-NEXT: store i32 3, ptr [[TMP65]], align 4 |
| // CHECK3-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1 |
| // CHECK3-NEXT: store i32 1, ptr [[TMP66]], align 4 |
| // CHECK3-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2 |
| // CHECK3-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4 |
| // CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3 |
| // CHECK3-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4 |
| // CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4 |
| // CHECK3-NEXT: store ptr @.offload_sizes.5, ptr [[TMP69]], align 4 |
| // CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5 |
| // CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP70]], align 4 |
| // CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP71]], align 4 |
| // CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP72]], align 4 |
| // CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8 |
| // CHECK3-NEXT: store i64 123, ptr [[TMP73]], align 8 |
| // CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9 |
| // CHECK3-NEXT: store i64 0, ptr [[TMP74]], align 8 |
| // CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10 |
| // CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4 |
| // CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11 |
| // CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4 |
| // CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12 |
| // CHECK3-NEXT: store i32 0, ptr [[TMP77]], align 4 |
| // CHECK3-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]]) |
| // CHECK3-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0 |
| // CHECK3-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]] |
| // CHECK3: omp_offload.failed24: |
| // CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT25]] |
| // CHECK3: omp_offload.cont25: |
| // CHECK3-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 4 |
| // CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr [[A26]], ptr [[TMP81]], align 4 |
| // CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP82]], align 4 |
| // CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0 |
| // CHECK3-NEXT: store i32 3, ptr [[TMP85]], align 4 |
| // CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1 |
| // CHECK3-NEXT: store i32 1, ptr [[TMP86]], align 4 |
| // CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2 |
| // CHECK3-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 4 |
| // CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3 |
| // CHECK3-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 4 |
| // CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4 |
| // CHECK3-NEXT: store ptr @.offload_sizes.7, ptr [[TMP89]], align 4 |
| // CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5 |
| // CHECK3-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP90]], align 4 |
| // CHECK3-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP91]], align 4 |
| // CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7 |
| // CHECK3-NEXT: store ptr null, ptr [[TMP92]], align 4 |
| // CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8 |
| // CHECK3-NEXT: store i64 123, ptr [[TMP93]], align 8 |
| // CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9 |
| // CHECK3-NEXT: store i64 0, ptr [[TMP94]], align 8 |
| // CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10 |
| // CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4 |
| // CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11 |
| // CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4 |
| // CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12 |
| // CHECK3-NEXT: store i32 0, ptr [[TMP97]], align 4 |
| // CHECK3-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]]) |
| // CHECK3-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0 |
| // CHECK3-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]] |
| // CHECK3: omp_offload.failed32: |
| // CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT33]] |
| // CHECK3: omp_offload.cont33: |
| // CHECK3-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4 |
| // CHECK3-NEXT: ret i32 [[TMP100]] |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36 |
| // CHECK3-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1:[0-9]+]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined, ptr [[TMP0]]) |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined |
| // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK3-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK3-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK3-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK3: cond.true: |
| // CHECK3-NEXT: br label [[COND_END:%.*]] |
| // CHECK3: cond.false: |
| // CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: br label [[COND_END]] |
| // CHECK3: cond.end: |
| // CHECK3-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK3-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK3: omp.inner.for.cond: |
| // CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP9:![0-9]+]] |
| // CHECK3-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK3-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK3-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK3: omp.inner.for.body: |
| // CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP9]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK3: omp.inner.for.inc: |
| // CHECK3-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK3-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]] |
| // CHECK3: omp.inner.for.end: |
| // CHECK3-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK3: omp.loop.exit: |
| // CHECK3-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK3-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK3: .omp.final.then: |
| // CHECK3-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK3-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK3: .omp.final.done: |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined.omp_outlined |
| // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK3-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 122 |
| // CHECK3-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK3: cond.true: |
| // CHECK3-NEXT: br label [[COND_END:%.*]] |
| // CHECK3: cond.false: |
| // CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: br label [[COND_END]] |
| // CHECK3: cond.end: |
| // CHECK3-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK3-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK3: omp.inner.for.cond: |
| // CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13:![0-9]+]] |
| // CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK3-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK3-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK3: omp.inner.for.body: |
| // CHECK3-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK3-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK3-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i32 0, i32 [[TMP11]] |
| // CHECK3-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK3-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK3: omp.body.continue: |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK3: omp.inner.for.inc: |
| // CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK3-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK3-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP14:![0-9]+]] |
| // CHECK3: omp.inner.for.end: |
| // CHECK3-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK3: omp.loop.exit: |
| // CHECK3-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK3-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK3-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK3: .omp.final.then: |
| // CHECK3-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK3-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK3: .omp.final.done: |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40 |
| // CHECK3-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined, ptr [[TMP0]]) |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined |
| // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK3-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK3-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK3-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK3: cond.true: |
| // CHECK3-NEXT: br label [[COND_END:%.*]] |
| // CHECK3: cond.false: |
| // CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: br label [[COND_END]] |
| // CHECK3: cond.end: |
| // CHECK3-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK3-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK3: omp.inner.for.cond: |
| // CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18:![0-9]+]] |
| // CHECK3-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK3-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK3-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK3: omp.inner.for.body: |
| // CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP18]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK3: omp.inner.for.inc: |
| // CHECK3-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK3-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]] |
| // CHECK3: omp.inner.for.end: |
| // CHECK3-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK3: omp.loop.exit: |
| // CHECK3-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK3-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK3: .omp.final.then: |
| // CHECK3-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK3-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK3: .omp.final.done: |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined.omp_outlined |
| // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK3-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 122 |
| // CHECK3-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK3: cond.true: |
| // CHECK3-NEXT: br label [[COND_END:%.*]] |
| // CHECK3: cond.false: |
| // CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: br label [[COND_END]] |
| // CHECK3: cond.end: |
| // CHECK3-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK3-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK3: omp.inner.for.cond: |
| // CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP21:![0-9]+]] |
| // CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK3-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK3-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK3: omp.inner.for.body: |
| // CHECK3-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK3-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK3-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i32 0, i32 [[TMP11]] |
| // CHECK3-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK3-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK3: omp.body.continue: |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK3: omp.inner.for.inc: |
| // CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK3-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK3-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP22:![0-9]+]] |
| // CHECK3: omp.inner.for.end: |
| // CHECK3-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK3: omp.loop.exit: |
| // CHECK3-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK3-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK3-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK3: .omp.final.then: |
| // CHECK3-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK3-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK3: .omp.final.done: |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44 |
| // CHECK3-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined, ptr [[TMP0]]) |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined |
| // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK3-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK3-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK3-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK3: cond.true: |
| // CHECK3-NEXT: br label [[COND_END:%.*]] |
| // CHECK3: cond.false: |
| // CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: br label [[COND_END]] |
| // CHECK3: cond.end: |
| // CHECK3-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK3-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK3: omp.inner.for.cond: |
| // CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP24:![0-9]+]] |
| // CHECK3-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK3-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK3-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK3: omp.inner.for.body: |
| // CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP24]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK3: omp.inner.for.inc: |
| // CHECK3-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK3-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP25:![0-9]+]] |
| // CHECK3: omp.inner.for.end: |
| // CHECK3-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK3: omp.loop.exit: |
| // CHECK3-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK3-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK3: .omp.final.then: |
| // CHECK3-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK3-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK3: .omp.final.done: |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined.omp_outlined |
| // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK3-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 61) |
| // CHECK3-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK3: omp.dispatch.cond: |
| // CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], [[TMP6]] |
| // CHECK3-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK3: cond.true: |
| // CHECK3-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: br label [[COND_END:%.*]] |
| // CHECK3: cond.false: |
| // CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: br label [[COND_END]] |
| // CHECK3: cond.end: |
| // CHECK3-NEXT: [[COND:%.*]] = phi i32 [ [[TMP7]], [[COND_TRUE]] ], [ [[TMP8]], [[COND_FALSE]] ] |
| // CHECK3-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP9]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK3-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4 |
| // CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP10]], [[TMP11]] |
| // CHECK3-NEXT: br i1 [[CMP1]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK3: omp.dispatch.body: |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK3: omp.inner.for.cond: |
| // CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP27:![0-9]+]] |
| // CHECK3-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK3-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]] |
| // CHECK3-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK3: omp.inner.for.body: |
| // CHECK3-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK3-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP14]], 1 |
| // CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK3-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP15:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i32 0, i32 [[TMP15]] |
| // CHECK3-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK3-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK3: omp.body.continue: |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK3: omp.inner.for.inc: |
| // CHECK3-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP16]], 1 |
| // CHECK3-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP28:![0-9]+]] |
| // CHECK3: omp.inner.for.end: |
| // CHECK3-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK3: omp.dispatch.inc: |
| // CHECK3-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: [[ADD4:%.*]] = add nsw i32 [[TMP17]], [[TMP18]] |
| // CHECK3-NEXT: store i32 [[ADD4]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: [[ADD5:%.*]] = add nsw i32 [[TMP19]], [[TMP20]] |
| // CHECK3-NEXT: store i32 [[ADD5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK3: omp.dispatch.end: |
| // CHECK3-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK3-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0 |
| // CHECK3-NEXT: br i1 [[TMP22]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK3: .omp.final.then: |
| // CHECK3-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK3-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK3: .omp.final.done: |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49 |
| // CHECK3-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined, ptr [[TMP0]]) |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined |
| // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK3-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK3-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK3-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK3: cond.true: |
| // CHECK3-NEXT: br label [[COND_END:%.*]] |
| // CHECK3: cond.false: |
| // CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: br label [[COND_END]] |
| // CHECK3: cond.end: |
| // CHECK3-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK3-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK3: omp.inner.for.cond: |
| // CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP30:![0-9]+]] |
| // CHECK3-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK3-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK3-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK3: omp.inner.for.body: |
| // CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP30]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK3: omp.inner.for.inc: |
| // CHECK3-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK3-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP31:![0-9]+]] |
| // CHECK3: omp.inner.for.end: |
| // CHECK3-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK3: omp.loop.exit: |
| // CHECK3-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK3-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK3: .omp.final.then: |
| // CHECK3-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK3-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK3: .omp.final.done: |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined.omp_outlined |
| // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK3-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 35, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 1) |
| // CHECK3-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK3: omp.dispatch.cond: |
| // CHECK3-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK3-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK3-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK3: omp.dispatch.body: |
| // CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK3: omp.inner.for.cond: |
| // CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP33:![0-9]+]] |
| // CHECK3-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK3-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK3-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK3: omp.inner.for.body: |
| // CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK3-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK3-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i32 0, i32 [[TMP12]] |
| // CHECK3-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK3-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK3: omp.body.continue: |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK3: omp.inner.for.inc: |
| // CHECK3-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK3-NEXT: [[ADD1:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK3-NEXT: store i32 [[ADD1]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP34:![0-9]+]] |
| // CHECK3: omp.inner.for.end: |
| // CHECK3-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK3: omp.dispatch.inc: |
| // CHECK3-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK3: omp.dispatch.end: |
| // CHECK3-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK3-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK3: .omp.final.then: |
| // CHECK3-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK3-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK3: .omp.final.done: |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54 |
| // CHECK3-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined, ptr [[TMP0]]) |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined |
| // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK3-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK3-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK3-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK3: cond.true: |
| // CHECK3-NEXT: br label [[COND_END:%.*]] |
| // CHECK3: cond.false: |
| // CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: br label [[COND_END]] |
| // CHECK3: cond.end: |
| // CHECK3-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK3-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK3: omp.inner.for.cond: |
| // CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP36:![0-9]+]] |
| // CHECK3-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK3-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK3-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK3: omp.inner.for.body: |
| // CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP36]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK3: omp.inner.for.inc: |
| // CHECK3-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK3-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP37:![0-9]+]] |
| // CHECK3: omp.inner.for.end: |
| // CHECK3-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK3: omp.loop.exit: |
| // CHECK3-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK3-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK3: .omp.final.then: |
| // CHECK3-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK3-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK3: .omp.final.done: |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined.omp_outlined |
| // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK3-NEXT: entry: |
| // CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK3-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK3-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK3-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 35, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 61) |
| // CHECK3-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK3: omp.dispatch.cond: |
| // CHECK3-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK3-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK3-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK3: omp.dispatch.body: |
| // CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK3-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK3: omp.inner.for.cond: |
| // CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP39:![0-9]+]] |
| // CHECK3-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK3-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK3-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK3: omp.inner.for.body: |
| // CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK3-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK3-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK3-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i32 0, i32 [[TMP12]] |
| // CHECK3-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK3-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK3: omp.body.continue: |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK3: omp.inner.for.inc: |
| // CHECK3-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK3-NEXT: [[ADD1:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK3-NEXT: store i32 [[ADD1]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK3-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP40:![0-9]+]] |
| // CHECK3: omp.inner.for.end: |
| // CHECK3-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK3: omp.dispatch.inc: |
| // CHECK3-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK3: omp.dispatch.end: |
| // CHECK3-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK3-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK3-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK3: .omp.final.then: |
| // CHECK3-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK3-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK3: .omp.final.done: |
| // CHECK3-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@_Z21teams_template_structv |
| // CHECK5-SAME: () #[[ATTR0:[0-9]+]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[V:%.*]] = alloca [[STRUCT_SS:%.*]], align 4 |
| // CHECK5-NEXT: [[CALL:%.*]] = call noundef signext i32 @_ZN2SSIiLi123ELx456EE3fooEv(ptr noundef nonnull align 4 dereferenceable(496) [[V]]) |
| // CHECK5-NEXT: ret i32 [[CALL]] |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@_ZN2SSIiLi123ELx456EE3fooEv |
| // CHECK5-SAME: (ptr noundef nonnull align 4 dereferenceable(496) [[THIS:%.*]]) #[[ATTR0]] comdat { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_BASEPTRS3:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_PTRS4:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_MAPPERS5:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[_TMP6:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_BASEPTRS11:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_PTRS12:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_MAPPERS13:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[_TMP14:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_BASEPTRS19:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_PTRS20:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_MAPPERS21:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[_TMP22:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_BASEPTRS27:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_PTRS28:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[DOTOFFLOAD_MAPPERS29:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK5-NEXT: [[_TMP30:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP0]], align 8 |
| // CHECK5-NEXT: [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK5-NEXT: store ptr [[A]], ptr [[TMP1]], align 8 |
| // CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP2]], align 8 |
| // CHECK5-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK5-NEXT: store i32 3, ptr [[TMP5]], align 4 |
| // CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK5-NEXT: store i32 1, ptr [[TMP6]], align 4 |
| // CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK5-NEXT: store ptr [[TMP3]], ptr [[TMP7]], align 8 |
| // CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK5-NEXT: store ptr [[TMP4]], ptr [[TMP8]], align 8 |
| // CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK5-NEXT: store ptr @.offload_sizes, ptr [[TMP9]], align 8 |
| // CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK5-NEXT: store ptr @.offload_maptypes, ptr [[TMP10]], align 8 |
| // CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP11]], align 8 |
| // CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP12]], align 8 |
| // CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK5-NEXT: store i64 123, ptr [[TMP13]], align 8 |
| // CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK5-NEXT: store i64 0, ptr [[TMP14]], align 8 |
| // CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4 |
| // CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4 |
| // CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK5-NEXT: store i32 0, ptr [[TMP17]], align 4 |
| // CHECK5-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK5-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0 |
| // CHECK5-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK5: omp_offload.failed: |
| // CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]] |
| // CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK5: omp_offload.cont: |
| // CHECK5-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0 |
| // CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 8 |
| // CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0 |
| // CHECK5-NEXT: store ptr [[A2]], ptr [[TMP21]], align 8 |
| // CHECK5-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i64 0, i64 0 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP22]], align 8 |
| // CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0 |
| // CHECK5-NEXT: store i32 3, ptr [[TMP25]], align 4 |
| // CHECK5-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1 |
| // CHECK5-NEXT: store i32 1, ptr [[TMP26]], align 4 |
| // CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2 |
| // CHECK5-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8 |
| // CHECK5-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3 |
| // CHECK5-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8 |
| // CHECK5-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4 |
| // CHECK5-NEXT: store ptr @.offload_sizes.1, ptr [[TMP29]], align 8 |
| // CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5 |
| // CHECK5-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP30]], align 8 |
| // CHECK5-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP31]], align 8 |
| // CHECK5-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP32]], align 8 |
| // CHECK5-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8 |
| // CHECK5-NEXT: store i64 123, ptr [[TMP33]], align 8 |
| // CHECK5-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9 |
| // CHECK5-NEXT: store i64 0, ptr [[TMP34]], align 8 |
| // CHECK5-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10 |
| // CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4 |
| // CHECK5-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11 |
| // CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4 |
| // CHECK5-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12 |
| // CHECK5-NEXT: store i32 0, ptr [[TMP37]], align 4 |
| // CHECK5-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]]) |
| // CHECK5-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0 |
| // CHECK5-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]] |
| // CHECK5: omp_offload.failed8: |
| // CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT9]] |
| // CHECK5: omp_offload.cont9: |
| // CHECK5-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0 |
| // CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 8 |
| // CHECK5-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0 |
| // CHECK5-NEXT: store ptr [[A10]], ptr [[TMP41]], align 8 |
| // CHECK5-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i64 0, i64 0 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP42]], align 8 |
| // CHECK5-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0 |
| // CHECK5-NEXT: store i32 3, ptr [[TMP45]], align 4 |
| // CHECK5-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1 |
| // CHECK5-NEXT: store i32 1, ptr [[TMP46]], align 4 |
| // CHECK5-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2 |
| // CHECK5-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 8 |
| // CHECK5-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3 |
| // CHECK5-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 8 |
| // CHECK5-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4 |
| // CHECK5-NEXT: store ptr @.offload_sizes.3, ptr [[TMP49]], align 8 |
| // CHECK5-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5 |
| // CHECK5-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP50]], align 8 |
| // CHECK5-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP51]], align 8 |
| // CHECK5-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP52]], align 8 |
| // CHECK5-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8 |
| // CHECK5-NEXT: store i64 123, ptr [[TMP53]], align 8 |
| // CHECK5-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9 |
| // CHECK5-NEXT: store i64 0, ptr [[TMP54]], align 8 |
| // CHECK5-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10 |
| // CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4 |
| // CHECK5-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11 |
| // CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4 |
| // CHECK5-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12 |
| // CHECK5-NEXT: store i32 0, ptr [[TMP57]], align 4 |
| // CHECK5-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]]) |
| // CHECK5-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0 |
| // CHECK5-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]] |
| // CHECK5: omp_offload.failed16: |
| // CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT17]] |
| // CHECK5: omp_offload.cont17: |
| // CHECK5-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0 |
| // CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 8 |
| // CHECK5-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0 |
| // CHECK5-NEXT: store ptr [[A18]], ptr [[TMP61]], align 8 |
| // CHECK5-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i64 0, i64 0 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP62]], align 8 |
| // CHECK5-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0 |
| // CHECK5-NEXT: store i32 3, ptr [[TMP65]], align 4 |
| // CHECK5-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1 |
| // CHECK5-NEXT: store i32 1, ptr [[TMP66]], align 4 |
| // CHECK5-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2 |
| // CHECK5-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 8 |
| // CHECK5-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3 |
| // CHECK5-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 8 |
| // CHECK5-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4 |
| // CHECK5-NEXT: store ptr @.offload_sizes.5, ptr [[TMP69]], align 8 |
| // CHECK5-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5 |
| // CHECK5-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP70]], align 8 |
| // CHECK5-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP71]], align 8 |
| // CHECK5-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP72]], align 8 |
| // CHECK5-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8 |
| // CHECK5-NEXT: store i64 123, ptr [[TMP73]], align 8 |
| // CHECK5-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9 |
| // CHECK5-NEXT: store i64 0, ptr [[TMP74]], align 8 |
| // CHECK5-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10 |
| // CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4 |
| // CHECK5-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11 |
| // CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4 |
| // CHECK5-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12 |
| // CHECK5-NEXT: store i32 0, ptr [[TMP77]], align 4 |
| // CHECK5-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]]) |
| // CHECK5-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0 |
| // CHECK5-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]] |
| // CHECK5: omp_offload.failed24: |
| // CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT25]] |
| // CHECK5: omp_offload.cont25: |
| // CHECK5-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0 |
| // CHECK5-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 8 |
| // CHECK5-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0 |
| // CHECK5-NEXT: store ptr [[A26]], ptr [[TMP81]], align 8 |
| // CHECK5-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i64 0, i64 0 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP82]], align 8 |
| // CHECK5-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0 |
| // CHECK5-NEXT: store i32 3, ptr [[TMP85]], align 4 |
| // CHECK5-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1 |
| // CHECK5-NEXT: store i32 1, ptr [[TMP86]], align 4 |
| // CHECK5-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2 |
| // CHECK5-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 8 |
| // CHECK5-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3 |
| // CHECK5-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 8 |
| // CHECK5-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4 |
| // CHECK5-NEXT: store ptr @.offload_sizes.7, ptr [[TMP89]], align 8 |
| // CHECK5-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5 |
| // CHECK5-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP90]], align 8 |
| // CHECK5-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP91]], align 8 |
| // CHECK5-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7 |
| // CHECK5-NEXT: store ptr null, ptr [[TMP92]], align 8 |
| // CHECK5-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8 |
| // CHECK5-NEXT: store i64 123, ptr [[TMP93]], align 8 |
| // CHECK5-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9 |
| // CHECK5-NEXT: store i64 0, ptr [[TMP94]], align 8 |
| // CHECK5-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10 |
| // CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4 |
| // CHECK5-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11 |
| // CHECK5-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4 |
| // CHECK5-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12 |
| // CHECK5-NEXT: store i32 0, ptr [[TMP97]], align 4 |
| // CHECK5-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]]) |
| // CHECK5-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0 |
| // CHECK5-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]] |
| // CHECK5: omp_offload.failed32: |
| // CHECK5-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK5-NEXT: br label [[OMP_OFFLOAD_CONT33]] |
| // CHECK5: omp_offload.cont33: |
| // CHECK5-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i64 0, i64 0 |
| // CHECK5-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4 |
| // CHECK5-NEXT: ret i32 [[TMP100]] |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36 |
| // CHECK5-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1:[0-9]+]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined, ptr [[TMP0]]) |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined |
| // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK5-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK5-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK5-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK5-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK5: cond.true: |
| // CHECK5-NEXT: br label [[COND_END:%.*]] |
| // CHECK5: cond.false: |
| // CHECK5-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: br label [[COND_END]] |
| // CHECK5: cond.end: |
| // CHECK5-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK5-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK5: omp.inner.for.cond: |
| // CHECK5-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP8:![0-9]+]] |
| // CHECK5-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK5-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK5-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK5: omp.inner.for.body: |
| // CHECK5-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK5-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK5-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK5-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK5-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP8]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK5: omp.inner.for.inc: |
| // CHECK5-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK5-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK5-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK5-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP8]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]] |
| // CHECK5: omp.inner.for.end: |
| // CHECK5-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK5: omp.loop.exit: |
| // CHECK5-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK5-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK5-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK5: .omp.final.then: |
| // CHECK5-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK5-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK5: .omp.final.done: |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined.omp_outlined |
| // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK5-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK5-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK5-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK5-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK5-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK5-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 122 |
| // CHECK5-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK5: cond.true: |
| // CHECK5-NEXT: br label [[COND_END:%.*]] |
| // CHECK5: cond.false: |
| // CHECK5-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: br label [[COND_END]] |
| // CHECK5: cond.end: |
| // CHECK5-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK5-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK5: omp.inner.for.cond: |
| // CHECK5-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12:![0-9]+]] |
| // CHECK5-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK5-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK5-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK5: omp.inner.for.body: |
| // CHECK5-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK5-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK5-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK5-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK5-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64 |
| // CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK5-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK5-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK5: omp.body.continue: |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK5: omp.inner.for.inc: |
| // CHECK5-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK5-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK5-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]] |
| // CHECK5: omp.inner.for.end: |
| // CHECK5-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK5: omp.loop.exit: |
| // CHECK5-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK5-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK5-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK5: .omp.final.then: |
| // CHECK5-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK5-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK5: .omp.final.done: |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40 |
| // CHECK5-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined, ptr [[TMP0]]) |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined |
| // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK5-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK5-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK5-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK5-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK5: cond.true: |
| // CHECK5-NEXT: br label [[COND_END:%.*]] |
| // CHECK5: cond.false: |
| // CHECK5-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: br label [[COND_END]] |
| // CHECK5: cond.end: |
| // CHECK5-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK5-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK5: omp.inner.for.cond: |
| // CHECK5-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17:![0-9]+]] |
| // CHECK5-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK5-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK5-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK5: omp.inner.for.body: |
| // CHECK5-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK5-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK5-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK5-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK5-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP17]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK5: omp.inner.for.inc: |
| // CHECK5-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK5-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK5-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK5-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP18:![0-9]+]] |
| // CHECK5: omp.inner.for.end: |
| // CHECK5-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK5: omp.loop.exit: |
| // CHECK5-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK5-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK5-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK5: .omp.final.then: |
| // CHECK5-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK5-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK5: .omp.final.done: |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined.omp_outlined |
| // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK5-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK5-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK5-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK5-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK5-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK5-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 122 |
| // CHECK5-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK5: cond.true: |
| // CHECK5-NEXT: br label [[COND_END:%.*]] |
| // CHECK5: cond.false: |
| // CHECK5-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: br label [[COND_END]] |
| // CHECK5: cond.end: |
| // CHECK5-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK5-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK5: omp.inner.for.cond: |
| // CHECK5-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20:![0-9]+]] |
| // CHECK5-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK5-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK5-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK5: omp.inner.for.body: |
| // CHECK5-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK5-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK5-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK5-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK5-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64 |
| // CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK5-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK5-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK5: omp.body.continue: |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK5: omp.inner.for.inc: |
| // CHECK5-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK5-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK5-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP21:![0-9]+]] |
| // CHECK5: omp.inner.for.end: |
| // CHECK5-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK5: omp.loop.exit: |
| // CHECK5-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK5-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK5-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK5: .omp.final.then: |
| // CHECK5-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK5-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK5: .omp.final.done: |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44 |
| // CHECK5-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined, ptr [[TMP0]]) |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined |
| // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK5-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK5-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK5-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK5-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK5: cond.true: |
| // CHECK5-NEXT: br label [[COND_END:%.*]] |
| // CHECK5: cond.false: |
| // CHECK5-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: br label [[COND_END]] |
| // CHECK5: cond.end: |
| // CHECK5-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK5-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK5: omp.inner.for.cond: |
| // CHECK5-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23:![0-9]+]] |
| // CHECK5-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK5-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK5-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK5: omp.inner.for.body: |
| // CHECK5-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK5-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK5-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK5-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK5-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP23]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK5: omp.inner.for.inc: |
| // CHECK5-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK5-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK5-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK5-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP24:![0-9]+]] |
| // CHECK5: omp.inner.for.end: |
| // CHECK5-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK5: omp.loop.exit: |
| // CHECK5-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK5-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK5-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK5: .omp.final.then: |
| // CHECK5-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK5-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK5: .omp.final.done: |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined.omp_outlined |
| // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK5-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK5-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK5-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK5-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK5-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 61) |
| // CHECK5-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK5: omp.dispatch.cond: |
| // CHECK5-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV2:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], [[CONV2]] |
| // CHECK5-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK5: cond.true: |
| // CHECK5-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV3:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK5-NEXT: br label [[COND_END:%.*]] |
| // CHECK5: cond.false: |
| // CHECK5-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: br label [[COND_END]] |
| // CHECK5: cond.end: |
| // CHECK5-NEXT: [[COND:%.*]] = phi i32 [ [[CONV3]], [[COND_TRUE]] ], [ [[TMP8]], [[COND_FALSE]] ] |
| // CHECK5-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[TMP9]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK5-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4 |
| // CHECK5-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[CMP4:%.*]] = icmp sle i32 [[TMP10]], [[TMP11]] |
| // CHECK5-NEXT: br i1 [[CMP4]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK5: omp.dispatch.body: |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK5: omp.inner.for.cond: |
| // CHECK5-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26:![0-9]+]] |
| // CHECK5-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK5-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]] |
| // CHECK5-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK5: omp.inner.for.body: |
| // CHECK5-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK5-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP14]], 1 |
| // CHECK5-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK5-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP15:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK5-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP15]] to i64 |
| // CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK5-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK5-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK5: omp.body.continue: |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK5: omp.inner.for.inc: |
| // CHECK5-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK5-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP16]], 1 |
| // CHECK5-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP27:![0-9]+]] |
| // CHECK5: omp.inner.for.end: |
| // CHECK5-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK5: omp.dispatch.inc: |
| // CHECK5-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP17]], [[TMP18]] |
| // CHECK5-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: [[ADD8:%.*]] = add nsw i32 [[TMP19]], [[TMP20]] |
| // CHECK5-NEXT: store i32 [[ADD8]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK5: omp.dispatch.end: |
| // CHECK5-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK5-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0 |
| // CHECK5-NEXT: br i1 [[TMP22]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK5: .omp.final.then: |
| // CHECK5-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK5-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK5: .omp.final.done: |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49 |
| // CHECK5-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined, ptr [[TMP0]]) |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined |
| // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK5-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK5-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK5-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK5-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK5: cond.true: |
| // CHECK5-NEXT: br label [[COND_END:%.*]] |
| // CHECK5: cond.false: |
| // CHECK5-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: br label [[COND_END]] |
| // CHECK5: cond.end: |
| // CHECK5-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK5-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK5: omp.inner.for.cond: |
| // CHECK5-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29:![0-9]+]] |
| // CHECK5-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK5-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK5-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK5: omp.inner.for.body: |
| // CHECK5-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK5-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK5-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK5-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK5-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP29]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK5: omp.inner.for.inc: |
| // CHECK5-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK5-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK5-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK5-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP30:![0-9]+]] |
| // CHECK5: omp.inner.for.end: |
| // CHECK5-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK5: omp.loop.exit: |
| // CHECK5-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK5-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK5-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK5: .omp.final.then: |
| // CHECK5-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK5-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK5: .omp.final.done: |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined.omp_outlined |
| // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK5-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK5-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK5-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK5-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK5-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 1073741859, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 1) |
| // CHECK5-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK5: omp.dispatch.cond: |
| // CHECK5-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK5-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK5-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK5: omp.dispatch.body: |
| // CHECK5-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK5: omp.inner.for.cond: |
| // CHECK5-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32:![0-9]+]] |
| // CHECK5-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK5-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK5-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK5: omp.inner.for.body: |
| // CHECK5-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK5-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK5-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK5-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK5-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP12]] to i64 |
| // CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK5-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK5-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK5: omp.body.continue: |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK5: omp.inner.for.inc: |
| // CHECK5-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK5-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK5-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP33:![0-9]+]] |
| // CHECK5: omp.inner.for.end: |
| // CHECK5-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK5: omp.dispatch.inc: |
| // CHECK5-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK5: omp.dispatch.end: |
| // CHECK5-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK5-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK5: .omp.final.then: |
| // CHECK5-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK5-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK5: .omp.final.done: |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54 |
| // CHECK5-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined, ptr [[TMP0]]) |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined |
| // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK5-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK5-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK5-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK5-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK5: cond.true: |
| // CHECK5-NEXT: br label [[COND_END:%.*]] |
| // CHECK5: cond.false: |
| // CHECK5-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: br label [[COND_END]] |
| // CHECK5: cond.end: |
| // CHECK5-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK5-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK5-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK5: omp.inner.for.cond: |
| // CHECK5-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35:![0-9]+]] |
| // CHECK5-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK5-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK5-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK5: omp.inner.for.body: |
| // CHECK5-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK5-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK5-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK5-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK5-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP35]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK5: omp.inner.for.inc: |
| // CHECK5-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK5-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK5-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK5-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP36:![0-9]+]] |
| // CHECK5: omp.inner.for.end: |
| // CHECK5-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK5: omp.loop.exit: |
| // CHECK5-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK5-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK5-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK5: .omp.final.then: |
| // CHECK5-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK5-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK5: .omp.final.done: |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK5-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined.omp_outlined |
| // CHECK5-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK5-NEXT: entry: |
| // CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK5-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK5-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK5-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK5-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK5-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK5-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK5-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK5-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK5-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK5-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK5-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK5-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 1073741859, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 61) |
| // CHECK5-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK5: omp.dispatch.cond: |
| // CHECK5-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK5-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK5-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK5: omp.dispatch.body: |
| // CHECK5-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK5-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK5: omp.inner.for.cond: |
| // CHECK5-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38:![0-9]+]] |
| // CHECK5-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK5-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK5-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK5: omp.inner.for.body: |
| // CHECK5-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK5-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK5-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK5-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK5-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK5-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP12]] to i64 |
| // CHECK5-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK5-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK5-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK5: omp.body.continue: |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK5: omp.inner.for.inc: |
| // CHECK5-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK5-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK5-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK5-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP39:![0-9]+]] |
| // CHECK5: omp.inner.for.end: |
| // CHECK5-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK5: omp.dispatch.inc: |
| // CHECK5-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK5: omp.dispatch.end: |
| // CHECK5-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK5-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK5-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK5: .omp.final.then: |
| // CHECK5-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK5-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK5: .omp.final.done: |
| // CHECK5-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@_Z21teams_template_structv |
| // CHECK7-SAME: () #[[ATTR0:[0-9]+]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[V:%.*]] = alloca [[STRUCT_SS:%.*]], align 4 |
| // CHECK7-NEXT: [[CALL:%.*]] = call noundef i32 @_ZN2SSIiLi123ELx456EE3fooEv(ptr noundef nonnull align 4 dereferenceable(496) [[V]]) |
| // CHECK7-NEXT: ret i32 [[CALL]] |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@_ZN2SSIiLi123ELx456EE3fooEv |
| // CHECK7-SAME: (ptr noundef nonnull align 4 dereferenceable(496) [[THIS:%.*]]) #[[ATTR0]] comdat align 2 { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK7-NEXT: [[DOTOFFLOAD_BASEPTRS3:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[DOTOFFLOAD_PTRS4:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[DOTOFFLOAD_MAPPERS5:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[_TMP6:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK7-NEXT: [[DOTOFFLOAD_BASEPTRS11:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[DOTOFFLOAD_PTRS12:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[DOTOFFLOAD_MAPPERS13:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[_TMP14:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK7-NEXT: [[DOTOFFLOAD_BASEPTRS19:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[DOTOFFLOAD_PTRS20:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[DOTOFFLOAD_MAPPERS21:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[_TMP22:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[KERNEL_ARGS23:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK7-NEXT: [[DOTOFFLOAD_BASEPTRS27:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[DOTOFFLOAD_PTRS28:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[DOTOFFLOAD_MAPPERS29:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK7-NEXT: [[_TMP30:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP0]], align 4 |
| // CHECK7-NEXT: [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr [[A]], ptr [[TMP1]], align 4 |
| // CHECK7-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP2]], align 4 |
| // CHECK7-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK7-NEXT: store i32 3, ptr [[TMP5]], align 4 |
| // CHECK7-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK7-NEXT: store i32 1, ptr [[TMP6]], align 4 |
| // CHECK7-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK7-NEXT: store ptr [[TMP3]], ptr [[TMP7]], align 4 |
| // CHECK7-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK7-NEXT: store ptr [[TMP4]], ptr [[TMP8]], align 4 |
| // CHECK7-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK7-NEXT: store ptr @.offload_sizes, ptr [[TMP9]], align 4 |
| // CHECK7-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK7-NEXT: store ptr @.offload_maptypes, ptr [[TMP10]], align 4 |
| // CHECK7-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP11]], align 4 |
| // CHECK7-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP12]], align 4 |
| // CHECK7-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK7-NEXT: store i64 123, ptr [[TMP13]], align 8 |
| // CHECK7-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK7-NEXT: store i64 0, ptr [[TMP14]], align 8 |
| // CHECK7-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4 |
| // CHECK7-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4 |
| // CHECK7-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK7-NEXT: store i32 0, ptr [[TMP17]], align 4 |
| // CHECK7-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK7-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0 |
| // CHECK7-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK7: omp_offload.failed: |
| // CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36(ptr [[THIS1]]) #[[ATTR2:[0-9]+]] |
| // CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK7: omp_offload.cont: |
| // CHECK7-NEXT: [[A2:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP20]], align 4 |
| // CHECK7-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr [[A2]], ptr [[TMP21]], align 4 |
| // CHECK7-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS5]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP22]], align 4 |
| // CHECK7-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS3]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS4]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0 |
| // CHECK7-NEXT: store i32 3, ptr [[TMP25]], align 4 |
| // CHECK7-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1 |
| // CHECK7-NEXT: store i32 1, ptr [[TMP26]], align 4 |
| // CHECK7-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2 |
| // CHECK7-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4 |
| // CHECK7-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3 |
| // CHECK7-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4 |
| // CHECK7-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4 |
| // CHECK7-NEXT: store ptr @.offload_sizes.1, ptr [[TMP29]], align 4 |
| // CHECK7-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5 |
| // CHECK7-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP30]], align 4 |
| // CHECK7-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP31]], align 4 |
| // CHECK7-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP32]], align 4 |
| // CHECK7-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8 |
| // CHECK7-NEXT: store i64 123, ptr [[TMP33]], align 8 |
| // CHECK7-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9 |
| // CHECK7-NEXT: store i64 0, ptr [[TMP34]], align 8 |
| // CHECK7-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10 |
| // CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4 |
| // CHECK7-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11 |
| // CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4 |
| // CHECK7-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12 |
| // CHECK7-NEXT: store i32 0, ptr [[TMP37]], align 4 |
| // CHECK7-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.region_id, ptr [[KERNEL_ARGS7]]) |
| // CHECK7-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0 |
| // CHECK7-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]] |
| // CHECK7: omp_offload.failed8: |
| // CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT9]] |
| // CHECK7: omp_offload.cont9: |
| // CHECK7-NEXT: [[A10:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP40:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP40]], align 4 |
| // CHECK7-NEXT: [[TMP41:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr [[A10]], ptr [[TMP41]], align 4 |
| // CHECK7-NEXT: [[TMP42:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS13]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP42]], align 4 |
| // CHECK7-NEXT: [[TMP43:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS11]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP44:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS12]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0 |
| // CHECK7-NEXT: store i32 3, ptr [[TMP45]], align 4 |
| // CHECK7-NEXT: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1 |
| // CHECK7-NEXT: store i32 1, ptr [[TMP46]], align 4 |
| // CHECK7-NEXT: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2 |
| // CHECK7-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4 |
| // CHECK7-NEXT: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3 |
| // CHECK7-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4 |
| // CHECK7-NEXT: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4 |
| // CHECK7-NEXT: store ptr @.offload_sizes.3, ptr [[TMP49]], align 4 |
| // CHECK7-NEXT: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5 |
| // CHECK7-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP50]], align 4 |
| // CHECK7-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP51]], align 4 |
| // CHECK7-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP52]], align 4 |
| // CHECK7-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8 |
| // CHECK7-NEXT: store i64 123, ptr [[TMP53]], align 8 |
| // CHECK7-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9 |
| // CHECK7-NEXT: store i64 0, ptr [[TMP54]], align 8 |
| // CHECK7-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10 |
| // CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP55]], align 4 |
| // CHECK7-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11 |
| // CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4 |
| // CHECK7-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12 |
| // CHECK7-NEXT: store i32 0, ptr [[TMP57]], align 4 |
| // CHECK7-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.region_id, ptr [[KERNEL_ARGS15]]) |
| // CHECK7-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0 |
| // CHECK7-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]] |
| // CHECK7: omp_offload.failed16: |
| // CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT17]] |
| // CHECK7: omp_offload.cont17: |
| // CHECK7-NEXT: [[A18:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP60:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP60]], align 4 |
| // CHECK7-NEXT: [[TMP61:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr [[A18]], ptr [[TMP61]], align 4 |
| // CHECK7-NEXT: [[TMP62:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS21]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP62]], align 4 |
| // CHECK7-NEXT: [[TMP63:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS19]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP64:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS20]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 0 |
| // CHECK7-NEXT: store i32 3, ptr [[TMP65]], align 4 |
| // CHECK7-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 1 |
| // CHECK7-NEXT: store i32 1, ptr [[TMP66]], align 4 |
| // CHECK7-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 2 |
| // CHECK7-NEXT: store ptr [[TMP63]], ptr [[TMP67]], align 4 |
| // CHECK7-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 3 |
| // CHECK7-NEXT: store ptr [[TMP64]], ptr [[TMP68]], align 4 |
| // CHECK7-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 4 |
| // CHECK7-NEXT: store ptr @.offload_sizes.5, ptr [[TMP69]], align 4 |
| // CHECK7-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 5 |
| // CHECK7-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP70]], align 4 |
| // CHECK7-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 6 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP71]], align 4 |
| // CHECK7-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 7 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP72]], align 4 |
| // CHECK7-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 8 |
| // CHECK7-NEXT: store i64 123, ptr [[TMP73]], align 8 |
| // CHECK7-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 9 |
| // CHECK7-NEXT: store i64 0, ptr [[TMP74]], align 8 |
| // CHECK7-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 10 |
| // CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP75]], align 4 |
| // CHECK7-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 11 |
| // CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP76]], align 4 |
| // CHECK7-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS23]], i32 0, i32 12 |
| // CHECK7-NEXT: store i32 0, ptr [[TMP77]], align 4 |
| // CHECK7-NEXT: [[TMP78:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.region_id, ptr [[KERNEL_ARGS23]]) |
| // CHECK7-NEXT: [[TMP79:%.*]] = icmp ne i32 [[TMP78]], 0 |
| // CHECK7-NEXT: br i1 [[TMP79]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]] |
| // CHECK7: omp_offload.failed24: |
| // CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT25]] |
| // CHECK7: omp_offload.cont25: |
| // CHECK7-NEXT: [[A26:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP80:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr [[THIS1]], ptr [[TMP80]], align 4 |
| // CHECK7-NEXT: [[TMP81:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr [[A26]], ptr [[TMP81]], align 4 |
| // CHECK7-NEXT: [[TMP82:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS29]], i32 0, i32 0 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP82]], align 4 |
| // CHECK7-NEXT: [[TMP83:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS27]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP84:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS28]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP85:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0 |
| // CHECK7-NEXT: store i32 3, ptr [[TMP85]], align 4 |
| // CHECK7-NEXT: [[TMP86:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1 |
| // CHECK7-NEXT: store i32 1, ptr [[TMP86]], align 4 |
| // CHECK7-NEXT: [[TMP87:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2 |
| // CHECK7-NEXT: store ptr [[TMP83]], ptr [[TMP87]], align 4 |
| // CHECK7-NEXT: [[TMP88:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3 |
| // CHECK7-NEXT: store ptr [[TMP84]], ptr [[TMP88]], align 4 |
| // CHECK7-NEXT: [[TMP89:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4 |
| // CHECK7-NEXT: store ptr @.offload_sizes.7, ptr [[TMP89]], align 4 |
| // CHECK7-NEXT: [[TMP90:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5 |
| // CHECK7-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP90]], align 4 |
| // CHECK7-NEXT: [[TMP91:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP91]], align 4 |
| // CHECK7-NEXT: [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7 |
| // CHECK7-NEXT: store ptr null, ptr [[TMP92]], align 4 |
| // CHECK7-NEXT: [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8 |
| // CHECK7-NEXT: store i64 123, ptr [[TMP93]], align 8 |
| // CHECK7-NEXT: [[TMP94:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9 |
| // CHECK7-NEXT: store i64 0, ptr [[TMP94]], align 8 |
| // CHECK7-NEXT: [[TMP95:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10 |
| // CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP95]], align 4 |
| // CHECK7-NEXT: [[TMP96:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11 |
| // CHECK7-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP96]], align 4 |
| // CHECK7-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12 |
| // CHECK7-NEXT: store i32 0, ptr [[TMP97]], align 4 |
| // CHECK7-NEXT: [[TMP98:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.region_id, ptr [[KERNEL_ARGS31]]) |
| // CHECK7-NEXT: [[TMP99:%.*]] = icmp ne i32 [[TMP98]], 0 |
| // CHECK7-NEXT: br i1 [[TMP99]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]] |
| // CHECK7: omp_offload.failed32: |
| // CHECK7-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54(ptr [[THIS1]]) #[[ATTR2]] |
| // CHECK7-NEXT: br label [[OMP_OFFLOAD_CONT33]] |
| // CHECK7: omp_offload.cont33: |
| // CHECK7-NEXT: [[A34:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A34]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP100:%.*]] = load i32, ptr [[ARRAYIDX]], align 4 |
| // CHECK7-NEXT: ret i32 [[TMP100]] |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36 |
| // CHECK7-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1:[0-9]+]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined, ptr [[TMP0]]) |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined |
| // CHECK7-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK7-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK7-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK7-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK7-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK7: cond.true: |
| // CHECK7-NEXT: br label [[COND_END:%.*]] |
| // CHECK7: cond.false: |
| // CHECK7-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: br label [[COND_END]] |
| // CHECK7: cond.end: |
| // CHECK7-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK7-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK7: omp.inner.for.cond: |
| // CHECK7-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP9:![0-9]+]] |
| // CHECK7-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK7-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK7-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK7: omp.inner.for.body: |
| // CHECK7-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK7-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK7-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP9]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK7: omp.inner.for.inc: |
| // CHECK7-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK7-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK7-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK7-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]] |
| // CHECK7: omp.inner.for.end: |
| // CHECK7-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK7: omp.loop.exit: |
| // CHECK7-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK7-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK7-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK7: .omp.final.then: |
| // CHECK7-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK7-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK7: .omp.final.done: |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l36.omp_outlined.omp_outlined |
| // CHECK7-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK7-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK7-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 122 |
| // CHECK7-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK7: cond.true: |
| // CHECK7-NEXT: br label [[COND_END:%.*]] |
| // CHECK7: cond.false: |
| // CHECK7-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: br label [[COND_END]] |
| // CHECK7: cond.end: |
| // CHECK7-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK7-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK7: omp.inner.for.cond: |
| // CHECK7-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13:![0-9]+]] |
| // CHECK7-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK7-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK7-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK7: omp.inner.for.body: |
| // CHECK7-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK7-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK7-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK7-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK7-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i32 0, i32 [[TMP11]] |
| // CHECK7-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK7-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK7: omp.body.continue: |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK7: omp.inner.for.inc: |
| // CHECK7-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK7-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK7-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP14:![0-9]+]] |
| // CHECK7: omp.inner.for.end: |
| // CHECK7-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK7: omp.loop.exit: |
| // CHECK7-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK7-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK7-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK7: .omp.final.then: |
| // CHECK7-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK7-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK7: .omp.final.done: |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40 |
| // CHECK7-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined, ptr [[TMP0]]) |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined |
| // CHECK7-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK7-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK7-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK7-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK7-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK7: cond.true: |
| // CHECK7-NEXT: br label [[COND_END:%.*]] |
| // CHECK7: cond.false: |
| // CHECK7-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: br label [[COND_END]] |
| // CHECK7: cond.end: |
| // CHECK7-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK7-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK7: omp.inner.for.cond: |
| // CHECK7-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18:![0-9]+]] |
| // CHECK7-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK7-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK7-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK7: omp.inner.for.body: |
| // CHECK7-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK7-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK7-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP18]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK7: omp.inner.for.inc: |
| // CHECK7-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK7-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK7-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK7-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]] |
| // CHECK7: omp.inner.for.end: |
| // CHECK7-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK7: omp.loop.exit: |
| // CHECK7-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK7-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK7-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK7: .omp.final.then: |
| // CHECK7-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK7-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK7: .omp.final.done: |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l40.omp_outlined.omp_outlined |
| // CHECK7-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK7-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK7-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 122 |
| // CHECK7-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK7: cond.true: |
| // CHECK7-NEXT: br label [[COND_END:%.*]] |
| // CHECK7: cond.false: |
| // CHECK7-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: br label [[COND_END]] |
| // CHECK7: cond.end: |
| // CHECK7-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK7-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK7: omp.inner.for.cond: |
| // CHECK7-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP21:![0-9]+]] |
| // CHECK7-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK7-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK7-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK7: omp.inner.for.body: |
| // CHECK7-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK7-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK7-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK7-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK7-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i32 0, i32 [[TMP11]] |
| // CHECK7-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK7-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK7: omp.body.continue: |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK7: omp.inner.for.inc: |
| // CHECK7-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK7-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK7-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP22:![0-9]+]] |
| // CHECK7: omp.inner.for.end: |
| // CHECK7-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK7: omp.loop.exit: |
| // CHECK7-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK7-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK7-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK7: .omp.final.then: |
| // CHECK7-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK7-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK7: .omp.final.done: |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44 |
| // CHECK7-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined, ptr [[TMP0]]) |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined |
| // CHECK7-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK7-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK7-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK7-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK7-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK7: cond.true: |
| // CHECK7-NEXT: br label [[COND_END:%.*]] |
| // CHECK7: cond.false: |
| // CHECK7-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: br label [[COND_END]] |
| // CHECK7: cond.end: |
| // CHECK7-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK7-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK7: omp.inner.for.cond: |
| // CHECK7-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP24:![0-9]+]] |
| // CHECK7-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK7-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK7-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK7: omp.inner.for.body: |
| // CHECK7-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK7-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK7-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP24]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK7: omp.inner.for.inc: |
| // CHECK7-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK7-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK7-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK7-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP25:![0-9]+]] |
| // CHECK7: omp.inner.for.end: |
| // CHECK7-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK7: omp.loop.exit: |
| // CHECK7-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK7-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK7-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK7: .omp.final.then: |
| // CHECK7-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK7-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK7: .omp.final.done: |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l44.omp_outlined.omp_outlined |
| // CHECK7-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK7-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 61) |
| // CHECK7-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK7: omp.dispatch.cond: |
| // CHECK7-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], [[TMP6]] |
| // CHECK7-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK7: cond.true: |
| // CHECK7-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: br label [[COND_END:%.*]] |
| // CHECK7: cond.false: |
| // CHECK7-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: br label [[COND_END]] |
| // CHECK7: cond.end: |
| // CHECK7-NEXT: [[COND:%.*]] = phi i32 [ [[TMP7]], [[COND_TRUE]] ], [ [[TMP8]], [[COND_FALSE]] ] |
| // CHECK7-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP9]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK7-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4 |
| // CHECK7-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP10]], [[TMP11]] |
| // CHECK7-NEXT: br i1 [[CMP1]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK7: omp.dispatch.body: |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK7: omp.inner.for.cond: |
| // CHECK7-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP27:![0-9]+]] |
| // CHECK7-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK7-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP12]], [[TMP13]] |
| // CHECK7-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK7: omp.inner.for.body: |
| // CHECK7-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK7-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP14]], 1 |
| // CHECK7-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK7-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK7-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP15:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i32 0, i32 [[TMP15]] |
| // CHECK7-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK7-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK7: omp.body.continue: |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK7: omp.inner.for.inc: |
| // CHECK7-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK7-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP16]], 1 |
| // CHECK7-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP28:![0-9]+]] |
| // CHECK7: omp.inner.for.end: |
| // CHECK7-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK7: omp.dispatch.inc: |
| // CHECK7-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: [[ADD4:%.*]] = add nsw i32 [[TMP17]], [[TMP18]] |
| // CHECK7-NEXT: store i32 [[ADD4]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: [[ADD5:%.*]] = add nsw i32 [[TMP19]], [[TMP20]] |
| // CHECK7-NEXT: store i32 [[ADD5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK7: omp.dispatch.end: |
| // CHECK7-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK7-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0 |
| // CHECK7-NEXT: br i1 [[TMP22]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK7: .omp.final.then: |
| // CHECK7-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK7-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK7: .omp.final.done: |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49 |
| // CHECK7-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined, ptr [[TMP0]]) |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined |
| // CHECK7-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK7-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK7-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK7-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK7-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK7: cond.true: |
| // CHECK7-NEXT: br label [[COND_END:%.*]] |
| // CHECK7: cond.false: |
| // CHECK7-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: br label [[COND_END]] |
| // CHECK7: cond.end: |
| // CHECK7-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK7-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK7: omp.inner.for.cond: |
| // CHECK7-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP30:![0-9]+]] |
| // CHECK7-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK7-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK7-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK7: omp.inner.for.body: |
| // CHECK7-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK7-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK7-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP30]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK7: omp.inner.for.inc: |
| // CHECK7-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK7-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK7-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK7-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP31:![0-9]+]] |
| // CHECK7: omp.inner.for.end: |
| // CHECK7-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK7: omp.loop.exit: |
| // CHECK7-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK7-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK7-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK7: .omp.final.then: |
| // CHECK7-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK7-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK7: .omp.final.done: |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l49.omp_outlined.omp_outlined |
| // CHECK7-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK7-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 1073741859, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 1) |
| // CHECK7-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK7: omp.dispatch.cond: |
| // CHECK7-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK7-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK7-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK7: omp.dispatch.body: |
| // CHECK7-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK7: omp.inner.for.cond: |
| // CHECK7-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP33:![0-9]+]] |
| // CHECK7-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK7-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK7-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK7: omp.inner.for.body: |
| // CHECK7-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK7-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK7-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK7-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK7-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i32 0, i32 [[TMP12]] |
| // CHECK7-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK7-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK7: omp.body.continue: |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK7: omp.inner.for.inc: |
| // CHECK7-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK7-NEXT: [[ADD1:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK7-NEXT: store i32 [[ADD1]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP33]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP34:![0-9]+]] |
| // CHECK7: omp.inner.for.end: |
| // CHECK7-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK7: omp.dispatch.inc: |
| // CHECK7-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK7: omp.dispatch.end: |
| // CHECK7-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK7-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK7: .omp.final.then: |
| // CHECK7-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK7-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK7: .omp.final.done: |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54 |
| // CHECK7-SAME: (ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined, ptr [[TMP0]]) |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined |
| // CHECK7-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK7-NEXT: store i32 122, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK7-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK7-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 122 |
| // CHECK7-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK7: cond.true: |
| // CHECK7-NEXT: br label [[COND_END:%.*]] |
| // CHECK7: cond.false: |
| // CHECK7-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: br label [[COND_END]] |
| // CHECK7: cond.end: |
| // CHECK7-NEXT: [[COND:%.*]] = phi i32 [ 122, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK7-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK7-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK7: omp.inner.for.cond: |
| // CHECK7-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP36:![0-9]+]] |
| // CHECK7-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK7-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK7-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK7: omp.inner.for.body: |
| // CHECK7-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK7-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK7-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP36]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK7: omp.inner.for.inc: |
| // CHECK7-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK7-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK7-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK7-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP36]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP37:![0-9]+]] |
| // CHECK7: omp.inner.for.end: |
| // CHECK7-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK7: omp.loop.exit: |
| // CHECK7-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK7-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK7-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK7: .omp.final.then: |
| // CHECK7-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK7-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK7: .omp.final.done: |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK7-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SSIiLi123ELx456EE3fooEv_l54.omp_outlined.omp_outlined |
| // CHECK7-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef [[THIS:%.*]]) #[[ATTR1]] { |
| // CHECK7-NEXT: entry: |
| // CHECK7-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK7-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK7-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK7-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK7-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK7-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 1073741859, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 61) |
| // CHECK7-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK7: omp.dispatch.cond: |
| // CHECK7-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK7-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK7-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK7: omp.dispatch.body: |
| // CHECK7-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK7-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK7: omp.inner.for.cond: |
| // CHECK7-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP39:![0-9]+]] |
| // CHECK7-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK7-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK7-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK7: omp.inner.for.body: |
| // CHECK7-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK7-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK7-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK7-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK7-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[TMP0]], i32 0, i32 0 |
| // CHECK7-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK7-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i32 0, i32 [[TMP12]] |
| // CHECK7-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK7-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK7: omp.body.continue: |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK7: omp.inner.for.inc: |
| // CHECK7-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK7-NEXT: [[ADD1:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK7-NEXT: store i32 [[ADD1]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP39]] |
| // CHECK7-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP40:![0-9]+]] |
| // CHECK7: omp.inner.for.end: |
| // CHECK7-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK7: omp.dispatch.inc: |
| // CHECK7-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK7: omp.dispatch.end: |
| // CHECK7-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK7-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK7-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK7: .omp.final.then: |
| // CHECK7-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK7-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK7: .omp.final.done: |
| // CHECK7-NEXT: ret void |
| // |
| // |
| // CHECK9-LABEL: define {{[^@]+}}@_Z21teams_template_structv |
| // CHECK9-SAME: () #[[ATTR0:[0-9]+]] { |
| // CHECK9-NEXT: entry: |
| // CHECK9-NEXT: [[V:%.*]] = alloca [[STRUCT_SS:%.*]], align 4 |
| // CHECK9-NEXT: [[CALL:%.*]] = call noundef signext i32 @_ZN2SSIiLi123ELx456EE3fooEv(ptr noundef nonnull align 4 dereferenceable(496) [[V]]) |
| // CHECK9-NEXT: ret i32 [[CALL]] |
| // |
| // |
| // CHECK9-LABEL: define {{[^@]+}}@_ZN2SSIiLi123ELx456EE3fooEv |
| // CHECK9-SAME: (ptr noundef nonnull align 4 dereferenceable(496) [[THIS:%.*]]) #[[ATTR0]] comdat { |
| // CHECK9-NEXT: entry: |
| // CHECK9-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK9-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[_TMP3:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_LB4:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_UB5:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_IV6:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[I7:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[_TMP20:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_LB21:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_UB22:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_IV23:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[I24:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[_TMP37:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_LB38:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_UB39:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_IV40:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[I41:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[_TMP54:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_LB55:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_UB56:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[DOTOMP_IV57:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: [[I58:%.*]] = alloca i32, align 4 |
| // CHECK9-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8 |
| // CHECK9-NEXT: [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8 |
| // CHECK9-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK9-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK9-NEXT: [[TMP0:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK9-NEXT: store i32 [[TMP0]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK9: omp.inner.for.cond: |
| // CHECK9-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP2:![0-9]+]] |
| // CHECK9-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK9-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP1]], [[TMP2]] |
| // CHECK9-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK9: omp.inner.for.body: |
| // CHECK9-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK9-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP3]], 1 |
| // CHECK9-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK9-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK9-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK9-NEXT: [[TMP4:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK9-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP4]] to i64 |
| // CHECK9-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK9-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK9-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK9: omp.body.continue: |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK9: omp.inner.for.inc: |
| // CHECK9-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK9-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP5]], 1 |
| // CHECK9-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]] |
| // CHECK9: omp.inner.for.end: |
| // CHECK9-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK9-NEXT: store i32 0, ptr [[DOTOMP_LB4]], align 4 |
| // CHECK9-NEXT: store i32 122, ptr [[DOTOMP_UB5]], align 4 |
| // CHECK9-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_LB4]], align 4 |
| // CHECK9-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_IV6]], align 4 |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_COND8:%.*]] |
| // CHECK9: omp.inner.for.cond8: |
| // CHECK9-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV6]], align 4, !llvm.access.group [[ACC_GRP6:![0-9]+]] |
| // CHECK9-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB5]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK9-NEXT: [[CMP9:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]] |
| // CHECK9-NEXT: br i1 [[CMP9]], label [[OMP_INNER_FOR_BODY10:%.*]], label [[OMP_INNER_FOR_END19:%.*]] |
| // CHECK9: omp.inner.for.body10: |
| // CHECK9-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV6]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK9-NEXT: [[MUL11:%.*]] = mul nsw i32 [[TMP9]], 1 |
| // CHECK9-NEXT: [[ADD12:%.*]] = add nsw i32 0, [[MUL11]] |
| // CHECK9-NEXT: store i32 [[ADD12]], ptr [[I7]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK9-NEXT: [[A13:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK9-NEXT: [[TMP10:%.*]] = load i32, ptr [[I7]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK9-NEXT: [[IDXPROM14:%.*]] = sext i32 [[TMP10]] to i64 |
| // CHECK9-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds [123 x i32], ptr [[A13]], i64 0, i64 [[IDXPROM14]] |
| // CHECK9-NEXT: store i32 0, ptr [[ARRAYIDX15]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK9-NEXT: br label [[OMP_BODY_CONTINUE16:%.*]] |
| // CHECK9: omp.body.continue16: |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_INC17:%.*]] |
| // CHECK9: omp.inner.for.inc17: |
| // CHECK9-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV6]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK9-NEXT: [[ADD18:%.*]] = add nsw i32 [[TMP11]], 1 |
| // CHECK9-NEXT: store i32 [[ADD18]], ptr [[DOTOMP_IV6]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_COND8]], !llvm.loop [[LOOP7:![0-9]+]] |
| // CHECK9: omp.inner.for.end19: |
| // CHECK9-NEXT: store i32 123, ptr [[I7]], align 4 |
| // CHECK9-NEXT: store i32 0, ptr [[DOTOMP_LB21]], align 4 |
| // CHECK9-NEXT: store i32 122, ptr [[DOTOMP_UB22]], align 4 |
| // CHECK9-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_LB21]], align 4 |
| // CHECK9-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV23]], align 4 |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_COND25:%.*]] |
| // CHECK9: omp.inner.for.cond25: |
| // CHECK9-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV23]], align 4, !llvm.access.group [[ACC_GRP9:![0-9]+]] |
| // CHECK9-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_UB22]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK9-NEXT: [[CMP26:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK9-NEXT: br i1 [[CMP26]], label [[OMP_INNER_FOR_BODY27:%.*]], label [[OMP_INNER_FOR_END36:%.*]] |
| // CHECK9: omp.inner.for.body27: |
| // CHECK9-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV23]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK9-NEXT: [[MUL28:%.*]] = mul nsw i32 [[TMP15]], 1 |
| // CHECK9-NEXT: [[ADD29:%.*]] = add nsw i32 0, [[MUL28]] |
| // CHECK9-NEXT: store i32 [[ADD29]], ptr [[I24]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK9-NEXT: [[A30:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK9-NEXT: [[TMP16:%.*]] = load i32, ptr [[I24]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK9-NEXT: [[IDXPROM31:%.*]] = sext i32 [[TMP16]] to i64 |
| // CHECK9-NEXT: [[ARRAYIDX32:%.*]] = getelementptr inbounds [123 x i32], ptr [[A30]], i64 0, i64 [[IDXPROM31]] |
| // CHECK9-NEXT: store i32 0, ptr [[ARRAYIDX32]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK9-NEXT: br label [[OMP_BODY_CONTINUE33:%.*]] |
| // CHECK9: omp.body.continue33: |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_INC34:%.*]] |
| // CHECK9: omp.inner.for.inc34: |
| // CHECK9-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV23]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK9-NEXT: [[ADD35:%.*]] = add nsw i32 [[TMP17]], 1 |
| // CHECK9-NEXT: store i32 [[ADD35]], ptr [[DOTOMP_IV23]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_COND25]], !llvm.loop [[LOOP10:![0-9]+]] |
| // CHECK9: omp.inner.for.end36: |
| // CHECK9-NEXT: store i32 123, ptr [[I24]], align 4 |
| // CHECK9-NEXT: store i32 0, ptr [[DOTOMP_LB38]], align 4 |
| // CHECK9-NEXT: store i32 122, ptr [[DOTOMP_UB39]], align 4 |
| // CHECK9-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB38]], align 4 |
| // CHECK9-NEXT: store i32 [[TMP18]], ptr [[DOTOMP_IV40]], align 4 |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_COND42:%.*]] |
| // CHECK9: omp.inner.for.cond42: |
| // CHECK9-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV40]], align 4, !llvm.access.group [[ACC_GRP12:![0-9]+]] |
| // CHECK9-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB39]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK9-NEXT: [[CMP43:%.*]] = icmp sle i32 [[TMP19]], [[TMP20]] |
| // CHECK9-NEXT: br i1 [[CMP43]], label [[OMP_INNER_FOR_BODY44:%.*]], label [[OMP_INNER_FOR_END53:%.*]] |
| // CHECK9: omp.inner.for.body44: |
| // CHECK9-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV40]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK9-NEXT: [[MUL45:%.*]] = mul nsw i32 [[TMP21]], 1 |
| // CHECK9-NEXT: [[ADD46:%.*]] = add nsw i32 0, [[MUL45]] |
| // CHECK9-NEXT: store i32 [[ADD46]], ptr [[I41]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK9-NEXT: [[A47:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK9-NEXT: [[TMP22:%.*]] = load i32, ptr [[I41]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK9-NEXT: [[IDXPROM48:%.*]] = sext i32 [[TMP22]] to i64 |
| // CHECK9-NEXT: [[ARRAYIDX49:%.*]] = getelementptr inbounds [123 x i32], ptr [[A47]], i64 0, i64 [[IDXPROM48]] |
| // CHECK9-NEXT: store i32 0, ptr [[ARRAYIDX49]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK9-NEXT: br label [[OMP_BODY_CONTINUE50:%.*]] |
| // CHECK9: omp.body.continue50: |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_INC51:%.*]] |
| // CHECK9: omp.inner.for.inc51: |
| // CHECK9-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV40]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK9-NEXT: [[ADD52:%.*]] = add nsw i32 [[TMP23]], 1 |
| // CHECK9-NEXT: store i32 [[ADD52]], ptr [[DOTOMP_IV40]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_COND42]], !llvm.loop [[LOOP13:![0-9]+]] |
| // CHECK9: omp.inner.for.end53: |
| // CHECK9-NEXT: store i32 123, ptr [[I41]], align 4 |
| // CHECK9-NEXT: store i32 0, ptr [[DOTOMP_LB55]], align 4 |
| // CHECK9-NEXT: store i32 122, ptr [[DOTOMP_UB56]], align 4 |
| // CHECK9-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTOMP_LB55]], align 4 |
| // CHECK9-NEXT: store i32 [[TMP24]], ptr [[DOTOMP_IV57]], align 4 |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_COND59:%.*]] |
| // CHECK9: omp.inner.for.cond59: |
| // CHECK9-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IV57]], align 4, !llvm.access.group [[ACC_GRP15:![0-9]+]] |
| // CHECK9-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTOMP_UB56]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK9-NEXT: [[CMP60:%.*]] = icmp sle i32 [[TMP25]], [[TMP26]] |
| // CHECK9-NEXT: br i1 [[CMP60]], label [[OMP_INNER_FOR_BODY61:%.*]], label [[OMP_INNER_FOR_END70:%.*]] |
| // CHECK9: omp.inner.for.body61: |
| // CHECK9-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IV57]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK9-NEXT: [[MUL62:%.*]] = mul nsw i32 [[TMP27]], 1 |
| // CHECK9-NEXT: [[ADD63:%.*]] = add nsw i32 0, [[MUL62]] |
| // CHECK9-NEXT: store i32 [[ADD63]], ptr [[I58]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK9-NEXT: [[A64:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK9-NEXT: [[TMP28:%.*]] = load i32, ptr [[I58]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK9-NEXT: [[IDXPROM65:%.*]] = sext i32 [[TMP28]] to i64 |
| // CHECK9-NEXT: [[ARRAYIDX66:%.*]] = getelementptr inbounds [123 x i32], ptr [[A64]], i64 0, i64 [[IDXPROM65]] |
| // CHECK9-NEXT: store i32 0, ptr [[ARRAYIDX66]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK9-NEXT: br label [[OMP_BODY_CONTINUE67:%.*]] |
| // CHECK9: omp.body.continue67: |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_INC68:%.*]] |
| // CHECK9: omp.inner.for.inc68: |
| // CHECK9-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IV57]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK9-NEXT: [[ADD69:%.*]] = add nsw i32 [[TMP29]], 1 |
| // CHECK9-NEXT: store i32 [[ADD69]], ptr [[DOTOMP_IV57]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK9-NEXT: br label [[OMP_INNER_FOR_COND59]], !llvm.loop [[LOOP16:![0-9]+]] |
| // CHECK9: omp.inner.for.end70: |
| // CHECK9-NEXT: store i32 123, ptr [[I58]], align 4 |
| // CHECK9-NEXT: [[A71:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK9-NEXT: [[ARRAYIDX72:%.*]] = getelementptr inbounds [123 x i32], ptr [[A71]], i64 0, i64 0 |
| // CHECK9-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARRAYIDX72]], align 4 |
| // CHECK9-NEXT: ret i32 [[TMP30]] |
| // |
| // |
| // CHECK11-LABEL: define {{[^@]+}}@_Z21teams_template_structv |
| // CHECK11-SAME: () #[[ATTR0:[0-9]+]] { |
| // CHECK11-NEXT: entry: |
| // CHECK11-NEXT: [[V:%.*]] = alloca [[STRUCT_SS:%.*]], align 4 |
| // CHECK11-NEXT: [[CALL:%.*]] = call noundef i32 @_ZN2SSIiLi123ELx456EE3fooEv(ptr noundef nonnull align 4 dereferenceable(496) [[V]]) |
| // CHECK11-NEXT: ret i32 [[CALL]] |
| // |
| // |
| // CHECK11-LABEL: define {{[^@]+}}@_ZN2SSIiLi123ELx456EE3fooEv |
| // CHECK11-SAME: (ptr noundef nonnull align 4 dereferenceable(496) [[THIS:%.*]]) #[[ATTR0]] comdat align 2 { |
| // CHECK11-NEXT: entry: |
| // CHECK11-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK11-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[_TMP3:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_LB4:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_UB5:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_IV6:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[I7:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[_TMP19:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_LB20:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_UB21:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_IV22:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[I23:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[_TMP35:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_LB36:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_UB37:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_IV38:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[I39:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[_TMP51:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_LB52:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_UB53:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[DOTOMP_IV54:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: [[I55:%.*]] = alloca i32, align 4 |
| // CHECK11-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4 |
| // CHECK11-NEXT: [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4 |
| // CHECK11-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK11-NEXT: store i32 122, ptr [[DOTOMP_UB]], align 4 |
| // CHECK11-NEXT: [[TMP0:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK11-NEXT: store i32 [[TMP0]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK11: omp.inner.for.cond: |
| // CHECK11-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP3:![0-9]+]] |
| // CHECK11-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK11-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP1]], [[TMP2]] |
| // CHECK11-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK11: omp.inner.for.body: |
| // CHECK11-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK11-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP3]], 1 |
| // CHECK11-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK11-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK11-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK11-NEXT: [[TMP4:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK11-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [123 x i32], ptr [[A]], i32 0, i32 [[TMP4]] |
| // CHECK11-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK11-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK11: omp.body.continue: |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK11: omp.inner.for.inc: |
| // CHECK11-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK11-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP5]], 1 |
| // CHECK11-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]] |
| // CHECK11: omp.inner.for.end: |
| // CHECK11-NEXT: store i32 123, ptr [[I]], align 4 |
| // CHECK11-NEXT: store i32 0, ptr [[DOTOMP_LB4]], align 4 |
| // CHECK11-NEXT: store i32 122, ptr [[DOTOMP_UB5]], align 4 |
| // CHECK11-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_LB4]], align 4 |
| // CHECK11-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_IV6]], align 4 |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_COND8:%.*]] |
| // CHECK11: omp.inner.for.cond8: |
| // CHECK11-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV6]], align 4, !llvm.access.group [[ACC_GRP7:![0-9]+]] |
| // CHECK11-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB5]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK11-NEXT: [[CMP9:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]] |
| // CHECK11-NEXT: br i1 [[CMP9]], label [[OMP_INNER_FOR_BODY10:%.*]], label [[OMP_INNER_FOR_END18:%.*]] |
| // CHECK11: omp.inner.for.body10: |
| // CHECK11-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV6]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK11-NEXT: [[MUL11:%.*]] = mul nsw i32 [[TMP9]], 1 |
| // CHECK11-NEXT: [[ADD12:%.*]] = add nsw i32 0, [[MUL11]] |
| // CHECK11-NEXT: store i32 [[ADD12]], ptr [[I7]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK11-NEXT: [[A13:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK11-NEXT: [[TMP10:%.*]] = load i32, ptr [[I7]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK11-NEXT: [[ARRAYIDX14:%.*]] = getelementptr inbounds [123 x i32], ptr [[A13]], i32 0, i32 [[TMP10]] |
| // CHECK11-NEXT: store i32 0, ptr [[ARRAYIDX14]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK11-NEXT: br label [[OMP_BODY_CONTINUE15:%.*]] |
| // CHECK11: omp.body.continue15: |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_INC16:%.*]] |
| // CHECK11: omp.inner.for.inc16: |
| // CHECK11-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV6]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK11-NEXT: [[ADD17:%.*]] = add nsw i32 [[TMP11]], 1 |
| // CHECK11-NEXT: store i32 [[ADD17]], ptr [[DOTOMP_IV6]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_COND8]], !llvm.loop [[LOOP8:![0-9]+]] |
| // CHECK11: omp.inner.for.end18: |
| // CHECK11-NEXT: store i32 123, ptr [[I7]], align 4 |
| // CHECK11-NEXT: store i32 0, ptr [[DOTOMP_LB20]], align 4 |
| // CHECK11-NEXT: store i32 122, ptr [[DOTOMP_UB21]], align 4 |
| // CHECK11-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_LB20]], align 4 |
| // CHECK11-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV22]], align 4 |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_COND24:%.*]] |
| // CHECK11: omp.inner.for.cond24: |
| // CHECK11-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV22]], align 4, !llvm.access.group [[ACC_GRP10:![0-9]+]] |
| // CHECK11-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_UB21]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK11-NEXT: [[CMP25:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK11-NEXT: br i1 [[CMP25]], label [[OMP_INNER_FOR_BODY26:%.*]], label [[OMP_INNER_FOR_END34:%.*]] |
| // CHECK11: omp.inner.for.body26: |
| // CHECK11-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV22]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK11-NEXT: [[MUL27:%.*]] = mul nsw i32 [[TMP15]], 1 |
| // CHECK11-NEXT: [[ADD28:%.*]] = add nsw i32 0, [[MUL27]] |
| // CHECK11-NEXT: store i32 [[ADD28]], ptr [[I23]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK11-NEXT: [[A29:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK11-NEXT: [[TMP16:%.*]] = load i32, ptr [[I23]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK11-NEXT: [[ARRAYIDX30:%.*]] = getelementptr inbounds [123 x i32], ptr [[A29]], i32 0, i32 [[TMP16]] |
| // CHECK11-NEXT: store i32 0, ptr [[ARRAYIDX30]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK11-NEXT: br label [[OMP_BODY_CONTINUE31:%.*]] |
| // CHECK11: omp.body.continue31: |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_INC32:%.*]] |
| // CHECK11: omp.inner.for.inc32: |
| // CHECK11-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV22]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK11-NEXT: [[ADD33:%.*]] = add nsw i32 [[TMP17]], 1 |
| // CHECK11-NEXT: store i32 [[ADD33]], ptr [[DOTOMP_IV22]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_COND24]], !llvm.loop [[LOOP11:![0-9]+]] |
| // CHECK11: omp.inner.for.end34: |
| // CHECK11-NEXT: store i32 123, ptr [[I23]], align 4 |
| // CHECK11-NEXT: store i32 0, ptr [[DOTOMP_LB36]], align 4 |
| // CHECK11-NEXT: store i32 122, ptr [[DOTOMP_UB37]], align 4 |
| // CHECK11-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB36]], align 4 |
| // CHECK11-NEXT: store i32 [[TMP18]], ptr [[DOTOMP_IV38]], align 4 |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_COND40:%.*]] |
| // CHECK11: omp.inner.for.cond40: |
| // CHECK11-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV38]], align 4, !llvm.access.group [[ACC_GRP13:![0-9]+]] |
| // CHECK11-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB37]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK11-NEXT: [[CMP41:%.*]] = icmp sle i32 [[TMP19]], [[TMP20]] |
| // CHECK11-NEXT: br i1 [[CMP41]], label [[OMP_INNER_FOR_BODY42:%.*]], label [[OMP_INNER_FOR_END50:%.*]] |
| // CHECK11: omp.inner.for.body42: |
| // CHECK11-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV38]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK11-NEXT: [[MUL43:%.*]] = mul nsw i32 [[TMP21]], 1 |
| // CHECK11-NEXT: [[ADD44:%.*]] = add nsw i32 0, [[MUL43]] |
| // CHECK11-NEXT: store i32 [[ADD44]], ptr [[I39]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK11-NEXT: [[A45:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK11-NEXT: [[TMP22:%.*]] = load i32, ptr [[I39]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK11-NEXT: [[ARRAYIDX46:%.*]] = getelementptr inbounds [123 x i32], ptr [[A45]], i32 0, i32 [[TMP22]] |
| // CHECK11-NEXT: store i32 0, ptr [[ARRAYIDX46]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK11-NEXT: br label [[OMP_BODY_CONTINUE47:%.*]] |
| // CHECK11: omp.body.continue47: |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_INC48:%.*]] |
| // CHECK11: omp.inner.for.inc48: |
| // CHECK11-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV38]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK11-NEXT: [[ADD49:%.*]] = add nsw i32 [[TMP23]], 1 |
| // CHECK11-NEXT: store i32 [[ADD49]], ptr [[DOTOMP_IV38]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_COND40]], !llvm.loop [[LOOP14:![0-9]+]] |
| // CHECK11: omp.inner.for.end50: |
| // CHECK11-NEXT: store i32 123, ptr [[I39]], align 4 |
| // CHECK11-NEXT: store i32 0, ptr [[DOTOMP_LB52]], align 4 |
| // CHECK11-NEXT: store i32 122, ptr [[DOTOMP_UB53]], align 4 |
| // CHECK11-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTOMP_LB52]], align 4 |
| // CHECK11-NEXT: store i32 [[TMP24]], ptr [[DOTOMP_IV54]], align 4 |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_COND56:%.*]] |
| // CHECK11: omp.inner.for.cond56: |
| // CHECK11-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IV54]], align 4, !llvm.access.group [[ACC_GRP16:![0-9]+]] |
| // CHECK11-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTOMP_UB53]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK11-NEXT: [[CMP57:%.*]] = icmp sle i32 [[TMP25]], [[TMP26]] |
| // CHECK11-NEXT: br i1 [[CMP57]], label [[OMP_INNER_FOR_BODY58:%.*]], label [[OMP_INNER_FOR_END66:%.*]] |
| // CHECK11: omp.inner.for.body58: |
| // CHECK11-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IV54]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK11-NEXT: [[MUL59:%.*]] = mul nsw i32 [[TMP27]], 1 |
| // CHECK11-NEXT: [[ADD60:%.*]] = add nsw i32 0, [[MUL59]] |
| // CHECK11-NEXT: store i32 [[ADD60]], ptr [[I55]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK11-NEXT: [[A61:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK11-NEXT: [[TMP28:%.*]] = load i32, ptr [[I55]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK11-NEXT: [[ARRAYIDX62:%.*]] = getelementptr inbounds [123 x i32], ptr [[A61]], i32 0, i32 [[TMP28]] |
| // CHECK11-NEXT: store i32 0, ptr [[ARRAYIDX62]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK11-NEXT: br label [[OMP_BODY_CONTINUE63:%.*]] |
| // CHECK11: omp.body.continue63: |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_INC64:%.*]] |
| // CHECK11: omp.inner.for.inc64: |
| // CHECK11-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IV54]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK11-NEXT: [[ADD65:%.*]] = add nsw i32 [[TMP29]], 1 |
| // CHECK11-NEXT: store i32 [[ADD65]], ptr [[DOTOMP_IV54]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK11-NEXT: br label [[OMP_INNER_FOR_COND56]], !llvm.loop [[LOOP17:![0-9]+]] |
| // CHECK11: omp.inner.for.end66: |
| // CHECK11-NEXT: store i32 123, ptr [[I55]], align 4 |
| // CHECK11-NEXT: [[A67:%.*]] = getelementptr inbounds [[STRUCT_SS]], ptr [[THIS1]], i32 0, i32 0 |
| // CHECK11-NEXT: [[ARRAYIDX68:%.*]] = getelementptr inbounds [123 x i32], ptr [[A67]], i32 0, i32 0 |
| // CHECK11-NEXT: [[TMP30:%.*]] = load i32, ptr [[ARRAYIDX68]], align 4 |
| // CHECK11-NEXT: ret i32 [[TMP30]] |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@main |
| // CHECK13-SAME: (i32 noundef signext [[ARGC:%.*]], ptr noundef [[ARGV:%.*]]) #[[ATTR0:[0-9]+]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[RETVAL:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[ARGV_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[N:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[SAVED_STACK:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[__VLA_EXPR0:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 8 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK13-NEXT: [[N_CASTED3:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_BASEPTRS4:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_PTRS5:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_MAPPERS6:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_SIZES7:%.*]] = alloca [3 x i64], align 8 |
| // CHECK13-NEXT: [[_TMP8:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_9:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_10:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_18:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[N_CASTED19:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_BASEPTRS20:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_PTRS21:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_MAPPERS22:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_SIZES23:%.*]] = alloca [4 x i64], align 8 |
| // CHECK13-NEXT: [[_TMP24:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_25:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_26:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK13-NEXT: [[N_CASTED34:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_BASEPTRS35:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_PTRS36:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_MAPPERS37:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_SIZES38:%.*]] = alloca [3 x i64], align 8 |
| // CHECK13-NEXT: [[_TMP39:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_40:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_41:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_49:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[N_CASTED50:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED51:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_BASEPTRS52:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_PTRS53:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_MAPPERS54:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_SIZES55:%.*]] = alloca [4 x i64], align 8 |
| // CHECK13-NEXT: [[_TMP56:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_57:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_58:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK13-NEXT: store i32 0, ptr [[RETVAL]], align 4 |
| // CHECK13-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK13-NEXT: store ptr [[ARGV]], ptr [[ARGV_ADDR]], align 8 |
| // CHECK13-NEXT: store i32 100, ptr [[N]], align 4 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK13-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| // CHECK13-NEXT: [[TMP2:%.*]] = call ptr @llvm.stacksave.p0() |
| // CHECK13-NEXT: store ptr [[TMP2]], ptr [[SAVED_STACK]], align 8 |
| // CHECK13-NEXT: [[VLA:%.*]] = alloca i32, i64 [[TMP1]], align 4 |
| // CHECK13-NEXT: store i64 [[TMP1]], ptr [[__VLA_EXPR0]], align 8 |
| // CHECK13-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP3]], ptr [[N_CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK13-NEXT: [[TMP5:%.*]] = mul nuw i64 [[TMP1]], 4 |
| // CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes, i64 24, i1 false) |
| // CHECK13-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK13-NEXT: store i64 [[TMP4]], ptr [[TMP6]], align 8 |
| // CHECK13-NEXT: [[TMP7:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK13-NEXT: store i64 [[TMP4]], ptr [[TMP7]], align 8 |
| // CHECK13-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP8]], align 8 |
| // CHECK13-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP9]], align 8 |
| // CHECK13-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP10]], align 8 |
| // CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP11]], align 8 |
| // CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP12]], align 8 |
| // CHECK13-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP13]], align 8 |
| // CHECK13-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 2 |
| // CHECK13-NEXT: store i64 [[TMP5]], ptr [[TMP14]], align 8 |
| // CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP15]], align 8 |
| // CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP19:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP19]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP20]], 0 |
| // CHECK13-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK13-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK13-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], 1 |
| // CHECK13-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64 |
| // CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP23]], align 4 |
| // CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP24]], align 4 |
| // CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[TMP16]], ptr [[TMP25]], align 8 |
| // CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK13-NEXT: store ptr [[TMP17]], ptr [[TMP26]], align 8 |
| // CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK13-NEXT: store ptr [[TMP18]], ptr [[TMP27]], align 8 |
| // CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK13-NEXT: store ptr @.offload_maptypes, ptr [[TMP28]], align 8 |
| // CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP29]], align 8 |
| // CHECK13-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP30]], align 8 |
| // CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK13-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8 |
| // CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK13-NEXT: store i64 0, ptr [[TMP32]], align 8 |
| // CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4 |
| // CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4 |
| // CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK13-NEXT: store i32 0, ptr [[TMP35]], align 4 |
| // CHECK13-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK13-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0 |
| // CHECK13-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK13: omp_offload.failed: |
| // CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]] |
| // CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK13: omp_offload.cont: |
| // CHECK13-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4 |
| // CHECK13-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8 |
| // CHECK13-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4 |
| // CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.1, i64 24, i1 false) |
| // CHECK13-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0 |
| // CHECK13-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8 |
| // CHECK13-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0 |
| // CHECK13-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8 |
| // CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP43]], align 8 |
| // CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8 |
| // CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8 |
| // CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP46]], align 8 |
| // CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8 |
| // CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8 |
| // CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2 |
| // CHECK13-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8 |
| // CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP50]], align 8 |
| // CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4 |
| // CHECK13-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4 |
| // CHECK13-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0 |
| // CHECK13-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1 |
| // CHECK13-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1 |
| // CHECK13-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4 |
| // CHECK13-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4 |
| // CHECK13-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1 |
| // CHECK13-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64 |
| // CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP58]], align 4 |
| // CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP59]], align 4 |
| // CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8 |
| // CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3 |
| // CHECK13-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8 |
| // CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4 |
| // CHECK13-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8 |
| // CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5 |
| // CHECK13-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP63]], align 8 |
| // CHECK13-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP64]], align 8 |
| // CHECK13-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP65]], align 8 |
| // CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8 |
| // CHECK13-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8 |
| // CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9 |
| // CHECK13-NEXT: store i64 0, ptr [[TMP67]], align 8 |
| // CHECK13-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4 |
| // CHECK13-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4 |
| // CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12 |
| // CHECK13-NEXT: store i32 0, ptr [[TMP70]], align 4 |
| // CHECK13-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]]) |
| // CHECK13-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0 |
| // CHECK13-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]] |
| // CHECK13: omp_offload.failed16: |
| // CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]] |
| // CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT17]] |
| // CHECK13: omp_offload.cont17: |
| // CHECK13-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR_18]], align 4 |
| // CHECK13-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP74]], ptr [[N_CASTED19]], align 4 |
| // CHECK13-NEXT: [[TMP75:%.*]] = load i64, ptr [[N_CASTED19]], align 8 |
| // CHECK13-NEXT: [[TMP76:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP77:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK13-NEXT: [[TMP78:%.*]] = mul nuw i64 [[TMP1]], 4 |
| // CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES23]], ptr align 8 @.offload_sizes.3, i64 32, i1 false) |
| // CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0 |
| // CHECK13-NEXT: store i64 [[TMP75]], ptr [[TMP79]], align 8 |
| // CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0 |
| // CHECK13-NEXT: store i64 [[TMP75]], ptr [[TMP80]], align 8 |
| // CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP81]], align 8 |
| // CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP82]], align 8 |
| // CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP83]], align 8 |
| // CHECK13-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP84]], align 8 |
| // CHECK13-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8 |
| // CHECK13-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP86]], align 8 |
| // CHECK13-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2 |
| // CHECK13-NEXT: store i64 [[TMP78]], ptr [[TMP87]], align 8 |
| // CHECK13-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP88]], align 8 |
| // CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3 |
| // CHECK13-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8 |
| // CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3 |
| // CHECK13-NEXT: store i64 [[TMP77]], ptr [[TMP90]], align 8 |
| // CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP91]], align 8 |
| // CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP95:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP95]], ptr [[DOTCAPTURE_EXPR_25]], align 4 |
| // CHECK13-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4 |
| // CHECK13-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP96]], 0 |
| // CHECK13-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1 |
| // CHECK13-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1 |
| // CHECK13-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4 |
| // CHECK13-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4 |
| // CHECK13-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP97]], 1 |
| // CHECK13-NEXT: [[TMP98:%.*]] = zext i32 [[ADD30]] to i64 |
| // CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP99]], align 4 |
| // CHECK13-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1 |
| // CHECK13-NEXT: store i32 4, ptr [[TMP100]], align 4 |
| // CHECK13-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8 |
| // CHECK13-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3 |
| // CHECK13-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8 |
| // CHECK13-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4 |
| // CHECK13-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 8 |
| // CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5 |
| // CHECK13-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP104]], align 8 |
| // CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP105]], align 8 |
| // CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP106]], align 8 |
| // CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8 |
| // CHECK13-NEXT: store i64 [[TMP98]], ptr [[TMP107]], align 8 |
| // CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9 |
| // CHECK13-NEXT: store i64 0, ptr [[TMP108]], align 8 |
| // CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4 |
| // CHECK13-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4 |
| // CHECK13-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12 |
| // CHECK13-NEXT: store i32 0, ptr [[TMP111]], align 4 |
| // CHECK13-NEXT: [[TMP112:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]]) |
| // CHECK13-NEXT: [[TMP113:%.*]] = icmp ne i32 [[TMP112]], 0 |
| // CHECK13-NEXT: br i1 [[TMP113]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]] |
| // CHECK13: omp_offload.failed32: |
| // CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i64 [[TMP75]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP77]]) #[[ATTR3]] |
| // CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT33]] |
| // CHECK13: omp_offload.cont33: |
| // CHECK13-NEXT: [[TMP114:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP114]], ptr [[N_CASTED34]], align 4 |
| // CHECK13-NEXT: [[TMP115:%.*]] = load i64, ptr [[N_CASTED34]], align 8 |
| // CHECK13-NEXT: [[TMP116:%.*]] = mul nuw i64 [[TMP1]], 4 |
| // CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES38]], ptr align 8 @.offload_sizes.5, i64 24, i1 false) |
| // CHECK13-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0 |
| // CHECK13-NEXT: store i64 [[TMP115]], ptr [[TMP117]], align 8 |
| // CHECK13-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0 |
| // CHECK13-NEXT: store i64 [[TMP115]], ptr [[TMP118]], align 8 |
| // CHECK13-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 0 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP119]], align 8 |
| // CHECK13-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP120]], align 8 |
| // CHECK13-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP121]], align 8 |
| // CHECK13-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 1 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP122]], align 8 |
| // CHECK13-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP123]], align 8 |
| // CHECK13-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP124]], align 8 |
| // CHECK13-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2 |
| // CHECK13-NEXT: store i64 [[TMP116]], ptr [[TMP125]], align 8 |
| // CHECK13-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 2 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP126]], align 8 |
| // CHECK13-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP130:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP130]], ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK13-NEXT: [[TMP131:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK13-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP131]], 0 |
| // CHECK13-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1 |
| // CHECK13-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1 |
| // CHECK13-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK13-NEXT: [[TMP132:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK13-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP132]], 1 |
| // CHECK13-NEXT: [[TMP133:%.*]] = zext i32 [[ADD45]] to i64 |
| // CHECK13-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP134]], align 4 |
| // CHECK13-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP135]], align 4 |
| // CHECK13-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[TMP127]], ptr [[TMP136]], align 8 |
| // CHECK13-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3 |
| // CHECK13-NEXT: store ptr [[TMP128]], ptr [[TMP137]], align 8 |
| // CHECK13-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4 |
| // CHECK13-NEXT: store ptr [[TMP129]], ptr [[TMP138]], align 8 |
| // CHECK13-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5 |
| // CHECK13-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP139]], align 8 |
| // CHECK13-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP140]], align 8 |
| // CHECK13-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP141]], align 8 |
| // CHECK13-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8 |
| // CHECK13-NEXT: store i64 [[TMP133]], ptr [[TMP142]], align 8 |
| // CHECK13-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9 |
| // CHECK13-NEXT: store i64 0, ptr [[TMP143]], align 8 |
| // CHECK13-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4 |
| // CHECK13-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP145]], align 4 |
| // CHECK13-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12 |
| // CHECK13-NEXT: store i32 0, ptr [[TMP146]], align 4 |
| // CHECK13-NEXT: [[TMP147:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]]) |
| // CHECK13-NEXT: [[TMP148:%.*]] = icmp ne i32 [[TMP147]], 0 |
| // CHECK13-NEXT: br i1 [[TMP148]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]] |
| // CHECK13: omp_offload.failed47: |
| // CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i64 [[TMP115]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]] |
| // CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT48]] |
| // CHECK13: omp_offload.cont48: |
| // CHECK13-NEXT: [[TMP149:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP149]], ptr [[DOTCAPTURE_EXPR_49]], align 4 |
| // CHECK13-NEXT: [[TMP150:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP150]], ptr [[N_CASTED50]], align 4 |
| // CHECK13-NEXT: [[TMP151:%.*]] = load i64, ptr [[N_CASTED50]], align 8 |
| // CHECK13-NEXT: [[TMP152:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4 |
| // CHECK13-NEXT: [[TMP153:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 8 |
| // CHECK13-NEXT: [[TMP154:%.*]] = mul nuw i64 [[TMP1]], 4 |
| // CHECK13-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES55]], ptr align 8 @.offload_sizes.7, i64 32, i1 false) |
| // CHECK13-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0 |
| // CHECK13-NEXT: store i64 [[TMP151]], ptr [[TMP155]], align 8 |
| // CHECK13-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0 |
| // CHECK13-NEXT: store i64 [[TMP151]], ptr [[TMP156]], align 8 |
| // CHECK13-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 0 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP157]], align 8 |
| // CHECK13-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP158]], align 8 |
| // CHECK13-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP1]], ptr [[TMP159]], align 8 |
| // CHECK13-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 1 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP160]], align 8 |
| // CHECK13-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP161]], align 8 |
| // CHECK13-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[VLA]], ptr [[TMP162]], align 8 |
| // CHECK13-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2 |
| // CHECK13-NEXT: store i64 [[TMP154]], ptr [[TMP163]], align 8 |
| // CHECK13-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 2 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP164]], align 8 |
| // CHECK13-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3 |
| // CHECK13-NEXT: store i64 [[TMP153]], ptr [[TMP165]], align 8 |
| // CHECK13-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3 |
| // CHECK13-NEXT: store i64 [[TMP153]], ptr [[TMP166]], align 8 |
| // CHECK13-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 3 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP167]], align 8 |
| // CHECK13-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP171:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP171]], ptr [[DOTCAPTURE_EXPR_57]], align 4 |
| // CHECK13-NEXT: [[TMP172:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4 |
| // CHECK13-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP172]], 0 |
| // CHECK13-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1 |
| // CHECK13-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1 |
| // CHECK13-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4 |
| // CHECK13-NEXT: [[TMP173:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4 |
| // CHECK13-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP173]], 1 |
| // CHECK13-NEXT: [[TMP174:%.*]] = zext i32 [[ADD62]] to i64 |
| // CHECK13-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP175]], align 4 |
| // CHECK13-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1 |
| // CHECK13-NEXT: store i32 4, ptr [[TMP176]], align 4 |
| // CHECK13-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[TMP168]], ptr [[TMP177]], align 8 |
| // CHECK13-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3 |
| // CHECK13-NEXT: store ptr [[TMP169]], ptr [[TMP178]], align 8 |
| // CHECK13-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4 |
| // CHECK13-NEXT: store ptr [[TMP170]], ptr [[TMP179]], align 8 |
| // CHECK13-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5 |
| // CHECK13-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP180]], align 8 |
| // CHECK13-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP181]], align 8 |
| // CHECK13-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP182]], align 8 |
| // CHECK13-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8 |
| // CHECK13-NEXT: store i64 [[TMP174]], ptr [[TMP183]], align 8 |
| // CHECK13-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9 |
| // CHECK13-NEXT: store i64 0, ptr [[TMP184]], align 8 |
| // CHECK13-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP185]], align 4 |
| // CHECK13-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP186]], align 4 |
| // CHECK13-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12 |
| // CHECK13-NEXT: store i32 0, ptr [[TMP187]], align 4 |
| // CHECK13-NEXT: [[TMP188:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]]) |
| // CHECK13-NEXT: [[TMP189:%.*]] = icmp ne i32 [[TMP188]], 0 |
| // CHECK13-NEXT: br i1 [[TMP189]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]] |
| // CHECK13: omp_offload.failed64: |
| // CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i64 [[TMP151]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP153]]) #[[ATTR3]] |
| // CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT65]] |
| // CHECK13: omp_offload.cont65: |
| // CHECK13-NEXT: [[TMP190:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4 |
| // CHECK13-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP190]]) |
| // CHECK13-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4 |
| // CHECK13-NEXT: [[TMP191:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8 |
| // CHECK13-NEXT: call void @llvm.stackrestore.p0(ptr [[TMP191]]) |
| // CHECK13-NEXT: [[TMP192:%.*]] = load i32, ptr [[RETVAL]], align 4 |
| // CHECK13-NEXT: ret i32 [[TMP192]] |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139 |
| // CHECK13-SAME: (i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2:[0-9]+]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined, i64 [[TMP3]], i64 [[TMP0]], ptr [[TMP1]]) |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK13-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK13-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK13-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK13: omp.precond.then: |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK13-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13:![0-9]+]] |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK13-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK13-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK13-NEXT: [[TMP16:%.*]] = zext i32 [[TMP15]] to i64 |
| // CHECK13-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK13-NEXT: [[TMP18:%.*]] = zext i32 [[TMP17]] to i64 |
| // CHECK13-NEXT: [[TMP19:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK13-NEXT: store i32 [[TMP19]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK13-NEXT: [[TMP20:%.*]] = load i64, ptr [[N_CASTED]], align 8, !llvm.access.group [[ACC_GRP13]] |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined.omp_outlined, i64 [[TMP16]], i64 [[TMP18]], i64 [[TMP20]], i64 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP13]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK13-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP14:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]]) |
| // CHECK13-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0 |
| // CHECK13-NEXT: br i1 [[TMP26]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP27]], 0 |
| // CHECK13-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK13-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK13: omp.precond.end: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK13-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK13-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK13-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK13: omp.precond.then: |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV3:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK13-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK13-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17:![0-9]+]] |
| // CHECK13-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK13-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK13-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK13-NEXT: [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK13-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP18]] to i64 |
| // CHECK13-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[IDXPROM]] |
| // CHECK13-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK13-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK13: omp.body.continue: |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK13-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK13-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP18:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK13-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK13-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB8:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK13-NEXT: [[DIV9:%.*]] = sdiv i32 [[SUB8]], 1 |
| // CHECK13-NEXT: [[MUL10:%.*]] = mul nsw i32 [[DIV9]], 1 |
| // CHECK13-NEXT: [[ADD11:%.*]] = add nsw i32 0, [[MUL10]] |
| // CHECK13-NEXT: store i32 [[ADD11]], ptr [[I4]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK13: omp.precond.end: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143 |
| // CHECK13-SAME: (i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined, i64 [[TMP3]], i64 [[TMP0]], ptr [[TMP1]]) |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK13-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK13-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK13-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK13: omp.precond.then: |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK13-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22:![0-9]+]] |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK13-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK13-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK13-NEXT: [[TMP16:%.*]] = zext i32 [[TMP15]] to i64 |
| // CHECK13-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK13-NEXT: [[TMP18:%.*]] = zext i32 [[TMP17]] to i64 |
| // CHECK13-NEXT: [[TMP19:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK13-NEXT: store i32 [[TMP19]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK13-NEXT: [[TMP20:%.*]] = load i64, ptr [[N_CASTED]], align 8, !llvm.access.group [[ACC_GRP22]] |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined.omp_outlined, i64 [[TMP16]], i64 [[TMP18]], i64 [[TMP20]], i64 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP22]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK13-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP23:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]]) |
| // CHECK13-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0 |
| // CHECK13-NEXT: br i1 [[TMP26]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP27]], 0 |
| // CHECK13-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK13-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK13: omp.precond.end: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK13-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK13-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK13-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK13: omp.precond.then: |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV3:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK13-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK13-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25:![0-9]+]] |
| // CHECK13-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK13-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK13-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK13-NEXT: [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK13-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP18]] to i64 |
| // CHECK13-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[IDXPROM]] |
| // CHECK13-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK13-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK13: omp.body.continue: |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK13-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK13-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP26:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK13-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK13-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB8:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK13-NEXT: [[DIV9:%.*]] = sdiv i32 [[SUB8]], 1 |
| // CHECK13-NEXT: [[MUL10:%.*]] = mul nsw i32 [[DIV9]], 1 |
| // CHECK13-NEXT: [[ADD11:%.*]] = add nsw i32 0, [[MUL10]] |
| // CHECK13-NEXT: store i32 [[ADD11]], ptr [[I4]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK13: omp.precond.end: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147 |
| // CHECK13-SAME: (i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined, i64 [[TMP3]], i64 [[TMP0]], ptr [[TMP1]], i64 [[TMP5]]) |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK13-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK13-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK13-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK13: omp.precond.then: |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP8]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP6]]) |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]] |
| // CHECK13-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP13]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28:![0-9]+]] |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP15]], 1 |
| // CHECK13-NEXT: [[CMP6:%.*]] = icmp slt i32 [[TMP14]], [[ADD]] |
| // CHECK13-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP17:%.*]] = zext i32 [[TMP16]] to i64 |
| // CHECK13-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP19:%.*]] = zext i32 [[TMP18]] to i64 |
| // CHECK13-NEXT: [[TMP20:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: store i32 [[TMP20]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i64, ptr [[N_CASTED]], align 8, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: store i32 [[TMP22]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP23:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined.omp_outlined, i64 [[TMP17]], i64 [[TMP19]], i64 [[TMP21]], i64 [[TMP0]], ptr [[TMP1]], i64 [[TMP23]]), !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP24]], [[TMP25]] |
| // CHECK13-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[ADD8:%.*]] = add nsw i32 [[TMP26]], [[TMP27]] |
| // CHECK13-NEXT: store i32 [[ADD8]], ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP28]], [[TMP29]] |
| // CHECK13-NEXT: store i32 [[ADD9]], ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP30:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP31:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[CMP10:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]] |
| // CHECK13-NEXT: br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]] |
| // CHECK13: cond.true11: |
| // CHECK13-NEXT: [[TMP32:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: br label [[COND_END13:%.*]] |
| // CHECK13: cond.false12: |
| // CHECK13-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: br label [[COND_END13]] |
| // CHECK13: cond.end13: |
| // CHECK13-NEXT: [[COND14:%.*]] = phi i32 [ [[TMP32]], [[COND_TRUE11]] ], [ [[TMP33]], [[COND_FALSE12]] ] |
| // CHECK13-NEXT: store i32 [[COND14]], ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: store i32 [[TMP34]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP29:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: [[TMP35:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP36:%.*]] = load i32, ptr [[TMP35]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP36]]) |
| // CHECK13-NEXT: [[TMP37:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP38:%.*]] = icmp ne i32 [[TMP37]], 0 |
| // CHECK13-NEXT: br i1 [[TMP38]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: [[TMP39:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[SUB15:%.*]] = sub nsw i32 [[TMP39]], 0 |
| // CHECK13-NEXT: [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1 |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV16]], 1 |
| // CHECK13-NEXT: [[ADD17:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD17]], ptr [[I4]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK13: omp.precond.end: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I5:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK13-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK13-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK13-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK13: omp.precond.then: |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV4:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK13-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[CONV4]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: [[CMP6:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK13-NEXT: br i1 [[CMP6]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP31:![0-9]+]] |
| // CHECK13-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK13-NEXT: [[CMP7:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK13-NEXT: br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK13-NEXT: [[TMP18:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK13-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP18]] to i64 |
| // CHECK13-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[IDXPROM]] |
| // CHECK13-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK13-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK13: omp.body.continue: |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK13-NEXT: [[ADD8:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK13-NEXT: store i32 [[ADD8]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP32:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK13-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK13-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[SUB9:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK13-NEXT: [[DIV10:%.*]] = sdiv i32 [[SUB9]], 1 |
| // CHECK13-NEXT: [[MUL11:%.*]] = mul nsw i32 [[DIV10]], 1 |
| // CHECK13-NEXT: [[ADD12:%.*]] = add nsw i32 0, [[MUL11]] |
| // CHECK13-NEXT: store i32 [[ADD12]], ptr [[I5]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK13: omp.precond.end: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151 |
| // CHECK13-SAME: (i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined, i64 [[TMP3]], i64 [[TMP0]], ptr [[TMP1]]) |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK13-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK13-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK13-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK13: omp.precond.then: |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK13-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP34:![0-9]+]] |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK13-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK13-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK13-NEXT: [[TMP16:%.*]] = zext i32 [[TMP15]] to i64 |
| // CHECK13-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK13-NEXT: [[TMP18:%.*]] = zext i32 [[TMP17]] to i64 |
| // CHECK13-NEXT: [[TMP19:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK13-NEXT: store i32 [[TMP19]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK13-NEXT: [[TMP20:%.*]] = load i64, ptr [[N_CASTED]], align 8, !llvm.access.group [[ACC_GRP34]] |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined.omp_outlined, i64 [[TMP16]], i64 [[TMP18]], i64 [[TMP20]], i64 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP34]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK13-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP35:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]]) |
| // CHECK13-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0 |
| // CHECK13-NEXT: br i1 [[TMP26]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP27]], 0 |
| // CHECK13-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK13-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK13: omp.precond.end: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK13-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK13-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK13-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK13: omp.precond.then: |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV3:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK13-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP11]], i32 35, i32 [[TMP8]], i32 [[TMP9]], i32 1, i32 1) |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK13: omp.dispatch.cond: |
| // CHECK13-NEXT: [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4 |
| // CHECK13-NEXT: [[TMP14:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP13]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK13-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK13-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK13: omp.dispatch.body: |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP15]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP37:![0-9]+]] |
| // CHECK13-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK13-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]] |
| // CHECK13-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1 |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK13-NEXT: [[TMP19:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK13-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP19]] to i64 |
| // CHECK13-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[IDXPROM]] |
| // CHECK13-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK13-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK13: omp.body.continue: |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK13-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP20]], 1 |
| // CHECK13-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP38:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK13: omp.dispatch.inc: |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK13: omp.dispatch.end: |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0 |
| // CHECK13-NEXT: br i1 [[TMP22]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP23]], 0 |
| // CHECK13-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK13-NEXT: [[MUL9:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK13-NEXT: [[ADD10:%.*]] = add nsw i32 0, [[MUL9]] |
| // CHECK13-NEXT: store i32 [[ADD10]], ptr [[I4]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK13: omp.precond.end: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155 |
| // CHECK13-SAME: (i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined, i64 [[TMP3]], i64 [[TMP0]], ptr [[TMP1]], i64 [[TMP5]]) |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK13-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK13-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK13-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK13: omp.precond.then: |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK13-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP40:![0-9]+]] |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK13-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: [[TMP16:%.*]] = zext i32 [[TMP15]] to i64 |
| // CHECK13-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: [[TMP18:%.*]] = zext i32 [[TMP17]] to i64 |
| // CHECK13-NEXT: [[TMP19:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: store i32 [[TMP19]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: [[TMP20:%.*]] = load i64, ptr [[N_CASTED]], align 8, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: [[TMP22:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined.omp_outlined, i64 [[TMP16]], i64 [[TMP18]], i64 [[TMP20]], i64 [[TMP0]], ptr [[TMP1]], i64 [[TMP22]]), !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP23]], [[TMP24]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP41:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: [[TMP25:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP26:%.*]] = load i32, ptr [[TMP25]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP26]]) |
| // CHECK13-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0 |
| // CHECK13-NEXT: br i1 [[TMP28]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP29]], 0 |
| // CHECK13-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK13-NEXT: [[ADD9:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD9]], ptr [[I4]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK13: omp.precond.end: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I5:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK13-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK13-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK13-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK13: omp.precond.then: |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV4:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK13-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[CONV4]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP11:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP12]], i32 35, i32 [[TMP9]], i32 [[TMP10]], i32 1, i32 [[TMP8]]) |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK13: omp.dispatch.cond: |
| // CHECK13-NEXT: [[TMP13:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[TMP13]], align 4 |
| // CHECK13-NEXT: [[TMP15:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP14]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK13-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0 |
| // CHECK13-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK13: omp.dispatch.body: |
| // CHECK13-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP43:![0-9]+]] |
| // CHECK13-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK13-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]] |
| // CHECK13-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1 |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK13-NEXT: [[TMP20:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK13-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP20]] to i64 |
| // CHECK13-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[IDXPROM]] |
| // CHECK13-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK13-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK13: omp.body.continue: |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK13-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1 |
| // CHECK13-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP44:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK13: omp.dispatch.inc: |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK13: omp.dispatch.end: |
| // CHECK13-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK13-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK13-NEXT: [[SUB8:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK13-NEXT: [[DIV9:%.*]] = sdiv i32 [[SUB8]], 1 |
| // CHECK13-NEXT: [[MUL10:%.*]] = mul nsw i32 [[DIV9]], 1 |
| // CHECK13-NEXT: [[ADD11:%.*]] = add nsw i32 0, [[MUL10]] |
| // CHECK13-NEXT: store i32 [[ADD11]], ptr [[I5]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK13: omp.precond.end: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@_Z5tmainIiLi10EEiT_ |
| // CHECK13-SAME: (i32 noundef signext [[ARGC:%.*]]) #[[ATTR5:[0-9]+]] comdat { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[A:%.*]] = alloca [10 x i32], align 4 |
| // CHECK13-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_BASEPTRS1:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_PTRS2:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_MAPPERS3:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK13-NEXT: [[_TMP4:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_BASEPTRS8:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_PTRS9:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_MAPPERS10:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK13-NEXT: [[_TMP11:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_BASEPTRS15:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_PTRS16:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_MAPPERS17:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK13-NEXT: [[_TMP18:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR_22:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED23:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_BASEPTRS24:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_PTRS25:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK13-NEXT: [[DOTOFFLOAD_MAPPERS26:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK13-NEXT: [[_TMP27:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK13-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK13-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK13-NEXT: [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[TMP0]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[TMP1]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP2]], align 8 |
| // CHECK13-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP5]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK13-NEXT: store i32 1, ptr [[TMP6]], align 4 |
| // CHECK13-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[TMP3]], ptr [[TMP7]], align 8 |
| // CHECK13-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK13-NEXT: store ptr [[TMP4]], ptr [[TMP8]], align 8 |
| // CHECK13-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK13-NEXT: store ptr @.offload_sizes.9, ptr [[TMP9]], align 8 |
| // CHECK13-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK13-NEXT: store ptr @.offload_maptypes.10, ptr [[TMP10]], align 8 |
| // CHECK13-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP11]], align 8 |
| // CHECK13-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP12]], align 8 |
| // CHECK13-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK13-NEXT: store i64 10, ptr [[TMP13]], align 8 |
| // CHECK13-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK13-NEXT: store i64 0, ptr [[TMP14]], align 8 |
| // CHECK13-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4 |
| // CHECK13-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4 |
| // CHECK13-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK13-NEXT: store i32 0, ptr [[TMP17]], align 4 |
| // CHECK13-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK13-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0 |
| // CHECK13-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK13: omp_offload.failed: |
| // CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]] |
| // CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK13: omp_offload.cont: |
| // CHECK13-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[TMP20]], align 8 |
| // CHECK13-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[TMP21]], align 8 |
| // CHECK13-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP22]], align 8 |
| // CHECK13-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP25]], align 4 |
| // CHECK13-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1 |
| // CHECK13-NEXT: store i32 1, ptr [[TMP26]], align 4 |
| // CHECK13-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8 |
| // CHECK13-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3 |
| // CHECK13-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8 |
| // CHECK13-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4 |
| // CHECK13-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 8 |
| // CHECK13-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5 |
| // CHECK13-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 8 |
| // CHECK13-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP31]], align 8 |
| // CHECK13-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP32]], align 8 |
| // CHECK13-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8 |
| // CHECK13-NEXT: store i64 10, ptr [[TMP33]], align 8 |
| // CHECK13-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9 |
| // CHECK13-NEXT: store i64 0, ptr [[TMP34]], align 8 |
| // CHECK13-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4 |
| // CHECK13-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4 |
| // CHECK13-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12 |
| // CHECK13-NEXT: store i32 0, ptr [[TMP37]], align 4 |
| // CHECK13-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]]) |
| // CHECK13-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0 |
| // CHECK13-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]] |
| // CHECK13: omp_offload.failed6: |
| // CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]] |
| // CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT7]] |
| // CHECK13: omp_offload.cont7: |
| // CHECK13-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP42:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK13-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[TMP43]], align 8 |
| // CHECK13-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[TMP44]], align 8 |
| // CHECK13-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP45]], align 8 |
| // CHECK13-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP42]], ptr [[TMP46]], align 8 |
| // CHECK13-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP42]], ptr [[TMP47]], align 8 |
| // CHECK13-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP48]], align 8 |
| // CHECK13-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP51]], align 4 |
| // CHECK13-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1 |
| // CHECK13-NEXT: store i32 2, ptr [[TMP52]], align 4 |
| // CHECK13-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8 |
| // CHECK13-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3 |
| // CHECK13-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 8 |
| // CHECK13-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4 |
| // CHECK13-NEXT: store ptr @.offload_sizes.13, ptr [[TMP55]], align 8 |
| // CHECK13-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5 |
| // CHECK13-NEXT: store ptr @.offload_maptypes.14, ptr [[TMP56]], align 8 |
| // CHECK13-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP57]], align 8 |
| // CHECK13-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP58]], align 8 |
| // CHECK13-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8 |
| // CHECK13-NEXT: store i64 10, ptr [[TMP59]], align 8 |
| // CHECK13-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9 |
| // CHECK13-NEXT: store i64 0, ptr [[TMP60]], align 8 |
| // CHECK13-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4 |
| // CHECK13-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4 |
| // CHECK13-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12 |
| // CHECK13-NEXT: store i32 0, ptr [[TMP63]], align 4 |
| // CHECK13-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]]) |
| // CHECK13-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0 |
| // CHECK13-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]] |
| // CHECK13: omp_offload.failed13: |
| // CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i64 [[TMP42]]) #[[ATTR3]] |
| // CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT14]] |
| // CHECK13: omp_offload.cont14: |
| // CHECK13-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[TMP66]], align 8 |
| // CHECK13-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[TMP67]], align 8 |
| // CHECK13-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP68]], align 8 |
| // CHECK13-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP71]], align 4 |
| // CHECK13-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1 |
| // CHECK13-NEXT: store i32 1, ptr [[TMP72]], align 4 |
| // CHECK13-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 8 |
| // CHECK13-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3 |
| // CHECK13-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 8 |
| // CHECK13-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4 |
| // CHECK13-NEXT: store ptr @.offload_sizes.15, ptr [[TMP75]], align 8 |
| // CHECK13-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5 |
| // CHECK13-NEXT: store ptr @.offload_maptypes.16, ptr [[TMP76]], align 8 |
| // CHECK13-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP77]], align 8 |
| // CHECK13-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP78]], align 8 |
| // CHECK13-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8 |
| // CHECK13-NEXT: store i64 10, ptr [[TMP79]], align 8 |
| // CHECK13-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9 |
| // CHECK13-NEXT: store i64 0, ptr [[TMP80]], align 8 |
| // CHECK13-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4 |
| // CHECK13-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4 |
| // CHECK13-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12 |
| // CHECK13-NEXT: store i32 0, ptr [[TMP83]], align 4 |
| // CHECK13-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]]) |
| // CHECK13-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0 |
| // CHECK13-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]] |
| // CHECK13: omp_offload.failed20: |
| // CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]] |
| // CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT21]] |
| // CHECK13: omp_offload.cont21: |
| // CHECK13-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4 |
| // CHECK13-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4 |
| // CHECK13-NEXT: [[TMP88:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 8 |
| // CHECK13-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[TMP89]], align 8 |
| // CHECK13-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[TMP90]], align 8 |
| // CHECK13-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP91]], align 8 |
| // CHECK13-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP88]], ptr [[TMP92]], align 8 |
| // CHECK13-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1 |
| // CHECK13-NEXT: store i64 [[TMP88]], ptr [[TMP93]], align 8 |
| // CHECK13-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP94]], align 8 |
| // CHECK13-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0 |
| // CHECK13-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0 |
| // CHECK13-NEXT: store i32 3, ptr [[TMP97]], align 4 |
| // CHECK13-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1 |
| // CHECK13-NEXT: store i32 2, ptr [[TMP98]], align 4 |
| // CHECK13-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2 |
| // CHECK13-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 8 |
| // CHECK13-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3 |
| // CHECK13-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 8 |
| // CHECK13-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4 |
| // CHECK13-NEXT: store ptr @.offload_sizes.17, ptr [[TMP101]], align 8 |
| // CHECK13-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5 |
| // CHECK13-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP102]], align 8 |
| // CHECK13-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP103]], align 8 |
| // CHECK13-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7 |
| // CHECK13-NEXT: store ptr null, ptr [[TMP104]], align 8 |
| // CHECK13-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8 |
| // CHECK13-NEXT: store i64 10, ptr [[TMP105]], align 8 |
| // CHECK13-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9 |
| // CHECK13-NEXT: store i64 0, ptr [[TMP106]], align 8 |
| // CHECK13-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4 |
| // CHECK13-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11 |
| // CHECK13-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4 |
| // CHECK13-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12 |
| // CHECK13-NEXT: store i32 0, ptr [[TMP109]], align 4 |
| // CHECK13-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]]) |
| // CHECK13-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0 |
| // CHECK13-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]] |
| // CHECK13: omp_offload.failed29: |
| // CHECK13-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i64 [[TMP88]]) #[[ATTR3]] |
| // CHECK13-NEXT: br label [[OMP_OFFLOAD_CONT30]] |
| // CHECK13: omp_offload.cont30: |
| // CHECK13-NEXT: ret i32 0 |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112 |
| // CHECK13-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined, ptr [[TMP0]]) |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK13-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP46:![0-9]+]] |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK13-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK13-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK13-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK13-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP46]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP47:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK13-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK13-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 9 |
| // CHECK13-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP49:![0-9]+]] |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK13-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK13-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK13-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64 |
| // CHECK13-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]] |
| // CHECK13-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK13-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK13: omp.body.continue: |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK13-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK13-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP50:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK13-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116 |
| // CHECK13-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined, ptr [[TMP0]]) |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK13-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP52:![0-9]+]] |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK13-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK13-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK13-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK13-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP52]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP53:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK13-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK13-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 9 |
| // CHECK13-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP55:![0-9]+]] |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK13-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK13-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK13-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64 |
| // CHECK13-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]] |
| // CHECK13-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK13-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK13: omp.body.continue: |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK13-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK13-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP56:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK13-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120 |
| // CHECK13-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined, ptr [[TMP0]], i64 [[TMP2]]) |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK13-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP58:![0-9]+]] |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK13-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK13-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK13-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK13-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK13-NEXT: store i32 [[TMP12]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8, !llvm.access.group [[ACC_GRP58]] |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]], i64 [[TMP13]]), !llvm.access.group [[ACC_GRP58]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP14]], [[TMP15]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP59:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK13-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0 |
| // CHECK13-NEXT: br i1 [[TMP17]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK13-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP5]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP3]]) |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK13: omp.dispatch.cond: |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV2:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP6]], [[CONV2]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV3:%.*]] = trunc i64 [[TMP8]] to i32 |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ [[CONV3]], [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP10]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[CMP4:%.*]] = icmp sle i32 [[TMP11]], [[TMP12]] |
| // CHECK13-NEXT: br i1 [[CMP4]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK13: omp.dispatch.body: |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP61:![0-9]+]] |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK13-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK13-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP15]], 1 |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK13-NEXT: [[TMP16:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK13-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP16]] to i64 |
| // CHECK13-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]] |
| // CHECK13-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK13-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK13: omp.body.continue: |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK13-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP17]], 1 |
| // CHECK13-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP62:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK13: omp.dispatch.inc: |
| // CHECK13-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP18]], [[TMP19]] |
| // CHECK13-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: [[ADD8:%.*]] = add nsw i32 [[TMP20]], [[TMP21]] |
| // CHECK13-NEXT: store i32 [[ADD8]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK13: omp.dispatch.end: |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP5]]) |
| // CHECK13-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK13-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124 |
| // CHECK13-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined, ptr [[TMP0]]) |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK13-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP64:![0-9]+]] |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK13-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK13-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK13-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK13-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP64]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP65:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK13-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK13-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 35, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 1) |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK13: omp.dispatch.cond: |
| // CHECK13-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK13-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK13-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK13: omp.dispatch.body: |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP67:![0-9]+]] |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK13-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP12]] to i64 |
| // CHECK13-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]] |
| // CHECK13-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK13-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK13: omp.body.continue: |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK13-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK13-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP68:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK13: omp.dispatch.inc: |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK13: omp.dispatch.end: |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK13-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128 |
| // CHECK13-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined, ptr [[TMP0]], i64 [[TMP2]]) |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK13-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK13: cond.true: |
| // CHECK13-NEXT: br label [[COND_END:%.*]] |
| // CHECK13: cond.false: |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: br label [[COND_END]] |
| // CHECK13: cond.end: |
| // CHECK13-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK13-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP70:![0-9]+]] |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK13-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK13-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK13-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK13-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK13-NEXT: store i32 [[TMP12]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8, !llvm.access.group [[ACC_GRP70]] |
| // CHECK13-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]], i64 [[TMP13]]), !llvm.access.group [[ACC_GRP70]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP14]], [[TMP15]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP71:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK13: omp.loop.exit: |
| // CHECK13-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK13-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0 |
| // CHECK13-NEXT: br i1 [[TMP17]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK13-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined.omp_outlined |
| // CHECK13-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK13-NEXT: entry: |
| // CHECK13-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK13-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK13-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK13-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK13-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK13-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK13-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK13-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK13-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK13-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK13-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK13-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP7]], i32 35, i32 [[TMP4]], i32 [[TMP5]], i32 1, i32 [[TMP3]]) |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK13: omp.dispatch.cond: |
| // CHECK13-NEXT: [[TMP8:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP7]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK13-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP8]], 0 |
| // CHECK13-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK13: omp.dispatch.body: |
| // CHECK13-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK13-NEXT: store i32 [[TMP9]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK13: omp.inner.for.cond: |
| // CHECK13-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP73:![0-9]+]] |
| // CHECK13-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK13-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP10]], [[TMP11]] |
| // CHECK13-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK13: omp.inner.for.body: |
| // CHECK13-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK13-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1 |
| // CHECK13-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK13-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK13-NEXT: [[TMP13:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK13-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP13]] to i64 |
| // CHECK13-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]] |
| // CHECK13-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK13-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK13: omp.body.continue: |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK13: omp.inner.for.inc: |
| // CHECK13-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK13-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP14]], 1 |
| // CHECK13-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK13-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP74:![0-9]+]] |
| // CHECK13: omp.inner.for.end: |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK13: omp.dispatch.inc: |
| // CHECK13-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK13: omp.dispatch.end: |
| // CHECK13-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK13-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0 |
| // CHECK13-NEXT: br i1 [[TMP16]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK13: .omp.final.then: |
| // CHECK13-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK13-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK13: .omp.final.done: |
| // CHECK13-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@main |
| // CHECK15-SAME: (i32 noundef [[ARGC:%.*]], ptr noundef [[ARGV:%.*]]) #[[ATTR0:[0-9]+]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[RETVAL:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[ARGV_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[N:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[SAVED_STACK:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[__VLA_EXPR0:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK15-NEXT: [[N_CASTED3:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_BASEPTRS4:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_PTRS5:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_MAPPERS6:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_SIZES7:%.*]] = alloca [3 x i64], align 4 |
| // CHECK15-NEXT: [[_TMP8:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_9:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_10:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_18:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_CASTED19:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_BASEPTRS20:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_PTRS21:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_MAPPERS22:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_SIZES23:%.*]] = alloca [4 x i64], align 4 |
| // CHECK15-NEXT: [[_TMP24:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_25:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_26:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK15-NEXT: [[N_CASTED34:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_BASEPTRS35:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_PTRS36:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_MAPPERS37:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_SIZES38:%.*]] = alloca [3 x i64], align 4 |
| // CHECK15-NEXT: [[_TMP39:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_40:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_41:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_49:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_CASTED50:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED51:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_BASEPTRS52:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_PTRS53:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_MAPPERS54:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_SIZES55:%.*]] = alloca [4 x i64], align 4 |
| // CHECK15-NEXT: [[_TMP56:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_57:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_58:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK15-NEXT: store i32 0, ptr [[RETVAL]], align 4 |
| // CHECK15-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[ARGV]], ptr [[ARGV_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 100, ptr [[N]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = call ptr @llvm.stacksave.p0() |
| // CHECK15-NEXT: store ptr [[TMP1]], ptr [[SAVED_STACK]], align 4 |
| // CHECK15-NEXT: [[VLA:%.*]] = alloca i32, i32 [[TMP0]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP0]], ptr [[__VLA_EXPR0]], align 4 |
| // CHECK15-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = mul nuw i32 [[TMP0]], 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = sext i32 [[TMP4]] to i64 |
| // CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes, i32 24, i1 false) |
| // CHECK15-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 [[TMP3]], ptr [[TMP6]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 [[TMP3]], ptr [[TMP7]], align 4 |
| // CHECK15-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP8]], align 4 |
| // CHECK15-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP9]], align 4 |
| // CHECK15-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP10]], align 4 |
| // CHECK15-NEXT: [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP11]], align 4 |
| // CHECK15-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP12]], align 4 |
| // CHECK15-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP13]], align 4 |
| // CHECK15-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 2 |
| // CHECK15-NEXT: store i64 [[TMP5]], ptr [[TMP14]], align 4 |
| // CHECK15-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP15]], align 4 |
| // CHECK15-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP19]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP20]], 0 |
| // CHECK15-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK15-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK15-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], 1 |
| // CHECK15-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64 |
| // CHECK15-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP23]], align 4 |
| // CHECK15-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP24]], align 4 |
| // CHECK15-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[TMP16]], ptr [[TMP25]], align 4 |
| // CHECK15-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr [[TMP17]], ptr [[TMP26]], align 4 |
| // CHECK15-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK15-NEXT: store ptr [[TMP18]], ptr [[TMP27]], align 4 |
| // CHECK15-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK15-NEXT: store ptr @.offload_maptypes, ptr [[TMP28]], align 4 |
| // CHECK15-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP29]], align 4 |
| // CHECK15-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP30]], align 4 |
| // CHECK15-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK15-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8 |
| // CHECK15-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK15-NEXT: store i64 0, ptr [[TMP32]], align 8 |
| // CHECK15-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4 |
| // CHECK15-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4 |
| // CHECK15-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK15-NEXT: store i32 0, ptr [[TMP35]], align 4 |
| // CHECK15-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK15-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0 |
| // CHECK15-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK15: omp_offload.failed: |
| // CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]] |
| // CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK15: omp_offload.cont: |
| // CHECK15-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4 |
| // CHECK15-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4 |
| // CHECK15-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4 |
| // CHECK15-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64 |
| // CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.1, i32 24, i1 false) |
| // CHECK15-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4 |
| // CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4 |
| // CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP44]], align 4 |
| // CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4 |
| // CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4 |
| // CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP47]], align 4 |
| // CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4 |
| // CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4 |
| // CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2 |
| // CHECK15-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4 |
| // CHECK15-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP51]], align 4 |
| // CHECK15-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4 |
| // CHECK15-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4 |
| // CHECK15-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0 |
| // CHECK15-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1 |
| // CHECK15-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1 |
| // CHECK15-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4 |
| // CHECK15-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4 |
| // CHECK15-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1 |
| // CHECK15-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64 |
| // CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP59]], align 4 |
| // CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP60]], align 4 |
| // CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4 |
| // CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4 |
| // CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4 |
| // CHECK15-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4 |
| // CHECK15-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5 |
| // CHECK15-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP64]], align 4 |
| // CHECK15-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP65]], align 4 |
| // CHECK15-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP66]], align 4 |
| // CHECK15-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8 |
| // CHECK15-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8 |
| // CHECK15-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9 |
| // CHECK15-NEXT: store i64 0, ptr [[TMP68]], align 8 |
| // CHECK15-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4 |
| // CHECK15-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4 |
| // CHECK15-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12 |
| // CHECK15-NEXT: store i32 0, ptr [[TMP71]], align 4 |
| // CHECK15-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]]) |
| // CHECK15-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0 |
| // CHECK15-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]] |
| // CHECK15: omp_offload.failed16: |
| // CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]] |
| // CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT17]] |
| // CHECK15: omp_offload.cont17: |
| // CHECK15-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_18]], align 4 |
| // CHECK15-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP75]], ptr [[N_CASTED19]], align 4 |
| // CHECK15-NEXT: [[TMP76:%.*]] = load i32, ptr [[N_CASTED19]], align 4 |
| // CHECK15-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP78:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP0]], 4 |
| // CHECK15-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64 |
| // CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES23]], ptr align 4 @.offload_sizes.3, i32 32, i1 false) |
| // CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 [[TMP76]], ptr [[TMP81]], align 4 |
| // CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 [[TMP76]], ptr [[TMP82]], align 4 |
| // CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP83]], align 4 |
| // CHECK15-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP84]], align 4 |
| // CHECK15-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP85]], align 4 |
| // CHECK15-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP86]], align 4 |
| // CHECK15-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 4 |
| // CHECK15-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 4 |
| // CHECK15-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2 |
| // CHECK15-NEXT: store i64 [[TMP80]], ptr [[TMP89]], align 4 |
| // CHECK15-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP90]], align 4 |
| // CHECK15-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3 |
| // CHECK15-NEXT: store i32 [[TMP78]], ptr [[TMP91]], align 4 |
| // CHECK15-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3 |
| // CHECK15-NEXT: store i32 [[TMP78]], ptr [[TMP92]], align 4 |
| // CHECK15-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP93]], align 4 |
| // CHECK15-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP96:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP97:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP97]], ptr [[DOTCAPTURE_EXPR_25]], align 4 |
| // CHECK15-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4 |
| // CHECK15-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP98]], 0 |
| // CHECK15-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1 |
| // CHECK15-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1 |
| // CHECK15-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4 |
| // CHECK15-NEXT: [[TMP99:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4 |
| // CHECK15-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP99]], 1 |
| // CHECK15-NEXT: [[TMP100:%.*]] = zext i32 [[ADD30]] to i64 |
| // CHECK15-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP101]], align 4 |
| // CHECK15-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 4, ptr [[TMP102]], align 4 |
| // CHECK15-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4 |
| // CHECK15-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4 |
| // CHECK15-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4 |
| // CHECK15-NEXT: store ptr [[TMP96]], ptr [[TMP105]], align 4 |
| // CHECK15-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5 |
| // CHECK15-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP106]], align 4 |
| // CHECK15-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP107]], align 4 |
| // CHECK15-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP108]], align 4 |
| // CHECK15-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8 |
| // CHECK15-NEXT: store i64 [[TMP100]], ptr [[TMP109]], align 8 |
| // CHECK15-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9 |
| // CHECK15-NEXT: store i64 0, ptr [[TMP110]], align 8 |
| // CHECK15-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4 |
| // CHECK15-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP112]], align 4 |
| // CHECK15-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12 |
| // CHECK15-NEXT: store i32 0, ptr [[TMP113]], align 4 |
| // CHECK15-NEXT: [[TMP114:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]]) |
| // CHECK15-NEXT: [[TMP115:%.*]] = icmp ne i32 [[TMP114]], 0 |
| // CHECK15-NEXT: br i1 [[TMP115]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]] |
| // CHECK15: omp_offload.failed32: |
| // CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i32 [[TMP76]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP78]]) #[[ATTR3]] |
| // CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT33]] |
| // CHECK15: omp_offload.cont33: |
| // CHECK15-NEXT: [[TMP116:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP116]], ptr [[N_CASTED34]], align 4 |
| // CHECK15-NEXT: [[TMP117:%.*]] = load i32, ptr [[N_CASTED34]], align 4 |
| // CHECK15-NEXT: [[TMP118:%.*]] = mul nuw i32 [[TMP0]], 4 |
| // CHECK15-NEXT: [[TMP119:%.*]] = sext i32 [[TMP118]] to i64 |
| // CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES38]], ptr align 4 @.offload_sizes.5, i32 24, i1 false) |
| // CHECK15-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 [[TMP117]], ptr [[TMP120]], align 4 |
| // CHECK15-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 [[TMP117]], ptr [[TMP121]], align 4 |
| // CHECK15-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP122]], align 4 |
| // CHECK15-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP123]], align 4 |
| // CHECK15-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP124]], align 4 |
| // CHECK15-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 1 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP125]], align 4 |
| // CHECK15-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP126]], align 4 |
| // CHECK15-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP127]], align 4 |
| // CHECK15-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2 |
| // CHECK15-NEXT: store i64 [[TMP119]], ptr [[TMP128]], align 4 |
| // CHECK15-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP129]], align 4 |
| // CHECK15-NEXT: [[TMP130:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP131:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP132:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP133:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP133]], ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK15-NEXT: [[TMP134:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK15-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP134]], 0 |
| // CHECK15-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1 |
| // CHECK15-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1 |
| // CHECK15-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK15-NEXT: [[TMP135:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK15-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP135]], 1 |
| // CHECK15-NEXT: [[TMP136:%.*]] = zext i32 [[ADD45]] to i64 |
| // CHECK15-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP137]], align 4 |
| // CHECK15-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP138]], align 4 |
| // CHECK15-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[TMP130]], ptr [[TMP139]], align 4 |
| // CHECK15-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr [[TMP131]], ptr [[TMP140]], align 4 |
| // CHECK15-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4 |
| // CHECK15-NEXT: store ptr [[TMP132]], ptr [[TMP141]], align 4 |
| // CHECK15-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5 |
| // CHECK15-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP142]], align 4 |
| // CHECK15-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP143]], align 4 |
| // CHECK15-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP144]], align 4 |
| // CHECK15-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8 |
| // CHECK15-NEXT: store i64 [[TMP136]], ptr [[TMP145]], align 8 |
| // CHECK15-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9 |
| // CHECK15-NEXT: store i64 0, ptr [[TMP146]], align 8 |
| // CHECK15-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP147]], align 4 |
| // CHECK15-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP148]], align 4 |
| // CHECK15-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12 |
| // CHECK15-NEXT: store i32 0, ptr [[TMP149]], align 4 |
| // CHECK15-NEXT: [[TMP150:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]]) |
| // CHECK15-NEXT: [[TMP151:%.*]] = icmp ne i32 [[TMP150]], 0 |
| // CHECK15-NEXT: br i1 [[TMP151]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]] |
| // CHECK15: omp_offload.failed47: |
| // CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i32 [[TMP117]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]] |
| // CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT48]] |
| // CHECK15: omp_offload.cont48: |
| // CHECK15-NEXT: [[TMP152:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR_49]], align 4 |
| // CHECK15-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP153]], ptr [[N_CASTED50]], align 4 |
| // CHECK15-NEXT: [[TMP154:%.*]] = load i32, ptr [[N_CASTED50]], align 4 |
| // CHECK15-NEXT: [[TMP155:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP155]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4 |
| // CHECK15-NEXT: [[TMP156:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4 |
| // CHECK15-NEXT: [[TMP157:%.*]] = mul nuw i32 [[TMP0]], 4 |
| // CHECK15-NEXT: [[TMP158:%.*]] = sext i32 [[TMP157]] to i64 |
| // CHECK15-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES55]], ptr align 4 @.offload_sizes.7, i32 32, i1 false) |
| // CHECK15-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 [[TMP154]], ptr [[TMP159]], align 4 |
| // CHECK15-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 [[TMP154]], ptr [[TMP160]], align 4 |
| // CHECK15-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP161]], align 4 |
| // CHECK15-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP162]], align 4 |
| // CHECK15-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP0]], ptr [[TMP163]], align 4 |
| // CHECK15-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 1 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP164]], align 4 |
| // CHECK15-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP165]], align 4 |
| // CHECK15-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[VLA]], ptr [[TMP166]], align 4 |
| // CHECK15-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2 |
| // CHECK15-NEXT: store i64 [[TMP158]], ptr [[TMP167]], align 4 |
| // CHECK15-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP168]], align 4 |
| // CHECK15-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3 |
| // CHECK15-NEXT: store i32 [[TMP156]], ptr [[TMP169]], align 4 |
| // CHECK15-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3 |
| // CHECK15-NEXT: store i32 [[TMP156]], ptr [[TMP170]], align 4 |
| // CHECK15-NEXT: [[TMP171:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP171]], align 4 |
| // CHECK15-NEXT: [[TMP172:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP173:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP174:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP175:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP175]], ptr [[DOTCAPTURE_EXPR_57]], align 4 |
| // CHECK15-NEXT: [[TMP176:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4 |
| // CHECK15-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP176]], 0 |
| // CHECK15-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1 |
| // CHECK15-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1 |
| // CHECK15-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4 |
| // CHECK15-NEXT: [[TMP177:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4 |
| // CHECK15-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP177]], 1 |
| // CHECK15-NEXT: [[TMP178:%.*]] = zext i32 [[ADD62]] to i64 |
| // CHECK15-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP179]], align 4 |
| // CHECK15-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 4, ptr [[TMP180]], align 4 |
| // CHECK15-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[TMP172]], ptr [[TMP181]], align 4 |
| // CHECK15-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr [[TMP173]], ptr [[TMP182]], align 4 |
| // CHECK15-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4 |
| // CHECK15-NEXT: store ptr [[TMP174]], ptr [[TMP183]], align 4 |
| // CHECK15-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5 |
| // CHECK15-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP184]], align 4 |
| // CHECK15-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP185]], align 4 |
| // CHECK15-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP186]], align 4 |
| // CHECK15-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8 |
| // CHECK15-NEXT: store i64 [[TMP178]], ptr [[TMP187]], align 8 |
| // CHECK15-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9 |
| // CHECK15-NEXT: store i64 0, ptr [[TMP188]], align 8 |
| // CHECK15-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP189]], align 4 |
| // CHECK15-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP190]], align 4 |
| // CHECK15-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12 |
| // CHECK15-NEXT: store i32 0, ptr [[TMP191]], align 4 |
| // CHECK15-NEXT: [[TMP192:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]]) |
| // CHECK15-NEXT: [[TMP193:%.*]] = icmp ne i32 [[TMP192]], 0 |
| // CHECK15-NEXT: br i1 [[TMP193]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]] |
| // CHECK15: omp_offload.failed64: |
| // CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i32 [[TMP154]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP156]]) #[[ATTR3]] |
| // CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT65]] |
| // CHECK15: omp_offload.cont65: |
| // CHECK15-NEXT: [[TMP194:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4 |
| // CHECK15-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP194]]) |
| // CHECK15-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4 |
| // CHECK15-NEXT: [[TMP195:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4 |
| // CHECK15-NEXT: call void @llvm.stackrestore.p0(ptr [[TMP195]]) |
| // CHECK15-NEXT: [[TMP196:%.*]] = load i32, ptr [[RETVAL]], align 4 |
| // CHECK15-NEXT: ret i32 [[TMP196]] |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139 |
| // CHECK15-SAME: (i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2:[0-9]+]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined, i32 [[TMP3]], i32 [[TMP0]], ptr [[TMP1]]) |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK15-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK15-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK15-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK15: omp.precond.then: |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK15-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP14:![0-9]+]] |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK15-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK15-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK15-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK15-NEXT: [[TMP17:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK15-NEXT: store i32 [[TMP17]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK15-NEXT: [[TMP18:%.*]] = load i32, ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined.omp_outlined, i32 [[TMP15]], i32 [[TMP16]], i32 [[TMP18]], i32 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP14]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK15-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]]) |
| // CHECK15-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0 |
| // CHECK15-NEXT: br i1 [[TMP24]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP25]], 0 |
| // CHECK15-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK15-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK15: omp.precond.end: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK15-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK15-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK15-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK15: omp.precond.then: |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK15-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18:![0-9]+]] |
| // CHECK15-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK15-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK15-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK15-NEXT: [[TMP18:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK15-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 [[TMP18]] |
| // CHECK15-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK15-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK15: omp.body.continue: |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK15-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK15-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK15-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK15-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK15-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK15-NEXT: [[MUL9:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK15-NEXT: [[ADD10:%.*]] = add nsw i32 0, [[MUL9]] |
| // CHECK15-NEXT: store i32 [[ADD10]], ptr [[I3]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK15: omp.precond.end: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143 |
| // CHECK15-SAME: (i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined, i32 [[TMP3]], i32 [[TMP0]], ptr [[TMP1]]) |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK15-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK15-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK15-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK15: omp.precond.then: |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK15-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23:![0-9]+]] |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK15-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK15-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK15-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK15-NEXT: [[TMP17:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK15-NEXT: store i32 [[TMP17]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK15-NEXT: [[TMP18:%.*]] = load i32, ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined.omp_outlined, i32 [[TMP15]], i32 [[TMP16]], i32 [[TMP18]], i32 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP23]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK15-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP24:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]]) |
| // CHECK15-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0 |
| // CHECK15-NEXT: br i1 [[TMP24]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP25]], 0 |
| // CHECK15-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK15-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK15: omp.precond.end: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK15-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK15-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK15-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK15: omp.precond.then: |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK15-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26:![0-9]+]] |
| // CHECK15-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK15-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK15-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK15-NEXT: [[TMP18:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK15-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 [[TMP18]] |
| // CHECK15-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK15-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK15: omp.body.continue: |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK15-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK15-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP27:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK15-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK15-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK15-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK15-NEXT: [[MUL9:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK15-NEXT: [[ADD10:%.*]] = add nsw i32 0, [[MUL9]] |
| // CHECK15-NEXT: store i32 [[ADD10]], ptr [[I3]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK15: omp.precond.end: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147 |
| // CHECK15-SAME: (i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined, i32 [[TMP3]], i32 [[TMP0]], ptr [[TMP1]], i32 [[TMP5]]) |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK15-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK15-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK15-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK15: omp.precond.then: |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP8]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP6]]) |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]] |
| // CHECK15-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP13]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29:![0-9]+]] |
| // CHECK15-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP15]], 1 |
| // CHECK15-NEXT: [[CMP6:%.*]] = icmp slt i32 [[TMP14]], [[ADD]] |
| // CHECK15-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP18:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: store i32 [[TMP18]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: store i32 [[TMP20]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined.omp_outlined, i32 [[TMP16]], i32 [[TMP17]], i32 [[TMP19]], i32 [[TMP0]], ptr [[TMP1]], i32 [[TMP21]]), !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP22]], [[TMP23]] |
| // CHECK15-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[ADD8:%.*]] = add nsw i32 [[TMP24]], [[TMP25]] |
| // CHECK15-NEXT: store i32 [[ADD8]], ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP26]], [[TMP27]] |
| // CHECK15-NEXT: store i32 [[ADD9]], ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[CMP10:%.*]] = icmp sgt i32 [[TMP28]], [[TMP29]] |
| // CHECK15-NEXT: br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]] |
| // CHECK15: cond.true11: |
| // CHECK15-NEXT: [[TMP30:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: br label [[COND_END13:%.*]] |
| // CHECK15: cond.false12: |
| // CHECK15-NEXT: [[TMP31:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: br label [[COND_END13]] |
| // CHECK15: cond.end13: |
| // CHECK15-NEXT: [[COND14:%.*]] = phi i32 [ [[TMP30]], [[COND_TRUE11]] ], [ [[TMP31]], [[COND_FALSE12]] ] |
| // CHECK15-NEXT: store i32 [[COND14]], ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: [[TMP32:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: store i32 [[TMP32]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP30:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP34]]) |
| // CHECK15-NEXT: [[TMP35:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0 |
| // CHECK15-NEXT: br i1 [[TMP36]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: [[TMP37:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[SUB15:%.*]] = sub nsw i32 [[TMP37]], 0 |
| // CHECK15-NEXT: [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1 |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV16]], 1 |
| // CHECK15-NEXT: [[ADD17:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD17]], ptr [[I4]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK15: omp.precond.end: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK15-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK15-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK15-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK15: omp.precond.then: |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK15-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32:![0-9]+]] |
| // CHECK15-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK15-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK15-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK15-NEXT: [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK15-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 [[TMP18]] |
| // CHECK15-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK15-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK15: omp.body.continue: |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK15-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK15-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP33:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK15-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK15-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[SUB8:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK15-NEXT: [[DIV9:%.*]] = sdiv i32 [[SUB8]], 1 |
| // CHECK15-NEXT: [[MUL10:%.*]] = mul nsw i32 [[DIV9]], 1 |
| // CHECK15-NEXT: [[ADD11:%.*]] = add nsw i32 0, [[MUL10]] |
| // CHECK15-NEXT: store i32 [[ADD11]], ptr [[I4]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK15: omp.precond.end: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151 |
| // CHECK15-SAME: (i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined, i32 [[TMP3]], i32 [[TMP0]], ptr [[TMP1]]) |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK15-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK15-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK15-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK15: omp.precond.then: |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK15-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35:![0-9]+]] |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK15-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK15-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK15-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK15-NEXT: [[TMP17:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK15-NEXT: store i32 [[TMP17]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK15-NEXT: [[TMP18:%.*]] = load i32, ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined.omp_outlined, i32 [[TMP15]], i32 [[TMP16]], i32 [[TMP18]], i32 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP35]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK15-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP36:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]]) |
| // CHECK15-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0 |
| // CHECK15-NEXT: br i1 [[TMP24]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP25]], 0 |
| // CHECK15-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK15-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK15: omp.precond.end: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK15-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK15-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK15-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK15: omp.precond.then: |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP11]], i32 35, i32 [[TMP8]], i32 [[TMP9]], i32 1, i32 1) |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK15: omp.dispatch.cond: |
| // CHECK15-NEXT: [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4 |
| // CHECK15-NEXT: [[TMP14:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP13]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK15-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK15-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK15: omp.dispatch.body: |
| // CHECK15-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP15]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38:![0-9]+]] |
| // CHECK15-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK15-NEXT: [[CMP4:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]] |
| // CHECK15-NEXT: br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1 |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK15-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 [[TMP19]] |
| // CHECK15-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK15-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK15: omp.body.continue: |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK15-NEXT: [[ADD5:%.*]] = add nsw i32 [[TMP20]], 1 |
| // CHECK15-NEXT: store i32 [[ADD5]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP39:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK15: omp.dispatch.inc: |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK15: omp.dispatch.end: |
| // CHECK15-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0 |
| // CHECK15-NEXT: br i1 [[TMP22]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP23]], 0 |
| // CHECK15-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK15-NEXT: [[MUL8:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK15-NEXT: [[ADD9:%.*]] = add nsw i32 0, [[MUL8]] |
| // CHECK15-NEXT: store i32 [[ADD9]], ptr [[I3]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK15: omp.precond.end: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155 |
| // CHECK15-SAME: (i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined, i32 [[TMP3]], i32 [[TMP0]], ptr [[TMP1]], i32 [[TMP5]]) |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK15-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK15-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK15-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK15: omp.precond.then: |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK15-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP41:![0-9]+]] |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK15-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: [[TMP17:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: store i32 [[TMP17]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: [[TMP18:%.*]] = load i32, ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: store i32 [[TMP19]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined.omp_outlined, i32 [[TMP15]], i32 [[TMP16]], i32 [[TMP18]], i32 [[TMP0]], ptr [[TMP1]], i32 [[TMP20]]), !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP42:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]]) |
| // CHECK15-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0 |
| // CHECK15-NEXT: br i1 [[TMP26]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP27]], 0 |
| // CHECK15-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK15-NEXT: [[ADD9:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD9]], ptr [[I4]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK15: omp.precond.end: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK15-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK15-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK15-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK15: omp.precond.then: |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP11:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP12]], i32 35, i32 [[TMP9]], i32 [[TMP10]], i32 1, i32 [[TMP8]]) |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK15: omp.dispatch.cond: |
| // CHECK15-NEXT: [[TMP13:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[TMP13]], align 4 |
| // CHECK15-NEXT: [[TMP15:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP14]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK15-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0 |
| // CHECK15-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK15: omp.dispatch.body: |
| // CHECK15-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP44:![0-9]+]] |
| // CHECK15-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK15-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]] |
| // CHECK15-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1 |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK15-NEXT: [[TMP20:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK15-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 [[TMP20]] |
| // CHECK15-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK15-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK15: omp.body.continue: |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK15-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP21]], 1 |
| // CHECK15-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP45:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK15: omp.dispatch.inc: |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK15: omp.dispatch.end: |
| // CHECK15-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK15-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK15-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK15-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK15-NEXT: [[MUL9:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK15-NEXT: [[ADD10:%.*]] = add nsw i32 0, [[MUL9]] |
| // CHECK15-NEXT: store i32 [[ADD10]], ptr [[I4]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK15: omp.precond.end: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@_Z5tmainIiLi10EEiT_ |
| // CHECK15-SAME: (i32 noundef [[ARGC:%.*]]) #[[ATTR5:[0-9]+]] comdat { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A:%.*]] = alloca [10 x i32], align 4 |
| // CHECK15-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK15-NEXT: [[DOTOFFLOAD_BASEPTRS1:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_PTRS2:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_MAPPERS3:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK15-NEXT: [[_TMP4:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_BASEPTRS8:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_PTRS9:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_MAPPERS10:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK15-NEXT: [[_TMP11:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK15-NEXT: [[DOTOFFLOAD_BASEPTRS15:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_PTRS16:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_MAPPERS17:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK15-NEXT: [[_TMP18:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR_22:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED23:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_BASEPTRS24:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_PTRS25:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK15-NEXT: [[DOTOFFLOAD_MAPPERS26:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK15-NEXT: [[_TMP27:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK15-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[TMP0]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[TMP1]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP2]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP5]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 1, ptr [[TMP6]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[TMP3]], ptr [[TMP7]], align 4 |
| // CHECK15-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr [[TMP4]], ptr [[TMP8]], align 4 |
| // CHECK15-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK15-NEXT: store ptr @.offload_sizes.9, ptr [[TMP9]], align 4 |
| // CHECK15-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK15-NEXT: store ptr @.offload_maptypes.10, ptr [[TMP10]], align 4 |
| // CHECK15-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP11]], align 4 |
| // CHECK15-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP12]], align 4 |
| // CHECK15-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK15-NEXT: store i64 10, ptr [[TMP13]], align 8 |
| // CHECK15-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK15-NEXT: store i64 0, ptr [[TMP14]], align 8 |
| // CHECK15-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4 |
| // CHECK15-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4 |
| // CHECK15-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK15-NEXT: store i32 0, ptr [[TMP17]], align 4 |
| // CHECK15-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK15-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0 |
| // CHECK15-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK15: omp_offload.failed: |
| // CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]] |
| // CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK15: omp_offload.cont: |
| // CHECK15-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[TMP20]], align 4 |
| // CHECK15-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[TMP21]], align 4 |
| // CHECK15-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP22]], align 4 |
| // CHECK15-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP25]], align 4 |
| // CHECK15-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 1, ptr [[TMP26]], align 4 |
| // CHECK15-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4 |
| // CHECK15-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4 |
| // CHECK15-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4 |
| // CHECK15-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 4 |
| // CHECK15-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5 |
| // CHECK15-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 4 |
| // CHECK15-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP31]], align 4 |
| // CHECK15-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP32]], align 4 |
| // CHECK15-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8 |
| // CHECK15-NEXT: store i64 10, ptr [[TMP33]], align 8 |
| // CHECK15-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9 |
| // CHECK15-NEXT: store i64 0, ptr [[TMP34]], align 8 |
| // CHECK15-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4 |
| // CHECK15-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4 |
| // CHECK15-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12 |
| // CHECK15-NEXT: store i32 0, ptr [[TMP37]], align 4 |
| // CHECK15-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]]) |
| // CHECK15-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0 |
| // CHECK15-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]] |
| // CHECK15: omp_offload.failed6: |
| // CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]] |
| // CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT7]] |
| // CHECK15: omp_offload.cont7: |
| // CHECK15-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP42:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[TMP43]], align 4 |
| // CHECK15-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[TMP44]], align 4 |
| // CHECK15-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP45]], align 4 |
| // CHECK15-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP42]], ptr [[TMP46]], align 4 |
| // CHECK15-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP42]], ptr [[TMP47]], align 4 |
| // CHECK15-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP48]], align 4 |
| // CHECK15-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP51]], align 4 |
| // CHECK15-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 2, ptr [[TMP52]], align 4 |
| // CHECK15-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4 |
| // CHECK15-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 4 |
| // CHECK15-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4 |
| // CHECK15-NEXT: store ptr @.offload_sizes.13, ptr [[TMP55]], align 4 |
| // CHECK15-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5 |
| // CHECK15-NEXT: store ptr @.offload_maptypes.14, ptr [[TMP56]], align 4 |
| // CHECK15-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP57]], align 4 |
| // CHECK15-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP58]], align 4 |
| // CHECK15-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8 |
| // CHECK15-NEXT: store i64 10, ptr [[TMP59]], align 8 |
| // CHECK15-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9 |
| // CHECK15-NEXT: store i64 0, ptr [[TMP60]], align 8 |
| // CHECK15-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4 |
| // CHECK15-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4 |
| // CHECK15-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12 |
| // CHECK15-NEXT: store i32 0, ptr [[TMP63]], align 4 |
| // CHECK15-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]]) |
| // CHECK15-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0 |
| // CHECK15-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]] |
| // CHECK15: omp_offload.failed13: |
| // CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i32 [[TMP42]]) #[[ATTR3]] |
| // CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT14]] |
| // CHECK15: omp_offload.cont14: |
| // CHECK15-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[TMP66]], align 4 |
| // CHECK15-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[TMP67]], align 4 |
| // CHECK15-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP68]], align 4 |
| // CHECK15-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP71]], align 4 |
| // CHECK15-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 1, ptr [[TMP72]], align 4 |
| // CHECK15-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 4 |
| // CHECK15-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 4 |
| // CHECK15-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4 |
| // CHECK15-NEXT: store ptr @.offload_sizes.15, ptr [[TMP75]], align 4 |
| // CHECK15-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5 |
| // CHECK15-NEXT: store ptr @.offload_maptypes.16, ptr [[TMP76]], align 4 |
| // CHECK15-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP77]], align 4 |
| // CHECK15-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP78]], align 4 |
| // CHECK15-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8 |
| // CHECK15-NEXT: store i64 10, ptr [[TMP79]], align 8 |
| // CHECK15-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9 |
| // CHECK15-NEXT: store i64 0, ptr [[TMP80]], align 8 |
| // CHECK15-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4 |
| // CHECK15-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4 |
| // CHECK15-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12 |
| // CHECK15-NEXT: store i32 0, ptr [[TMP83]], align 4 |
| // CHECK15-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]]) |
| // CHECK15-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0 |
| // CHECK15-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]] |
| // CHECK15: omp_offload.failed20: |
| // CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]] |
| // CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT21]] |
| // CHECK15: omp_offload.cont21: |
| // CHECK15-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4 |
| // CHECK15-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4 |
| // CHECK15-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4 |
| // CHECK15-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[TMP89]], align 4 |
| // CHECK15-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[TMP90]], align 4 |
| // CHECK15-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP91]], align 4 |
| // CHECK15-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP88]], ptr [[TMP92]], align 4 |
| // CHECK15-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 [[TMP88]], ptr [[TMP93]], align 4 |
| // CHECK15-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP94]], align 4 |
| // CHECK15-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0 |
| // CHECK15-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0 |
| // CHECK15-NEXT: store i32 3, ptr [[TMP97]], align 4 |
| // CHECK15-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1 |
| // CHECK15-NEXT: store i32 2, ptr [[TMP98]], align 4 |
| // CHECK15-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2 |
| // CHECK15-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 4 |
| // CHECK15-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3 |
| // CHECK15-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 4 |
| // CHECK15-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4 |
| // CHECK15-NEXT: store ptr @.offload_sizes.17, ptr [[TMP101]], align 4 |
| // CHECK15-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5 |
| // CHECK15-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP102]], align 4 |
| // CHECK15-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP103]], align 4 |
| // CHECK15-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7 |
| // CHECK15-NEXT: store ptr null, ptr [[TMP104]], align 4 |
| // CHECK15-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8 |
| // CHECK15-NEXT: store i64 10, ptr [[TMP105]], align 8 |
| // CHECK15-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9 |
| // CHECK15-NEXT: store i64 0, ptr [[TMP106]], align 8 |
| // CHECK15-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4 |
| // CHECK15-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11 |
| // CHECK15-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4 |
| // CHECK15-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12 |
| // CHECK15-NEXT: store i32 0, ptr [[TMP109]], align 4 |
| // CHECK15-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]]) |
| // CHECK15-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0 |
| // CHECK15-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]] |
| // CHECK15: omp_offload.failed29: |
| // CHECK15-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i32 [[TMP88]]) #[[ATTR3]] |
| // CHECK15-NEXT: br label [[OMP_OFFLOAD_CONT30]] |
| // CHECK15: omp_offload.cont30: |
| // CHECK15-NEXT: ret i32 0 |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112 |
| // CHECK15-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined, ptr [[TMP0]]) |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK15-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP47:![0-9]+]] |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK15-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK15-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP47]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP48:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK15-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 9 |
| // CHECK15-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP50:![0-9]+]] |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK15-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK15-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK15-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP11]] |
| // CHECK15-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK15-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK15: omp.body.continue: |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK15-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK15-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP51:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK15-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116 |
| // CHECK15-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined, ptr [[TMP0]]) |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK15-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP53:![0-9]+]] |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK15-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK15-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP53]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP54:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK15-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 9 |
| // CHECK15-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP56:![0-9]+]] |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK15-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK15-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK15-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP11]] |
| // CHECK15-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK15-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK15: omp.body.continue: |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK15-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK15-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP57:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK15-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120 |
| // CHECK15-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined, ptr [[TMP0]], i32 [[TMP2]]) |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK15-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP59:![0-9]+]] |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK15-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK15-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK15-NEXT: store i32 [[TMP10]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]], i32 [[TMP11]]), !llvm.access.group [[ACC_GRP59]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP60:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK15-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP5]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP3]]) |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK15: omp.dispatch.cond: |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ [[TMP8]], [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP10]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP11]], [[TMP12]] |
| // CHECK15-NEXT: br i1 [[CMP1]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK15: omp.dispatch.body: |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP62:![0-9]+]] |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK15-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK15-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP15]], 1 |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK15-NEXT: [[TMP16:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK15-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP16]] |
| // CHECK15-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK15-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK15: omp.body.continue: |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK15-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP17]], 1 |
| // CHECK15-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP63:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK15: omp.dispatch.inc: |
| // CHECK15-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: [[ADD4:%.*]] = add nsw i32 [[TMP18]], [[TMP19]] |
| // CHECK15-NEXT: store i32 [[ADD4]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: [[ADD5:%.*]] = add nsw i32 [[TMP20]], [[TMP21]] |
| // CHECK15-NEXT: store i32 [[ADD5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK15: omp.dispatch.end: |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP5]]) |
| // CHECK15-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK15-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124 |
| // CHECK15-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined, ptr [[TMP0]]) |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK15-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP65:![0-9]+]] |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK15-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK15-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP65]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP66:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK15-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 35, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 1) |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK15: omp.dispatch.cond: |
| // CHECK15-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK15-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK15-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK15: omp.dispatch.body: |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP68:![0-9]+]] |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK15-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP12]] |
| // CHECK15-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK15-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK15: omp.body.continue: |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK15-NEXT: [[ADD1:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK15-NEXT: store i32 [[ADD1]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP69:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK15: omp.dispatch.inc: |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK15: omp.dispatch.end: |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK15-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128 |
| // CHECK15-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined, ptr [[TMP0]], i32 [[TMP2]]) |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK15-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK15: cond.true: |
| // CHECK15-NEXT: br label [[COND_END:%.*]] |
| // CHECK15: cond.false: |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: br label [[COND_END]] |
| // CHECK15: cond.end: |
| // CHECK15-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK15-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP71:![0-9]+]] |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK15-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK15-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK15-NEXT: store i32 [[TMP10]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK15-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]], i32 [[TMP11]]), !llvm.access.group [[ACC_GRP71]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP72:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK15: omp.loop.exit: |
| // CHECK15-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK15-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK15-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined.omp_outlined |
| // CHECK15-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK15-NEXT: entry: |
| // CHECK15-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK15-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK15-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK15-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK15-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK15-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK15-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP7]], i32 35, i32 [[TMP4]], i32 [[TMP5]], i32 1, i32 [[TMP3]]) |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK15: omp.dispatch.cond: |
| // CHECK15-NEXT: [[TMP8:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP7]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK15-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP8]], 0 |
| // CHECK15-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK15: omp.dispatch.body: |
| // CHECK15-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK15-NEXT: store i32 [[TMP9]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK15: omp.inner.for.cond: |
| // CHECK15-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP74:![0-9]+]] |
| // CHECK15-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK15-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP10]], [[TMP11]] |
| // CHECK15-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK15: omp.inner.for.body: |
| // CHECK15-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK15-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1 |
| // CHECK15-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK15-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK15-NEXT: [[TMP13:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK15-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP13]] |
| // CHECK15-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK15-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK15: omp.body.continue: |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK15: omp.inner.for.inc: |
| // CHECK15-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK15-NEXT: [[ADD1:%.*]] = add nsw i32 [[TMP14]], 1 |
| // CHECK15-NEXT: store i32 [[ADD1]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK15-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP75:![0-9]+]] |
| // CHECK15: omp.inner.for.end: |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK15: omp.dispatch.inc: |
| // CHECK15-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK15: omp.dispatch.end: |
| // CHECK15-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK15-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0 |
| // CHECK15-NEXT: br i1 [[TMP16]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK15: .omp.final.then: |
| // CHECK15-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK15-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK15: .omp.final.done: |
| // CHECK15-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@main |
| // CHECK17-SAME: (i32 noundef signext [[ARGC:%.*]], ptr noundef [[ARGV:%.*]]) #[[ATTR0:[0-9]+]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[RETVAL:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[ARGV_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[N:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[SAVED_STACK:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[__VLA_EXPR0:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 8 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK17-NEXT: [[N_CASTED3:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_BASEPTRS4:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_PTRS5:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_MAPPERS6:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_SIZES7:%.*]] = alloca [3 x i64], align 8 |
| // CHECK17-NEXT: [[_TMP8:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_9:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_10:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_18:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[N_CASTED19:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_BASEPTRS20:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_PTRS21:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_MAPPERS22:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_SIZES23:%.*]] = alloca [4 x i64], align 8 |
| // CHECK17-NEXT: [[_TMP24:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_25:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_26:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK17-NEXT: [[N_CASTED34:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_BASEPTRS35:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_PTRS36:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_MAPPERS37:%.*]] = alloca [3 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_SIZES38:%.*]] = alloca [3 x i64], align 8 |
| // CHECK17-NEXT: [[_TMP39:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_40:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_41:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_49:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[N_CASTED50:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED51:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_BASEPTRS52:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_PTRS53:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_MAPPERS54:%.*]] = alloca [4 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_SIZES55:%.*]] = alloca [4 x i64], align 8 |
| // CHECK17-NEXT: [[_TMP56:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_57:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_58:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK17-NEXT: store i32 0, ptr [[RETVAL]], align 4 |
| // CHECK17-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK17-NEXT: store ptr [[ARGV]], ptr [[ARGV_ADDR]], align 8 |
| // CHECK17-NEXT: store i32 100, ptr [[N]], align 4 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK17-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| // CHECK17-NEXT: [[TMP2:%.*]] = call ptr @llvm.stacksave.p0() |
| // CHECK17-NEXT: store ptr [[TMP2]], ptr [[SAVED_STACK]], align 8 |
| // CHECK17-NEXT: [[VLA:%.*]] = alloca i32, i64 [[TMP1]], align 4 |
| // CHECK17-NEXT: store i64 [[TMP1]], ptr [[__VLA_EXPR0]], align 8 |
| // CHECK17-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP3]], ptr [[N_CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK17-NEXT: [[TMP5:%.*]] = mul nuw i64 [[TMP1]], 4 |
| // CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes, i64 24, i1 false) |
| // CHECK17-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK17-NEXT: store i64 [[TMP4]], ptr [[TMP6]], align 8 |
| // CHECK17-NEXT: [[TMP7:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK17-NEXT: store i64 [[TMP4]], ptr [[TMP7]], align 8 |
| // CHECK17-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP8]], align 8 |
| // CHECK17-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP9]], align 8 |
| // CHECK17-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP10]], align 8 |
| // CHECK17-NEXT: [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP11]], align 8 |
| // CHECK17-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP12]], align 8 |
| // CHECK17-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP13]], align 8 |
| // CHECK17-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 2 |
| // CHECK17-NEXT: store i64 [[TMP5]], ptr [[TMP14]], align 8 |
| // CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP15]], align 8 |
| // CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP19:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP19]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP20]], 0 |
| // CHECK17-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK17-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK17-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], 1 |
| // CHECK17-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64 |
| // CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP23]], align 4 |
| // CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP24]], align 4 |
| // CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[TMP16]], ptr [[TMP25]], align 8 |
| // CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK17-NEXT: store ptr [[TMP17]], ptr [[TMP26]], align 8 |
| // CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK17-NEXT: store ptr [[TMP18]], ptr [[TMP27]], align 8 |
| // CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK17-NEXT: store ptr @.offload_maptypes, ptr [[TMP28]], align 8 |
| // CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP29]], align 8 |
| // CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP30]], align 8 |
| // CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK17-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8 |
| // CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK17-NEXT: store i64 0, ptr [[TMP32]], align 8 |
| // CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4 |
| // CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4 |
| // CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK17-NEXT: store i32 0, ptr [[TMP35]], align 4 |
| // CHECK17-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK17-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0 |
| // CHECK17-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK17: omp_offload.failed: |
| // CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i64 [[TMP4]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3:[0-9]+]] |
| // CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK17: omp_offload.cont: |
| // CHECK17-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4 |
| // CHECK17-NEXT: [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8 |
| // CHECK17-NEXT: [[TMP40:%.*]] = mul nuw i64 [[TMP1]], 4 |
| // CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES7]], ptr align 8 @.offload_sizes.1, i64 24, i1 false) |
| // CHECK17-NEXT: [[TMP41:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0 |
| // CHECK17-NEXT: store i64 [[TMP39]], ptr [[TMP41]], align 8 |
| // CHECK17-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0 |
| // CHECK17-NEXT: store i64 [[TMP39]], ptr [[TMP42]], align 8 |
| // CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP43]], align 8 |
| // CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP44]], align 8 |
| // CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP45]], align 8 |
| // CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP46]], align 8 |
| // CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP47]], align 8 |
| // CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 8 |
| // CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2 |
| // CHECK17-NEXT: store i64 [[TMP40]], ptr [[TMP49]], align 8 |
| // CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP50]], align 8 |
| // CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP54:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP54]], ptr [[DOTCAPTURE_EXPR_9]], align 4 |
| // CHECK17-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4 |
| // CHECK17-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP55]], 0 |
| // CHECK17-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1 |
| // CHECK17-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1 |
| // CHECK17-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4 |
| // CHECK17-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4 |
| // CHECK17-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP56]], 1 |
| // CHECK17-NEXT: [[TMP57:%.*]] = zext i32 [[ADD14]] to i64 |
| // CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP58]], align 4 |
| // CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP59]], align 4 |
| // CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[TMP51]], ptr [[TMP60]], align 8 |
| // CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3 |
| // CHECK17-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 8 |
| // CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4 |
| // CHECK17-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 8 |
| // CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5 |
| // CHECK17-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP63]], align 8 |
| // CHECK17-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP64]], align 8 |
| // CHECK17-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP65]], align 8 |
| // CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8 |
| // CHECK17-NEXT: store i64 [[TMP57]], ptr [[TMP66]], align 8 |
| // CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9 |
| // CHECK17-NEXT: store i64 0, ptr [[TMP67]], align 8 |
| // CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP68]], align 4 |
| // CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4 |
| // CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12 |
| // CHECK17-NEXT: store i32 0, ptr [[TMP70]], align 4 |
| // CHECK17-NEXT: [[TMP71:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]]) |
| // CHECK17-NEXT: [[TMP72:%.*]] = icmp ne i32 [[TMP71]], 0 |
| // CHECK17-NEXT: br i1 [[TMP72]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]] |
| // CHECK17: omp_offload.failed16: |
| // CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i64 [[TMP39]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]] |
| // CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT17]] |
| // CHECK17: omp_offload.cont17: |
| // CHECK17-NEXT: [[TMP73:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP73]], ptr [[DOTCAPTURE_EXPR_18]], align 4 |
| // CHECK17-NEXT: [[TMP74:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP74]], ptr [[N_CASTED19]], align 4 |
| // CHECK17-NEXT: [[TMP75:%.*]] = load i64, ptr [[N_CASTED19]], align 8 |
| // CHECK17-NEXT: [[TMP76:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP76]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP77:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK17-NEXT: [[TMP78:%.*]] = mul nuw i64 [[TMP1]], 4 |
| // CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES23]], ptr align 8 @.offload_sizes.3, i64 32, i1 false) |
| // CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0 |
| // CHECK17-NEXT: store i64 [[TMP75]], ptr [[TMP79]], align 8 |
| // CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0 |
| // CHECK17-NEXT: store i64 [[TMP75]], ptr [[TMP80]], align 8 |
| // CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 0 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP81]], align 8 |
| // CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP82]], align 8 |
| // CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP83]], align 8 |
| // CHECK17-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 1 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP84]], align 8 |
| // CHECK17-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP85]], align 8 |
| // CHECK17-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP86]], align 8 |
| // CHECK17-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2 |
| // CHECK17-NEXT: store i64 [[TMP78]], ptr [[TMP87]], align 8 |
| // CHECK17-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 2 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP88]], align 8 |
| // CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3 |
| // CHECK17-NEXT: store i64 [[TMP77]], ptr [[TMP89]], align 8 |
| // CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3 |
| // CHECK17-NEXT: store i64 [[TMP77]], ptr [[TMP90]], align 8 |
| // CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i64 0, i64 3 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP91]], align 8 |
| // CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP95:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP95]], ptr [[DOTCAPTURE_EXPR_25]], align 4 |
| // CHECK17-NEXT: [[TMP96:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4 |
| // CHECK17-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP96]], 0 |
| // CHECK17-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1 |
| // CHECK17-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1 |
| // CHECK17-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4 |
| // CHECK17-NEXT: [[TMP97:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4 |
| // CHECK17-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP97]], 1 |
| // CHECK17-NEXT: [[TMP98:%.*]] = zext i32 [[ADD30]] to i64 |
| // CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP99]], align 4 |
| // CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1 |
| // CHECK17-NEXT: store i32 4, ptr [[TMP100]], align 4 |
| // CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[TMP92]], ptr [[TMP101]], align 8 |
| // CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3 |
| // CHECK17-NEXT: store ptr [[TMP93]], ptr [[TMP102]], align 8 |
| // CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4 |
| // CHECK17-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 8 |
| // CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5 |
| // CHECK17-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP104]], align 8 |
| // CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP105]], align 8 |
| // CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP106]], align 8 |
| // CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8 |
| // CHECK17-NEXT: store i64 [[TMP98]], ptr [[TMP107]], align 8 |
| // CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9 |
| // CHECK17-NEXT: store i64 0, ptr [[TMP108]], align 8 |
| // CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP109]], align 4 |
| // CHECK17-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP110]], align 4 |
| // CHECK17-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12 |
| // CHECK17-NEXT: store i32 0, ptr [[TMP111]], align 4 |
| // CHECK17-NEXT: [[TMP112:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]]) |
| // CHECK17-NEXT: [[TMP113:%.*]] = icmp ne i32 [[TMP112]], 0 |
| // CHECK17-NEXT: br i1 [[TMP113]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]] |
| // CHECK17: omp_offload.failed32: |
| // CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i64 [[TMP75]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP77]]) #[[ATTR3]] |
| // CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT33]] |
| // CHECK17: omp_offload.cont33: |
| // CHECK17-NEXT: [[TMP114:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP114]], ptr [[N_CASTED34]], align 4 |
| // CHECK17-NEXT: [[TMP115:%.*]] = load i64, ptr [[N_CASTED34]], align 8 |
| // CHECK17-NEXT: [[TMP116:%.*]] = mul nuw i64 [[TMP1]], 4 |
| // CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES38]], ptr align 8 @.offload_sizes.5, i64 24, i1 false) |
| // CHECK17-NEXT: [[TMP117:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0 |
| // CHECK17-NEXT: store i64 [[TMP115]], ptr [[TMP117]], align 8 |
| // CHECK17-NEXT: [[TMP118:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0 |
| // CHECK17-NEXT: store i64 [[TMP115]], ptr [[TMP118]], align 8 |
| // CHECK17-NEXT: [[TMP119:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 0 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP119]], align 8 |
| // CHECK17-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP120]], align 8 |
| // CHECK17-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP121]], align 8 |
| // CHECK17-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 1 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP122]], align 8 |
| // CHECK17-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP123]], align 8 |
| // CHECK17-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP124]], align 8 |
| // CHECK17-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2 |
| // CHECK17-NEXT: store i64 [[TMP116]], ptr [[TMP125]], align 8 |
| // CHECK17-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i64 0, i64 2 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP126]], align 8 |
| // CHECK17-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP130:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP130]], ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK17-NEXT: [[TMP131:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK17-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP131]], 0 |
| // CHECK17-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1 |
| // CHECK17-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1 |
| // CHECK17-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK17-NEXT: [[TMP132:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK17-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP132]], 1 |
| // CHECK17-NEXT: [[TMP133:%.*]] = zext i32 [[ADD45]] to i64 |
| // CHECK17-NEXT: [[TMP134:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP134]], align 4 |
| // CHECK17-NEXT: [[TMP135:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP135]], align 4 |
| // CHECK17-NEXT: [[TMP136:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[TMP127]], ptr [[TMP136]], align 8 |
| // CHECK17-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3 |
| // CHECK17-NEXT: store ptr [[TMP128]], ptr [[TMP137]], align 8 |
| // CHECK17-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4 |
| // CHECK17-NEXT: store ptr [[TMP129]], ptr [[TMP138]], align 8 |
| // CHECK17-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5 |
| // CHECK17-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP139]], align 8 |
| // CHECK17-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP140]], align 8 |
| // CHECK17-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP141]], align 8 |
| // CHECK17-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8 |
| // CHECK17-NEXT: store i64 [[TMP133]], ptr [[TMP142]], align 8 |
| // CHECK17-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9 |
| // CHECK17-NEXT: store i64 0, ptr [[TMP143]], align 8 |
| // CHECK17-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP144]], align 4 |
| // CHECK17-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP145]], align 4 |
| // CHECK17-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12 |
| // CHECK17-NEXT: store i32 0, ptr [[TMP146]], align 4 |
| // CHECK17-NEXT: [[TMP147:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]]) |
| // CHECK17-NEXT: [[TMP148:%.*]] = icmp ne i32 [[TMP147]], 0 |
| // CHECK17-NEXT: br i1 [[TMP148]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]] |
| // CHECK17: omp_offload.failed47: |
| // CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i64 [[TMP115]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR3]] |
| // CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT48]] |
| // CHECK17: omp_offload.cont48: |
| // CHECK17-NEXT: [[TMP149:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP149]], ptr [[DOTCAPTURE_EXPR_49]], align 4 |
| // CHECK17-NEXT: [[TMP150:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP150]], ptr [[N_CASTED50]], align 4 |
| // CHECK17-NEXT: [[TMP151:%.*]] = load i64, ptr [[N_CASTED50]], align 8 |
| // CHECK17-NEXT: [[TMP152:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4 |
| // CHECK17-NEXT: [[TMP153:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 8 |
| // CHECK17-NEXT: [[TMP154:%.*]] = mul nuw i64 [[TMP1]], 4 |
| // CHECK17-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES55]], ptr align 8 @.offload_sizes.7, i64 32, i1 false) |
| // CHECK17-NEXT: [[TMP155:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0 |
| // CHECK17-NEXT: store i64 [[TMP151]], ptr [[TMP155]], align 8 |
| // CHECK17-NEXT: [[TMP156:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0 |
| // CHECK17-NEXT: store i64 [[TMP151]], ptr [[TMP156]], align 8 |
| // CHECK17-NEXT: [[TMP157:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 0 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP157]], align 8 |
| // CHECK17-NEXT: [[TMP158:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP158]], align 8 |
| // CHECK17-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP1]], ptr [[TMP159]], align 8 |
| // CHECK17-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 1 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP160]], align 8 |
| // CHECK17-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP161]], align 8 |
| // CHECK17-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[VLA]], ptr [[TMP162]], align 8 |
| // CHECK17-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2 |
| // CHECK17-NEXT: store i64 [[TMP154]], ptr [[TMP163]], align 8 |
| // CHECK17-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 2 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP164]], align 8 |
| // CHECK17-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3 |
| // CHECK17-NEXT: store i64 [[TMP153]], ptr [[TMP165]], align 8 |
| // CHECK17-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3 |
| // CHECK17-NEXT: store i64 [[TMP153]], ptr [[TMP166]], align 8 |
| // CHECK17-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i64 0, i64 3 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP167]], align 8 |
| // CHECK17-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP171:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP171]], ptr [[DOTCAPTURE_EXPR_57]], align 4 |
| // CHECK17-NEXT: [[TMP172:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4 |
| // CHECK17-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP172]], 0 |
| // CHECK17-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1 |
| // CHECK17-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1 |
| // CHECK17-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4 |
| // CHECK17-NEXT: [[TMP173:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4 |
| // CHECK17-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP173]], 1 |
| // CHECK17-NEXT: [[TMP174:%.*]] = zext i32 [[ADD62]] to i64 |
| // CHECK17-NEXT: [[TMP175:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP175]], align 4 |
| // CHECK17-NEXT: [[TMP176:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1 |
| // CHECK17-NEXT: store i32 4, ptr [[TMP176]], align 4 |
| // CHECK17-NEXT: [[TMP177:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[TMP168]], ptr [[TMP177]], align 8 |
| // CHECK17-NEXT: [[TMP178:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3 |
| // CHECK17-NEXT: store ptr [[TMP169]], ptr [[TMP178]], align 8 |
| // CHECK17-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4 |
| // CHECK17-NEXT: store ptr [[TMP170]], ptr [[TMP179]], align 8 |
| // CHECK17-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5 |
| // CHECK17-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP180]], align 8 |
| // CHECK17-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP181]], align 8 |
| // CHECK17-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP182]], align 8 |
| // CHECK17-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8 |
| // CHECK17-NEXT: store i64 [[TMP174]], ptr [[TMP183]], align 8 |
| // CHECK17-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9 |
| // CHECK17-NEXT: store i64 0, ptr [[TMP184]], align 8 |
| // CHECK17-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP185]], align 4 |
| // CHECK17-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP186]], align 4 |
| // CHECK17-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12 |
| // CHECK17-NEXT: store i32 0, ptr [[TMP187]], align 4 |
| // CHECK17-NEXT: [[TMP188:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]]) |
| // CHECK17-NEXT: [[TMP189:%.*]] = icmp ne i32 [[TMP188]], 0 |
| // CHECK17-NEXT: br i1 [[TMP189]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]] |
| // CHECK17: omp_offload.failed64: |
| // CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i64 [[TMP151]], i64 [[TMP1]], ptr [[VLA]], i64 [[TMP153]]) #[[ATTR3]] |
| // CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT65]] |
| // CHECK17: omp_offload.cont65: |
| // CHECK17-NEXT: [[TMP190:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4 |
| // CHECK17-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP190]]) |
| // CHECK17-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4 |
| // CHECK17-NEXT: [[TMP191:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8 |
| // CHECK17-NEXT: call void @llvm.stackrestore.p0(ptr [[TMP191]]) |
| // CHECK17-NEXT: [[TMP192:%.*]] = load i32, ptr [[RETVAL]], align 4 |
| // CHECK17-NEXT: ret i32 [[TMP192]] |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139 |
| // CHECK17-SAME: (i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2:[0-9]+]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined, i64 [[TMP3]], i64 [[TMP0]], ptr [[TMP1]]) |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK17-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK17-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK17-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK17: omp.precond.then: |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK17-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13:![0-9]+]] |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK17-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK17-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK17-NEXT: [[TMP16:%.*]] = zext i32 [[TMP15]] to i64 |
| // CHECK17-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK17-NEXT: [[TMP18:%.*]] = zext i32 [[TMP17]] to i64 |
| // CHECK17-NEXT: [[TMP19:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK17-NEXT: store i32 [[TMP19]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK17-NEXT: [[TMP20:%.*]] = load i64, ptr [[N_CASTED]], align 8, !llvm.access.group [[ACC_GRP13]] |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined.omp_outlined, i64 [[TMP16]], i64 [[TMP18]], i64 [[TMP20]], i64 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP13]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK17-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP14:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]]) |
| // CHECK17-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0 |
| // CHECK17-NEXT: br i1 [[TMP26]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP27]], 0 |
| // CHECK17-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK17-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK17: omp.precond.end: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK17-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK17-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK17-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK17: omp.precond.then: |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV3:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK17-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK17-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17:![0-9]+]] |
| // CHECK17-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK17-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK17-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK17-NEXT: [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK17-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP18]] to i64 |
| // CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[IDXPROM]] |
| // CHECK17-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK17-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK17: omp.body.continue: |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK17-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK17-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP17]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP18:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK17-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK17-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB8:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK17-NEXT: [[DIV9:%.*]] = sdiv i32 [[SUB8]], 1 |
| // CHECK17-NEXT: [[MUL10:%.*]] = mul nsw i32 [[DIV9]], 1 |
| // CHECK17-NEXT: [[ADD11:%.*]] = add nsw i32 0, [[MUL10]] |
| // CHECK17-NEXT: store i32 [[ADD11]], ptr [[I4]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK17: omp.precond.end: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143 |
| // CHECK17-SAME: (i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined, i64 [[TMP3]], i64 [[TMP0]], ptr [[TMP1]]) |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK17-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK17-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK17-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK17: omp.precond.then: |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK17-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22:![0-9]+]] |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK17-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK17-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK17-NEXT: [[TMP16:%.*]] = zext i32 [[TMP15]] to i64 |
| // CHECK17-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK17-NEXT: [[TMP18:%.*]] = zext i32 [[TMP17]] to i64 |
| // CHECK17-NEXT: [[TMP19:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK17-NEXT: store i32 [[TMP19]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK17-NEXT: [[TMP20:%.*]] = load i64, ptr [[N_CASTED]], align 8, !llvm.access.group [[ACC_GRP22]] |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined.omp_outlined, i64 [[TMP16]], i64 [[TMP18]], i64 [[TMP20]], i64 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP22]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK17-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP23:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]]) |
| // CHECK17-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0 |
| // CHECK17-NEXT: br i1 [[TMP26]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP27]], 0 |
| // CHECK17-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK17-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK17: omp.precond.end: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK17-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK17-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK17-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK17: omp.precond.then: |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV3:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK17-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK17-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25:![0-9]+]] |
| // CHECK17-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK17-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK17-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK17-NEXT: [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK17-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP18]] to i64 |
| // CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[IDXPROM]] |
| // CHECK17-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK17-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK17: omp.body.continue: |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK17-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK17-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP26:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK17-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK17-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB8:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK17-NEXT: [[DIV9:%.*]] = sdiv i32 [[SUB8]], 1 |
| // CHECK17-NEXT: [[MUL10:%.*]] = mul nsw i32 [[DIV9]], 1 |
| // CHECK17-NEXT: [[ADD11:%.*]] = add nsw i32 0, [[MUL10]] |
| // CHECK17-NEXT: store i32 [[ADD11]], ptr [[I4]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK17: omp.precond.end: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147 |
| // CHECK17-SAME: (i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined, i64 [[TMP3]], i64 [[TMP0]], ptr [[TMP1]], i64 [[TMP5]]) |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK17-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK17-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK17-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK17: omp.precond.then: |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP8]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP6]]) |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]] |
| // CHECK17-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP13]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28:![0-9]+]] |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP15]], 1 |
| // CHECK17-NEXT: [[CMP6:%.*]] = icmp slt i32 [[TMP14]], [[ADD]] |
| // CHECK17-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP17:%.*]] = zext i32 [[TMP16]] to i64 |
| // CHECK17-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP19:%.*]] = zext i32 [[TMP18]] to i64 |
| // CHECK17-NEXT: [[TMP20:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: store i32 [[TMP20]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i64, ptr [[N_CASTED]], align 8, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: store i32 [[TMP22]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP23:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined.omp_outlined, i64 [[TMP17]], i64 [[TMP19]], i64 [[TMP21]], i64 [[TMP0]], ptr [[TMP1]], i64 [[TMP23]]), !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP24]], [[TMP25]] |
| // CHECK17-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[ADD8:%.*]] = add nsw i32 [[TMP26]], [[TMP27]] |
| // CHECK17-NEXT: store i32 [[ADD8]], ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP28]], [[TMP29]] |
| // CHECK17-NEXT: store i32 [[ADD9]], ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP30:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP31:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[CMP10:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]] |
| // CHECK17-NEXT: br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]] |
| // CHECK17: cond.true11: |
| // CHECK17-NEXT: [[TMP32:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: br label [[COND_END13:%.*]] |
| // CHECK17: cond.false12: |
| // CHECK17-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: br label [[COND_END13]] |
| // CHECK17: cond.end13: |
| // CHECK17-NEXT: [[COND14:%.*]] = phi i32 [ [[TMP32]], [[COND_TRUE11]] ], [ [[TMP33]], [[COND_FALSE12]] ] |
| // CHECK17-NEXT: store i32 [[COND14]], ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: store i32 [[TMP34]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP29:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: [[TMP35:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP36:%.*]] = load i32, ptr [[TMP35]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP36]]) |
| // CHECK17-NEXT: [[TMP37:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP38:%.*]] = icmp ne i32 [[TMP37]], 0 |
| // CHECK17-NEXT: br i1 [[TMP38]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: [[TMP39:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[SUB15:%.*]] = sub nsw i32 [[TMP39]], 0 |
| // CHECK17-NEXT: [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1 |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV16]], 1 |
| // CHECK17-NEXT: [[ADD17:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD17]], ptr [[I4]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK17: omp.precond.end: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I5:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK17-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK17-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK17-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK17: omp.precond.then: |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV4:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK17-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[CONV4]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: [[CMP6:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK17-NEXT: br i1 [[CMP6]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP31:![0-9]+]] |
| // CHECK17-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK17-NEXT: [[CMP7:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK17-NEXT: br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK17-NEXT: [[TMP18:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK17-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP18]] to i64 |
| // CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[IDXPROM]] |
| // CHECK17-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK17-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK17: omp.body.continue: |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK17-NEXT: [[ADD8:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK17-NEXT: store i32 [[ADD8]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP32:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK17-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK17-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[SUB9:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK17-NEXT: [[DIV10:%.*]] = sdiv i32 [[SUB9]], 1 |
| // CHECK17-NEXT: [[MUL11:%.*]] = mul nsw i32 [[DIV10]], 1 |
| // CHECK17-NEXT: [[ADD12:%.*]] = add nsw i32 0, [[MUL11]] |
| // CHECK17-NEXT: store i32 [[ADD12]], ptr [[I5]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK17: omp.precond.end: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151 |
| // CHECK17-SAME: (i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined, i64 [[TMP3]], i64 [[TMP0]], ptr [[TMP1]]) |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK17-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK17-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK17-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK17: omp.precond.then: |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK17-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP34:![0-9]+]] |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK17-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK17-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK17-NEXT: [[TMP16:%.*]] = zext i32 [[TMP15]] to i64 |
| // CHECK17-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK17-NEXT: [[TMP18:%.*]] = zext i32 [[TMP17]] to i64 |
| // CHECK17-NEXT: [[TMP19:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK17-NEXT: store i32 [[TMP19]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK17-NEXT: [[TMP20:%.*]] = load i64, ptr [[N_CASTED]], align 8, !llvm.access.group [[ACC_GRP34]] |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined.omp_outlined, i64 [[TMP16]], i64 [[TMP18]], i64 [[TMP20]], i64 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP34]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK17-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP34]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP35:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]]) |
| // CHECK17-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0 |
| // CHECK17-NEXT: br i1 [[TMP26]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP27]], 0 |
| // CHECK17-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK17-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK17: omp.precond.end: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK17-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK17-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK17-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK17: omp.precond.then: |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV3:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK17-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP11]], i32 1073741859, i32 [[TMP8]], i32 [[TMP9]], i32 1, i32 1) |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK17: omp.dispatch.cond: |
| // CHECK17-NEXT: [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4 |
| // CHECK17-NEXT: [[TMP14:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP13]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK17-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK17-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK17: omp.dispatch.body: |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP15]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP37:![0-9]+]] |
| // CHECK17-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK17-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]] |
| // CHECK17-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1 |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK17-NEXT: [[TMP19:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK17-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP19]] to i64 |
| // CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[IDXPROM]] |
| // CHECK17-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK17-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK17: omp.body.continue: |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK17-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP20]], 1 |
| // CHECK17-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP37]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP38:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK17: omp.dispatch.inc: |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK17: omp.dispatch.end: |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0 |
| // CHECK17-NEXT: br i1 [[TMP22]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP23]], 0 |
| // CHECK17-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK17-NEXT: [[MUL9:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK17-NEXT: [[ADD10:%.*]] = add nsw i32 0, [[MUL9]] |
| // CHECK17-NEXT: store i32 [[ADD10]], ptr [[I4]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK17: omp.precond.end: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155 |
| // CHECK17-SAME: (i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i64, ptr [[N_CASTED]], align 8 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined, i64 [[TMP3]], i64 [[TMP0]], ptr [[TMP1]], i64 [[TMP5]]) |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[N_CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK17-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK17-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK17-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK17: omp.precond.then: |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK17-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP40:![0-9]+]] |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK17-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: [[TMP16:%.*]] = zext i32 [[TMP15]] to i64 |
| // CHECK17-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: [[TMP18:%.*]] = zext i32 [[TMP17]] to i64 |
| // CHECK17-NEXT: [[TMP19:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: store i32 [[TMP19]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: [[TMP20:%.*]] = load i64, ptr [[N_CASTED]], align 8, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: [[TMP22:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined.omp_outlined, i64 [[TMP16]], i64 [[TMP18]], i64 [[TMP20]], i64 [[TMP0]], ptr [[TMP1]], i64 [[TMP22]]), !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP23]], [[TMP24]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP40]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP41:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: [[TMP25:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP26:%.*]] = load i32, ptr [[TMP25]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP26]]) |
| // CHECK17-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP28:%.*]] = icmp ne i32 [[TMP27]], 0 |
| // CHECK17-NEXT: br i1 [[TMP28]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP29]], 0 |
| // CHECK17-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK17-NEXT: [[ADD9:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD9]], ptr [[I4]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK17: omp.precond.end: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], i64 noundef [[N:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[N_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I5:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK17-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK17-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK17-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK17: omp.precond.then: |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV:%.*]] = trunc i64 [[TMP6]] to i32 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV4:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK17-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[CONV4]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP11:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP12]], i32 1073741859, i32 [[TMP9]], i32 [[TMP10]], i32 1, i32 [[TMP8]]) |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK17: omp.dispatch.cond: |
| // CHECK17-NEXT: [[TMP13:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[TMP13]], align 4 |
| // CHECK17-NEXT: [[TMP15:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP14]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK17-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0 |
| // CHECK17-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK17: omp.dispatch.body: |
| // CHECK17-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP43:![0-9]+]] |
| // CHECK17-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK17-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]] |
| // CHECK17-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1 |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK17-NEXT: [[TMP20:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK17-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP20]] to i64 |
| // CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i64 [[IDXPROM]] |
| // CHECK17-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK17-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK17: omp.body.continue: |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK17-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP21]], 1 |
| // CHECK17-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP43]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP44:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK17: omp.dispatch.inc: |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK17: omp.dispatch.end: |
| // CHECK17-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK17-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK17-NEXT: [[SUB8:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK17-NEXT: [[DIV9:%.*]] = sdiv i32 [[SUB8]], 1 |
| // CHECK17-NEXT: [[MUL10:%.*]] = mul nsw i32 [[DIV9]], 1 |
| // CHECK17-NEXT: [[ADD11:%.*]] = add nsw i32 0, [[MUL10]] |
| // CHECK17-NEXT: store i32 [[ADD11]], ptr [[I5]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK17: omp.precond.end: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@_Z5tmainIiLi10EEiT_ |
| // CHECK17-SAME: (i32 noundef signext [[ARGC:%.*]]) #[[ATTR5:[0-9]+]] comdat { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[A:%.*]] = alloca [10 x i32], align 4 |
| // CHECK17-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_BASEPTRS1:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_PTRS2:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_MAPPERS3:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK17-NEXT: [[_TMP4:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_BASEPTRS8:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_PTRS9:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_MAPPERS10:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK17-NEXT: [[_TMP11:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_BASEPTRS15:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_PTRS16:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_MAPPERS17:%.*]] = alloca [1 x ptr], align 8 |
| // CHECK17-NEXT: [[_TMP18:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR_22:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED23:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_BASEPTRS24:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_PTRS25:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK17-NEXT: [[DOTOFFLOAD_MAPPERS26:%.*]] = alloca [2 x ptr], align 8 |
| // CHECK17-NEXT: [[_TMP27:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK17-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK17-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK17-NEXT: [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[TMP0]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[TMP1]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP2]], align 8 |
| // CHECK17-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP5]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK17-NEXT: store i32 1, ptr [[TMP6]], align 4 |
| // CHECK17-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[TMP3]], ptr [[TMP7]], align 8 |
| // CHECK17-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK17-NEXT: store ptr [[TMP4]], ptr [[TMP8]], align 8 |
| // CHECK17-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK17-NEXT: store ptr @.offload_sizes.9, ptr [[TMP9]], align 8 |
| // CHECK17-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK17-NEXT: store ptr @.offload_maptypes.10, ptr [[TMP10]], align 8 |
| // CHECK17-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP11]], align 8 |
| // CHECK17-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP12]], align 8 |
| // CHECK17-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK17-NEXT: store i64 10, ptr [[TMP13]], align 8 |
| // CHECK17-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK17-NEXT: store i64 0, ptr [[TMP14]], align 8 |
| // CHECK17-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4 |
| // CHECK17-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4 |
| // CHECK17-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK17-NEXT: store i32 0, ptr [[TMP17]], align 4 |
| // CHECK17-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK17-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0 |
| // CHECK17-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK17: omp_offload.failed: |
| // CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]] |
| // CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK17: omp_offload.cont: |
| // CHECK17-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[TMP20]], align 8 |
| // CHECK17-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[TMP21]], align 8 |
| // CHECK17-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP22]], align 8 |
| // CHECK17-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP25]], align 4 |
| // CHECK17-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1 |
| // CHECK17-NEXT: store i32 1, ptr [[TMP26]], align 4 |
| // CHECK17-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 8 |
| // CHECK17-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3 |
| // CHECK17-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 8 |
| // CHECK17-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4 |
| // CHECK17-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 8 |
| // CHECK17-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5 |
| // CHECK17-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 8 |
| // CHECK17-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP31]], align 8 |
| // CHECK17-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP32]], align 8 |
| // CHECK17-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8 |
| // CHECK17-NEXT: store i64 10, ptr [[TMP33]], align 8 |
| // CHECK17-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9 |
| // CHECK17-NEXT: store i64 0, ptr [[TMP34]], align 8 |
| // CHECK17-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4 |
| // CHECK17-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4 |
| // CHECK17-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12 |
| // CHECK17-NEXT: store i32 0, ptr [[TMP37]], align 4 |
| // CHECK17-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]]) |
| // CHECK17-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0 |
| // CHECK17-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]] |
| // CHECK17: omp_offload.failed6: |
| // CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]] |
| // CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT7]] |
| // CHECK17: omp_offload.cont7: |
| // CHECK17-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP42:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK17-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[TMP43]], align 8 |
| // CHECK17-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[TMP44]], align 8 |
| // CHECK17-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 0 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP45]], align 8 |
| // CHECK17-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP42]], ptr [[TMP46]], align 8 |
| // CHECK17-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP42]], ptr [[TMP47]], align 8 |
| // CHECK17-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i64 0, i64 1 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP48]], align 8 |
| // CHECK17-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP51]], align 4 |
| // CHECK17-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1 |
| // CHECK17-NEXT: store i32 2, ptr [[TMP52]], align 4 |
| // CHECK17-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 8 |
| // CHECK17-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3 |
| // CHECK17-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 8 |
| // CHECK17-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4 |
| // CHECK17-NEXT: store ptr @.offload_sizes.13, ptr [[TMP55]], align 8 |
| // CHECK17-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5 |
| // CHECK17-NEXT: store ptr @.offload_maptypes.14, ptr [[TMP56]], align 8 |
| // CHECK17-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP57]], align 8 |
| // CHECK17-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP58]], align 8 |
| // CHECK17-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8 |
| // CHECK17-NEXT: store i64 10, ptr [[TMP59]], align 8 |
| // CHECK17-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9 |
| // CHECK17-NEXT: store i64 0, ptr [[TMP60]], align 8 |
| // CHECK17-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4 |
| // CHECK17-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4 |
| // CHECK17-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12 |
| // CHECK17-NEXT: store i32 0, ptr [[TMP63]], align 4 |
| // CHECK17-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]]) |
| // CHECK17-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0 |
| // CHECK17-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]] |
| // CHECK17: omp_offload.failed13: |
| // CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i64 [[TMP42]]) #[[ATTR3]] |
| // CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT14]] |
| // CHECK17: omp_offload.cont14: |
| // CHECK17-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[TMP66]], align 8 |
| // CHECK17-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[TMP67]], align 8 |
| // CHECK17-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i64 0, i64 0 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP68]], align 8 |
| // CHECK17-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP71]], align 4 |
| // CHECK17-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1 |
| // CHECK17-NEXT: store i32 1, ptr [[TMP72]], align 4 |
| // CHECK17-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 8 |
| // CHECK17-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3 |
| // CHECK17-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 8 |
| // CHECK17-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4 |
| // CHECK17-NEXT: store ptr @.offload_sizes.15, ptr [[TMP75]], align 8 |
| // CHECK17-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5 |
| // CHECK17-NEXT: store ptr @.offload_maptypes.16, ptr [[TMP76]], align 8 |
| // CHECK17-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP77]], align 8 |
| // CHECK17-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP78]], align 8 |
| // CHECK17-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8 |
| // CHECK17-NEXT: store i64 10, ptr [[TMP79]], align 8 |
| // CHECK17-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9 |
| // CHECK17-NEXT: store i64 0, ptr [[TMP80]], align 8 |
| // CHECK17-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4 |
| // CHECK17-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4 |
| // CHECK17-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12 |
| // CHECK17-NEXT: store i32 0, ptr [[TMP83]], align 4 |
| // CHECK17-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]]) |
| // CHECK17-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0 |
| // CHECK17-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]] |
| // CHECK17: omp_offload.failed20: |
| // CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]] |
| // CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT21]] |
| // CHECK17: omp_offload.cont21: |
| // CHECK17-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4 |
| // CHECK17-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4 |
| // CHECK17-NEXT: [[TMP88:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 8 |
| // CHECK17-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[TMP89]], align 8 |
| // CHECK17-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[TMP90]], align 8 |
| // CHECK17-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 0 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP91]], align 8 |
| // CHECK17-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP88]], ptr [[TMP92]], align 8 |
| // CHECK17-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1 |
| // CHECK17-NEXT: store i64 [[TMP88]], ptr [[TMP93]], align 8 |
| // CHECK17-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i64 0, i64 1 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP94]], align 8 |
| // CHECK17-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0 |
| // CHECK17-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0 |
| // CHECK17-NEXT: store i32 3, ptr [[TMP97]], align 4 |
| // CHECK17-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1 |
| // CHECK17-NEXT: store i32 2, ptr [[TMP98]], align 4 |
| // CHECK17-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2 |
| // CHECK17-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 8 |
| // CHECK17-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3 |
| // CHECK17-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 8 |
| // CHECK17-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4 |
| // CHECK17-NEXT: store ptr @.offload_sizes.17, ptr [[TMP101]], align 8 |
| // CHECK17-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5 |
| // CHECK17-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP102]], align 8 |
| // CHECK17-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP103]], align 8 |
| // CHECK17-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7 |
| // CHECK17-NEXT: store ptr null, ptr [[TMP104]], align 8 |
| // CHECK17-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8 |
| // CHECK17-NEXT: store i64 10, ptr [[TMP105]], align 8 |
| // CHECK17-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9 |
| // CHECK17-NEXT: store i64 0, ptr [[TMP106]], align 8 |
| // CHECK17-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4 |
| // CHECK17-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11 |
| // CHECK17-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4 |
| // CHECK17-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12 |
| // CHECK17-NEXT: store i32 0, ptr [[TMP109]], align 4 |
| // CHECK17-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]]) |
| // CHECK17-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0 |
| // CHECK17-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]] |
| // CHECK17: omp_offload.failed29: |
| // CHECK17-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i64 [[TMP88]]) #[[ATTR3]] |
| // CHECK17-NEXT: br label [[OMP_OFFLOAD_CONT30]] |
| // CHECK17: omp_offload.cont30: |
| // CHECK17-NEXT: ret i32 0 |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112 |
| // CHECK17-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined, ptr [[TMP0]]) |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK17-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP46:![0-9]+]] |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK17-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK17-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK17-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK17-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP46]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP46]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP47:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK17-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK17-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 9 |
| // CHECK17-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP49:![0-9]+]] |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK17-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK17-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK17-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64 |
| // CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]] |
| // CHECK17-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK17-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK17: omp.body.continue: |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK17-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK17-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP49]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP50:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK17-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116 |
| // CHECK17-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined, ptr [[TMP0]]) |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK17-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP52:![0-9]+]] |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK17-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK17-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK17-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK17-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP52]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP52]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP53:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK17-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK17-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 9 |
| // CHECK17-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP55:![0-9]+]] |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK17-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK17-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK17-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64 |
| // CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]] |
| // CHECK17-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK17-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK17: omp.body.continue: |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK17-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK17-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP55]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP56:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK17-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120 |
| // CHECK17-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined, ptr [[TMP0]], i64 [[TMP2]]) |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK17-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP58:![0-9]+]] |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK17-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK17-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK17-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK17-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK17-NEXT: store i32 [[TMP12]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8, !llvm.access.group [[ACC_GRP58]] |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]], i64 [[TMP13]]), !llvm.access.group [[ACC_GRP58]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP14]], [[TMP15]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP58]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP59:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK17-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0 |
| // CHECK17-NEXT: br i1 [[TMP17]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK17-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP5]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP3]]) |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK17: omp.dispatch.cond: |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV2:%.*]] = trunc i64 [[TMP7]] to i32 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP6]], [[CONV2]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV3:%.*]] = trunc i64 [[TMP8]] to i32 |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ [[CONV3]], [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP10]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[CMP4:%.*]] = icmp sle i32 [[TMP11]], [[TMP12]] |
| // CHECK17-NEXT: br i1 [[CMP4]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK17: omp.dispatch.body: |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP61:![0-9]+]] |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK17-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK17-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP15]], 1 |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK17-NEXT: [[TMP16:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK17-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP16]] to i64 |
| // CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]] |
| // CHECK17-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK17-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK17: omp.body.continue: |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK17-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP17]], 1 |
| // CHECK17-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP61]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP62:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK17: omp.dispatch.inc: |
| // CHECK17-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP18]], [[TMP19]] |
| // CHECK17-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: [[ADD8:%.*]] = add nsw i32 [[TMP20]], [[TMP21]] |
| // CHECK17-NEXT: store i32 [[ADD8]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK17: omp.dispatch.end: |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP5]]) |
| // CHECK17-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK17-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124 |
| // CHECK17-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined, ptr [[TMP0]]) |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK17-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP64:![0-9]+]] |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK17-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK17-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK17-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK17-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP64]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP64]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP65:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK17-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK17-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 1073741859, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 1) |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK17: omp.dispatch.cond: |
| // CHECK17-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK17-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK17-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK17: omp.dispatch.body: |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP67:![0-9]+]] |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK17-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP12]] to i64 |
| // CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]] |
| // CHECK17-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK17-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK17: omp.body.continue: |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK17-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK17-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP67]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP68:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK17: omp.dispatch.inc: |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK17: omp.dispatch.end: |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK17-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128 |
| // CHECK17-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8 |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined, ptr [[TMP0]], i64 [[TMP2]]) |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK17-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK17: cond.true: |
| // CHECK17-NEXT: br label [[COND_END:%.*]] |
| // CHECK17: cond.false: |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: br label [[COND_END]] |
| // CHECK17: cond.end: |
| // CHECK17-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK17-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP70:![0-9]+]] |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK17-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK17-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK17-NEXT: [[TMP9:%.*]] = zext i32 [[TMP8]] to i64 |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK17-NEXT: [[TMP11:%.*]] = zext i32 [[TMP10]] to i64 |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK17-NEXT: store i32 [[TMP12]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8, !llvm.access.group [[ACC_GRP70]] |
| // CHECK17-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined.omp_outlined, i64 [[TMP9]], i64 [[TMP11]], ptr [[TMP0]], i64 [[TMP13]]), !llvm.access.group [[ACC_GRP70]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP14]], [[TMP15]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP70]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP71:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK17: omp.loop.exit: |
| // CHECK17-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK17-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP17:%.*]] = icmp ne i32 [[TMP16]], 0 |
| // CHECK17-NEXT: br i1 [[TMP17]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK17-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined.omp_outlined |
| // CHECK17-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[DOTPREVIOUS_LB_:%.*]], i64 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK17-NEXT: entry: |
| // CHECK17-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK17-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8 |
| // CHECK17-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK17-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP1:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV:%.*]] = trunc i64 [[TMP1]] to i32 |
| // CHECK17-NEXT: [[TMP2:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8 |
| // CHECK17-NEXT: [[CONV1:%.*]] = trunc i64 [[TMP2]] to i32 |
| // CHECK17-NEXT: store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[CONV1]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK17-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK17-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK17-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8 |
| // CHECK17-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK17-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP7]], i32 1073741859, i32 [[TMP4]], i32 [[TMP5]], i32 1, i32 [[TMP3]]) |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK17: omp.dispatch.cond: |
| // CHECK17-NEXT: [[TMP8:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP7]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK17-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP8]], 0 |
| // CHECK17-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK17: omp.dispatch.body: |
| // CHECK17-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK17-NEXT: store i32 [[TMP9]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK17: omp.inner.for.cond: |
| // CHECK17-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP73:![0-9]+]] |
| // CHECK17-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK17-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP10]], [[TMP11]] |
| // CHECK17-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK17: omp.inner.for.body: |
| // CHECK17-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK17-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1 |
| // CHECK17-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK17-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK17-NEXT: [[TMP13:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK17-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP13]] to i64 |
| // CHECK17-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]] |
| // CHECK17-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK17-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK17: omp.body.continue: |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK17: omp.inner.for.inc: |
| // CHECK17-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK17-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP14]], 1 |
| // CHECK17-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP73]] |
| // CHECK17-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP74:![0-9]+]] |
| // CHECK17: omp.inner.for.end: |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK17: omp.dispatch.inc: |
| // CHECK17-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK17: omp.dispatch.end: |
| // CHECK17-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK17-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0 |
| // CHECK17-NEXT: br i1 [[TMP16]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK17: .omp.final.then: |
| // CHECK17-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK17-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK17: .omp.final.done: |
| // CHECK17-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@main |
| // CHECK19-SAME: (i32 noundef [[ARGC:%.*]], ptr noundef [[ARGV:%.*]]) #[[ATTR0:[0-9]+]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[RETVAL:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[ARGV_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[N:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[SAVED_STACK:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[__VLA_EXPR0:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [3 x i64], align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK19-NEXT: [[N_CASTED3:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_BASEPTRS4:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_PTRS5:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_MAPPERS6:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_SIZES7:%.*]] = alloca [3 x i64], align 4 |
| // CHECK19-NEXT: [[_TMP8:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_9:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_10:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[KERNEL_ARGS15:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_18:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_CASTED19:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_BASEPTRS20:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_PTRS21:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_MAPPERS22:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_SIZES23:%.*]] = alloca [4 x i64], align 4 |
| // CHECK19-NEXT: [[_TMP24:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_25:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_26:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[KERNEL_ARGS31:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK19-NEXT: [[N_CASTED34:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_BASEPTRS35:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_PTRS36:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_MAPPERS37:%.*]] = alloca [3 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_SIZES38:%.*]] = alloca [3 x i64], align 4 |
| // CHECK19-NEXT: [[_TMP39:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_40:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_41:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[KERNEL_ARGS46:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_49:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_CASTED50:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED51:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_BASEPTRS52:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_PTRS53:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_MAPPERS54:%.*]] = alloca [4 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_SIZES55:%.*]] = alloca [4 x i64], align 4 |
| // CHECK19-NEXT: [[_TMP56:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_57:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_58:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[KERNEL_ARGS63:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK19-NEXT: store i32 0, ptr [[RETVAL]], align 4 |
| // CHECK19-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[ARGV]], ptr [[ARGV_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 100, ptr [[N]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = call ptr @llvm.stacksave.p0() |
| // CHECK19-NEXT: store ptr [[TMP1]], ptr [[SAVED_STACK]], align 4 |
| // CHECK19-NEXT: [[VLA:%.*]] = alloca i32, i32 [[TMP0]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP0]], ptr [[__VLA_EXPR0]], align 4 |
| // CHECK19-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = mul nuw i32 [[TMP0]], 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = sext i32 [[TMP4]] to i64 |
| // CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes, i32 24, i1 false) |
| // CHECK19-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 [[TMP3]], ptr [[TMP6]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 [[TMP3]], ptr [[TMP7]], align 4 |
| // CHECK19-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP8]], align 4 |
| // CHECK19-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP9]], align 4 |
| // CHECK19-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP10]], align 4 |
| // CHECK19-NEXT: [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP11]], align 4 |
| // CHECK19-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP12]], align 4 |
| // CHECK19-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP13]], align 4 |
| // CHECK19-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 2 |
| // CHECK19-NEXT: store i64 [[TMP5]], ptr [[TMP14]], align 4 |
| // CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP15]], align 4 |
| // CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP18:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP19]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP20]], 0 |
| // CHECK19-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK19-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK19-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], 1 |
| // CHECK19-NEXT: [[TMP22:%.*]] = zext i32 [[ADD]] to i64 |
| // CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP23]], align 4 |
| // CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP24]], align 4 |
| // CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[TMP16]], ptr [[TMP25]], align 4 |
| // CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr [[TMP17]], ptr [[TMP26]], align 4 |
| // CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK19-NEXT: store ptr [[TMP18]], ptr [[TMP27]], align 4 |
| // CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK19-NEXT: store ptr @.offload_maptypes, ptr [[TMP28]], align 4 |
| // CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP29]], align 4 |
| // CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP30]], align 4 |
| // CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK19-NEXT: store i64 [[TMP22]], ptr [[TMP31]], align 8 |
| // CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK19-NEXT: store i64 0, ptr [[TMP32]], align 8 |
| // CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4 |
| // CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4 |
| // CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK19-NEXT: store i32 0, ptr [[TMP35]], align 4 |
| // CHECK19-NEXT: [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3:[0-9]+]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK19-NEXT: [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0 |
| // CHECK19-NEXT: br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK19: omp_offload.failed: |
| // CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139(i32 [[TMP3]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3:[0-9]+]] |
| // CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK19: omp_offload.cont: |
| // CHECK19-NEXT: [[TMP38:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP38]], ptr [[N_CASTED3]], align 4 |
| // CHECK19-NEXT: [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4 |
| // CHECK19-NEXT: [[TMP40:%.*]] = mul nuw i32 [[TMP0]], 4 |
| // CHECK19-NEXT: [[TMP41:%.*]] = sext i32 [[TMP40]] to i64 |
| // CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES7]], ptr align 4 @.offload_sizes.1, i32 24, i1 false) |
| // CHECK19-NEXT: [[TMP42:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 [[TMP39]], ptr [[TMP42]], align 4 |
| // CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 [[TMP39]], ptr [[TMP43]], align 4 |
| // CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP44]], align 4 |
| // CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP45]], align 4 |
| // CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP46]], align 4 |
| // CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP47]], align 4 |
| // CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP48]], align 4 |
| // CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP49]], align 4 |
| // CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 2 |
| // CHECK19-NEXT: store i64 [[TMP41]], ptr [[TMP50]], align 4 |
| // CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP51]], align 4 |
| // CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES7]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP55:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP55]], ptr [[DOTCAPTURE_EXPR_9]], align 4 |
| // CHECK19-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4 |
| // CHECK19-NEXT: [[SUB11:%.*]] = sub nsw i32 [[TMP56]], 0 |
| // CHECK19-NEXT: [[DIV12:%.*]] = sdiv i32 [[SUB11]], 1 |
| // CHECK19-NEXT: [[SUB13:%.*]] = sub nsw i32 [[DIV12]], 1 |
| // CHECK19-NEXT: store i32 [[SUB13]], ptr [[DOTCAPTURE_EXPR_10]], align 4 |
| // CHECK19-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_10]], align 4 |
| // CHECK19-NEXT: [[ADD14:%.*]] = add nsw i32 [[TMP57]], 1 |
| // CHECK19-NEXT: [[TMP58:%.*]] = zext i32 [[ADD14]] to i64 |
| // CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP59]], align 4 |
| // CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP60]], align 4 |
| // CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[TMP52]], ptr [[TMP61]], align 4 |
| // CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr [[TMP53]], ptr [[TMP62]], align 4 |
| // CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 4 |
| // CHECK19-NEXT: store ptr [[TMP54]], ptr [[TMP63]], align 4 |
| // CHECK19-NEXT: [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 5 |
| // CHECK19-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP64]], align 4 |
| // CHECK19-NEXT: [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 6 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP65]], align 4 |
| // CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 7 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP66]], align 4 |
| // CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 8 |
| // CHECK19-NEXT: store i64 [[TMP58]], ptr [[TMP67]], align 8 |
| // CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 9 |
| // CHECK19-NEXT: store i64 0, ptr [[TMP68]], align 8 |
| // CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 10 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP69]], align 4 |
| // CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 11 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP70]], align 4 |
| // CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS15]], i32 0, i32 12 |
| // CHECK19-NEXT: store i32 0, ptr [[TMP71]], align 4 |
| // CHECK19-NEXT: [[TMP72:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.region_id, ptr [[KERNEL_ARGS15]]) |
| // CHECK19-NEXT: [[TMP73:%.*]] = icmp ne i32 [[TMP72]], 0 |
| // CHECK19-NEXT: br i1 [[TMP73]], label [[OMP_OFFLOAD_FAILED16:%.*]], label [[OMP_OFFLOAD_CONT17:%.*]] |
| // CHECK19: omp_offload.failed16: |
| // CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143(i32 [[TMP39]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]] |
| // CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT17]] |
| // CHECK19: omp_offload.cont17: |
| // CHECK19-NEXT: [[TMP74:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP74]], ptr [[DOTCAPTURE_EXPR_18]], align 4 |
| // CHECK19-NEXT: [[TMP75:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP75]], ptr [[N_CASTED19]], align 4 |
| // CHECK19-NEXT: [[TMP76:%.*]] = load i32, ptr [[N_CASTED19]], align 4 |
| // CHECK19-NEXT: [[TMP77:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_18]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP77]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP78:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP79:%.*]] = mul nuw i32 [[TMP0]], 4 |
| // CHECK19-NEXT: [[TMP80:%.*]] = sext i32 [[TMP79]] to i64 |
| // CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES23]], ptr align 4 @.offload_sizes.3, i32 32, i1 false) |
| // CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 [[TMP76]], ptr [[TMP81]], align 4 |
| // CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 [[TMP76]], ptr [[TMP82]], align 4 |
| // CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP83]], align 4 |
| // CHECK19-NEXT: [[TMP84:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP84]], align 4 |
| // CHECK19-NEXT: [[TMP85:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP85]], align 4 |
| // CHECK19-NEXT: [[TMP86:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 1 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP86]], align 4 |
| // CHECK19-NEXT: [[TMP87:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP87]], align 4 |
| // CHECK19-NEXT: [[TMP88:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP88]], align 4 |
| // CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 2 |
| // CHECK19-NEXT: store i64 [[TMP80]], ptr [[TMP89]], align 4 |
| // CHECK19-NEXT: [[TMP90:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP90]], align 4 |
| // CHECK19-NEXT: [[TMP91:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 3 |
| // CHECK19-NEXT: store i32 [[TMP78]], ptr [[TMP91]], align 4 |
| // CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 3 |
| // CHECK19-NEXT: store i32 [[TMP78]], ptr [[TMP92]], align 4 |
| // CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS22]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP93]], align 4 |
| // CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS20]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS21]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES23]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP97:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP97]], ptr [[DOTCAPTURE_EXPR_25]], align 4 |
| // CHECK19-NEXT: [[TMP98:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_25]], align 4 |
| // CHECK19-NEXT: [[SUB27:%.*]] = sub nsw i32 [[TMP98]], 0 |
| // CHECK19-NEXT: [[DIV28:%.*]] = sdiv i32 [[SUB27]], 1 |
| // CHECK19-NEXT: [[SUB29:%.*]] = sub nsw i32 [[DIV28]], 1 |
| // CHECK19-NEXT: store i32 [[SUB29]], ptr [[DOTCAPTURE_EXPR_26]], align 4 |
| // CHECK19-NEXT: [[TMP99:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_26]], align 4 |
| // CHECK19-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP99]], 1 |
| // CHECK19-NEXT: [[TMP100:%.*]] = zext i32 [[ADD30]] to i64 |
| // CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP101]], align 4 |
| // CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 4, ptr [[TMP102]], align 4 |
| // CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[TMP94]], ptr [[TMP103]], align 4 |
| // CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr [[TMP95]], ptr [[TMP104]], align 4 |
| // CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 4 |
| // CHECK19-NEXT: store ptr [[TMP96]], ptr [[TMP105]], align 4 |
| // CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 5 |
| // CHECK19-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP106]], align 4 |
| // CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 6 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP107]], align 4 |
| // CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 7 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP108]], align 4 |
| // CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 8 |
| // CHECK19-NEXT: store i64 [[TMP100]], ptr [[TMP109]], align 8 |
| // CHECK19-NEXT: [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 9 |
| // CHECK19-NEXT: store i64 0, ptr [[TMP110]], align 8 |
| // CHECK19-NEXT: [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 10 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4 |
| // CHECK19-NEXT: [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 11 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP112]], align 4 |
| // CHECK19-NEXT: [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS31]], i32 0, i32 12 |
| // CHECK19-NEXT: store i32 0, ptr [[TMP113]], align 4 |
| // CHECK19-NEXT: [[TMP114:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.region_id, ptr [[KERNEL_ARGS31]]) |
| // CHECK19-NEXT: [[TMP115:%.*]] = icmp ne i32 [[TMP114]], 0 |
| // CHECK19-NEXT: br i1 [[TMP115]], label [[OMP_OFFLOAD_FAILED32:%.*]], label [[OMP_OFFLOAD_CONT33:%.*]] |
| // CHECK19: omp_offload.failed32: |
| // CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147(i32 [[TMP76]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP78]]) #[[ATTR3]] |
| // CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT33]] |
| // CHECK19: omp_offload.cont33: |
| // CHECK19-NEXT: [[TMP116:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP116]], ptr [[N_CASTED34]], align 4 |
| // CHECK19-NEXT: [[TMP117:%.*]] = load i32, ptr [[N_CASTED34]], align 4 |
| // CHECK19-NEXT: [[TMP118:%.*]] = mul nuw i32 [[TMP0]], 4 |
| // CHECK19-NEXT: [[TMP119:%.*]] = sext i32 [[TMP118]] to i64 |
| // CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES38]], ptr align 4 @.offload_sizes.5, i32 24, i1 false) |
| // CHECK19-NEXT: [[TMP120:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 [[TMP117]], ptr [[TMP120]], align 4 |
| // CHECK19-NEXT: [[TMP121:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 [[TMP117]], ptr [[TMP121]], align 4 |
| // CHECK19-NEXT: [[TMP122:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP122]], align 4 |
| // CHECK19-NEXT: [[TMP123:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP123]], align 4 |
| // CHECK19-NEXT: [[TMP124:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP124]], align 4 |
| // CHECK19-NEXT: [[TMP125:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 1 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP125]], align 4 |
| // CHECK19-NEXT: [[TMP126:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP126]], align 4 |
| // CHECK19-NEXT: [[TMP127:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP127]], align 4 |
| // CHECK19-NEXT: [[TMP128:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 2 |
| // CHECK19-NEXT: store i64 [[TMP119]], ptr [[TMP128]], align 4 |
| // CHECK19-NEXT: [[TMP129:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS37]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP129]], align 4 |
| // CHECK19-NEXT: [[TMP130:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS35]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP131:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS36]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP132:%.*]] = getelementptr inbounds [3 x i64], ptr [[DOTOFFLOAD_SIZES38]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP133:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP133]], ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK19-NEXT: [[TMP134:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK19-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP134]], 0 |
| // CHECK19-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1 |
| // CHECK19-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1 |
| // CHECK19-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK19-NEXT: [[TMP135:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK19-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP135]], 1 |
| // CHECK19-NEXT: [[TMP136:%.*]] = zext i32 [[ADD45]] to i64 |
| // CHECK19-NEXT: [[TMP137:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP137]], align 4 |
| // CHECK19-NEXT: [[TMP138:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP138]], align 4 |
| // CHECK19-NEXT: [[TMP139:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[TMP130]], ptr [[TMP139]], align 4 |
| // CHECK19-NEXT: [[TMP140:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr [[TMP131]], ptr [[TMP140]], align 4 |
| // CHECK19-NEXT: [[TMP141:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 4 |
| // CHECK19-NEXT: store ptr [[TMP132]], ptr [[TMP141]], align 4 |
| // CHECK19-NEXT: [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 5 |
| // CHECK19-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP142]], align 4 |
| // CHECK19-NEXT: [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 6 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP143]], align 4 |
| // CHECK19-NEXT: [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 7 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP144]], align 4 |
| // CHECK19-NEXT: [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 8 |
| // CHECK19-NEXT: store i64 [[TMP136]], ptr [[TMP145]], align 8 |
| // CHECK19-NEXT: [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 9 |
| // CHECK19-NEXT: store i64 0, ptr [[TMP146]], align 8 |
| // CHECK19-NEXT: [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 10 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP147]], align 4 |
| // CHECK19-NEXT: [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 11 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP148]], align 4 |
| // CHECK19-NEXT: [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS46]], i32 0, i32 12 |
| // CHECK19-NEXT: store i32 0, ptr [[TMP149]], align 4 |
| // CHECK19-NEXT: [[TMP150:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.region_id, ptr [[KERNEL_ARGS46]]) |
| // CHECK19-NEXT: [[TMP151:%.*]] = icmp ne i32 [[TMP150]], 0 |
| // CHECK19-NEXT: br i1 [[TMP151]], label [[OMP_OFFLOAD_FAILED47:%.*]], label [[OMP_OFFLOAD_CONT48:%.*]] |
| // CHECK19: omp_offload.failed47: |
| // CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151(i32 [[TMP117]], i32 [[TMP0]], ptr [[VLA]]) #[[ATTR3]] |
| // CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT48]] |
| // CHECK19: omp_offload.cont48: |
| // CHECK19-NEXT: [[TMP152:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP152]], ptr [[DOTCAPTURE_EXPR_49]], align 4 |
| // CHECK19-NEXT: [[TMP153:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP153]], ptr [[N_CASTED50]], align 4 |
| // CHECK19-NEXT: [[TMP154:%.*]] = load i32, ptr [[N_CASTED50]], align 4 |
| // CHECK19-NEXT: [[TMP155:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_49]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP155]], ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4 |
| // CHECK19-NEXT: [[TMP156:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED51]], align 4 |
| // CHECK19-NEXT: [[TMP157:%.*]] = mul nuw i32 [[TMP0]], 4 |
| // CHECK19-NEXT: [[TMP158:%.*]] = sext i32 [[TMP157]] to i64 |
| // CHECK19-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES55]], ptr align 4 @.offload_sizes.7, i32 32, i1 false) |
| // CHECK19-NEXT: [[TMP159:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 [[TMP154]], ptr [[TMP159]], align 4 |
| // CHECK19-NEXT: [[TMP160:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 [[TMP154]], ptr [[TMP160]], align 4 |
| // CHECK19-NEXT: [[TMP161:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP161]], align 4 |
| // CHECK19-NEXT: [[TMP162:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP162]], align 4 |
| // CHECK19-NEXT: [[TMP163:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP0]], ptr [[TMP163]], align 4 |
| // CHECK19-NEXT: [[TMP164:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 1 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP164]], align 4 |
| // CHECK19-NEXT: [[TMP165:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP165]], align 4 |
| // CHECK19-NEXT: [[TMP166:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[VLA]], ptr [[TMP166]], align 4 |
| // CHECK19-NEXT: [[TMP167:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 2 |
| // CHECK19-NEXT: store i64 [[TMP158]], ptr [[TMP167]], align 4 |
| // CHECK19-NEXT: [[TMP168:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP168]], align 4 |
| // CHECK19-NEXT: [[TMP169:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 3 |
| // CHECK19-NEXT: store i32 [[TMP156]], ptr [[TMP169]], align 4 |
| // CHECK19-NEXT: [[TMP170:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 3 |
| // CHECK19-NEXT: store i32 [[TMP156]], ptr [[TMP170]], align 4 |
| // CHECK19-NEXT: [[TMP171:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS54]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP171]], align 4 |
| // CHECK19-NEXT: [[TMP172:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS52]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP173:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS53]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP174:%.*]] = getelementptr inbounds [4 x i64], ptr [[DOTOFFLOAD_SIZES55]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP175:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP175]], ptr [[DOTCAPTURE_EXPR_57]], align 4 |
| // CHECK19-NEXT: [[TMP176:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_57]], align 4 |
| // CHECK19-NEXT: [[SUB59:%.*]] = sub nsw i32 [[TMP176]], 0 |
| // CHECK19-NEXT: [[DIV60:%.*]] = sdiv i32 [[SUB59]], 1 |
| // CHECK19-NEXT: [[SUB61:%.*]] = sub nsw i32 [[DIV60]], 1 |
| // CHECK19-NEXT: store i32 [[SUB61]], ptr [[DOTCAPTURE_EXPR_58]], align 4 |
| // CHECK19-NEXT: [[TMP177:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_58]], align 4 |
| // CHECK19-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP177]], 1 |
| // CHECK19-NEXT: [[TMP178:%.*]] = zext i32 [[ADD62]] to i64 |
| // CHECK19-NEXT: [[TMP179:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP179]], align 4 |
| // CHECK19-NEXT: [[TMP180:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 4, ptr [[TMP180]], align 4 |
| // CHECK19-NEXT: [[TMP181:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[TMP172]], ptr [[TMP181]], align 4 |
| // CHECK19-NEXT: [[TMP182:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr [[TMP173]], ptr [[TMP182]], align 4 |
| // CHECK19-NEXT: [[TMP183:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 4 |
| // CHECK19-NEXT: store ptr [[TMP174]], ptr [[TMP183]], align 4 |
| // CHECK19-NEXT: [[TMP184:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 5 |
| // CHECK19-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP184]], align 4 |
| // CHECK19-NEXT: [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 6 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP185]], align 4 |
| // CHECK19-NEXT: [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 7 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP186]], align 4 |
| // CHECK19-NEXT: [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 8 |
| // CHECK19-NEXT: store i64 [[TMP178]], ptr [[TMP187]], align 8 |
| // CHECK19-NEXT: [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 9 |
| // CHECK19-NEXT: store i64 0, ptr [[TMP188]], align 8 |
| // CHECK19-NEXT: [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 10 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP189]], align 4 |
| // CHECK19-NEXT: [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 11 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP190]], align 4 |
| // CHECK19-NEXT: [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS63]], i32 0, i32 12 |
| // CHECK19-NEXT: store i32 0, ptr [[TMP191]], align 4 |
| // CHECK19-NEXT: [[TMP192:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.region_id, ptr [[KERNEL_ARGS63]]) |
| // CHECK19-NEXT: [[TMP193:%.*]] = icmp ne i32 [[TMP192]], 0 |
| // CHECK19-NEXT: br i1 [[TMP193]], label [[OMP_OFFLOAD_FAILED64:%.*]], label [[OMP_OFFLOAD_CONT65:%.*]] |
| // CHECK19: omp_offload.failed64: |
| // CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155(i32 [[TMP154]], i32 [[TMP0]], ptr [[VLA]], i32 [[TMP156]]) #[[ATTR3]] |
| // CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT65]] |
| // CHECK19: omp_offload.cont65: |
| // CHECK19-NEXT: [[TMP194:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4 |
| // CHECK19-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP194]]) |
| // CHECK19-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4 |
| // CHECK19-NEXT: [[TMP195:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4 |
| // CHECK19-NEXT: call void @llvm.stackrestore.p0(ptr [[TMP195]]) |
| // CHECK19-NEXT: [[TMP196:%.*]] = load i32, ptr [[RETVAL]], align 4 |
| // CHECK19-NEXT: ret i32 [[TMP196]] |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139 |
| // CHECK19-SAME: (i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2:[0-9]+]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined, i32 [[TMP3]], i32 [[TMP0]], ptr [[TMP1]]) |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK19-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK19-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK19-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK19: omp.precond.then: |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK19-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP14:![0-9]+]] |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK19-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK19-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK19-NEXT: store i32 [[TMP17]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK19-NEXT: [[TMP18:%.*]] = load i32, ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined.omp_outlined, i32 [[TMP15]], i32 [[TMP16]], i32 [[TMP18]], i32 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP14]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP14]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]]) |
| // CHECK19-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0 |
| // CHECK19-NEXT: br i1 [[TMP24]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP25]], 0 |
| // CHECK19-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK19-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK19: omp.precond.end: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l139.omp_outlined.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK19-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK19-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK19-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK19: omp.precond.then: |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK19-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18:![0-9]+]] |
| // CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK19-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK19-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK19-NEXT: [[TMP18:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 [[TMP18]] |
| // CHECK19-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK19-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK19: omp.body.continue: |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK19-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK19-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK19-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK19-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK19-NEXT: [[MUL9:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK19-NEXT: [[ADD10:%.*]] = add nsw i32 0, [[MUL9]] |
| // CHECK19-NEXT: store i32 [[ADD10]], ptr [[I3]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK19: omp.precond.end: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143 |
| // CHECK19-SAME: (i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined, i32 [[TMP3]], i32 [[TMP0]], ptr [[TMP1]]) |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK19-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK19-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK19-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK19: omp.precond.then: |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK19-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23:![0-9]+]] |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK19-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK19-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK19-NEXT: store i32 [[TMP17]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK19-NEXT: [[TMP18:%.*]] = load i32, ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined.omp_outlined, i32 [[TMP15]], i32 [[TMP16]], i32 [[TMP18]], i32 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP23]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP24:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]]) |
| // CHECK19-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0 |
| // CHECK19-NEXT: br i1 [[TMP24]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP25]], 0 |
| // CHECK19-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK19-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK19: omp.precond.end: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l143.omp_outlined.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK19-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK19-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK19-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK19: omp.precond.then: |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK19-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26:![0-9]+]] |
| // CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK19-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK19-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK19-NEXT: [[TMP18:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 [[TMP18]] |
| // CHECK19-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK19-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK19: omp.body.continue: |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK19-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK19-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP27:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK19-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK19-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK19-NEXT: [[MUL9:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK19-NEXT: [[ADD10:%.*]] = add nsw i32 0, [[MUL9]] |
| // CHECK19-NEXT: store i32 [[ADD10]], ptr [[I3]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK19: omp.precond.end: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147 |
| // CHECK19-SAME: (i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined, i32 [[TMP3]], i32 [[TMP0]], ptr [[TMP1]], i32 [[TMP5]]) |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK19-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK19-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK19-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK19: omp.precond.then: |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[TMP7]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP8]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP6]]) |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]] |
| // CHECK19-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ [[TMP11]], [[COND_TRUE]] ], [ [[TMP12]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP13]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29:![0-9]+]] |
| // CHECK19-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP15]], 1 |
| // CHECK19-NEXT: [[CMP6:%.*]] = icmp slt i32 [[TMP14]], [[ADD]] |
| // CHECK19-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP18:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: store i32 [[TMP18]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: store i32 [[TMP20]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined.omp_outlined, i32 [[TMP16]], i32 [[TMP17]], i32 [[TMP19]], i32 [[TMP0]], ptr [[TMP1]], i32 [[TMP21]]), !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP22]], [[TMP23]] |
| // CHECK19-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[ADD8:%.*]] = add nsw i32 [[TMP24]], [[TMP25]] |
| // CHECK19-NEXT: store i32 [[ADD8]], ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[ADD9:%.*]] = add nsw i32 [[TMP26]], [[TMP27]] |
| // CHECK19-NEXT: store i32 [[ADD9]], ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[CMP10:%.*]] = icmp sgt i32 [[TMP28]], [[TMP29]] |
| // CHECK19-NEXT: br i1 [[CMP10]], label [[COND_TRUE11:%.*]], label [[COND_FALSE12:%.*]] |
| // CHECK19: cond.true11: |
| // CHECK19-NEXT: [[TMP30:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: br label [[COND_END13:%.*]] |
| // CHECK19: cond.false12: |
| // CHECK19-NEXT: [[TMP31:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: br label [[COND_END13]] |
| // CHECK19: cond.end13: |
| // CHECK19-NEXT: [[COND14:%.*]] = phi i32 [ [[TMP30]], [[COND_TRUE11]] ], [ [[TMP31]], [[COND_FALSE12]] ] |
| // CHECK19-NEXT: store i32 [[COND14]], ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: [[TMP32:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: store i32 [[TMP32]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP30:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP34]]) |
| // CHECK19-NEXT: [[TMP35:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0 |
| // CHECK19-NEXT: br i1 [[TMP36]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: [[TMP37:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[SUB15:%.*]] = sub nsw i32 [[TMP37]], 0 |
| // CHECK19-NEXT: [[DIV16:%.*]] = sdiv i32 [[SUB15]], 1 |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV16]], 1 |
| // CHECK19-NEXT: [[ADD17:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD17]], ptr [[I4]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK19: omp.precond.end: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l147.omp_outlined.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK19-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK19-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK19-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK19: omp.precond.then: |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP9]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]] |
| // CHECK19-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32:![0-9]+]] |
| // CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK19-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]] |
| // CHECK19-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP17]], 1 |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK19-NEXT: [[TMP18:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 [[TMP18]] |
| // CHECK19-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK19-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK19: omp.body.continue: |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK19-NEXT: [[ADD7:%.*]] = add nsw i32 [[TMP19]], 1 |
| // CHECK19-NEXT: store i32 [[ADD7]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP32]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP33:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP21]]) |
| // CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK19-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[SUB8:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK19-NEXT: [[DIV9:%.*]] = sdiv i32 [[SUB8]], 1 |
| // CHECK19-NEXT: [[MUL10:%.*]] = mul nsw i32 [[DIV9]], 1 |
| // CHECK19-NEXT: [[ADD11:%.*]] = add nsw i32 0, [[MUL10]] |
| // CHECK19-NEXT: store i32 [[ADD11]], ptr [[I4]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK19: omp.precond.end: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151 |
| // CHECK19-SAME: (i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined, i32 [[TMP3]], i32 [[TMP0]], ptr [[TMP1]]) |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK19-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK19-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK19-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK19: omp.precond.then: |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK19-NEXT: br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35:![0-9]+]] |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK19-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK19-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK19-NEXT: store i32 [[TMP17]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK19-NEXT: [[TMP18:%.*]] = load i32, ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined.omp_outlined, i32 [[TMP15]], i32 [[TMP16]], i32 [[TMP18]], i32 [[TMP0]], ptr [[TMP1]]), !llvm.access.group [[ACC_GRP35]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP35]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP36:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]]) |
| // CHECK19-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP24:%.*]] = icmp ne i32 [[TMP23]], 0 |
| // CHECK19-NEXT: br i1 [[TMP24]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP25]], 0 |
| // CHECK19-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK19-NEXT: [[ADD8:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD8]], ptr [[I3]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK19: omp.precond.end: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l151.omp_outlined.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK19-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK19-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK19-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK19: omp.precond.then: |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP11]], i32 1073741859, i32 [[TMP8]], i32 [[TMP9]], i32 1, i32 1) |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK19: omp.dispatch.cond: |
| // CHECK19-NEXT: [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4 |
| // CHECK19-NEXT: [[TMP14:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP13]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK19-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK19-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK19: omp.dispatch.body: |
| // CHECK19-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP15]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38:![0-9]+]] |
| // CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK19-NEXT: [[CMP4:%.*]] = icmp sle i32 [[TMP16]], [[TMP17]] |
| // CHECK19-NEXT: br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP18]], 1 |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 [[TMP19]] |
| // CHECK19-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK19-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK19: omp.body.continue: |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK19-NEXT: [[ADD5:%.*]] = add nsw i32 [[TMP20]], 1 |
| // CHECK19-NEXT: store i32 [[ADD5]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP38]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP39:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK19: omp.dispatch.inc: |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK19: omp.dispatch.end: |
| // CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0 |
| // CHECK19-NEXT: br i1 [[TMP22]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP23]], 0 |
| // CHECK19-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK19-NEXT: [[MUL8:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK19-NEXT: [[ADD9:%.*]] = add nsw i32 0, [[MUL8]] |
| // CHECK19-NEXT: store i32 [[ADD9]], ptr [[I3]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK19: omp.precond.end: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155 |
| // CHECK19-SAME: (i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined, i32 [[TMP3]], i32 [[TMP0]], ptr [[TMP1]], i32 [[TMP5]]) |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK19-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK19-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK19-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK19: omp.precond.then: |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP7]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: [[CMP5:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]] |
| // CHECK19-NEXT: br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ [[TMP10]], [[COND_TRUE]] ], [ [[TMP11]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP12]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP41:![0-9]+]] |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: [[CMP6:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK19-NEXT: br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[N_ADDR]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: store i32 [[TMP17]], ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: [[TMP18:%.*]] = load i32, ptr [[N_CASTED]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: store i32 [[TMP19]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined.omp_outlined, i32 [[TMP15]], i32 [[TMP16]], i32 [[TMP18]], i32 [[TMP0]], ptr [[TMP1]], i32 [[TMP20]]), !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP41]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP42:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]]) |
| // CHECK19-NEXT: [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP26:%.*]] = icmp ne i32 [[TMP25]], 0 |
| // CHECK19-NEXT: br i1 [[TMP26]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP27]], 0 |
| // CHECK19-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK19-NEXT: [[ADD9:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD9]], ptr [[I4]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK19: omp.precond.end: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l155.omp_outlined.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], i32 noundef [[N:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I4:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load i32, ptr [[VLA_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK19-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK19-NEXT: [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK19-NEXT: store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP4]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]] |
| // CHECK19: omp.precond.then: |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP11:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP12]], i32 1073741859, i32 [[TMP9]], i32 [[TMP10]], i32 1, i32 [[TMP8]]) |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK19: omp.dispatch.cond: |
| // CHECK19-NEXT: [[TMP13:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[TMP13]], align 4 |
| // CHECK19-NEXT: [[TMP15:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP14]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK19-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP15]], 0 |
| // CHECK19-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK19: omp.dispatch.body: |
| // CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP44:![0-9]+]] |
| // CHECK19-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK19-NEXT: [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]] |
| // CHECK19-NEXT: br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1 |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 [[TMP20]] |
| // CHECK19-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK19-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK19: omp.body.continue: |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK19-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP21]], 1 |
| // CHECK19-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP44]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP45:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK19: omp.dispatch.inc: |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK19: omp.dispatch.end: |
| // CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK19-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK19-NEXT: [[SUB7:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK19-NEXT: [[DIV8:%.*]] = sdiv i32 [[SUB7]], 1 |
| // CHECK19-NEXT: [[MUL9:%.*]] = mul nsw i32 [[DIV8]], 1 |
| // CHECK19-NEXT: [[ADD10:%.*]] = add nsw i32 0, [[MUL9]] |
| // CHECK19-NEXT: store i32 [[ADD10]], ptr [[I4]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: br label [[OMP_PRECOND_END]] |
| // CHECK19: omp.precond.end: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@_Z5tmainIiLi10EEiT_ |
| // CHECK19-SAME: (i32 noundef [[ARGC:%.*]]) #[[ATTR5:[0-9]+]] comdat { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A:%.*]] = alloca [10 x i32], align 4 |
| // CHECK19-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8 |
| // CHECK19-NEXT: [[DOTOFFLOAD_BASEPTRS1:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_PTRS2:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_MAPPERS3:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK19-NEXT: [[_TMP4:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[KERNEL_ARGS5:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_BASEPTRS8:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_PTRS9:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_MAPPERS10:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK19-NEXT: [[_TMP11:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[KERNEL_ARGS12:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK19-NEXT: [[DOTOFFLOAD_BASEPTRS15:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_PTRS16:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_MAPPERS17:%.*]] = alloca [1 x ptr], align 4 |
| // CHECK19-NEXT: [[_TMP18:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[KERNEL_ARGS19:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR_22:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED23:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_BASEPTRS24:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_PTRS25:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK19-NEXT: [[DOTOFFLOAD_MAPPERS26:%.*]] = alloca [2 x ptr], align 4 |
| // CHECK19-NEXT: [[_TMP27:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8 |
| // CHECK19-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[TMP0]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[TMP1]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP2]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP5]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 1, ptr [[TMP6]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[TMP3]], ptr [[TMP7]], align 4 |
| // CHECK19-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr [[TMP4]], ptr [[TMP8]], align 4 |
| // CHECK19-NEXT: [[TMP9:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4 |
| // CHECK19-NEXT: store ptr @.offload_sizes.9, ptr [[TMP9]], align 4 |
| // CHECK19-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5 |
| // CHECK19-NEXT: store ptr @.offload_maptypes.10, ptr [[TMP10]], align 4 |
| // CHECK19-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP11]], align 4 |
| // CHECK19-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP12]], align 4 |
| // CHECK19-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8 |
| // CHECK19-NEXT: store i64 10, ptr [[TMP13]], align 8 |
| // CHECK19-NEXT: [[TMP14:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9 |
| // CHECK19-NEXT: store i64 0, ptr [[TMP14]], align 8 |
| // CHECK19-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP15]], align 4 |
| // CHECK19-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP16]], align 4 |
| // CHECK19-NEXT: [[TMP17:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12 |
| // CHECK19-NEXT: store i32 0, ptr [[TMP17]], align 4 |
| // CHECK19-NEXT: [[TMP18:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.region_id, ptr [[KERNEL_ARGS]]) |
| // CHECK19-NEXT: [[TMP19:%.*]] = icmp ne i32 [[TMP18]], 0 |
| // CHECK19-NEXT: br i1 [[TMP19]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]] |
| // CHECK19: omp_offload.failed: |
| // CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112(ptr [[A]]) #[[ATTR3]] |
| // CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT]] |
| // CHECK19: omp_offload.cont: |
| // CHECK19-NEXT: [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[TMP20]], align 4 |
| // CHECK19-NEXT: [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[TMP21]], align 4 |
| // CHECK19-NEXT: [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP22]], align 4 |
| // CHECK19-NEXT: [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP25]], align 4 |
| // CHECK19-NEXT: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 1, ptr [[TMP26]], align 4 |
| // CHECK19-NEXT: [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[TMP23]], ptr [[TMP27]], align 4 |
| // CHECK19-NEXT: [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr [[TMP24]], ptr [[TMP28]], align 4 |
| // CHECK19-NEXT: [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 4 |
| // CHECK19-NEXT: store ptr @.offload_sizes.11, ptr [[TMP29]], align 4 |
| // CHECK19-NEXT: [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 5 |
| // CHECK19-NEXT: store ptr @.offload_maptypes.12, ptr [[TMP30]], align 4 |
| // CHECK19-NEXT: [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 6 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP31]], align 4 |
| // CHECK19-NEXT: [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 7 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP32]], align 4 |
| // CHECK19-NEXT: [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 8 |
| // CHECK19-NEXT: store i64 10, ptr [[TMP33]], align 8 |
| // CHECK19-NEXT: [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 9 |
| // CHECK19-NEXT: store i64 0, ptr [[TMP34]], align 8 |
| // CHECK19-NEXT: [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 10 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP35]], align 4 |
| // CHECK19-NEXT: [[TMP36:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 11 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP36]], align 4 |
| // CHECK19-NEXT: [[TMP37:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS5]], i32 0, i32 12 |
| // CHECK19-NEXT: store i32 0, ptr [[TMP37]], align 4 |
| // CHECK19-NEXT: [[TMP38:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.region_id, ptr [[KERNEL_ARGS5]]) |
| // CHECK19-NEXT: [[TMP39:%.*]] = icmp ne i32 [[TMP38]], 0 |
| // CHECK19-NEXT: br i1 [[TMP39]], label [[OMP_OFFLOAD_FAILED6:%.*]], label [[OMP_OFFLOAD_CONT7:%.*]] |
| // CHECK19: omp_offload.failed6: |
| // CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116(ptr [[A]]) #[[ATTR3]] |
| // CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT7]] |
| // CHECK19: omp_offload.cont7: |
| // CHECK19-NEXT: [[TMP40:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP40]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP41]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP42:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[TMP43]], align 4 |
| // CHECK19-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[TMP44]], align 4 |
| // CHECK19-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP45]], align 4 |
| // CHECK19-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP42]], ptr [[TMP46]], align 4 |
| // CHECK19-NEXT: [[TMP47:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP42]], ptr [[TMP47]], align 4 |
| // CHECK19-NEXT: [[TMP48:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS10]], i32 0, i32 1 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP48]], align 4 |
| // CHECK19-NEXT: [[TMP49:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS8]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP50:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS9]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP51:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP51]], align 4 |
| // CHECK19-NEXT: [[TMP52:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 2, ptr [[TMP52]], align 4 |
| // CHECK19-NEXT: [[TMP53:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[TMP49]], ptr [[TMP53]], align 4 |
| // CHECK19-NEXT: [[TMP54:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr [[TMP50]], ptr [[TMP54]], align 4 |
| // CHECK19-NEXT: [[TMP55:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 4 |
| // CHECK19-NEXT: store ptr @.offload_sizes.13, ptr [[TMP55]], align 4 |
| // CHECK19-NEXT: [[TMP56:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 5 |
| // CHECK19-NEXT: store ptr @.offload_maptypes.14, ptr [[TMP56]], align 4 |
| // CHECK19-NEXT: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 6 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP57]], align 4 |
| // CHECK19-NEXT: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 7 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP58]], align 4 |
| // CHECK19-NEXT: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 8 |
| // CHECK19-NEXT: store i64 10, ptr [[TMP59]], align 8 |
| // CHECK19-NEXT: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 9 |
| // CHECK19-NEXT: store i64 0, ptr [[TMP60]], align 8 |
| // CHECK19-NEXT: [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 10 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP61]], align 4 |
| // CHECK19-NEXT: [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 11 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP62]], align 4 |
| // CHECK19-NEXT: [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS12]], i32 0, i32 12 |
| // CHECK19-NEXT: store i32 0, ptr [[TMP63]], align 4 |
| // CHECK19-NEXT: [[TMP64:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.region_id, ptr [[KERNEL_ARGS12]]) |
| // CHECK19-NEXT: [[TMP65:%.*]] = icmp ne i32 [[TMP64]], 0 |
| // CHECK19-NEXT: br i1 [[TMP65]], label [[OMP_OFFLOAD_FAILED13:%.*]], label [[OMP_OFFLOAD_CONT14:%.*]] |
| // CHECK19: omp_offload.failed13: |
| // CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120(ptr [[A]], i32 [[TMP42]]) #[[ATTR3]] |
| // CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT14]] |
| // CHECK19: omp_offload.cont14: |
| // CHECK19-NEXT: [[TMP66:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[TMP66]], align 4 |
| // CHECK19-NEXT: [[TMP67:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[TMP67]], align 4 |
| // CHECK19-NEXT: [[TMP68:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS17]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP68]], align 4 |
| // CHECK19-NEXT: [[TMP69:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS15]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP70:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS16]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP71]], align 4 |
| // CHECK19-NEXT: [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 1, ptr [[TMP72]], align 4 |
| // CHECK19-NEXT: [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[TMP69]], ptr [[TMP73]], align 4 |
| // CHECK19-NEXT: [[TMP74:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr [[TMP70]], ptr [[TMP74]], align 4 |
| // CHECK19-NEXT: [[TMP75:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 4 |
| // CHECK19-NEXT: store ptr @.offload_sizes.15, ptr [[TMP75]], align 4 |
| // CHECK19-NEXT: [[TMP76:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 5 |
| // CHECK19-NEXT: store ptr @.offload_maptypes.16, ptr [[TMP76]], align 4 |
| // CHECK19-NEXT: [[TMP77:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 6 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP77]], align 4 |
| // CHECK19-NEXT: [[TMP78:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 7 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP78]], align 4 |
| // CHECK19-NEXT: [[TMP79:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 8 |
| // CHECK19-NEXT: store i64 10, ptr [[TMP79]], align 8 |
| // CHECK19-NEXT: [[TMP80:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 9 |
| // CHECK19-NEXT: store i64 0, ptr [[TMP80]], align 8 |
| // CHECK19-NEXT: [[TMP81:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 10 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP81]], align 4 |
| // CHECK19-NEXT: [[TMP82:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 11 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP82]], align 4 |
| // CHECK19-NEXT: [[TMP83:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS19]], i32 0, i32 12 |
| // CHECK19-NEXT: store i32 0, ptr [[TMP83]], align 4 |
| // CHECK19-NEXT: [[TMP84:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.region_id, ptr [[KERNEL_ARGS19]]) |
| // CHECK19-NEXT: [[TMP85:%.*]] = icmp ne i32 [[TMP84]], 0 |
| // CHECK19-NEXT: br i1 [[TMP85]], label [[OMP_OFFLOAD_FAILED20:%.*]], label [[OMP_OFFLOAD_CONT21:%.*]] |
| // CHECK19: omp_offload.failed20: |
| // CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124(ptr [[A]]) #[[ATTR3]] |
| // CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT21]] |
| // CHECK19: omp_offload.cont21: |
| // CHECK19-NEXT: [[TMP86:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP86]], ptr [[DOTCAPTURE_EXPR_22]], align 4 |
| // CHECK19-NEXT: [[TMP87:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP87]], ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4 |
| // CHECK19-NEXT: [[TMP88:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED23]], align 4 |
| // CHECK19-NEXT: [[TMP89:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[TMP89]], align 4 |
| // CHECK19-NEXT: [[TMP90:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[TMP90]], align 4 |
| // CHECK19-NEXT: [[TMP91:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 0 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP91]], align 4 |
| // CHECK19-NEXT: [[TMP92:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP88]], ptr [[TMP92]], align 4 |
| // CHECK19-NEXT: [[TMP93:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 [[TMP88]], ptr [[TMP93]], align 4 |
| // CHECK19-NEXT: [[TMP94:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS26]], i32 0, i32 1 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP94]], align 4 |
| // CHECK19-NEXT: [[TMP95:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS24]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP96:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS25]], i32 0, i32 0 |
| // CHECK19-NEXT: [[TMP97:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0 |
| // CHECK19-NEXT: store i32 3, ptr [[TMP97]], align 4 |
| // CHECK19-NEXT: [[TMP98:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1 |
| // CHECK19-NEXT: store i32 2, ptr [[TMP98]], align 4 |
| // CHECK19-NEXT: [[TMP99:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2 |
| // CHECK19-NEXT: store ptr [[TMP95]], ptr [[TMP99]], align 4 |
| // CHECK19-NEXT: [[TMP100:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3 |
| // CHECK19-NEXT: store ptr [[TMP96]], ptr [[TMP100]], align 4 |
| // CHECK19-NEXT: [[TMP101:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4 |
| // CHECK19-NEXT: store ptr @.offload_sizes.17, ptr [[TMP101]], align 4 |
| // CHECK19-NEXT: [[TMP102:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5 |
| // CHECK19-NEXT: store ptr @.offload_maptypes.18, ptr [[TMP102]], align 4 |
| // CHECK19-NEXT: [[TMP103:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP103]], align 4 |
| // CHECK19-NEXT: [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7 |
| // CHECK19-NEXT: store ptr null, ptr [[TMP104]], align 4 |
| // CHECK19-NEXT: [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8 |
| // CHECK19-NEXT: store i64 10, ptr [[TMP105]], align 8 |
| // CHECK19-NEXT: [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9 |
| // CHECK19-NEXT: store i64 0, ptr [[TMP106]], align 8 |
| // CHECK19-NEXT: [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP107]], align 4 |
| // CHECK19-NEXT: [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11 |
| // CHECK19-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP108]], align 4 |
| // CHECK19-NEXT: [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12 |
| // CHECK19-NEXT: store i32 0, ptr [[TMP109]], align 4 |
| // CHECK19-NEXT: [[TMP110:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.region_id, ptr [[KERNEL_ARGS28]]) |
| // CHECK19-NEXT: [[TMP111:%.*]] = icmp ne i32 [[TMP110]], 0 |
| // CHECK19-NEXT: br i1 [[TMP111]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]] |
| // CHECK19: omp_offload.failed29: |
| // CHECK19-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128(ptr [[A]], i32 [[TMP88]]) #[[ATTR3]] |
| // CHECK19-NEXT: br label [[OMP_OFFLOAD_CONT30]] |
| // CHECK19: omp_offload.cont30: |
| // CHECK19-NEXT: ret i32 0 |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112 |
| // CHECK19-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined, ptr [[TMP0]]) |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK19-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP47:![0-9]+]] |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK19-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK19-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP47]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP47]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP48:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK19-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l112.omp_outlined.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 9 |
| // CHECK19-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP50:![0-9]+]] |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK19-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK19-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP11]] |
| // CHECK19-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK19-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK19: omp.body.continue: |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK19-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK19-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP50]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP51:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK19-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116 |
| // CHECK19-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined, ptr [[TMP0]]) |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK19-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP53:![0-9]+]] |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK19-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK19-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP53]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP53]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP54:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK19-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l116.omp_outlined.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP4]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP5]], 9 |
| // CHECK19-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP6]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP56:![0-9]+]] |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK19-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK19-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP11]] |
| // CHECK19-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK19-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK19: omp.body.continue: |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK19-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK19-NEXT: store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP56]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP57:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP4]]) |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP14:%.*]] = icmp ne i32 [[TMP13]], 0 |
| // CHECK19-NEXT: br i1 [[TMP14]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120 |
| // CHECK19-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined, ptr [[TMP0]], i32 [[TMP2]]) |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK19-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP59:![0-9]+]] |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK19-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK19-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK19-NEXT: store i32 [[TMP10]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]], i32 [[TMP11]]), !llvm.access.group [[ACC_GRP59]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP59]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP60:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK19-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l120.omp_outlined.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP5]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP3]]) |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK19: omp.dispatch.cond: |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ [[TMP8]], [[COND_TRUE]] ], [ [[TMP9]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP10]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP11]], [[TMP12]] |
| // CHECK19-NEXT: br i1 [[CMP1]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK19: omp.dispatch.body: |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP62:![0-9]+]] |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK19-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP13]], [[TMP14]] |
| // CHECK19-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP15]], 1 |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK19-NEXT: [[TMP16:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP16]] |
| // CHECK19-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK19-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK19: omp.body.continue: |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK19-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP17]], 1 |
| // CHECK19-NEXT: store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP62]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP63:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK19: omp.dispatch.inc: |
| // CHECK19-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: [[ADD4:%.*]] = add nsw i32 [[TMP18]], [[TMP19]] |
| // CHECK19-NEXT: store i32 [[ADD4]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: [[ADD5:%.*]] = add nsw i32 [[TMP20]], [[TMP21]] |
| // CHECK19-NEXT: store i32 [[ADD5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK19: omp.dispatch.end: |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP5]]) |
| // CHECK19-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0 |
| // CHECK19-NEXT: br i1 [[TMP23]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124 |
| // CHECK19-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined, ptr [[TMP0]]) |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK19-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP65:![0-9]+]] |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK19-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK19-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]]), !llvm.access.group [[ACC_GRP65]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP10]], [[TMP11]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP65]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP66:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP13:%.*]] = icmp ne i32 [[TMP12]], 0 |
| // CHECK19-NEXT: br i1 [[TMP13]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l124.omp_outlined.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP6]], i32 1073741859, i32 [[TMP3]], i32 [[TMP4]], i32 1, i32 1) |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK19: omp.dispatch.cond: |
| // CHECK19-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP6]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK19-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0 |
| // CHECK19-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK19: omp.dispatch.body: |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP8]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP68:![0-9]+]] |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1 |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP12]] |
| // CHECK19-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK19-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK19: omp.body.continue: |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK19-NEXT: [[ADD1:%.*]] = add nsw i32 [[TMP13]], 1 |
| // CHECK19-NEXT: store i32 [[ADD1]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP68]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP69:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK19: omp.dispatch.inc: |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK19: omp.dispatch.end: |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK19-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128 |
| // CHECK19-SAME: (ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP1]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4 |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined, ptr [[TMP0]], i32 [[TMP2]]) |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 9, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP2]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1) |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP3]], 9 |
| // CHECK19-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]] |
| // CHECK19: cond.true: |
| // CHECK19-NEXT: br label [[COND_END:%.*]] |
| // CHECK19: cond.false: |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: br label [[COND_END]] |
| // CHECK19: cond.end: |
| // CHECK19-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP4]], [[COND_FALSE]] ] |
| // CHECK19-NEXT: store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP71:![0-9]+]] |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK19-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP6]], [[TMP7]] |
| // CHECK19-NEXT: br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK19-NEXT: store i32 [[TMP10]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK19-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined.omp_outlined, i32 [[TMP8]], i32 [[TMP9]], ptr [[TMP0]], i32 [[TMP11]]), !llvm.access.group [[ACC_GRP71]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP12]], [[TMP13]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP71]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP72:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_LOOP_EXIT:%.*]] |
| // CHECK19: omp.loop.exit: |
| // CHECK19-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP2]]) |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP15:%.*]] = icmp ne i32 [[TMP14]], 0 |
| // CHECK19-NEXT: br i1 [[TMP15]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK19-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiLi10EEiT__l128.omp_outlined.omp_outlined |
| // CHECK19-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[DOTPREVIOUS_LB_:%.*]], i32 noundef [[DOTPREVIOUS_UB_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[A:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] { |
| // CHECK19-NEXT: entry: |
| // CHECK19-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[A_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK19-NEXT: [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK19-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP1]], ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP2]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 4 |
| // CHECK19-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4 |
| // CHECK19-NEXT: [[TMP6:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4 |
| // CHECK19-NEXT: [[TMP7:%.*]] = load i32, ptr [[TMP6]], align 4 |
| // CHECK19-NEXT: call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP7]], i32 1073741859, i32 [[TMP4]], i32 [[TMP5]], i32 1, i32 [[TMP3]]) |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_COND:%.*]] |
| // CHECK19: omp.dispatch.cond: |
| // CHECK19-NEXT: [[TMP8:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP7]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]]) |
| // CHECK19-NEXT: [[TOBOOL:%.*]] = icmp ne i32 [[TMP8]], 0 |
| // CHECK19-NEXT: br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]] |
| // CHECK19: omp.dispatch.body: |
| // CHECK19-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK19-NEXT: store i32 [[TMP9]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK19: omp.inner.for.cond: |
| // CHECK19-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP74:![0-9]+]] |
| // CHECK19-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK19-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP10]], [[TMP11]] |
| // CHECK19-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK19: omp.inner.for.body: |
| // CHECK19-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK19-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP12]], 1 |
| // CHECK19-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK19-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK19-NEXT: [[TMP13:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK19-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 [[TMP13]] |
| // CHECK19-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK19-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK19: omp.body.continue: |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK19: omp.inner.for.inc: |
| // CHECK19-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK19-NEXT: [[ADD1:%.*]] = add nsw i32 [[TMP14]], 1 |
| // CHECK19-NEXT: store i32 [[ADD1]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP74]] |
| // CHECK19-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP75:![0-9]+]] |
| // CHECK19: omp.inner.for.end: |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_INC:%.*]] |
| // CHECK19: omp.dispatch.inc: |
| // CHECK19-NEXT: br label [[OMP_DISPATCH_COND]] |
| // CHECK19: omp.dispatch.end: |
| // CHECK19-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IS_LAST]], align 4 |
| // CHECK19-NEXT: [[TMP16:%.*]] = icmp ne i32 [[TMP15]], 0 |
| // CHECK19-NEXT: br i1 [[TMP16]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]] |
| // CHECK19: .omp.final.then: |
| // CHECK19-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK19-NEXT: br label [[DOTOMP_FINAL_DONE]] |
| // CHECK19: .omp.final.done: |
| // CHECK19-NEXT: ret void |
| // |
| // |
| // CHECK21-LABEL: define {{[^@]+}}@main |
| // CHECK21-SAME: (i32 noundef signext [[ARGC:%.*]], ptr noundef [[ARGV:%.*]]) #[[ATTR0:[0-9]+]] { |
| // CHECK21-NEXT: entry: |
| // CHECK21-NEXT: [[RETVAL:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[ARGV_ADDR:%.*]] = alloca ptr, align 8 |
| // CHECK21-NEXT: [[N:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[SAVED_STACK:%.*]] = alloca ptr, align 8 |
| // CHECK21-NEXT: [[__VLA_EXPR0:%.*]] = alloca i64, align 8 |
| // CHECK21-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[_TMP10:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_11:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_12:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_LB16:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_UB17:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I18:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_IV21:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I22:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_39:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[_TMP40:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_41:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_42:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_LB46:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_UB47:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I48:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_IV51:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I52:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[_TMP69:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_70:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_71:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_LB75:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_UB76:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I77:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_IV80:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I81:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_98:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[_TMP99:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_100:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_101:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_LB105:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_UB106:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I107:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_IV110:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I111:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[RETVAL]], align 4 |
| // CHECK21-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK21-NEXT: store ptr [[ARGV]], ptr [[ARGV_ADDR]], align 8 |
| // CHECK21-NEXT: store i32 100, ptr [[N]], align 4 |
| // CHECK21-NEXT: [[TMP0:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK21-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64 |
| // CHECK21-NEXT: [[TMP2:%.*]] = call ptr @llvm.stacksave.p0() |
| // CHECK21-NEXT: store ptr [[TMP2]], ptr [[SAVED_STACK]], align 8 |
| // CHECK21-NEXT: [[VLA:%.*]] = alloca i32, i64 [[TMP1]], align 4 |
| // CHECK21-NEXT: store i64 [[TMP1]], ptr [[__VLA_EXPR0]], align 8 |
| // CHECK21-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK21-NEXT: [[TMP3:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP3]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK21-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK21-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP4]], 0 |
| // CHECK21-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK21-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK21-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK21-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP5]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK21-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK21-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP6]] |
| // CHECK21-NEXT: br i1 [[CMP]], label [[SIMD_IF_THEN:%.*]], label [[SIMD_IF_END:%.*]] |
| // CHECK21: simd.if.then: |
| // CHECK21-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP7]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK21: omp.inner.for.cond: |
| // CHECK21-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP2:![0-9]+]] |
| // CHECK21-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK21-NEXT: [[CMP4:%.*]] = icmp sle i32 [[TMP8]], [[TMP9]] |
| // CHECK21-NEXT: br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK21: omp.inner.for.body: |
| // CHECK21-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK21-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP10]], 1 |
| // CHECK21-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK21-NEXT: store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK21-NEXT: [[TMP11:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK21-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64 |
| // CHECK21-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i64 [[IDXPROM]] |
| // CHECK21-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK21-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK21: omp.body.continue: |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK21: omp.inner.for.inc: |
| // CHECK21-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK21-NEXT: [[ADD5:%.*]] = add nsw i32 [[TMP12]], 1 |
| // CHECK21-NEXT: store i32 [[ADD5]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP2]] |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]] |
| // CHECK21: omp.inner.for.end: |
| // CHECK21-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK21-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP13]], 0 |
| // CHECK21-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK21-NEXT: [[MUL8:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK21-NEXT: [[ADD9:%.*]] = add nsw i32 0, [[MUL8]] |
| // CHECK21-NEXT: store i32 [[ADD9]], ptr [[I3]], align 4 |
| // CHECK21-NEXT: br label [[SIMD_IF_END]] |
| // CHECK21: simd.if.end: |
| // CHECK21-NEXT: [[TMP14:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP14]], ptr [[DOTCAPTURE_EXPR_11]], align 4 |
| // CHECK21-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_11]], align 4 |
| // CHECK21-NEXT: [[SUB13:%.*]] = sub nsw i32 [[TMP15]], 0 |
| // CHECK21-NEXT: [[DIV14:%.*]] = sdiv i32 [[SUB13]], 1 |
| // CHECK21-NEXT: [[SUB15:%.*]] = sub nsw i32 [[DIV14]], 1 |
| // CHECK21-NEXT: store i32 [[SUB15]], ptr [[DOTCAPTURE_EXPR_12]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[DOTOMP_LB16]], align 4 |
| // CHECK21-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP16]], ptr [[DOTOMP_UB17]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[I18]], align 4 |
| // CHECK21-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_11]], align 4 |
| // CHECK21-NEXT: [[CMP19:%.*]] = icmp slt i32 0, [[TMP17]] |
| // CHECK21-NEXT: br i1 [[CMP19]], label [[SIMD_IF_THEN20:%.*]], label [[SIMD_IF_END38:%.*]] |
| // CHECK21: simd.if.then20: |
| // CHECK21-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB16]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP18]], ptr [[DOTOMP_IV21]], align 4 |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND23:%.*]] |
| // CHECK21: omp.inner.for.cond23: |
| // CHECK21-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP6:![0-9]+]] |
| // CHECK21-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB17]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK21-NEXT: [[CMP24:%.*]] = icmp sle i32 [[TMP19]], [[TMP20]] |
| // CHECK21-NEXT: br i1 [[CMP24]], label [[OMP_INNER_FOR_BODY25:%.*]], label [[OMP_INNER_FOR_END33:%.*]] |
| // CHECK21: omp.inner.for.body25: |
| // CHECK21-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK21-NEXT: [[MUL26:%.*]] = mul nsw i32 [[TMP21]], 1 |
| // CHECK21-NEXT: [[ADD27:%.*]] = add nsw i32 0, [[MUL26]] |
| // CHECK21-NEXT: store i32 [[ADD27]], ptr [[I22]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK21-NEXT: [[TMP22:%.*]] = load i32, ptr [[I22]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK21-NEXT: [[IDXPROM28:%.*]] = sext i32 [[TMP22]] to i64 |
| // CHECK21-NEXT: [[ARRAYIDX29:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i64 [[IDXPROM28]] |
| // CHECK21-NEXT: store i32 0, ptr [[ARRAYIDX29]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK21-NEXT: br label [[OMP_BODY_CONTINUE30:%.*]] |
| // CHECK21: omp.body.continue30: |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_INC31:%.*]] |
| // CHECK21: omp.inner.for.inc31: |
| // CHECK21-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK21-NEXT: [[ADD32:%.*]] = add nsw i32 [[TMP23]], 1 |
| // CHECK21-NEXT: store i32 [[ADD32]], ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP6]] |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND23]], !llvm.loop [[LOOP7:![0-9]+]] |
| // CHECK21: omp.inner.for.end33: |
| // CHECK21-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_11]], align 4 |
| // CHECK21-NEXT: [[SUB34:%.*]] = sub nsw i32 [[TMP24]], 0 |
| // CHECK21-NEXT: [[DIV35:%.*]] = sdiv i32 [[SUB34]], 1 |
| // CHECK21-NEXT: [[MUL36:%.*]] = mul nsw i32 [[DIV35]], 1 |
| // CHECK21-NEXT: [[ADD37:%.*]] = add nsw i32 0, [[MUL36]] |
| // CHECK21-NEXT: store i32 [[ADD37]], ptr [[I22]], align 4 |
| // CHECK21-NEXT: br label [[SIMD_IF_END38]] |
| // CHECK21: simd.if.end38: |
| // CHECK21-NEXT: [[TMP25:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP25]], ptr [[DOTCAPTURE_EXPR_39]], align 4 |
| // CHECK21-NEXT: [[TMP26:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP26]], ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK21-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK21-NEXT: [[SUB43:%.*]] = sub nsw i32 [[TMP27]], 0 |
| // CHECK21-NEXT: [[DIV44:%.*]] = sdiv i32 [[SUB43]], 1 |
| // CHECK21-NEXT: [[SUB45:%.*]] = sub nsw i32 [[DIV44]], 1 |
| // CHECK21-NEXT: store i32 [[SUB45]], ptr [[DOTCAPTURE_EXPR_42]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[DOTOMP_LB46]], align 4 |
| // CHECK21-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_42]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP28]], ptr [[DOTOMP_UB47]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[I48]], align 4 |
| // CHECK21-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK21-NEXT: [[CMP49:%.*]] = icmp slt i32 0, [[TMP29]] |
| // CHECK21-NEXT: br i1 [[CMP49]], label [[SIMD_IF_THEN50:%.*]], label [[SIMD_IF_END68:%.*]] |
| // CHECK21: simd.if.then50: |
| // CHECK21-NEXT: [[TMP30:%.*]] = load i32, ptr [[DOTOMP_LB46]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP30]], ptr [[DOTOMP_IV51]], align 4 |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND53:%.*]] |
| // CHECK21: omp.inner.for.cond53: |
| // CHECK21-NEXT: [[TMP31:%.*]] = load i32, ptr [[DOTOMP_IV51]], align 4, !llvm.access.group [[ACC_GRP9:![0-9]+]] |
| // CHECK21-NEXT: [[TMP32:%.*]] = load i32, ptr [[DOTOMP_UB47]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK21-NEXT: [[CMP54:%.*]] = icmp sle i32 [[TMP31]], [[TMP32]] |
| // CHECK21-NEXT: br i1 [[CMP54]], label [[OMP_INNER_FOR_BODY55:%.*]], label [[OMP_INNER_FOR_END63:%.*]] |
| // CHECK21: omp.inner.for.body55: |
| // CHECK21-NEXT: [[TMP33:%.*]] = load i32, ptr [[DOTOMP_IV51]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK21-NEXT: [[MUL56:%.*]] = mul nsw i32 [[TMP33]], 1 |
| // CHECK21-NEXT: [[ADD57:%.*]] = add nsw i32 0, [[MUL56]] |
| // CHECK21-NEXT: store i32 [[ADD57]], ptr [[I52]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK21-NEXT: [[TMP34:%.*]] = load i32, ptr [[I52]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK21-NEXT: [[IDXPROM58:%.*]] = sext i32 [[TMP34]] to i64 |
| // CHECK21-NEXT: [[ARRAYIDX59:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i64 [[IDXPROM58]] |
| // CHECK21-NEXT: store i32 0, ptr [[ARRAYIDX59]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK21-NEXT: br label [[OMP_BODY_CONTINUE60:%.*]] |
| // CHECK21: omp.body.continue60: |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_INC61:%.*]] |
| // CHECK21: omp.inner.for.inc61: |
| // CHECK21-NEXT: [[TMP35:%.*]] = load i32, ptr [[DOTOMP_IV51]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK21-NEXT: [[ADD62:%.*]] = add nsw i32 [[TMP35]], 1 |
| // CHECK21-NEXT: store i32 [[ADD62]], ptr [[DOTOMP_IV51]], align 4, !llvm.access.group [[ACC_GRP9]] |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND53]], !llvm.loop [[LOOP10:![0-9]+]] |
| // CHECK21: omp.inner.for.end63: |
| // CHECK21-NEXT: [[TMP36:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK21-NEXT: [[SUB64:%.*]] = sub nsw i32 [[TMP36]], 0 |
| // CHECK21-NEXT: [[DIV65:%.*]] = sdiv i32 [[SUB64]], 1 |
| // CHECK21-NEXT: [[MUL66:%.*]] = mul nsw i32 [[DIV65]], 1 |
| // CHECK21-NEXT: [[ADD67:%.*]] = add nsw i32 0, [[MUL66]] |
| // CHECK21-NEXT: store i32 [[ADD67]], ptr [[I52]], align 4 |
| // CHECK21-NEXT: br label [[SIMD_IF_END68]] |
| // CHECK21: simd.if.end68: |
| // CHECK21-NEXT: [[TMP37:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP37]], ptr [[DOTCAPTURE_EXPR_70]], align 4 |
| // CHECK21-NEXT: [[TMP38:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_70]], align 4 |
| // CHECK21-NEXT: [[SUB72:%.*]] = sub nsw i32 [[TMP38]], 0 |
| // CHECK21-NEXT: [[DIV73:%.*]] = sdiv i32 [[SUB72]], 1 |
| // CHECK21-NEXT: [[SUB74:%.*]] = sub nsw i32 [[DIV73]], 1 |
| // CHECK21-NEXT: store i32 [[SUB74]], ptr [[DOTCAPTURE_EXPR_71]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[DOTOMP_LB75]], align 4 |
| // CHECK21-NEXT: [[TMP39:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_71]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP39]], ptr [[DOTOMP_UB76]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[I77]], align 4 |
| // CHECK21-NEXT: [[TMP40:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_70]], align 4 |
| // CHECK21-NEXT: [[CMP78:%.*]] = icmp slt i32 0, [[TMP40]] |
| // CHECK21-NEXT: br i1 [[CMP78]], label [[SIMD_IF_THEN79:%.*]], label [[SIMD_IF_END97:%.*]] |
| // CHECK21: simd.if.then79: |
| // CHECK21-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTOMP_LB75]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP41]], ptr [[DOTOMP_IV80]], align 4 |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND82:%.*]] |
| // CHECK21: omp.inner.for.cond82: |
| // CHECK21-NEXT: [[TMP42:%.*]] = load i32, ptr [[DOTOMP_IV80]], align 4, !llvm.access.group [[ACC_GRP12:![0-9]+]] |
| // CHECK21-NEXT: [[TMP43:%.*]] = load i32, ptr [[DOTOMP_UB76]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK21-NEXT: [[CMP83:%.*]] = icmp sle i32 [[TMP42]], [[TMP43]] |
| // CHECK21-NEXT: br i1 [[CMP83]], label [[OMP_INNER_FOR_BODY84:%.*]], label [[OMP_INNER_FOR_END92:%.*]] |
| // CHECK21: omp.inner.for.body84: |
| // CHECK21-NEXT: [[TMP44:%.*]] = load i32, ptr [[DOTOMP_IV80]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK21-NEXT: [[MUL85:%.*]] = mul nsw i32 [[TMP44]], 1 |
| // CHECK21-NEXT: [[ADD86:%.*]] = add nsw i32 0, [[MUL85]] |
| // CHECK21-NEXT: store i32 [[ADD86]], ptr [[I81]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK21-NEXT: [[TMP45:%.*]] = load i32, ptr [[I81]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK21-NEXT: [[IDXPROM87:%.*]] = sext i32 [[TMP45]] to i64 |
| // CHECK21-NEXT: [[ARRAYIDX88:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i64 [[IDXPROM87]] |
| // CHECK21-NEXT: store i32 0, ptr [[ARRAYIDX88]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK21-NEXT: br label [[OMP_BODY_CONTINUE89:%.*]] |
| // CHECK21: omp.body.continue89: |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_INC90:%.*]] |
| // CHECK21: omp.inner.for.inc90: |
| // CHECK21-NEXT: [[TMP46:%.*]] = load i32, ptr [[DOTOMP_IV80]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK21-NEXT: [[ADD91:%.*]] = add nsw i32 [[TMP46]], 1 |
| // CHECK21-NEXT: store i32 [[ADD91]], ptr [[DOTOMP_IV80]], align 4, !llvm.access.group [[ACC_GRP12]] |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND82]], !llvm.loop [[LOOP13:![0-9]+]] |
| // CHECK21: omp.inner.for.end92: |
| // CHECK21-NEXT: [[TMP47:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_70]], align 4 |
| // CHECK21-NEXT: [[SUB93:%.*]] = sub nsw i32 [[TMP47]], 0 |
| // CHECK21-NEXT: [[DIV94:%.*]] = sdiv i32 [[SUB93]], 1 |
| // CHECK21-NEXT: [[MUL95:%.*]] = mul nsw i32 [[DIV94]], 1 |
| // CHECK21-NEXT: [[ADD96:%.*]] = add nsw i32 0, [[MUL95]] |
| // CHECK21-NEXT: store i32 [[ADD96]], ptr [[I81]], align 4 |
| // CHECK21-NEXT: br label [[SIMD_IF_END97]] |
| // CHECK21: simd.if.end97: |
| // CHECK21-NEXT: [[TMP48:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP48]], ptr [[DOTCAPTURE_EXPR_98]], align 4 |
| // CHECK21-NEXT: [[TMP49:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP49]], ptr [[DOTCAPTURE_EXPR_100]], align 4 |
| // CHECK21-NEXT: [[TMP50:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_100]], align 4 |
| // CHECK21-NEXT: [[SUB102:%.*]] = sub nsw i32 [[TMP50]], 0 |
| // CHECK21-NEXT: [[DIV103:%.*]] = sdiv i32 [[SUB102]], 1 |
| // CHECK21-NEXT: [[SUB104:%.*]] = sub nsw i32 [[DIV103]], 1 |
| // CHECK21-NEXT: store i32 [[SUB104]], ptr [[DOTCAPTURE_EXPR_101]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[DOTOMP_LB105]], align 4 |
| // CHECK21-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_101]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP51]], ptr [[DOTOMP_UB106]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[I107]], align 4 |
| // CHECK21-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_100]], align 4 |
| // CHECK21-NEXT: [[CMP108:%.*]] = icmp slt i32 0, [[TMP52]] |
| // CHECK21-NEXT: br i1 [[CMP108]], label [[SIMD_IF_THEN109:%.*]], label [[SIMD_IF_END127:%.*]] |
| // CHECK21: simd.if.then109: |
| // CHECK21-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTOMP_LB105]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP53]], ptr [[DOTOMP_IV110]], align 4 |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND112:%.*]] |
| // CHECK21: omp.inner.for.cond112: |
| // CHECK21-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTOMP_IV110]], align 4, !llvm.access.group [[ACC_GRP15:![0-9]+]] |
| // CHECK21-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTOMP_UB106]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK21-NEXT: [[CMP113:%.*]] = icmp sle i32 [[TMP54]], [[TMP55]] |
| // CHECK21-NEXT: br i1 [[CMP113]], label [[OMP_INNER_FOR_BODY114:%.*]], label [[OMP_INNER_FOR_END122:%.*]] |
| // CHECK21: omp.inner.for.body114: |
| // CHECK21-NEXT: [[TMP56:%.*]] = load i32, ptr [[DOTOMP_IV110]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK21-NEXT: [[MUL115:%.*]] = mul nsw i32 [[TMP56]], 1 |
| // CHECK21-NEXT: [[ADD116:%.*]] = add nsw i32 0, [[MUL115]] |
| // CHECK21-NEXT: store i32 [[ADD116]], ptr [[I111]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK21-NEXT: [[TMP57:%.*]] = load i32, ptr [[I111]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK21-NEXT: [[IDXPROM117:%.*]] = sext i32 [[TMP57]] to i64 |
| // CHECK21-NEXT: [[ARRAYIDX118:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i64 [[IDXPROM117]] |
| // CHECK21-NEXT: store i32 0, ptr [[ARRAYIDX118]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK21-NEXT: br label [[OMP_BODY_CONTINUE119:%.*]] |
| // CHECK21: omp.body.continue119: |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_INC120:%.*]] |
| // CHECK21: omp.inner.for.inc120: |
| // CHECK21-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTOMP_IV110]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK21-NEXT: [[ADD121:%.*]] = add nsw i32 [[TMP58]], 1 |
| // CHECK21-NEXT: store i32 [[ADD121]], ptr [[DOTOMP_IV110]], align 4, !llvm.access.group [[ACC_GRP15]] |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND112]], !llvm.loop [[LOOP16:![0-9]+]] |
| // CHECK21: omp.inner.for.end122: |
| // CHECK21-NEXT: [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_100]], align 4 |
| // CHECK21-NEXT: [[SUB123:%.*]] = sub nsw i32 [[TMP59]], 0 |
| // CHECK21-NEXT: [[DIV124:%.*]] = sdiv i32 [[SUB123]], 1 |
| // CHECK21-NEXT: [[MUL125:%.*]] = mul nsw i32 [[DIV124]], 1 |
| // CHECK21-NEXT: [[ADD126:%.*]] = add nsw i32 0, [[MUL125]] |
| // CHECK21-NEXT: store i32 [[ADD126]], ptr [[I111]], align 4 |
| // CHECK21-NEXT: br label [[SIMD_IF_END127]] |
| // CHECK21: simd.if.end127: |
| // CHECK21-NEXT: [[TMP60:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4 |
| // CHECK21-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiLi10EEiT_(i32 noundef signext [[TMP60]]) |
| // CHECK21-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4 |
| // CHECK21-NEXT: [[TMP61:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8 |
| // CHECK21-NEXT: call void @llvm.stackrestore.p0(ptr [[TMP61]]) |
| // CHECK21-NEXT: [[TMP62:%.*]] = load i32, ptr [[RETVAL]], align 4 |
| // CHECK21-NEXT: ret i32 [[TMP62]] |
| // |
| // |
| // CHECK21-LABEL: define {{[^@]+}}@_Z5tmainIiLi10EEiT_ |
| // CHECK21-SAME: (i32 noundef signext [[ARGC:%.*]]) #[[ATTR2:[0-9]+]] comdat { |
| // CHECK21-NEXT: entry: |
| // CHECK21-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[A:%.*]] = alloca [10 x i32], align 4 |
| // CHECK21-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[_TMP2:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_LB3:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_UB4:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_IV5:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I6:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[_TMP18:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_LB19:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_UB20:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_IV21:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I22:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[_TMP34:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_LB35:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_UB36:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_IV37:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I38:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTCAPTURE_EXPR_50:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[_TMP51:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_LB52:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_UB53:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[DOTOMP_IV54:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: [[I55:%.*]] = alloca i32, align 4 |
| // CHECK21-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK21-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK21-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK21-NEXT: [[TMP0:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP0]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK21: omp.inner.for.cond: |
| // CHECK21-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18:![0-9]+]] |
| // CHECK21-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK21-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP1]], [[TMP2]] |
| // CHECK21-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK21: omp.inner.for.body: |
| // CHECK21-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK21-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP3]], 1 |
| // CHECK21-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK21-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK21-NEXT: [[TMP4:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK21-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP4]] to i64 |
| // CHECK21-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i64 0, i64 [[IDXPROM]] |
| // CHECK21-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK21-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK21: omp.body.continue: |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK21: omp.inner.for.inc: |
| // CHECK21-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK21-NEXT: [[ADD1:%.*]] = add nsw i32 [[TMP5]], 1 |
| // CHECK21-NEXT: store i32 [[ADD1]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP18]] |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP19:![0-9]+]] |
| // CHECK21: omp.inner.for.end: |
| // CHECK21-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[DOTOMP_LB3]], align 4 |
| // CHECK21-NEXT: store i32 9, ptr [[DOTOMP_UB4]], align 4 |
| // CHECK21-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_LB3]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_IV5]], align 4 |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND7:%.*]] |
| // CHECK21: omp.inner.for.cond7: |
| // CHECK21-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV5]], align 4, !llvm.access.group [[ACC_GRP21:![0-9]+]] |
| // CHECK21-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB4]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK21-NEXT: [[CMP8:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]] |
| // CHECK21-NEXT: br i1 [[CMP8]], label [[OMP_INNER_FOR_BODY9:%.*]], label [[OMP_INNER_FOR_END17:%.*]] |
| // CHECK21: omp.inner.for.body9: |
| // CHECK21-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV5]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK21-NEXT: [[MUL10:%.*]] = mul nsw i32 [[TMP9]], 1 |
| // CHECK21-NEXT: [[ADD11:%.*]] = add nsw i32 0, [[MUL10]] |
| // CHECK21-NEXT: store i32 [[ADD11]], ptr [[I6]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK21-NEXT: [[TMP10:%.*]] = load i32, ptr [[I6]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK21-NEXT: [[IDXPROM12:%.*]] = sext i32 [[TMP10]] to i64 |
| // CHECK21-NEXT: [[ARRAYIDX13:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i64 0, i64 [[IDXPROM12]] |
| // CHECK21-NEXT: store i32 0, ptr [[ARRAYIDX13]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK21-NEXT: br label [[OMP_BODY_CONTINUE14:%.*]] |
| // CHECK21: omp.body.continue14: |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_INC15:%.*]] |
| // CHECK21: omp.inner.for.inc15: |
| // CHECK21-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV5]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK21-NEXT: [[ADD16:%.*]] = add nsw i32 [[TMP11]], 1 |
| // CHECK21-NEXT: store i32 [[ADD16]], ptr [[DOTOMP_IV5]], align 4, !llvm.access.group [[ACC_GRP21]] |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND7]], !llvm.loop [[LOOP22:![0-9]+]] |
| // CHECK21: omp.inner.for.end17: |
| // CHECK21-NEXT: store i32 10, ptr [[I6]], align 4 |
| // CHECK21-NEXT: [[TMP12:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP12]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[DOTOMP_LB19]], align 4 |
| // CHECK21-NEXT: store i32 9, ptr [[DOTOMP_UB20]], align 4 |
| // CHECK21-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_LB19]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP13]], ptr [[DOTOMP_IV21]], align 4 |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND23:%.*]] |
| // CHECK21: omp.inner.for.cond23: |
| // CHECK21-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP24:![0-9]+]] |
| // CHECK21-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB20]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK21-NEXT: [[CMP24:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]] |
| // CHECK21-NEXT: br i1 [[CMP24]], label [[OMP_INNER_FOR_BODY25:%.*]], label [[OMP_INNER_FOR_END33:%.*]] |
| // CHECK21: omp.inner.for.body25: |
| // CHECK21-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK21-NEXT: [[MUL26:%.*]] = mul nsw i32 [[TMP16]], 1 |
| // CHECK21-NEXT: [[ADD27:%.*]] = add nsw i32 0, [[MUL26]] |
| // CHECK21-NEXT: store i32 [[ADD27]], ptr [[I22]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK21-NEXT: [[TMP17:%.*]] = load i32, ptr [[I22]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK21-NEXT: [[IDXPROM28:%.*]] = sext i32 [[TMP17]] to i64 |
| // CHECK21-NEXT: [[ARRAYIDX29:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i64 0, i64 [[IDXPROM28]] |
| // CHECK21-NEXT: store i32 0, ptr [[ARRAYIDX29]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK21-NEXT: br label [[OMP_BODY_CONTINUE30:%.*]] |
| // CHECK21: omp.body.continue30: |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_INC31:%.*]] |
| // CHECK21: omp.inner.for.inc31: |
| // CHECK21-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK21-NEXT: [[ADD32:%.*]] = add nsw i32 [[TMP18]], 1 |
| // CHECK21-NEXT: store i32 [[ADD32]], ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP24]] |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND23]], !llvm.loop [[LOOP25:![0-9]+]] |
| // CHECK21: omp.inner.for.end33: |
| // CHECK21-NEXT: store i32 10, ptr [[I22]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[DOTOMP_LB35]], align 4 |
| // CHECK21-NEXT: store i32 9, ptr [[DOTOMP_UB36]], align 4 |
| // CHECK21-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_LB35]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP19]], ptr [[DOTOMP_IV37]], align 4 |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND39:%.*]] |
| // CHECK21: omp.inner.for.cond39: |
| // CHECK21-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV37]], align 4, !llvm.access.group [[ACC_GRP27:![0-9]+]] |
| // CHECK21-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB36]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK21-NEXT: [[CMP40:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]] |
| // CHECK21-NEXT: br i1 [[CMP40]], label [[OMP_INNER_FOR_BODY41:%.*]], label [[OMP_INNER_FOR_END49:%.*]] |
| // CHECK21: omp.inner.for.body41: |
| // CHECK21-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV37]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK21-NEXT: [[MUL42:%.*]] = mul nsw i32 [[TMP22]], 1 |
| // CHECK21-NEXT: [[ADD43:%.*]] = add nsw i32 0, [[MUL42]] |
| // CHECK21-NEXT: store i32 [[ADD43]], ptr [[I38]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK21-NEXT: [[TMP23:%.*]] = load i32, ptr [[I38]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK21-NEXT: [[IDXPROM44:%.*]] = sext i32 [[TMP23]] to i64 |
| // CHECK21-NEXT: [[ARRAYIDX45:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i64 0, i64 [[IDXPROM44]] |
| // CHECK21-NEXT: store i32 0, ptr [[ARRAYIDX45]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK21-NEXT: br label [[OMP_BODY_CONTINUE46:%.*]] |
| // CHECK21: omp.body.continue46: |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_INC47:%.*]] |
| // CHECK21: omp.inner.for.inc47: |
| // CHECK21-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTOMP_IV37]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK21-NEXT: [[ADD48:%.*]] = add nsw i32 [[TMP24]], 1 |
| // CHECK21-NEXT: store i32 [[ADD48]], ptr [[DOTOMP_IV37]], align 4, !llvm.access.group [[ACC_GRP27]] |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND39]], !llvm.loop [[LOOP28:![0-9]+]] |
| // CHECK21: omp.inner.for.end49: |
| // CHECK21-NEXT: store i32 10, ptr [[I38]], align 4 |
| // CHECK21-NEXT: [[TMP25:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP25]], ptr [[DOTCAPTURE_EXPR_50]], align 4 |
| // CHECK21-NEXT: store i32 0, ptr [[DOTOMP_LB52]], align 4 |
| // CHECK21-NEXT: store i32 9, ptr [[DOTOMP_UB53]], align 4 |
| // CHECK21-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTOMP_LB52]], align 4 |
| // CHECK21-NEXT: store i32 [[TMP26]], ptr [[DOTOMP_IV54]], align 4 |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND56:%.*]] |
| // CHECK21: omp.inner.for.cond56: |
| // CHECK21-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IV54]], align 4, !llvm.access.group [[ACC_GRP30:![0-9]+]] |
| // CHECK21-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTOMP_UB53]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK21-NEXT: [[CMP57:%.*]] = icmp sle i32 [[TMP27]], [[TMP28]] |
| // CHECK21-NEXT: br i1 [[CMP57]], label [[OMP_INNER_FOR_BODY58:%.*]], label [[OMP_INNER_FOR_END66:%.*]] |
| // CHECK21: omp.inner.for.body58: |
| // CHECK21-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IV54]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK21-NEXT: [[MUL59:%.*]] = mul nsw i32 [[TMP29]], 1 |
| // CHECK21-NEXT: [[ADD60:%.*]] = add nsw i32 0, [[MUL59]] |
| // CHECK21-NEXT: store i32 [[ADD60]], ptr [[I55]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK21-NEXT: [[TMP30:%.*]] = load i32, ptr [[I55]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK21-NEXT: [[IDXPROM61:%.*]] = sext i32 [[TMP30]] to i64 |
| // CHECK21-NEXT: [[ARRAYIDX62:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i64 0, i64 [[IDXPROM61]] |
| // CHECK21-NEXT: store i32 0, ptr [[ARRAYIDX62]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK21-NEXT: br label [[OMP_BODY_CONTINUE63:%.*]] |
| // CHECK21: omp.body.continue63: |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_INC64:%.*]] |
| // CHECK21: omp.inner.for.inc64: |
| // CHECK21-NEXT: [[TMP31:%.*]] = load i32, ptr [[DOTOMP_IV54]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK21-NEXT: [[ADD65:%.*]] = add nsw i32 [[TMP31]], 1 |
| // CHECK21-NEXT: store i32 [[ADD65]], ptr [[DOTOMP_IV54]], align 4, !llvm.access.group [[ACC_GRP30]] |
| // CHECK21-NEXT: br label [[OMP_INNER_FOR_COND56]], !llvm.loop [[LOOP31:![0-9]+]] |
| // CHECK21: omp.inner.for.end66: |
| // CHECK21-NEXT: store i32 10, ptr [[I55]], align 4 |
| // CHECK21-NEXT: ret i32 0 |
| // |
| // |
| // CHECK23-LABEL: define {{[^@]+}}@main |
| // CHECK23-SAME: (i32 noundef [[ARGC:%.*]], ptr noundef [[ARGV:%.*]]) #[[ATTR0:[0-9]+]] { |
| // CHECK23-NEXT: entry: |
| // CHECK23-NEXT: [[RETVAL:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[ARGV_ADDR:%.*]] = alloca ptr, align 4 |
| // CHECK23-NEXT: [[N:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[SAVED_STACK:%.*]] = alloca ptr, align 4 |
| // CHECK23-NEXT: [[__VLA_EXPR0:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I3:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[_TMP10:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_11:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_12:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_LB16:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_UB17:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I18:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_IV21:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I22:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_38:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[_TMP39:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_40:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_41:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_LB45:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_UB46:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I47:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_IV50:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I51:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[_TMP67:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_68:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_69:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_LB73:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_UB74:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I75:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_IV78:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I79:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_95:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[_TMP96:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_97:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_98:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_LB102:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_UB103:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I104:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_IV107:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I108:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[RETVAL]], align 4 |
| // CHECK23-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK23-NEXT: store ptr [[ARGV]], ptr [[ARGV_ADDR]], align 4 |
| // CHECK23-NEXT: store i32 100, ptr [[N]], align 4 |
| // CHECK23-NEXT: [[TMP0:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK23-NEXT: [[TMP1:%.*]] = call ptr @llvm.stacksave.p0() |
| // CHECK23-NEXT: store ptr [[TMP1]], ptr [[SAVED_STACK]], align 4 |
| // CHECK23-NEXT: [[VLA:%.*]] = alloca i32, i32 [[TMP0]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP0]], ptr [[__VLA_EXPR0]], align 4 |
| // CHECK23-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK23-NEXT: [[TMP2:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP2]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK23-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK23-NEXT: [[SUB:%.*]] = sub nsw i32 [[TMP3]], 0 |
| // CHECK23-NEXT: [[DIV:%.*]] = sdiv i32 [[SUB]], 1 |
| // CHECK23-NEXT: [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1 |
| // CHECK23-NEXT: store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK23-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP4]], ptr [[DOTOMP_UB]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[I]], align 4 |
| // CHECK23-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK23-NEXT: [[CMP:%.*]] = icmp slt i32 0, [[TMP5]] |
| // CHECK23-NEXT: br i1 [[CMP]], label [[SIMD_IF_THEN:%.*]], label [[SIMD_IF_END:%.*]] |
| // CHECK23: simd.if.then: |
| // CHECK23-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK23: omp.inner.for.cond: |
| // CHECK23-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP3:![0-9]+]] |
| // CHECK23-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK23-NEXT: [[CMP4:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]] |
| // CHECK23-NEXT: br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK23: omp.inner.for.body: |
| // CHECK23-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK23-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP9]], 1 |
| // CHECK23-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK23-NEXT: store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK23-NEXT: [[TMP10:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK23-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i32 [[TMP10]] |
| // CHECK23-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK23-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK23: omp.body.continue: |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK23: omp.inner.for.inc: |
| // CHECK23-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK23-NEXT: [[ADD5:%.*]] = add nsw i32 [[TMP11]], 1 |
| // CHECK23-NEXT: store i32 [[ADD5]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP3]] |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]] |
| // CHECK23: omp.inner.for.end: |
| // CHECK23-NEXT: [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK23-NEXT: [[SUB6:%.*]] = sub nsw i32 [[TMP12]], 0 |
| // CHECK23-NEXT: [[DIV7:%.*]] = sdiv i32 [[SUB6]], 1 |
| // CHECK23-NEXT: [[MUL8:%.*]] = mul nsw i32 [[DIV7]], 1 |
| // CHECK23-NEXT: [[ADD9:%.*]] = add nsw i32 0, [[MUL8]] |
| // CHECK23-NEXT: store i32 [[ADD9]], ptr [[I3]], align 4 |
| // CHECK23-NEXT: br label [[SIMD_IF_END]] |
| // CHECK23: simd.if.end: |
| // CHECK23-NEXT: [[TMP13:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP13]], ptr [[DOTCAPTURE_EXPR_11]], align 4 |
| // CHECK23-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_11]], align 4 |
| // CHECK23-NEXT: [[SUB13:%.*]] = sub nsw i32 [[TMP14]], 0 |
| // CHECK23-NEXT: [[DIV14:%.*]] = sdiv i32 [[SUB13]], 1 |
| // CHECK23-NEXT: [[SUB15:%.*]] = sub nsw i32 [[DIV14]], 1 |
| // CHECK23-NEXT: store i32 [[SUB15]], ptr [[DOTCAPTURE_EXPR_12]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[DOTOMP_LB16]], align 4 |
| // CHECK23-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_12]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP15]], ptr [[DOTOMP_UB17]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[I18]], align 4 |
| // CHECK23-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_11]], align 4 |
| // CHECK23-NEXT: [[CMP19:%.*]] = icmp slt i32 0, [[TMP16]] |
| // CHECK23-NEXT: br i1 [[CMP19]], label [[SIMD_IF_THEN20:%.*]], label [[SIMD_IF_END37:%.*]] |
| // CHECK23: simd.if.then20: |
| // CHECK23-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB16]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP17]], ptr [[DOTOMP_IV21]], align 4 |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND23:%.*]] |
| // CHECK23: omp.inner.for.cond23: |
| // CHECK23-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP7:![0-9]+]] |
| // CHECK23-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB17]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK23-NEXT: [[CMP24:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]] |
| // CHECK23-NEXT: br i1 [[CMP24]], label [[OMP_INNER_FOR_BODY25:%.*]], label [[OMP_INNER_FOR_END32:%.*]] |
| // CHECK23: omp.inner.for.body25: |
| // CHECK23-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK23-NEXT: [[MUL26:%.*]] = mul nsw i32 [[TMP20]], 1 |
| // CHECK23-NEXT: [[ADD27:%.*]] = add nsw i32 0, [[MUL26]] |
| // CHECK23-NEXT: store i32 [[ADD27]], ptr [[I22]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK23-NEXT: [[TMP21:%.*]] = load i32, ptr [[I22]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK23-NEXT: [[ARRAYIDX28:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i32 [[TMP21]] |
| // CHECK23-NEXT: store i32 0, ptr [[ARRAYIDX28]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK23-NEXT: br label [[OMP_BODY_CONTINUE29:%.*]] |
| // CHECK23: omp.body.continue29: |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_INC30:%.*]] |
| // CHECK23: omp.inner.for.inc30: |
| // CHECK23-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK23-NEXT: [[ADD31:%.*]] = add nsw i32 [[TMP22]], 1 |
| // CHECK23-NEXT: store i32 [[ADD31]], ptr [[DOTOMP_IV21]], align 4, !llvm.access.group [[ACC_GRP7]] |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND23]], !llvm.loop [[LOOP8:![0-9]+]] |
| // CHECK23: omp.inner.for.end32: |
| // CHECK23-NEXT: [[TMP23:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_11]], align 4 |
| // CHECK23-NEXT: [[SUB33:%.*]] = sub nsw i32 [[TMP23]], 0 |
| // CHECK23-NEXT: [[DIV34:%.*]] = sdiv i32 [[SUB33]], 1 |
| // CHECK23-NEXT: [[MUL35:%.*]] = mul nsw i32 [[DIV34]], 1 |
| // CHECK23-NEXT: [[ADD36:%.*]] = add nsw i32 0, [[MUL35]] |
| // CHECK23-NEXT: store i32 [[ADD36]], ptr [[I22]], align 4 |
| // CHECK23-NEXT: br label [[SIMD_IF_END37]] |
| // CHECK23: simd.if.end37: |
| // CHECK23-NEXT: [[TMP24:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP24]], ptr [[DOTCAPTURE_EXPR_38]], align 4 |
| // CHECK23-NEXT: [[TMP25:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP25]], ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK23-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK23-NEXT: [[SUB42:%.*]] = sub nsw i32 [[TMP26]], 0 |
| // CHECK23-NEXT: [[DIV43:%.*]] = sdiv i32 [[SUB42]], 1 |
| // CHECK23-NEXT: [[SUB44:%.*]] = sub nsw i32 [[DIV43]], 1 |
| // CHECK23-NEXT: store i32 [[SUB44]], ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[DOTOMP_LB45]], align 4 |
| // CHECK23-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_41]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP27]], ptr [[DOTOMP_UB46]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[I47]], align 4 |
| // CHECK23-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK23-NEXT: [[CMP48:%.*]] = icmp slt i32 0, [[TMP28]] |
| // CHECK23-NEXT: br i1 [[CMP48]], label [[SIMD_IF_THEN49:%.*]], label [[SIMD_IF_END66:%.*]] |
| // CHECK23: simd.if.then49: |
| // CHECK23-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTOMP_LB45]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP29]], ptr [[DOTOMP_IV50]], align 4 |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND52:%.*]] |
| // CHECK23: omp.inner.for.cond52: |
| // CHECK23-NEXT: [[TMP30:%.*]] = load i32, ptr [[DOTOMP_IV50]], align 4, !llvm.access.group [[ACC_GRP10:![0-9]+]] |
| // CHECK23-NEXT: [[TMP31:%.*]] = load i32, ptr [[DOTOMP_UB46]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK23-NEXT: [[CMP53:%.*]] = icmp sle i32 [[TMP30]], [[TMP31]] |
| // CHECK23-NEXT: br i1 [[CMP53]], label [[OMP_INNER_FOR_BODY54:%.*]], label [[OMP_INNER_FOR_END61:%.*]] |
| // CHECK23: omp.inner.for.body54: |
| // CHECK23-NEXT: [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV50]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK23-NEXT: [[MUL55:%.*]] = mul nsw i32 [[TMP32]], 1 |
| // CHECK23-NEXT: [[ADD56:%.*]] = add nsw i32 0, [[MUL55]] |
| // CHECK23-NEXT: store i32 [[ADD56]], ptr [[I51]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK23-NEXT: [[TMP33:%.*]] = load i32, ptr [[I51]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK23-NEXT: [[ARRAYIDX57:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i32 [[TMP33]] |
| // CHECK23-NEXT: store i32 0, ptr [[ARRAYIDX57]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK23-NEXT: br label [[OMP_BODY_CONTINUE58:%.*]] |
| // CHECK23: omp.body.continue58: |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_INC59:%.*]] |
| // CHECK23: omp.inner.for.inc59: |
| // CHECK23-NEXT: [[TMP34:%.*]] = load i32, ptr [[DOTOMP_IV50]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK23-NEXT: [[ADD60:%.*]] = add nsw i32 [[TMP34]], 1 |
| // CHECK23-NEXT: store i32 [[ADD60]], ptr [[DOTOMP_IV50]], align 4, !llvm.access.group [[ACC_GRP10]] |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND52]], !llvm.loop [[LOOP11:![0-9]+]] |
| // CHECK23: omp.inner.for.end61: |
| // CHECK23-NEXT: [[TMP35:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_40]], align 4 |
| // CHECK23-NEXT: [[SUB62:%.*]] = sub nsw i32 [[TMP35]], 0 |
| // CHECK23-NEXT: [[DIV63:%.*]] = sdiv i32 [[SUB62]], 1 |
| // CHECK23-NEXT: [[MUL64:%.*]] = mul nsw i32 [[DIV63]], 1 |
| // CHECK23-NEXT: [[ADD65:%.*]] = add nsw i32 0, [[MUL64]] |
| // CHECK23-NEXT: store i32 [[ADD65]], ptr [[I51]], align 4 |
| // CHECK23-NEXT: br label [[SIMD_IF_END66]] |
| // CHECK23: simd.if.end66: |
| // CHECK23-NEXT: [[TMP36:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP36]], ptr [[DOTCAPTURE_EXPR_68]], align 4 |
| // CHECK23-NEXT: [[TMP37:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_68]], align 4 |
| // CHECK23-NEXT: [[SUB70:%.*]] = sub nsw i32 [[TMP37]], 0 |
| // CHECK23-NEXT: [[DIV71:%.*]] = sdiv i32 [[SUB70]], 1 |
| // CHECK23-NEXT: [[SUB72:%.*]] = sub nsw i32 [[DIV71]], 1 |
| // CHECK23-NEXT: store i32 [[SUB72]], ptr [[DOTCAPTURE_EXPR_69]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[DOTOMP_LB73]], align 4 |
| // CHECK23-NEXT: [[TMP38:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_69]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP38]], ptr [[DOTOMP_UB74]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[I75]], align 4 |
| // CHECK23-NEXT: [[TMP39:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_68]], align 4 |
| // CHECK23-NEXT: [[CMP76:%.*]] = icmp slt i32 0, [[TMP39]] |
| // CHECK23-NEXT: br i1 [[CMP76]], label [[SIMD_IF_THEN77:%.*]], label [[SIMD_IF_END94:%.*]] |
| // CHECK23: simd.if.then77: |
| // CHECK23-NEXT: [[TMP40:%.*]] = load i32, ptr [[DOTOMP_LB73]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP40]], ptr [[DOTOMP_IV78]], align 4 |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND80:%.*]] |
| // CHECK23: omp.inner.for.cond80: |
| // CHECK23-NEXT: [[TMP41:%.*]] = load i32, ptr [[DOTOMP_IV78]], align 4, !llvm.access.group [[ACC_GRP13:![0-9]+]] |
| // CHECK23-NEXT: [[TMP42:%.*]] = load i32, ptr [[DOTOMP_UB74]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK23-NEXT: [[CMP81:%.*]] = icmp sle i32 [[TMP41]], [[TMP42]] |
| // CHECK23-NEXT: br i1 [[CMP81]], label [[OMP_INNER_FOR_BODY82:%.*]], label [[OMP_INNER_FOR_END89:%.*]] |
| // CHECK23: omp.inner.for.body82: |
| // CHECK23-NEXT: [[TMP43:%.*]] = load i32, ptr [[DOTOMP_IV78]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK23-NEXT: [[MUL83:%.*]] = mul nsw i32 [[TMP43]], 1 |
| // CHECK23-NEXT: [[ADD84:%.*]] = add nsw i32 0, [[MUL83]] |
| // CHECK23-NEXT: store i32 [[ADD84]], ptr [[I79]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK23-NEXT: [[TMP44:%.*]] = load i32, ptr [[I79]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK23-NEXT: [[ARRAYIDX85:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i32 [[TMP44]] |
| // CHECK23-NEXT: store i32 0, ptr [[ARRAYIDX85]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK23-NEXT: br label [[OMP_BODY_CONTINUE86:%.*]] |
| // CHECK23: omp.body.continue86: |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_INC87:%.*]] |
| // CHECK23: omp.inner.for.inc87: |
| // CHECK23-NEXT: [[TMP45:%.*]] = load i32, ptr [[DOTOMP_IV78]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK23-NEXT: [[ADD88:%.*]] = add nsw i32 [[TMP45]], 1 |
| // CHECK23-NEXT: store i32 [[ADD88]], ptr [[DOTOMP_IV78]], align 4, !llvm.access.group [[ACC_GRP13]] |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND80]], !llvm.loop [[LOOP14:![0-9]+]] |
| // CHECK23: omp.inner.for.end89: |
| // CHECK23-NEXT: [[TMP46:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_68]], align 4 |
| // CHECK23-NEXT: [[SUB90:%.*]] = sub nsw i32 [[TMP46]], 0 |
| // CHECK23-NEXT: [[DIV91:%.*]] = sdiv i32 [[SUB90]], 1 |
| // CHECK23-NEXT: [[MUL92:%.*]] = mul nsw i32 [[DIV91]], 1 |
| // CHECK23-NEXT: [[ADD93:%.*]] = add nsw i32 0, [[MUL92]] |
| // CHECK23-NEXT: store i32 [[ADD93]], ptr [[I79]], align 4 |
| // CHECK23-NEXT: br label [[SIMD_IF_END94]] |
| // CHECK23: simd.if.end94: |
| // CHECK23-NEXT: [[TMP47:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP47]], ptr [[DOTCAPTURE_EXPR_95]], align 4 |
| // CHECK23-NEXT: [[TMP48:%.*]] = load i32, ptr [[N]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP48]], ptr [[DOTCAPTURE_EXPR_97]], align 4 |
| // CHECK23-NEXT: [[TMP49:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_97]], align 4 |
| // CHECK23-NEXT: [[SUB99:%.*]] = sub nsw i32 [[TMP49]], 0 |
| // CHECK23-NEXT: [[DIV100:%.*]] = sdiv i32 [[SUB99]], 1 |
| // CHECK23-NEXT: [[SUB101:%.*]] = sub nsw i32 [[DIV100]], 1 |
| // CHECK23-NEXT: store i32 [[SUB101]], ptr [[DOTCAPTURE_EXPR_98]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[DOTOMP_LB102]], align 4 |
| // CHECK23-NEXT: [[TMP50:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_98]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP50]], ptr [[DOTOMP_UB103]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[I104]], align 4 |
| // CHECK23-NEXT: [[TMP51:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_97]], align 4 |
| // CHECK23-NEXT: [[CMP105:%.*]] = icmp slt i32 0, [[TMP51]] |
| // CHECK23-NEXT: br i1 [[CMP105]], label [[SIMD_IF_THEN106:%.*]], label [[SIMD_IF_END123:%.*]] |
| // CHECK23: simd.if.then106: |
| // CHECK23-NEXT: [[TMP52:%.*]] = load i32, ptr [[DOTOMP_LB102]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP52]], ptr [[DOTOMP_IV107]], align 4 |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND109:%.*]] |
| // CHECK23: omp.inner.for.cond109: |
| // CHECK23-NEXT: [[TMP53:%.*]] = load i32, ptr [[DOTOMP_IV107]], align 4, !llvm.access.group [[ACC_GRP16:![0-9]+]] |
| // CHECK23-NEXT: [[TMP54:%.*]] = load i32, ptr [[DOTOMP_UB103]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK23-NEXT: [[CMP110:%.*]] = icmp sle i32 [[TMP53]], [[TMP54]] |
| // CHECK23-NEXT: br i1 [[CMP110]], label [[OMP_INNER_FOR_BODY111:%.*]], label [[OMP_INNER_FOR_END118:%.*]] |
| // CHECK23: omp.inner.for.body111: |
| // CHECK23-NEXT: [[TMP55:%.*]] = load i32, ptr [[DOTOMP_IV107]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK23-NEXT: [[MUL112:%.*]] = mul nsw i32 [[TMP55]], 1 |
| // CHECK23-NEXT: [[ADD113:%.*]] = add nsw i32 0, [[MUL112]] |
| // CHECK23-NEXT: store i32 [[ADD113]], ptr [[I108]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK23-NEXT: [[TMP56:%.*]] = load i32, ptr [[I108]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK23-NEXT: [[ARRAYIDX114:%.*]] = getelementptr inbounds i32, ptr [[VLA]], i32 [[TMP56]] |
| // CHECK23-NEXT: store i32 0, ptr [[ARRAYIDX114]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK23-NEXT: br label [[OMP_BODY_CONTINUE115:%.*]] |
| // CHECK23: omp.body.continue115: |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_INC116:%.*]] |
| // CHECK23: omp.inner.for.inc116: |
| // CHECK23-NEXT: [[TMP57:%.*]] = load i32, ptr [[DOTOMP_IV107]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK23-NEXT: [[ADD117:%.*]] = add nsw i32 [[TMP57]], 1 |
| // CHECK23-NEXT: store i32 [[ADD117]], ptr [[DOTOMP_IV107]], align 4, !llvm.access.group [[ACC_GRP16]] |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND109]], !llvm.loop [[LOOP17:![0-9]+]] |
| // CHECK23: omp.inner.for.end118: |
| // CHECK23-NEXT: [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_97]], align 4 |
| // CHECK23-NEXT: [[SUB119:%.*]] = sub nsw i32 [[TMP58]], 0 |
| // CHECK23-NEXT: [[DIV120:%.*]] = sdiv i32 [[SUB119]], 1 |
| // CHECK23-NEXT: [[MUL121:%.*]] = mul nsw i32 [[DIV120]], 1 |
| // CHECK23-NEXT: [[ADD122:%.*]] = add nsw i32 0, [[MUL121]] |
| // CHECK23-NEXT: store i32 [[ADD122]], ptr [[I108]], align 4 |
| // CHECK23-NEXT: br label [[SIMD_IF_END123]] |
| // CHECK23: simd.if.end123: |
| // CHECK23-NEXT: [[TMP59:%.*]] = load i32, ptr [[ARGC_ADDR]], align 4 |
| // CHECK23-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiLi10EEiT_(i32 noundef [[TMP59]]) |
| // CHECK23-NEXT: store i32 [[CALL]], ptr [[RETVAL]], align 4 |
| // CHECK23-NEXT: [[TMP60:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4 |
| // CHECK23-NEXT: call void @llvm.stackrestore.p0(ptr [[TMP60]]) |
| // CHECK23-NEXT: [[TMP61:%.*]] = load i32, ptr [[RETVAL]], align 4 |
| // CHECK23-NEXT: ret i32 [[TMP61]] |
| // |
| // |
| // CHECK23-LABEL: define {{[^@]+}}@_Z5tmainIiLi10EEiT_ |
| // CHECK23-SAME: (i32 noundef [[ARGC:%.*]]) #[[ATTR2:[0-9]+]] comdat { |
| // CHECK23-NEXT: entry: |
| // CHECK23-NEXT: [[ARGC_ADDR:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[A:%.*]] = alloca [10 x i32], align 4 |
| // CHECK23-NEXT: [[M:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[TMP:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[_TMP2:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_LB3:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_UB4:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_IV5:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I6:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[_TMP17:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_LB18:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_UB19:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_IV20:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I21:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[_TMP32:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_LB33:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_UB34:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_IV35:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I36:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTCAPTURE_EXPR_47:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[_TMP48:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_LB49:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_UB50:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[DOTOMP_IV51:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: [[I52:%.*]] = alloca i32, align 4 |
| // CHECK23-NEXT: store i32 [[ARGC]], ptr [[ARGC_ADDR]], align 4 |
| // CHECK23-NEXT: store i32 10, ptr [[M]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 4 |
| // CHECK23-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 4 |
| // CHECK23-NEXT: [[TMP0:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP0]], ptr [[DOTOMP_IV]], align 4 |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND:%.*]] |
| // CHECK23: omp.inner.for.cond: |
| // CHECK23-NEXT: [[TMP1:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19:![0-9]+]] |
| // CHECK23-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP19]] |
| // CHECK23-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP1]], [[TMP2]] |
| // CHECK23-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]] |
| // CHECK23: omp.inner.for.body: |
| // CHECK23-NEXT: [[TMP3:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]] |
| // CHECK23-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP3]], 1 |
| // CHECK23-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]] |
| // CHECK23-NEXT: store i32 [[ADD]], ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]] |
| // CHECK23-NEXT: [[TMP4:%.*]] = load i32, ptr [[I]], align 4, !llvm.access.group [[ACC_GRP19]] |
| // CHECK23-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i32 0, i32 [[TMP4]] |
| // CHECK23-NEXT: store i32 0, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP19]] |
| // CHECK23-NEXT: br label [[OMP_BODY_CONTINUE:%.*]] |
| // CHECK23: omp.body.continue: |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_INC:%.*]] |
| // CHECK23: omp.inner.for.inc: |
| // CHECK23-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]] |
| // CHECK23-NEXT: [[ADD1:%.*]] = add nsw i32 [[TMP5]], 1 |
| // CHECK23-NEXT: store i32 [[ADD1]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]] |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]] |
| // CHECK23: omp.inner.for.end: |
| // CHECK23-NEXT: store i32 10, ptr [[I]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[DOTOMP_LB3]], align 4 |
| // CHECK23-NEXT: store i32 9, ptr [[DOTOMP_UB4]], align 4 |
| // CHECK23-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_LB3]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_IV5]], align 4 |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND7:%.*]] |
| // CHECK23: omp.inner.for.cond7: |
| // CHECK23-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV5]], align 4, !llvm.access.group [[ACC_GRP22:![0-9]+]] |
| // CHECK23-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB4]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK23-NEXT: [[CMP8:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]] |
| // CHECK23-NEXT: br i1 [[CMP8]], label [[OMP_INNER_FOR_BODY9:%.*]], label [[OMP_INNER_FOR_END16:%.*]] |
| // CHECK23: omp.inner.for.body9: |
| // CHECK23-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV5]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK23-NEXT: [[MUL10:%.*]] = mul nsw i32 [[TMP9]], 1 |
| // CHECK23-NEXT: [[ADD11:%.*]] = add nsw i32 0, [[MUL10]] |
| // CHECK23-NEXT: store i32 [[ADD11]], ptr [[I6]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK23-NEXT: [[TMP10:%.*]] = load i32, ptr [[I6]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK23-NEXT: [[ARRAYIDX12:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i32 0, i32 [[TMP10]] |
| // CHECK23-NEXT: store i32 0, ptr [[ARRAYIDX12]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK23-NEXT: br label [[OMP_BODY_CONTINUE13:%.*]] |
| // CHECK23: omp.body.continue13: |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_INC14:%.*]] |
| // CHECK23: omp.inner.for.inc14: |
| // CHECK23-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV5]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK23-NEXT: [[ADD15:%.*]] = add nsw i32 [[TMP11]], 1 |
| // CHECK23-NEXT: store i32 [[ADD15]], ptr [[DOTOMP_IV5]], align 4, !llvm.access.group [[ACC_GRP22]] |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND7]], !llvm.loop [[LOOP23:![0-9]+]] |
| // CHECK23: omp.inner.for.end16: |
| // CHECK23-NEXT: store i32 10, ptr [[I6]], align 4 |
| // CHECK23-NEXT: [[TMP12:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP12]], ptr [[DOTCAPTURE_EXPR_]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[DOTOMP_LB18]], align 4 |
| // CHECK23-NEXT: store i32 9, ptr [[DOTOMP_UB19]], align 4 |
| // CHECK23-NEXT: [[TMP13:%.*]] = load i32, ptr [[DOTOMP_LB18]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP13]], ptr [[DOTOMP_IV20]], align 4 |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND22:%.*]] |
| // CHECK23: omp.inner.for.cond22: |
| // CHECK23-NEXT: [[TMP14:%.*]] = load i32, ptr [[DOTOMP_IV20]], align 4, !llvm.access.group [[ACC_GRP25:![0-9]+]] |
| // CHECK23-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB19]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK23-NEXT: [[CMP23:%.*]] = icmp sle i32 [[TMP14]], [[TMP15]] |
| // CHECK23-NEXT: br i1 [[CMP23]], label [[OMP_INNER_FOR_BODY24:%.*]], label [[OMP_INNER_FOR_END31:%.*]] |
| // CHECK23: omp.inner.for.body24: |
| // CHECK23-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_IV20]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK23-NEXT: [[MUL25:%.*]] = mul nsw i32 [[TMP16]], 1 |
| // CHECK23-NEXT: [[ADD26:%.*]] = add nsw i32 0, [[MUL25]] |
| // CHECK23-NEXT: store i32 [[ADD26]], ptr [[I21]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK23-NEXT: [[TMP17:%.*]] = load i32, ptr [[I21]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK23-NEXT: [[ARRAYIDX27:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i32 0, i32 [[TMP17]] |
| // CHECK23-NEXT: store i32 0, ptr [[ARRAYIDX27]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK23-NEXT: br label [[OMP_BODY_CONTINUE28:%.*]] |
| // CHECK23: omp.body.continue28: |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_INC29:%.*]] |
| // CHECK23: omp.inner.for.inc29: |
| // CHECK23-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV20]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK23-NEXT: [[ADD30:%.*]] = add nsw i32 [[TMP18]], 1 |
| // CHECK23-NEXT: store i32 [[ADD30]], ptr [[DOTOMP_IV20]], align 4, !llvm.access.group [[ACC_GRP25]] |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND22]], !llvm.loop [[LOOP26:![0-9]+]] |
| // CHECK23: omp.inner.for.end31: |
| // CHECK23-NEXT: store i32 10, ptr [[I21]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[DOTOMP_LB33]], align 4 |
| // CHECK23-NEXT: store i32 9, ptr [[DOTOMP_UB34]], align 4 |
| // CHECK23-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_LB33]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP19]], ptr [[DOTOMP_IV35]], align 4 |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND37:%.*]] |
| // CHECK23: omp.inner.for.cond37: |
| // CHECK23-NEXT: [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV35]], align 4, !llvm.access.group [[ACC_GRP28:![0-9]+]] |
| // CHECK23-NEXT: [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB34]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK23-NEXT: [[CMP38:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]] |
| // CHECK23-NEXT: br i1 [[CMP38]], label [[OMP_INNER_FOR_BODY39:%.*]], label [[OMP_INNER_FOR_END46:%.*]] |
| // CHECK23: omp.inner.for.body39: |
| // CHECK23-NEXT: [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV35]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK23-NEXT: [[MUL40:%.*]] = mul nsw i32 [[TMP22]], 1 |
| // CHECK23-NEXT: [[ADD41:%.*]] = add nsw i32 0, [[MUL40]] |
| // CHECK23-NEXT: store i32 [[ADD41]], ptr [[I36]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK23-NEXT: [[TMP23:%.*]] = load i32, ptr [[I36]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK23-NEXT: [[ARRAYIDX42:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i32 0, i32 [[TMP23]] |
| // CHECK23-NEXT: store i32 0, ptr [[ARRAYIDX42]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK23-NEXT: br label [[OMP_BODY_CONTINUE43:%.*]] |
| // CHECK23: omp.body.continue43: |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_INC44:%.*]] |
| // CHECK23: omp.inner.for.inc44: |
| // CHECK23-NEXT: [[TMP24:%.*]] = load i32, ptr [[DOTOMP_IV35]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK23-NEXT: [[ADD45:%.*]] = add nsw i32 [[TMP24]], 1 |
| // CHECK23-NEXT: store i32 [[ADD45]], ptr [[DOTOMP_IV35]], align 4, !llvm.access.group [[ACC_GRP28]] |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND37]], !llvm.loop [[LOOP29:![0-9]+]] |
| // CHECK23: omp.inner.for.end46: |
| // CHECK23-NEXT: store i32 10, ptr [[I36]], align 4 |
| // CHECK23-NEXT: [[TMP25:%.*]] = load i32, ptr [[M]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP25]], ptr [[DOTCAPTURE_EXPR_47]], align 4 |
| // CHECK23-NEXT: store i32 0, ptr [[DOTOMP_LB49]], align 4 |
| // CHECK23-NEXT: store i32 9, ptr [[DOTOMP_UB50]], align 4 |
| // CHECK23-NEXT: [[TMP26:%.*]] = load i32, ptr [[DOTOMP_LB49]], align 4 |
| // CHECK23-NEXT: store i32 [[TMP26]], ptr [[DOTOMP_IV51]], align 4 |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND53:%.*]] |
| // CHECK23: omp.inner.for.cond53: |
| // CHECK23-NEXT: [[TMP27:%.*]] = load i32, ptr [[DOTOMP_IV51]], align 4, !llvm.access.group [[ACC_GRP31:![0-9]+]] |
| // CHECK23-NEXT: [[TMP28:%.*]] = load i32, ptr [[DOTOMP_UB50]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK23-NEXT: [[CMP54:%.*]] = icmp sle i32 [[TMP27]], [[TMP28]] |
| // CHECK23-NEXT: br i1 [[CMP54]], label [[OMP_INNER_FOR_BODY55:%.*]], label [[OMP_INNER_FOR_END62:%.*]] |
| // CHECK23: omp.inner.for.body55: |
| // CHECK23-NEXT: [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IV51]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK23-NEXT: [[MUL56:%.*]] = mul nsw i32 [[TMP29]], 1 |
| // CHECK23-NEXT: [[ADD57:%.*]] = add nsw i32 0, [[MUL56]] |
| // CHECK23-NEXT: store i32 [[ADD57]], ptr [[I52]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK23-NEXT: [[TMP30:%.*]] = load i32, ptr [[I52]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK23-NEXT: [[ARRAYIDX58:%.*]] = getelementptr inbounds [10 x i32], ptr [[A]], i32 0, i32 [[TMP30]] |
| // CHECK23-NEXT: store i32 0, ptr [[ARRAYIDX58]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK23-NEXT: br label [[OMP_BODY_CONTINUE59:%.*]] |
| // CHECK23: omp.body.continue59: |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_INC60:%.*]] |
| // CHECK23: omp.inner.for.inc60: |
| // CHECK23-NEXT: [[TMP31:%.*]] = load i32, ptr [[DOTOMP_IV51]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK23-NEXT: [[ADD61:%.*]] = add nsw i32 [[TMP31]], 1 |
| // CHECK23-NEXT: store i32 [[ADD61]], ptr [[DOTOMP_IV51]], align 4, !llvm.access.group [[ACC_GRP31]] |
| // CHECK23-NEXT: br label [[OMP_INNER_FOR_COND53]], !llvm.loop [[LOOP32:![0-9]+]] |
| // CHECK23: omp.inner.for.end62: |
| // CHECK23-NEXT: store i32 10, ptr [[I52]], align 4 |
| // CHECK23-NEXT: ret i32 0 |
| // |
