clang/test/OpenMP/nvptx_target_parallel_num_threads_codegen.cpp

   1 // 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 _
   2 // Test target codegen - host bc file has to be created first.
   3 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc
   4 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK1
   5 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc
   6 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK2
   7 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK2
   8
   9 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc
  10 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK1
  11 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc
  12 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK2
  13 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK2
  14
  15 // expected-no-diagnostics
  16 #ifndef HEADER
  17 #define HEADER
  18
  19 template<typename tx>
  20 tx ftemplate(int n) {
  21   tx a = 0;
  22   short aa = 0;
  23   tx b[10];
  24
  25   #pragma omp target parallel map(tofrom: aa) num_threads(1024)
  26   {
  27     aa += 1;
  28   }
  29
  30   #pragma omp target parallel map(tofrom:a, aa, b) if(target: n>40) num_threads(n)
  31   {
  32     a += 1;
  33     aa += 1;
  34     b[2] += 1;
  35   }
  36
  37   return a;
  38 }
  39
  40 int bar(int n){
  41   int a = 0;
  42
  43   a += ftemplate<int>(n);
  44
  45   return a;
  46 }
  47
  48 #endif
  49 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l25
  50 // CHECK1-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[AA:%.*]]) #[[ATTR0:[0-9]+]] {
  51 // CHECK1-NEXT:  entry:
  52 // CHECK1-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
  53 // CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 8
  54 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
  55 // CHECK1-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
  56 // CHECK1-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
  57 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
  58 // CHECK1-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l25_kernel_environment, ptr [[DYN_PTR]])
  59 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
  60 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
  61 // CHECK1:       user_code.entry:
  62 // CHECK1-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
  63 // CHECK1-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
  64 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 8
  65 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 1024, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l25_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 1)
  66 // CHECK1-NEXT:    call void @__kmpc_target_deinit()
  67 // CHECK1-NEXT:    ret void
  68 // CHECK1:       worker.exit:
  69 // CHECK1-NEXT:    ret void
  70 //
  71 //
  72 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l25_omp_outlined
  73 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[AA:%.*]]) #[[ATTR1:[0-9]+]] {
  74 // CHECK1-NEXT:  entry:
  75 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
  76 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
  77 // CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 8
  78 // CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
  79 // CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
  80 // CHECK1-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
  81 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
  82 // CHECK1-NEXT:    [[TMP1:%.*]] = load i16, ptr [[TMP0]], align 2
  83 // CHECK1-NEXT:    [[CONV:%.*]] = sext i16 [[TMP1]] to i32
  84 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[CONV]], 1
  85 // CHECK1-NEXT:    [[CONV1:%.*]] = trunc i32 [[ADD]] to i16
  86 // CHECK1-NEXT:    store i16 [[CONV1]], ptr [[TMP0]], align 2
  87 // CHECK1-NEXT:    ret void
  88 //
  89 //
  90 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30
  91 // CHECK1-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]], i64 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR4:[0-9]+]] {
  92 // CHECK1-NEXT:  entry:
  93 // CHECK1-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
  94 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
  95 // CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 8
  96 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
  97 // CHECK1-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i64, align 8
  98 // CHECK1-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x ptr], align 8
  99 // CHECK1-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
 100 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 101 // CHECK1-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
 102 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 103 // CHECK1-NEXT:    store i64 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 8
 104 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 105 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
 106 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 107 // CHECK1-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30_kernel_environment, ptr [[DYN_PTR]])
 108 // CHECK1-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP3]], -1
 109 // CHECK1-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 110 // CHECK1:       user_code.entry:
 111 // CHECK1-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 112 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 113 // CHECK1-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
 114 // CHECK1-NEXT:    store ptr [[TMP0]], ptr [[TMP6]], align 8
 115 // CHECK1-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 1
 116 // CHECK1-NEXT:    store ptr [[TMP1]], ptr [[TMP7]], align 8
 117 // CHECK1-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 2
 118 // CHECK1-NEXT:    store ptr [[TMP2]], ptr [[TMP8]], align 8
 119 // CHECK1-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP4]], i32 1, i32 [[TMP5]], i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 3)
 120 // CHECK1-NEXT:    call void @__kmpc_target_deinit()
 121 // CHECK1-NEXT:    ret void
 122 // CHECK1:       worker.exit:
 123 // CHECK1-NEXT:    ret void
 124 //
 125 //
 126 // CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30_omp_outlined
 127 // CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR1]] {
 128 // CHECK1-NEXT:  entry:
 129 // CHECK1-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 130 // CHECK1-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
 131 // CHECK1-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 8
 132 // CHECK1-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 8
 133 // CHECK1-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 8
 134 // CHECK1-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
 135 // CHECK1-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
 136 // CHECK1-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 8
 137 // CHECK1-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 8
 138 // CHECK1-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 8
 139 // CHECK1-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
 140 // CHECK1-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[AA_ADDR]], align 8
 141 // CHECK1-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 8
 142 // CHECK1-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP0]], align 4
 143 // CHECK1-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 1
 144 // CHECK1-NEXT:    store i32 [[ADD]], ptr [[TMP0]], align 4
 145 // CHECK1-NEXT:    [[TMP4:%.*]] = load i16, ptr [[TMP1]], align 2
 146 // CHECK1-NEXT:    [[CONV:%.*]] = sext i16 [[TMP4]] to i32
 147 // CHECK1-NEXT:    [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
 148 // CHECK1-NEXT:    [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
 149 // CHECK1-NEXT:    store i16 [[CONV2]], ptr [[TMP1]], align 2
 150 // CHECK1-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP2]], i64 0, i64 2
 151 // CHECK1-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 152 // CHECK1-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP5]], 1
 153 // CHECK1-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4
 154 // CHECK1-NEXT:    ret void
 155 //
 156 //
 157 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l25
 158 // CHECK2-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[AA:%.*]]) #[[ATTR0:[0-9]+]] {
 159 // CHECK2-NEXT:  entry:
 160 // CHECK2-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
 161 // CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
 162 // CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 4
 163 // CHECK2-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
 164 // CHECK2-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 165 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
 166 // CHECK2-NEXT:    [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l25_kernel_environment, ptr [[DYN_PTR]])
 167 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
 168 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 169 // CHECK2:       user_code.entry:
 170 // CHECK2-NEXT:    [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
 171 // CHECK2-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
 172 // CHECK2-NEXT:    store ptr [[TMP0]], ptr [[TMP3]], align 4
 173 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 1024, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l25_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 1)
 174 // CHECK2-NEXT:    call void @__kmpc_target_deinit()
 175 // CHECK2-NEXT:    ret void
 176 // CHECK2:       worker.exit:
 177 // CHECK2-NEXT:    ret void
 178 //
 179 //
 180 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l25_omp_outlined
 181 // CHECK2-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[AA:%.*]]) #[[ATTR1:[0-9]+]] {
 182 // CHECK2-NEXT:  entry:
 183 // CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 184 // CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 185 // CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
 186 // CHECK2-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 187 // CHECK2-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 188 // CHECK2-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 189 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
 190 // CHECK2-NEXT:    [[TMP1:%.*]] = load i16, ptr [[TMP0]], align 2
 191 // CHECK2-NEXT:    [[CONV:%.*]] = sext i16 [[TMP1]] to i32
 192 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[CONV]], 1
 193 // CHECK2-NEXT:    [[CONV1:%.*]] = trunc i32 [[ADD]] to i16
 194 // CHECK2-NEXT:    store i16 [[CONV1]], ptr [[TMP0]], align 2
 195 // CHECK2-NEXT:    ret void
 196 //
 197 //
 198 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30
 199 // CHECK2-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]], i32 noundef [[DOTCAPTURE_EXPR_:%.*]]) #[[ATTR4:[0-9]+]] {
 200 // CHECK2-NEXT:  entry:
 201 // CHECK2-NEXT:    [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
 202 // CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 203 // CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
 204 // CHECK2-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 205 // CHECK2-NEXT:    [[DOTCAPTURE_EXPR__ADDR:%.*]] = alloca i32, align 4
 206 // CHECK2-NEXT:    [[CAPTURED_VARS_ADDRS:%.*]] = alloca [3 x ptr], align 4
 207 // CHECK2-NEXT:    store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
 208 // CHECK2-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 209 // CHECK2-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 210 // CHECK2-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 211 // CHECK2-NEXT:    store i32 [[DOTCAPTURE_EXPR_]], ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 212 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 213 // CHECK2-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
 214 // CHECK2-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 215 // CHECK2-NEXT:    [[TMP3:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30_kernel_environment, ptr [[DYN_PTR]])
 216 // CHECK2-NEXT:    [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP3]], -1
 217 // CHECK2-NEXT:    br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
 218 // CHECK2:       user_code.entry:
 219 // CHECK2-NEXT:    [[TMP4:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
 220 // CHECK2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[DOTCAPTURE_EXPR__ADDR]], align 4
 221 // CHECK2-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 0
 222 // CHECK2-NEXT:    store ptr [[TMP0]], ptr [[TMP6]], align 4
 223 // CHECK2-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 1
 224 // CHECK2-NEXT:    store ptr [[TMP1]], ptr [[TMP7]], align 4
 225 // CHECK2-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [3 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i32 0, i32 2
 226 // CHECK2-NEXT:    store ptr [[TMP2]], ptr [[TMP8]], align 4
 227 // CHECK2-NEXT:    call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP4]], i32 1, i32 [[TMP5]], i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i32 3)
 228 // CHECK2-NEXT:    call void @__kmpc_target_deinit()
 229 // CHECK2-NEXT:    ret void
 230 // CHECK2:       worker.exit:
 231 // CHECK2-NEXT:    ret void
 232 //
 233 //
 234 // CHECK2-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l30_omp_outlined
 235 // CHECK2-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[A:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR1]] {
 236 // CHECK2-NEXT:  entry:
 237 // CHECK2-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
 238 // CHECK2-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
 239 // CHECK2-NEXT:    [[A_ADDR:%.*]] = alloca ptr, align 4
 240 // CHECK2-NEXT:    [[AA_ADDR:%.*]] = alloca ptr, align 4
 241 // CHECK2-NEXT:    [[B_ADDR:%.*]] = alloca ptr, align 4
 242 // CHECK2-NEXT:    store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
 243 // CHECK2-NEXT:    store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
 244 // CHECK2-NEXT:    store ptr [[A]], ptr [[A_ADDR]], align 4
 245 // CHECK2-NEXT:    store ptr [[AA]], ptr [[AA_ADDR]], align 4
 246 // CHECK2-NEXT:    store ptr [[B]], ptr [[B_ADDR]], align 4
 247 // CHECK2-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 4
 248 // CHECK2-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[AA_ADDR]], align 4
 249 // CHECK2-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[B_ADDR]], align 4
 250 // CHECK2-NEXT:    [[TMP3:%.*]] = load i32, ptr [[TMP0]], align 4
 251 // CHECK2-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP3]], 1
 252 // CHECK2-NEXT:    store i32 [[ADD]], ptr [[TMP0]], align 4
 253 // CHECK2-NEXT:    [[TMP4:%.*]] = load i16, ptr [[TMP1]], align 2
 254 // CHECK2-NEXT:    [[CONV:%.*]] = sext i16 [[TMP4]] to i32
 255 // CHECK2-NEXT:    [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
 256 // CHECK2-NEXT:    [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
 257 // CHECK2-NEXT:    store i16 [[CONV2]], ptr [[TMP1]], align 2
 258 // CHECK2-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP2]], i32 0, i32 2
 259 // CHECK2-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
 260 // CHECK2-NEXT:    [[ADD3:%.*]] = add nsw i32 [[TMP5]], 1
 261 // CHECK2-NEXT:    store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4
 262 // CHECK2-NEXT:    ret void
 263 //