test/OpenMP/distribute_parallel_for_codegen.cpp - llvm-project/clang - Git at Google

 // 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 _
 // Test host code gen
 // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
 // RUN: %clang_cc1 -DLAMBDA -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 -DLAMBDA -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK1
 // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK3
 // RUN: %clang_cc1 -DLAMBDA -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 -DLAMBDA -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK3

 // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
 // RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -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 -DLAMBDA -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
 // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
 // RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -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 -DLAMBDA -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"

 // RUN: %clang_cc1  -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK9
 // RUN: %clang_cc1  -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  -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK9
 // RUN: %clang_cc1  -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK11
 // RUN: %clang_cc1  -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  -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK11

 // RUN: %clang_cc1  -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
 // RUN: %clang_cc1  -fopenmp-simd -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  -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
 // RUN: %clang_cc1  -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
 // RUN: %clang_cc1  -fopenmp-simd -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  -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
 // expected-no-diagnostics
 #ifndef HEADER
 #define HEADER


 template <typename T>
 T tmain() {
   T *a, *b, *c;
   int n = 10000;
   int ch = 100;

   // no schedule clauses
   #pragma omp target
   #pragma omp teams
   #pragma omp distribute parallel for
   for (int i = 0; i < n; ++i) {
     #pragma omp cancel for
     a[i] = b[i] + c[i];
   }

   // dist_schedule: static no chunk
   #pragma omp target
   #pragma omp teams
   #pragma omp distribute parallel for dist_schedule(static)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];
   }

   // dist_schedule: static chunk
   #pragma omp target
   #pragma omp teams
   #pragma omp distribute parallel for dist_schedule(static, ch)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];
   }

   // schedule: static no chunk
   #pragma omp target
   #pragma omp teams
   #pragma omp distribute parallel for schedule(static)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];
   }

   // schedule: static chunk
   #pragma omp target
   #pragma omp teams
   #pragma omp distribute parallel for schedule(static, ch)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];
   }

   // schedule: dynamic no chunk
   #pragma omp target
   #pragma omp teams
   #pragma omp distribute parallel for schedule(dynamic)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];
   }

   // schedule: dynamic chunk
   #pragma omp target
   #pragma omp teams
   #pragma omp distribute parallel for schedule(dynamic, ch)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];
   }

   return T();
 }

 int main() {
   double *a, *b, *c;
   int n = 10000;
   int ch = 100;

 #ifdef LAMBDA
   [&]() {


     // no schedule clauses
     #pragma omp target
     #pragma omp teams

     #pragma omp distribute parallel for
     for (int i = 0; i < n; ++i) {
       a[i] = b[i] + c[i];


       // check EUB for distribute

       // initialize omp.iv

       // check exit condition

       // check that PrevLB and PrevUB are passed to the 'for'
       // check that distlb and distub are properly passed to fork_call

       // increment by stride (distInc - 'parallel for' executes the whole chunk) and latch


       // implementation of 'parallel for'


       // initialize lb and ub to PrevLB and PrevUB

       // PrevEUB is only used when 'for' has a chunked schedule, otherwise EUB is used
       // In this case we use EUB

       // initialize omp.iv

       // check exit condition

       // check that PrevLB and PrevUB are passed to the 'for'

       // check stride 1 for 'for' in 'distribute parallel for'


       [&]() {
 	a[i] = b[i] + c[i];
       }();
     }

     // dist_schedule: static no chunk (same sa default - no dist_schedule)
     #pragma omp target
     #pragma omp teams

     #pragma omp distribute parallel for dist_schedule(static)
     for (int i = 0; i < n; ++i) {
       a[i] = b[i] + c[i];


       // check EUB for distribute

       // initialize omp.iv

       // check exit condition

       // check that PrevLB and PrevUB are passed to the 'for'
       // check that distlb and distub are properly passed to fork_call

       // increment by stride (distInc - 'parallel for' executes the whole chunk) and latch


       // implementation of 'parallel for'


       // initialize lb and ub to PrevLB and PrevUB

       // PrevEUB is only used when 'for' has a chunked schedule, otherwise EUB is used
       // In this case we use EUB

       // initialize omp.iv

       // check exit condition

       // check that PrevLB and PrevUB are passed to the 'for'

       // check stride 1 for 'for' in 'distribute parallel for'

       [&]() {
 	a[i] = b[i] + c[i];
       }();
     }

     // dist_schedule: static chunk
     #pragma omp target
     #pragma omp teams

     #pragma omp distribute parallel for dist_schedule(static, ch)
     for (int i = 0; i < n; ++i) {
       a[i] = b[i] + c[i];


       // check EUB for distribute

       // initialize omp.iv

       // check exit condition

       // check that PrevLB and PrevUB are passed to the 'for'
       // check that distlb and distub are properly passed to fork_call

       // check DistInc

       // Update UB

       // Store LB in IV


       // loop exit

       // skip implementation of 'parallel for': using default scheduling and was tested above
       [&]() {
 	a[i] = b[i] + c[i];
       }();
     }

     // schedule: static no chunk
     #pragma omp target
     #pragma omp teams

     #pragma omp distribute parallel for schedule(static)
     for (int i = 0; i < n; ++i) {
       a[i] = b[i] + c[i];

       // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

       // 'parallel for' implementation is the same as the case without schedule clase (static no chunk is the default)


       // initialize lb and ub to PrevLB and PrevUB

       // PrevEUB is only used when 'for' has a chunked schedule, otherwise EUB is used
       // In this case we use EUB

       // initialize omp.iv

       // check exit condition

       // check that PrevLB and PrevUB are passed to the 'for'

       // check stride 1 for 'for' in 'distribute parallel for'


       [&]() {
 	a[i] = b[i] + c[i];
       }();
     }

     // schedule: static chunk
     #pragma omp target
     #pragma omp teams

     #pragma omp distribute parallel for schedule(static, ch)
     for (int i = 0; i < n; ++i) {
       a[i] = b[i] + c[i];
       // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

       // 'parallel for' implementation using outer and inner loops and PrevEUB

       // initialize lb and ub to PrevLB and PrevUB

       // check PrevEUB (using PrevUB instead of NumIt as upper bound)

       // initialize omp.iv (IV = LB)

       // outer loop: while (IV < UB) {


       // skip body branch

       // IV = IV + 1 and inner loop latch

       // check NextLB and NextUB


       [&]() {
 	a[i] = b[i] + c[i];
       }();
     }

     // schedule: dynamic no chunk
     #pragma omp target
     #pragma omp teams

     #pragma omp distribute parallel for schedule(dynamic)
     for (int i = 0; i < n; ++i) {
       a[i] = b[i] + c[i];
       // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

       // 'parallel for' implementation using outer and inner loops and PrevEUB

       // initialize lb and ub to PrevLB and PrevUB


       // initialize omp.iv (IV = LB)


       // skip body branch

       // IV = IV + 1 and inner loop latch

       // check NextLB and NextUB


       [&]() {
 	a[i] = b[i] + c[i];
       }();
     }

     // schedule: dynamic chunk
     #pragma omp target
     #pragma omp teams

     #pragma omp distribute parallel for schedule(dynamic, ch)
     for (int i = 0; i < n; ++i) {
       a[i] = b[i] + c[i];
       // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

       // 'parallel for' implementation using outer and inner loops and PrevEUB

       // initialize lb and ub to PrevLB and PrevUB


       // initialize omp.iv (IV = LB)


       // skip body branch

       // IV = IV + 1 and inner loop latch

       // check NextLB and NextUB


       [&]() {
 	a[i] = b[i] + c[i];
       }();
     }
   }();
   return 0;
 #else


   // no schedule clauses
   #pragma omp target
   #pragma omp teams

   #pragma omp distribute parallel for
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];


     // check EUB for distribute

     // initialize omp.iv

     // check exit condition

     // check that PrevLB and PrevUB are passed to the 'for'
     // check that distlb and distub are properly passed to fork_call

     // increment by stride (distInc - 'parallel for' executes the whole chunk) and latch


     // implementation of 'parallel for'


     // initialize lb and ub to PrevLB and PrevUB

     // PrevEUB is only used when 'for' has a chunked schedule, otherwise EUB is used
     // In this case we use EUB

     // initialize omp.iv

     // check exit condition

     // check that PrevLB and PrevUB are passed to the 'for'

     // check stride 1 for 'for' in 'distribute parallel for'

   }

   // dist_schedule: static no chunk
   #pragma omp target
   #pragma omp teams

   #pragma omp distribute parallel for dist_schedule(static)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];


     // check EUB for distribute

     // initialize omp.iv

     // check exit condition

     // check that PrevLB and PrevUB are passed to the 'for'
     // check that distlb and distub are properly passed to fork_call

     // increment by stride (distInc - 'parallel for' executes the whole chunk) and latch


     // implementation of 'parallel for'


     // initialize lb and ub to PrevLB and PrevUB

     // PrevEUB is only used when 'for' has a chunked schedule, otherwise EUB is used
     // In this case we use EUB

     // initialize omp.iv

     // check exit condition

     // check that PrevLB and PrevUB are passed to the 'for'

     // check stride 1 for 'for' in 'distribute parallel for'

   }

   // dist_schedule: static chunk
   #pragma omp target
   #pragma omp teams

   #pragma omp distribute parallel for dist_schedule(static, ch)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];

     // unlike the previous tests, in this one we have a outer and inner loop for 'distribute'

     // check EUB for distribute

     // initialize omp.iv

     // check exit condition

     // check that PrevLB and PrevUB are passed to the 'for'
     // check that distlb and distub are properly passed to fork_call

     // check DistInc

     // Update UB

     // Store LB in IV


     // loop exit

     // skip implementation of 'parallel for': using default scheduling and was tested above
   }

   // schedule: static no chunk
   #pragma omp target
   #pragma omp teams

   #pragma omp distribute parallel for schedule(static)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];

     // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

     // 'parallel for' implementation is the same as the case without schedule clase (static no chunk is the default)


     // initialize lb and ub to PrevLB and PrevUB

     // PrevEUB is only used when 'for' has a chunked schedule, otherwise EUB is used
     // In this case we use EUB

     // initialize omp.iv

     // check exit condition

     // check that PrevLB and PrevUB are passed to the 'for'

     // check stride 1 for 'for' in 'distribute parallel for'

   }

   // schedule: static chunk
   #pragma omp target
   #pragma omp teams

   #pragma omp distribute parallel for schedule(static, ch)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];
     // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

     // 'parallel for' implementation using outer and inner loops and PrevEUB

     // initialize lb and ub to PrevLB and PrevUB

     // check PrevEUB (using PrevUB instead of NumIt as upper bound)

     // initialize omp.iv (IV = LB)

     // outer loop: while (IV < UB) {


     // skip body branch

     // IV = IV + 1 and inner loop latch

     // check NextLB and NextUB


   }

   // schedule: dynamic no chunk
   #pragma omp target
   #pragma omp teams

   #pragma omp distribute parallel for schedule(dynamic)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];
     // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

     // 'parallel for' implementation using outer and inner loops and PrevEUB

     // initialize lb and ub to PrevLB and PrevUB


     // initialize omp.iv (IV = LB)


     // skip body branch

     // IV = IV + 1 and inner loop latch

     // check NextLB and NextUB


   }

   // schedule: dynamic chunk
   #pragma omp target
   #pragma omp teams

   #pragma omp distribute parallel for schedule(dynamic, ch)
   for (int i = 0; i < n; ++i) {
     a[i] = b[i] + c[i];
     // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

     // 'parallel for' implementation using outer and inner loops and PrevEUB

     // initialize lb and ub to PrevLB and PrevUB


     // initialize omp.iv (IV = LB)


     // skip body branch

     // IV = IV + 1 and inner loop latch

     // check NextLB and NextUB


   }

   return tmain<int>();
 #endif
 }

 // check code


 // check EUB for distribute

 // initialize omp.iv

 // check exit condition

 // check that PrevLB and PrevUB are passed to the 'for'
 // check that distlb and distub are properly passed to fork_call

 // increment by stride (distInc - 'parallel for' executes the whole chunk) and latch


 // implementation of 'parallel for'


 // initialize lb and ub to PrevLB and PrevUB

 // PrevEUB is only used when 'for' has a chunked schedule, otherwise EUB is used
 // In this case we use EUB

 // initialize omp.iv

 // check exit condition

 // check that PrevLB and PrevUB are passed to the 'for'

 // check stride 1 for 'for' in 'distribute parallel for'


 // check EUB for distribute

 // initialize omp.iv

 // check exit condition

 // check that PrevLB and PrevUB are passed to the 'for'
 // check that distlb and distub are properly passed to fork_call

 // increment by stride (distInc - 'parallel for' executes the whole chunk) and latch


 // implementation of 'parallel for'


 // initialize lb and ub to PrevLB and PrevUB

 // PrevEUB is only used when 'for' has a chunked schedule, otherwise EUB is used
 // In this case we use EUB

 // initialize omp.iv

 // check exit condition

 // check that PrevLB and PrevUB are passed to the 'for'

 // check stride 1 for 'for' in 'distribute parallel for'


 // unlike the previous tests, in this one we have a outer and inner loop for 'distribute'

 // check EUB for distribute

 // initialize omp.iv

 // check exit condition

 // check that PrevLB and PrevUB are passed to the 'for'
 // check that distlb and distub are properly passed to fork_call

 // check DistInc

 // Update UB

 // Store LB in IV


 // loop exit

 // skip implementation of 'parallel for': using default scheduling and was tested above


 // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

 // 'parallel for' implementation is the same as the case without schedule clase (static no chunk is the default)


 // initialize lb and ub to PrevLB and PrevUB

 // PrevEUB is only used when 'for' has a chunked schedule, otherwise EUB is used
 // In this case we use EUB

 // initialize omp.iv

 // check exit condition

 // check that PrevLB and PrevUB are passed to the 'for'

 // check stride 1 for 'for' in 'distribute parallel for'


 // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

 // 'parallel for' implementation using outer and inner loops and PrevEUB

 // initialize lb and ub to PrevLB and PrevUB

 // check PrevEUB (using PrevUB instead of NumIt as upper bound)

 // initialize omp.iv (IV = LB)

 // outer loop: while (IV < UB) {


 // skip body branch

 // IV = IV + 1 and inner loop latch

 // check NextLB and NextUB


 // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

 // 'parallel for' implementation using outer and inner loops and PrevEUB

 // initialize lb and ub to PrevLB and PrevUB


 // initialize omp.iv (IV = LB)


 // skip body branch

 // IV = IV + 1 and inner loop latch

 // check NextLB and NextUB


 // skip rest of implementation of 'distribute' as it is tested above for default dist_schedule case

 // 'parallel for' implementation using outer and inner loops and PrevEUB

 // initialize lb and ub to PrevLB and PrevUB


 // initialize omp.iv (IV = LB)


 // skip body branch

 // IV = IV + 1 and inner loop latch

 // check NextLB and NextUB


 #endif
 // CHECK1-LABEL: define {{[^@]+}}@main
 // CHECK1-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[A:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[N:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[CH:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON:%.*]], align 8
 // CHECK1-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 // CHECK1-NEXT:    store i32 10000, ptr [[N]], align 4
 // CHECK1-NEXT:    store i32 100, ptr [[CH]], align 4
 // CHECK1-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK1-NEXT:    store ptr [[N]], ptr [[TMP0]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK1-NEXT:    store ptr [[A]], ptr [[TMP1]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK1-NEXT:    store ptr [[B]], ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK1-NEXT:    store ptr [[C]], ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[REF_TMP]], i32 0, i32 4
 // CHECK1-NEXT:    store ptr [[CH]], ptr [[TMP4]], align 8
 // CHECK1-NEXT:    call void @"_ZZ4mainENK3$_0clEv"(ptr noundef nonnull align 8 dereferenceable(40) [[REF_TMP]])
 // CHECK1-NEXT:    ret i32 0
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l117
 // CHECK1-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2:[0-9]+]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3:[0-9]+]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l117.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l117.omp_outlined
 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = 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 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP7]], 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:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK1-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l117.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l117.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I4:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_0:%.*]], align 8
 // 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 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK1-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[CONV3]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK1-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i64 [[IDXPROM]]
 // CHECK1-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
 // CHECK1-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i64 [[IDXPROM7]]
 // CHECK1-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX8]], align 8
 // CHECK1-NEXT:    [[ADD9:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK1-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK1-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i64 [[IDXPROM10]]
 // CHECK1-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX11]], align 8
 // CHECK1-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP28]], align 8
 // CHECK1-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK1-NEXT:    store ptr [[I4]], ptr [[TMP29]], align 8
 // CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP30]], align 8
 // CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP31]], align 8
 // CHECK1-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(ptr noundef nonnull align 8 dereferenceable(32) [[REF_TMP]])
 // 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:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP32]], 1
 // CHECK1-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP34]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l160
 // CHECK1-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l160.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l160.omp_outlined
 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = 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 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP7]], 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:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK1-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l160.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l160.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I4:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_1:%.*]], align 8
 // 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 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK1-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[CONV3]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK1-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i64 [[IDXPROM]]
 // CHECK1-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
 // CHECK1-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i64 [[IDXPROM7]]
 // CHECK1-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX8]], align 8
 // CHECK1-NEXT:    [[ADD9:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK1-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK1-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i64 [[IDXPROM10]]
 // CHECK1-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX11]], align 8
 // CHECK1-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP28]], align 8
 // CHECK1-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK1-NEXT:    store ptr [[I4]], ptr [[TMP29]], align 8
 // CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP30]], align 8
 // CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP31]], align 8
 // CHECK1-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE0_clEv"(ptr noundef nonnull align 8 dereferenceable(32) [[REF_TMP]])
 // 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:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP32]], 1
 // CHECK1-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP34]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l202
 // CHECK1-SAME: (i64 noundef [[CH:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[CH_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store i64 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l202.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l202.omp_outlined
 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = 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 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP7]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP8]], 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:    [[TMP9:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP9]])
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK1-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP21]] to i64
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l202.omp_outlined.omp_outlined, i64 [[TMP20]], i64 [[TMP22]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]])
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
 // CHECK1-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK1-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK1-NEXT:    store i32 [[ADD7]], ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP27]], [[TMP28]]
 // CHECK1-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP30:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP29]], [[TMP30]]
 // CHECK1-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
 // CHECK1:       cond.true10:
 // CHECK1-NEXT:    [[TMP31:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    br label [[COND_END12:%.*]]
 // CHECK1:       cond.false11:
 // CHECK1-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END12]]
 // CHECK1:       cond.end12:
 // CHECK1-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP31]], [[COND_TRUE10]] ], [ [[TMP32]], [[COND_FALSE11]] ]
 // CHECK1-NEXT:    store i32 [[COND13]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP33:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP33]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP35:%.*]] = load i32, ptr [[TMP34]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP35]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l202.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I4:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_2:%.*]], align 8
 // 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 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK1-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[CONV3]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK1-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i64 [[IDXPROM]]
 // CHECK1-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
 // CHECK1-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i64 [[IDXPROM7]]
 // CHECK1-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX8]], align 8
 // CHECK1-NEXT:    [[ADD9:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK1-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK1-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i64 [[IDXPROM10]]
 // CHECK1-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX11]], align 8
 // CHECK1-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[CLASS_ANON_2]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP28]], align 8
 // CHECK1-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[CLASS_ANON_2]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK1-NEXT:    store ptr [[I4]], ptr [[TMP29]], align 8
 // CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_2]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP30]], align 8
 // CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_2]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP31]], align 8
 // CHECK1-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE1_clEv"(ptr noundef nonnull align 8 dereferenceable(32) [[REF_TMP]])
 // 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:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP32]], 1
 // CHECK1-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP34]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l235
 // CHECK1-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l235.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l235.omp_outlined
 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = 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 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP7]], 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:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK1-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l235.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l235.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I4:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_3:%.*]], align 8
 // 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 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK1-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[CONV3]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK1-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i64 [[IDXPROM]]
 // CHECK1-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
 // CHECK1-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i64 [[IDXPROM7]]
 // CHECK1-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX8]], align 8
 // CHECK1-NEXT:    [[ADD9:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK1-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK1-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK1-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i64 [[IDXPROM10]]
 // CHECK1-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX11]], align 8
 // CHECK1-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[CLASS_ANON_3]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP28]], align 8
 // CHECK1-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[CLASS_ANON_3]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK1-NEXT:    store ptr [[I4]], ptr [[TMP29]], align 8
 // CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_3]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP30]], align 8
 // CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_3]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP31]], align 8
 // CHECK1-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE2_clEv"(ptr noundef nonnull align 8 dereferenceable(32) [[REF_TMP]])
 // 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:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP32]], 1
 // CHECK1-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP34]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l267
 // CHECK1-SAME: (i64 noundef [[CH:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[CH_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store i64 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l267.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l267.omp_outlined
 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I4:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8
 // 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 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    store i32 [[TMP9]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK1-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP21]] to i64
 // CHECK1-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    store i32 [[TMP23]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l267.omp_outlined.omp_outlined, i64 [[TMP20]], i64 [[TMP22]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i64 [[TMP24]])
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
 // CHECK1-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP28:%.*]] = load i32, ptr [[TMP27]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP28]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l267.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I5:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_4:%.*]], align 8
 // 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 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK1-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[CONV4]], 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:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK1-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP12]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP10]])
 // CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK1:       omp.dispatch.cond:
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV6:%.*]] = trunc i64 [[TMP14]] to i32
 // CHECK1-NEXT:    [[CMP7:%.*]] = icmp sgt i32 [[TMP13]], [[CONV6]]
 // CHECK1-NEXT:    br i1 [[CMP7]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV8:%.*]] = trunc i64 [[TMP15]] to i32
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[CONV8]], [[COND_TRUE]] ], [ [[TMP16]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[CMP9:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK1-NEXT:    br i1 [[CMP9]], 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:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[CMP10:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
 // CHECK1-NEXT:    br i1 [[CMP10]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[I5]], align 4
 // CHECK1-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I5]], align 4
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i64 [[IDXPROM]]
 // CHECK1-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX]], align 8
 // CHECK1-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK1-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I5]], align 4
 // CHECK1-NEXT:    [[IDXPROM11:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX12:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i64 [[IDXPROM11]]
 // CHECK1-NEXT:    [[TMP28:%.*]] = load double, ptr [[ARRAYIDX12]], align 8
 // CHECK1-NEXT:    [[ADD13:%.*]] = fadd double [[TMP25]], [[TMP28]]
 // CHECK1-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK1-NEXT:    [[TMP30:%.*]] = load i32, ptr [[I5]], align 4
 // CHECK1-NEXT:    [[IDXPROM14:%.*]] = sext i32 [[TMP30]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX15:%.*]] = getelementptr inbounds double, ptr [[TMP29]], i64 [[IDXPROM14]]
 // CHECK1-NEXT:    store double [[ADD13]], ptr [[ARRAYIDX15]], align 8
 // CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_4]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP31]], align 8
 // CHECK1-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[CLASS_ANON_4]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK1-NEXT:    store ptr [[I5]], ptr [[TMP32]], align 8
 // CHECK1-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [[CLASS_ANON_4]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP33]], align 8
 // CHECK1-NEXT:    [[TMP34:%.*]] = getelementptr inbounds [[CLASS_ANON_4]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP34]], align 8
 // CHECK1-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE3_clEv"(ptr noundef nonnull align 8 dereferenceable(32) [[REF_TMP]])
 // 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:    [[TMP35:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[ADD16:%.*]] = add nsw i32 [[TMP35]], 1
 // CHECK1-NEXT:    store i32 [[ADD16]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK1:       omp.dispatch.inc:
 // CHECK1-NEXT:    [[TMP36:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    [[TMP37:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    [[ADD17:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
 // CHECK1-NEXT:    store i32 [[ADD17]], ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    [[TMP38:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP39:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    [[ADD18:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
 // CHECK1-NEXT:    store i32 [[ADD18]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK1:       omp.dispatch.end:
 // CHECK1-NEXT:    [[TMP40:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP41:%.*]] = load i32, ptr [[TMP40]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP41]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l300
 // CHECK1-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l300.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l300.omp_outlined
 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = 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 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP7]], 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:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK1-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK1-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l300.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l300.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I4:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_5:%.*]], align 8
 // 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 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK1-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[CONV3]], 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:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4
 // CHECK1-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP13]], i32 35, i32 [[TMP10]], i32 [[TMP11]], i32 1, i32 1)
 // CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK1:       omp.dispatch.cond:
 // CHECK1-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[TMP14]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP15]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP16]], 0
 // CHECK1-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK1:       omp.dispatch.body:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12:![0-9]+]]
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP2]], align 8, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[TMP22:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP22]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP21]], i64 [[IDXPROM]]
 // CHECK1-NEXT:    [[TMP23:%.*]] = load double, ptr [[ARRAYIDX]], align 8, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[TMP3]], align 8, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[TMP25:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[IDXPROM6:%.*]] = sext i32 [[TMP25]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds double, ptr [[TMP24]], i64 [[IDXPROM6]]
 // CHECK1-NEXT:    [[TMP26:%.*]] = load double, ptr [[ARRAYIDX7]], align 8, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[ADD8:%.*]] = fadd double [[TMP23]], [[TMP26]]
 // CHECK1-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[TMP1]], align 8, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[TMP28:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[IDXPROM9:%.*]] = sext i32 [[TMP28]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds double, ptr [[TMP27]], i64 [[IDXPROM9]]
 // CHECK1-NEXT:    store double [[ADD8]], ptr [[ARRAYIDX10]], align 8, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[CLASS_ANON_5]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP29]], align 8, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_5]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK1-NEXT:    store ptr [[I4]], ptr [[TMP30]], align 8, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_5]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP31]], align 8, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[CLASS_ANON_5]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP32]], align 8, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE4_clEv"(ptr noundef nonnull align 8 dereferenceable(32) [[REF_TMP]]), !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:    [[TMP33:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP12]]
 // CHECK1-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP33]], 1
 // CHECK1-NEXT:    store i32 [[ADD11]], 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_DISPATCH_INC:%.*]]
 // CHECK1:       omp.dispatch.inc:
 // CHECK1-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK1:       omp.dispatch.end:
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l329
 // CHECK1-SAME: (i64 noundef [[CH:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[CH_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    store i64 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l329.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l329.omp_outlined
 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR2]] {
 // CHECK1-NEXT:  entry:
 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I4:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8
 // 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 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK1-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    store i32 [[TMP9]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK1-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK1:       cond.true:
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    br label [[COND_END:%.*]]
 // CHECK1:       cond.false:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    br label [[COND_END]]
 // CHECK1:       cond.end:
 // CHECK1-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK1-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK1-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK1-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK1-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP21]] to i64
 // CHECK1-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK1-NEXT:    store i32 [[TMP23]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK1-NEXT:    [[TMP24:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
 // CHECK1-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l329.omp_outlined.omp_outlined, i64 [[TMP20]], i64 [[TMP22]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i64 [[TMP24]])
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK1:       omp.inner.for.inc:
 // CHECK1-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK1:       omp.inner.for.end:
 // CHECK1-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK1:       omp.loop.exit:
 // CHECK1-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP28:%.*]] = load i32, ptr [[TMP27]], align 4
 // CHECK1-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP28]])
 // CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l329.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8
 // CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[I:%.*]] = 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:    [[I5:%.*]] = alloca i32, align 4
 // CHECK1-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_6:%.*]], align 8
 // 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 [[N]], ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK1-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK1-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK1-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK1-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK1-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK1:       omp.precond.then:
 // CHECK1-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK1-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK1-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK1-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK1-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[CONV4]], 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:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK1-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK1-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP14:%.*]] = load i32, ptr [[TMP13]], align 4
 // CHECK1-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP14]], i32 35, i32 [[TMP11]], i32 [[TMP12]], i32 1, i32 [[TMP10]])
 // CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK1:       omp.dispatch.cond:
 // CHECK1-NEXT:    [[TMP15:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK1-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP15]], align 4
 // CHECK1-NEXT:    [[TMP17:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP16]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP17]], 0
 // CHECK1-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK1:       omp.dispatch.body:
 // CHECK1-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK1-NEXT:    store i32 [[TMP18]], ptr [[DOTOMP_IV]], align 4
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK1:       omp.inner.for.cond:
 // CHECK1-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP15:![0-9]+]]
 // CHECK1-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP19]], [[TMP20]]
 // CHECK1-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK1:       omp.inner.for.body:
 // CHECK1-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP21]], 1
 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[TMP22:%.*]] = load ptr, ptr [[TMP2]], align 8, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[TMP23:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP22]], i64 [[IDXPROM]]
 // CHECK1-NEXT:    [[TMP24:%.*]] = load double, ptr [[ARRAYIDX]], align 8, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[TMP3]], align 8, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[TMP26:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP26]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP25]], i64 [[IDXPROM7]]
 // CHECK1-NEXT:    [[TMP27:%.*]] = load double, ptr [[ARRAYIDX8]], align 8, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[ADD9:%.*]] = fadd double [[TMP24]], [[TMP27]]
 // CHECK1-NEXT:    [[TMP28:%.*]] = load ptr, ptr [[TMP1]], align 8, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[TMP29:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP29]] to i64
 // CHECK1-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds double, ptr [[TMP28]], i64 [[IDXPROM10]]
 // CHECK1-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX11]], align 8, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_6]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP30]], align 8, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_6]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK1-NEXT:    store ptr [[I5]], ptr [[TMP31]], align 8, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[CLASS_ANON_6]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP32]], align 8, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [[CLASS_ANON_6]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK1-NEXT:    store ptr [[TMP3]], ptr [[TMP33]], align 8, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE5_clEv"(ptr noundef nonnull align 8 dereferenceable(32) [[REF_TMP]]), !llvm.access.group [[ACC_GRP15]]
 // 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:    [[TMP34:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP34]], 1
 // CHECK1-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP15]]
 // CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP16:![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:    br label [[OMP_PRECOND_END]]
 // CHECK1:       omp.precond.end:
 // CHECK1-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@main
 // CHECK3-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[A:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[N:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[CH:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON:%.*]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 // CHECK3-NEXT:    store i32 10000, ptr [[N]], align 4
 // CHECK3-NEXT:    store i32 100, ptr [[CH]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK3-NEXT:    store ptr [[N]], ptr [[TMP0]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK3-NEXT:    store ptr [[A]], ptr [[TMP1]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK3-NEXT:    store ptr [[B]], ptr [[TMP2]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK3-NEXT:    store ptr [[C]], ptr [[TMP3]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[CLASS_ANON]], ptr [[REF_TMP]], i32 0, i32 4
 // CHECK3-NEXT:    store ptr [[CH]], ptr [[TMP4]], align 4
 // CHECK3-NEXT:    call void @"_ZZ4mainENK3$_0clEv"(ptr noundef nonnull align 4 dereferenceable(20) [[REF_TMP]])
 // CHECK3-NEXT:    ret i32 0
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l117
 // CHECK3-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2:[0-9]+]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3:[0-9]+]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l117.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l117.omp_outlined
 // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = 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 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP7]], 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:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK3-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l117.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l117.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_0:%.*]], 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 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP9]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK3-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i32 [[TMP21]]
 // CHECK3-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 4
 // CHECK3-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK3-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK3-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX6]], align 4
 // CHECK3-NEXT:    [[ADD7:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK3-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK3-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK3-NEXT:    store double [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // CHECK3-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[TMP28]], align 4
 // CHECK3-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK3-NEXT:    store ptr [[I3]], ptr [[TMP29]], align 4
 // CHECK3-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK3-NEXT:    store ptr [[TMP2]], ptr [[TMP30]], align 4
 // CHECK3-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK3-NEXT:    store ptr [[TMP3]], ptr [[TMP31]], align 4
 // CHECK3-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(ptr noundef nonnull align 4 dereferenceable(16) [[REF_TMP]])
 // 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:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP32]], 1
 // CHECK3-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP34]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l160
 // CHECK3-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l160.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l160.omp_outlined
 // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = 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 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP7]], 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:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK3-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l160.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l160.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_1:%.*]], 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 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP9]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK3-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i32 [[TMP21]]
 // CHECK3-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 4
 // CHECK3-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK3-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK3-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX6]], align 4
 // CHECK3-NEXT:    [[ADD7:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK3-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK3-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK3-NEXT:    store double [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // CHECK3-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[TMP28]], align 4
 // CHECK3-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK3-NEXT:    store ptr [[I3]], ptr [[TMP29]], align 4
 // CHECK3-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK3-NEXT:    store ptr [[TMP2]], ptr [[TMP30]], align 4
 // CHECK3-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK3-NEXT:    store ptr [[TMP3]], ptr [[TMP31]], align 4
 // CHECK3-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE0_clEv"(ptr noundef nonnull align 4 dereferenceable(16) [[REF_TMP]])
 // 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:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP32]], 1
 // CHECK3-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP34]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l202
 // CHECK3-SAME: (i32 noundef [[CH:%.*]], i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[CH_ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store i32 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l202.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l202.omp_outlined
 // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = 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 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP7]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP8]], 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:    [[TMP9:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP9]])
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l202.omp_outlined.omp_outlined, i32 [[TMP19]], i32 [[TMP20]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]])
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
 // CHECK3-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK3-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK3-NEXT:    store i32 [[ADD7]], ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK3-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
 // CHECK3-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
 // CHECK3:       cond.true10:
 // CHECK3-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    br label [[COND_END12:%.*]]
 // CHECK3:       cond.false11:
 // CHECK3-NEXT:    [[TMP30:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END12]]
 // CHECK3:       cond.end12:
 // CHECK3-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP29]], [[COND_TRUE10]] ], [ [[TMP30]], [[COND_FALSE11]] ]
 // CHECK3-NEXT:    store i32 [[COND13]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP31:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP31]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP32:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP33:%.*]] = load i32, ptr [[TMP32]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP33]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l202.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_2:%.*]], 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 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP9]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK3-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i32 [[TMP21]]
 // CHECK3-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 4
 // CHECK3-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK3-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK3-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX6]], align 4
 // CHECK3-NEXT:    [[ADD7:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK3-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK3-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK3-NEXT:    store double [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // CHECK3-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[CLASS_ANON_2]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[TMP28]], align 4
 // CHECK3-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[CLASS_ANON_2]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK3-NEXT:    store ptr [[I3]], ptr [[TMP29]], align 4
 // CHECK3-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_2]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK3-NEXT:    store ptr [[TMP2]], ptr [[TMP30]], align 4
 // CHECK3-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_2]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK3-NEXT:    store ptr [[TMP3]], ptr [[TMP31]], align 4
 // CHECK3-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE1_clEv"(ptr noundef nonnull align 4 dereferenceable(16) [[REF_TMP]])
 // 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:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP32]], 1
 // CHECK3-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP34]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l235
 // CHECK3-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l235.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l235.omp_outlined
 // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = 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 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP7]], 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:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK3-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l235.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l235.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_3:%.*]], 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 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP9]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK3-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i32 [[TMP21]]
 // CHECK3-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 4
 // CHECK3-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK3-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK3-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX6]], align 4
 // CHECK3-NEXT:    [[ADD7:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK3-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK3-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK3-NEXT:    store double [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // CHECK3-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[CLASS_ANON_3]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[TMP28]], align 4
 // CHECK3-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[CLASS_ANON_3]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK3-NEXT:    store ptr [[I3]], ptr [[TMP29]], align 4
 // CHECK3-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_3]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK3-NEXT:    store ptr [[TMP2]], ptr [[TMP30]], align 4
 // CHECK3-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_3]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK3-NEXT:    store ptr [[TMP3]], ptr [[TMP31]], align 4
 // CHECK3-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE2_clEv"(ptr noundef nonnull align 4 dereferenceable(16) [[REF_TMP]])
 // 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:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP32]], 1
 // CHECK3-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP34]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l267
 // CHECK3-SAME: (i32 noundef [[CH:%.*]], i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[CH_ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store i32 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l267.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l267.omp_outlined
 // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I4:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = 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 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    store i32 [[TMP9]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK3-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK3-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l267.omp_outlined.omp_outlined, i32 [[TMP19]], i32 [[TMP20]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i32 [[TMP22]])
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
 // CHECK3-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP26:%.*]] = load i32, ptr [[TMP25]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP26]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l267.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I4:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_4:%.*]], 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 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP9]], 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:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP12]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP10]])
 // CHECK3-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK3:       omp.dispatch.cond:
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP13]], [[TMP14]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP15]], [[COND_TRUE]] ], [ [[TMP16]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK3-NEXT:    br i1 [[CMP6]], 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:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[CMP7:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
 // CHECK3-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK3-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK3-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK3-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX]], align 4
 // CHECK3-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK3-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK3-NEXT:    [[TMP28:%.*]] = load double, ptr [[ARRAYIDX8]], align 4
 // CHECK3-NEXT:    [[ADD9:%.*]] = fadd double [[TMP25]], [[TMP28]]
 // CHECK3-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK3-NEXT:    [[TMP30:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK3-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds double, ptr [[TMP29]], i32 [[TMP30]]
 // CHECK3-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX10]], align 4
 // CHECK3-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_4]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[TMP31]], align 4
 // CHECK3-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[CLASS_ANON_4]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK3-NEXT:    store ptr [[I4]], ptr [[TMP32]], align 4
 // CHECK3-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [[CLASS_ANON_4]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK3-NEXT:    store ptr [[TMP2]], ptr [[TMP33]], align 4
 // CHECK3-NEXT:    [[TMP34:%.*]] = getelementptr inbounds [[CLASS_ANON_4]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK3-NEXT:    store ptr [[TMP3]], ptr [[TMP34]], align 4
 // CHECK3-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE3_clEv"(ptr noundef nonnull align 4 dereferenceable(16) [[REF_TMP]])
 // 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:    [[TMP35:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP35]], 1
 // CHECK3-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK3:       omp.dispatch.inc:
 // CHECK3-NEXT:    [[TMP36:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    [[TMP37:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP36]], [[TMP37]]
 // CHECK3-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    [[TMP38:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP39:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP38]], [[TMP39]]
 // CHECK3-NEXT:    store i32 [[ADD13]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK3:       omp.dispatch.end:
 // CHECK3-NEXT:    [[TMP40:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP41:%.*]] = load i32, ptr [[TMP40]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP41]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l300
 // CHECK3-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l300.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l300.omp_outlined
 // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = 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 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP7]], 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:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK3-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK3-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l300.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l300.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I3:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_5:%.*]], 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 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP9]], 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:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4
 // CHECK3-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP13]], i32 35, i32 [[TMP10]], i32 [[TMP11]], i32 1, i32 1)
 // CHECK3-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK3:       omp.dispatch.cond:
 // CHECK3-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[TMP14]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP15]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK3-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP16]], 0
 // CHECK3-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK3:       omp.dispatch.body:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13:![0-9]+]]
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK3-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP2]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[TMP22:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP21]], i32 [[TMP22]]
 // CHECK3-NEXT:    [[TMP23:%.*]] = load double, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[TMP3]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[TMP25:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds double, ptr [[TMP24]], i32 [[TMP25]]
 // CHECK3-NEXT:    [[TMP26:%.*]] = load double, ptr [[ARRAYIDX5]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[ADD6:%.*]] = fadd double [[TMP23]], [[TMP26]]
 // CHECK3-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[TMP1]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[TMP28:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds double, ptr [[TMP27]], i32 [[TMP28]]
 // CHECK3-NEXT:    store double [[ADD6]], ptr [[ARRAYIDX7]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[CLASS_ANON_5]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[TMP29]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_5]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK3-NEXT:    store ptr [[I3]], ptr [[TMP30]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_5]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK3-NEXT:    store ptr [[TMP2]], ptr [[TMP31]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[CLASS_ANON_5]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK3-NEXT:    store ptr [[TMP3]], ptr [[TMP32]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE4_clEv"(ptr noundef nonnull align 4 dereferenceable(16) [[REF_TMP]]), !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:    [[TMP33:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP13]]
 // CHECK3-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP33]], 1
 // CHECK3-NEXT:    store i32 [[ADD8]], 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_DISPATCH_INC:%.*]]
 // CHECK3:       omp.dispatch.inc:
 // CHECK3-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK3:       omp.dispatch.end:
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l329
 // CHECK3-SAME: (i32 noundef [[CH:%.*]], i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[CH_ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    store i32 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l329.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l329.omp_outlined
 // CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR2]] {
 // CHECK3-NEXT:  entry:
 // CHECK3-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I4:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = 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 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK3-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    store i32 [[TMP9]], 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:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK3:       cond.true:
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    br label [[COND_END:%.*]]
 // CHECK3:       cond.false:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    br label [[COND_END]]
 // CHECK3:       cond.end:
 // CHECK3-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK3-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK3-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK3-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK3-NEXT:    store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK3-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK3-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l329.omp_outlined.omp_outlined, i32 [[TMP19]], i32 [[TMP20]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i32 [[TMP22]])
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK3:       omp.inner.for.inc:
 // CHECK3-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK3:       omp.inner.for.end:
 // CHECK3-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK3:       omp.loop.exit:
 // CHECK3-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP26:%.*]] = load i32, ptr [[TMP25]], align 4
 // CHECK3-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP26]])
 // CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l329.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 nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR2]] {
 // 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:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[I:%.*]] = 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:    [[I4:%.*]] = alloca i32, align 4
 // CHECK3-NEXT:    [[REF_TMP:%.*]] = alloca [[CLASS_ANON_6:%.*]], 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 [[N]], ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK3-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK3-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK3-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK3-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK3-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK3:       omp.precond.then:
 // CHECK3-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK3-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK3-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP9]], 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:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK3-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP14:%.*]] = load i32, ptr [[TMP13]], align 4
 // CHECK3-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP14]], i32 35, i32 [[TMP11]], i32 [[TMP12]], i32 1, i32 [[TMP10]])
 // CHECK3-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK3:       omp.dispatch.cond:
 // CHECK3-NEXT:    [[TMP15:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK3-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP15]], align 4
 // CHECK3-NEXT:    [[TMP17:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP16]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK3-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP17]], 0
 // CHECK3-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK3:       omp.dispatch.body:
 // CHECK3-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK3-NEXT:    store i32 [[TMP18]], ptr [[DOTOMP_IV]], align 4
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK3:       omp.inner.for.cond:
 // CHECK3-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16:![0-9]+]]
 // CHECK3-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP19]], [[TMP20]]
 // CHECK3-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK3:       omp.inner.for.body:
 // CHECK3-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP21]], 1
 // CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK3-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[TMP22:%.*]] = load ptr, ptr [[TMP2]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[TMP23:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP22]], i32 [[TMP23]]
 // CHECK3-NEXT:    [[TMP24:%.*]] = load double, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[TMP3]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[TMP26:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds double, ptr [[TMP25]], i32 [[TMP26]]
 // CHECK3-NEXT:    [[TMP27:%.*]] = load double, ptr [[ARRAYIDX6]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[ADD7:%.*]] = fadd double [[TMP24]], [[TMP27]]
 // CHECK3-NEXT:    [[TMP28:%.*]] = load ptr, ptr [[TMP1]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[TMP29:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP28]], i32 [[TMP29]]
 // CHECK3-NEXT:    store double [[ADD7]], ptr [[ARRAYIDX8]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[CLASS_ANON_6]], ptr [[REF_TMP]], i32 0, i32 0
 // CHECK3-NEXT:    store ptr [[TMP1]], ptr [[TMP30]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[CLASS_ANON_6]], ptr [[REF_TMP]], i32 0, i32 1
 // CHECK3-NEXT:    store ptr [[I4]], ptr [[TMP31]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[CLASS_ANON_6]], ptr [[REF_TMP]], i32 0, i32 2
 // CHECK3-NEXT:    store ptr [[TMP2]], ptr [[TMP32]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [[CLASS_ANON_6]], ptr [[REF_TMP]], i32 0, i32 3
 // CHECK3-NEXT:    store ptr [[TMP3]], ptr [[TMP33]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    call void @"_ZZZ4mainENK3$_0clEvENKUlvE5_clEv"(ptr noundef nonnull align 4 dereferenceable(16) [[REF_TMP]]), !llvm.access.group [[ACC_GRP16]]
 // 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:    [[TMP34:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP34]], 1
 // CHECK3-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP16]]
 // CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP17:![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:    br label [[OMP_PRECOND_END]]
 // CHECK3:       omp.precond.end:
 // CHECK3-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@main
 // CHECK9-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[A:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[CH:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
 // CHECK9-NEXT:    [[N_CASTED3:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS4:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS5:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS6:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP7:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_8:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_9:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    [[CH_CASTED:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_CASTED17:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS18:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS19:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS20:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP21:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_22:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_23:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    [[N_CASTED31:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS32:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS33:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS34:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP35:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_36:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_37:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    [[CH_CASTED45:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_CASTED46:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS47:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS48:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS49:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP50:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_51:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_52:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    [[N_CASTED60:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS61:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS62:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS63:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP64:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_65:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_66:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    [[CH_CASTED74:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_CASTED75:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS76:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS77:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS78:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP79:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_80:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_81:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 // CHECK9-NEXT:    store i32 10000, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 100, ptr [[CH]], align 4
 // CHECK9-NEXT:    [[TMP0:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP0]], ptr [[N_CASTED]], align 4
 // CHECK9-NEXT:    [[TMP1:%.*]] = load i64, ptr [[N_CASTED]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP1]], ptr [[TMP5]], align 8
 // CHECK9-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP1]], ptr [[TMP6]], align 8
 // CHECK9-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP7]], align 8
 // CHECK9-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP2]], ptr [[TMP8]], align 8
 // CHECK9-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP2]], ptr [[TMP9]], align 8
 // CHECK9-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP10]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP3]], ptr [[TMP11]], align 8
 // CHECK9-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP3]], ptr [[TMP12]], align 8
 // CHECK9-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP13]], align 8
 // CHECK9-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP4]], ptr [[TMP14]], align 8
 // CHECK9-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP4]], ptr [[TMP15]], align 8
 // CHECK9-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP16]], align 8
 // CHECK9-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP19]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP20]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], 1
 // CHECK9-NEXT:    [[TMP22:%.*]] = zext i32 [[ADD]] to i64
 // CHECK9-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP23]], align 4
 // CHECK9-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 4, ptr [[TMP24]], align 4
 // CHECK9-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP17]], ptr [[TMP25]], align 8
 // CHECK9-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP18]], ptr [[TMP26]], align 8
 // CHECK9-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes, ptr [[TMP27]], align 8
 // CHECK9-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes, ptr [[TMP28]], align 8
 // CHECK9-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP29]], align 8
 // CHECK9-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP30]], align 8
 // CHECK9-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP22]], ptr [[TMP31]], align 8
 // CHECK9-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP32]], align 8
 // CHECK9-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
 // CHECK9-NEXT:    [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
 // CHECK9-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP35]], align 4
 // CHECK9-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_l369.region_id, ptr [[KERNEL_ARGS]])
 // CHECK9-NEXT:    [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
 // CHECK9-NEXT:    br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 // CHECK9:       omp_offload.failed:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369(i64 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2:[0-9]+]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 // CHECK9:       omp_offload.cont:
 // CHECK9-NEXT:    [[TMP38:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
 // CHECK9-NEXT:    [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
 // CHECK9-NEXT:    [[TMP40:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP41:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP42:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP39]], ptr [[TMP43]], align 8
 // CHECK9-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP39]], ptr [[TMP44]], align 8
 // CHECK9-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP45]], align 8
 // CHECK9-NEXT:    [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP40]], ptr [[TMP46]], align 8
 // CHECK9-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP40]], ptr [[TMP47]], align 8
 // CHECK9-NEXT:    [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP48]], align 8
 // CHECK9-NEXT:    [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP41]], ptr [[TMP49]], align 8
 // CHECK9-NEXT:    [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP41]], ptr [[TMP50]], align 8
 // CHECK9-NEXT:    [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP51]], align 8
 // CHECK9-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP42]], ptr [[TMP52]], align 8
 // CHECK9-NEXT:    [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP42]], ptr [[TMP53]], align 8
 // CHECK9-NEXT:    [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP54]], align 8
 // CHECK9-NEXT:    [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP57:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
 // CHECK9-NEXT:    [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
 // CHECK9-NEXT:    [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
 // CHECK9-NEXT:    [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
 // CHECK9-NEXT:    [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
 // CHECK9-NEXT:    store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
 // CHECK9-NEXT:    [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
 // CHECK9-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
 // CHECK9-NEXT:    [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
 // CHECK9-NEXT:    [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP61]], align 4
 // CHECK9-NEXT:    [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 4, ptr [[TMP62]], align 4
 // CHECK9-NEXT:    [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP55]], ptr [[TMP63]], align 8
 // CHECK9-NEXT:    [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP56]], ptr [[TMP64]], align 8
 // CHECK9-NEXT:    [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.1, ptr [[TMP65]], align 8
 // CHECK9-NEXT:    [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.2, ptr [[TMP66]], align 8
 // CHECK9-NEXT:    [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP67]], align 8
 // CHECK9-NEXT:    [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP68]], align 8
 // CHECK9-NEXT:    [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP60]], ptr [[TMP69]], align 8
 // CHECK9-NEXT:    [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP70]], align 8
 // CHECK9-NEXT:    [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
 // CHECK9-NEXT:    [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP72]], align 4
 // CHECK9-NEXT:    [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP73]], align 4
 // CHECK9-NEXT:    [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.region_id, ptr [[KERNEL_ARGS14]])
 // CHECK9-NEXT:    [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
 // CHECK9-NEXT:    br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
 // CHECK9:       omp_offload.failed15:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408(i64 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT16]]
 // CHECK9:       omp_offload.cont16:
 // CHECK9-NEXT:    [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK9-NEXT:    store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
 // CHECK9-NEXT:    [[TMP77:%.*]] = load i64, ptr [[CH_CASTED]], align 8
 // CHECK9-NEXT:    [[TMP78:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
 // CHECK9-NEXT:    [[TMP79:%.*]] = load i64, ptr [[N_CASTED17]], align 8
 // CHECK9-NEXT:    [[TMP80:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP81:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP82:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP77]], ptr [[TMP83]], align 8
 // CHECK9-NEXT:    [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP77]], ptr [[TMP84]], align 8
 // CHECK9-NEXT:    [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP85]], align 8
 // CHECK9-NEXT:    [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP79]], ptr [[TMP86]], align 8
 // CHECK9-NEXT:    [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP79]], ptr [[TMP87]], align 8
 // CHECK9-NEXT:    [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP88]], align 8
 // CHECK9-NEXT:    [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP80]], ptr [[TMP89]], align 8
 // CHECK9-NEXT:    [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP80]], ptr [[TMP90]], align 8
 // CHECK9-NEXT:    [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP91]], align 8
 // CHECK9-NEXT:    [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP81]], ptr [[TMP92]], align 8
 // CHECK9-NEXT:    [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP81]], ptr [[TMP93]], align 8
 // CHECK9-NEXT:    [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP94]], align 8
 // CHECK9-NEXT:    [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP82]], ptr [[TMP95]], align 8
 // CHECK9-NEXT:    [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP82]], ptr [[TMP96]], align 8
 // CHECK9-NEXT:    [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 4
 // CHECK9-NEXT:    store ptr null, ptr [[TMP97]], align 8
 // CHECK9-NEXT:    [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP100:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
 // CHECK9-NEXT:    [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
 // CHECK9-NEXT:    [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
 // CHECK9-NEXT:    [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
 // CHECK9-NEXT:    [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
 // CHECK9-NEXT:    store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
 // CHECK9-NEXT:    [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
 // CHECK9-NEXT:    [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
 // CHECK9-NEXT:    [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
 // CHECK9-NEXT:    [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP104]], align 4
 // CHECK9-NEXT:    [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 5, ptr [[TMP105]], align 4
 // CHECK9-NEXT:    [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP98]], ptr [[TMP106]], align 8
 // CHECK9-NEXT:    [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP99]], ptr [[TMP107]], align 8
 // CHECK9-NEXT:    [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.3, ptr [[TMP108]], align 8
 // CHECK9-NEXT:    [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.4, ptr [[TMP109]], align 8
 // CHECK9-NEXT:    [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP110]], align 8
 // CHECK9-NEXT:    [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP111]], align 8
 // CHECK9-NEXT:    [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP103]], ptr [[TMP112]], align 8
 // CHECK9-NEXT:    [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP113]], align 8
 // CHECK9-NEXT:    [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
 // CHECK9-NEXT:    [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP115]], align 4
 // CHECK9-NEXT:    [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP116]], align 4
 // CHECK9-NEXT:    [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.region_id, ptr [[KERNEL_ARGS28]])
 // CHECK9-NEXT:    [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
 // CHECK9-NEXT:    br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
 // CHECK9:       omp_offload.failed29:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447(i64 [[TMP77]], i64 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT30]]
 // CHECK9:       omp_offload.cont30:
 // CHECK9-NEXT:    [[TMP119:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
 // CHECK9-NEXT:    [[TMP120:%.*]] = load i64, ptr [[N_CASTED31]], align 8
 // CHECK9-NEXT:    [[TMP121:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP122:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP123:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP120]], ptr [[TMP124]], align 8
 // CHECK9-NEXT:    [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP120]], ptr [[TMP125]], align 8
 // CHECK9-NEXT:    [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP126]], align 8
 // CHECK9-NEXT:    [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP121]], ptr [[TMP127]], align 8
 // CHECK9-NEXT:    [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP121]], ptr [[TMP128]], align 8
 // CHECK9-NEXT:    [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP129]], align 8
 // CHECK9-NEXT:    [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP122]], ptr [[TMP130]], align 8
 // CHECK9-NEXT:    [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP122]], ptr [[TMP131]], align 8
 // CHECK9-NEXT:    [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP132]], align 8
 // CHECK9-NEXT:    [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP123]], ptr [[TMP133]], align 8
 // CHECK9-NEXT:    [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP123]], ptr [[TMP134]], align 8
 // CHECK9-NEXT:    [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP135]], align 8
 // CHECK9-NEXT:    [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP138:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
 // CHECK9-NEXT:    [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
 // CHECK9-NEXT:    [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
 // CHECK9-NEXT:    [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
 // CHECK9-NEXT:    [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
 // CHECK9-NEXT:    store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
 // CHECK9-NEXT:    [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
 // CHECK9-NEXT:    [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
 // CHECK9-NEXT:    [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
 // CHECK9-NEXT:    [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP142]], align 4
 // CHECK9-NEXT:    [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 4, ptr [[TMP143]], align 4
 // CHECK9-NEXT:    [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP136]], ptr [[TMP144]], align 8
 // CHECK9-NEXT:    [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP137]], ptr [[TMP145]], align 8
 // CHECK9-NEXT:    [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.5, ptr [[TMP146]], align 8
 // CHECK9-NEXT:    [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.6, ptr [[TMP147]], align 8
 // CHECK9-NEXT:    [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP148]], align 8
 // CHECK9-NEXT:    [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP149]], align 8
 // CHECK9-NEXT:    [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP141]], ptr [[TMP150]], align 8
 // CHECK9-NEXT:    [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP151]], align 8
 // CHECK9-NEXT:    [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
 // CHECK9-NEXT:    [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP153]], align 4
 // CHECK9-NEXT:    [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP154]], align 4
 // CHECK9-NEXT:    [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.region_id, ptr [[KERNEL_ARGS42]])
 // CHECK9-NEXT:    [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
 // CHECK9-NEXT:    br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
 // CHECK9:       omp_offload.failed43:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478(i64 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT44]]
 // CHECK9:       omp_offload.cont44:
 // CHECK9-NEXT:    [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK9-NEXT:    store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
 // CHECK9-NEXT:    [[TMP158:%.*]] = load i64, ptr [[CH_CASTED45]], align 8
 // CHECK9-NEXT:    [[TMP159:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
 // CHECK9-NEXT:    [[TMP160:%.*]] = load i64, ptr [[N_CASTED46]], align 8
 // CHECK9-NEXT:    [[TMP161:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP162:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP163:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP158]], ptr [[TMP164]], align 8
 // CHECK9-NEXT:    [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP158]], ptr [[TMP165]], align 8
 // CHECK9-NEXT:    [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP166]], align 8
 // CHECK9-NEXT:    [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP160]], ptr [[TMP167]], align 8
 // CHECK9-NEXT:    [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP160]], ptr [[TMP168]], align 8
 // CHECK9-NEXT:    [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP169]], align 8
 // CHECK9-NEXT:    [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP161]], ptr [[TMP170]], align 8
 // CHECK9-NEXT:    [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP161]], ptr [[TMP171]], align 8
 // CHECK9-NEXT:    [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP172]], align 8
 // CHECK9-NEXT:    [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP162]], ptr [[TMP173]], align 8
 // CHECK9-NEXT:    [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP162]], ptr [[TMP174]], align 8
 // CHECK9-NEXT:    [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP175]], align 8
 // CHECK9-NEXT:    [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP163]], ptr [[TMP176]], align 8
 // CHECK9-NEXT:    [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP163]], ptr [[TMP177]], align 8
 // CHECK9-NEXT:    [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 4
 // CHECK9-NEXT:    store ptr null, ptr [[TMP178]], align 8
 // CHECK9-NEXT:    [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP181:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
 // CHECK9-NEXT:    [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
 // CHECK9-NEXT:    [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
 // CHECK9-NEXT:    [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
 // CHECK9-NEXT:    [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
 // CHECK9-NEXT:    store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
 // CHECK9-NEXT:    [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
 // CHECK9-NEXT:    [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
 // CHECK9-NEXT:    [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
 // CHECK9-NEXT:    [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP185]], align 4
 // CHECK9-NEXT:    [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 5, ptr [[TMP186]], align 4
 // CHECK9-NEXT:    [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP179]], ptr [[TMP187]], align 8
 // CHECK9-NEXT:    [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP180]], ptr [[TMP188]], align 8
 // CHECK9-NEXT:    [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.7, ptr [[TMP189]], align 8
 // CHECK9-NEXT:    [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.8, ptr [[TMP190]], align 8
 // CHECK9-NEXT:    [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP191]], align 8
 // CHECK9-NEXT:    [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP192]], align 8
 // CHECK9-NEXT:    [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP184]], ptr [[TMP193]], align 8
 // CHECK9-NEXT:    [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP194]], align 8
 // CHECK9-NEXT:    [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
 // CHECK9-NEXT:    [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP196]], align 4
 // CHECK9-NEXT:    [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP197]], align 4
 // CHECK9-NEXT:    [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.region_id, ptr [[KERNEL_ARGS57]])
 // CHECK9-NEXT:    [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
 // CHECK9-NEXT:    br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
 // CHECK9:       omp_offload.failed58:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506(i64 [[TMP158]], i64 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT59]]
 // CHECK9:       omp_offload.cont59:
 // CHECK9-NEXT:    [[TMP200:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
 // CHECK9-NEXT:    [[TMP201:%.*]] = load i64, ptr [[N_CASTED60]], align 8
 // CHECK9-NEXT:    [[TMP202:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP203:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP204:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP201]], ptr [[TMP205]], align 8
 // CHECK9-NEXT:    [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP201]], ptr [[TMP206]], align 8
 // CHECK9-NEXT:    [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP207]], align 8
 // CHECK9-NEXT:    [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP202]], ptr [[TMP208]], align 8
 // CHECK9-NEXT:    [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP202]], ptr [[TMP209]], align 8
 // CHECK9-NEXT:    [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP210]], align 8
 // CHECK9-NEXT:    [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP203]], ptr [[TMP211]], align 8
 // CHECK9-NEXT:    [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP203]], ptr [[TMP212]], align 8
 // CHECK9-NEXT:    [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP213]], align 8
 // CHECK9-NEXT:    [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP204]], ptr [[TMP214]], align 8
 // CHECK9-NEXT:    [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP204]], ptr [[TMP215]], align 8
 // CHECK9-NEXT:    [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP216]], align 8
 // CHECK9-NEXT:    [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP219:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
 // CHECK9-NEXT:    [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
 // CHECK9-NEXT:    [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
 // CHECK9-NEXT:    [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
 // CHECK9-NEXT:    [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
 // CHECK9-NEXT:    store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
 // CHECK9-NEXT:    [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
 // CHECK9-NEXT:    [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
 // CHECK9-NEXT:    [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
 // CHECK9-NEXT:    [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP223]], align 4
 // CHECK9-NEXT:    [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 4, ptr [[TMP224]], align 4
 // CHECK9-NEXT:    [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP217]], ptr [[TMP225]], align 8
 // CHECK9-NEXT:    [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP218]], ptr [[TMP226]], align 8
 // CHECK9-NEXT:    [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.9, ptr [[TMP227]], align 8
 // CHECK9-NEXT:    [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.10, ptr [[TMP228]], align 8
 // CHECK9-NEXT:    [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP229]], align 8
 // CHECK9-NEXT:    [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP230]], align 8
 // CHECK9-NEXT:    [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP222]], ptr [[TMP231]], align 8
 // CHECK9-NEXT:    [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP232]], align 8
 // CHECK9-NEXT:    [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
 // CHECK9-NEXT:    [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP234]], align 4
 // CHECK9-NEXT:    [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP235]], align 4
 // CHECK9-NEXT:    [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.region_id, ptr [[KERNEL_ARGS71]])
 // CHECK9-NEXT:    [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
 // CHECK9-NEXT:    br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
 // CHECK9:       omp_offload.failed72:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536(i64 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT73]]
 // CHECK9:       omp_offload.cont73:
 // CHECK9-NEXT:    [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK9-NEXT:    store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
 // CHECK9-NEXT:    [[TMP239:%.*]] = load i64, ptr [[CH_CASTED74]], align 8
 // CHECK9-NEXT:    [[TMP240:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
 // CHECK9-NEXT:    [[TMP241:%.*]] = load i64, ptr [[N_CASTED75]], align 8
 // CHECK9-NEXT:    [[TMP242:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP243:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP244:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP239]], ptr [[TMP245]], align 8
 // CHECK9-NEXT:    [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP239]], ptr [[TMP246]], align 8
 // CHECK9-NEXT:    [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP247]], align 8
 // CHECK9-NEXT:    [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP241]], ptr [[TMP248]], align 8
 // CHECK9-NEXT:    [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP241]], ptr [[TMP249]], align 8
 // CHECK9-NEXT:    [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP250]], align 8
 // CHECK9-NEXT:    [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP242]], ptr [[TMP251]], align 8
 // CHECK9-NEXT:    [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP242]], ptr [[TMP252]], align 8
 // CHECK9-NEXT:    [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP253]], align 8
 // CHECK9-NEXT:    [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP243]], ptr [[TMP254]], align 8
 // CHECK9-NEXT:    [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP243]], ptr [[TMP255]], align 8
 // CHECK9-NEXT:    [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP256]], align 8
 // CHECK9-NEXT:    [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP244]], ptr [[TMP257]], align 8
 // CHECK9-NEXT:    [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP244]], ptr [[TMP258]], align 8
 // CHECK9-NEXT:    [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 4
 // CHECK9-NEXT:    store ptr null, ptr [[TMP259]], align 8
 // CHECK9-NEXT:    [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP262:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
 // CHECK9-NEXT:    [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
 // CHECK9-NEXT:    [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
 // CHECK9-NEXT:    [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
 // CHECK9-NEXT:    [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
 // CHECK9-NEXT:    store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
 // CHECK9-NEXT:    [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
 // CHECK9-NEXT:    [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
 // CHECK9-NEXT:    [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
 // CHECK9-NEXT:    [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP266]], align 4
 // CHECK9-NEXT:    [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 5, ptr [[TMP267]], align 4
 // CHECK9-NEXT:    [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP260]], ptr [[TMP268]], align 8
 // CHECK9-NEXT:    [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP261]], ptr [[TMP269]], align 8
 // CHECK9-NEXT:    [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.11, ptr [[TMP270]], align 8
 // CHECK9-NEXT:    [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.12, ptr [[TMP271]], align 8
 // CHECK9-NEXT:    [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP272]], align 8
 // CHECK9-NEXT:    [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP273]], align 8
 // CHECK9-NEXT:    [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP265]], ptr [[TMP274]], align 8
 // CHECK9-NEXT:    [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP275]], align 8
 // CHECK9-NEXT:    [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
 // CHECK9-NEXT:    [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP277]], align 4
 // CHECK9-NEXT:    [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP278]], align 4
 // CHECK9-NEXT:    [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.region_id, ptr [[KERNEL_ARGS86]])
 // CHECK9-NEXT:    [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
 // CHECK9-NEXT:    br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
 // CHECK9:       omp_offload.failed87:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562(i64 [[TMP239]], i64 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT88]]
 // CHECK9:       omp_offload.cont88:
 // CHECK9-NEXT:    [[CALL:%.*]] = call noundef signext i32 @_Z5tmainIiET_v()
 // CHECK9-NEXT:    ret i32 [[CALL]]
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369
 // CHECK9-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1:[0-9]+]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i64 [[IDXPROM7]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX8]], align 8
 // CHECK9-NEXT:    [[ADD9:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i64 [[IDXPROM10]]
 // CHECK9-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX11]], align 8
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK9-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408
 // CHECK9-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i64 [[IDXPROM7]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX8]], align 8
 // CHECK9-NEXT:    [[ADD9:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i64 [[IDXPROM10]]
 // CHECK9-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX11]], align 8
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK9-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447
 // CHECK9-SAME: (i64 noundef [[CH:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP7]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP9]])
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.omp_outlined.omp_outlined, i64 [[TMP20]], i64 [[TMP22]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK9-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK9-NEXT:    store i32 [[ADD7]], ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP27]], [[TMP28]]
 // CHECK9-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP30:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP29]], [[TMP30]]
 // CHECK9-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
 // CHECK9:       cond.true10:
 // CHECK9-NEXT:    [[TMP31:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END12:%.*]]
 // CHECK9:       cond.false11:
 // CHECK9-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END12]]
 // CHECK9:       cond.end12:
 // CHECK9-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP31]], [[COND_TRUE10]] ], [ [[TMP32]], [[COND_FALSE11]] ]
 // CHECK9-NEXT:    store i32 [[COND13]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP33:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP33]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP35:%.*]] = load i32, ptr [[TMP34]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP35]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i64 [[IDXPROM7]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX8]], align 8
 // CHECK9-NEXT:    [[ADD9:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i64 [[IDXPROM10]]
 // CHECK9-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX11]], align 8
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK9-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478
 // CHECK9-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 8
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i64 [[IDXPROM7]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX8]], align 8
 // CHECK9-NEXT:    [[ADD9:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i64 [[IDXPROM10]]
 // CHECK9-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX11]], align 8
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK9-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506
 // CHECK9-SAME: (i64 noundef [[CH:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    store i32 [[TMP23]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.omp_outlined.omp_outlined, i64 [[TMP20]], i64 [[TMP22]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i64 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP28:%.*]] = load i32, ptr [[TMP27]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP28]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I5:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV4]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP12]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP10]])
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK9:       omp.dispatch.cond:
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV6:%.*]] = trunc i64 [[TMP14]] to i32
 // CHECK9-NEXT:    [[CMP7:%.*]] = icmp sgt i32 [[TMP13]], [[CONV6]]
 // CHECK9-NEXT:    br i1 [[CMP7]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV8:%.*]] = trunc i64 [[TMP15]] to i32
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[CONV8]], [[COND_TRUE]] ], [ [[TMP16]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP9:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK9-NEXT:    br i1 [[CMP9]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK9:       omp.dispatch.body:
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP10:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
 // CHECK9-NEXT:    br i1 [[CMP10]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I5]], align 4
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I5]], align 4
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX]], align 8
 // CHECK9-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I5]], align 4
 // CHECK9-NEXT:    [[IDXPROM11:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX12:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i64 [[IDXPROM11]]
 // CHECK9-NEXT:    [[TMP28:%.*]] = load double, ptr [[ARRAYIDX12]], align 8
 // CHECK9-NEXT:    [[ADD13:%.*]] = fadd double [[TMP25]], [[TMP28]]
 // CHECK9-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK9-NEXT:    [[TMP30:%.*]] = load i32, ptr [[I5]], align 4
 // CHECK9-NEXT:    [[IDXPROM14:%.*]] = sext i32 [[TMP30]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX15:%.*]] = getelementptr inbounds double, ptr [[TMP29]], i64 [[IDXPROM14]]
 // CHECK9-NEXT:    store double [[ADD13]], ptr [[ARRAYIDX15]], align 8
 // 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:    [[TMP31:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[ADD16:%.*]] = add nsw i32 [[TMP31]], 1
 // CHECK9-NEXT:    store i32 [[ADD16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK9:       omp.dispatch.inc:
 // CHECK9-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP33:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD17:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
 // CHECK9-NEXT:    store i32 [[ADD17]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP34:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP35:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD18:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
 // CHECK9-NEXT:    store i32 [[ADD18]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK9:       omp.dispatch.end:
 // CHECK9-NEXT:    [[TMP36:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP37:%.*]] = load i32, ptr [[TMP36]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP37]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536
 // CHECK9-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4
 // CHECK9-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP13]], i32 35, i32 [[TMP10]], i32 [[TMP11]], i32 1, i32 1)
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK9:       omp.dispatch.cond:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[TMP14]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP15]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK9-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP16]], 0
 // CHECK9-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK9:       omp.dispatch.body:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19:![0-9]+]]
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP2]], align 8, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP22]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP21]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP23:%.*]] = load double, ptr [[ARRAYIDX]], align 8, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[TMP3]], align 8, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[IDXPROM6:%.*]] = sext i32 [[TMP25]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds double, ptr [[TMP24]], i64 [[IDXPROM6]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load double, ptr [[ARRAYIDX7]], align 8, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[ADD8:%.*]] = fadd double [[TMP23]], [[TMP26]]
 // CHECK9-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[TMP1]], align 8, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[TMP28:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[IDXPROM9:%.*]] = sext i32 [[TMP28]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds double, ptr [[TMP27]], i64 [[IDXPROM9]]
 // CHECK9-NEXT:    store double [[ADD8]], ptr [[ARRAYIDX10]], align 8, !llvm.access.group [[ACC_GRP19]]
 // 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:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP29]], 1
 // CHECK9-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP19]]
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP20:![0-9]+]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK9:       omp.dispatch.inc:
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK9:       omp.dispatch.end:
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562
 // CHECK9-SAME: (i64 noundef [[CH:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    store i32 [[TMP23]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.omp_outlined.omp_outlined, i64 [[TMP20]], i64 [[TMP22]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i64 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP28:%.*]] = load i32, ptr [[TMP27]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP28]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I5:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV4]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[TMP13]], align 4
 // CHECK9-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP14]], i32 35, i32 [[TMP11]], i32 [[TMP12]], i32 1, i32 [[TMP10]])
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK9:       omp.dispatch.cond:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP15]], align 4
 // CHECK9-NEXT:    [[TMP17:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP16]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK9-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP17]], 0
 // CHECK9-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK9:       omp.dispatch.body:
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP18]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22:![0-9]+]]
 // CHECK9-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP19]], [[TMP20]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP21]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[TMP22:%.*]] = load ptr, ptr [[TMP2]], align 8, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[TMP23:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP22]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP24:%.*]] = load double, ptr [[ARRAYIDX]], align 8, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[TMP3]], align 8, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP26]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP25]], i64 [[IDXPROM7]]
 // CHECK9-NEXT:    [[TMP27:%.*]] = load double, ptr [[ARRAYIDX8]], align 8, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[ADD9:%.*]] = fadd double [[TMP24]], [[TMP27]]
 // CHECK9-NEXT:    [[TMP28:%.*]] = load ptr, ptr [[TMP1]], align 8, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[TMP29:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP29]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds double, ptr [[TMP28]], i64 [[IDXPROM10]]
 // CHECK9-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX11]], align 8, !llvm.access.group [[ACC_GRP22]]
 // 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:    [[TMP30:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP30]], 1
 // CHECK9-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP22]]
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP23:![0-9]+]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK9:       omp.dispatch.inc:
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK9:       omp.dispatch.end:
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@_Z5tmainIiET_v
 // CHECK9-SAME: () #[[ATTR3:[0-9]+]] comdat {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[A:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[CH:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[N_CASTED:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
 // CHECK9-NEXT:    [[N_CASTED3:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS4:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS5:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS6:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP7:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_8:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_9:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    [[CH_CASTED:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_CASTED17:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS18:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS19:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS20:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP21:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_22:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_23:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    [[N_CASTED31:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS32:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS33:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS34:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP35:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_36:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_37:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    [[CH_CASTED45:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_CASTED46:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS47:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS48:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS49:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP50:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_51:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_52:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    [[N_CASTED60:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS61:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS62:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS63:%.*]] = alloca [4 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP64:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_65:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_66:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    [[CH_CASTED74:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_CASTED75:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_BASEPTRS76:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_PTRS77:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[DOTOFFLOAD_MAPPERS78:%.*]] = alloca [5 x ptr], align 8
 // CHECK9-NEXT:    [[_TMP79:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_80:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_81:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK9-NEXT:    store i32 10000, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 100, ptr [[CH]], align 4
 // CHECK9-NEXT:    [[TMP0:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP0]], ptr [[N_CASTED]], align 4
 // CHECK9-NEXT:    [[TMP1:%.*]] = load i64, ptr [[N_CASTED]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP1]], ptr [[TMP5]], align 8
 // CHECK9-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP1]], ptr [[TMP6]], align 8
 // CHECK9-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP7]], align 8
 // CHECK9-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP2]], ptr [[TMP8]], align 8
 // CHECK9-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP2]], ptr [[TMP9]], align 8
 // CHECK9-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP10]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP3]], ptr [[TMP11]], align 8
 // CHECK9-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP3]], ptr [[TMP12]], align 8
 // CHECK9-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP13]], align 8
 // CHECK9-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP4]], ptr [[TMP14]], align 8
 // CHECK9-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP4]], ptr [[TMP15]], align 8
 // CHECK9-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP16]], align 8
 // CHECK9-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP19]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP20]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], 1
 // CHECK9-NEXT:    [[TMP22:%.*]] = zext i32 [[ADD]] to i64
 // CHECK9-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP23]], align 4
 // CHECK9-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 4, ptr [[TMP24]], align 4
 // CHECK9-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP17]], ptr [[TMP25]], align 8
 // CHECK9-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP18]], ptr [[TMP26]], align 8
 // CHECK9-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.13, ptr [[TMP27]], align 8
 // CHECK9-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.14, ptr [[TMP28]], align 8
 // CHECK9-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP29]], align 8
 // CHECK9-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP30]], align 8
 // CHECK9-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP22]], ptr [[TMP31]], align 8
 // CHECK9-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP32]], align 8
 // CHECK9-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
 // CHECK9-NEXT:    [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
 // CHECK9-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP35]], align 4
 // CHECK9-NEXT:    [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.region_id, ptr [[KERNEL_ARGS]])
 // CHECK9-NEXT:    [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
 // CHECK9-NEXT:    br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 // CHECK9:       omp_offload.failed:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42(i64 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 // CHECK9:       omp_offload.cont:
 // CHECK9-NEXT:    [[TMP38:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
 // CHECK9-NEXT:    [[TMP39:%.*]] = load i64, ptr [[N_CASTED3]], align 8
 // CHECK9-NEXT:    [[TMP40:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP41:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP42:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP39]], ptr [[TMP43]], align 8
 // CHECK9-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP39]], ptr [[TMP44]], align 8
 // CHECK9-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP45]], align 8
 // CHECK9-NEXT:    [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP40]], ptr [[TMP46]], align 8
 // CHECK9-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP40]], ptr [[TMP47]], align 8
 // CHECK9-NEXT:    [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP48]], align 8
 // CHECK9-NEXT:    [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP41]], ptr [[TMP49]], align 8
 // CHECK9-NEXT:    [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP41]], ptr [[TMP50]], align 8
 // CHECK9-NEXT:    [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP51]], align 8
 // CHECK9-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP42]], ptr [[TMP52]], align 8
 // CHECK9-NEXT:    [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP42]], ptr [[TMP53]], align 8
 // CHECK9-NEXT:    [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP54]], align 8
 // CHECK9-NEXT:    [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP57:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
 // CHECK9-NEXT:    [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
 // CHECK9-NEXT:    [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
 // CHECK9-NEXT:    [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
 // CHECK9-NEXT:    [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
 // CHECK9-NEXT:    store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
 // CHECK9-NEXT:    [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
 // CHECK9-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
 // CHECK9-NEXT:    [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
 // CHECK9-NEXT:    [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP61]], align 4
 // CHECK9-NEXT:    [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 4, ptr [[TMP62]], align 4
 // CHECK9-NEXT:    [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP55]], ptr [[TMP63]], align 8
 // CHECK9-NEXT:    [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP56]], ptr [[TMP64]], align 8
 // CHECK9-NEXT:    [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.15, ptr [[TMP65]], align 8
 // CHECK9-NEXT:    [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.16, ptr [[TMP66]], align 8
 // CHECK9-NEXT:    [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP67]], align 8
 // CHECK9-NEXT:    [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP68]], align 8
 // CHECK9-NEXT:    [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP60]], ptr [[TMP69]], align 8
 // CHECK9-NEXT:    [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP70]], align 8
 // CHECK9-NEXT:    [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
 // CHECK9-NEXT:    [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP72]], align 4
 // CHECK9-NEXT:    [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP73]], align 4
 // CHECK9-NEXT:    [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.region_id, ptr [[KERNEL_ARGS14]])
 // CHECK9-NEXT:    [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
 // CHECK9-NEXT:    br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
 // CHECK9:       omp_offload.failed15:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51(i64 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT16]]
 // CHECK9:       omp_offload.cont16:
 // CHECK9-NEXT:    [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK9-NEXT:    store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
 // CHECK9-NEXT:    [[TMP77:%.*]] = load i64, ptr [[CH_CASTED]], align 8
 // CHECK9-NEXT:    [[TMP78:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
 // CHECK9-NEXT:    [[TMP79:%.*]] = load i64, ptr [[N_CASTED17]], align 8
 // CHECK9-NEXT:    [[TMP80:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP81:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP82:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP77]], ptr [[TMP83]], align 8
 // CHECK9-NEXT:    [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP77]], ptr [[TMP84]], align 8
 // CHECK9-NEXT:    [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP85]], align 8
 // CHECK9-NEXT:    [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP79]], ptr [[TMP86]], align 8
 // CHECK9-NEXT:    [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP79]], ptr [[TMP87]], align 8
 // CHECK9-NEXT:    [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP88]], align 8
 // CHECK9-NEXT:    [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP80]], ptr [[TMP89]], align 8
 // CHECK9-NEXT:    [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP80]], ptr [[TMP90]], align 8
 // CHECK9-NEXT:    [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP91]], align 8
 // CHECK9-NEXT:    [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP81]], ptr [[TMP92]], align 8
 // CHECK9-NEXT:    [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP81]], ptr [[TMP93]], align 8
 // CHECK9-NEXT:    [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP94]], align 8
 // CHECK9-NEXT:    [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP82]], ptr [[TMP95]], align 8
 // CHECK9-NEXT:    [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP82]], ptr [[TMP96]], align 8
 // CHECK9-NEXT:    [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i64 0, i64 4
 // CHECK9-NEXT:    store ptr null, ptr [[TMP97]], align 8
 // CHECK9-NEXT:    [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP100:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
 // CHECK9-NEXT:    [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
 // CHECK9-NEXT:    [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
 // CHECK9-NEXT:    [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
 // CHECK9-NEXT:    [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
 // CHECK9-NEXT:    store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
 // CHECK9-NEXT:    [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
 // CHECK9-NEXT:    [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
 // CHECK9-NEXT:    [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
 // CHECK9-NEXT:    [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP104]], align 4
 // CHECK9-NEXT:    [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 5, ptr [[TMP105]], align 4
 // CHECK9-NEXT:    [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP98]], ptr [[TMP106]], align 8
 // CHECK9-NEXT:    [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP99]], ptr [[TMP107]], align 8
 // CHECK9-NEXT:    [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.17, ptr [[TMP108]], align 8
 // CHECK9-NEXT:    [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.18, ptr [[TMP109]], align 8
 // CHECK9-NEXT:    [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP110]], align 8
 // CHECK9-NEXT:    [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP111]], align 8
 // CHECK9-NEXT:    [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP103]], ptr [[TMP112]], align 8
 // CHECK9-NEXT:    [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP113]], align 8
 // CHECK9-NEXT:    [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
 // CHECK9-NEXT:    [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP115]], align 4
 // CHECK9-NEXT:    [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP116]], align 4
 // CHECK9-NEXT:    [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.region_id, ptr [[KERNEL_ARGS28]])
 // CHECK9-NEXT:    [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
 // CHECK9-NEXT:    br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
 // CHECK9:       omp_offload.failed29:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59(i64 [[TMP77]], i64 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT30]]
 // CHECK9:       omp_offload.cont30:
 // CHECK9-NEXT:    [[TMP119:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
 // CHECK9-NEXT:    [[TMP120:%.*]] = load i64, ptr [[N_CASTED31]], align 8
 // CHECK9-NEXT:    [[TMP121:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP122:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP123:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP120]], ptr [[TMP124]], align 8
 // CHECK9-NEXT:    [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP120]], ptr [[TMP125]], align 8
 // CHECK9-NEXT:    [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP126]], align 8
 // CHECK9-NEXT:    [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP121]], ptr [[TMP127]], align 8
 // CHECK9-NEXT:    [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP121]], ptr [[TMP128]], align 8
 // CHECK9-NEXT:    [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP129]], align 8
 // CHECK9-NEXT:    [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP122]], ptr [[TMP130]], align 8
 // CHECK9-NEXT:    [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP122]], ptr [[TMP131]], align 8
 // CHECK9-NEXT:    [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP132]], align 8
 // CHECK9-NEXT:    [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP123]], ptr [[TMP133]], align 8
 // CHECK9-NEXT:    [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP123]], ptr [[TMP134]], align 8
 // CHECK9-NEXT:    [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP135]], align 8
 // CHECK9-NEXT:    [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP138:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
 // CHECK9-NEXT:    [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
 // CHECK9-NEXT:    [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
 // CHECK9-NEXT:    [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
 // CHECK9-NEXT:    [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
 // CHECK9-NEXT:    store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
 // CHECK9-NEXT:    [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
 // CHECK9-NEXT:    [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
 // CHECK9-NEXT:    [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
 // CHECK9-NEXT:    [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP142]], align 4
 // CHECK9-NEXT:    [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 4, ptr [[TMP143]], align 4
 // CHECK9-NEXT:    [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP136]], ptr [[TMP144]], align 8
 // CHECK9-NEXT:    [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP137]], ptr [[TMP145]], align 8
 // CHECK9-NEXT:    [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.19, ptr [[TMP146]], align 8
 // CHECK9-NEXT:    [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.20, ptr [[TMP147]], align 8
 // CHECK9-NEXT:    [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP148]], align 8
 // CHECK9-NEXT:    [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP149]], align 8
 // CHECK9-NEXT:    [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP141]], ptr [[TMP150]], align 8
 // CHECK9-NEXT:    [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP151]], align 8
 // CHECK9-NEXT:    [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
 // CHECK9-NEXT:    [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP153]], align 4
 // CHECK9-NEXT:    [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP154]], align 4
 // CHECK9-NEXT:    [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.region_id, ptr [[KERNEL_ARGS42]])
 // CHECK9-NEXT:    [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
 // CHECK9-NEXT:    br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
 // CHECK9:       omp_offload.failed43:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67(i64 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT44]]
 // CHECK9:       omp_offload.cont44:
 // CHECK9-NEXT:    [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK9-NEXT:    store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
 // CHECK9-NEXT:    [[TMP158:%.*]] = load i64, ptr [[CH_CASTED45]], align 8
 // CHECK9-NEXT:    [[TMP159:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
 // CHECK9-NEXT:    [[TMP160:%.*]] = load i64, ptr [[N_CASTED46]], align 8
 // CHECK9-NEXT:    [[TMP161:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP162:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP163:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP158]], ptr [[TMP164]], align 8
 // CHECK9-NEXT:    [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP158]], ptr [[TMP165]], align 8
 // CHECK9-NEXT:    [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP166]], align 8
 // CHECK9-NEXT:    [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP160]], ptr [[TMP167]], align 8
 // CHECK9-NEXT:    [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP160]], ptr [[TMP168]], align 8
 // CHECK9-NEXT:    [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP169]], align 8
 // CHECK9-NEXT:    [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP161]], ptr [[TMP170]], align 8
 // CHECK9-NEXT:    [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP161]], ptr [[TMP171]], align 8
 // CHECK9-NEXT:    [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP172]], align 8
 // CHECK9-NEXT:    [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP162]], ptr [[TMP173]], align 8
 // CHECK9-NEXT:    [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP162]], ptr [[TMP174]], align 8
 // CHECK9-NEXT:    [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP175]], align 8
 // CHECK9-NEXT:    [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP163]], ptr [[TMP176]], align 8
 // CHECK9-NEXT:    [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP163]], ptr [[TMP177]], align 8
 // CHECK9-NEXT:    [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i64 0, i64 4
 // CHECK9-NEXT:    store ptr null, ptr [[TMP178]], align 8
 // CHECK9-NEXT:    [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP181:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
 // CHECK9-NEXT:    [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
 // CHECK9-NEXT:    [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
 // CHECK9-NEXT:    [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
 // CHECK9-NEXT:    [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
 // CHECK9-NEXT:    store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
 // CHECK9-NEXT:    [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
 // CHECK9-NEXT:    [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
 // CHECK9-NEXT:    [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
 // CHECK9-NEXT:    [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP185]], align 4
 // CHECK9-NEXT:    [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 5, ptr [[TMP186]], align 4
 // CHECK9-NEXT:    [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP179]], ptr [[TMP187]], align 8
 // CHECK9-NEXT:    [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP180]], ptr [[TMP188]], align 8
 // CHECK9-NEXT:    [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.21, ptr [[TMP189]], align 8
 // CHECK9-NEXT:    [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.22, ptr [[TMP190]], align 8
 // CHECK9-NEXT:    [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP191]], align 8
 // CHECK9-NEXT:    [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP192]], align 8
 // CHECK9-NEXT:    [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP184]], ptr [[TMP193]], align 8
 // CHECK9-NEXT:    [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP194]], align 8
 // CHECK9-NEXT:    [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
 // CHECK9-NEXT:    [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP196]], align 4
 // CHECK9-NEXT:    [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP197]], align 4
 // CHECK9-NEXT:    [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.region_id, ptr [[KERNEL_ARGS57]])
 // CHECK9-NEXT:    [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
 // CHECK9-NEXT:    br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
 // CHECK9:       omp_offload.failed58:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75(i64 [[TMP158]], i64 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT59]]
 // CHECK9:       omp_offload.cont59:
 // CHECK9-NEXT:    [[TMP200:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
 // CHECK9-NEXT:    [[TMP201:%.*]] = load i64, ptr [[N_CASTED60]], align 8
 // CHECK9-NEXT:    [[TMP202:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP203:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP204:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP201]], ptr [[TMP205]], align 8
 // CHECK9-NEXT:    [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP201]], ptr [[TMP206]], align 8
 // CHECK9-NEXT:    [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP207]], align 8
 // CHECK9-NEXT:    [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP202]], ptr [[TMP208]], align 8
 // CHECK9-NEXT:    [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
 // CHECK9-NEXT:    store ptr [[TMP202]], ptr [[TMP209]], align 8
 // CHECK9-NEXT:    [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP210]], align 8
 // CHECK9-NEXT:    [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP203]], ptr [[TMP211]], align 8
 // CHECK9-NEXT:    [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP203]], ptr [[TMP212]], align 8
 // CHECK9-NEXT:    [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP213]], align 8
 // CHECK9-NEXT:    [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP204]], ptr [[TMP214]], align 8
 // CHECK9-NEXT:    [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP204]], ptr [[TMP215]], align 8
 // CHECK9-NEXT:    [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP216]], align 8
 // CHECK9-NEXT:    [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP219:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
 // CHECK9-NEXT:    [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
 // CHECK9-NEXT:    [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
 // CHECK9-NEXT:    [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
 // CHECK9-NEXT:    [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
 // CHECK9-NEXT:    store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
 // CHECK9-NEXT:    [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
 // CHECK9-NEXT:    [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
 // CHECK9-NEXT:    [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
 // CHECK9-NEXT:    [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP223]], align 4
 // CHECK9-NEXT:    [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 4, ptr [[TMP224]], align 4
 // CHECK9-NEXT:    [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP217]], ptr [[TMP225]], align 8
 // CHECK9-NEXT:    [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP218]], ptr [[TMP226]], align 8
 // CHECK9-NEXT:    [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.23, ptr [[TMP227]], align 8
 // CHECK9-NEXT:    [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.24, ptr [[TMP228]], align 8
 // CHECK9-NEXT:    [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP229]], align 8
 // CHECK9-NEXT:    [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP230]], align 8
 // CHECK9-NEXT:    [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP222]], ptr [[TMP231]], align 8
 // CHECK9-NEXT:    [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP232]], align 8
 // CHECK9-NEXT:    [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
 // CHECK9-NEXT:    [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP234]], align 4
 // CHECK9-NEXT:    [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP235]], align 4
 // CHECK9-NEXT:    [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.region_id, ptr [[KERNEL_ARGS71]])
 // CHECK9-NEXT:    [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
 // CHECK9-NEXT:    br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
 // CHECK9:       omp_offload.failed72:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83(i64 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT73]]
 // CHECK9:       omp_offload.cont73:
 // CHECK9-NEXT:    [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK9-NEXT:    store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
 // CHECK9-NEXT:    [[TMP239:%.*]] = load i64, ptr [[CH_CASTED74]], align 8
 // CHECK9-NEXT:    [[TMP240:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
 // CHECK9-NEXT:    [[TMP241:%.*]] = load i64, ptr [[N_CASTED75]], align 8
 // CHECK9-NEXT:    [[TMP242:%.*]] = load ptr, ptr [[A]], align 8
 // CHECK9-NEXT:    [[TMP243:%.*]] = load ptr, ptr [[B]], align 8
 // CHECK9-NEXT:    [[TMP244:%.*]] = load ptr, ptr [[C]], align 8
 // CHECK9-NEXT:    [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP239]], ptr [[TMP245]], align 8
 // CHECK9-NEXT:    [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
 // CHECK9-NEXT:    store i64 [[TMP239]], ptr [[TMP246]], align 8
 // CHECK9-NEXT:    [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 0
 // CHECK9-NEXT:    store ptr null, ptr [[TMP247]], align 8
 // CHECK9-NEXT:    [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP241]], ptr [[TMP248]], align 8
 // CHECK9-NEXT:    [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
 // CHECK9-NEXT:    store i64 [[TMP241]], ptr [[TMP249]], align 8
 // CHECK9-NEXT:    [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 1
 // CHECK9-NEXT:    store ptr null, ptr [[TMP250]], align 8
 // CHECK9-NEXT:    [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP242]], ptr [[TMP251]], align 8
 // CHECK9-NEXT:    [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP242]], ptr [[TMP252]], align 8
 // CHECK9-NEXT:    [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 2
 // CHECK9-NEXT:    store ptr null, ptr [[TMP253]], align 8
 // CHECK9-NEXT:    [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP243]], ptr [[TMP254]], align 8
 // CHECK9-NEXT:    [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP243]], ptr [[TMP255]], align 8
 // CHECK9-NEXT:    [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 3
 // CHECK9-NEXT:    store ptr null, ptr [[TMP256]], align 8
 // CHECK9-NEXT:    [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP244]], ptr [[TMP257]], align 8
 // CHECK9-NEXT:    [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr [[TMP244]], ptr [[TMP258]], align 8
 // CHECK9-NEXT:    [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i64 0, i64 4
 // CHECK9-NEXT:    store ptr null, ptr [[TMP259]], align 8
 // CHECK9-NEXT:    [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
 // CHECK9-NEXT:    [[TMP262:%.*]] = load i32, ptr [[N]], align 4
 // CHECK9-NEXT:    store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
 // CHECK9-NEXT:    [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
 // CHECK9-NEXT:    [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
 // CHECK9-NEXT:    [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
 // CHECK9-NEXT:    [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
 // CHECK9-NEXT:    store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
 // CHECK9-NEXT:    [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
 // CHECK9-NEXT:    [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
 // CHECK9-NEXT:    [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
 // CHECK9-NEXT:    [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
 // CHECK9-NEXT:    store i32 3, ptr [[TMP266]], align 4
 // CHECK9-NEXT:    [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
 // CHECK9-NEXT:    store i32 5, ptr [[TMP267]], align 4
 // CHECK9-NEXT:    [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
 // CHECK9-NEXT:    store ptr [[TMP260]], ptr [[TMP268]], align 8
 // CHECK9-NEXT:    [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
 // CHECK9-NEXT:    store ptr [[TMP261]], ptr [[TMP269]], align 8
 // CHECK9-NEXT:    [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
 // CHECK9-NEXT:    store ptr @.offload_sizes.25, ptr [[TMP270]], align 8
 // CHECK9-NEXT:    [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
 // CHECK9-NEXT:    store ptr @.offload_maptypes.26, ptr [[TMP271]], align 8
 // CHECK9-NEXT:    [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
 // CHECK9-NEXT:    store ptr null, ptr [[TMP272]], align 8
 // CHECK9-NEXT:    [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
 // CHECK9-NEXT:    store ptr null, ptr [[TMP273]], align 8
 // CHECK9-NEXT:    [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
 // CHECK9-NEXT:    store i64 [[TMP265]], ptr [[TMP274]], align 8
 // CHECK9-NEXT:    [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
 // CHECK9-NEXT:    store i64 0, ptr [[TMP275]], align 8
 // CHECK9-NEXT:    [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
 // CHECK9-NEXT:    [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
 // CHECK9-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP277]], align 4
 // CHECK9-NEXT:    [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
 // CHECK9-NEXT:    store i32 0, ptr [[TMP278]], align 4
 // CHECK9-NEXT:    [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.region_id, ptr [[KERNEL_ARGS86]])
 // CHECK9-NEXT:    [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
 // CHECK9-NEXT:    br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
 // CHECK9:       omp_offload.failed87:
 // CHECK9-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91(i64 [[TMP239]], i64 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
 // CHECK9-NEXT:    br label [[OMP_OFFLOAD_CONT88]]
 // CHECK9:       omp_offload.cont88:
 // CHECK9-NEXT:    ret i32 0
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42
 // CHECK9-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = call i32 @__kmpc_cancel(ptr @[[GLOB3]], i32 [[TMP21]], i32 2)
 // CHECK9-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
 // CHECK9-NEXT:    br i1 [[TMP23]], label [[DOTCANCEL_EXIT:%.*]], label [[DOTCANCEL_CONTINUE:%.*]]
 // CHECK9:       .cancel.exit:
 // CHECK9-NEXT:    br label [[CANCEL_EXIT:%.*]]
 // CHECK9:       .cancel.continue:
 // CHECK9-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP25]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP24]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK9-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK9-NEXT:    [[TMP28:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP28]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[TMP27]], i64 [[IDXPROM7]]
 // CHECK9-NEXT:    [[TMP29:%.*]] = load i32, ptr [[ARRAYIDX8]], align 4
 // CHECK9-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP26]], [[TMP29]]
 // CHECK9-NEXT:    [[TMP30:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK9-NEXT:    [[TMP31:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP31]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds i32, ptr [[TMP30]], i64 [[IDXPROM10]]
 // CHECK9-NEXT:    store i32 [[ADD9]], ptr [[ARRAYIDX11]], align 4
 // 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:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP32]], 1
 // CHECK9-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP34]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       cancel.exit:
 // CHECK9-NEXT:    [[TMP35:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP36:%.*]] = load i32, ptr [[TMP35]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP36]])
 // CHECK9-NEXT:    br label [[CANCEL_CONT:%.*]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    br label [[CANCEL_CONT]]
 // CHECK9:       cancel.cont:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51
 // CHECK9-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP20]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[TMP23]], i64 [[IDXPROM7]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX8]], align 4
 // CHECK9-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP22]], [[TMP25]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds i32, ptr [[TMP26]], i64 [[IDXPROM10]]
 // CHECK9-NEXT:    store i32 [[ADD9]], ptr [[ARRAYIDX11]], align 4
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK9-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59
 // CHECK9-SAME: (i64 noundef [[CH:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP7]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP9]])
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.omp_outlined.omp_outlined, i64 [[TMP20]], i64 [[TMP22]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK9-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK9-NEXT:    store i32 [[ADD7]], ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP27]], [[TMP28]]
 // CHECK9-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP30:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP29]], [[TMP30]]
 // CHECK9-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
 // CHECK9:       cond.true10:
 // CHECK9-NEXT:    [[TMP31:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END12:%.*]]
 // CHECK9:       cond.false11:
 // CHECK9-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END12]]
 // CHECK9:       cond.end12:
 // CHECK9-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP31]], [[COND_TRUE10]] ], [ [[TMP32]], [[COND_FALSE11]] ]
 // CHECK9-NEXT:    store i32 [[COND13]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP33:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP33]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP34:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP35:%.*]] = load i32, ptr [[TMP34]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP35]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP20]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[TMP23]], i64 [[IDXPROM7]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX8]], align 4
 // CHECK9-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP22]], [[TMP25]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds i32, ptr [[TMP26]], i64 [[IDXPROM10]]
 // CHECK9-NEXT:    store i32 [[ADD9]], ptr [[ARRAYIDX11]], align 4
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK9-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67
 // CHECK9-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP20]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[TMP23]], i64 [[IDXPROM7]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX8]], align 4
 // CHECK9-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP22]], [[TMP25]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK9-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds i32, ptr [[TMP26]], i64 [[IDXPROM10]]
 // CHECK9-NEXT:    store i32 [[ADD9]], ptr [[ARRAYIDX11]], align 4
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK9-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75
 // CHECK9-SAME: (i64 noundef [[CH:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    store i32 [[TMP23]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.omp_outlined.omp_outlined, i64 [[TMP20]], i64 [[TMP22]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i64 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP28:%.*]] = load i32, ptr [[TMP27]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP28]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I5:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV4]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP12]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP10]])
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK9:       omp.dispatch.cond:
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV6:%.*]] = trunc i64 [[TMP14]] to i32
 // CHECK9-NEXT:    [[CMP7:%.*]] = icmp sgt i32 [[TMP13]], [[CONV6]]
 // CHECK9-NEXT:    br i1 [[CMP7]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV8:%.*]] = trunc i64 [[TMP15]] to i32
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[CONV8]], [[COND_TRUE]] ], [ [[TMP16]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP9:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK9-NEXT:    br i1 [[CMP9]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK9:       omp.dispatch.body:
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[CMP10:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
 // CHECK9-NEXT:    br i1 [[CMP10]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I5]], align 4
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP2]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I5]], align 4
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP24]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP23]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK9-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP3]], align 8
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I5]], align 4
 // CHECK9-NEXT:    [[IDXPROM11:%.*]] = sext i32 [[TMP27]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX12:%.*]] = getelementptr inbounds i32, ptr [[TMP26]], i64 [[IDXPROM11]]
 // CHECK9-NEXT:    [[TMP28:%.*]] = load i32, ptr [[ARRAYIDX12]], align 4
 // CHECK9-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP25]], [[TMP28]]
 // CHECK9-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[TMP1]], align 8
 // CHECK9-NEXT:    [[TMP30:%.*]] = load i32, ptr [[I5]], align 4
 // CHECK9-NEXT:    [[IDXPROM14:%.*]] = sext i32 [[TMP30]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX15:%.*]] = getelementptr inbounds i32, ptr [[TMP29]], i64 [[IDXPROM14]]
 // CHECK9-NEXT:    store i32 [[ADD13]], ptr [[ARRAYIDX15]], align 4
 // 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:    [[TMP31:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[ADD16:%.*]] = add nsw i32 [[TMP31]], 1
 // CHECK9-NEXT:    store i32 [[ADD16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK9:       omp.dispatch.inc:
 // CHECK9-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP33:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD17:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
 // CHECK9-NEXT:    store i32 [[ADD17]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP34:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP35:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD18:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
 // CHECK9-NEXT:    store i32 [[ADD18]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK9:       omp.dispatch.end:
 // CHECK9-NEXT:    [[TMP36:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP37:%.*]] = load i32, ptr [[TMP36]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP37]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83
 // CHECK9-SAME: (i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK9-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.omp_outlined.omp_outlined, i64 [[TMP18]], i64 [[TMP20]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV3:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV3]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4
 // CHECK9-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP13]], i32 35, i32 [[TMP10]], i32 [[TMP11]], i32 1, i32 1)
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK9:       omp.dispatch.cond:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[TMP14]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP15]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK9-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP16]], 0
 // CHECK9-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK9:       omp.dispatch.body:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25:![0-9]+]]
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP2]], align 8, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[TMP22:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP22]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP21]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP23:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[TMP3]], align 8, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[IDXPROM6:%.*]] = sext i32 [[TMP25]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds i32, ptr [[TMP24]], i64 [[IDXPROM6]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load i32, ptr [[ARRAYIDX7]], align 4, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP23]], [[TMP26]]
 // CHECK9-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[TMP1]], align 8, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[TMP28:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[IDXPROM9:%.*]] = sext i32 [[TMP28]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, ptr [[TMP27]], i64 [[IDXPROM9]]
 // CHECK9-NEXT:    store i32 [[ADD8]], ptr [[ARRAYIDX10]], align 4, !llvm.access.group [[ACC_GRP25]]
 // 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:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP29]], 1
 // CHECK9-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP25]]
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP26:![0-9]+]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK9:       omp.dispatch.inc:
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK9:       omp.dispatch.end:
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91
 // CHECK9-SAME: (i64 noundef [[CH:%.*]], i64 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    store i64 [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.omp_outlined
 // CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK9-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK9-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK9:       cond.true:
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    br label [[COND_END:%.*]]
 // CHECK9:       cond.false:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    br label [[COND_END]]
 // CHECK9:       cond.end:
 // CHECK9-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK9-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK9-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK9-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP21]] to i64
 // CHECK9-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK9-NEXT:    store i32 [[TMP23]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i64, ptr [[DOTCAPTURE_EXPR__CASTED]], align 8
 // CHECK9-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.omp_outlined.omp_outlined, i64 [[TMP20]], i64 [[TMP22]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i64 [[TMP24]])
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK9:       omp.inner.for.inc:
 // CHECK9-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK9:       omp.loop.exit:
 // CHECK9-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP28:%.*]] = load i32, ptr [[TMP27]], align 4
 // CHECK9-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP28]])
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.omp_outlined.omp_outlined
 // CHECK9-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(4) [[N:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[A:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[B:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[C:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR1]] {
 // CHECK9-NEXT:  entry:
 // CHECK9-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 8
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8
 // CHECK9-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    [[I5:%.*]] = alloca i32, align 4
 // CHECK9-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
 // CHECK9-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK9-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK9-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK9-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK9-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK9-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK9-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK9-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK9:       omp.precond.then:
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK9-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP8:%.*]] = load i64, ptr [[DOTPREVIOUS_LB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV:%.*]] = trunc i64 [[TMP8]] to i32
 // CHECK9-NEXT:    [[TMP9:%.*]] = load i64, ptr [[DOTPREVIOUS_UB__ADDR]], align 8
 // CHECK9-NEXT:    [[CONV4:%.*]] = trunc i64 [[TMP9]] to i32
 // CHECK9-NEXT:    store i32 [[CONV]], ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[CONV4]], ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK9-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK9-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK9-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK9-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP14:%.*]] = load i32, ptr [[TMP13]], align 4
 // CHECK9-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP14]], i32 35, i32 [[TMP11]], i32 [[TMP12]], i32 1, i32 [[TMP10]])
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK9:       omp.dispatch.cond:
 // CHECK9-NEXT:    [[TMP15:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
 // CHECK9-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP15]], align 4
 // CHECK9-NEXT:    [[TMP17:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP16]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK9-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP17]], 0
 // CHECK9-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK9:       omp.dispatch.body:
 // CHECK9-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK9-NEXT:    store i32 [[TMP18]], ptr [[DOTOMP_IV]], align 4
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK9:       omp.inner.for.cond:
 // CHECK9-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28:![0-9]+]]
 // CHECK9-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP19]], [[TMP20]]
 // CHECK9-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK9:       omp.inner.for.body:
 // CHECK9-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP21]], 1
 // CHECK9-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK9-NEXT:    store i32 [[ADD]], ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[TMP22:%.*]] = load ptr, ptr [[TMP2]], align 8, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[TMP23:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP23]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP22]], i64 [[IDXPROM]]
 // CHECK9-NEXT:    [[TMP24:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[TMP3]], align 8, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[TMP26:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP26]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[TMP25]], i64 [[IDXPROM7]]
 // CHECK9-NEXT:    [[TMP27:%.*]] = load i32, ptr [[ARRAYIDX8]], align 4, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP24]], [[TMP27]]
 // CHECK9-NEXT:    [[TMP28:%.*]] = load ptr, ptr [[TMP1]], align 8, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[TMP29:%.*]] = load i32, ptr [[I5]], align 4, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[IDXPROM10:%.*]] = sext i32 [[TMP29]] to i64
 // CHECK9-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds i32, ptr [[TMP28]], i64 [[IDXPROM10]]
 // CHECK9-NEXT:    store i32 [[ADD9]], ptr [[ARRAYIDX11]], align 4, !llvm.access.group [[ACC_GRP28]]
 // 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:    [[TMP30:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP30]], 1
 // CHECK9-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP28]]
 // CHECK9-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP29:![0-9]+]]
 // CHECK9:       omp.inner.for.end:
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK9:       omp.dispatch.inc:
 // CHECK9-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK9:       omp.dispatch.end:
 // CHECK9-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK9:       omp.precond.end:
 // CHECK9-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@main
 // CHECK11-SAME: () #[[ATTR0:[0-9]+]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[RETVAL:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[CH:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
 // CHECK11-NEXT:    [[N_CASTED3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS4:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS5:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS6:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP7:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_8:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_9:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    [[CH_CASTED:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_CASTED17:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS18:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS19:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS20:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP21:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_22:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_23:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    [[N_CASTED31:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS32:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS33:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS34:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP35:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_36:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_37:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    [[CH_CASTED45:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_CASTED46:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS47:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS48:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS49:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP50:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_51:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_52:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    [[N_CASTED60:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS61:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS62:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS63:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP64:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_65:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_66:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    [[CH_CASTED74:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_CASTED75:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS76:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS77:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS78:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP79:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_80:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_81:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    store i32 0, ptr [[RETVAL]], align 4
 // CHECK11-NEXT:    store i32 10000, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 100, ptr [[CH]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP0]], ptr [[N_CASTED]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_CASTED]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP1]], ptr [[TMP5]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP1]], ptr [[TMP6]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP7]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP2]], ptr [[TMP8]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP2]], ptr [[TMP9]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP3]], ptr [[TMP11]], align 4
 // CHECK11-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP3]], ptr [[TMP12]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP13]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP4]], ptr [[TMP14]], align 4
 // CHECK11-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP4]], ptr [[TMP15]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP16]], align 4
 // CHECK11-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP19]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP20]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], 1
 // CHECK11-NEXT:    [[TMP22:%.*]] = zext i32 [[ADD]] to i64
 // CHECK11-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP23]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 4, ptr [[TMP24]], align 4
 // CHECK11-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP17]], ptr [[TMP25]], align 4
 // CHECK11-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP18]], ptr [[TMP26]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes, ptr [[TMP27]], align 4
 // CHECK11-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes, ptr [[TMP28]], align 4
 // CHECK11-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP29]], align 4
 // CHECK11-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP30]], align 4
 // CHECK11-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP22]], ptr [[TMP31]], align 8
 // CHECK11-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP32]], align 8
 // CHECK11-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
 // CHECK11-NEXT:    [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
 // CHECK11-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP35]], align 4
 // CHECK11-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_l369.region_id, ptr [[KERNEL_ARGS]])
 // CHECK11-NEXT:    [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
 // CHECK11-NEXT:    br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 // CHECK11:       omp_offload.failed:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369(i32 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2:[0-9]+]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 // CHECK11:       omp_offload.cont:
 // CHECK11-NEXT:    [[TMP38:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
 // CHECK11-NEXT:    [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
 // CHECK11-NEXT:    [[TMP40:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP41:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP42:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP39]], ptr [[TMP43]], align 4
 // CHECK11-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP39]], ptr [[TMP44]], align 4
 // CHECK11-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP45]], align 4
 // CHECK11-NEXT:    [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP40]], ptr [[TMP46]], align 4
 // CHECK11-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP40]], ptr [[TMP47]], align 4
 // CHECK11-NEXT:    [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP48]], align 4
 // CHECK11-NEXT:    [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP41]], ptr [[TMP49]], align 4
 // CHECK11-NEXT:    [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP41]], ptr [[TMP50]], align 4
 // CHECK11-NEXT:    [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP51]], align 4
 // CHECK11-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP42]], ptr [[TMP52]], align 4
 // CHECK11-NEXT:    [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP42]], ptr [[TMP53]], align 4
 // CHECK11-NEXT:    [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP54]], align 4
 // CHECK11-NEXT:    [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP57:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
 // CHECK11-NEXT:    [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
 // CHECK11-NEXT:    [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
 // CHECK11-NEXT:    [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
 // CHECK11-NEXT:    [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
 // CHECK11-NEXT:    store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
 // CHECK11-NEXT:    [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
 // CHECK11-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
 // CHECK11-NEXT:    [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
 // CHECK11-NEXT:    [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP61]], align 4
 // CHECK11-NEXT:    [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 4, ptr [[TMP62]], align 4
 // CHECK11-NEXT:    [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP55]], ptr [[TMP63]], align 4
 // CHECK11-NEXT:    [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP56]], ptr [[TMP64]], align 4
 // CHECK11-NEXT:    [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.1, ptr [[TMP65]], align 4
 // CHECK11-NEXT:    [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.2, ptr [[TMP66]], align 4
 // CHECK11-NEXT:    [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP67]], align 4
 // CHECK11-NEXT:    [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP68]], align 4
 // CHECK11-NEXT:    [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP60]], ptr [[TMP69]], align 8
 // CHECK11-NEXT:    [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP70]], align 8
 // CHECK11-NEXT:    [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
 // CHECK11-NEXT:    [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP72]], align 4
 // CHECK11-NEXT:    [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP73]], align 4
 // CHECK11-NEXT:    [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.region_id, ptr [[KERNEL_ARGS14]])
 // CHECK11-NEXT:    [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
 // CHECK11-NEXT:    br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
 // CHECK11:       omp_offload.failed15:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408(i32 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT16]]
 // CHECK11:       omp_offload.cont16:
 // CHECK11-NEXT:    [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK11-NEXT:    store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
 // CHECK11-NEXT:    [[TMP77:%.*]] = load i32, ptr [[CH_CASTED]], align 4
 // CHECK11-NEXT:    [[TMP78:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
 // CHECK11-NEXT:    [[TMP79:%.*]] = load i32, ptr [[N_CASTED17]], align 4
 // CHECK11-NEXT:    [[TMP80:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP81:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP82:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP77]], ptr [[TMP83]], align 4
 // CHECK11-NEXT:    [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP77]], ptr [[TMP84]], align 4
 // CHECK11-NEXT:    [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP85]], align 4
 // CHECK11-NEXT:    [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP79]], ptr [[TMP86]], align 4
 // CHECK11-NEXT:    [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP79]], ptr [[TMP87]], align 4
 // CHECK11-NEXT:    [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP88]], align 4
 // CHECK11-NEXT:    [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP80]], ptr [[TMP89]], align 4
 // CHECK11-NEXT:    [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP80]], ptr [[TMP90]], align 4
 // CHECK11-NEXT:    [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP91]], align 4
 // CHECK11-NEXT:    [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP81]], ptr [[TMP92]], align 4
 // CHECK11-NEXT:    [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP81]], ptr [[TMP93]], align 4
 // CHECK11-NEXT:    [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP94]], align 4
 // CHECK11-NEXT:    [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP82]], ptr [[TMP95]], align 4
 // CHECK11-NEXT:    [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP82]], ptr [[TMP96]], align 4
 // CHECK11-NEXT:    [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr null, ptr [[TMP97]], align 4
 // CHECK11-NEXT:    [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP100:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
 // CHECK11-NEXT:    [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
 // CHECK11-NEXT:    [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
 // CHECK11-NEXT:    [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
 // CHECK11-NEXT:    [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
 // CHECK11-NEXT:    store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
 // CHECK11-NEXT:    [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
 // CHECK11-NEXT:    [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
 // CHECK11-NEXT:    [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
 // CHECK11-NEXT:    [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP104]], align 4
 // CHECK11-NEXT:    [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 5, ptr [[TMP105]], align 4
 // CHECK11-NEXT:    [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP98]], ptr [[TMP106]], align 4
 // CHECK11-NEXT:    [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP99]], ptr [[TMP107]], align 4
 // CHECK11-NEXT:    [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.3, ptr [[TMP108]], align 4
 // CHECK11-NEXT:    [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.4, ptr [[TMP109]], align 4
 // CHECK11-NEXT:    [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP110]], align 4
 // CHECK11-NEXT:    [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP111]], align 4
 // CHECK11-NEXT:    [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP103]], ptr [[TMP112]], align 8
 // CHECK11-NEXT:    [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP113]], align 8
 // CHECK11-NEXT:    [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
 // CHECK11-NEXT:    [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP115]], align 4
 // CHECK11-NEXT:    [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP116]], align 4
 // CHECK11-NEXT:    [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.region_id, ptr [[KERNEL_ARGS28]])
 // CHECK11-NEXT:    [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
 // CHECK11-NEXT:    br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
 // CHECK11:       omp_offload.failed29:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447(i32 [[TMP77]], i32 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT30]]
 // CHECK11:       omp_offload.cont30:
 // CHECK11-NEXT:    [[TMP119:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
 // CHECK11-NEXT:    [[TMP120:%.*]] = load i32, ptr [[N_CASTED31]], align 4
 // CHECK11-NEXT:    [[TMP121:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP122:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP123:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP120]], ptr [[TMP124]], align 4
 // CHECK11-NEXT:    [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP120]], ptr [[TMP125]], align 4
 // CHECK11-NEXT:    [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP126]], align 4
 // CHECK11-NEXT:    [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP121]], ptr [[TMP127]], align 4
 // CHECK11-NEXT:    [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP121]], ptr [[TMP128]], align 4
 // CHECK11-NEXT:    [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP129]], align 4
 // CHECK11-NEXT:    [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP122]], ptr [[TMP130]], align 4
 // CHECK11-NEXT:    [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP122]], ptr [[TMP131]], align 4
 // CHECK11-NEXT:    [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP132]], align 4
 // CHECK11-NEXT:    [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP123]], ptr [[TMP133]], align 4
 // CHECK11-NEXT:    [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP123]], ptr [[TMP134]], align 4
 // CHECK11-NEXT:    [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP135]], align 4
 // CHECK11-NEXT:    [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP138:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
 // CHECK11-NEXT:    [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
 // CHECK11-NEXT:    [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
 // CHECK11-NEXT:    [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
 // CHECK11-NEXT:    [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
 // CHECK11-NEXT:    store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
 // CHECK11-NEXT:    [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
 // CHECK11-NEXT:    [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
 // CHECK11-NEXT:    [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
 // CHECK11-NEXT:    [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP142]], align 4
 // CHECK11-NEXT:    [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 4, ptr [[TMP143]], align 4
 // CHECK11-NEXT:    [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP136]], ptr [[TMP144]], align 4
 // CHECK11-NEXT:    [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP137]], ptr [[TMP145]], align 4
 // CHECK11-NEXT:    [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.5, ptr [[TMP146]], align 4
 // CHECK11-NEXT:    [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.6, ptr [[TMP147]], align 4
 // CHECK11-NEXT:    [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP148]], align 4
 // CHECK11-NEXT:    [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP149]], align 4
 // CHECK11-NEXT:    [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP141]], ptr [[TMP150]], align 8
 // CHECK11-NEXT:    [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP151]], align 8
 // CHECK11-NEXT:    [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
 // CHECK11-NEXT:    [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP153]], align 4
 // CHECK11-NEXT:    [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP154]], align 4
 // CHECK11-NEXT:    [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.region_id, ptr [[KERNEL_ARGS42]])
 // CHECK11-NEXT:    [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
 // CHECK11-NEXT:    br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
 // CHECK11:       omp_offload.failed43:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478(i32 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT44]]
 // CHECK11:       omp_offload.cont44:
 // CHECK11-NEXT:    [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK11-NEXT:    store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
 // CHECK11-NEXT:    [[TMP158:%.*]] = load i32, ptr [[CH_CASTED45]], align 4
 // CHECK11-NEXT:    [[TMP159:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
 // CHECK11-NEXT:    [[TMP160:%.*]] = load i32, ptr [[N_CASTED46]], align 4
 // CHECK11-NEXT:    [[TMP161:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP162:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP163:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP158]], ptr [[TMP164]], align 4
 // CHECK11-NEXT:    [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP158]], ptr [[TMP165]], align 4
 // CHECK11-NEXT:    [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP166]], align 4
 // CHECK11-NEXT:    [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP160]], ptr [[TMP167]], align 4
 // CHECK11-NEXT:    [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP160]], ptr [[TMP168]], align 4
 // CHECK11-NEXT:    [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP169]], align 4
 // CHECK11-NEXT:    [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP161]], ptr [[TMP170]], align 4
 // CHECK11-NEXT:    [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP161]], ptr [[TMP171]], align 4
 // CHECK11-NEXT:    [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP172]], align 4
 // CHECK11-NEXT:    [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP162]], ptr [[TMP173]], align 4
 // CHECK11-NEXT:    [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP162]], ptr [[TMP174]], align 4
 // CHECK11-NEXT:    [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP175]], align 4
 // CHECK11-NEXT:    [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP163]], ptr [[TMP176]], align 4
 // CHECK11-NEXT:    [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP163]], ptr [[TMP177]], align 4
 // CHECK11-NEXT:    [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr null, ptr [[TMP178]], align 4
 // CHECK11-NEXT:    [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP181:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
 // CHECK11-NEXT:    [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
 // CHECK11-NEXT:    [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
 // CHECK11-NEXT:    [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
 // CHECK11-NEXT:    [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
 // CHECK11-NEXT:    store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
 // CHECK11-NEXT:    [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
 // CHECK11-NEXT:    [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
 // CHECK11-NEXT:    [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
 // CHECK11-NEXT:    [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP185]], align 4
 // CHECK11-NEXT:    [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 5, ptr [[TMP186]], align 4
 // CHECK11-NEXT:    [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP179]], ptr [[TMP187]], align 4
 // CHECK11-NEXT:    [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP180]], ptr [[TMP188]], align 4
 // CHECK11-NEXT:    [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.7, ptr [[TMP189]], align 4
 // CHECK11-NEXT:    [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.8, ptr [[TMP190]], align 4
 // CHECK11-NEXT:    [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP191]], align 4
 // CHECK11-NEXT:    [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP192]], align 4
 // CHECK11-NEXT:    [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP184]], ptr [[TMP193]], align 8
 // CHECK11-NEXT:    [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP194]], align 8
 // CHECK11-NEXT:    [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
 // CHECK11-NEXT:    [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP196]], align 4
 // CHECK11-NEXT:    [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP197]], align 4
 // CHECK11-NEXT:    [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.region_id, ptr [[KERNEL_ARGS57]])
 // CHECK11-NEXT:    [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
 // CHECK11-NEXT:    br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
 // CHECK11:       omp_offload.failed58:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506(i32 [[TMP158]], i32 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT59]]
 // CHECK11:       omp_offload.cont59:
 // CHECK11-NEXT:    [[TMP200:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
 // CHECK11-NEXT:    [[TMP201:%.*]] = load i32, ptr [[N_CASTED60]], align 4
 // CHECK11-NEXT:    [[TMP202:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP203:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP204:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP201]], ptr [[TMP205]], align 4
 // CHECK11-NEXT:    [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP201]], ptr [[TMP206]], align 4
 // CHECK11-NEXT:    [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP207]], align 4
 // CHECK11-NEXT:    [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP202]], ptr [[TMP208]], align 4
 // CHECK11-NEXT:    [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP202]], ptr [[TMP209]], align 4
 // CHECK11-NEXT:    [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP210]], align 4
 // CHECK11-NEXT:    [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP203]], ptr [[TMP211]], align 4
 // CHECK11-NEXT:    [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP203]], ptr [[TMP212]], align 4
 // CHECK11-NEXT:    [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP213]], align 4
 // CHECK11-NEXT:    [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP204]], ptr [[TMP214]], align 4
 // CHECK11-NEXT:    [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP204]], ptr [[TMP215]], align 4
 // CHECK11-NEXT:    [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP216]], align 4
 // CHECK11-NEXT:    [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP219:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
 // CHECK11-NEXT:    [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
 // CHECK11-NEXT:    [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
 // CHECK11-NEXT:    [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
 // CHECK11-NEXT:    [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
 // CHECK11-NEXT:    store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
 // CHECK11-NEXT:    [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
 // CHECK11-NEXT:    [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
 // CHECK11-NEXT:    [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
 // CHECK11-NEXT:    [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP223]], align 4
 // CHECK11-NEXT:    [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 4, ptr [[TMP224]], align 4
 // CHECK11-NEXT:    [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP217]], ptr [[TMP225]], align 4
 // CHECK11-NEXT:    [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP218]], ptr [[TMP226]], align 4
 // CHECK11-NEXT:    [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.9, ptr [[TMP227]], align 4
 // CHECK11-NEXT:    [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.10, ptr [[TMP228]], align 4
 // CHECK11-NEXT:    [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP229]], align 4
 // CHECK11-NEXT:    [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP230]], align 4
 // CHECK11-NEXT:    [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP222]], ptr [[TMP231]], align 8
 // CHECK11-NEXT:    [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP232]], align 8
 // CHECK11-NEXT:    [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
 // CHECK11-NEXT:    [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP234]], align 4
 // CHECK11-NEXT:    [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP235]], align 4
 // CHECK11-NEXT:    [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.region_id, ptr [[KERNEL_ARGS71]])
 // CHECK11-NEXT:    [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
 // CHECK11-NEXT:    br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
 // CHECK11:       omp_offload.failed72:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536(i32 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT73]]
 // CHECK11:       omp_offload.cont73:
 // CHECK11-NEXT:    [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK11-NEXT:    store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
 // CHECK11-NEXT:    [[TMP239:%.*]] = load i32, ptr [[CH_CASTED74]], align 4
 // CHECK11-NEXT:    [[TMP240:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
 // CHECK11-NEXT:    [[TMP241:%.*]] = load i32, ptr [[N_CASTED75]], align 4
 // CHECK11-NEXT:    [[TMP242:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP243:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP244:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP239]], ptr [[TMP245]], align 4
 // CHECK11-NEXT:    [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP239]], ptr [[TMP246]], align 4
 // CHECK11-NEXT:    [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP247]], align 4
 // CHECK11-NEXT:    [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP241]], ptr [[TMP248]], align 4
 // CHECK11-NEXT:    [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP241]], ptr [[TMP249]], align 4
 // CHECK11-NEXT:    [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP250]], align 4
 // CHECK11-NEXT:    [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP242]], ptr [[TMP251]], align 4
 // CHECK11-NEXT:    [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP242]], ptr [[TMP252]], align 4
 // CHECK11-NEXT:    [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP253]], align 4
 // CHECK11-NEXT:    [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP243]], ptr [[TMP254]], align 4
 // CHECK11-NEXT:    [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP243]], ptr [[TMP255]], align 4
 // CHECK11-NEXT:    [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP256]], align 4
 // CHECK11-NEXT:    [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP244]], ptr [[TMP257]], align 4
 // CHECK11-NEXT:    [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP244]], ptr [[TMP258]], align 4
 // CHECK11-NEXT:    [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr null, ptr [[TMP259]], align 4
 // CHECK11-NEXT:    [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP262:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
 // CHECK11-NEXT:    [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
 // CHECK11-NEXT:    [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
 // CHECK11-NEXT:    [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
 // CHECK11-NEXT:    [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
 // CHECK11-NEXT:    store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
 // CHECK11-NEXT:    [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
 // CHECK11-NEXT:    [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
 // CHECK11-NEXT:    [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
 // CHECK11-NEXT:    [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP266]], align 4
 // CHECK11-NEXT:    [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 5, ptr [[TMP267]], align 4
 // CHECK11-NEXT:    [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP260]], ptr [[TMP268]], align 4
 // CHECK11-NEXT:    [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP261]], ptr [[TMP269]], align 4
 // CHECK11-NEXT:    [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.11, ptr [[TMP270]], align 4
 // CHECK11-NEXT:    [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.12, ptr [[TMP271]], align 4
 // CHECK11-NEXT:    [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP272]], align 4
 // CHECK11-NEXT:    [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP273]], align 4
 // CHECK11-NEXT:    [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP265]], ptr [[TMP274]], align 8
 // CHECK11-NEXT:    [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP275]], align 8
 // CHECK11-NEXT:    [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
 // CHECK11-NEXT:    [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP277]], align 4
 // CHECK11-NEXT:    [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP278]], align 4
 // CHECK11-NEXT:    [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.region_id, ptr [[KERNEL_ARGS86]])
 // CHECK11-NEXT:    [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
 // CHECK11-NEXT:    br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
 // CHECK11:       omp_offload.failed87:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562(i32 [[TMP239]], i32 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT88]]
 // CHECK11:       omp_offload.cont88:
 // CHECK11-NEXT:    [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
 // CHECK11-NEXT:    ret i32 [[CALL]]
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369
 // CHECK11-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1:[0-9]+]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l369.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i32 [[TMP21]]
 // CHECK11-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 4
 // CHECK11-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX6]], align 4
 // CHECK11-NEXT:    [[ADD7:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK11-NEXT:    store double [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK11-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408
 // CHECK11-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l408.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i32 [[TMP21]]
 // CHECK11-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 4
 // CHECK11-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX6]], align 4
 // CHECK11-NEXT:    [[ADD7:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK11-NEXT:    store double [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK11-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447
 // CHECK11-SAME: (i32 noundef [[CH:%.*]], i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP7]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP9]])
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.omp_outlined.omp_outlined, i32 [[TMP19]], i32 [[TMP20]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK11-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK11-NEXT:    store i32 [[ADD7]], ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK11-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
 // CHECK11-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
 // CHECK11:       cond.true10:
 // CHECK11-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END12:%.*]]
 // CHECK11:       cond.false11:
 // CHECK11-NEXT:    [[TMP30:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END12]]
 // CHECK11:       cond.end12:
 // CHECK11-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP29]], [[COND_TRUE10]] ], [ [[TMP30]], [[COND_FALSE11]] ]
 // CHECK11-NEXT:    store i32 [[COND13]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP31:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP31]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP32:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP33:%.*]] = load i32, ptr [[TMP32]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP33]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l447.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i32 [[TMP21]]
 // CHECK11-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 4
 // CHECK11-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX6]], align 4
 // CHECK11-NEXT:    [[ADD7:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK11-NEXT:    store double [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK11-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478
 // CHECK11-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l478.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP20]], i32 [[TMP21]]
 // CHECK11-NEXT:    [[TMP22:%.*]] = load double, ptr [[ARRAYIDX]], align 4
 // CHECK11-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX6]], align 4
 // CHECK11-NEXT:    [[ADD7:%.*]] = fadd double [[TMP22]], [[TMP25]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK11-NEXT:    store double [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK11-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506
 // CHECK11-SAME: (i32 noundef [[CH:%.*]], i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.omp_outlined.omp_outlined, i32 [[TMP19]], i32 [[TMP20]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP26:%.*]] = load i32, ptr [[TMP25]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP26]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l506.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP12]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP10]])
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK11:       omp.dispatch.cond:
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP13]], [[TMP14]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP15]], [[COND_TRUE]] ], [ [[TMP16]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK11-NEXT:    br i1 [[CMP6]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK11:       omp.dispatch.body:
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP7:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
 // CHECK11-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK11-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load double, ptr [[ARRAYIDX]], align 4
 // CHECK11-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK11-NEXT:    [[TMP28:%.*]] = load double, ptr [[ARRAYIDX8]], align 4
 // CHECK11-NEXT:    [[ADD9:%.*]] = fadd double [[TMP25]], [[TMP28]]
 // CHECK11-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    [[TMP30:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds double, ptr [[TMP29]], i32 [[TMP30]]
 // CHECK11-NEXT:    store double [[ADD9]], ptr [[ARRAYIDX10]], align 4
 // 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:    [[TMP31:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP31]], 1
 // CHECK11-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK11:       omp.dispatch.inc:
 // CHECK11-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP33:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
 // CHECK11-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP34:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP35:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
 // CHECK11-NEXT:    store i32 [[ADD13]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK11:       omp.dispatch.end:
 // CHECK11-NEXT:    [[TMP36:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP37:%.*]] = load i32, ptr [[TMP36]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP37]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536
 // CHECK11-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l536.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4
 // CHECK11-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP13]], i32 35, i32 [[TMP10]], i32 [[TMP11]], i32 1, i32 1)
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK11:       omp.dispatch.cond:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[TMP14]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP15]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK11-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP16]], 0
 // CHECK11-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK11:       omp.dispatch.body:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20:![0-9]+]]
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP2]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP21]], i32 [[TMP22]]
 // CHECK11-NEXT:    [[TMP23:%.*]] = load double, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[TMP3]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds double, ptr [[TMP24]], i32 [[TMP25]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load double, ptr [[ARRAYIDX5]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[ADD6:%.*]] = fadd double [[TMP23]], [[TMP26]]
 // CHECK11-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[TMP1]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[TMP28:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds double, ptr [[TMP27]], i32 [[TMP28]]
 // CHECK11-NEXT:    store double [[ADD6]], ptr [[ARRAYIDX7]], align 4, !llvm.access.group [[ACC_GRP20]]
 // 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:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP29]], 1
 // CHECK11-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP20]]
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP21:![0-9]+]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK11:       omp.dispatch.inc:
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK11:       omp.dispatch.end:
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562
 // CHECK11-SAME: (i32 noundef [[CH:%.*]], i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.omp_outlined.omp_outlined, i32 [[TMP19]], i32 [[TMP20]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP26:%.*]] = load i32, ptr [[TMP25]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP26]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l562.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[TMP13]], align 4
 // CHECK11-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP14]], i32 35, i32 [[TMP11]], i32 [[TMP12]], i32 1, i32 [[TMP10]])
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK11:       omp.dispatch.cond:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP15]], align 4
 // CHECK11-NEXT:    [[TMP17:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP16]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK11-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP17]], 0
 // CHECK11-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK11:       omp.dispatch.body:
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP18]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23:![0-9]+]]
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP19]], [[TMP20]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP21]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[TMP22:%.*]] = load ptr, ptr [[TMP2]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[TMP23:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[TMP22]], i32 [[TMP23]]
 // CHECK11-NEXT:    [[TMP24:%.*]] = load double, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[TMP3]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds double, ptr [[TMP25]], i32 [[TMP26]]
 // CHECK11-NEXT:    [[TMP27:%.*]] = load double, ptr [[ARRAYIDX6]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[ADD7:%.*]] = fadd double [[TMP24]], [[TMP27]]
 // CHECK11-NEXT:    [[TMP28:%.*]] = load ptr, ptr [[TMP1]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[TMP29:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds double, ptr [[TMP28]], i32 [[TMP29]]
 // CHECK11-NEXT:    store double [[ADD7]], ptr [[ARRAYIDX8]], align 4, !llvm.access.group [[ACC_GRP23]]
 // 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:    [[TMP30:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP30]], 1
 // CHECK11-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP23]]
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP24:![0-9]+]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK11:       omp.dispatch.inc:
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK11:       omp.dispatch.end:
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@_Z5tmainIiET_v
 // CHECK11-SAME: () #[[ATTR3:[0-9]+]] comdat {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[A:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[CH:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_CASTED:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
 // CHECK11-NEXT:    [[N_CASTED3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS4:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS5:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS6:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP7:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_8:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_9:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS14:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    [[CH_CASTED:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_CASTED17:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS18:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS19:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS20:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP21:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_22:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_23:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS28:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    [[N_CASTED31:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS32:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS33:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS34:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP35:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_36:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_37:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS42:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    [[CH_CASTED45:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_CASTED46:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS47:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS48:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS49:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP50:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_51:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_52:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS57:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    [[N_CASTED60:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS61:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS62:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS63:%.*]] = alloca [4 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP64:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_65:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_66:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS71:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    [[CH_CASTED74:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_CASTED75:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_BASEPTRS76:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_PTRS77:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[DOTOFFLOAD_MAPPERS78:%.*]] = alloca [5 x ptr], align 4
 // CHECK11-NEXT:    [[_TMP79:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_80:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_81:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[KERNEL_ARGS86:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK11-NEXT:    store i32 10000, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 100, ptr [[CH]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP0]], ptr [[N_CASTED]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load i32, ptr [[N_CASTED]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP1]], ptr [[TMP5]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP1]], ptr [[TMP6]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP7]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP2]], ptr [[TMP8]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP2]], ptr [[TMP9]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP3]], ptr [[TMP11]], align 4
 // CHECK11-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP3]], ptr [[TMP12]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP13]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP4]], ptr [[TMP14]], align 4
 // CHECK11-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP4]], ptr [[TMP15]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP16]], align 4
 // CHECK11-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP19]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP20]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP21]], 1
 // CHECK11-NEXT:    [[TMP22:%.*]] = zext i32 [[ADD]] to i64
 // CHECK11-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP23]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 4, ptr [[TMP24]], align 4
 // CHECK11-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP17]], ptr [[TMP25]], align 4
 // CHECK11-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP18]], ptr [[TMP26]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.13, ptr [[TMP27]], align 4
 // CHECK11-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.14, ptr [[TMP28]], align 4
 // CHECK11-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP29]], align 4
 // CHECK11-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP30]], align 4
 // CHECK11-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP22]], ptr [[TMP31]], align 8
 // CHECK11-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP32]], align 8
 // CHECK11-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP33]], align 4
 // CHECK11-NEXT:    [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP34]], align 4
 // CHECK11-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP35]], align 4
 // CHECK11-NEXT:    [[TMP36:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.region_id, ptr [[KERNEL_ARGS]])
 // CHECK11-NEXT:    [[TMP37:%.*]] = icmp ne i32 [[TMP36]], 0
 // CHECK11-NEXT:    br i1 [[TMP37]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 // CHECK11:       omp_offload.failed:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42(i32 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 // CHECK11:       omp_offload.cont:
 // CHECK11-NEXT:    [[TMP38:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP38]], ptr [[N_CASTED3]], align 4
 // CHECK11-NEXT:    [[TMP39:%.*]] = load i32, ptr [[N_CASTED3]], align 4
 // CHECK11-NEXT:    [[TMP40:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP41:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP42:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP39]], ptr [[TMP43]], align 4
 // CHECK11-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP39]], ptr [[TMP44]], align 4
 // CHECK11-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP45]], align 4
 // CHECK11-NEXT:    [[TMP46:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP40]], ptr [[TMP46]], align 4
 // CHECK11-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP40]], ptr [[TMP47]], align 4
 // CHECK11-NEXT:    [[TMP48:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP48]], align 4
 // CHECK11-NEXT:    [[TMP49:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP41]], ptr [[TMP49]], align 4
 // CHECK11-NEXT:    [[TMP50:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP41]], ptr [[TMP50]], align 4
 // CHECK11-NEXT:    [[TMP51:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP51]], align 4
 // CHECK11-NEXT:    [[TMP52:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP42]], ptr [[TMP52]], align 4
 // CHECK11-NEXT:    [[TMP53:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP42]], ptr [[TMP53]], align 4
 // CHECK11-NEXT:    [[TMP54:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP54]], align 4
 // CHECK11-NEXT:    [[TMP55:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP56:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP57:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP57]], ptr [[DOTCAPTURE_EXPR_8]], align 4
 // CHECK11-NEXT:    [[TMP58:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_8]], align 4
 // CHECK11-NEXT:    [[SUB10:%.*]] = sub nsw i32 [[TMP58]], 0
 // CHECK11-NEXT:    [[DIV11:%.*]] = sdiv i32 [[SUB10]], 1
 // CHECK11-NEXT:    [[SUB12:%.*]] = sub nsw i32 [[DIV11]], 1
 // CHECK11-NEXT:    store i32 [[SUB12]], ptr [[DOTCAPTURE_EXPR_9]], align 4
 // CHECK11-NEXT:    [[TMP59:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_9]], align 4
 // CHECK11-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP59]], 1
 // CHECK11-NEXT:    [[TMP60:%.*]] = zext i32 [[ADD13]] to i64
 // CHECK11-NEXT:    [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP61]], align 4
 // CHECK11-NEXT:    [[TMP62:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 4, ptr [[TMP62]], align 4
 // CHECK11-NEXT:    [[TMP63:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP55]], ptr [[TMP63]], align 4
 // CHECK11-NEXT:    [[TMP64:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP56]], ptr [[TMP64]], align 4
 // CHECK11-NEXT:    [[TMP65:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.15, ptr [[TMP65]], align 4
 // CHECK11-NEXT:    [[TMP66:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.16, ptr [[TMP66]], align 4
 // CHECK11-NEXT:    [[TMP67:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP67]], align 4
 // CHECK11-NEXT:    [[TMP68:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP68]], align 4
 // CHECK11-NEXT:    [[TMP69:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP60]], ptr [[TMP69]], align 8
 // CHECK11-NEXT:    [[TMP70:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP70]], align 8
 // CHECK11-NEXT:    [[TMP71:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP71]], align 4
 // CHECK11-NEXT:    [[TMP72:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP72]], align 4
 // CHECK11-NEXT:    [[TMP73:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS14]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP73]], align 4
 // CHECK11-NEXT:    [[TMP74:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.region_id, ptr [[KERNEL_ARGS14]])
 // CHECK11-NEXT:    [[TMP75:%.*]] = icmp ne i32 [[TMP74]], 0
 // CHECK11-NEXT:    br i1 [[TMP75]], label [[OMP_OFFLOAD_FAILED15:%.*]], label [[OMP_OFFLOAD_CONT16:%.*]]
 // CHECK11:       omp_offload.failed15:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51(i32 [[TMP39]], ptr [[TMP40]], ptr [[TMP41]], ptr [[TMP42]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT16]]
 // CHECK11:       omp_offload.cont16:
 // CHECK11-NEXT:    [[TMP76:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK11-NEXT:    store i32 [[TMP76]], ptr [[CH_CASTED]], align 4
 // CHECK11-NEXT:    [[TMP77:%.*]] = load i32, ptr [[CH_CASTED]], align 4
 // CHECK11-NEXT:    [[TMP78:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP78]], ptr [[N_CASTED17]], align 4
 // CHECK11-NEXT:    [[TMP79:%.*]] = load i32, ptr [[N_CASTED17]], align 4
 // CHECK11-NEXT:    [[TMP80:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP81:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP82:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP83:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP77]], ptr [[TMP83]], align 4
 // CHECK11-NEXT:    [[TMP84:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP77]], ptr [[TMP84]], align 4
 // CHECK11-NEXT:    [[TMP85:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP85]], align 4
 // CHECK11-NEXT:    [[TMP86:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP79]], ptr [[TMP86]], align 4
 // CHECK11-NEXT:    [[TMP87:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP79]], ptr [[TMP87]], align 4
 // CHECK11-NEXT:    [[TMP88:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP88]], align 4
 // CHECK11-NEXT:    [[TMP89:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP80]], ptr [[TMP89]], align 4
 // CHECK11-NEXT:    [[TMP90:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP80]], ptr [[TMP90]], align 4
 // CHECK11-NEXT:    [[TMP91:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP91]], align 4
 // CHECK11-NEXT:    [[TMP92:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP81]], ptr [[TMP92]], align 4
 // CHECK11-NEXT:    [[TMP93:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP81]], ptr [[TMP93]], align 4
 // CHECK11-NEXT:    [[TMP94:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP94]], align 4
 // CHECK11-NEXT:    [[TMP95:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP82]], ptr [[TMP95]], align 4
 // CHECK11-NEXT:    [[TMP96:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP82]], ptr [[TMP96]], align 4
 // CHECK11-NEXT:    [[TMP97:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS20]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr null, ptr [[TMP97]], align 4
 // CHECK11-NEXT:    [[TMP98:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS18]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP99:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS19]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP100:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP100]], ptr [[DOTCAPTURE_EXPR_22]], align 4
 // CHECK11-NEXT:    [[TMP101:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_22]], align 4
 // CHECK11-NEXT:    [[SUB24:%.*]] = sub nsw i32 [[TMP101]], 0
 // CHECK11-NEXT:    [[DIV25:%.*]] = sdiv i32 [[SUB24]], 1
 // CHECK11-NEXT:    [[SUB26:%.*]] = sub nsw i32 [[DIV25]], 1
 // CHECK11-NEXT:    store i32 [[SUB26]], ptr [[DOTCAPTURE_EXPR_23]], align 4
 // CHECK11-NEXT:    [[TMP102:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_23]], align 4
 // CHECK11-NEXT:    [[ADD27:%.*]] = add nsw i32 [[TMP102]], 1
 // CHECK11-NEXT:    [[TMP103:%.*]] = zext i32 [[ADD27]] to i64
 // CHECK11-NEXT:    [[TMP104:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP104]], align 4
 // CHECK11-NEXT:    [[TMP105:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 5, ptr [[TMP105]], align 4
 // CHECK11-NEXT:    [[TMP106:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP98]], ptr [[TMP106]], align 4
 // CHECK11-NEXT:    [[TMP107:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP99]], ptr [[TMP107]], align 4
 // CHECK11-NEXT:    [[TMP108:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.17, ptr [[TMP108]], align 4
 // CHECK11-NEXT:    [[TMP109:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.18, ptr [[TMP109]], align 4
 // CHECK11-NEXT:    [[TMP110:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP110]], align 4
 // CHECK11-NEXT:    [[TMP111:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP111]], align 4
 // CHECK11-NEXT:    [[TMP112:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP103]], ptr [[TMP112]], align 8
 // CHECK11-NEXT:    [[TMP113:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP113]], align 8
 // CHECK11-NEXT:    [[TMP114:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP114]], align 4
 // CHECK11-NEXT:    [[TMP115:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP115]], align 4
 // CHECK11-NEXT:    [[TMP116:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS28]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP116]], align 4
 // CHECK11-NEXT:    [[TMP117:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.region_id, ptr [[KERNEL_ARGS28]])
 // CHECK11-NEXT:    [[TMP118:%.*]] = icmp ne i32 [[TMP117]], 0
 // CHECK11-NEXT:    br i1 [[TMP118]], label [[OMP_OFFLOAD_FAILED29:%.*]], label [[OMP_OFFLOAD_CONT30:%.*]]
 // CHECK11:       omp_offload.failed29:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59(i32 [[TMP77]], i32 [[TMP79]], ptr [[TMP80]], ptr [[TMP81]], ptr [[TMP82]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT30]]
 // CHECK11:       omp_offload.cont30:
 // CHECK11-NEXT:    [[TMP119:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP119]], ptr [[N_CASTED31]], align 4
 // CHECK11-NEXT:    [[TMP120:%.*]] = load i32, ptr [[N_CASTED31]], align 4
 // CHECK11-NEXT:    [[TMP121:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP122:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP123:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP124:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP120]], ptr [[TMP124]], align 4
 // CHECK11-NEXT:    [[TMP125:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP120]], ptr [[TMP125]], align 4
 // CHECK11-NEXT:    [[TMP126:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP126]], align 4
 // CHECK11-NEXT:    [[TMP127:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP121]], ptr [[TMP127]], align 4
 // CHECK11-NEXT:    [[TMP128:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP121]], ptr [[TMP128]], align 4
 // CHECK11-NEXT:    [[TMP129:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP129]], align 4
 // CHECK11-NEXT:    [[TMP130:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP122]], ptr [[TMP130]], align 4
 // CHECK11-NEXT:    [[TMP131:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP122]], ptr [[TMP131]], align 4
 // CHECK11-NEXT:    [[TMP132:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP132]], align 4
 // CHECK11-NEXT:    [[TMP133:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP123]], ptr [[TMP133]], align 4
 // CHECK11-NEXT:    [[TMP134:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP123]], ptr [[TMP134]], align 4
 // CHECK11-NEXT:    [[TMP135:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS34]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP135]], align 4
 // CHECK11-NEXT:    [[TMP136:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS32]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP137:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS33]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP138:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP138]], ptr [[DOTCAPTURE_EXPR_36]], align 4
 // CHECK11-NEXT:    [[TMP139:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_36]], align 4
 // CHECK11-NEXT:    [[SUB38:%.*]] = sub nsw i32 [[TMP139]], 0
 // CHECK11-NEXT:    [[DIV39:%.*]] = sdiv i32 [[SUB38]], 1
 // CHECK11-NEXT:    [[SUB40:%.*]] = sub nsw i32 [[DIV39]], 1
 // CHECK11-NEXT:    store i32 [[SUB40]], ptr [[DOTCAPTURE_EXPR_37]], align 4
 // CHECK11-NEXT:    [[TMP140:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_37]], align 4
 // CHECK11-NEXT:    [[ADD41:%.*]] = add nsw i32 [[TMP140]], 1
 // CHECK11-NEXT:    [[TMP141:%.*]] = zext i32 [[ADD41]] to i64
 // CHECK11-NEXT:    [[TMP142:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP142]], align 4
 // CHECK11-NEXT:    [[TMP143:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 4, ptr [[TMP143]], align 4
 // CHECK11-NEXT:    [[TMP144:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP136]], ptr [[TMP144]], align 4
 // CHECK11-NEXT:    [[TMP145:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP137]], ptr [[TMP145]], align 4
 // CHECK11-NEXT:    [[TMP146:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.19, ptr [[TMP146]], align 4
 // CHECK11-NEXT:    [[TMP147:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.20, ptr [[TMP147]], align 4
 // CHECK11-NEXT:    [[TMP148:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP148]], align 4
 // CHECK11-NEXT:    [[TMP149:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP149]], align 4
 // CHECK11-NEXT:    [[TMP150:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP141]], ptr [[TMP150]], align 8
 // CHECK11-NEXT:    [[TMP151:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP151]], align 8
 // CHECK11-NEXT:    [[TMP152:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP152]], align 4
 // CHECK11-NEXT:    [[TMP153:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP153]], align 4
 // CHECK11-NEXT:    [[TMP154:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS42]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP154]], align 4
 // CHECK11-NEXT:    [[TMP155:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.region_id, ptr [[KERNEL_ARGS42]])
 // CHECK11-NEXT:    [[TMP156:%.*]] = icmp ne i32 [[TMP155]], 0
 // CHECK11-NEXT:    br i1 [[TMP156]], label [[OMP_OFFLOAD_FAILED43:%.*]], label [[OMP_OFFLOAD_CONT44:%.*]]
 // CHECK11:       omp_offload.failed43:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67(i32 [[TMP120]], ptr [[TMP121]], ptr [[TMP122]], ptr [[TMP123]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT44]]
 // CHECK11:       omp_offload.cont44:
 // CHECK11-NEXT:    [[TMP157:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK11-NEXT:    store i32 [[TMP157]], ptr [[CH_CASTED45]], align 4
 // CHECK11-NEXT:    [[TMP158:%.*]] = load i32, ptr [[CH_CASTED45]], align 4
 // CHECK11-NEXT:    [[TMP159:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP159]], ptr [[N_CASTED46]], align 4
 // CHECK11-NEXT:    [[TMP160:%.*]] = load i32, ptr [[N_CASTED46]], align 4
 // CHECK11-NEXT:    [[TMP161:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP162:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP163:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP164:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP158]], ptr [[TMP164]], align 4
 // CHECK11-NEXT:    [[TMP165:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP158]], ptr [[TMP165]], align 4
 // CHECK11-NEXT:    [[TMP166:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP166]], align 4
 // CHECK11-NEXT:    [[TMP167:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP160]], ptr [[TMP167]], align 4
 // CHECK11-NEXT:    [[TMP168:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP160]], ptr [[TMP168]], align 4
 // CHECK11-NEXT:    [[TMP169:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP169]], align 4
 // CHECK11-NEXT:    [[TMP170:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP161]], ptr [[TMP170]], align 4
 // CHECK11-NEXT:    [[TMP171:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP161]], ptr [[TMP171]], align 4
 // CHECK11-NEXT:    [[TMP172:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP172]], align 4
 // CHECK11-NEXT:    [[TMP173:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP162]], ptr [[TMP173]], align 4
 // CHECK11-NEXT:    [[TMP174:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP162]], ptr [[TMP174]], align 4
 // CHECK11-NEXT:    [[TMP175:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP175]], align 4
 // CHECK11-NEXT:    [[TMP176:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP163]], ptr [[TMP176]], align 4
 // CHECK11-NEXT:    [[TMP177:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP163]], ptr [[TMP177]], align 4
 // CHECK11-NEXT:    [[TMP178:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS49]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr null, ptr [[TMP178]], align 4
 // CHECK11-NEXT:    [[TMP179:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS47]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP180:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS48]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP181:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP181]], ptr [[DOTCAPTURE_EXPR_51]], align 4
 // CHECK11-NEXT:    [[TMP182:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_51]], align 4
 // CHECK11-NEXT:    [[SUB53:%.*]] = sub nsw i32 [[TMP182]], 0
 // CHECK11-NEXT:    [[DIV54:%.*]] = sdiv i32 [[SUB53]], 1
 // CHECK11-NEXT:    [[SUB55:%.*]] = sub nsw i32 [[DIV54]], 1
 // CHECK11-NEXT:    store i32 [[SUB55]], ptr [[DOTCAPTURE_EXPR_52]], align 4
 // CHECK11-NEXT:    [[TMP183:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_52]], align 4
 // CHECK11-NEXT:    [[ADD56:%.*]] = add nsw i32 [[TMP183]], 1
 // CHECK11-NEXT:    [[TMP184:%.*]] = zext i32 [[ADD56]] to i64
 // CHECK11-NEXT:    [[TMP185:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP185]], align 4
 // CHECK11-NEXT:    [[TMP186:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 5, ptr [[TMP186]], align 4
 // CHECK11-NEXT:    [[TMP187:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP179]], ptr [[TMP187]], align 4
 // CHECK11-NEXT:    [[TMP188:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP180]], ptr [[TMP188]], align 4
 // CHECK11-NEXT:    [[TMP189:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.21, ptr [[TMP189]], align 4
 // CHECK11-NEXT:    [[TMP190:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.22, ptr [[TMP190]], align 4
 // CHECK11-NEXT:    [[TMP191:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP191]], align 4
 // CHECK11-NEXT:    [[TMP192:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP192]], align 4
 // CHECK11-NEXT:    [[TMP193:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP184]], ptr [[TMP193]], align 8
 // CHECK11-NEXT:    [[TMP194:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP194]], align 8
 // CHECK11-NEXT:    [[TMP195:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP195]], align 4
 // CHECK11-NEXT:    [[TMP196:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP196]], align 4
 // CHECK11-NEXT:    [[TMP197:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS57]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP197]], align 4
 // CHECK11-NEXT:    [[TMP198:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.region_id, ptr [[KERNEL_ARGS57]])
 // CHECK11-NEXT:    [[TMP199:%.*]] = icmp ne i32 [[TMP198]], 0
 // CHECK11-NEXT:    br i1 [[TMP199]], label [[OMP_OFFLOAD_FAILED58:%.*]], label [[OMP_OFFLOAD_CONT59:%.*]]
 // CHECK11:       omp_offload.failed58:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75(i32 [[TMP158]], i32 [[TMP160]], ptr [[TMP161]], ptr [[TMP162]], ptr [[TMP163]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT59]]
 // CHECK11:       omp_offload.cont59:
 // CHECK11-NEXT:    [[TMP200:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP200]], ptr [[N_CASTED60]], align 4
 // CHECK11-NEXT:    [[TMP201:%.*]] = load i32, ptr [[N_CASTED60]], align 4
 // CHECK11-NEXT:    [[TMP202:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP203:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP204:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP205:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP201]], ptr [[TMP205]], align 4
 // CHECK11-NEXT:    [[TMP206:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP201]], ptr [[TMP206]], align 4
 // CHECK11-NEXT:    [[TMP207:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP207]], align 4
 // CHECK11-NEXT:    [[TMP208:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP202]], ptr [[TMP208]], align 4
 // CHECK11-NEXT:    [[TMP209:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr [[TMP202]], ptr [[TMP209]], align 4
 // CHECK11-NEXT:    [[TMP210:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP210]], align 4
 // CHECK11-NEXT:    [[TMP211:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP203]], ptr [[TMP211]], align 4
 // CHECK11-NEXT:    [[TMP212:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP203]], ptr [[TMP212]], align 4
 // CHECK11-NEXT:    [[TMP213:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP213]], align 4
 // CHECK11-NEXT:    [[TMP214:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP204]], ptr [[TMP214]], align 4
 // CHECK11-NEXT:    [[TMP215:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP204]], ptr [[TMP215]], align 4
 // CHECK11-NEXT:    [[TMP216:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS63]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP216]], align 4
 // CHECK11-NEXT:    [[TMP217:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS61]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP218:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS62]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP219:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP219]], ptr [[DOTCAPTURE_EXPR_65]], align 4
 // CHECK11-NEXT:    [[TMP220:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_65]], align 4
 // CHECK11-NEXT:    [[SUB67:%.*]] = sub nsw i32 [[TMP220]], 0
 // CHECK11-NEXT:    [[DIV68:%.*]] = sdiv i32 [[SUB67]], 1
 // CHECK11-NEXT:    [[SUB69:%.*]] = sub nsw i32 [[DIV68]], 1
 // CHECK11-NEXT:    store i32 [[SUB69]], ptr [[DOTCAPTURE_EXPR_66]], align 4
 // CHECK11-NEXT:    [[TMP221:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_66]], align 4
 // CHECK11-NEXT:    [[ADD70:%.*]] = add nsw i32 [[TMP221]], 1
 // CHECK11-NEXT:    [[TMP222:%.*]] = zext i32 [[ADD70]] to i64
 // CHECK11-NEXT:    [[TMP223:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP223]], align 4
 // CHECK11-NEXT:    [[TMP224:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 4, ptr [[TMP224]], align 4
 // CHECK11-NEXT:    [[TMP225:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP217]], ptr [[TMP225]], align 4
 // CHECK11-NEXT:    [[TMP226:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP218]], ptr [[TMP226]], align 4
 // CHECK11-NEXT:    [[TMP227:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.23, ptr [[TMP227]], align 4
 // CHECK11-NEXT:    [[TMP228:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.24, ptr [[TMP228]], align 4
 // CHECK11-NEXT:    [[TMP229:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP229]], align 4
 // CHECK11-NEXT:    [[TMP230:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP230]], align 4
 // CHECK11-NEXT:    [[TMP231:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP222]], ptr [[TMP231]], align 8
 // CHECK11-NEXT:    [[TMP232:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP232]], align 8
 // CHECK11-NEXT:    [[TMP233:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP233]], align 4
 // CHECK11-NEXT:    [[TMP234:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP234]], align 4
 // CHECK11-NEXT:    [[TMP235:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS71]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP235]], align 4
 // CHECK11-NEXT:    [[TMP236:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.region_id, ptr [[KERNEL_ARGS71]])
 // CHECK11-NEXT:    [[TMP237:%.*]] = icmp ne i32 [[TMP236]], 0
 // CHECK11-NEXT:    br i1 [[TMP237]], label [[OMP_OFFLOAD_FAILED72:%.*]], label [[OMP_OFFLOAD_CONT73:%.*]]
 // CHECK11:       omp_offload.failed72:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83(i32 [[TMP201]], ptr [[TMP202]], ptr [[TMP203]], ptr [[TMP204]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT73]]
 // CHECK11:       omp_offload.cont73:
 // CHECK11-NEXT:    [[TMP238:%.*]] = load i32, ptr [[CH]], align 4
 // CHECK11-NEXT:    store i32 [[TMP238]], ptr [[CH_CASTED74]], align 4
 // CHECK11-NEXT:    [[TMP239:%.*]] = load i32, ptr [[CH_CASTED74]], align 4
 // CHECK11-NEXT:    [[TMP240:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP240]], ptr [[N_CASTED75]], align 4
 // CHECK11-NEXT:    [[TMP241:%.*]] = load i32, ptr [[N_CASTED75]], align 4
 // CHECK11-NEXT:    [[TMP242:%.*]] = load ptr, ptr [[A]], align 4
 // CHECK11-NEXT:    [[TMP243:%.*]] = load ptr, ptr [[B]], align 4
 // CHECK11-NEXT:    [[TMP244:%.*]] = load ptr, ptr [[C]], align 4
 // CHECK11-NEXT:    [[TMP245:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP239]], ptr [[TMP245]], align 4
 // CHECK11-NEXT:    [[TMP246:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 [[TMP239]], ptr [[TMP246]], align 4
 // CHECK11-NEXT:    [[TMP247:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 0
 // CHECK11-NEXT:    store ptr null, ptr [[TMP247]], align 4
 // CHECK11-NEXT:    [[TMP248:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP241]], ptr [[TMP248]], align 4
 // CHECK11-NEXT:    [[TMP249:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 [[TMP241]], ptr [[TMP249]], align 4
 // CHECK11-NEXT:    [[TMP250:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 1
 // CHECK11-NEXT:    store ptr null, ptr [[TMP250]], align 4
 // CHECK11-NEXT:    [[TMP251:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP242]], ptr [[TMP251]], align 4
 // CHECK11-NEXT:    [[TMP252:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP242]], ptr [[TMP252]], align 4
 // CHECK11-NEXT:    [[TMP253:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr null, ptr [[TMP253]], align 4
 // CHECK11-NEXT:    [[TMP254:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP243]], ptr [[TMP254]], align 4
 // CHECK11-NEXT:    [[TMP255:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP243]], ptr [[TMP255]], align 4
 // CHECK11-NEXT:    [[TMP256:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr null, ptr [[TMP256]], align 4
 // CHECK11-NEXT:    [[TMP257:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP244]], ptr [[TMP257]], align 4
 // CHECK11-NEXT:    [[TMP258:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr [[TMP244]], ptr [[TMP258]], align 4
 // CHECK11-NEXT:    [[TMP259:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS78]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr null, ptr [[TMP259]], align 4
 // CHECK11-NEXT:    [[TMP260:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS76]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP261:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS77]], i32 0, i32 0
 // CHECK11-NEXT:    [[TMP262:%.*]] = load i32, ptr [[N]], align 4
 // CHECK11-NEXT:    store i32 [[TMP262]], ptr [[DOTCAPTURE_EXPR_80]], align 4
 // CHECK11-NEXT:    [[TMP263:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_80]], align 4
 // CHECK11-NEXT:    [[SUB82:%.*]] = sub nsw i32 [[TMP263]], 0
 // CHECK11-NEXT:    [[DIV83:%.*]] = sdiv i32 [[SUB82]], 1
 // CHECK11-NEXT:    [[SUB84:%.*]] = sub nsw i32 [[DIV83]], 1
 // CHECK11-NEXT:    store i32 [[SUB84]], ptr [[DOTCAPTURE_EXPR_81]], align 4
 // CHECK11-NEXT:    [[TMP264:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_81]], align 4
 // CHECK11-NEXT:    [[ADD85:%.*]] = add nsw i32 [[TMP264]], 1
 // CHECK11-NEXT:    [[TMP265:%.*]] = zext i32 [[ADD85]] to i64
 // CHECK11-NEXT:    [[TMP266:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 0
 // CHECK11-NEXT:    store i32 3, ptr [[TMP266]], align 4
 // CHECK11-NEXT:    [[TMP267:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 1
 // CHECK11-NEXT:    store i32 5, ptr [[TMP267]], align 4
 // CHECK11-NEXT:    [[TMP268:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 2
 // CHECK11-NEXT:    store ptr [[TMP260]], ptr [[TMP268]], align 4
 // CHECK11-NEXT:    [[TMP269:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 3
 // CHECK11-NEXT:    store ptr [[TMP261]], ptr [[TMP269]], align 4
 // CHECK11-NEXT:    [[TMP270:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 4
 // CHECK11-NEXT:    store ptr @.offload_sizes.25, ptr [[TMP270]], align 4
 // CHECK11-NEXT:    [[TMP271:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 5
 // CHECK11-NEXT:    store ptr @.offload_maptypes.26, ptr [[TMP271]], align 4
 // CHECK11-NEXT:    [[TMP272:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 6
 // CHECK11-NEXT:    store ptr null, ptr [[TMP272]], align 4
 // CHECK11-NEXT:    [[TMP273:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 7
 // CHECK11-NEXT:    store ptr null, ptr [[TMP273]], align 4
 // CHECK11-NEXT:    [[TMP274:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 8
 // CHECK11-NEXT:    store i64 [[TMP265]], ptr [[TMP274]], align 8
 // CHECK11-NEXT:    [[TMP275:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 9
 // CHECK11-NEXT:    store i64 0, ptr [[TMP275]], align 8
 // CHECK11-NEXT:    [[TMP276:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 10
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP276]], align 4
 // CHECK11-NEXT:    [[TMP277:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 11
 // CHECK11-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP277]], align 4
 // CHECK11-NEXT:    [[TMP278:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS86]], i32 0, i32 12
 // CHECK11-NEXT:    store i32 0, ptr [[TMP278]], align 4
 // CHECK11-NEXT:    [[TMP279:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB3]], i64 -1, i32 0, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.region_id, ptr [[KERNEL_ARGS86]])
 // CHECK11-NEXT:    [[TMP280:%.*]] = icmp ne i32 [[TMP279]], 0
 // CHECK11-NEXT:    br i1 [[TMP280]], label [[OMP_OFFLOAD_FAILED87:%.*]], label [[OMP_OFFLOAD_CONT88:%.*]]
 // CHECK11:       omp_offload.failed87:
 // CHECK11-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91(i32 [[TMP239]], i32 [[TMP241]], ptr [[TMP242]], ptr [[TMP243]], ptr [[TMP244]]) #[[ATTR2]]
 // CHECK11-NEXT:    br label [[OMP_OFFLOAD_CONT88]]
 // CHECK11:       omp_offload.cont88:
 // CHECK11-NEXT:    ret i32 0
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42
 // CHECK11-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l42.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[TMP20]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = call i32 @__kmpc_cancel(ptr @[[GLOB3]], i32 [[TMP21]], i32 2)
 // CHECK11-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
 // CHECK11-NEXT:    br i1 [[TMP23]], label [[DOTCANCEL_EXIT:%.*]], label [[DOTCANCEL_CONTINUE:%.*]]
 // CHECK11:       .cancel.exit:
 // CHECK11-NEXT:    br label [[CANCEL_EXIT:%.*]]
 // CHECK11:       .cancel.continue:
 // CHECK11-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK11-NEXT:    [[TMP25:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP24]], i32 [[TMP25]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK11-NEXT:    [[TMP28:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[TMP27]], i32 [[TMP28]]
 // CHECK11-NEXT:    [[TMP29:%.*]] = load i32, ptr [[ARRAYIDX6]], align 4
 // CHECK11-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP26]], [[TMP29]]
 // CHECK11-NEXT:    [[TMP30:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    [[TMP31:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[TMP30]], i32 [[TMP31]]
 // CHECK11-NEXT:    store i32 [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // 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:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP32]], 1
 // CHECK11-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP33:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP34:%.*]] = load i32, ptr [[TMP33]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP34]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       cancel.exit:
 // CHECK11-NEXT:    [[TMP35:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP36:%.*]] = load i32, ptr [[TMP35]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP36]])
 // CHECK11-NEXT:    br label [[CANCEL_CONT:%.*]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    br label [[CANCEL_CONT]]
 // CHECK11:       cancel.cont:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51
 // CHECK11-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l51.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP20]], i32 [[TMP21]]
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK11-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX6]], align 4
 // CHECK11-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP22]], [[TMP25]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK11-NEXT:    store i32 [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK11-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59
 // CHECK11-SAME: (i32 noundef [[CH:%.*]], i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP7]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 91, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP9]])
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP18]], 1
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp slt i32 [[TMP17]], [[ADD]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.omp_outlined.omp_outlined, i32 [[TMP19]], i32 [[TMP20]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP21]], [[TMP22]]
 // CHECK11-NEXT:    store i32 [[ADD6]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK11-NEXT:    store i32 [[ADD7]], ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP25:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP26:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP25]], [[TMP26]]
 // CHECK11-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP28:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP9:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
 // CHECK11-NEXT:    br i1 [[CMP9]], label [[COND_TRUE10:%.*]], label [[COND_FALSE11:%.*]]
 // CHECK11:       cond.true10:
 // CHECK11-NEXT:    [[TMP29:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END12:%.*]]
 // CHECK11:       cond.false11:
 // CHECK11-NEXT:    [[TMP30:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END12]]
 // CHECK11:       cond.end12:
 // CHECK11-NEXT:    [[COND13:%.*]] = phi i32 [ [[TMP29]], [[COND_TRUE10]] ], [ [[TMP30]], [[COND_FALSE11]] ]
 // CHECK11-NEXT:    store i32 [[COND13]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP31:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP31]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP32:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP33:%.*]] = load i32, ptr [[TMP32]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP33]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l59.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP20]], i32 [[TMP21]]
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK11-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX6]], align 4
 // CHECK11-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP22]], [[TMP25]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK11-NEXT:    store i32 [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK11-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67
 // CHECK11-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l67.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP11]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP19]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP20]], i32 [[TMP21]]
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK11-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX6]], align 4
 // CHECK11-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP22]], [[TMP25]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I3]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK11-NEXT:    store i32 [[ADD7]], ptr [[ARRAYIDX8]], align 4
 // 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:    [[TMP28:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP28]], 1
 // CHECK11-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP30]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75
 // CHECK11-SAME: (i32 noundef [[CH:%.*]], i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.omp_outlined.omp_outlined, i32 [[TMP19]], i32 [[TMP20]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP26:%.*]] = load i32, ptr [[TMP25]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP26]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l75.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB2]], i32 [[TMP12]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 [[TMP10]])
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK11:       omp.dispatch.cond:
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP13]], [[TMP14]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP15]], [[COND_TRUE]] ], [ [[TMP16]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK11-NEXT:    br i1 [[CMP6]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK11:       omp.dispatch.body:
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[CMP7:%.*]] = icmp sle i32 [[TMP20]], [[TMP21]]
 // CHECK11-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP22]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4
 // CHECK11-NEXT:    [[TMP23:%.*]] = load ptr, ptr [[TMP2]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP23]], i32 [[TMP24]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 // CHECK11-NEXT:    [[TMP26:%.*]] = load ptr, ptr [[TMP3]], align 4
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[TMP26]], i32 [[TMP27]]
 // CHECK11-NEXT:    [[TMP28:%.*]] = load i32, ptr [[ARRAYIDX8]], align 4
 // CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP25]], [[TMP28]]
 // CHECK11-NEXT:    [[TMP29:%.*]] = load ptr, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    [[TMP30:%.*]] = load i32, ptr [[I4]], align 4
 // CHECK11-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, ptr [[TMP29]], i32 [[TMP30]]
 // CHECK11-NEXT:    store i32 [[ADD9]], ptr [[ARRAYIDX10]], align 4
 // 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:    [[TMP31:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[ADD11:%.*]] = add nsw i32 [[TMP31]], 1
 // CHECK11-NEXT:    store i32 [[ADD11]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK11:       omp.dispatch.inc:
 // CHECK11-NEXT:    [[TMP32:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP33:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD12:%.*]] = add nsw i32 [[TMP32]], [[TMP33]]
 // CHECK11-NEXT:    store i32 [[ADD12]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP34:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP35:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD13:%.*]] = add nsw i32 [[TMP34]], [[TMP35]]
 // CHECK11-NEXT:    store i32 [[ADD13]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK11:       omp.dispatch.end:
 // CHECK11-NEXT:    [[TMP36:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP37:%.*]] = load i32, ptr [[TMP36]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP37]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83
 // CHECK11-SAME: (i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.omp_outlined, ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[TMP8]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP9]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP12]], [[COND_TRUE]] ], [ [[TMP13]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP14]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP15]], [[TMP16]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 6, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.omp_outlined.omp_outlined, i32 [[TMP17]], i32 [[TMP18]], ptr [[TMP0]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP19]], [[TMP20]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[TMP21]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l83.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I3:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB2:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB2]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP12:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[TMP12]], align 4
 // CHECK11-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP13]], i32 35, i32 [[TMP10]], i32 [[TMP11]], i32 1, i32 1)
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK11:       omp.dispatch.cond:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[TMP14]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP15]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK11-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP16]], 0
 // CHECK11-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK11:       omp.dispatch.body:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP17]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26:![0-9]+]]
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[CMP4:%.*]] = icmp sle i32 [[TMP18]], [[TMP19]]
 // CHECK11-NEXT:    br i1 [[CMP4]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP20]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[TMP21:%.*]] = load ptr, ptr [[TMP2]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP21]], i32 [[TMP22]]
 // CHECK11-NEXT:    [[TMP23:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[TMP24:%.*]] = load ptr, ptr [[TMP3]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, ptr [[TMP24]], i32 [[TMP25]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load i32, ptr [[ARRAYIDX5]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[ADD6:%.*]] = add nsw i32 [[TMP23]], [[TMP26]]
 // CHECK11-NEXT:    [[TMP27:%.*]] = load ptr, ptr [[TMP1]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[TMP28:%.*]] = load i32, ptr [[I3]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds i32, ptr [[TMP27]], i32 [[TMP28]]
 // CHECK11-NEXT:    store i32 [[ADD6]], ptr [[ARRAYIDX7]], align 4, !llvm.access.group [[ACC_GRP26]]
 // 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:    [[TMP29:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    [[ADD8:%.*]] = add nsw i32 [[TMP29]], 1
 // CHECK11-NEXT:    store i32 [[ADD8]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP26]]
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP27:![0-9]+]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK11:       omp.dispatch.inc:
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK11:       omp.dispatch.end:
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91
 // CHECK11-SAME: (i32 noundef [[CH:%.*]], i32 noundef [[N:%.*]], ptr noundef [[A:%.*]], ptr noundef [[B:%.*]], ptr noundef [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    store i32 [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_teams(ptr @[[GLOB3]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.omp_outlined, ptr [[CH_ADDR]], ptr [[N_ADDR]], ptr [[A_ADDR]], ptr [[B_ADDR]], ptr [[C_ADDR]])
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.omp_outlined
 // CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[CH:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[CH_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_COMB_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR__CASTED:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[CH]], ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[CH_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP5]], ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP1]], align 4
 // CHECK11-NEXT:    store i32 [[TMP6]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP7]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP8]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP11]], i32 92, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_COMB_LB]], ptr [[DOTOMP_COMB_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
 // CHECK11:       cond.true:
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    br label [[COND_END:%.*]]
 // CHECK11:       cond.false:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    br label [[COND_END]]
 // CHECK11:       cond.end:
 // CHECK11-NEXT:    [[COND:%.*]] = phi i32 [ [[TMP14]], [[COND_TRUE]] ], [ [[TMP15]], [[COND_FALSE]] ]
 // CHECK11-NEXT:    store i32 [[COND]], ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP16]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP17:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[CMP6:%.*]] = icmp sle i32 [[TMP17]], [[TMP18]]
 // CHECK11-NEXT:    br i1 [[CMP6]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_COMB_LB]], align 4
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_COMB_UB]], align 4
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_]], align 4
 // CHECK11-NEXT:    store i32 [[TMP21]], ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK11-NEXT:    [[TMP22:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__CASTED]], align 4
 // CHECK11-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB3]], i32 7, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.omp_outlined.omp_outlined, i32 [[TMP19]], i32 [[TMP20]], ptr [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], ptr [[TMP4]], i32 [[TMP22]])
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
 // CHECK11:       omp.inner.for.inc:
 // CHECK11-NEXT:    [[TMP23:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP23]], [[TMP24]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
 // CHECK11:       omp.loop.exit:
 // CHECK11-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP26:%.*]] = load i32, ptr [[TMP25]], align 4
 // CHECK11-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP26]])
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //
 //
 // CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_v_l91.omp_outlined.omp_outlined
 // CHECK11-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(4) [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[C:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR1]] {
 // CHECK11-NEXT:  entry:
 // CHECK11-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_LB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTPREVIOUS_UB__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[N_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[C_ADDR:%.*]] = alloca ptr, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[TMP:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    [[I4:%.*]] = alloca i32, align 4
 // CHECK11-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_LB_]], ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTPREVIOUS_UB_]], ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[N]], ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    store ptr [[C]], ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[N_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
 // CHECK11-NEXT:    [[TMP4:%.*]] = load i32, ptr [[TMP0]], align 4
 // CHECK11-NEXT:    store i32 [[TMP4]], ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP5]], 0
 // CHECK11-NEXT:    [[DIV:%.*]] = sdiv i32 [[SUB]], 1
 // CHECK11-NEXT:    [[SUB3:%.*]] = sub nsw i32 [[DIV]], 1
 // CHECK11-NEXT:    store i32 [[SUB3]], ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[I]], align 4
 // CHECK11-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_1]], align 4
 // CHECK11-NEXT:    [[CMP:%.*]] = icmp slt i32 0, [[TMP6]]
 // CHECK11-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
 // CHECK11:       omp.precond.then:
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR_2]], align 4
 // CHECK11-NEXT:    store i32 [[TMP7]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP8:%.*]] = load i32, ptr [[DOTPREVIOUS_LB__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTPREVIOUS_UB__ADDR]], align 4
 // CHECK11-NEXT:    store i32 [[TMP8]], ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP9]], ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
 // CHECK11-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
 // CHECK11-NEXT:    [[TMP10:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP11:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    [[TMP12:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
 // CHECK11-NEXT:    [[TMP13:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP14:%.*]] = load i32, ptr [[TMP13]], align 4
 // CHECK11-NEXT:    call void @__kmpc_dispatch_init_4(ptr @[[GLOB3]], i32 [[TMP14]], i32 35, i32 [[TMP11]], i32 [[TMP12]], i32 1, i32 [[TMP10]])
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
 // CHECK11:       omp.dispatch.cond:
 // CHECK11-NEXT:    [[TMP15:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
 // CHECK11-NEXT:    [[TMP16:%.*]] = load i32, ptr [[TMP15]], align 4
 // CHECK11-NEXT:    [[TMP17:%.*]] = call i32 @__kmpc_dispatch_next_4(ptr @[[GLOB3]], i32 [[TMP16]], ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]])
 // CHECK11-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP17]], 0
 // CHECK11-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
 // CHECK11:       omp.dispatch.body:
 // CHECK11-NEXT:    [[TMP18:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
 // CHECK11-NEXT:    store i32 [[TMP18]], ptr [[DOTOMP_IV]], align 4
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
 // CHECK11:       omp.inner.for.cond:
 // CHECK11-NEXT:    [[TMP19:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29:![0-9]+]]
 // CHECK11-NEXT:    [[TMP20:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[CMP5:%.*]] = icmp sle i32 [[TMP19]], [[TMP20]]
 // CHECK11-NEXT:    br i1 [[CMP5]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
 // CHECK11:       omp.inner.for.body:
 // CHECK11-NEXT:    [[TMP21:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP21]], 1
 // CHECK11-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
 // CHECK11-NEXT:    store i32 [[ADD]], ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[TMP22:%.*]] = load ptr, ptr [[TMP2]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[TMP23:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[TMP22]], i32 [[TMP23]]
 // CHECK11-NEXT:    [[TMP24:%.*]] = load i32, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[TMP25:%.*]] = load ptr, ptr [[TMP3]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[TMP26:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[TMP25]], i32 [[TMP26]]
 // CHECK11-NEXT:    [[TMP27:%.*]] = load i32, ptr [[ARRAYIDX6]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[ADD7:%.*]] = add nsw i32 [[TMP24]], [[TMP27]]
 // CHECK11-NEXT:    [[TMP28:%.*]] = load ptr, ptr [[TMP1]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[TMP29:%.*]] = load i32, ptr [[I4]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds i32, ptr [[TMP28]], i32 [[TMP29]]
 // CHECK11-NEXT:    store i32 [[ADD7]], ptr [[ARRAYIDX8]], align 4, !llvm.access.group [[ACC_GRP29]]
 // 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:    [[TMP30:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    [[ADD9:%.*]] = add nsw i32 [[TMP30]], 1
 // CHECK11-NEXT:    store i32 [[ADD9]], ptr [[DOTOMP_IV]], align 4, !llvm.access.group [[ACC_GRP29]]
 // CHECK11-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP30:![0-9]+]]
 // CHECK11:       omp.inner.for.end:
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
 // CHECK11:       omp.dispatch.inc:
 // CHECK11-NEXT:    br label [[OMP_DISPATCH_COND]]
 // CHECK11:       omp.dispatch.end:
 // CHECK11-NEXT:    br label [[OMP_PRECOND_END]]
 // CHECK11:       omp.precond.end:
 // CHECK11-NEXT:    ret void
 //