// 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 _
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --check-prefixes=CHECK1,CHECK1-NORMAL
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefixes=CHECK2,CHECK2-NORMAL

// RUN: %clang_cc1 -verify -fopenmp -fopenmp-enable-irbuilder -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --check-prefixes=CHECK1,CHECK1-IRBUILDER
// RUN: %clang_cc1 -fopenmp -fopenmp-enable-irbuilder -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp -fopenmp-enable-irbuilder -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefixes=CHECK2,CHECK2-IRBUILDER

// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -fopenmp-version=45 -o - | FileCheck %s --check-prefixes=CHECK3,CHECK3-NORMAL
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -fopenmp-version=45 -o %t %s
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefixes=CHECK4,CHECK4-NORMAL

// RUN: %clang_cc1 -verify -fopenmp -fopenmp-enable-irbuilder -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -fopenmp-version=45 -o - | FileCheck %s --check-prefixes=CHECK3,CHECK3-IRBUILDER
// RUN: %clang_cc1 -fopenmp -fopenmp-enable-irbuilder -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -fopenmp-version=45 -o %t %s
// RUN: %clang_cc1 -fopenmp -fopenmp-enable-irbuilder -fopenmp-version=45 -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefixes=CHECK4,CHECK4-IRBUILDER

// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --check-prefix=CHECK5
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK6
// expected-no-diagnostics
#ifndef HEADER
#define HEADER

void static_not_chunked(float *a, float *b, float *c, float *d) {
  #pragma omp for schedule(static) ordered

// Loop header

  for (int i = 32000000; i > 33; i += -7) {
// Start of body: calculate i from IV:

// ... start of ordered region ...
// ... loop body ...
// End of body: store into a[i]:
// ... end of ordered region ...
    #pragma omp ordered
    a[i] = b[i] * c[i] * d[i];
  }
}

void dynamic1(float *a, float *b, float *c, float *d) {
  #pragma omp for schedule(dynamic) ordered

// Loop header

  for (unsigned long long i = 131071; i < 2147483647; i += 127) {
// Start of body: calculate i from IV:

// ... start of ordered region ...
// ... loop body ...
// End of body: store into a[i]:
// ... end of ordered region ...
    #pragma omp ordered threads
    a[i] = b[i] * c[i] * d[i];

// ... end iteration for ordered loop ...
  }
}

void test_auto(float *a, float *b, float *c, float *d) {
  unsigned int x = 0;
  unsigned int y = 0;
  #pragma omp for schedule(auto) collapse(2) ordered

// Loop header

// FIXME: When the iteration count of some nested loop is not a known constant,
// we should pre-calculate it, like we do for the total number of iterations!
  for (char i = static_cast<char>(y); i <= '9'; ++i)
    for (x = 11; x > 0; --x) {
// Start of body: indices are calculated from IV:

// ... start of ordered region ...
// ... loop body ...
// End of body: store into a[i]:
// ... end of ordered region ...
    #pragma omp ordered
    a[i] = b[i] * c[i] * d[i];

// ... end iteration for ordered loop ...
  }
}

void runtime(float *a, float *b, float *c, float *d) {
  int x = 0;
  #pragma omp for collapse(2) schedule(runtime) ordered

// Loop header

  for (unsigned char i = '0' ; i <= '9'; ++i)
    for (x = -10; x < 10; ++x) {
// Start of body: indices are calculated from IV:

// ... start of ordered region ...
// ... loop body ...
// End of body: store into a[i]:
// ... end of ordered region ...
    #pragma omp ordered threads
    a[i] = b[i] * c[i] * d[i];

// ... end iteration for ordered loop ...
  }
}

float f[10];
void foo_simd(int low, int up) {
#pragma omp simd
  for (int i = low; i < up; ++i) {
    f[i] = 0.0;
#pragma omp ordered simd
    f[i] = 1.0;
  }
#pragma omp for simd ordered
  for (int i = low; i < up; ++i) {
    f[i] = 0.0;
#pragma omp ordered simd
    f[i] = 1.0;
  }
}


#endif // HEADER

// CHECK1-LABEL: define {{[^@]+}}@_Z18static_not_chunkedPfS_S_S_
// CHECK1-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
// CHECK1-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK1-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK1-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK1-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 4571423, i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK1-IRBUILDER-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP0]], i32 66, i32 0, i32 4571423, i32 1, i32 1)
// CHECK1-NORMAL-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 66, i32 0, i32 4571423, i32 1, i32 1)
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK1:       omp.dispatch.cond:
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK1-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK1:       omp.dispatch.body:
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP3]], [[TMP4]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP5]], 7
// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 32000000, [[MUL]]
// CHECK1-NEXT:    store i32 [[SUB]], i32* [[I]], align 4
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK1-NEXT:    [[TMP6:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[I]], align 4
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP7]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP6]], i64 [[IDXPROM]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK1-NEXT:    [[TMP9:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4
// CHECK1-NEXT:    [[IDXPROM1:%.*]] = sext i32 [[TMP10]] to i64
// CHECK1-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP9]], i64 [[IDXPROM1]]
// CHECK1-NEXT:    [[TMP11:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK1-NEXT:    [[MUL3:%.*]] = fmul float [[TMP8]], [[TMP11]]
// CHECK1-NEXT:    [[TMP12:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
// CHECK1-NEXT:    [[IDXPROM4:%.*]] = sext i32 [[TMP13]] to i64
// CHECK1-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[TMP12]], i64 [[IDXPROM4]]
// CHECK1-NEXT:    [[TMP14:%.*]] = load float, float* [[ARRAYIDX5]], align 4
// CHECK1-NEXT:    [[MUL6:%.*]] = fmul float [[MUL3]], [[TMP14]]
// CHECK1-NEXT:    [[TMP15:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I]], align 4
// CHECK1-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP16]] to i64
// CHECK1-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[IDXPROM7]]
// CHECK1-NEXT:    store float [[MUL6]], float* [[ARRAYIDX8]], align 4
// CHECK1-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP17]], 1
// CHECK1-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK1-NEXT:    call void @__kmpc_dispatch_fini_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    br label [[OMP_DISPATCH_COND]]
// CHECK1:       omp.dispatch.end:
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP0]])
// CHECK1-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP0]])
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z8dynamic1PfS_S_S_
// CHECK1-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i64, align 8
// CHECK1-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK1-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK1-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK1-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK1-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK1-NEXT:    store i64 16908287, i64* [[DOTOMP_UB]], align 8
// CHECK1-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6:[0-9]+]])
// CHECK1-NEXT:    call void @__kmpc_dispatch_init_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 1073741891, i64 0, i64 16908287, i64 1, i64 1)
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK1:       omp.dispatch.cond:
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6]])
// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]])
// CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK1-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK1:       omp.dispatch.body:
// CHECK1-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK1-NEXT:    store i64 [[TMP2]], i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP3:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[TMP4:%.*]] = load i64, i64* [[DOTOMP_UB]], align 8
// CHECK1-NEXT:    [[ADD:%.*]] = add i64 [[TMP4]], 1
// CHECK1-NEXT:    [[CMP:%.*]] = icmp ult i64 [[TMP3]], [[ADD]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[MUL:%.*]] = mul i64 [[TMP5]], 127
// CHECK1-NEXT:    [[ADD1:%.*]] = add i64 131071, [[MUL]]
// CHECK1-NEXT:    store i64 [[ADD1]], i64* [[I]], align 8
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK1-NEXT:    [[TMP6:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK1-NEXT:    [[TMP7:%.*]] = load i64, i64* [[I]], align 8
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP6]], i64 [[TMP7]]
// CHECK1-NEXT:    [[TMP8:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK1-NEXT:    [[TMP9:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[TMP10:%.*]] = load i64, i64* [[I]], align 8
// CHECK1-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP9]], i64 [[TMP10]]
// CHECK1-NEXT:    [[TMP11:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK1-NEXT:    [[MUL3:%.*]] = fmul float [[TMP8]], [[TMP11]]
// CHECK1-NEXT:    [[TMP12:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK1-NEXT:    [[TMP13:%.*]] = load i64, i64* [[I]], align 8
// CHECK1-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[TMP12]], i64 [[TMP13]]
// CHECK1-NEXT:    [[TMP14:%.*]] = load float, float* [[ARRAYIDX4]], align 4
// CHECK1-NEXT:    [[MUL5:%.*]] = fmul float [[MUL3]], [[TMP14]]
// CHECK1-NEXT:    [[TMP15:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK1-NEXT:    [[TMP16:%.*]] = load i64, i64* [[I]], align 8
// CHECK1-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[TMP16]]
// CHECK1-NEXT:    store float [[MUL5]], float* [[ARRAYIDX6]], align 4
// CHECK1-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[ADD7:%.*]] = add i64 [[TMP17]], 1
// CHECK1-NEXT:    store i64 [[ADD7]], i64* [[DOTOMP_IV]], align 8
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6]])
// CHECK1-NEXT:    call void @__kmpc_dispatch_fini_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    br label [[OMP_DISPATCH_COND]]
// CHECK1:       omp.dispatch.end:
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK1-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z9test_autoPfS_S_S_
// CHECK1-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[Y:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i8, align 1
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i8, align 1
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK1-NEXT:    [[X6:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I8:%.*]] = alloca i8, align 1
// CHECK1-NEXT:    [[X9:%.*]] = alloca i32, align 4
// CHECK1-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK1-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK1-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK1-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[Y]], align 4
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[Y]], align 4
// CHECK1-NEXT:    [[CONV:%.*]] = trunc i32 [[TMP1]] to i8
// CHECK1-NEXT:    store i8 [[CONV]], i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK1-NEXT:    [[TMP2:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK1-NEXT:    [[CONV3:%.*]] = sext i8 [[TMP2]] to i32
// CHECK1-NEXT:    [[SUB:%.*]] = sub i32 57, [[CONV3]]
// CHECK1-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], 1
// CHECK1-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK1-NEXT:    [[CONV4:%.*]] = zext i32 [[DIV]] to i64
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV4]], 11
// CHECK1-NEXT:    [[SUB5:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK1-NEXT:    store i64 [[SUB5]], i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK1-NEXT:    [[TMP3:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK1-NEXT:    store i8 [[TMP3]], i8* [[I]], align 1
// CHECK1-NEXT:    store i32 11, i32* [[X6]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK1-NEXT:    [[CONV7:%.*]] = sext i8 [[TMP4]] to i32
// CHECK1-NEXT:    [[CMP:%.*]] = icmp sle i32 [[CONV7]], 57
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK1-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK1-NEXT:    store i64 [[TMP5]], i64* [[DOTOMP_UB]], align 8
// CHECK1-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8:[0-9]+]])
// CHECK1-NEXT:    call void @__kmpc_dispatch_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 1073741894, i64 0, i64 [[TMP6]], i64 1, i64 1)
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK1:       omp.dispatch.cond:
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8]])
// CHECK1-NEXT:    [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]])
// CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0
// CHECK1-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK1:       omp.dispatch.body:
// CHECK1-NEXT:    [[TMP8:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK1-NEXT:    store i64 [[TMP8]], i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP9:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_UB]], align 8
// CHECK1-NEXT:    [[CMP10:%.*]] = icmp sle i64 [[TMP9]], [[TMP10]]
// CHECK1-NEXT:    br i1 [[CMP10]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP11:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK1-NEXT:    [[CONV11:%.*]] = sext i8 [[TMP11]] to i64
// CHECK1-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[DIV12:%.*]] = sdiv i64 [[TMP12]], 11
// CHECK1-NEXT:    [[MUL13:%.*]] = mul nsw i64 [[DIV12]], 1
// CHECK1-NEXT:    [[ADD14:%.*]] = add nsw i64 [[CONV11]], [[MUL13]]
// CHECK1-NEXT:    [[CONV15:%.*]] = trunc i64 [[ADD14]] to i8
// CHECK1-NEXT:    store i8 [[CONV15]], i8* [[I8]], align 1
// CHECK1-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[DIV16:%.*]] = sdiv i64 [[TMP14]], 11
// CHECK1-NEXT:    [[MUL17:%.*]] = mul nsw i64 [[DIV16]], 11
// CHECK1-NEXT:    [[SUB18:%.*]] = sub nsw i64 [[TMP13]], [[MUL17]]
// CHECK1-NEXT:    [[MUL19:%.*]] = mul nsw i64 [[SUB18]], 1
// CHECK1-NEXT:    [[SUB20:%.*]] = sub nsw i64 11, [[MUL19]]
// CHECK1-NEXT:    [[CONV21:%.*]] = trunc i64 [[SUB20]] to i32
// CHECK1-NEXT:    store i32 [[CONV21]], i32* [[X9]], align 4
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK1-NEXT:    [[TMP15:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK1-NEXT:    [[TMP16:%.*]] = load i8, i8* [[I8]], align 1
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i8 [[TMP16]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[IDXPROM]]
// CHECK1-NEXT:    [[TMP17:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK1-NEXT:    [[TMP18:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[TMP19:%.*]] = load i8, i8* [[I8]], align 1
// CHECK1-NEXT:    [[IDXPROM22:%.*]] = sext i8 [[TMP19]] to i64
// CHECK1-NEXT:    [[ARRAYIDX23:%.*]] = getelementptr inbounds float, float* [[TMP18]], i64 [[IDXPROM22]]
// CHECK1-NEXT:    [[TMP20:%.*]] = load float, float* [[ARRAYIDX23]], align 4
// CHECK1-NEXT:    [[MUL24:%.*]] = fmul float [[TMP17]], [[TMP20]]
// CHECK1-NEXT:    [[TMP21:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK1-NEXT:    [[TMP22:%.*]] = load i8, i8* [[I8]], align 1
// CHECK1-NEXT:    [[IDXPROM25:%.*]] = sext i8 [[TMP22]] to i64
// CHECK1-NEXT:    [[ARRAYIDX26:%.*]] = getelementptr inbounds float, float* [[TMP21]], i64 [[IDXPROM25]]
// CHECK1-NEXT:    [[TMP23:%.*]] = load float, float* [[ARRAYIDX26]], align 4
// CHECK1-NEXT:    [[MUL27:%.*]] = fmul float [[MUL24]], [[TMP23]]
// CHECK1-NEXT:    [[TMP24:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK1-NEXT:    [[TMP25:%.*]] = load i8, i8* [[I8]], align 1
// CHECK1-NEXT:    [[IDXPROM28:%.*]] = sext i8 [[TMP25]] to i64
// CHECK1-NEXT:    [[ARRAYIDX29:%.*]] = getelementptr inbounds float, float* [[TMP24]], i64 [[IDXPROM28]]
// CHECK1-NEXT:    store float [[MUL27]], float* [[ARRAYIDX29]], align 4
// CHECK1-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK1-NEXT:    [[ADD30:%.*]] = add nsw i64 [[TMP26]], 1
// CHECK1-NEXT:    store i64 [[ADD30]], i64* [[DOTOMP_IV]], align 8
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8]])
// CHECK1-NEXT:    call void @__kmpc_dispatch_fini_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    br label [[OMP_DISPATCH_COND]]
// CHECK1:       omp.dispatch.end:
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK1-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z7runtimePfS_S_S_
// CHECK1-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK1-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[TMP:%.*]] = alloca i8, align 1
// CHECK1-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK1-NEXT:    [[X2:%.*]] = alloca i32, align 4
// CHECK1-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK1-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK1-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK1-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK1-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 199, i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10:[0-9]+]])
// CHECK1-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 1073741893, i32 0, i32 199, i32 1, i32 1)
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK1:       omp.dispatch.cond:
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10]])
// CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK1-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK1:       omp.dispatch.body:
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP3]], [[TMP4]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP5]], 20
// CHECK1-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 48, [[MUL]]
// CHECK1-NEXT:    [[CONV:%.*]] = trunc i32 [[ADD]] to i8
// CHECK1-NEXT:    store i8 [[CONV]], i8* [[I]], align 1
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[DIV3:%.*]] = sdiv i32 [[TMP7]], 20
// CHECK1-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[DIV3]], 20
// CHECK1-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[MUL4]]
// CHECK1-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK1-NEXT:    [[ADD6:%.*]] = add nsw i32 -10, [[MUL5]]
// CHECK1-NEXT:    store i32 [[ADD6]], i32* [[X2]], align 4
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK1-NEXT:    [[TMP8:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK1-NEXT:    [[TMP9:%.*]] = load i8, i8* [[I]], align 1
// CHECK1-NEXT:    [[IDXPROM:%.*]] = zext i8 [[TMP9]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP8]], i64 [[IDXPROM]]
// CHECK1-NEXT:    [[TMP10:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK1-NEXT:    [[TMP11:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK1-NEXT:    [[TMP12:%.*]] = load i8, i8* [[I]], align 1
// CHECK1-NEXT:    [[IDXPROM7:%.*]] = zext i8 [[TMP12]] to i64
// CHECK1-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[TMP11]], i64 [[IDXPROM7]]
// CHECK1-NEXT:    [[TMP13:%.*]] = load float, float* [[ARRAYIDX8]], align 4
// CHECK1-NEXT:    [[MUL9:%.*]] = fmul float [[TMP10]], [[TMP13]]
// CHECK1-NEXT:    [[TMP14:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK1-NEXT:    [[TMP15:%.*]] = load i8, i8* [[I]], align 1
// CHECK1-NEXT:    [[IDXPROM10:%.*]] = zext i8 [[TMP15]] to i64
// CHECK1-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds float, float* [[TMP14]], i64 [[IDXPROM10]]
// CHECK1-NEXT:    [[TMP16:%.*]] = load float, float* [[ARRAYIDX11]], align 4
// CHECK1-NEXT:    [[MUL12:%.*]] = fmul float [[MUL9]], [[TMP16]]
// CHECK1-NEXT:    [[TMP17:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK1-NEXT:    [[TMP18:%.*]] = load i8, i8* [[I]], align 1
// CHECK1-NEXT:    [[IDXPROM13:%.*]] = zext i8 [[TMP18]] to i64
// CHECK1-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr inbounds float, float* [[TMP17]], i64 [[IDXPROM13]]
// CHECK1-NEXT:    store float [[MUL12]], float* [[ARRAYIDX14]], align 4
// CHECK1-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK1-NEXT:    store i32 [[ADD15]], i32* [[DOTOMP_IV]], align 4
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10]])
// CHECK1-NEXT:    call void @__kmpc_dispatch_fini_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    br label [[OMP_DISPATCH_COND]]
// CHECK1:       omp.dispatch.end:
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK1-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z8foo_simdii
// CHECK1-SAME: (i32 [[LOW:%.*]], i32 [[UP:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[LOW_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[UP_ADDR:%.*]] = 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:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTOMP_IV16:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[_TMP17:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_18:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_19:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[DOTCAPTURE_EXPR_20:%.*]] = alloca i32, align 4
// CHECK1-NEXT:    [[I26:%.*]] = 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:    [[I28:%.*]] = alloca i32, align 4
// CHECK1-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT:    store i32 [[LOW]], i32* [[LOW_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[UP]], i32* [[UP_ADDR]], align 4
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[TMP2:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[SUB:%.*]] = sub i32 [[TMP3]], [[TMP4]]
// CHECK1-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK1-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK1-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK1-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK1-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK1-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    store i32 [[TMP5]], i32* [[I]], align 4
// CHECK1-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP6]], [[TMP7]]
// CHECK1-NEXT:    br i1 [[CMP]], label [[SIMD_IF_THEN:%.*]], label [[SIMD_IF_END:%.*]]
// CHECK1:       simd.if.then:
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IV]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1:       omp.inner.for.cond:
// CHECK1-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK1-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group !3
// CHECK1-NEXT:    [[ADD6:%.*]] = add i32 [[TMP9]], 1
// CHECK1-NEXT:    [[CMP7:%.*]] = icmp ult i32 [[TMP8]], [[ADD6]]
// CHECK1-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1:       omp.inner.for.body:
// CHECK1-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !llvm.access.group !3
// CHECK1-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK1-NEXT:    [[MUL:%.*]] = mul i32 [[TMP11]], 1
// CHECK1-NEXT:    [[ADD8:%.*]] = add i32 [[TMP10]], [[MUL]]
// CHECK1-NEXT:    store i32 [[ADD8]], i32* [[I5]], align 4, !llvm.access.group !3
// CHECK1-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I5]], align 4, !llvm.access.group !3
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP12]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK1-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX]], align 4, !llvm.access.group !3
// CHECK1-NEXT:    call void @__captured_stmt(i32* [[I5]]), !llvm.access.group !3
// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK1:       omp.body.continue:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1:       omp.inner.for.inc:
// CHECK1-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK1-NEXT:    [[ADD9:%.*]] = add i32 [[TMP13]], 1
// CHECK1-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
// CHECK1:       omp.inner.for.end:
// CHECK1-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK1-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK1-NEXT:    [[SUB10:%.*]] = sub i32 [[TMP15]], [[TMP16]]
// CHECK1-NEXT:    [[SUB11:%.*]] = sub i32 [[SUB10]], 1
// CHECK1-NEXT:    [[ADD12:%.*]] = add i32 [[SUB11]], 1
// CHECK1-NEXT:    [[DIV13:%.*]] = udiv i32 [[ADD12]], 1
// CHECK1-NEXT:    [[MUL14:%.*]] = mul i32 [[DIV13]], 1
// CHECK1-NEXT:    [[ADD15:%.*]] = add i32 [[TMP14]], [[MUL14]]
// CHECK1-NEXT:    store i32 [[ADD15]], i32* [[I5]], align 4
// CHECK1-NEXT:    br label [[SIMD_IF_END]]
// CHECK1:       simd.if.end:
// CHECK1-NEXT:    [[TMP17:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP17]], i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK1-NEXT:    [[TMP18:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK1-NEXT:    store i32 [[TMP18]], i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK1-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK1-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK1-NEXT:    [[SUB21:%.*]] = sub i32 [[TMP19]], [[TMP20]]
// CHECK1-NEXT:    [[SUB22:%.*]] = sub i32 [[SUB21]], 1
// CHECK1-NEXT:    [[ADD23:%.*]] = add i32 [[SUB22]], 1
// CHECK1-NEXT:    [[DIV24:%.*]] = udiv i32 [[ADD23]], 1
// CHECK1-NEXT:    [[SUB25:%.*]] = sub i32 [[DIV24]], 1
// CHECK1-NEXT:    store i32 [[SUB25]], i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK1-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK1-NEXT:    store i32 [[TMP21]], i32* [[I26]], align 4
// CHECK1-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK1-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK1-NEXT:    [[CMP27:%.*]] = icmp slt i32 [[TMP22]], [[TMP23]]
// CHECK1-NEXT:    br i1 [[CMP27]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK1:       omp.precond.then:
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK1-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_UB]], align 4
// CHECK1-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK1-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12:[0-9]+]])
// CHECK1-NEXT:    call void @__kmpc_dispatch_init_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 66, i32 0, i32 [[TMP25]], i32 1, i32 1)
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK1:       omp.dispatch.cond:
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12]])
// CHECK1-NEXT:    [[TMP26:%.*]] = call i32 @__kmpc_dispatch_next_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK1-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK1-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK1:       omp.dispatch.body:
// CHECK1-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK1-NEXT:    store i32 [[TMP27]], i32* [[DOTOMP_IV16]], align 4
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND29:%.*]]
// CHECK1-IRBUILDER:       omp.inner.for.cond30:
// CHECK1-NORMAL:       omp.inner.for.cond29:
// CHECK1-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK1-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !7
// CHECK1-NEXT:    [[ADD30:%.*]] = add i32 [[TMP29]], 1
// CHECK1-NEXT:    [[CMP31:%.*]] = icmp ult i32 [[TMP28]], [[ADD30]]
// CHECK1-NEXT:    br i1 [[CMP31]], label [[OMP_INNER_FOR_BODY32:%.*]], label [[OMP_INNER_FOR_END40:%.*]]
// CHECK1-IRBUILDER:       omp.inner.for.body33:
// CHECK1-NORMAL:       omp.inner.for.body32:
// CHECK1-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4, !llvm.access.group !7
// CHECK1-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK1-NEXT:    [[MUL33:%.*]] = mul i32 [[TMP31]], 1
// CHECK1-NEXT:    [[ADD34:%.*]] = add i32 [[TMP30]], [[MUL33]]
// CHECK1-NEXT:    store i32 [[ADD34]], i32* [[I28]], align 4, !llvm.access.group !7
// CHECK1-NEXT:    [[TMP32:%.*]] = load i32, i32* [[I28]], align 4, !llvm.access.group !7
// CHECK1-NEXT:    [[IDXPROM35:%.*]] = sext i32 [[TMP32]] to i64
// CHECK1-NEXT:    [[ARRAYIDX36:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM35]]
// CHECK1-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX36]], align 4, !llvm.access.group !7
// CHECK1-NEXT:    call void @__captured_stmt.1(i32* [[I28]]), !llvm.access.group !7
// CHECK1-NEXT:    br label [[OMP_BODY_CONTINUE37:%.*]]
// CHECK1-IRBUILDER:       omp.body.continue38:
// CHECK1-NORMAL:       omp.body.continue37:
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_INC38:%.*]]
// CHECK1-IRBUILDER:       omp.inner.for.inc39:
// CHECK1-NORMAL:       omp.inner.for.inc38:
// CHECK1-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK1-NEXT:    [[ADD39:%.*]] = add i32 [[TMP33]], 1
// CHECK1-NEXT:    store i32 [[ADD39]], i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12]])
// CHECK1-NEXT:    call void @__kmpc_dispatch_fini_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]]), !llvm.access.group !7
// CHECK1-NEXT:    br label [[OMP_INNER_FOR_COND29]], !llvm.loop [[LOOP8:![0-9]+]]
// CHECK1-IRBUILDER:       omp.inner.for.end42:
// CHECK1-NORMAL:       omp.inner.for.end40:
// CHECK1-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK1:       omp.dispatch.inc:
// CHECK1-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK1:       omp.dispatch.end:
// CHECK1-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK1-NEXT:    [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
// CHECK1-NEXT:    br i1 [[TMP35]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK1:       .omp.final.then:
// CHECK1-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK1-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK1-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK1-NEXT:    [[SUB41:%.*]] = sub i32 [[TMP37]], [[TMP38]]
// CHECK1-NEXT:    [[SUB42:%.*]] = sub i32 [[SUB41]], 1
// CHECK1-NEXT:    [[ADD43:%.*]] = add i32 [[SUB42]], 1
// CHECK1-NEXT:    [[DIV44:%.*]] = udiv i32 [[ADD43]], 1
// CHECK1-NEXT:    [[MUL45:%.*]] = mul i32 [[DIV44]], 1
// CHECK1-NEXT:    [[ADD46:%.*]] = add i32 [[TMP36]], [[MUL45]]
// CHECK1-NEXT:    store i32 [[ADD46]], i32* [[I28]], align 4
// CHECK1-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK1:       .omp.final.done:
// CHECK1-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK1:       omp.precond.end:
// CHECK1-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK1-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__captured_stmt
// CHECK1-SAME: (i32* nonnull align 4 dereferenceable(4) [[I:%.*]]) #[[ATTR3:[0-9]+]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[I_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    store i32* [[I]], i32** [[I_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[I_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP1]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK1-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX]], align 4
// CHECK1-NEXT:    ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__captured_stmt.1
// CHECK1-SAME: (i32* nonnull align 4 dereferenceable(4) [[I:%.*]]) #[[ATTR3]] {
// CHECK1-NEXT:  entry:
// CHECK1-NEXT:    [[I_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT:    store i32* [[I]], i32** [[I_ADDR]], align 8
// CHECK1-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[I_ADDR]], align 8
// CHECK1-NEXT:    [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK1-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP1]] to i64
// CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK1-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX]], align 4
// CHECK1-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_Z18static_not_chunkedPfS_S_S_
// CHECK2-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
// CHECK2-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK2-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK2-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK2-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 4571423, i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK2-IRBUILDER-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP0]], i32 66, i32 0, i32 4571423, i32 1, i32 1)
// CHECK2-NORMAL-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 66, i32 0, i32 4571423, i32 1, i32 1)
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK2:       omp.dispatch.cond:
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK2-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK2-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK2:       omp.dispatch.body:
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP3]], [[TMP4]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP5]], 7
// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 32000000, [[MUL]]
// CHECK2-NEXT:    store i32 [[SUB]], i32* [[I]], align 4
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    [[TMP6:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[I]], align 4
// CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP7]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP6]], i64 [[IDXPROM]]
// CHECK2-NEXT:    [[TMP8:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK2-NEXT:    [[TMP9:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4
// CHECK2-NEXT:    [[IDXPROM1:%.*]] = sext i32 [[TMP10]] to i64
// CHECK2-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP9]], i64 [[IDXPROM1]]
// CHECK2-NEXT:    [[TMP11:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK2-NEXT:    [[MUL3:%.*]] = fmul float [[TMP8]], [[TMP11]]
// CHECK2-NEXT:    [[TMP12:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
// CHECK2-NEXT:    [[IDXPROM4:%.*]] = sext i32 [[TMP13]] to i64
// CHECK2-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[TMP12]], i64 [[IDXPROM4]]
// CHECK2-NEXT:    [[TMP14:%.*]] = load float, float* [[ARRAYIDX5]], align 4
// CHECK2-NEXT:    [[MUL6:%.*]] = fmul float [[MUL3]], [[TMP14]]
// CHECK2-NEXT:    [[TMP15:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I]], align 4
// CHECK2-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP16]] to i64
// CHECK2-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[IDXPROM7]]
// CHECK2-NEXT:    store float [[MUL6]], float* [[ARRAYIDX8]], align 4
// CHECK2-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK2:       omp.body.continue:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP17]], 1
// CHECK2-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK2-NEXT:    call void @__kmpc_dispatch_fini_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK2:       omp.dispatch.inc:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK2:       omp.dispatch.end:
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP0]])
// CHECK2-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP0]])
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_Z8dynamic1PfS_S_S_
// CHECK2-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i64, align 8
// CHECK2-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK2-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK2-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK2-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK2-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK2-NEXT:    store i64 16908287, i64* [[DOTOMP_UB]], align 8
// CHECK2-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6:[0-9]+]])
// CHECK2-NEXT:    call void @__kmpc_dispatch_init_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 1073741891, i64 0, i64 16908287, i64 1, i64 1)
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK2:       omp.dispatch.cond:
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6]])
// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]])
// CHECK2-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK2-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK2:       omp.dispatch.body:
// CHECK2-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK2-NEXT:    store i64 [[TMP2]], i64* [[DOTOMP_IV]], align 8
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP3:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK2-NEXT:    [[TMP4:%.*]] = load i64, i64* [[DOTOMP_UB]], align 8
// CHECK2-NEXT:    [[ADD:%.*]] = add i64 [[TMP4]], 1
// CHECK2-NEXT:    [[CMP:%.*]] = icmp ult i64 [[TMP3]], [[ADD]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK2-NEXT:    [[MUL:%.*]] = mul i64 [[TMP5]], 127
// CHECK2-NEXT:    [[ADD1:%.*]] = add i64 131071, [[MUL]]
// CHECK2-NEXT:    store i64 [[ADD1]], i64* [[I]], align 8
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    [[TMP6:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK2-NEXT:    [[TMP7:%.*]] = load i64, i64* [[I]], align 8
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP6]], i64 [[TMP7]]
// CHECK2-NEXT:    [[TMP8:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK2-NEXT:    [[TMP9:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[TMP10:%.*]] = load i64, i64* [[I]], align 8
// CHECK2-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP9]], i64 [[TMP10]]
// CHECK2-NEXT:    [[TMP11:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK2-NEXT:    [[MUL3:%.*]] = fmul float [[TMP8]], [[TMP11]]
// CHECK2-NEXT:    [[TMP12:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK2-NEXT:    [[TMP13:%.*]] = load i64, i64* [[I]], align 8
// CHECK2-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[TMP12]], i64 [[TMP13]]
// CHECK2-NEXT:    [[TMP14:%.*]] = load float, float* [[ARRAYIDX4]], align 4
// CHECK2-NEXT:    [[MUL5:%.*]] = fmul float [[MUL3]], [[TMP14]]
// CHECK2-NEXT:    [[TMP15:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK2-NEXT:    [[TMP16:%.*]] = load i64, i64* [[I]], align 8
// CHECK2-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[TMP16]]
// CHECK2-NEXT:    store float [[MUL5]], float* [[ARRAYIDX6]], align 4
// CHECK2-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK2:       omp.body.continue:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK2-NEXT:    [[ADD7:%.*]] = add i64 [[TMP17]], 1
// CHECK2-NEXT:    store i64 [[ADD7]], i64* [[DOTOMP_IV]], align 8
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6]])
// CHECK2-NEXT:    call void @__kmpc_dispatch_fini_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK2:       omp.dispatch.inc:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK2:       omp.dispatch.end:
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK2-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_Z9test_autoPfS_S_S_
// CHECK2-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[Y:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i8, align 1
// CHECK2-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i8, align 1
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK2-NEXT:    [[X6:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I8:%.*]] = alloca i8, align 1
// CHECK2-NEXT:    [[X9:%.*]] = alloca i32, align 4
// CHECK2-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK2-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK2-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK2-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[Y]], align 4
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[Y]], align 4
// CHECK2-NEXT:    [[CONV:%.*]] = trunc i32 [[TMP1]] to i8
// CHECK2-NEXT:    store i8 [[CONV]], i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK2-NEXT:    [[TMP2:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK2-NEXT:    [[CONV3:%.*]] = sext i8 [[TMP2]] to i32
// CHECK2-NEXT:    [[SUB:%.*]] = sub i32 57, [[CONV3]]
// CHECK2-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], 1
// CHECK2-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK2-NEXT:    [[CONV4:%.*]] = zext i32 [[DIV]] to i64
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV4]], 11
// CHECK2-NEXT:    [[SUB5:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK2-NEXT:    store i64 [[SUB5]], i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK2-NEXT:    [[TMP3:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK2-NEXT:    store i8 [[TMP3]], i8* [[I]], align 1
// CHECK2-NEXT:    store i32 11, i32* [[X6]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK2-NEXT:    [[CONV7:%.*]] = sext i8 [[TMP4]] to i32
// CHECK2-NEXT:    [[CMP:%.*]] = icmp sle i32 [[CONV7]], 57
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK2-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK2-NEXT:    store i64 [[TMP5]], i64* [[DOTOMP_UB]], align 8
// CHECK2-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8:[0-9]+]])
// CHECK2-NEXT:    call void @__kmpc_dispatch_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 1073741894, i64 0, i64 [[TMP6]], i64 1, i64 1)
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK2:       omp.dispatch.cond:
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8]])
// CHECK2-NEXT:    [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]])
// CHECK2-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0
// CHECK2-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK2:       omp.dispatch.body:
// CHECK2-NEXT:    [[TMP8:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK2-NEXT:    store i64 [[TMP8]], i64* [[DOTOMP_IV]], align 8
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP9:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK2-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_UB]], align 8
// CHECK2-NEXT:    [[CMP10:%.*]] = icmp sle i64 [[TMP9]], [[TMP10]]
// CHECK2-NEXT:    br i1 [[CMP10]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP11:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK2-NEXT:    [[CONV11:%.*]] = sext i8 [[TMP11]] to i64
// CHECK2-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK2-NEXT:    [[DIV12:%.*]] = sdiv i64 [[TMP12]], 11
// CHECK2-NEXT:    [[MUL13:%.*]] = mul nsw i64 [[DIV12]], 1
// CHECK2-NEXT:    [[ADD14:%.*]] = add nsw i64 [[CONV11]], [[MUL13]]
// CHECK2-NEXT:    [[CONV15:%.*]] = trunc i64 [[ADD14]] to i8
// CHECK2-NEXT:    store i8 [[CONV15]], i8* [[I8]], align 1
// CHECK2-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK2-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK2-NEXT:    [[DIV16:%.*]] = sdiv i64 [[TMP14]], 11
// CHECK2-NEXT:    [[MUL17:%.*]] = mul nsw i64 [[DIV16]], 11
// CHECK2-NEXT:    [[SUB18:%.*]] = sub nsw i64 [[TMP13]], [[MUL17]]
// CHECK2-NEXT:    [[MUL19:%.*]] = mul nsw i64 [[SUB18]], 1
// CHECK2-NEXT:    [[SUB20:%.*]] = sub nsw i64 11, [[MUL19]]
// CHECK2-NEXT:    [[CONV21:%.*]] = trunc i64 [[SUB20]] to i32
// CHECK2-NEXT:    store i32 [[CONV21]], i32* [[X9]], align 4
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    [[TMP15:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK2-NEXT:    [[TMP16:%.*]] = load i8, i8* [[I8]], align 1
// CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i8 [[TMP16]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[IDXPROM]]
// CHECK2-NEXT:    [[TMP17:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK2-NEXT:    [[TMP18:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[TMP19:%.*]] = load i8, i8* [[I8]], align 1
// CHECK2-NEXT:    [[IDXPROM22:%.*]] = sext i8 [[TMP19]] to i64
// CHECK2-NEXT:    [[ARRAYIDX23:%.*]] = getelementptr inbounds float, float* [[TMP18]], i64 [[IDXPROM22]]
// CHECK2-NEXT:    [[TMP20:%.*]] = load float, float* [[ARRAYIDX23]], align 4
// CHECK2-NEXT:    [[MUL24:%.*]] = fmul float [[TMP17]], [[TMP20]]
// CHECK2-NEXT:    [[TMP21:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK2-NEXT:    [[TMP22:%.*]] = load i8, i8* [[I8]], align 1
// CHECK2-NEXT:    [[IDXPROM25:%.*]] = sext i8 [[TMP22]] to i64
// CHECK2-NEXT:    [[ARRAYIDX26:%.*]] = getelementptr inbounds float, float* [[TMP21]], i64 [[IDXPROM25]]
// CHECK2-NEXT:    [[TMP23:%.*]] = load float, float* [[ARRAYIDX26]], align 4
// CHECK2-NEXT:    [[MUL27:%.*]] = fmul float [[MUL24]], [[TMP23]]
// CHECK2-NEXT:    [[TMP24:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK2-NEXT:    [[TMP25:%.*]] = load i8, i8* [[I8]], align 1
// CHECK2-NEXT:    [[IDXPROM28:%.*]] = sext i8 [[TMP25]] to i64
// CHECK2-NEXT:    [[ARRAYIDX29:%.*]] = getelementptr inbounds float, float* [[TMP24]], i64 [[IDXPROM28]]
// CHECK2-NEXT:    store float [[MUL27]], float* [[ARRAYIDX29]], align 4
// CHECK2-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK2:       omp.body.continue:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK2-NEXT:    [[ADD30:%.*]] = add nsw i64 [[TMP26]], 1
// CHECK2-NEXT:    store i64 [[ADD30]], i64* [[DOTOMP_IV]], align 8
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8]])
// CHECK2-NEXT:    call void @__kmpc_dispatch_fini_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK2:       omp.dispatch.inc:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK2:       omp.dispatch.end:
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK2-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_Z7runtimePfS_S_S_
// CHECK2-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK2-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i8, align 1
// CHECK2-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK2-NEXT:    [[X2:%.*]] = alloca i32, align 4
// CHECK2-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK2-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK2-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK2-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK2-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 199, i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10:[0-9]+]])
// CHECK2-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 1073741893, i32 0, i32 199, i32 1, i32 1)
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK2:       omp.dispatch.cond:
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10]])
// CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK2-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK2-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK2:       omp.dispatch.body:
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP3]], [[TMP4]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP5]], 20
// CHECK2-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 48, [[MUL]]
// CHECK2-NEXT:    [[CONV:%.*]] = trunc i32 [[ADD]] to i8
// CHECK2-NEXT:    store i8 [[CONV]], i8* [[I]], align 1
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[DIV3:%.*]] = sdiv i32 [[TMP7]], 20
// CHECK2-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[DIV3]], 20
// CHECK2-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[MUL4]]
// CHECK2-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK2-NEXT:    [[ADD6:%.*]] = add nsw i32 -10, [[MUL5]]
// CHECK2-NEXT:    store i32 [[ADD6]], i32* [[X2]], align 4
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    [[TMP8:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK2-NEXT:    [[TMP9:%.*]] = load i8, i8* [[I]], align 1
// CHECK2-NEXT:    [[IDXPROM:%.*]] = zext i8 [[TMP9]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP8]], i64 [[IDXPROM]]
// CHECK2-NEXT:    [[TMP10:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK2-NEXT:    [[TMP11:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK2-NEXT:    [[TMP12:%.*]] = load i8, i8* [[I]], align 1
// CHECK2-NEXT:    [[IDXPROM7:%.*]] = zext i8 [[TMP12]] to i64
// CHECK2-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[TMP11]], i64 [[IDXPROM7]]
// CHECK2-NEXT:    [[TMP13:%.*]] = load float, float* [[ARRAYIDX8]], align 4
// CHECK2-NEXT:    [[MUL9:%.*]] = fmul float [[TMP10]], [[TMP13]]
// CHECK2-NEXT:    [[TMP14:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK2-NEXT:    [[TMP15:%.*]] = load i8, i8* [[I]], align 1
// CHECK2-NEXT:    [[IDXPROM10:%.*]] = zext i8 [[TMP15]] to i64
// CHECK2-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds float, float* [[TMP14]], i64 [[IDXPROM10]]
// CHECK2-NEXT:    [[TMP16:%.*]] = load float, float* [[ARRAYIDX11]], align 4
// CHECK2-NEXT:    [[MUL12:%.*]] = fmul float [[MUL9]], [[TMP16]]
// CHECK2-NEXT:    [[TMP17:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK2-NEXT:    [[TMP18:%.*]] = load i8, i8* [[I]], align 1
// CHECK2-NEXT:    [[IDXPROM13:%.*]] = zext i8 [[TMP18]] to i64
// CHECK2-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr inbounds float, float* [[TMP17]], i64 [[IDXPROM13]]
// CHECK2-NEXT:    store float [[MUL12]], float* [[ARRAYIDX14]], align 4
// CHECK2-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK2:       omp.body.continue:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK2-NEXT:    store i32 [[ADD15]], i32* [[DOTOMP_IV]], align 4
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10]])
// CHECK2-NEXT:    call void @__kmpc_dispatch_fini_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK2:       omp.dispatch.inc:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK2:       omp.dispatch.end:
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK2-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_Z8foo_simdii
// CHECK2-SAME: (i32 [[LOW:%.*]], i32 [[UP:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[LOW_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[UP_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IV16:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[_TMP17:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_18:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_19:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTCAPTURE_EXPR_20:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I26:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK2-NEXT:    [[I28:%.*]] = alloca i32, align 4
// CHECK2-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT:    store i32 [[LOW]], i32* [[LOW_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[UP]], i32* [[UP_ADDR]], align 4
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[TMP2:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[SUB:%.*]] = sub i32 [[TMP3]], [[TMP4]]
// CHECK2-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK2-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK2-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK2-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK2-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK2-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    store i32 [[TMP5]], i32* [[I]], align 4
// CHECK2-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP6]], [[TMP7]]
// CHECK2-NEXT:    br i1 [[CMP]], label [[SIMD_IF_THEN:%.*]], label [[SIMD_IF_END:%.*]]
// CHECK2:       simd.if.then:
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IV]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2:       omp.inner.for.cond:
// CHECK2-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK2-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group !3
// CHECK2-NEXT:    [[ADD6:%.*]] = add i32 [[TMP9]], 1
// CHECK2-NEXT:    [[CMP7:%.*]] = icmp ult i32 [[TMP8]], [[ADD6]]
// CHECK2-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2:       omp.inner.for.body:
// CHECK2-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !llvm.access.group !3
// CHECK2-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK2-NEXT:    [[MUL:%.*]] = mul i32 [[TMP11]], 1
// CHECK2-NEXT:    [[ADD8:%.*]] = add i32 [[TMP10]], [[MUL]]
// CHECK2-NEXT:    store i32 [[ADD8]], i32* [[I5]], align 4, !llvm.access.group !3
// CHECK2-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I5]], align 4, !llvm.access.group !3
// CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP12]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK2-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX]], align 4, !llvm.access.group !3
// CHECK2-NEXT:    call void @__captured_stmt(i32* [[I5]]), !llvm.access.group !3
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK2:       omp.body.continue:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2:       omp.inner.for.inc:
// CHECK2-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK2-NEXT:    [[ADD9:%.*]] = add i32 [[TMP13]], 1
// CHECK2-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
// CHECK2:       omp.inner.for.end:
// CHECK2-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK2-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK2-NEXT:    [[SUB10:%.*]] = sub i32 [[TMP15]], [[TMP16]]
// CHECK2-NEXT:    [[SUB11:%.*]] = sub i32 [[SUB10]], 1
// CHECK2-NEXT:    [[ADD12:%.*]] = add i32 [[SUB11]], 1
// CHECK2-NEXT:    [[DIV13:%.*]] = udiv i32 [[ADD12]], 1
// CHECK2-NEXT:    [[MUL14:%.*]] = mul i32 [[DIV13]], 1
// CHECK2-NEXT:    [[ADD15:%.*]] = add i32 [[TMP14]], [[MUL14]]
// CHECK2-NEXT:    store i32 [[ADD15]], i32* [[I5]], align 4
// CHECK2-NEXT:    br label [[SIMD_IF_END]]
// CHECK2:       simd.if.end:
// CHECK2-NEXT:    [[TMP17:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP17]], i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK2-NEXT:    [[TMP18:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK2-NEXT:    store i32 [[TMP18]], i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK2-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK2-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK2-NEXT:    [[SUB21:%.*]] = sub i32 [[TMP19]], [[TMP20]]
// CHECK2-NEXT:    [[SUB22:%.*]] = sub i32 [[SUB21]], 1
// CHECK2-NEXT:    [[ADD23:%.*]] = add i32 [[SUB22]], 1
// CHECK2-NEXT:    [[DIV24:%.*]] = udiv i32 [[ADD23]], 1
// CHECK2-NEXT:    [[SUB25:%.*]] = sub i32 [[DIV24]], 1
// CHECK2-NEXT:    store i32 [[SUB25]], i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK2-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK2-NEXT:    store i32 [[TMP21]], i32* [[I26]], align 4
// CHECK2-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK2-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK2-NEXT:    [[CMP27:%.*]] = icmp slt i32 [[TMP22]], [[TMP23]]
// CHECK2-NEXT:    br i1 [[CMP27]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK2:       omp.precond.then:
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK2-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_UB]], align 4
// CHECK2-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK2-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12:[0-9]+]])
// CHECK2-NEXT:    call void @__kmpc_dispatch_init_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 66, i32 0, i32 [[TMP25]], i32 1, i32 1)
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK2:       omp.dispatch.cond:
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12]])
// CHECK2-NEXT:    [[TMP26:%.*]] = call i32 @__kmpc_dispatch_next_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK2-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK2-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK2:       omp.dispatch.body:
// CHECK2-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK2-NEXT:    store i32 [[TMP27]], i32* [[DOTOMP_IV16]], align 4
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND29:%.*]]
// CHECK2-IRBUILDER:       omp.inner.for.cond30:
// CHECK2-NORMAL:       omp.inner.for.cond29:
// CHECK2-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK2-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !7
// CHECK2-NEXT:    [[ADD30:%.*]] = add i32 [[TMP29]], 1
// CHECK2-NEXT:    [[CMP31:%.*]] = icmp ult i32 [[TMP28]], [[ADD30]]
// CHECK2-NEXT:    br i1 [[CMP31]], label [[OMP_INNER_FOR_BODY32:%.*]], label [[OMP_INNER_FOR_END40:%.*]]
// CHECK2-IRBUILDER:       omp.inner.for.body33:
// CHECK2-NORMAL:       omp.inner.for.body32:
// CHECK2-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4, !llvm.access.group !7
// CHECK2-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK2-NEXT:    [[MUL33:%.*]] = mul i32 [[TMP31]], 1
// CHECK2-NEXT:    [[ADD34:%.*]] = add i32 [[TMP30]], [[MUL33]]
// CHECK2-NEXT:    store i32 [[ADD34]], i32* [[I28]], align 4, !llvm.access.group !7
// CHECK2-NEXT:    [[TMP32:%.*]] = load i32, i32* [[I28]], align 4, !llvm.access.group !7
// CHECK2-NEXT:    [[IDXPROM35:%.*]] = sext i32 [[TMP32]] to i64
// CHECK2-NEXT:    [[ARRAYIDX36:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM35]]
// CHECK2-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX36]], align 4, !llvm.access.group !7
// CHECK2-NEXT:    call void @__captured_stmt.1(i32* [[I28]]), !llvm.access.group !7
// CHECK2-NEXT:    br label [[OMP_BODY_CONTINUE37:%.*]]
// CHECK2-IRBUILDER:       omp.body.continue38:
// CHECK2-NORMAL:       omp.body.continue37:
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_INC38:%.*]]
// CHECK2-IRBUILDER:       omp.inner.for.inc39:
// CHECK2-NORMAL:       omp.inner.for.inc38:
// CHECK2-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK2-NEXT:    [[ADD39:%.*]] = add i32 [[TMP33]], 1
// CHECK2-NEXT:    store i32 [[ADD39]], i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12]])
// CHECK2-NEXT:    call void @__kmpc_dispatch_fini_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]]), !llvm.access.group !7
// CHECK2-NEXT:    br label [[OMP_INNER_FOR_COND29]], !llvm.loop [[LOOP8:![0-9]+]]
// CHECK2-IRBUILDER:       omp.inner.for.end42:
// CHECK2-NORMAL:       omp.inner.for.end40:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK2:       omp.dispatch.inc:
// CHECK2-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK2:       omp.dispatch.end:
// CHECK2-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK2-NEXT:    [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
// CHECK2-NEXT:    br i1 [[TMP35]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK2:       .omp.final.then:
// CHECK2-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK2-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK2-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK2-NEXT:    [[SUB41:%.*]] = sub i32 [[TMP37]], [[TMP38]]
// CHECK2-NEXT:    [[SUB42:%.*]] = sub i32 [[SUB41]], 1
// CHECK2-NEXT:    [[ADD43:%.*]] = add i32 [[SUB42]], 1
// CHECK2-NEXT:    [[DIV44:%.*]] = udiv i32 [[ADD43]], 1
// CHECK2-NEXT:    [[MUL45:%.*]] = mul i32 [[DIV44]], 1
// CHECK2-NEXT:    [[ADD46:%.*]] = add i32 [[TMP36]], [[MUL45]]
// CHECK2-NEXT:    store i32 [[ADD46]], i32* [[I28]], align 4
// CHECK2-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK2:       .omp.final.done:
// CHECK2-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK2:       omp.precond.end:
// CHECK2-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK2-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__captured_stmt
// CHECK2-SAME: (i32* nonnull align 4 dereferenceable(4) [[I:%.*]]) #[[ATTR3:[0-9]+]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[I_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    store i32* [[I]], i32** [[I_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[I_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP1]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK2-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX]], align 4
// CHECK2-NEXT:    ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__captured_stmt.1
// CHECK2-SAME: (i32* nonnull align 4 dereferenceable(4) [[I:%.*]]) #[[ATTR3]] {
// CHECK2-NEXT:  entry:
// CHECK2-NEXT:    [[I_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT:    store i32* [[I]], i32** [[I_ADDR]], align 8
// CHECK2-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[I_ADDR]], align 8
// CHECK2-NEXT:    [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK2-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP1]] to i64
// CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK2-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX]], align 4
// CHECK2-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z18static_not_chunkedPfS_S_S_
// CHECK3-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
// CHECK3-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK3-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK3-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK3-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 4571423, i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK3-IRBUILDER-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP0]], i32 66, i32 0, i32 4571423, i32 1, i32 1)
// CHECK3-NORMAL-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 66, i32 0, i32 4571423, i32 1, i32 1)
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK3:       omp.dispatch.cond:
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK3-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK3-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK3:       omp.dispatch.body:
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP3]], [[TMP4]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP5]], 7
// CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 32000000, [[MUL]]
// CHECK3-NEXT:    store i32 [[SUB]], i32* [[I]], align 4
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK3-NEXT:    [[TMP6:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[I]], align 4
// CHECK3-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP7]] to i64
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP6]], i64 [[IDXPROM]]
// CHECK3-NEXT:    [[TMP8:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK3-NEXT:    [[TMP9:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4
// CHECK3-NEXT:    [[IDXPROM1:%.*]] = sext i32 [[TMP10]] to i64
// CHECK3-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP9]], i64 [[IDXPROM1]]
// CHECK3-NEXT:    [[TMP11:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK3-NEXT:    [[MUL3:%.*]] = fmul float [[TMP8]], [[TMP11]]
// CHECK3-NEXT:    [[TMP12:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
// CHECK3-NEXT:    [[IDXPROM4:%.*]] = sext i32 [[TMP13]] to i64
// CHECK3-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[TMP12]], i64 [[IDXPROM4]]
// CHECK3-NEXT:    [[TMP14:%.*]] = load float, float* [[ARRAYIDX5]], align 4
// CHECK3-NEXT:    [[MUL6:%.*]] = fmul float [[MUL3]], [[TMP14]]
// CHECK3-NEXT:    [[TMP15:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I]], align 4
// CHECK3-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP16]] to i64
// CHECK3-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[IDXPROM7]]
// CHECK3-NEXT:    store float [[MUL6]], float* [[ARRAYIDX8]], align 4
// CHECK3-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP17]], 1
// CHECK3-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK3-NEXT:    call void @__kmpc_dispatch_fini_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    br label [[OMP_DISPATCH_COND]]
// CHECK3:       omp.dispatch.end:
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP0]])
// CHECK3-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP0]])
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z8dynamic1PfS_S_S_
// CHECK3-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i64, align 8
// CHECK3-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK3-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK3-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK3-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK3-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK3-NEXT:    store i64 16908287, i64* [[DOTOMP_UB]], align 8
// CHECK3-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6:[0-9]+]])
// CHECK3-NEXT:    call void @__kmpc_dispatch_init_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 67, i64 0, i64 16908287, i64 1, i64 1)
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK3:       omp.dispatch.cond:
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6]])
// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]])
// CHECK3-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK3-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK3:       omp.dispatch.body:
// CHECK3-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK3-NEXT:    store i64 [[TMP2]], i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP3:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[TMP4:%.*]] = load i64, i64* [[DOTOMP_UB]], align 8
// CHECK3-NEXT:    [[ADD:%.*]] = add i64 [[TMP4]], 1
// CHECK3-NEXT:    [[CMP:%.*]] = icmp ult i64 [[TMP3]], [[ADD]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[MUL:%.*]] = mul i64 [[TMP5]], 127
// CHECK3-NEXT:    [[ADD1:%.*]] = add i64 131071, [[MUL]]
// CHECK3-NEXT:    store i64 [[ADD1]], i64* [[I]], align 8
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK3-NEXT:    [[TMP6:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK3-NEXT:    [[TMP7:%.*]] = load i64, i64* [[I]], align 8
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP6]], i64 [[TMP7]]
// CHECK3-NEXT:    [[TMP8:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK3-NEXT:    [[TMP9:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK3-NEXT:    [[TMP10:%.*]] = load i64, i64* [[I]], align 8
// CHECK3-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP9]], i64 [[TMP10]]
// CHECK3-NEXT:    [[TMP11:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK3-NEXT:    [[MUL3:%.*]] = fmul float [[TMP8]], [[TMP11]]
// CHECK3-NEXT:    [[TMP12:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK3-NEXT:    [[TMP13:%.*]] = load i64, i64* [[I]], align 8
// CHECK3-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[TMP12]], i64 [[TMP13]]
// CHECK3-NEXT:    [[TMP14:%.*]] = load float, float* [[ARRAYIDX4]], align 4
// CHECK3-NEXT:    [[MUL5:%.*]] = fmul float [[MUL3]], [[TMP14]]
// CHECK3-NEXT:    [[TMP15:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK3-NEXT:    [[TMP16:%.*]] = load i64, i64* [[I]], align 8
// CHECK3-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[TMP16]]
// CHECK3-NEXT:    store float [[MUL5]], float* [[ARRAYIDX6]], align 4
// CHECK3-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[ADD7:%.*]] = add i64 [[TMP17]], 1
// CHECK3-NEXT:    store i64 [[ADD7]], i64* [[DOTOMP_IV]], align 8
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6]])
// CHECK3-NEXT:    call void @__kmpc_dispatch_fini_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    br label [[OMP_DISPATCH_COND]]
// CHECK3:       omp.dispatch.end:
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK3-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z9test_autoPfS_S_S_
// CHECK3-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[Y:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i8, align 1
// CHECK3-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i8, align 1
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK3-NEXT:    [[X6:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I8:%.*]] = alloca i8, align 1
// CHECK3-NEXT:    [[X9:%.*]] = alloca i32, align 4
// CHECK3-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK3-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK3-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK3-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK3-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[Y]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[Y]], align 4
// CHECK3-NEXT:    [[CONV:%.*]] = trunc i32 [[TMP1]] to i8
// CHECK3-NEXT:    store i8 [[CONV]], i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK3-NEXT:    [[TMP2:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK3-NEXT:    [[CONV3:%.*]] = sext i8 [[TMP2]] to i32
// CHECK3-NEXT:    [[SUB:%.*]] = sub i32 57, [[CONV3]]
// CHECK3-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], 1
// CHECK3-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK3-NEXT:    [[CONV4:%.*]] = zext i32 [[DIV]] to i64
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV4]], 11
// CHECK3-NEXT:    [[SUB5:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK3-NEXT:    store i64 [[SUB5]], i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK3-NEXT:    [[TMP3:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK3-NEXT:    store i8 [[TMP3]], i8* [[I]], align 1
// CHECK3-NEXT:    store i32 11, i32* [[X6]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK3-NEXT:    [[CONV7:%.*]] = sext i8 [[TMP4]] to i32
// CHECK3-NEXT:    [[CMP:%.*]] = icmp sle i32 [[CONV7]], 57
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK3-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK3-NEXT:    store i64 [[TMP5]], i64* [[DOTOMP_UB]], align 8
// CHECK3-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8:[0-9]+]])
// CHECK3-NEXT:    call void @__kmpc_dispatch_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 70, i64 0, i64 [[TMP6]], i64 1, i64 1)
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK3:       omp.dispatch.cond:
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8]])
// CHECK3-NEXT:    [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]])
// CHECK3-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0
// CHECK3-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK3:       omp.dispatch.body:
// CHECK3-NEXT:    [[TMP8:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK3-NEXT:    store i64 [[TMP8]], i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP9:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_UB]], align 8
// CHECK3-NEXT:    [[CMP10:%.*]] = icmp sle i64 [[TMP9]], [[TMP10]]
// CHECK3-NEXT:    br i1 [[CMP10]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP11:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK3-NEXT:    [[CONV11:%.*]] = sext i8 [[TMP11]] to i64
// CHECK3-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[DIV12:%.*]] = sdiv i64 [[TMP12]], 11
// CHECK3-NEXT:    [[MUL13:%.*]] = mul nsw i64 [[DIV12]], 1
// CHECK3-NEXT:    [[ADD14:%.*]] = add nsw i64 [[CONV11]], [[MUL13]]
// CHECK3-NEXT:    [[CONV15:%.*]] = trunc i64 [[ADD14]] to i8
// CHECK3-NEXT:    store i8 [[CONV15]], i8* [[I8]], align 1
// CHECK3-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[DIV16:%.*]] = sdiv i64 [[TMP14]], 11
// CHECK3-NEXT:    [[MUL17:%.*]] = mul nsw i64 [[DIV16]], 11
// CHECK3-NEXT:    [[SUB18:%.*]] = sub nsw i64 [[TMP13]], [[MUL17]]
// CHECK3-NEXT:    [[MUL19:%.*]] = mul nsw i64 [[SUB18]], 1
// CHECK3-NEXT:    [[SUB20:%.*]] = sub nsw i64 11, [[MUL19]]
// CHECK3-NEXT:    [[CONV21:%.*]] = trunc i64 [[SUB20]] to i32
// CHECK3-NEXT:    store i32 [[CONV21]], i32* [[X9]], align 4
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK3-NEXT:    [[TMP15:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK3-NEXT:    [[TMP16:%.*]] = load i8, i8* [[I8]], align 1
// CHECK3-NEXT:    [[IDXPROM:%.*]] = sext i8 [[TMP16]] to i64
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[IDXPROM]]
// CHECK3-NEXT:    [[TMP17:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK3-NEXT:    [[TMP18:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK3-NEXT:    [[TMP19:%.*]] = load i8, i8* [[I8]], align 1
// CHECK3-NEXT:    [[IDXPROM22:%.*]] = sext i8 [[TMP19]] to i64
// CHECK3-NEXT:    [[ARRAYIDX23:%.*]] = getelementptr inbounds float, float* [[TMP18]], i64 [[IDXPROM22]]
// CHECK3-NEXT:    [[TMP20:%.*]] = load float, float* [[ARRAYIDX23]], align 4
// CHECK3-NEXT:    [[MUL24:%.*]] = fmul float [[TMP17]], [[TMP20]]
// CHECK3-NEXT:    [[TMP21:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK3-NEXT:    [[TMP22:%.*]] = load i8, i8* [[I8]], align 1
// CHECK3-NEXT:    [[IDXPROM25:%.*]] = sext i8 [[TMP22]] to i64
// CHECK3-NEXT:    [[ARRAYIDX26:%.*]] = getelementptr inbounds float, float* [[TMP21]], i64 [[IDXPROM25]]
// CHECK3-NEXT:    [[TMP23:%.*]] = load float, float* [[ARRAYIDX26]], align 4
// CHECK3-NEXT:    [[MUL27:%.*]] = fmul float [[MUL24]], [[TMP23]]
// CHECK3-NEXT:    [[TMP24:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK3-NEXT:    [[TMP25:%.*]] = load i8, i8* [[I8]], align 1
// CHECK3-NEXT:    [[IDXPROM28:%.*]] = sext i8 [[TMP25]] to i64
// CHECK3-NEXT:    [[ARRAYIDX29:%.*]] = getelementptr inbounds float, float* [[TMP24]], i64 [[IDXPROM28]]
// CHECK3-NEXT:    store float [[MUL27]], float* [[ARRAYIDX29]], align 4
// CHECK3-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK3-NEXT:    [[ADD30:%.*]] = add nsw i64 [[TMP26]], 1
// CHECK3-NEXT:    store i64 [[ADD30]], i64* [[DOTOMP_IV]], align 8
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8]])
// CHECK3-NEXT:    call void @__kmpc_dispatch_fini_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    br label [[OMP_DISPATCH_COND]]
// CHECK3:       omp.dispatch.end:
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK3-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z7runtimePfS_S_S_
// CHECK3-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK3-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[TMP:%.*]] = alloca i8, align 1
// CHECK3-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK3-NEXT:    [[X2:%.*]] = alloca i32, align 4
// CHECK3-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK3-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK3-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK3-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK3-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 199, i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10:[0-9]+]])
// CHECK3-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 69, i32 0, i32 199, i32 1, i32 1)
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK3:       omp.dispatch.cond:
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10]])
// CHECK3-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK3-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK3-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK3:       omp.dispatch.body:
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP3]], [[TMP4]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP5]], 20
// CHECK3-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add nsw i32 48, [[MUL]]
// CHECK3-NEXT:    [[CONV:%.*]] = trunc i32 [[ADD]] to i8
// CHECK3-NEXT:    store i8 [[CONV]], i8* [[I]], align 1
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[DIV3:%.*]] = sdiv i32 [[TMP7]], 20
// CHECK3-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[DIV3]], 20
// CHECK3-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[MUL4]]
// CHECK3-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK3-NEXT:    [[ADD6:%.*]] = add nsw i32 -10, [[MUL5]]
// CHECK3-NEXT:    store i32 [[ADD6]], i32* [[X2]], align 4
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK3-NEXT:    [[TMP8:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK3-NEXT:    [[TMP9:%.*]] = load i8, i8* [[I]], align 1
// CHECK3-NEXT:    [[IDXPROM:%.*]] = zext i8 [[TMP9]] to i64
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP8]], i64 [[IDXPROM]]
// CHECK3-NEXT:    [[TMP10:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK3-NEXT:    [[TMP11:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK3-NEXT:    [[TMP12:%.*]] = load i8, i8* [[I]], align 1
// CHECK3-NEXT:    [[IDXPROM7:%.*]] = zext i8 [[TMP12]] to i64
// CHECK3-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[TMP11]], i64 [[IDXPROM7]]
// CHECK3-NEXT:    [[TMP13:%.*]] = load float, float* [[ARRAYIDX8]], align 4
// CHECK3-NEXT:    [[MUL9:%.*]] = fmul float [[TMP10]], [[TMP13]]
// CHECK3-NEXT:    [[TMP14:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK3-NEXT:    [[TMP15:%.*]] = load i8, i8* [[I]], align 1
// CHECK3-NEXT:    [[IDXPROM10:%.*]] = zext i8 [[TMP15]] to i64
// CHECK3-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds float, float* [[TMP14]], i64 [[IDXPROM10]]
// CHECK3-NEXT:    [[TMP16:%.*]] = load float, float* [[ARRAYIDX11]], align 4
// CHECK3-NEXT:    [[MUL12:%.*]] = fmul float [[MUL9]], [[TMP16]]
// CHECK3-NEXT:    [[TMP17:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK3-NEXT:    [[TMP18:%.*]] = load i8, i8* [[I]], align 1
// CHECK3-NEXT:    [[IDXPROM13:%.*]] = zext i8 [[TMP18]] to i64
// CHECK3-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr inbounds float, float* [[TMP17]], i64 [[IDXPROM13]]
// CHECK3-NEXT:    store float [[MUL12]], float* [[ARRAYIDX14]], align 4
// CHECK3-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK3-NEXT:    store i32 [[ADD15]], i32* [[DOTOMP_IV]], align 4
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10]])
// CHECK3-NEXT:    call void @__kmpc_dispatch_fini_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// 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:    br label [[OMP_DISPATCH_COND]]
// CHECK3:       omp.dispatch.end:
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK3-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z8foo_simdii
// CHECK3-SAME: (i32 [[LOW:%.*]], i32 [[UP:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[LOW_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[UP_ADDR:%.*]] = 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:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTOMP_IV16:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[_TMP17:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_18:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_19:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[DOTCAPTURE_EXPR_20:%.*]] = alloca i32, align 4
// CHECK3-NEXT:    [[I26:%.*]] = 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:    [[I28:%.*]] = alloca i32, align 4
// CHECK3-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT:    store i32 [[LOW]], i32* [[LOW_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[UP]], i32* [[UP_ADDR]], align 4
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[TMP2:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[SUB:%.*]] = sub i32 [[TMP3]], [[TMP4]]
// CHECK3-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK3-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK3-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK3-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK3-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK3-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    store i32 [[TMP5]], i32* [[I]], align 4
// CHECK3-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP6]], [[TMP7]]
// CHECK3-NEXT:    br i1 [[CMP]], label [[SIMD_IF_THEN:%.*]], label [[SIMD_IF_END:%.*]]
// CHECK3:       simd.if.then:
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IV]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3:       omp.inner.for.cond:
// CHECK3-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK3-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group !3
// CHECK3-NEXT:    [[ADD6:%.*]] = add i32 [[TMP9]], 1
// CHECK3-NEXT:    [[CMP7:%.*]] = icmp ult i32 [[TMP8]], [[ADD6]]
// CHECK3-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3:       omp.inner.for.body:
// CHECK3-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !llvm.access.group !3
// CHECK3-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK3-NEXT:    [[MUL:%.*]] = mul i32 [[TMP11]], 1
// CHECK3-NEXT:    [[ADD8:%.*]] = add i32 [[TMP10]], [[MUL]]
// CHECK3-NEXT:    store i32 [[ADD8]], i32* [[I5]], align 4, !llvm.access.group !3
// CHECK3-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I5]], align 4, !llvm.access.group !3
// CHECK3-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP12]] to i64
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK3-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX]], align 4, !llvm.access.group !3
// CHECK3-NEXT:    call void @__captured_stmt(i32* [[I5]]), !llvm.access.group !3
// CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK3:       omp.body.continue:
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3:       omp.inner.for.inc:
// CHECK3-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK3-NEXT:    [[ADD9:%.*]] = add i32 [[TMP13]], 1
// CHECK3-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
// CHECK3:       omp.inner.for.end:
// CHECK3-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK3-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK3-NEXT:    [[SUB10:%.*]] = sub i32 [[TMP15]], [[TMP16]]
// CHECK3-NEXT:    [[SUB11:%.*]] = sub i32 [[SUB10]], 1
// CHECK3-NEXT:    [[ADD12:%.*]] = add i32 [[SUB11]], 1
// CHECK3-NEXT:    [[DIV13:%.*]] = udiv i32 [[ADD12]], 1
// CHECK3-NEXT:    [[MUL14:%.*]] = mul i32 [[DIV13]], 1
// CHECK3-NEXT:    [[ADD15:%.*]] = add i32 [[TMP14]], [[MUL14]]
// CHECK3-NEXT:    store i32 [[ADD15]], i32* [[I5]], align 4
// CHECK3-NEXT:    br label [[SIMD_IF_END]]
// CHECK3:       simd.if.end:
// CHECK3-NEXT:    [[TMP17:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP17]], i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK3-NEXT:    [[TMP18:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK3-NEXT:    store i32 [[TMP18]], i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK3-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK3-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK3-NEXT:    [[SUB21:%.*]] = sub i32 [[TMP19]], [[TMP20]]
// CHECK3-NEXT:    [[SUB22:%.*]] = sub i32 [[SUB21]], 1
// CHECK3-NEXT:    [[ADD23:%.*]] = add i32 [[SUB22]], 1
// CHECK3-NEXT:    [[DIV24:%.*]] = udiv i32 [[ADD23]], 1
// CHECK3-NEXT:    [[SUB25:%.*]] = sub i32 [[DIV24]], 1
// CHECK3-NEXT:    store i32 [[SUB25]], i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK3-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK3-NEXT:    store i32 [[TMP21]], i32* [[I26]], align 4
// CHECK3-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK3-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK3-NEXT:    [[CMP27:%.*]] = icmp slt i32 [[TMP22]], [[TMP23]]
// CHECK3-NEXT:    br i1 [[CMP27]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK3:       omp.precond.then:
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK3-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_UB]], align 4
// CHECK3-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK3-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12:[0-9]+]])
// CHECK3-NEXT:    call void @__kmpc_dispatch_init_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 66, i32 0, i32 [[TMP25]], i32 1, i32 1)
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK3:       omp.dispatch.cond:
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12]])
// CHECK3-NEXT:    [[TMP26:%.*]] = call i32 @__kmpc_dispatch_next_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK3-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK3-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK3:       omp.dispatch.body:
// CHECK3-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK3-NEXT:    store i32 [[TMP27]], i32* [[DOTOMP_IV16]], align 4
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND29:%.*]]
// CHECK3-IRBUILDER:       omp.inner.for.cond30:
// CHECK3-NORMAL:       omp.inner.for.cond29:
// CHECK3-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK3-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !7
// CHECK3-NEXT:    [[ADD30:%.*]] = add i32 [[TMP29]], 1
// CHECK3-NEXT:    [[CMP31:%.*]] = icmp ult i32 [[TMP28]], [[ADD30]]
// CHECK3-NEXT:    br i1 [[CMP31]], label [[OMP_INNER_FOR_BODY32:%.*]], label [[OMP_INNER_FOR_END40:%.*]]
// CHECK3-IRBUILDER:       omp.inner.for.body33:
// CHECK3-NORMAL:       omp.inner.for.body32:
// CHECK3-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4, !llvm.access.group !7
// CHECK3-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK3-NEXT:    [[MUL33:%.*]] = mul i32 [[TMP31]], 1
// CHECK3-NEXT:    [[ADD34:%.*]] = add i32 [[TMP30]], [[MUL33]]
// CHECK3-NEXT:    store i32 [[ADD34]], i32* [[I28]], align 4, !llvm.access.group !7
// CHECK3-NEXT:    [[TMP32:%.*]] = load i32, i32* [[I28]], align 4, !llvm.access.group !7
// CHECK3-NEXT:    [[IDXPROM35:%.*]] = sext i32 [[TMP32]] to i64
// CHECK3-NEXT:    [[ARRAYIDX36:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM35]]
// CHECK3-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX36]], align 4, !llvm.access.group !7
// CHECK3-NEXT:    call void @__captured_stmt.1(i32* [[I28]]), !llvm.access.group !7
// CHECK3-NEXT:    br label [[OMP_BODY_CONTINUE37:%.*]]
// CHECK3-IRBUILDER:       omp.body.continue38:
// CHECK3-NORMAL:       omp.body.continue37:
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_INC38:%.*]]
// CHECK3-IRBUILDER:       omp.inner.for.inc39:
// CHECK3-NORMAL:       omp.inner.for.inc38:
// CHECK3-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK3-NEXT:    [[ADD39:%.*]] = add i32 [[TMP33]], 1
// CHECK3-NEXT:    store i32 [[ADD39]], i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12]])
// CHECK3-NEXT:    call void @__kmpc_dispatch_fini_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]]), !llvm.access.group !7
// CHECK3-NEXT:    br label [[OMP_INNER_FOR_COND29]], !llvm.loop [[LOOP8:![0-9]+]]
// CHECK3-IRBUILDER:       omp.inner.for.end42:
// CHECK3-NORMAL:       omp.inner.for.end40:
// CHECK3-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK3:       omp.dispatch.inc:
// CHECK3-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK3:       omp.dispatch.end:
// CHECK3-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK3-NEXT:    [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
// CHECK3-NEXT:    br i1 [[TMP35]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK3:       .omp.final.then:
// CHECK3-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK3-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK3-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK3-NEXT:    [[SUB41:%.*]] = sub i32 [[TMP37]], [[TMP38]]
// CHECK3-NEXT:    [[SUB42:%.*]] = sub i32 [[SUB41]], 1
// CHECK3-NEXT:    [[ADD43:%.*]] = add i32 [[SUB42]], 1
// CHECK3-NEXT:    [[DIV44:%.*]] = udiv i32 [[ADD43]], 1
// CHECK3-NEXT:    [[MUL45:%.*]] = mul i32 [[DIV44]], 1
// CHECK3-NEXT:    [[ADD46:%.*]] = add i32 [[TMP36]], [[MUL45]]
// CHECK3-NEXT:    store i32 [[ADD46]], i32* [[I28]], align 4
// CHECK3-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK3:       .omp.final.done:
// CHECK3-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK3:       omp.precond.end:
// CHECK3-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK3-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__captured_stmt
// CHECK3-SAME: (i32* nonnull align 4 dereferenceable(4) [[I:%.*]]) #[[ATTR3:[0-9]+]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[I_ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    store i32* [[I]], i32** [[I_ADDR]], align 8
// CHECK3-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[I_ADDR]], align 8
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK3-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP1]] to i64
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK3-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX]], align 4
// CHECK3-NEXT:    ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__captured_stmt.1
// CHECK3-SAME: (i32* nonnull align 4 dereferenceable(4) [[I:%.*]]) #[[ATTR3]] {
// CHECK3-NEXT:  entry:
// CHECK3-NEXT:    [[I_ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT:    store i32* [[I]], i32** [[I_ADDR]], align 8
// CHECK3-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[I_ADDR]], align 8
// CHECK3-NEXT:    [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK3-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP1]] to i64
// CHECK3-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK3-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX]], align 4
// CHECK3-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_Z18static_not_chunkedPfS_S_S_
// CHECK4-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
// CHECK4-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK4-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK4-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK4-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 4571423, i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]])
// CHECK4-IRBUILDER-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP0]], i32 66, i32 0, i32 4571423, i32 1, i32 1)
// CHECK4-NORMAL-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 66, i32 0, i32 4571423, i32 1, i32 1)
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK4:       omp.dispatch.cond:
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK4-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK4-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK4-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK4:       omp.dispatch.body:
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP3]], [[TMP4]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP5]], 7
// CHECK4-NEXT:    [[SUB:%.*]] = sub nsw i32 32000000, [[MUL]]
// CHECK4-NEXT:    store i32 [[SUB]], i32* [[I]], align 4
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    [[TMP6:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[I]], align 4
// CHECK4-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP7]] to i64
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP6]], i64 [[IDXPROM]]
// CHECK4-NEXT:    [[TMP8:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK4-NEXT:    [[TMP9:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[I]], align 4
// CHECK4-NEXT:    [[IDXPROM1:%.*]] = sext i32 [[TMP10]] to i64
// CHECK4-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP9]], i64 [[IDXPROM1]]
// CHECK4-NEXT:    [[TMP11:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK4-NEXT:    [[MUL3:%.*]] = fmul float [[TMP8]], [[TMP11]]
// CHECK4-NEXT:    [[TMP12:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[I]], align 4
// CHECK4-NEXT:    [[IDXPROM4:%.*]] = sext i32 [[TMP13]] to i64
// CHECK4-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[TMP12]], i64 [[IDXPROM4]]
// CHECK4-NEXT:    [[TMP14:%.*]] = load float, float* [[ARRAYIDX5]], align 4
// CHECK4-NEXT:    [[MUL6:%.*]] = fmul float [[MUL3]], [[TMP14]]
// CHECK4-NEXT:    [[TMP15:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[I]], align 4
// CHECK4-NEXT:    [[IDXPROM7:%.*]] = sext i32 [[TMP16]] to i64
// CHECK4-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[IDXPROM7]]
// CHECK4-NEXT:    store float [[MUL6]], float* [[ARRAYIDX8]], align 4
// CHECK4-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK4:       omp.body.continue:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP17]], 1
// CHECK4-NEXT:    store i32 [[ADD]], i32* [[DOTOMP_IV]], align 4
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3]])
// CHECK4-NEXT:    call void @__kmpc_dispatch_fini_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK4:       omp.dispatch.inc:
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK4:       omp.dispatch.end:
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4:[0-9]+]], i32 [[TMP0]])
// CHECK4-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP0]])
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_Z8dynamic1PfS_S_S_
// CHECK4-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i64, align 8
// CHECK4-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK4-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK4-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK4-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK4-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK4-NEXT:    store i64 16908287, i64* [[DOTOMP_UB]], align 8
// CHECK4-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6:[0-9]+]])
// CHECK4-NEXT:    call void @__kmpc_dispatch_init_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 67, i64 0, i64 16908287, i64 1, i64 1)
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK4:       omp.dispatch.cond:
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6]])
// CHECK4-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]])
// CHECK4-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK4-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK4:       omp.dispatch.body:
// CHECK4-NEXT:    [[TMP2:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK4-NEXT:    store i64 [[TMP2]], i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP3:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[TMP4:%.*]] = load i64, i64* [[DOTOMP_UB]], align 8
// CHECK4-NEXT:    [[ADD:%.*]] = add i64 [[TMP4]], 1
// CHECK4-NEXT:    [[CMP:%.*]] = icmp ult i64 [[TMP3]], [[ADD]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[MUL:%.*]] = mul i64 [[TMP5]], 127
// CHECK4-NEXT:    [[ADD1:%.*]] = add i64 131071, [[MUL]]
// CHECK4-NEXT:    store i64 [[ADD1]], i64* [[I]], align 8
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    [[TMP6:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK4-NEXT:    [[TMP7:%.*]] = load i64, i64* [[I]], align 8
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP6]], i64 [[TMP7]]
// CHECK4-NEXT:    [[TMP8:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK4-NEXT:    [[TMP9:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK4-NEXT:    [[TMP10:%.*]] = load i64, i64* [[I]], align 8
// CHECK4-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP9]], i64 [[TMP10]]
// CHECK4-NEXT:    [[TMP11:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK4-NEXT:    [[MUL3:%.*]] = fmul float [[TMP8]], [[TMP11]]
// CHECK4-NEXT:    [[TMP12:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK4-NEXT:    [[TMP13:%.*]] = load i64, i64* [[I]], align 8
// CHECK4-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[TMP12]], i64 [[TMP13]]
// CHECK4-NEXT:    [[TMP14:%.*]] = load float, float* [[ARRAYIDX4]], align 4
// CHECK4-NEXT:    [[MUL5:%.*]] = fmul float [[MUL3]], [[TMP14]]
// CHECK4-NEXT:    [[TMP15:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK4-NEXT:    [[TMP16:%.*]] = load i64, i64* [[I]], align 8
// CHECK4-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[TMP16]]
// CHECK4-NEXT:    store float [[MUL5]], float* [[ARRAYIDX6]], align 4
// CHECK4-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK4:       omp.body.continue:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP17:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[ADD7:%.*]] = add i64 [[TMP17]], 1
// CHECK4-NEXT:    store i64 [[ADD7]], i64* [[DOTOMP_IV]], align 8
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB6]])
// CHECK4-NEXT:    call void @__kmpc_dispatch_fini_8u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK4:       omp.dispatch.inc:
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK4:       omp.dispatch.end:
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK4-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_Z9test_autoPfS_S_S_
// CHECK4-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[Y:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i8, align 1
// CHECK4-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i8, align 1
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK4-NEXT:    [[X6:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_LB:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_UB:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i64, align 8
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I8:%.*]] = alloca i8, align 1
// CHECK4-NEXT:    [[X9:%.*]] = alloca i32, align 4
// CHECK4-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK4-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK4-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK4-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK4-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[Y]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[Y]], align 4
// CHECK4-NEXT:    [[CONV:%.*]] = trunc i32 [[TMP1]] to i8
// CHECK4-NEXT:    store i8 [[CONV]], i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK4-NEXT:    [[TMP2:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK4-NEXT:    [[CONV3:%.*]] = sext i8 [[TMP2]] to i32
// CHECK4-NEXT:    [[SUB:%.*]] = sub i32 57, [[CONV3]]
// CHECK4-NEXT:    [[ADD:%.*]] = add i32 [[SUB]], 1
// CHECK4-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK4-NEXT:    [[CONV4:%.*]] = zext i32 [[DIV]] to i64
// CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i64 [[CONV4]], 11
// CHECK4-NEXT:    [[SUB5:%.*]] = sub nsw i64 [[MUL]], 1
// CHECK4-NEXT:    store i64 [[SUB5]], i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK4-NEXT:    [[TMP3:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK4-NEXT:    store i8 [[TMP3]], i8* [[I]], align 1
// CHECK4-NEXT:    store i32 11, i32* [[X6]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK4-NEXT:    [[CONV7:%.*]] = sext i8 [[TMP4]] to i32
// CHECK4-NEXT:    [[CMP:%.*]] = icmp sle i32 [[CONV7]], 57
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK4:       omp.precond.then:
// CHECK4-NEXT:    store i64 0, i64* [[DOTOMP_LB]], align 8
// CHECK4-NEXT:    [[TMP5:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK4-NEXT:    store i64 [[TMP5]], i64* [[DOTOMP_UB]], align 8
// CHECK4-NEXT:    store i64 1, i64* [[DOTOMP_STRIDE]], align 8
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP6:%.*]] = load i64, i64* [[DOTCAPTURE_EXPR_2]], align 8
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8:[0-9]+]])
// CHECK4-NEXT:    call void @__kmpc_dispatch_init_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 70, i64 0, i64 [[TMP6]], i64 1, i64 1)
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK4:       omp.dispatch.cond:
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8]])
// CHECK4-NEXT:    [[TMP7:%.*]] = call i32 @__kmpc_dispatch_next_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i64* [[DOTOMP_LB]], i64* [[DOTOMP_UB]], i64* [[DOTOMP_STRIDE]])
// CHECK4-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP7]], 0
// CHECK4-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK4:       omp.dispatch.body:
// CHECK4-NEXT:    [[TMP8:%.*]] = load i64, i64* [[DOTOMP_LB]], align 8
// CHECK4-NEXT:    store i64 [[TMP8]], i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP9:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[TMP10:%.*]] = load i64, i64* [[DOTOMP_UB]], align 8
// CHECK4-NEXT:    [[CMP10:%.*]] = icmp sle i64 [[TMP9]], [[TMP10]]
// CHECK4-NEXT:    br i1 [[CMP10]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP11:%.*]] = load i8, i8* [[DOTCAPTURE_EXPR_]], align 1
// CHECK4-NEXT:    [[CONV11:%.*]] = sext i8 [[TMP11]] to i64
// CHECK4-NEXT:    [[TMP12:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[DIV12:%.*]] = sdiv i64 [[TMP12]], 11
// CHECK4-NEXT:    [[MUL13:%.*]] = mul nsw i64 [[DIV12]], 1
// CHECK4-NEXT:    [[ADD14:%.*]] = add nsw i64 [[CONV11]], [[MUL13]]
// CHECK4-NEXT:    [[CONV15:%.*]] = trunc i64 [[ADD14]] to i8
// CHECK4-NEXT:    store i8 [[CONV15]], i8* [[I8]], align 1
// CHECK4-NEXT:    [[TMP13:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[TMP14:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[DIV16:%.*]] = sdiv i64 [[TMP14]], 11
// CHECK4-NEXT:    [[MUL17:%.*]] = mul nsw i64 [[DIV16]], 11
// CHECK4-NEXT:    [[SUB18:%.*]] = sub nsw i64 [[TMP13]], [[MUL17]]
// CHECK4-NEXT:    [[MUL19:%.*]] = mul nsw i64 [[SUB18]], 1
// CHECK4-NEXT:    [[SUB20:%.*]] = sub nsw i64 11, [[MUL19]]
// CHECK4-NEXT:    [[CONV21:%.*]] = trunc i64 [[SUB20]] to i32
// CHECK4-NEXT:    store i32 [[CONV21]], i32* [[X9]], align 4
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    [[TMP15:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK4-NEXT:    [[TMP16:%.*]] = load i8, i8* [[I8]], align 1
// CHECK4-NEXT:    [[IDXPROM:%.*]] = sext i8 [[TMP16]] to i64
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP15]], i64 [[IDXPROM]]
// CHECK4-NEXT:    [[TMP17:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK4-NEXT:    [[TMP18:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK4-NEXT:    [[TMP19:%.*]] = load i8, i8* [[I8]], align 1
// CHECK4-NEXT:    [[IDXPROM22:%.*]] = sext i8 [[TMP19]] to i64
// CHECK4-NEXT:    [[ARRAYIDX23:%.*]] = getelementptr inbounds float, float* [[TMP18]], i64 [[IDXPROM22]]
// CHECK4-NEXT:    [[TMP20:%.*]] = load float, float* [[ARRAYIDX23]], align 4
// CHECK4-NEXT:    [[MUL24:%.*]] = fmul float [[TMP17]], [[TMP20]]
// CHECK4-NEXT:    [[TMP21:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK4-NEXT:    [[TMP22:%.*]] = load i8, i8* [[I8]], align 1
// CHECK4-NEXT:    [[IDXPROM25:%.*]] = sext i8 [[TMP22]] to i64
// CHECK4-NEXT:    [[ARRAYIDX26:%.*]] = getelementptr inbounds float, float* [[TMP21]], i64 [[IDXPROM25]]
// CHECK4-NEXT:    [[TMP23:%.*]] = load float, float* [[ARRAYIDX26]], align 4
// CHECK4-NEXT:    [[MUL27:%.*]] = fmul float [[MUL24]], [[TMP23]]
// CHECK4-NEXT:    [[TMP24:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK4-NEXT:    [[TMP25:%.*]] = load i8, i8* [[I8]], align 1
// CHECK4-NEXT:    [[IDXPROM28:%.*]] = sext i8 [[TMP25]] to i64
// CHECK4-NEXT:    [[ARRAYIDX29:%.*]] = getelementptr inbounds float, float* [[TMP24]], i64 [[IDXPROM28]]
// CHECK4-NEXT:    store float [[MUL27]], float* [[ARRAYIDX29]], align 4
// CHECK4-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK4:       omp.body.continue:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP26:%.*]] = load i64, i64* [[DOTOMP_IV]], align 8
// CHECK4-NEXT:    [[ADD30:%.*]] = add nsw i64 [[TMP26]], 1
// CHECK4-NEXT:    store i64 [[ADD30]], i64* [[DOTOMP_IV]], align 8
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB8]])
// CHECK4-NEXT:    call void @__kmpc_dispatch_fini_8(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK4:       omp.dispatch.inc:
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK4:       omp.dispatch.end:
// CHECK4-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.end:
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK4-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_Z7runtimePfS_S_S_
// CHECK4-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK4-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i8, align 1
// CHECK4-NEXT:    [[_TMP1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK4-NEXT:    [[X2:%.*]] = alloca i32, align 4
// CHECK4-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK4-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK4-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK4-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK4-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 199, i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10:[0-9]+]])
// CHECK4-NEXT:    call void @__kmpc_dispatch_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 69, i32 0, i32 199, i32 1, i32 1)
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK4:       omp.dispatch.cond:
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10]])
// CHECK4-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_dispatch_next_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK4-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK4-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK4:       omp.dispatch.body:
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp sle i32 [[TMP3]], [[TMP4]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[DIV:%.*]] = sdiv i32 [[TMP5]], 20
// CHECK4-NEXT:    [[MUL:%.*]] = mul nsw i32 [[DIV]], 1
// CHECK4-NEXT:    [[ADD:%.*]] = add nsw i32 48, [[MUL]]
// CHECK4-NEXT:    [[CONV:%.*]] = trunc i32 [[ADD]] to i8
// CHECK4-NEXT:    store i8 [[CONV]], i8* [[I]], align 1
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[DIV3:%.*]] = sdiv i32 [[TMP7]], 20
// CHECK4-NEXT:    [[MUL4:%.*]] = mul nsw i32 [[DIV3]], 20
// CHECK4-NEXT:    [[SUB:%.*]] = sub nsw i32 [[TMP6]], [[MUL4]]
// CHECK4-NEXT:    [[MUL5:%.*]] = mul nsw i32 [[SUB]], 1
// CHECK4-NEXT:    [[ADD6:%.*]] = add nsw i32 -10, [[MUL5]]
// CHECK4-NEXT:    store i32 [[ADD6]], i32* [[X2]], align 4
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-NEXT:    call void @__kmpc_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    [[TMP8:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK4-NEXT:    [[TMP9:%.*]] = load i8, i8* [[I]], align 1
// CHECK4-NEXT:    [[IDXPROM:%.*]] = zext i8 [[TMP9]] to i64
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP8]], i64 [[IDXPROM]]
// CHECK4-NEXT:    [[TMP10:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK4-NEXT:    [[TMP11:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK4-NEXT:    [[TMP12:%.*]] = load i8, i8* [[I]], align 1
// CHECK4-NEXT:    [[IDXPROM7:%.*]] = zext i8 [[TMP12]] to i64
// CHECK4-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[TMP11]], i64 [[IDXPROM7]]
// CHECK4-NEXT:    [[TMP13:%.*]] = load float, float* [[ARRAYIDX8]], align 4
// CHECK4-NEXT:    [[MUL9:%.*]] = fmul float [[TMP10]], [[TMP13]]
// CHECK4-NEXT:    [[TMP14:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK4-NEXT:    [[TMP15:%.*]] = load i8, i8* [[I]], align 1
// CHECK4-NEXT:    [[IDXPROM10:%.*]] = zext i8 [[TMP15]] to i64
// CHECK4-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds float, float* [[TMP14]], i64 [[IDXPROM10]]
// CHECK4-NEXT:    [[TMP16:%.*]] = load float, float* [[ARRAYIDX11]], align 4
// CHECK4-NEXT:    [[MUL12:%.*]] = fmul float [[MUL9]], [[TMP16]]
// CHECK4-NEXT:    [[TMP17:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK4-NEXT:    [[TMP18:%.*]] = load i8, i8* [[I]], align 1
// CHECK4-NEXT:    [[IDXPROM13:%.*]] = zext i8 [[TMP18]] to i64
// CHECK4-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr inbounds float, float* [[TMP17]], i64 [[IDXPROM13]]
// CHECK4-NEXT:    store float [[MUL12]], float* [[ARRAYIDX14]], align 4
// CHECK4-NEXT:    call void @__kmpc_end_ordered(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK4:       omp.body.continue:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    [[ADD15:%.*]] = add nsw i32 [[TMP19]], 1
// CHECK4-NEXT:    store i32 [[ADD15]], i32* [[DOTOMP_IV]], align 4
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB10]])
// CHECK4-NEXT:    call void @__kmpc_dispatch_fini_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK4:       omp.dispatch.inc:
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK4:       omp.dispatch.end:
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK4-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_Z8foo_simdii
// CHECK4-SAME: (i32 [[LOW:%.*]], i32 [[UP:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[LOW_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[UP_ADDR:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IV16:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[_TMP17:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_18:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_19:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTCAPTURE_EXPR_20:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I26:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK4-NEXT:    [[I28:%.*]] = alloca i32, align 4
// CHECK4-NORMAL-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-NEXT:    store i32 [[LOW]], i32* [[LOW_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[UP]], i32* [[UP_ADDR]], align 4
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[TMP2:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP2]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[SUB:%.*]] = sub i32 [[TMP3]], [[TMP4]]
// CHECK4-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK4-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK4-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK4-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK4-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK4-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    store i32 [[TMP5]], i32* [[I]], align 4
// CHECK4-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP6]], [[TMP7]]
// CHECK4-NEXT:    br i1 [[CMP]], label [[SIMD_IF_THEN:%.*]], label [[SIMD_IF_END:%.*]]
// CHECK4:       simd.if.then:
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IV]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4:       omp.inner.for.cond:
// CHECK4-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK4-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group !3
// CHECK4-NEXT:    [[ADD6:%.*]] = add i32 [[TMP9]], 1
// CHECK4-NEXT:    [[CMP7:%.*]] = icmp ult i32 [[TMP8]], [[ADD6]]
// CHECK4-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4:       omp.inner.for.body:
// CHECK4-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !llvm.access.group !3
// CHECK4-NEXT:    [[TMP11:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK4-NEXT:    [[MUL:%.*]] = mul i32 [[TMP11]], 1
// CHECK4-NEXT:    [[ADD8:%.*]] = add i32 [[TMP10]], [[MUL]]
// CHECK4-NEXT:    store i32 [[ADD8]], i32* [[I5]], align 4, !llvm.access.group !3
// CHECK4-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I5]], align 4, !llvm.access.group !3
// CHECK4-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP12]] to i64
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK4-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX]], align 4, !llvm.access.group !3
// CHECK4-NEXT:    call void @__captured_stmt(i32* [[I5]]), !llvm.access.group !3
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK4:       omp.body.continue:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4:       omp.inner.for.inc:
// CHECK4-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK4-NEXT:    [[ADD9:%.*]] = add i32 [[TMP13]], 1
// CHECK4-NEXT:    store i32 [[ADD9]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !3
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
// CHECK4:       omp.inner.for.end:
// CHECK4-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK4-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK4-NEXT:    [[SUB10:%.*]] = sub i32 [[TMP15]], [[TMP16]]
// CHECK4-NEXT:    [[SUB11:%.*]] = sub i32 [[SUB10]], 1
// CHECK4-NEXT:    [[ADD12:%.*]] = add i32 [[SUB11]], 1
// CHECK4-NEXT:    [[DIV13:%.*]] = udiv i32 [[ADD12]], 1
// CHECK4-NEXT:    [[MUL14:%.*]] = mul i32 [[DIV13]], 1
// CHECK4-NEXT:    [[ADD15:%.*]] = add i32 [[TMP14]], [[MUL14]]
// CHECK4-NEXT:    store i32 [[ADD15]], i32* [[I5]], align 4
// CHECK4-NEXT:    br label [[SIMD_IF_END]]
// CHECK4:       simd.if.end:
// CHECK4-NEXT:    [[TMP17:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP17]], i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK4-NEXT:    [[TMP18:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK4-NEXT:    store i32 [[TMP18]], i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK4-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK4-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK4-NEXT:    [[SUB21:%.*]] = sub i32 [[TMP19]], [[TMP20]]
// CHECK4-NEXT:    [[SUB22:%.*]] = sub i32 [[SUB21]], 1
// CHECK4-NEXT:    [[ADD23:%.*]] = add i32 [[SUB22]], 1
// CHECK4-NEXT:    [[DIV24:%.*]] = udiv i32 [[ADD23]], 1
// CHECK4-NEXT:    [[SUB25:%.*]] = sub i32 [[DIV24]], 1
// CHECK4-NEXT:    store i32 [[SUB25]], i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK4-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK4-NEXT:    store i32 [[TMP21]], i32* [[I26]], align 4
// CHECK4-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK4-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK4-NEXT:    [[CMP27:%.*]] = icmp slt i32 [[TMP22]], [[TMP23]]
// CHECK4-NEXT:    br i1 [[CMP27]], label [[OMP_PRECOND_THEN:%.*]], label [[OMP_PRECOND_END:%.*]]
// CHECK4:       omp.precond.then:
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK4-NEXT:    store i32 [[TMP24]], i32* [[DOTOMP_UB]], align 4
// CHECK4-NEXT:    store i32 1, i32* [[DOTOMP_STRIDE]], align 4
// CHECK4-NEXT:    store i32 0, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12:[0-9]+]])
// CHECK4-NEXT:    call void @__kmpc_dispatch_init_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32 66, i32 0, i32 [[TMP25]], i32 1, i32 1)
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND:%.*]]
// CHECK4:       omp.dispatch.cond:
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12]])
// CHECK4-NEXT:    [[TMP26:%.*]] = call i32 @__kmpc_dispatch_next_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i32* [[DOTOMP_IS_LAST]], i32* [[DOTOMP_LB]], i32* [[DOTOMP_UB]], i32* [[DOTOMP_STRIDE]])
// CHECK4-NEXT:    [[TOBOOL:%.*]] = icmp ne i32 [[TMP26]], 0
// CHECK4-NEXT:    br i1 [[TOBOOL]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]
// CHECK4:       omp.dispatch.body:
// CHECK4-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK4-NEXT:    store i32 [[TMP27]], i32* [[DOTOMP_IV16]], align 4
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND29:%.*]]
// CHECK4-IRBUILDER:       omp.inner.for.cond30:
// CHECK4-NORMAL:       omp.inner.for.cond29:
// CHECK4-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK4-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !7
// CHECK4-NEXT:    [[ADD30:%.*]] = add i32 [[TMP29]], 1
// CHECK4-NEXT:    [[CMP31:%.*]] = icmp ult i32 [[TMP28]], [[ADD30]]
// CHECK4-NEXT:    br i1 [[CMP31]], label [[OMP_INNER_FOR_BODY32:%.*]], label [[OMP_INNER_FOR_END40:%.*]]
// CHECK4-IRBUILDER:       omp.inner.for.body33:
// CHECK4-NORMAL:       omp.inner.for.body32:
// CHECK4-NEXT:    [[TMP30:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4, !llvm.access.group !7
// CHECK4-NEXT:    [[TMP31:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK4-NEXT:    [[MUL33:%.*]] = mul i32 [[TMP31]], 1
// CHECK4-NEXT:    [[ADD34:%.*]] = add i32 [[TMP30]], [[MUL33]]
// CHECK4-NEXT:    store i32 [[ADD34]], i32* [[I28]], align 4, !llvm.access.group !7
// CHECK4-NEXT:    [[TMP32:%.*]] = load i32, i32* [[I28]], align 4, !llvm.access.group !7
// CHECK4-NEXT:    [[IDXPROM35:%.*]] = sext i32 [[TMP32]] to i64
// CHECK4-NEXT:    [[ARRAYIDX36:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM35]]
// CHECK4-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX36]], align 4, !llvm.access.group !7
// CHECK4-NEXT:    call void @__captured_stmt.1(i32* [[I28]]), !llvm.access.group !7
// CHECK4-NEXT:    br label [[OMP_BODY_CONTINUE37:%.*]]
// CHECK4-IRBUILDER:       omp.body.continue38:
// CHECK4-NORMAL:       omp.body.continue37:
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_INC38:%.*]]
// CHECK4-IRBUILDER:       omp.inner.for.inc39:
// CHECK4-NORMAL:       omp.inner.for.inc38:
// CHECK4-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK4-NEXT:    [[ADD39:%.*]] = add i32 [[TMP33]], 1
// CHECK4-NEXT:    store i32 [[ADD39]], i32* [[DOTOMP_IV16]], align 4, !llvm.access.group !7
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB12]])
// CHECK4-NEXT:    call void @__kmpc_dispatch_fini_4u(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]]), !llvm.access.group !7
// CHECK4-NEXT:    br label [[OMP_INNER_FOR_COND29]], !llvm.loop [[LOOP8:![0-9]+]]
// CHECK4-IRBUILDER:       omp.inner.for.end42:
// CHECK4-NORMAL:       omp.inner.for.end40:
// CHECK4-NEXT:    br label [[OMP_DISPATCH_INC:%.*]]
// CHECK4:       omp.dispatch.inc:
// CHECK4-NEXT:    br label [[OMP_DISPATCH_COND]]
// CHECK4:       omp.dispatch.end:
// CHECK4-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTOMP_IS_LAST]], align 4
// CHECK4-NEXT:    [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
// CHECK4-NEXT:    br i1 [[TMP35]], label [[DOTOMP_FINAL_THEN:%.*]], label [[DOTOMP_FINAL_DONE:%.*]]
// CHECK4:       .omp.final.then:
// CHECK4-NEXT:    [[TMP36:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK4-NEXT:    [[TMP37:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK4-NEXT:    [[TMP38:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_18]], align 4
// CHECK4-NEXT:    [[SUB41:%.*]] = sub i32 [[TMP37]], [[TMP38]]
// CHECK4-NEXT:    [[SUB42:%.*]] = sub i32 [[SUB41]], 1
// CHECK4-NEXT:    [[ADD43:%.*]] = add i32 [[SUB42]], 1
// CHECK4-NEXT:    [[DIV44:%.*]] = udiv i32 [[ADD43]], 1
// CHECK4-NEXT:    [[MUL45:%.*]] = mul i32 [[DIV44]], 1
// CHECK4-NEXT:    [[ADD46:%.*]] = add i32 [[TMP36]], [[MUL45]]
// CHECK4-NEXT:    store i32 [[ADD46]], i32* [[I28]], align 4
// CHECK4-NEXT:    br label [[DOTOMP_FINAL_DONE]]
// CHECK4:       .omp.final.done:
// CHECK4-NEXT:    br label [[OMP_PRECOND_END]]
// CHECK4:       omp.precond.end:
// CHECK4-IRBUILDER-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-IRBUILDER-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB4]], i32 [[TMP0]])
// CHECK4-NORMAL-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP0]])
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__captured_stmt
// CHECK4-SAME: (i32* nonnull align 4 dereferenceable(4) [[I:%.*]]) #[[ATTR3:[0-9]+]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[I_ADDR:%.*]] = alloca i32*, align 8
// CHECK4-NEXT:    store i32* [[I]], i32** [[I_ADDR]], align 8
// CHECK4-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[I_ADDR]], align 8
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK4-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP1]] to i64
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK4-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX]], align 4
// CHECK4-NEXT:    ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__captured_stmt.1
// CHECK4-SAME: (i32* nonnull align 4 dereferenceable(4) [[I:%.*]]) #[[ATTR3]] {
// CHECK4-NEXT:  entry:
// CHECK4-NEXT:    [[I_ADDR:%.*]] = alloca i32*, align 8
// CHECK4-NEXT:    store i32* [[I]], i32** [[I_ADDR]], align 8
// CHECK4-NEXT:    [[TMP0:%.*]] = load i32*, i32** [[I_ADDR]], align 8
// CHECK4-NEXT:    [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK4-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP1]] to i64
// CHECK4-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK4-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX]], align 4
// CHECK4-NEXT:    ret void
//
//
// CHECK5-LABEL: define {{[^@]+}}@_Z18static_not_chunkedPfS_S_S_
// CHECK5-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK5-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK5-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK5-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK5-NEXT:    store i32 32000000, i32* [[I]], align 4
// CHECK5-NEXT:    br label [[FOR_COND:%.*]]
// CHECK5:       for.cond:
// CHECK5-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP0]], 33
// CHECK5-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK5:       for.body:
// CHECK5-NEXT:    [[TMP1:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[I]], align 4
// CHECK5-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP2]] to i64
// CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP1]], i64 [[IDXPROM]]
// CHECK5-NEXT:    [[TMP3:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[I]], align 4
// CHECK5-NEXT:    [[IDXPROM1:%.*]] = sext i32 [[TMP5]] to i64
// CHECK5-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP4]], i64 [[IDXPROM1]]
// CHECK5-NEXT:    [[TMP6:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK5-NEXT:    [[MUL:%.*]] = fmul float [[TMP3]], [[TMP6]]
// CHECK5-NEXT:    [[TMP7:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[I]], align 4
// CHECK5-NEXT:    [[IDXPROM3:%.*]] = sext i32 [[TMP8]] to i64
// CHECK5-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[TMP7]], i64 [[IDXPROM3]]
// CHECK5-NEXT:    [[TMP9:%.*]] = load float, float* [[ARRAYIDX4]], align 4
// CHECK5-NEXT:    [[MUL5:%.*]] = fmul float [[MUL]], [[TMP9]]
// CHECK5-NEXT:    [[TMP10:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK5-NEXT:    [[IDXPROM6:%.*]] = sext i32 [[TMP11]] to i64
// CHECK5-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds float, float* [[TMP10]], i64 [[IDXPROM6]]
// CHECK5-NEXT:    store float [[MUL5]], float* [[ARRAYIDX7]], align 4
// CHECK5-NEXT:    br label [[FOR_INC:%.*]]
// CHECK5:       for.inc:
// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
// CHECK5-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP12]], -7
// CHECK5-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK5-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP2:![0-9]+]]
// CHECK5:       for.end:
// CHECK5-NEXT:    ret void
//
//
// CHECK5-LABEL: define {{[^@]+}}@_Z8dynamic1PfS_S_S_
// CHECK5-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[I:%.*]] = alloca i64, align 8
// CHECK5-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK5-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK5-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK5-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK5-NEXT:    store i64 131071, i64* [[I]], align 8
// CHECK5-NEXT:    br label [[FOR_COND:%.*]]
// CHECK5:       for.cond:
// CHECK5-NEXT:    [[TMP0:%.*]] = load i64, i64* [[I]], align 8
// CHECK5-NEXT:    [[CMP:%.*]] = icmp ult i64 [[TMP0]], 2147483647
// CHECK5-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK5:       for.body:
// CHECK5-NEXT:    [[TMP1:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK5-NEXT:    [[TMP2:%.*]] = load i64, i64* [[I]], align 8
// CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP1]], i64 [[TMP2]]
// CHECK5-NEXT:    [[TMP3:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK5-NEXT:    [[TMP5:%.*]] = load i64, i64* [[I]], align 8
// CHECK5-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds float, float* [[TMP4]], i64 [[TMP5]]
// CHECK5-NEXT:    [[TMP6:%.*]] = load float, float* [[ARRAYIDX1]], align 4
// CHECK5-NEXT:    [[MUL:%.*]] = fmul float [[TMP3]], [[TMP6]]
// CHECK5-NEXT:    [[TMP7:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK5-NEXT:    [[TMP8:%.*]] = load i64, i64* [[I]], align 8
// CHECK5-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP7]], i64 [[TMP8]]
// CHECK5-NEXT:    [[TMP9:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK5-NEXT:    [[MUL3:%.*]] = fmul float [[MUL]], [[TMP9]]
// CHECK5-NEXT:    [[TMP10:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK5-NEXT:    [[TMP11:%.*]] = load i64, i64* [[I]], align 8
// CHECK5-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[TMP10]], i64 [[TMP11]]
// CHECK5-NEXT:    store float [[MUL3]], float* [[ARRAYIDX4]], align 4
// CHECK5-NEXT:    br label [[FOR_INC:%.*]]
// CHECK5:       for.inc:
// CHECK5-NEXT:    [[TMP12:%.*]] = load i64, i64* [[I]], align 8
// CHECK5-NEXT:    [[ADD:%.*]] = add i64 [[TMP12]], 127
// CHECK5-NEXT:    store i64 [[ADD]], i64* [[I]], align 8
// CHECK5-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
// CHECK5:       for.end:
// CHECK5-NEXT:    ret void
//
//
// CHECK5-LABEL: define {{[^@]+}}@_Z9test_autoPfS_S_S_
// CHECK5-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[Y:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK5-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK5-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK5-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK5-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK5-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[Y]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load i32, i32* [[Y]], align 4
// CHECK5-NEXT:    [[CONV:%.*]] = trunc i32 [[TMP0]] to i8
// CHECK5-NEXT:    store i8 [[CONV]], i8* [[I]], align 1
// CHECK5-NEXT:    br label [[FOR_COND:%.*]]
// CHECK5:       for.cond:
// CHECK5-NEXT:    [[TMP1:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[CONV1:%.*]] = sext i8 [[TMP1]] to i32
// CHECK5-NEXT:    [[CMP:%.*]] = icmp sle i32 [[CONV1]], 57
// CHECK5-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END13:%.*]]
// CHECK5:       for.body:
// CHECK5-NEXT:    store i32 11, i32* [[X]], align 4
// CHECK5-NEXT:    br label [[FOR_COND2:%.*]]
// CHECK5:       for.cond2:
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[X]], align 4
// CHECK5-NEXT:    [[CMP3:%.*]] = icmp ugt i32 [[TMP2]], 0
// CHECK5-NEXT:    br i1 [[CMP3]], label [[FOR_BODY4:%.*]], label [[FOR_END:%.*]]
// CHECK5:       for.body4:
// CHECK5-NEXT:    [[TMP3:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK5-NEXT:    [[TMP4:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[IDXPROM:%.*]] = sext i8 [[TMP4]] to i64
// CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP3]], i64 [[IDXPROM]]
// CHECK5-NEXT:    [[TMP5:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK5-NEXT:    [[TMP6:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK5-NEXT:    [[TMP7:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[IDXPROM5:%.*]] = sext i8 [[TMP7]] to i64
// CHECK5-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds float, float* [[TMP6]], i64 [[IDXPROM5]]
// CHECK5-NEXT:    [[TMP8:%.*]] = load float, float* [[ARRAYIDX6]], align 4
// CHECK5-NEXT:    [[MUL:%.*]] = fmul float [[TMP5]], [[TMP8]]
// CHECK5-NEXT:    [[TMP9:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK5-NEXT:    [[TMP10:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[IDXPROM7:%.*]] = sext i8 [[TMP10]] to i64
// CHECK5-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[TMP9]], i64 [[IDXPROM7]]
// CHECK5-NEXT:    [[TMP11:%.*]] = load float, float* [[ARRAYIDX8]], align 4
// CHECK5-NEXT:    [[MUL9:%.*]] = fmul float [[MUL]], [[TMP11]]
// CHECK5-NEXT:    [[TMP12:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK5-NEXT:    [[TMP13:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[IDXPROM10:%.*]] = sext i8 [[TMP13]] to i64
// CHECK5-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds float, float* [[TMP12]], i64 [[IDXPROM10]]
// CHECK5-NEXT:    store float [[MUL9]], float* [[ARRAYIDX11]], align 4
// CHECK5-NEXT:    br label [[FOR_INC:%.*]]
// CHECK5:       for.inc:
// CHECK5-NEXT:    [[TMP14:%.*]] = load i32, i32* [[X]], align 4
// CHECK5-NEXT:    [[DEC:%.*]] = add i32 [[TMP14]], -1
// CHECK5-NEXT:    store i32 [[DEC]], i32* [[X]], align 4
// CHECK5-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP5:![0-9]+]]
// CHECK5:       for.end:
// CHECK5-NEXT:    br label [[FOR_INC12:%.*]]
// CHECK5:       for.inc12:
// CHECK5-NEXT:    [[TMP15:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[INC:%.*]] = add i8 [[TMP15]], 1
// CHECK5-NEXT:    store i8 [[INC]], i8* [[I]], align 1
// CHECK5-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
// CHECK5:       for.end13:
// CHECK5-NEXT:    ret void
//
//
// CHECK5-LABEL: define {{[^@]+}}@_Z7runtimePfS_S_S_
// CHECK5-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK5-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK5-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK5-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK5-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK5-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK5-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK5-NEXT:    store i8 48, i8* [[I]], align 1
// CHECK5-NEXT:    br label [[FOR_COND:%.*]]
// CHECK5:       for.cond:
// CHECK5-NEXT:    [[TMP0:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[CONV:%.*]] = zext i8 [[TMP0]] to i32
// CHECK5-NEXT:    [[CMP:%.*]] = icmp sle i32 [[CONV]], 57
// CHECK5-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END13:%.*]]
// CHECK5:       for.body:
// CHECK5-NEXT:    store i32 -10, i32* [[X]], align 4
// CHECK5-NEXT:    br label [[FOR_COND1:%.*]]
// CHECK5:       for.cond1:
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32, i32* [[X]], align 4
// CHECK5-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP1]], 10
// CHECK5-NEXT:    br i1 [[CMP2]], label [[FOR_BODY3:%.*]], label [[FOR_END:%.*]]
// CHECK5:       for.body3:
// CHECK5-NEXT:    [[TMP2:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK5-NEXT:    [[TMP3:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[IDXPROM:%.*]] = zext i8 [[TMP3]] to i64
// CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP2]], i64 [[IDXPROM]]
// CHECK5-NEXT:    [[TMP4:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK5-NEXT:    [[TMP5:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK5-NEXT:    [[TMP6:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[IDXPROM4:%.*]] = zext i8 [[TMP6]] to i64
// CHECK5-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[TMP5]], i64 [[IDXPROM4]]
// CHECK5-NEXT:    [[TMP7:%.*]] = load float, float* [[ARRAYIDX5]], align 4
// CHECK5-NEXT:    [[MUL:%.*]] = fmul float [[TMP4]], [[TMP7]]
// CHECK5-NEXT:    [[TMP8:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK5-NEXT:    [[TMP9:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[IDXPROM6:%.*]] = zext i8 [[TMP9]] to i64
// CHECK5-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds float, float* [[TMP8]], i64 [[IDXPROM6]]
// CHECK5-NEXT:    [[TMP10:%.*]] = load float, float* [[ARRAYIDX7]], align 4
// CHECK5-NEXT:    [[MUL8:%.*]] = fmul float [[MUL]], [[TMP10]]
// CHECK5-NEXT:    [[TMP11:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK5-NEXT:    [[TMP12:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[IDXPROM9:%.*]] = zext i8 [[TMP12]] to i64
// CHECK5-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds float, float* [[TMP11]], i64 [[IDXPROM9]]
// CHECK5-NEXT:    store float [[MUL8]], float* [[ARRAYIDX10]], align 4
// CHECK5-NEXT:    br label [[FOR_INC:%.*]]
// CHECK5:       for.inc:
// CHECK5-NEXT:    [[TMP13:%.*]] = load i32, i32* [[X]], align 4
// CHECK5-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK5-NEXT:    store i32 [[INC]], i32* [[X]], align 4
// CHECK5-NEXT:    br label [[FOR_COND1]], !llvm.loop [[LOOP7:![0-9]+]]
// CHECK5:       for.end:
// CHECK5-NEXT:    br label [[FOR_INC11:%.*]]
// CHECK5:       for.inc11:
// CHECK5-NEXT:    [[TMP14:%.*]] = load i8, i8* [[I]], align 1
// CHECK5-NEXT:    [[INC12:%.*]] = add i8 [[TMP14]], 1
// CHECK5-NEXT:    store i8 [[INC12]], i8* [[I]], align 1
// CHECK5-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP8:![0-9]+]]
// CHECK5:       for.end13:
// CHECK5-NEXT:    ret void
//
//
// CHECK5-LABEL: define {{[^@]+}}@_Z8foo_simdii
// CHECK5-SAME: (i32 [[LOW:%.*]], i32 [[UP:%.*]]) #[[ATTR0]] {
// CHECK5-NEXT:  entry:
// CHECK5-NEXT:    [[LOW_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[UP_ADDR:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[_TMP18:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_19:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_20:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTCAPTURE_EXPR_21:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I27:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[DOTOMP_IV30:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    [[I31:%.*]] = alloca i32, align 4
// CHECK5-NEXT:    store i32 [[LOW]], i32* [[LOW_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[UP]], i32* [[UP_ADDR]], align 4
// CHECK5-NEXT:    [[TMP0:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP0]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[TMP1:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[SUB:%.*]] = sub i32 [[TMP2]], [[TMP3]]
// CHECK5-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK5-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK5-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK5-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK5-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK5-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    store i32 [[TMP4]], i32* [[I]], align 4
// CHECK5-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP5]], [[TMP6]]
// CHECK5-NEXT:    br i1 [[CMP]], label [[SIMD_IF_THEN:%.*]], label [[SIMD_IF_END:%.*]]
// CHECK5:       simd.if.then:
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_IV]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK5:       omp.inner.for.cond:
// CHECK5-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
// CHECK5-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group !9
// CHECK5-NEXT:    [[ADD6:%.*]] = add i32 [[TMP8]], 1
// CHECK5-NEXT:    [[CMP7:%.*]] = icmp ult i32 [[TMP7]], [[ADD6]]
// CHECK5-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK5:       omp.inner.for.body:
// CHECK5-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !llvm.access.group !9
// CHECK5-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
// CHECK5-NEXT:    [[MUL:%.*]] = mul i32 [[TMP10]], 1
// CHECK5-NEXT:    [[ADD8:%.*]] = add i32 [[TMP9]], [[MUL]]
// CHECK5-NEXT:    store i32 [[ADD8]], i32* [[I5]], align 4, !llvm.access.group !9
// CHECK5-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I5]], align 4, !llvm.access.group !9
// CHECK5-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
// CHECK5-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK5-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX]], align 4, !llvm.access.group !9
// CHECK5-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I5]], align 4, !llvm.access.group !9
// CHECK5-NEXT:    [[IDXPROM9:%.*]] = sext i32 [[TMP12]] to i64
// CHECK5-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM9]]
// CHECK5-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX10]], align 4, !llvm.access.group !9
// CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK5:       omp.body.continue:
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK5:       omp.inner.for.inc:
// CHECK5-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
// CHECK5-NEXT:    [[ADD11:%.*]] = add i32 [[TMP13]], 1
// CHECK5-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
// CHECK5:       omp.inner.for.end:
// CHECK5-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK5-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK5-NEXT:    [[SUB12:%.*]] = sub i32 [[TMP15]], [[TMP16]]
// CHECK5-NEXT:    [[SUB13:%.*]] = sub i32 [[SUB12]], 1
// CHECK5-NEXT:    [[ADD14:%.*]] = add i32 [[SUB13]], 1
// CHECK5-NEXT:    [[DIV15:%.*]] = udiv i32 [[ADD14]], 1
// CHECK5-NEXT:    [[MUL16:%.*]] = mul i32 [[DIV15]], 1
// CHECK5-NEXT:    [[ADD17:%.*]] = add i32 [[TMP14]], [[MUL16]]
// CHECK5-NEXT:    store i32 [[ADD17]], i32* [[I5]], align 4
// CHECK5-NEXT:    br label [[SIMD_IF_END]]
// CHECK5:       simd.if.end:
// CHECK5-NEXT:    [[TMP17:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP17]], i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK5-NEXT:    [[TMP18:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK5-NEXT:    store i32 [[TMP18]], i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK5-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK5-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK5-NEXT:    [[SUB22:%.*]] = sub i32 [[TMP19]], [[TMP20]]
// CHECK5-NEXT:    [[SUB23:%.*]] = sub i32 [[SUB22]], 1
// CHECK5-NEXT:    [[ADD24:%.*]] = add i32 [[SUB23]], 1
// CHECK5-NEXT:    [[DIV25:%.*]] = udiv i32 [[ADD24]], 1
// CHECK5-NEXT:    [[SUB26:%.*]] = sub i32 [[DIV25]], 1
// CHECK5-NEXT:    store i32 [[SUB26]], i32* [[DOTCAPTURE_EXPR_21]], align 4
// CHECK5-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_21]], align 4
// CHECK5-NEXT:    store i32 [[TMP21]], i32* [[DOTOMP_UB]], align 4
// CHECK5-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK5-NEXT:    store i32 [[TMP22]], i32* [[I27]], align 4
// CHECK5-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK5-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK5-NEXT:    [[CMP28:%.*]] = icmp slt i32 [[TMP23]], [[TMP24]]
// CHECK5-NEXT:    br i1 [[CMP28]], label [[SIMD_IF_THEN29:%.*]], label [[SIMD_IF_END52:%.*]]
// CHECK5:       simd.if.then29:
// CHECK5-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK5-NEXT:    store i32 [[TMP25]], i32* [[DOTOMP_IV30]], align 4
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND32:%.*]]
// CHECK5:       omp.inner.for.cond32:
// CHECK5-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_IV30]], align 4, !llvm.access.group !13
// CHECK5-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !13
// CHECK5-NEXT:    [[ADD33:%.*]] = add i32 [[TMP27]], 1
// CHECK5-NEXT:    [[CMP34:%.*]] = icmp ult i32 [[TMP26]], [[ADD33]]
// CHECK5-NEXT:    br i1 [[CMP34]], label [[OMP_INNER_FOR_BODY35:%.*]], label [[OMP_INNER_FOR_END45:%.*]]
// CHECK5:       omp.inner.for.body35:
// CHECK5-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4, !llvm.access.group !13
// CHECK5-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV30]], align 4, !llvm.access.group !13
// CHECK5-NEXT:    [[MUL36:%.*]] = mul i32 [[TMP29]], 1
// CHECK5-NEXT:    [[ADD37:%.*]] = add i32 [[TMP28]], [[MUL36]]
// CHECK5-NEXT:    store i32 [[ADD37]], i32* [[I31]], align 4, !llvm.access.group !13
// CHECK5-NEXT:    [[TMP30:%.*]] = load i32, i32* [[I31]], align 4, !llvm.access.group !13
// CHECK5-NEXT:    [[IDXPROM38:%.*]] = sext i32 [[TMP30]] to i64
// CHECK5-NEXT:    [[ARRAYIDX39:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM38]]
// CHECK5-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX39]], align 4, !llvm.access.group !13
// CHECK5-NEXT:    [[TMP31:%.*]] = load i32, i32* [[I31]], align 4, !llvm.access.group !13
// CHECK5-NEXT:    [[IDXPROM40:%.*]] = sext i32 [[TMP31]] to i64
// CHECK5-NEXT:    [[ARRAYIDX41:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM40]]
// CHECK5-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX41]], align 4, !llvm.access.group !13
// CHECK5-NEXT:    br label [[OMP_BODY_CONTINUE42:%.*]]
// CHECK5:       omp.body.continue42:
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_INC43:%.*]]
// CHECK5:       omp.inner.for.inc43:
// CHECK5-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV30]], align 4, !llvm.access.group !13
// CHECK5-NEXT:    [[ADD44:%.*]] = add i32 [[TMP32]], 1
// CHECK5-NEXT:    store i32 [[ADD44]], i32* [[DOTOMP_IV30]], align 4, !llvm.access.group !13
// CHECK5-NEXT:    br label [[OMP_INNER_FOR_COND32]], !llvm.loop [[LOOP14:![0-9]+]]
// CHECK5:       omp.inner.for.end45:
// CHECK5-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK5-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK5-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK5-NEXT:    [[SUB46:%.*]] = sub i32 [[TMP34]], [[TMP35]]
// CHECK5-NEXT:    [[SUB47:%.*]] = sub i32 [[SUB46]], 1
// CHECK5-NEXT:    [[ADD48:%.*]] = add i32 [[SUB47]], 1
// CHECK5-NEXT:    [[DIV49:%.*]] = udiv i32 [[ADD48]], 1
// CHECK5-NEXT:    [[MUL50:%.*]] = mul i32 [[DIV49]], 1
// CHECK5-NEXT:    [[ADD51:%.*]] = add i32 [[TMP33]], [[MUL50]]
// CHECK5-NEXT:    store i32 [[ADD51]], i32* [[I31]], align 4
// CHECK5-NEXT:    br label [[SIMD_IF_END52]]
// CHECK5:       simd.if.end52:
// CHECK5-NEXT:    ret void
//
//
// CHECK6-LABEL: define {{[^@]+}}@_Z18static_not_chunkedPfS_S_S_
// CHECK6-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK6-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK6-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK6-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK6-NEXT:    store i32 32000000, i32* [[I]], align 4
// CHECK6-NEXT:    br label [[FOR_COND:%.*]]
// CHECK6:       for.cond:
// CHECK6-NEXT:    [[TMP0:%.*]] = load i32, i32* [[I]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP0]], 33
// CHECK6-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK6:       for.body:
// CHECK6-NEXT:    [[TMP1:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[I]], align 4
// CHECK6-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP2]] to i64
// CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP1]], i64 [[IDXPROM]]
// CHECK6-NEXT:    [[TMP3:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[I]], align 4
// CHECK6-NEXT:    [[IDXPROM1:%.*]] = sext i32 [[TMP5]] to i64
// CHECK6-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP4]], i64 [[IDXPROM1]]
// CHECK6-NEXT:    [[TMP6:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK6-NEXT:    [[MUL:%.*]] = fmul float [[TMP3]], [[TMP6]]
// CHECK6-NEXT:    [[TMP7:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[I]], align 4
// CHECK6-NEXT:    [[IDXPROM3:%.*]] = sext i32 [[TMP8]] to i64
// CHECK6-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[TMP7]], i64 [[IDXPROM3]]
// CHECK6-NEXT:    [[TMP9:%.*]] = load float, float* [[ARRAYIDX4]], align 4
// CHECK6-NEXT:    [[MUL5:%.*]] = fmul float [[MUL]], [[TMP9]]
// CHECK6-NEXT:    [[TMP10:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I]], align 4
// CHECK6-NEXT:    [[IDXPROM6:%.*]] = sext i32 [[TMP11]] to i64
// CHECK6-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds float, float* [[TMP10]], i64 [[IDXPROM6]]
// CHECK6-NEXT:    store float [[MUL5]], float* [[ARRAYIDX7]], align 4
// CHECK6-NEXT:    br label [[FOR_INC:%.*]]
// CHECK6:       for.inc:
// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I]], align 4
// CHECK6-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP12]], -7
// CHECK6-NEXT:    store i32 [[ADD]], i32* [[I]], align 4
// CHECK6-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP2:![0-9]+]]
// CHECK6:       for.end:
// CHECK6-NEXT:    ret void
//
//
// CHECK6-LABEL: define {{[^@]+}}@_Z8dynamic1PfS_S_S_
// CHECK6-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[I:%.*]] = alloca i64, align 8
// CHECK6-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK6-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK6-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK6-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK6-NEXT:    store i64 131071, i64* [[I]], align 8
// CHECK6-NEXT:    br label [[FOR_COND:%.*]]
// CHECK6:       for.cond:
// CHECK6-NEXT:    [[TMP0:%.*]] = load i64, i64* [[I]], align 8
// CHECK6-NEXT:    [[CMP:%.*]] = icmp ult i64 [[TMP0]], 2147483647
// CHECK6-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// CHECK6:       for.body:
// CHECK6-NEXT:    [[TMP1:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK6-NEXT:    [[TMP2:%.*]] = load i64, i64* [[I]], align 8
// CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP1]], i64 [[TMP2]]
// CHECK6-NEXT:    [[TMP3:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK6-NEXT:    [[TMP5:%.*]] = load i64, i64* [[I]], align 8
// CHECK6-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds float, float* [[TMP4]], i64 [[TMP5]]
// CHECK6-NEXT:    [[TMP6:%.*]] = load float, float* [[ARRAYIDX1]], align 4
// CHECK6-NEXT:    [[MUL:%.*]] = fmul float [[TMP3]], [[TMP6]]
// CHECK6-NEXT:    [[TMP7:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK6-NEXT:    [[TMP8:%.*]] = load i64, i64* [[I]], align 8
// CHECK6-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[TMP7]], i64 [[TMP8]]
// CHECK6-NEXT:    [[TMP9:%.*]] = load float, float* [[ARRAYIDX2]], align 4
// CHECK6-NEXT:    [[MUL3:%.*]] = fmul float [[MUL]], [[TMP9]]
// CHECK6-NEXT:    [[TMP10:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK6-NEXT:    [[TMP11:%.*]] = load i64, i64* [[I]], align 8
// CHECK6-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds float, float* [[TMP10]], i64 [[TMP11]]
// CHECK6-NEXT:    store float [[MUL3]], float* [[ARRAYIDX4]], align 4
// CHECK6-NEXT:    br label [[FOR_INC:%.*]]
// CHECK6:       for.inc:
// CHECK6-NEXT:    [[TMP12:%.*]] = load i64, i64* [[I]], align 8
// CHECK6-NEXT:    [[ADD:%.*]] = add i64 [[TMP12]], 127
// CHECK6-NEXT:    store i64 [[ADD]], i64* [[I]], align 8
// CHECK6-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
// CHECK6:       for.end:
// CHECK6-NEXT:    ret void
//
//
// CHECK6-LABEL: define {{[^@]+}}@_Z9test_autoPfS_S_S_
// CHECK6-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[Y:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK6-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK6-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK6-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK6-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK6-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[Y]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load i32, i32* [[Y]], align 4
// CHECK6-NEXT:    [[CONV:%.*]] = trunc i32 [[TMP0]] to i8
// CHECK6-NEXT:    store i8 [[CONV]], i8* [[I]], align 1
// CHECK6-NEXT:    br label [[FOR_COND:%.*]]
// CHECK6:       for.cond:
// CHECK6-NEXT:    [[TMP1:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[CONV1:%.*]] = sext i8 [[TMP1]] to i32
// CHECK6-NEXT:    [[CMP:%.*]] = icmp sle i32 [[CONV1]], 57
// CHECK6-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END13:%.*]]
// CHECK6:       for.body:
// CHECK6-NEXT:    store i32 11, i32* [[X]], align 4
// CHECK6-NEXT:    br label [[FOR_COND2:%.*]]
// CHECK6:       for.cond2:
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[X]], align 4
// CHECK6-NEXT:    [[CMP3:%.*]] = icmp ugt i32 [[TMP2]], 0
// CHECK6-NEXT:    br i1 [[CMP3]], label [[FOR_BODY4:%.*]], label [[FOR_END:%.*]]
// CHECK6:       for.body4:
// CHECK6-NEXT:    [[TMP3:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK6-NEXT:    [[TMP4:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[IDXPROM:%.*]] = sext i8 [[TMP4]] to i64
// CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP3]], i64 [[IDXPROM]]
// CHECK6-NEXT:    [[TMP5:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK6-NEXT:    [[TMP6:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK6-NEXT:    [[TMP7:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[IDXPROM5:%.*]] = sext i8 [[TMP7]] to i64
// CHECK6-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds float, float* [[TMP6]], i64 [[IDXPROM5]]
// CHECK6-NEXT:    [[TMP8:%.*]] = load float, float* [[ARRAYIDX6]], align 4
// CHECK6-NEXT:    [[MUL:%.*]] = fmul float [[TMP5]], [[TMP8]]
// CHECK6-NEXT:    [[TMP9:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK6-NEXT:    [[TMP10:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[IDXPROM7:%.*]] = sext i8 [[TMP10]] to i64
// CHECK6-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr inbounds float, float* [[TMP9]], i64 [[IDXPROM7]]
// CHECK6-NEXT:    [[TMP11:%.*]] = load float, float* [[ARRAYIDX8]], align 4
// CHECK6-NEXT:    [[MUL9:%.*]] = fmul float [[MUL]], [[TMP11]]
// CHECK6-NEXT:    [[TMP12:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK6-NEXT:    [[TMP13:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[IDXPROM10:%.*]] = sext i8 [[TMP13]] to i64
// CHECK6-NEXT:    [[ARRAYIDX11:%.*]] = getelementptr inbounds float, float* [[TMP12]], i64 [[IDXPROM10]]
// CHECK6-NEXT:    store float [[MUL9]], float* [[ARRAYIDX11]], align 4
// CHECK6-NEXT:    br label [[FOR_INC:%.*]]
// CHECK6:       for.inc:
// CHECK6-NEXT:    [[TMP14:%.*]] = load i32, i32* [[X]], align 4
// CHECK6-NEXT:    [[DEC:%.*]] = add i32 [[TMP14]], -1
// CHECK6-NEXT:    store i32 [[DEC]], i32* [[X]], align 4
// CHECK6-NEXT:    br label [[FOR_COND2]], !llvm.loop [[LOOP5:![0-9]+]]
// CHECK6:       for.end:
// CHECK6-NEXT:    br label [[FOR_INC12:%.*]]
// CHECK6:       for.inc12:
// CHECK6-NEXT:    [[TMP15:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[INC:%.*]] = add i8 [[TMP15]], 1
// CHECK6-NEXT:    store i8 [[INC]], i8* [[I]], align 1
// CHECK6-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
// CHECK6:       for.end13:
// CHECK6-NEXT:    ret void
//
//
// CHECK6-LABEL: define {{[^@]+}}@_Z7runtimePfS_S_S_
// CHECK6-SAME: (float* [[A:%.*]], float* [[B:%.*]], float* [[C:%.*]], float* [[D:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[A_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[B_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[C_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[D_ADDR:%.*]] = alloca float*, align 8
// CHECK6-NEXT:    [[X:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i8, align 1
// CHECK6-NEXT:    store float* [[A]], float** [[A_ADDR]], align 8
// CHECK6-NEXT:    store float* [[B]], float** [[B_ADDR]], align 8
// CHECK6-NEXT:    store float* [[C]], float** [[C_ADDR]], align 8
// CHECK6-NEXT:    store float* [[D]], float** [[D_ADDR]], align 8
// CHECK6-NEXT:    store i32 0, i32* [[X]], align 4
// CHECK6-NEXT:    store i8 48, i8* [[I]], align 1
// CHECK6-NEXT:    br label [[FOR_COND:%.*]]
// CHECK6:       for.cond:
// CHECK6-NEXT:    [[TMP0:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[CONV:%.*]] = zext i8 [[TMP0]] to i32
// CHECK6-NEXT:    [[CMP:%.*]] = icmp sle i32 [[CONV]], 57
// CHECK6-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END13:%.*]]
// CHECK6:       for.body:
// CHECK6-NEXT:    store i32 -10, i32* [[X]], align 4
// CHECK6-NEXT:    br label [[FOR_COND1:%.*]]
// CHECK6:       for.cond1:
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32, i32* [[X]], align 4
// CHECK6-NEXT:    [[CMP2:%.*]] = icmp slt i32 [[TMP1]], 10
// CHECK6-NEXT:    br i1 [[CMP2]], label [[FOR_BODY3:%.*]], label [[FOR_END:%.*]]
// CHECK6:       for.body3:
// CHECK6-NEXT:    [[TMP2:%.*]] = load float*, float** [[B_ADDR]], align 8
// CHECK6-NEXT:    [[TMP3:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[IDXPROM:%.*]] = zext i8 [[TMP3]] to i64
// CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[TMP2]], i64 [[IDXPROM]]
// CHECK6-NEXT:    [[TMP4:%.*]] = load float, float* [[ARRAYIDX]], align 4
// CHECK6-NEXT:    [[TMP5:%.*]] = load float*, float** [[C_ADDR]], align 8
// CHECK6-NEXT:    [[TMP6:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[IDXPROM4:%.*]] = zext i8 [[TMP6]] to i64
// CHECK6-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds float, float* [[TMP5]], i64 [[IDXPROM4]]
// CHECK6-NEXT:    [[TMP7:%.*]] = load float, float* [[ARRAYIDX5]], align 4
// CHECK6-NEXT:    [[MUL:%.*]] = fmul float [[TMP4]], [[TMP7]]
// CHECK6-NEXT:    [[TMP8:%.*]] = load float*, float** [[D_ADDR]], align 8
// CHECK6-NEXT:    [[TMP9:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[IDXPROM6:%.*]] = zext i8 [[TMP9]] to i64
// CHECK6-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr inbounds float, float* [[TMP8]], i64 [[IDXPROM6]]
// CHECK6-NEXT:    [[TMP10:%.*]] = load float, float* [[ARRAYIDX7]], align 4
// CHECK6-NEXT:    [[MUL8:%.*]] = fmul float [[MUL]], [[TMP10]]
// CHECK6-NEXT:    [[TMP11:%.*]] = load float*, float** [[A_ADDR]], align 8
// CHECK6-NEXT:    [[TMP12:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[IDXPROM9:%.*]] = zext i8 [[TMP12]] to i64
// CHECK6-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds float, float* [[TMP11]], i64 [[IDXPROM9]]
// CHECK6-NEXT:    store float [[MUL8]], float* [[ARRAYIDX10]], align 4
// CHECK6-NEXT:    br label [[FOR_INC:%.*]]
// CHECK6:       for.inc:
// CHECK6-NEXT:    [[TMP13:%.*]] = load i32, i32* [[X]], align 4
// CHECK6-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP13]], 1
// CHECK6-NEXT:    store i32 [[INC]], i32* [[X]], align 4
// CHECK6-NEXT:    br label [[FOR_COND1]], !llvm.loop [[LOOP7:![0-9]+]]
// CHECK6:       for.end:
// CHECK6-NEXT:    br label [[FOR_INC11:%.*]]
// CHECK6:       for.inc11:
// CHECK6-NEXT:    [[TMP14:%.*]] = load i8, i8* [[I]], align 1
// CHECK6-NEXT:    [[INC12:%.*]] = add i8 [[TMP14]], 1
// CHECK6-NEXT:    store i8 [[INC12]], i8* [[I]], align 1
// CHECK6-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP8:![0-9]+]]
// CHECK6:       for.end13:
// CHECK6-NEXT:    ret void
//
//
// CHECK6-LABEL: define {{[^@]+}}@_Z8foo_simdii
// CHECK6-SAME: (i32 [[LOW:%.*]], i32 [[UP:%.*]]) #[[ATTR0]] {
// CHECK6-NEXT:  entry:
// CHECK6-NEXT:    [[LOW_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[UP_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_1:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_2:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I5:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[_TMP18:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_19:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_20:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTCAPTURE_EXPR_21:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I27:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[DOTOMP_IV30:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    [[I31:%.*]] = alloca i32, align 4
// CHECK6-NEXT:    store i32 [[LOW]], i32* [[LOW_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[UP]], i32* [[UP_ADDR]], align 4
// CHECK6-NEXT:    [[TMP0:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP0]], i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[TMP1:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP1]], i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    [[TMP2:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    [[TMP3:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[SUB:%.*]] = sub i32 [[TMP2]], [[TMP3]]
// CHECK6-NEXT:    [[SUB3:%.*]] = sub i32 [[SUB]], 1
// CHECK6-NEXT:    [[ADD:%.*]] = add i32 [[SUB3]], 1
// CHECK6-NEXT:    [[DIV:%.*]] = udiv i32 [[ADD]], 1
// CHECK6-NEXT:    [[SUB4:%.*]] = sub i32 [[DIV]], 1
// CHECK6-NEXT:    store i32 [[SUB4]], i32* [[DOTCAPTURE_EXPR_2]], align 4
// CHECK6-NEXT:    [[TMP4:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    store i32 [[TMP4]], i32* [[I]], align 4
// CHECK6-NEXT:    [[TMP5:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[TMP6:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP5]], [[TMP6]]
// CHECK6-NEXT:    br i1 [[CMP]], label [[SIMD_IF_THEN:%.*]], label [[SIMD_IF_END:%.*]]
// CHECK6:       simd.if.then:
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_IV]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK6:       omp.inner.for.cond:
// CHECK6-NEXT:    [[TMP7:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
// CHECK6-NEXT:    [[TMP8:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_2]], align 4, !llvm.access.group !9
// CHECK6-NEXT:    [[ADD6:%.*]] = add i32 [[TMP8]], 1
// CHECK6-NEXT:    [[CMP7:%.*]] = icmp ult i32 [[TMP7]], [[ADD6]]
// CHECK6-NEXT:    br i1 [[CMP7]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK6:       omp.inner.for.body:
// CHECK6-NEXT:    [[TMP9:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4, !llvm.access.group !9
// CHECK6-NEXT:    [[TMP10:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
// CHECK6-NEXT:    [[MUL:%.*]] = mul i32 [[TMP10]], 1
// CHECK6-NEXT:    [[ADD8:%.*]] = add i32 [[TMP9]], [[MUL]]
// CHECK6-NEXT:    store i32 [[ADD8]], i32* [[I5]], align 4, !llvm.access.group !9
// CHECK6-NEXT:    [[TMP11:%.*]] = load i32, i32* [[I5]], align 4, !llvm.access.group !9
// CHECK6-NEXT:    [[IDXPROM:%.*]] = sext i32 [[TMP11]] to i64
// CHECK6-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM]]
// CHECK6-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX]], align 4, !llvm.access.group !9
// CHECK6-NEXT:    [[TMP12:%.*]] = load i32, i32* [[I5]], align 4, !llvm.access.group !9
// CHECK6-NEXT:    [[IDXPROM9:%.*]] = sext i32 [[TMP12]] to i64
// CHECK6-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM9]]
// CHECK6-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX10]], align 4, !llvm.access.group !9
// CHECK6-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK6:       omp.body.continue:
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK6:       omp.inner.for.inc:
// CHECK6-NEXT:    [[TMP13:%.*]] = load i32, i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
// CHECK6-NEXT:    [[ADD11:%.*]] = add i32 [[TMP13]], 1
// CHECK6-NEXT:    store i32 [[ADD11]], i32* [[DOTOMP_IV]], align 4, !llvm.access.group !9
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
// CHECK6:       omp.inner.for.end:
// CHECK6-NEXT:    [[TMP14:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[TMP15:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_1]], align 4
// CHECK6-NEXT:    [[TMP16:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_]], align 4
// CHECK6-NEXT:    [[SUB12:%.*]] = sub i32 [[TMP15]], [[TMP16]]
// CHECK6-NEXT:    [[SUB13:%.*]] = sub i32 [[SUB12]], 1
// CHECK6-NEXT:    [[ADD14:%.*]] = add i32 [[SUB13]], 1
// CHECK6-NEXT:    [[DIV15:%.*]] = udiv i32 [[ADD14]], 1
// CHECK6-NEXT:    [[MUL16:%.*]] = mul i32 [[DIV15]], 1
// CHECK6-NEXT:    [[ADD17:%.*]] = add i32 [[TMP14]], [[MUL16]]
// CHECK6-NEXT:    store i32 [[ADD17]], i32* [[I5]], align 4
// CHECK6-NEXT:    br label [[SIMD_IF_END]]
// CHECK6:       simd.if.end:
// CHECK6-NEXT:    [[TMP17:%.*]] = load i32, i32* [[LOW_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP17]], i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK6-NEXT:    [[TMP18:%.*]] = load i32, i32* [[UP_ADDR]], align 4
// CHECK6-NEXT:    store i32 [[TMP18]], i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK6-NEXT:    [[TMP19:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK6-NEXT:    [[TMP20:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK6-NEXT:    [[SUB22:%.*]] = sub i32 [[TMP19]], [[TMP20]]
// CHECK6-NEXT:    [[SUB23:%.*]] = sub i32 [[SUB22]], 1
// CHECK6-NEXT:    [[ADD24:%.*]] = add i32 [[SUB23]], 1
// CHECK6-NEXT:    [[DIV25:%.*]] = udiv i32 [[ADD24]], 1
// CHECK6-NEXT:    [[SUB26:%.*]] = sub i32 [[DIV25]], 1
// CHECK6-NEXT:    store i32 [[SUB26]], i32* [[DOTCAPTURE_EXPR_21]], align 4
// CHECK6-NEXT:    store i32 0, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    [[TMP21:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_21]], align 4
// CHECK6-NEXT:    store i32 [[TMP21]], i32* [[DOTOMP_UB]], align 4
// CHECK6-NEXT:    [[TMP22:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK6-NEXT:    store i32 [[TMP22]], i32* [[I27]], align 4
// CHECK6-NEXT:    [[TMP23:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK6-NEXT:    [[TMP24:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK6-NEXT:    [[CMP28:%.*]] = icmp slt i32 [[TMP23]], [[TMP24]]
// CHECK6-NEXT:    br i1 [[CMP28]], label [[SIMD_IF_THEN29:%.*]], label [[SIMD_IF_END52:%.*]]
// CHECK6:       simd.if.then29:
// CHECK6-NEXT:    [[TMP25:%.*]] = load i32, i32* [[DOTOMP_LB]], align 4
// CHECK6-NEXT:    store i32 [[TMP25]], i32* [[DOTOMP_IV30]], align 4
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND32:%.*]]
// CHECK6:       omp.inner.for.cond32:
// CHECK6-NEXT:    [[TMP26:%.*]] = load i32, i32* [[DOTOMP_IV30]], align 4, !llvm.access.group !13
// CHECK6-NEXT:    [[TMP27:%.*]] = load i32, i32* [[DOTOMP_UB]], align 4, !llvm.access.group !13
// CHECK6-NEXT:    [[ADD33:%.*]] = add i32 [[TMP27]], 1
// CHECK6-NEXT:    [[CMP34:%.*]] = icmp ult i32 [[TMP26]], [[ADD33]]
// CHECK6-NEXT:    br i1 [[CMP34]], label [[OMP_INNER_FOR_BODY35:%.*]], label [[OMP_INNER_FOR_END45:%.*]]
// CHECK6:       omp.inner.for.body35:
// CHECK6-NEXT:    [[TMP28:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4, !llvm.access.group !13
// CHECK6-NEXT:    [[TMP29:%.*]] = load i32, i32* [[DOTOMP_IV30]], align 4, !llvm.access.group !13
// CHECK6-NEXT:    [[MUL36:%.*]] = mul i32 [[TMP29]], 1
// CHECK6-NEXT:    [[ADD37:%.*]] = add i32 [[TMP28]], [[MUL36]]
// CHECK6-NEXT:    store i32 [[ADD37]], i32* [[I31]], align 4, !llvm.access.group !13
// CHECK6-NEXT:    [[TMP30:%.*]] = load i32, i32* [[I31]], align 4, !llvm.access.group !13
// CHECK6-NEXT:    [[IDXPROM38:%.*]] = sext i32 [[TMP30]] to i64
// CHECK6-NEXT:    [[ARRAYIDX39:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM38]]
// CHECK6-NEXT:    store float 0.000000e+00, float* [[ARRAYIDX39]], align 4, !llvm.access.group !13
// CHECK6-NEXT:    [[TMP31:%.*]] = load i32, i32* [[I31]], align 4, !llvm.access.group !13
// CHECK6-NEXT:    [[IDXPROM40:%.*]] = sext i32 [[TMP31]] to i64
// CHECK6-NEXT:    [[ARRAYIDX41:%.*]] = getelementptr inbounds [10 x float], [10 x float]* @f, i64 0, i64 [[IDXPROM40]]
// CHECK6-NEXT:    store float 1.000000e+00, float* [[ARRAYIDX41]], align 4, !llvm.access.group !13
// CHECK6-NEXT:    br label [[OMP_BODY_CONTINUE42:%.*]]
// CHECK6:       omp.body.continue42:
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_INC43:%.*]]
// CHECK6:       omp.inner.for.inc43:
// CHECK6-NEXT:    [[TMP32:%.*]] = load i32, i32* [[DOTOMP_IV30]], align 4, !llvm.access.group !13
// CHECK6-NEXT:    [[ADD44:%.*]] = add i32 [[TMP32]], 1
// CHECK6-NEXT:    store i32 [[ADD44]], i32* [[DOTOMP_IV30]], align 4, !llvm.access.group !13
// CHECK6-NEXT:    br label [[OMP_INNER_FOR_COND32]], !llvm.loop [[LOOP14:![0-9]+]]
// CHECK6:       omp.inner.for.end45:
// CHECK6-NEXT:    [[TMP33:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK6-NEXT:    [[TMP34:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_20]], align 4
// CHECK6-NEXT:    [[TMP35:%.*]] = load i32, i32* [[DOTCAPTURE_EXPR_19]], align 4
// CHECK6-NEXT:    [[SUB46:%.*]] = sub i32 [[TMP34]], [[TMP35]]
// CHECK6-NEXT:    [[SUB47:%.*]] = sub i32 [[SUB46]], 1
// CHECK6-NEXT:    [[ADD48:%.*]] = add i32 [[SUB47]], 1
// CHECK6-NEXT:    [[DIV49:%.*]] = udiv i32 [[ADD48]], 1
// CHECK6-NEXT:    [[MUL50:%.*]] = mul i32 [[DIV49]], 1
// CHECK6-NEXT:    [[ADD51:%.*]] = add i32 [[TMP33]], [[MUL50]]
// CHECK6-NEXT:    store i32 [[ADD51]], i32* [[I31]], align 4
// CHECK6-NEXT:    br label [[SIMD_IF_END52]]
// CHECK6:       simd.if.end52:
// CHECK6-NEXT:    ret void
//