clang-p2996/clang/test/OpenMP/target_parallel_generic_loop_codegen-1.cpp

// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _
// Test host codegen.
// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s

// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s

// Test target parallel for codegen - host bc file has to be created first.
// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK
// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s -check-prefix=TCHECK
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK

// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s

// Check that no target code is emitted if no omptests flag was provided.
// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK-NTARGET

// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s

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

// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s

// Test target parallel for codegen - host bc file has to be created first.
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK

// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s -check-prefix=TCHECK
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK

// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s

// Check that no target code is emitted if no omptests flag was provided.
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK-NTARGET-OMP-DEFAULT

// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s

// expected-no-diagnostics
#ifndef HEADER
#define HEADER





// We have 7 target regions





// We have 4 initializers, one for the 500 priority, another one for 501, or more for the default priority, and the last one for the offloading registration function.


extern int *R;

struct SA {
  int arr[4];
  void foo() {
    int a = *R;
    a += 1;
    *R = a;
  }
  SA() {
    int a = *R;
    a += 2;
    *R = a;
  }
  ~SA() {
    int a = *R;
    a += 3;
    *R = a;
  }
};

struct SB {
  int arr[8];
  void foo() {
    int a = *R;
    #pragma omp target parallel loop
    for (int i = 0; i < 10; ++i)
      a += 4;
    *R = a;
  }
  SB() {
    int a = *R;
    a += 5;
    *R = a;
  }
  ~SB() {
    int a = *R;
    a += 6;
    *R = a;
  }
};

struct SC {
  int arr[16];
  void foo() {
    int a = *R;
    a += 7;
    *R = a;
  }
  SC() {
    int a = *R;
    #pragma omp target parallel loop
    for (int i = 0; i < 10; ++i)
      a += 8;
    *R = a;
  }
  ~SC() {
    int a = *R;
    a += 9;
    *R = a;
  }
};

struct SD {
  int arr[32];
  void foo() {
    int a = *R;
    a += 10;
    *R = a;
  }
  SD() {
    int a = *R;
    a += 11;
    *R = a;
  }
  ~SD() {
    int a = *R;
    #pragma omp target parallel loop
    for (int i = 0; i < 10; ++i)
      a += 12;
    *R = a;
  }
};

struct SE {
  int arr[64];
  void foo() {
    int a = *R;
    #pragma omp target parallel loop if(target: 0)
    for (int i = 0; i < 10; ++i)
      a += 13;
    *R = a;
  }
  SE() {
    int a = *R;
    #pragma omp target parallel loop
    for (int i = 0; i < 10; ++i)
      a += 14;
    *R = a;
  }
  ~SE() {
    int a = *R;
    #pragma omp target parallel loop
    for (int i = 0; i < 10; ++i)
      a += 15;
    *R = a;
  }
};

template <int x>
struct ST {
  int arr[128 + x];
  void foo() {
    int a = *R;
    #pragma omp target parallel loop
    for (int i = 0; i < 10; ++i)
      a += 16 + x;
    *R = a;
  }
  ST() {
    int a = *R;
    #pragma omp target parallel loop
    for (int i = 0; i < 10; ++i)
      a += 17 + x;
    *R = a;
  }
  ~ST() {
    int a = *R;
    #pragma omp target parallel loop
    for (int i = 0; i < 10; ++i)
      a += 18 + x;
    *R = a;
  }
};

// We have to make sure we us all the target regions:




// We have 2 initializers with priority 500

// We have 1 initializers with priority 501

// We have 6 initializers with default priority

static __attribute__((init_priority(500))) SA a1;
SA a2;
SB __attribute__((init_priority(500))) b1;
SB __attribute__((init_priority(501))) b2;
static SC c1;
SD d1;
SE e1;
ST<100> t1;
ST<1000> t2;


int bar(int a){
  int r = a;

  a1.foo();
  a2.foo();
  b1.foo();
  b2.foo();
  c1.foo();
  d1.foo();
  e1.foo();
  t1.foo();
  t2.foo();

  #pragma omp target parallel loop
  for (int i = 0; i < 10; ++i)
    ++r;

  return r + *R;
}

// Check metadata is properly generated:


#endif
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK-SAME: () #[[ATTR3:[0-9]+]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @_ZL2a1, ptr @__dso_handle) #[[ATTR2:[0-9]+]]
// CHECK-NEXT:    ret void
//
//
//
//
//
//
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// CHECK-SAME: () #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @a2, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
// CHECK-SAME: () #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NEXT:    ret void
//
//
//
//
//
//
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init.3
// CHECK-SAME: () #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b2, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init.4
// CHECK-SAME: () #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @_ZN2SCC1Ev(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SCD1Ev, ptr @_ZL2c1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NEXT:    ret void
//
//
//
//
//
//
//
//
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init.7
// CHECK-SAME: () #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @_ZN2SDC1Ev(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SDD1Ev, ptr @d1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NEXT:    ret void
//
//
//
//
//
//
//
//
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init.10
// CHECK-SAME: () #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @_ZN2SEC1Ev(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SED1Ev, ptr @e1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init.15
// CHECK-SAME: () #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @_ZN2STILi100EEC1Ev(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi100EED1Ev, ptr @t1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
// CHECK-LABEL: define {{[^@]+}}@__cxx_global_var_init.20
// CHECK-SAME: () #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @_ZN2STILi1000EEC1Ev(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi1000EED1Ev, ptr @t2, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
// CHECK-LABEL: define {{[^@]+}}@_GLOBAL__I_000500
// CHECK-SAME: () #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @__cxx_global_var_init()
// CHECK-NEXT:    call void @__cxx_global_var_init.2()
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@_GLOBAL__I_000501
// CHECK-SAME: () #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @__cxx_global_var_init.3()
// CHECK-NEXT:    ret void
//
//
// CHECK-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_target_parallel_generic_loop_codegen_1.cpp
// CHECK-SAME: () #[[ATTR3]] {
// CHECK-NEXT:  entry:
// CHECK-NEXT:    call void @__cxx_global_var_init.1()
// CHECK-NEXT:    call void @__cxx_global_var_init.4()
// CHECK-NEXT:    call void @__cxx_global_var_init.7()
// CHECK-NEXT:    call void @__cxx_global_var_init.10()
// CHECK-NEXT:    call void @__cxx_global_var_init.15()
// CHECK-NEXT:    call void @__cxx_global_var_init.20()
// CHECK-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@__cxx_global_var_init
// SIMD-ONLY0-SAME: () #[[ATTR0:[0-9]+]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @_ZL2a1, ptr @__dso_handle) #[[ATTR2:[0-9]+]]
// SIMD-ONLY0-NEXT:    ret void
//
//
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// SIMD-ONLY0-SAME: () #[[ATTR0]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @a2, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY0-NEXT:    ret void
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
// SIMD-ONLY0-SAME: () #[[ATTR0]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY0-NEXT:    ret void
//
//
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@__cxx_global_var_init.3
// SIMD-ONLY0-SAME: () #[[ATTR0]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b2, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY0-NEXT:    ret void
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@__cxx_global_var_init.4
// SIMD-ONLY0-SAME: () #[[ATTR0]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @_ZN2SCC1Ev(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SCD1Ev, ptr @_ZL2c1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY0-NEXT:    ret void
//
//
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@__cxx_global_var_init.5
// SIMD-ONLY0-SAME: () #[[ATTR0]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @_ZN2SDC1Ev(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SDD1Ev, ptr @d1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY0-NEXT:    ret void
//
//
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@__cxx_global_var_init.6
// SIMD-ONLY0-SAME: () #[[ATTR0]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @_ZN2SEC1Ev(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SED1Ev, ptr @e1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY0-NEXT:    ret void
//
//
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@__cxx_global_var_init.7
// SIMD-ONLY0-SAME: () #[[ATTR0]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @_ZN2STILi100EEC1Ev(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi100EED1Ev, ptr @t1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY0-NEXT:    ret void
//
//
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@__cxx_global_var_init.8
// SIMD-ONLY0-SAME: () #[[ATTR0]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @_ZN2STILi1000EEC1Ev(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// SIMD-ONLY0-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi1000EED1Ev, ptr @t2, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY0-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@_GLOBAL__I_000500
// SIMD-ONLY0-SAME: () #[[ATTR0]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @__cxx_global_var_init()
// SIMD-ONLY0-NEXT:    call void @__cxx_global_var_init.2()
// SIMD-ONLY0-NEXT:    ret void
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@_GLOBAL__I_000501
// SIMD-ONLY0-SAME: () #[[ATTR0]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @__cxx_global_var_init.3()
// SIMD-ONLY0-NEXT:    ret void
//
//
// SIMD-ONLY0-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_target_parallel_generic_loop_codegen_1.cpp
// SIMD-ONLY0-SAME: () #[[ATTR0]] {
// SIMD-ONLY0-NEXT:  entry:
// SIMD-ONLY0-NEXT:    call void @__cxx_global_var_init.1()
// SIMD-ONLY0-NEXT:    call void @__cxx_global_var_init.4()
// SIMD-ONLY0-NEXT:    call void @__cxx_global_var_init.5()
// SIMD-ONLY0-NEXT:    call void @__cxx_global_var_init.6()
// SIMD-ONLY0-NEXT:    call void @__cxx_global_var_init.7()
// SIMD-ONLY0-NEXT:    call void @__cxx_global_var_init.8()
// SIMD-ONLY0-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@__cxx_global_var_init
// SIMD-ONLY1-SAME: () #[[ATTR0:[0-9]+]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// SIMD-ONLY1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @_ZL2a1, ptr @__dso_handle) #[[ATTR2:[0-9]+]]
// SIMD-ONLY1-NEXT:    ret void
//
//
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// SIMD-ONLY1-SAME: () #[[ATTR0]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// SIMD-ONLY1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @a2, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY1-NEXT:    ret void
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
// SIMD-ONLY1-SAME: () #[[ATTR0]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// SIMD-ONLY1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY1-NEXT:    ret void
//
//
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@__cxx_global_var_init.3
// SIMD-ONLY1-SAME: () #[[ATTR0]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// SIMD-ONLY1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b2, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY1-NEXT:    ret void
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@__cxx_global_var_init.4
// SIMD-ONLY1-SAME: () #[[ATTR0]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @_ZN2SCC1Ev(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// SIMD-ONLY1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SCD1Ev, ptr @_ZL2c1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY1-NEXT:    ret void
//
//
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@__cxx_global_var_init.5
// SIMD-ONLY1-SAME: () #[[ATTR0]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @_ZN2SDC1Ev(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// SIMD-ONLY1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SDD1Ev, ptr @d1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY1-NEXT:    ret void
//
//
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@__cxx_global_var_init.6
// SIMD-ONLY1-SAME: () #[[ATTR0]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @_ZN2SEC1Ev(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// SIMD-ONLY1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SED1Ev, ptr @e1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY1-NEXT:    ret void
//
//
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@__cxx_global_var_init.7
// SIMD-ONLY1-SAME: () #[[ATTR0]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @_ZN2STILi100EEC1Ev(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// SIMD-ONLY1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi100EED1Ev, ptr @t1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY1-NEXT:    ret void
//
//
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@__cxx_global_var_init.8
// SIMD-ONLY1-SAME: () #[[ATTR0]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @_ZN2STILi1000EEC1Ev(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// SIMD-ONLY1-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi1000EED1Ev, ptr @t2, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY1-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@_GLOBAL__I_000500
// SIMD-ONLY1-SAME: () #[[ATTR0]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @__cxx_global_var_init()
// SIMD-ONLY1-NEXT:    call void @__cxx_global_var_init.2()
// SIMD-ONLY1-NEXT:    ret void
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@_GLOBAL__I_000501
// SIMD-ONLY1-SAME: () #[[ATTR0]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @__cxx_global_var_init.3()
// SIMD-ONLY1-NEXT:    ret void
//
//
// SIMD-ONLY1-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_target_parallel_generic_loop_codegen_1.cpp
// SIMD-ONLY1-SAME: () #[[ATTR0]] {
// SIMD-ONLY1-NEXT:  entry:
// SIMD-ONLY1-NEXT:    call void @__cxx_global_var_init.1()
// SIMD-ONLY1-NEXT:    call void @__cxx_global_var_init.4()
// SIMD-ONLY1-NEXT:    call void @__cxx_global_var_init.5()
// SIMD-ONLY1-NEXT:    call void @__cxx_global_var_init.6()
// SIMD-ONLY1-NEXT:    call void @__cxx_global_var_init.7()
// SIMD-ONLY1-NEXT:    call void @__cxx_global_var_init.8()
// SIMD-ONLY1-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_Z3bari
// CHECK-NTARGET-SAME: (i32 noundef signext [[A:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[R:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[R_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[R]], align 4
// CHECK-NTARGET-NEXT:    call void @_ZN2SA3fooEv(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// CHECK-NTARGET-NEXT:    call void @_ZN2SA3fooEv(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// CHECK-NTARGET-NEXT:    call void @_ZN2SB3fooEv(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// CHECK-NTARGET-NEXT:    call void @_ZN2SB3fooEv(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// CHECK-NTARGET-NEXT:    call void @_ZN2SC3fooEv(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// CHECK-NTARGET-NEXT:    call void @_ZN2SD3fooEv(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// CHECK-NTARGET-NEXT:    call void @_ZN2SE3fooEv(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// CHECK-NTARGET-NEXT:    call void @_ZN2STILi100EE3fooEv(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// CHECK-NTARGET-NEXT:    call void @_ZN2STILi1000EE3fooEv(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[R]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[R_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i64, ptr [[R_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267(i64 [[TMP2]]) #[[ATTR2:[0-9]+]]
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[R]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], [[TMP5]]
// CHECK-NTARGET-NEXT:    ret i32 [[ADD]]
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SA3fooEv
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SB3fooEv
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SC3fooEv
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 7
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SD3fooEv
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 10
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SE3fooEv
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2STILi100EE3fooEv
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2STILi1000EE3fooEv
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267
// CHECK-NTARGET-SAME: (i64 noundef [[R:%.*]]) #[[ATTR1:[0-9]+]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[R_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[R_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[R]], ptr [[R_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[R_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[R_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[R_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2:[0-9]+]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[R:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[R_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[R]], ptr [[R_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[R_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK-NTARGET-NEXT:    store i32 [[INC]], ptr [[R_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK-NTARGET-SAME: () #[[ATTR3:[0-9]+]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @_ZL2a1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SAC1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2SAC2Ev(ptr noundef nonnull align 4 dereferenceable(16) [[THIS1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SAD1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2SAD2Ev(ptr noundef nonnull align 4 dereferenceable(16) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SAC2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 2
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SAD2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 3
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// CHECK-NTARGET-SAME: () #[[ATTR3]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @a2, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
// CHECK-NTARGET-SAME: () #[[ATTR3]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SBC1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2SBC2Ev(ptr noundef nonnull align 4 dereferenceable(32) [[THIS1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SBD1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2SBD2Ev(ptr noundef nonnull align 4 dereferenceable(32) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SBC2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 5
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SBD2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 6
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@__cxx_global_var_init.3
// CHECK-NTARGET-SAME: () #[[ATTR3]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b2, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@__cxx_global_var_init.4
// CHECK-NTARGET-SAME: () #[[ATTR3]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @_ZN2SCC1Ev(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SCD1Ev, ptr @_ZL2c1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SCC1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2SCC2Ev(ptr noundef nonnull align 4 dereferenceable(64) [[THIS1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SCD1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2SCD2Ev(ptr noundef nonnull align 4 dereferenceable(64) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SCC2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 8
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SCD2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@__cxx_global_var_init.5
// CHECK-NTARGET-SAME: () #[[ATTR3]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @_ZN2SDC1Ev(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SDD1Ev, ptr @d1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SDC1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2SDC2Ev(ptr noundef nonnull align 4 dereferenceable(128) [[THIS1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SDD1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2SDD2Ev(ptr noundef nonnull align 4 dereferenceable(128) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SDC2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 11
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SDD2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 12
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@__cxx_global_var_init.6
// CHECK-NTARGET-SAME: () #[[ATTR3]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @_ZN2SEC1Ev(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SED1Ev, ptr @e1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SEC1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2SEC2Ev(ptr noundef nonnull align 4 dereferenceable(256) [[THIS1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SED1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2SED2Ev(ptr noundef nonnull align 4 dereferenceable(256) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SEC2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 14
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2SED2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 15
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@__cxx_global_var_init.7
// CHECK-NTARGET-SAME: () #[[ATTR3]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @_ZN2STILi100EEC1Ev(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi100EED1Ev, ptr @t1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2STILi100EEC1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2STILi100EEC2Ev(ptr noundef nonnull align 4 dereferenceable(912) [[THIS1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2STILi100EED1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2STILi100EED2Ev(ptr noundef nonnull align 4 dereferenceable(912) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2STILi100EEC2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 117
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2STILi100EED2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 118
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@__cxx_global_var_init.8
// CHECK-NTARGET-SAME: () #[[ATTR3]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @_ZN2STILi1000EEC1Ev(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi1000EED1Ev, ptr @t2, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2STILi1000EEC1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2STILi1000EEC2Ev(ptr noundef nonnull align 4 dereferenceable(4512) [[THIS1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2STILi1000EED1Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    call void @_ZN2STILi1000EED2Ev(ptr noundef nonnull align 4 dereferenceable(4512) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2STILi1000EEC2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 1017
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_ZN2STILi1000EED2Ev
// CHECK-NTARGET-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR0]] comdat {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 1018
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 4
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 13
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 116
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211
// CHECK-NTARGET-SAME: (i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211.omp_outlined
// CHECK-NTARGET-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET:       cond.true:
// CHECK-NTARGET-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET:       cond.false:
// CHECK-NTARGET-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    br label [[COND_END]]
// CHECK-NTARGET:       cond.end:
// CHECK-NTARGET-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET:       omp.inner.for.cond:
// CHECK-NTARGET-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET:       omp.inner.for.body:
// CHECK-NTARGET-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 1016
// CHECK-NTARGET-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET:       omp.body.continue:
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET:       omp.inner.for.inc:
// CHECK-NTARGET-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET:       omp.inner.for.end:
// CHECK-NTARGET-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET:       omp.loop.exit:
// CHECK-NTARGET-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_GLOBAL__I_000500
// CHECK-NTARGET-SAME: () #[[ATTR3]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @__cxx_global_var_init()
// CHECK-NTARGET-NEXT:    call void @__cxx_global_var_init.2()
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_GLOBAL__I_000501
// CHECK-NTARGET-SAME: () #[[ATTR3]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @__cxx_global_var_init.3()
// CHECK-NTARGET-NEXT:    ret void
//
//
// CHECK-NTARGET-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_target_parallel_generic_loop_codegen_1.cpp
// CHECK-NTARGET-SAME: () #[[ATTR3]] {
// CHECK-NTARGET-NEXT:  entry:
// CHECK-NTARGET-NEXT:    call void @__cxx_global_var_init.1()
// CHECK-NTARGET-NEXT:    call void @__cxx_global_var_init.4()
// CHECK-NTARGET-NEXT:    call void @__cxx_global_var_init.5()
// CHECK-NTARGET-NEXT:    call void @__cxx_global_var_init.6()
// CHECK-NTARGET-NEXT:    call void @__cxx_global_var_init.7()
// CHECK-NTARGET-NEXT:    call void @__cxx_global_var_init.8()
// CHECK-NTARGET-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@__cxx_global_var_init
// SIMD-ONLY2-SAME: () #[[ATTR0:[0-9]+]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @_ZL2a1, ptr @__dso_handle) #[[ATTR2:[0-9]+]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SAC1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2SAC2Ev(ptr noundef nonnull align 4 dereferenceable(16) [[THIS1]])
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SAD1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2SAD2Ev(ptr noundef nonnull align 4 dereferenceable(16) [[THIS1]]) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// SIMD-ONLY2-SAME: () #[[ATTR0]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @a2, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
// SIMD-ONLY2-SAME: () #[[ATTR0]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SBC1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2SBC2Ev(ptr noundef nonnull align 4 dereferenceable(32) [[THIS1]])
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SBD1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2SBD2Ev(ptr noundef nonnull align 4 dereferenceable(32) [[THIS1]]) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@__cxx_global_var_init.3
// SIMD-ONLY2-SAME: () #[[ATTR0]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b2, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@__cxx_global_var_init.4
// SIMD-ONLY2-SAME: () #[[ATTR0]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @_ZN2SCC1Ev(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SCD1Ev, ptr @_ZL2c1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SCC1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2SCC2Ev(ptr noundef nonnull align 4 dereferenceable(64) [[THIS1]])
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SCD1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2SCD2Ev(ptr noundef nonnull align 4 dereferenceable(64) [[THIS1]]) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@__cxx_global_var_init.5
// SIMD-ONLY2-SAME: () #[[ATTR0]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @_ZN2SDC1Ev(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SDD1Ev, ptr @d1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SDC1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2SDC2Ev(ptr noundef nonnull align 4 dereferenceable(128) [[THIS1]])
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SDD1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2SDD2Ev(ptr noundef nonnull align 4 dereferenceable(128) [[THIS1]]) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@__cxx_global_var_init.6
// SIMD-ONLY2-SAME: () #[[ATTR0]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @_ZN2SEC1Ev(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SED1Ev, ptr @e1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SEC1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2SEC2Ev(ptr noundef nonnull align 4 dereferenceable(256) [[THIS1]])
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SED1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2SED2Ev(ptr noundef nonnull align 4 dereferenceable(256) [[THIS1]]) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@__cxx_global_var_init.7
// SIMD-ONLY2-SAME: () #[[ATTR0]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @_ZN2STILi100EEC1Ev(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi100EED1Ev, ptr @t1, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2STILi100EEC1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2STILi100EEC2Ev(ptr noundef nonnull align 4 dereferenceable(912) [[THIS1]])
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2STILi100EED1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2STILi100EED2Ev(ptr noundef nonnull align 4 dereferenceable(912) [[THIS1]]) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@__cxx_global_var_init.8
// SIMD-ONLY2-SAME: () #[[ATTR0]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @_ZN2STILi1000EEC1Ev(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi1000EED1Ev, ptr @t2, ptr @__dso_handle) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2STILi1000EEC1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2STILi1000EEC2Ev(ptr noundef nonnull align 4 dereferenceable(4512) [[THIS1]])
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2STILi1000EED1Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    call void @_ZN2STILi1000EED2Ev(ptr noundef nonnull align 4 dereferenceable(4512) [[THIS1]]) #[[ATTR2]]
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_Z3bari
// SIMD-ONLY2-SAME: (i32 noundef signext [[A:%.*]]) #[[ATTR1]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[R:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP0]], ptr [[R]], align 4
// SIMD-ONLY2-NEXT:    call void @_ZN2SA3fooEv(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// SIMD-ONLY2-NEXT:    call void @_ZN2SA3fooEv(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// SIMD-ONLY2-NEXT:    call void @_ZN2SB3fooEv(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// SIMD-ONLY2-NEXT:    call void @_ZN2SB3fooEv(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// SIMD-ONLY2-NEXT:    call void @_ZN2SC3fooEv(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// SIMD-ONLY2-NEXT:    call void @_ZN2SD3fooEv(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// SIMD-ONLY2-NEXT:    call void @_ZN2SE3fooEv(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// SIMD-ONLY2-NEXT:    call void @_ZN2STILi100EE3fooEv(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// SIMD-ONLY2-NEXT:    call void @_ZN2STILi1000EE3fooEv(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP1]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[R]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP2]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[R]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC1:%.*]] = add nsw i32 [[TMP3]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC1]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP2:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[R]], align 4
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP4]], [[TMP6]]
// SIMD-ONLY2-NEXT:    ret i32 [[ADD]]
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SA3fooEv
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 1
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SB3fooEv
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 4
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP4:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SC3fooEv
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 7
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SD3fooEv
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SE3fooEv
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 13
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP5:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2STILi100EE3fooEv
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 116
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP6:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2STILi1000EE3fooEv
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 1016
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP7:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SAC2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 2
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SAD2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 3
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SBC2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 5
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SBD2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 6
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SCC2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 8
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP8:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SCD2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 9
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SDC2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 11
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SDD2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 12
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP9:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SEC2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 14
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP10:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2SED2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 15
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP11:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2STILi100EEC2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 117
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP12:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2STILi100EED2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 118
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP13:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2STILi1000EEC2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 1017
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP14:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_ZN2STILi1000EED2Ev
// SIMD-ONLY2-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// SIMD-ONLY2-NEXT:    [[A:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    [[I:%.*]] = alloca i32, align 4
// SIMD-ONLY2-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// SIMD-ONLY2-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// SIMD-ONLY2-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    store i32 0, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND:%.*]]
// SIMD-ONLY2:       for.cond:
// SIMD-ONLY2-NEXT:    [[TMP2:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[CMP:%.*]] = icmp slt i32 [[TMP2]], 10
// SIMD-ONLY2-NEXT:    br i1 [[CMP]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]
// SIMD-ONLY2:       for.body:
// SIMD-ONLY2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 1018
// SIMD-ONLY2-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_INC:%.*]]
// SIMD-ONLY2:       for.inc:
// SIMD-ONLY2-NEXT:    [[TMP4:%.*]] = load i32, ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// SIMD-ONLY2-NEXT:    store i32 [[INC]], ptr [[I]], align 4
// SIMD-ONLY2-NEXT:    br label [[FOR_COND]], !llvm.loop [[LOOP15:![0-9]+]]
// SIMD-ONLY2:       for.end:
// SIMD-ONLY2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[A]], align 4
// SIMD-ONLY2-NEXT:    [[TMP6:%.*]] = load ptr, ptr @R, align 8
// SIMD-ONLY2-NEXT:    store i32 [[TMP5]], ptr [[TMP6]], align 4
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_GLOBAL__I_000500
// SIMD-ONLY2-SAME: () #[[ATTR0]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @__cxx_global_var_init()
// SIMD-ONLY2-NEXT:    call void @__cxx_global_var_init.2()
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_GLOBAL__I_000501
// SIMD-ONLY2-SAME: () #[[ATTR0]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @__cxx_global_var_init.3()
// SIMD-ONLY2-NEXT:    ret void
//
//
// SIMD-ONLY2-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_target_parallel_generic_loop_codegen_1.cpp
// SIMD-ONLY2-SAME: () #[[ATTR0]] {
// SIMD-ONLY2-NEXT:  entry:
// SIMD-ONLY2-NEXT:    call void @__cxx_global_var_init.1()
// SIMD-ONLY2-NEXT:    call void @__cxx_global_var_init.4()
// SIMD-ONLY2-NEXT:    call void @__cxx_global_var_init.5()
// SIMD-ONLY2-NEXT:    call void @__cxx_global_var_init.6()
// SIMD-ONLY2-NEXT:    call void @__cxx_global_var_init.7()
// SIMD-ONLY2-NEXT:    call void @__cxx_global_var_init.8()
// SIMD-ONLY2-NEXT:    ret void
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init
// OMP-DEFAULT-SAME: () #[[ATTR0:[0-9]+]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @_ZL2a1, ptr @__dso_handle) #[[ATTR2:[0-9]+]]
// OMP-DEFAULT-NEXT:    ret void
//
//
//
//
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// OMP-DEFAULT-SAME: () #[[ATTR0]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @a2, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEFAULT-NEXT:    ret void
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
// OMP-DEFAULT-SAME: () #[[ATTR0]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b1, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEFAULT-NEXT:    ret void
//
//
//
//
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.3
// OMP-DEFAULT-SAME: () #[[ATTR0]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b2, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEFAULT-NEXT:    ret void
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.4
// OMP-DEFAULT-SAME: () #[[ATTR0]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @_ZN2SCC1Ev(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SCD1Ev, ptr @_ZL2c1, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEFAULT-NEXT:    ret void
//
//
//
//
//
//
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.5
// OMP-DEFAULT-SAME: () #[[ATTR0]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @_ZN2SDC1Ev(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SDD1Ev, ptr @d1, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEFAULT-NEXT:    ret void
//
//
//
//
//
//
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.8
// OMP-DEFAULT-SAME: () #[[ATTR0]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @_ZN2SEC1Ev(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SED1Ev, ptr @e1, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEFAULT-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.13
// OMP-DEFAULT-SAME: () #[[ATTR0]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @_ZN2STILi100EEC1Ev(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi100EED1Ev, ptr @t1, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEFAULT-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.18
// OMP-DEFAULT-SAME: () #[[ATTR0]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @_ZN2STILi1000EEC1Ev(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi1000EED1Ev, ptr @t2, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEFAULT-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@_GLOBAL__I_000500
// OMP-DEFAULT-SAME: () #[[ATTR0]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init()
// OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.2()
// OMP-DEFAULT-NEXT:    ret void
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@_GLOBAL__I_000501
// OMP-DEFAULT-SAME: () #[[ATTR0]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.3()
// OMP-DEFAULT-NEXT:    ret void
//
//
// OMP-DEFAULT-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_target_parallel_generic_loop_codegen_1.cpp
// OMP-DEFAULT-SAME: () #[[ATTR0]] {
// OMP-DEFAULT-NEXT:  entry:
// OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.1()
// OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.4()
// OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.5()
// OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.8()
// OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.13()
// OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.18()
// OMP-DEFAULT-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init
// OMP-DEfAULT-SAME: () #[[ATTR0:[0-9]+]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @_ZL2a1, ptr @__dso_handle) #[[ATTR2:[0-9]+]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SAC1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2SAC2Ev(ptr noundef nonnull align 4 dereferenceable(16) [[THIS1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SAD1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2SAD2Ev(ptr noundef nonnull align 4 dereferenceable(16) [[THIS1]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SAC2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 2
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SAD2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 3
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// OMP-DEfAULT-SAME: () #[[ATTR0]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @a2, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
// OMP-DEfAULT-SAME: () #[[ATTR0]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b1, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SBC1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2SBC2Ev(ptr noundef nonnull align 4 dereferenceable(32) [[THIS1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SBD1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2SBD2Ev(ptr noundef nonnull align 4 dereferenceable(32) [[THIS1]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SBC2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 5
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SBD2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 6
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.3
// OMP-DEfAULT-SAME: () #[[ATTR0]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b2, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.4
// OMP-DEfAULT-SAME: () #[[ATTR0]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @_ZN2SCC1Ev(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SCD1Ev, ptr @_ZL2c1, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SCC1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2SCC2Ev(ptr noundef nonnull align 4 dereferenceable(64) [[THIS1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SCD1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2SCD2Ev(ptr noundef nonnull align 4 dereferenceable(64) [[THIS1]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SCC2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP6]], align 4
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP8]], ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP16]], align 4
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP18]], align 8
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP21]], align 4
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2:[0-9]+]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP24]], ptr [[TMP25]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3:[0-9]+]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 8
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SCD2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.5
// OMP-DEfAULT-SAME: () #[[ATTR0]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @_ZN2SDC1Ev(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SDD1Ev, ptr @d1, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SDC1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2SDC2Ev(ptr noundef nonnull align 4 dereferenceable(128) [[THIS1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SDD1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2SDD2Ev(ptr noundef nonnull align 4 dereferenceable(128) [[THIS1]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SDC2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 11
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SDD2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP6]], align 4
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP8]], ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes.6, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes.7, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP16]], align 4
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP18]], align 8
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP21]], align 4
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP24]], ptr [[TMP25]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 12
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.8
// OMP-DEfAULT-SAME: () #[[ATTR0]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @_ZN2SEC1Ev(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SED1Ev, ptr @e1, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SEC1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2SEC2Ev(ptr noundef nonnull align 4 dereferenceable(256) [[THIS1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SED1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2SED2Ev(ptr noundef nonnull align 4 dereferenceable(256) [[THIS1]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SEC2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP6]], align 4
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP8]], ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes.9, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes.10, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP16]], align 4
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP18]], align 8
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP21]], align 4
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP24]], ptr [[TMP25]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 14
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SED2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP6]], align 4
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP8]], ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes.11, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes.12, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP16]], align 4
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP18]], align 8
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP21]], align 4
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP24]], ptr [[TMP25]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 15
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.13
// OMP-DEfAULT-SAME: () #[[ATTR0]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @_ZN2STILi100EEC1Ev(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi100EED1Ev, ptr @t1, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2STILi100EEC1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2STILi100EEC2Ev(ptr noundef nonnull align 4 dereferenceable(912) [[THIS1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2STILi100EED1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2STILi100EED2Ev(ptr noundef nonnull align 4 dereferenceable(912) [[THIS1]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2STILi100EEC2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP6]], align 4
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP8]], ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes.14, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes.15, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP16]], align 4
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP18]], align 8
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP21]], align 4
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP24]], ptr [[TMP25]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 117
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2STILi100EED2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP6]], align 4
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP8]], ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes.16, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes.17, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP16]], align 4
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP18]], align 8
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP21]], align 4
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP24]], ptr [[TMP25]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 118
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.18
// OMP-DEfAULT-SAME: () #[[ATTR0]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @_ZN2STILi1000EEC1Ev(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi1000EED1Ev, ptr @t2, ptr @__dso_handle) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2STILi1000EEC1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2STILi1000EEC2Ev(ptr noundef nonnull align 4 dereferenceable(4512) [[THIS1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2STILi1000EED1Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2STILi1000EED2Ev(ptr noundef nonnull align 4 dereferenceable(4512) [[THIS1]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2STILi1000EEC2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP6]], align 4
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP8]], ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes.19, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes.20, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP16]], align 4
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP18]], align 8
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP21]], align 4
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP24]], ptr [[TMP25]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 1017
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2STILi1000EED2Ev
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP6]], align 4
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP8]], ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes.21, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes.22, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP16]], align 4
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP18]], align 8
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP21]], align 4
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP24]], ptr [[TMP25]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 1018
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_Z3bari
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR1]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[R:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[R_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[R]], align 4
// OMP-DEfAULT-NEXT:    call void @_ZN2SA3fooEv(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// OMP-DEfAULT-NEXT:    call void @_ZN2SA3fooEv(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// OMP-DEfAULT-NEXT:    call void @_ZN2SB3fooEv(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// OMP-DEfAULT-NEXT:    call void @_ZN2SB3fooEv(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// OMP-DEfAULT-NEXT:    call void @_ZN2SC3fooEv(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// OMP-DEfAULT-NEXT:    call void @_ZN2SD3fooEv(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// OMP-DEfAULT-NEXT:    call void @_ZN2SE3fooEv(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// OMP-DEfAULT-NEXT:    call void @_ZN2STILi100EE3fooEv(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// OMP-DEfAULT-NEXT:    call void @_ZN2STILi1000EE3fooEv(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[R]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[R_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[R_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[TMP3]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP8]], align 4
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP6]], ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes.23, ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes.24, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP16]], align 8
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267(i32 [[TMP2]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = load i32, ptr [[R]], align 4
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load i32, ptr [[TMP24]], align 4
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP23]], [[TMP25]]
// OMP-DEfAULT-NEXT:    ret i32 [[ADD]]
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SA3fooEv
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SB3fooEv
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP6]], align 4
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP8]], ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes.25, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes.26, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP16]], align 4
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP18]], align 8
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP21]], align 4
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP24]], ptr [[TMP25]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SC3fooEv
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 7
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SD3fooEv
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 10
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2SE3fooEv
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2STILi100EE3fooEv
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP6]], align 4
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP8]], ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes.27, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes.28, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP16]], align 4
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP18]], align 8
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP21]], align 4
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP24]], ptr [[TMP25]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_ZN2STILi1000EE3fooEv
// OMP-DEfAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) #[[ATTR1]] comdat align 2 {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
// OMP-DEfAULT-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
// OMP-DEfAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP5]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP6]], align 4
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
// OMP-DEfAULT-NEXT:    store i32 3, ptr [[TMP9]], align 4
// OMP-DEfAULT-NEXT:    [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[TMP10]], align 4
// OMP-DEfAULT-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
// OMP-DEfAULT-NEXT:    store ptr [[TMP7]], ptr [[TMP11]], align 4
// OMP-DEfAULT-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
// OMP-DEfAULT-NEXT:    store ptr [[TMP8]], ptr [[TMP12]], align 4
// OMP-DEfAULT-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
// OMP-DEfAULT-NEXT:    store ptr @.offload_sizes.29, ptr [[TMP13]], align 4
// OMP-DEfAULT-NEXT:    [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
// OMP-DEfAULT-NEXT:    store ptr @.offload_maptypes.30, ptr [[TMP14]], align 4
// OMP-DEfAULT-NEXT:    [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP15]], align 4
// OMP-DEfAULT-NEXT:    [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
// OMP-DEfAULT-NEXT:    store ptr null, ptr [[TMP16]], align 4
// OMP-DEfAULT-NEXT:    [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP17]], align 8
// OMP-DEfAULT-NEXT:    [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
// OMP-DEfAULT-NEXT:    store i64 0, ptr [[TMP18]], align 8
// OMP-DEfAULT-NEXT:    [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
// OMP-DEfAULT-NEXT:    store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP19]], align 4
// OMP-DEfAULT-NEXT:    [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
// OMP-DEfAULT-NEXT:    store [3 x i32] zeroinitializer, ptr [[TMP20]], align 4
// OMP-DEfAULT-NEXT:    [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[TMP21]], align 4
// OMP-DEfAULT-NEXT:    [[TMP22:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB2]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211.region_id, ptr [[KERNEL_ARGS]])
// OMP-DEfAULT-NEXT:    [[TMP23:%.*]] = icmp ne i32 [[TMP22]], 0
// OMP-DEfAULT-NEXT:    br i1 [[TMP23]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
// OMP-DEfAULT:       omp_offload.failed:
// OMP-DEfAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211(i32 [[TMP3]]) #[[ATTR2]]
// OMP-DEfAULT-NEXT:    br label [[OMP_OFFLOAD_CONT]]
// OMP-DEfAULT:       omp_offload.cont:
// OMP-DEfAULT-NEXT:    [[TMP24:%.*]] = load i32, ptr [[A]], align 4
// OMP-DEfAULT-NEXT:    [[TMP25:%.*]] = load ptr, ptr @R, align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP24]], ptr [[TMP25]], align 4
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267
// OMP-DEfAULT-SAME: (i32 noundef [[R:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[R_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[R_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[R]], ptr [[R_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[R_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[R_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[R_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[R:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[R_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[R]], ptr [[R_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[R_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP8]], 1
// OMP-DEfAULT-NEXT:    store i32 [[INC]], ptr [[R_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 4
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 13
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 116
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211
// OMP-DEfAULT-SAME: (i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[A_CASTED:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// OMP-DEfAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211.omp_outlined, i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211.omp_outlined
// OMP-DEfAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR3]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// OMP-DEfAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// OMP-DEfAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// OMP-DEfAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// OMP-DEfAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// OMP-DEfAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// OMP-DEfAULT:       cond.true:
// OMP-DEfAULT-NEXT:    br label [[COND_END:%.*]]
// OMP-DEfAULT:       cond.false:
// OMP-DEfAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    br label [[COND_END]]
// OMP-DEfAULT:       cond.end:
// OMP-DEfAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// OMP-DEfAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// OMP-DEfAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// OMP-DEfAULT:       omp.inner.for.cond:
// OMP-DEfAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// OMP-DEfAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// OMP-DEfAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// OMP-DEfAULT:       omp.inner.for.body:
// OMP-DEfAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// OMP-DEfAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// OMP-DEfAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// OMP-DEfAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 1016
// OMP-DEfAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// OMP-DEfAULT:       omp.body.continue:
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// OMP-DEfAULT:       omp.inner.for.inc:
// OMP-DEfAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// OMP-DEfAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// OMP-DEfAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// OMP-DEfAULT:       omp.inner.for.end:
// OMP-DEfAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// OMP-DEfAULT:       omp.loop.exit:
// OMP-DEfAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_GLOBAL__I_000500
// OMP-DEfAULT-SAME: () #[[ATTR0]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @__cxx_global_var_init()
// OMP-DEfAULT-NEXT:    call void @__cxx_global_var_init.2()
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_GLOBAL__I_000501
// OMP-DEfAULT-SAME: () #[[ATTR0]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @__cxx_global_var_init.3()
// OMP-DEfAULT-NEXT:    ret void
//
//
// OMP-DEfAULT-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_target_parallel_generic_loop_codegen_1.cpp
// OMP-DEfAULT-SAME: () #[[ATTR0]] {
// OMP-DEfAULT-NEXT:  entry:
// OMP-DEfAULT-NEXT:    call void @__cxx_global_var_init.1()
// OMP-DEfAULT-NEXT:    call void @__cxx_global_var_init.4()
// OMP-DEfAULT-NEXT:    call void @__cxx_global_var_init.5()
// OMP-DEfAULT-NEXT:    call void @__cxx_global_var_init.8()
// OMP-DEfAULT-NEXT:    call void @__cxx_global_var_init.13()
// OMP-DEfAULT-NEXT:    call void @__cxx_global_var_init.18()
// OMP-DEfAULT-NEXT:    ret void
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0:[0-9]+]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @_ZL2a1, ptr @__dso_handle) #[[ATTR2:[0-9]+]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SAC1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SAC2Ev(ptr noundef nonnull align 4 dereferenceable(16) [[THIS1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SAD1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SAD2Ev(ptr noundef nonnull align 4 dereferenceable(16) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SAC2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 2
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SAD2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 3
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SAC1Ev(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SAD1Ev, ptr @a2, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SBC1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SBC2Ev(ptr noundef nonnull align 4 dereferenceable(32) [[THIS1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SBD1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SBD2Ev(ptr noundef nonnull align 4 dereferenceable(32) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SBC2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 5
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SBD2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 6
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.3
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SBC1Ev(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SBD1Ev, ptr @b2, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.4
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SCC1Ev(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SCD1Ev, ptr @_ZL2c1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SCC1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SCC2Ev(ptr noundef nonnull align 4 dereferenceable(64) [[THIS1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SCD1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SCD2Ev(ptr noundef nonnull align 4 dereferenceable(64) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SCC2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3:[0-9]+]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2:[0-9]+]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SCC1Ev_l148.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1:[0-9]+]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SCD2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.5
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SDC1Ev(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SDD1Ev, ptr @d1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SDC1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SDC2Ev(ptr noundef nonnull align 4 dereferenceable(128) [[THIS1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SDD1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SDD2Ev(ptr noundef nonnull align 4 dereferenceable(128) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SDC2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 11
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SDD2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SDD1Ev_l174.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 12
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.6
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SEC1Ev(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2SED1Ev, ptr @e1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SEC1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SEC2Ev(ptr noundef nonnull align 4 dereferenceable(256) [[THIS1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SED1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SED2Ev(ptr noundef nonnull align 4 dereferenceable(256) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SEC2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SEC1Ev_l192.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 14
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SED2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SED1Ev_l199.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 15
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.7
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2STILi100EEC1Ev(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi100EED1Ev, ptr @t1, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2STILi100EEC1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2STILi100EEC2Ev(ptr noundef nonnull align 4 dereferenceable(912) [[THIS1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2STILi100EED1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2STILi100EED2Ev(ptr noundef nonnull align 4 dereferenceable(912) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2STILi100EEC2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EEC1Ev_l218.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 117
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2STILi100EED2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EED1Ev_l225.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 118
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@__cxx_global_var_init.8
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2STILi1000EEC1Ev(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = call i32 @__cxa_atexit(ptr @_ZN2STILi1000EED1Ev, ptr @t2, ptr @__dso_handle) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2STILi1000EEC1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2STILi1000EEC2Ev(ptr noundef nonnull align 4 dereferenceable(4512) [[THIS1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2STILi1000EED1Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2STILi1000EED2Ev(ptr noundef nonnull align 4 dereferenceable(4512) [[THIS1]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2STILi1000EEC2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EEC1Ev_l218.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 1017
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2STILi1000EED2Ev
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EED1Ev_l225.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 1018
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_Z3bari
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i32 noundef signext [[A:%.*]]) #[[ATTR1]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[R:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[R_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[R]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SA3fooEv(ptr noundef nonnull align 4 dereferenceable(16) @_ZL2a1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SA3fooEv(ptr noundef nonnull align 4 dereferenceable(16) @a2)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SB3fooEv(ptr noundef nonnull align 4 dereferenceable(32) @b1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SB3fooEv(ptr noundef nonnull align 4 dereferenceable(32) @b2)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SC3fooEv(ptr noundef nonnull align 4 dereferenceable(64) @_ZL2c1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SD3fooEv(ptr noundef nonnull align 4 dereferenceable(128) @d1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2SE3fooEv(ptr noundef nonnull align 4 dereferenceable(256) @e1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2STILi100EE3fooEv(ptr noundef nonnull align 4 dereferenceable(912) @t1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @_ZN2STILi1000EE3fooEv(ptr noundef nonnull align 4 dereferenceable(4512) @t2)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[R]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[R_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i64, ptr [[R_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267(i64 [[TMP2]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[R]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[TMP4]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], [[TMP5]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret i32 [[ADD]]
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SA3fooEv
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(16) [[THIS:%.*]]) #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SB3fooEv
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(32) [[THIS:%.*]]) #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SC3fooEv
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(64) [[THIS:%.*]]) #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 7
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SD3fooEv
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(128) [[THIS:%.*]]) #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], 10
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP3]], ptr [[TMP4]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2SE3fooEv
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(256) [[THIS:%.*]]) #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2STILi100EE3fooEv
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(912) [[THIS:%.*]]) #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_ZN2STILi1000EE3fooEv
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noundef nonnull align 4 dereferenceable(4512) [[THIS:%.*]]) #[[ATTR1]] comdat {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP1]], ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP2]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211(i64 [[TMP3]]) #[[ATTR2]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[A]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load ptr, ptr @R, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[TMP5]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[R:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[R_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[R_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[R]], ptr [[R_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[R_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[R_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[R_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3bari_l267.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[R:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[R_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[R]], ptr [[R_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[R_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[INC:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[INC]], ptr [[R_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SB3fooEv_l122.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2SE3fooEv_l185.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 13
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi100EE3fooEv_l211.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 116
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211
// CHECK-NTARGET-OMP-DEFAULT-SAME: (i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB2]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211.omp_outlined, i64 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2STILi1000EE3fooEv_l211.omp_outlined
// CHECK-NTARGET-OMP-DEFAULT-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR3]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IV:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_LB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_UB:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_STRIDE:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[I:%.*]] = alloca i32, align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 9, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 1, ptr [[DOTOMP_STRIDE]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 0, ptr [[DOTOMP_IS_LAST]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP1:%.*]] = load i32, ptr [[TMP0]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_init_4(ptr @[[GLOB1]], i32 [[TMP1]], i32 34, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP2:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP:%.*]] = icmp sgt i32 [[TMP2]], 9
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.true:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.false:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP3:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[COND_END]]
// CHECK-NTARGET-OMP-DEFAULT:       cond.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP3]], [[COND_FALSE]] ]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[COND]], ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP4:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[TMP4]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.cond:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP6:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[CMP1:%.*]] = icmp sle i32 [[TMP5]], [[TMP6]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br i1 [[CMP1]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.body:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP7]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD:%.*]] = add nsw i32 0, [[MUL]]
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD]], ptr [[I]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD2:%.*]] = add nsw i32 [[TMP8]], 1016
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD2]], ptr [[A_ADDR]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_BODY_CONTINUE:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.body.continue:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.inc:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP9]], 1
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    store i32 [[ADD3]], ptr [[DOTOMP_IV]], align 4
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_INNER_FOR_COND]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.inner.for.end:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    br label [[OMP_LOOP_EXIT:%.*]]
// CHECK-NTARGET-OMP-DEFAULT:       omp.loop.exit:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__kmpc_for_static_fini(ptr @[[GLOB1]], i32 [[TMP1]])
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_GLOBAL__I_000500
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init()
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.2()
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_GLOBAL__I_000501
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.3()
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//
// CHECK-NTARGET-OMP-DEFAULT-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_target_parallel_generic_loop_codegen_1.cpp
// CHECK-NTARGET-OMP-DEFAULT-SAME: () #[[ATTR0]] {
// CHECK-NTARGET-OMP-DEFAULT-NEXT:  entry:
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.1()
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.4()
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.5()
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.6()
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.7()
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    call void @__cxx_global_var_init.8()
// CHECK-NTARGET-OMP-DEFAULT-NEXT:    ret void
//
//// NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
// TCHECK: {{.*}}