clang/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp

   1 // Test host codegen.
   2 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP50 --check-prefix OMP50-64
   3 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP45
   4 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
   5 // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP50 --check-prefix OMP50-64
   6 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP50 --check-prefix OMP50-32
   7 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
   8 // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP50 --check-prefix OMP50-32
   9
  10
  11 // 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
  12 // 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 -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix OMP45
  13 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP45
  14 // 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
  15 // 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 -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix OMP45
  16
  17
  18 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
  19 // 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
  20 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
  21 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
  22 // 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
  23 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
  24 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
  25
  26 // Test target codegen - host bc file has to be created first.
  27 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
  28 // RUN: %clang_cc1 -verify -Wno-vla -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 --check-prefix TCHECK-64
  29 // 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
  30 // 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 -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
  31 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
  32 // RUN: %clang_cc1 -verify -Wno-vla -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 --check-prefix TCHECK-32
  33 // 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
  34 // 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 -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
  35
  36 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
  37 // RUN: %clang_cc1 -verify -Wno-vla -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
  38 // 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
  39 // 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 -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
  40 // RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
  41 // RUN: %clang_cc1 -verify -Wno-vla -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
  42 // 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
  43 // 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 -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
  44 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
  45
  46 // expected-no-diagnostics
  47 #ifndef HEADER
  48 #define HEADER
  49
  50 // CHECK-DAG: [[TT:%.+]] = type { i64, i8 }
  51 // CHECK-DAG: [[ENTTY:%.+]] = type { ptr, ptr, i[[SZ:32|64]], i32, i32 }
  52
  53 // TCHECK: [[ENTTY:%.+]] = type { ptr, ptr, i{{32|64}}, i32, i32 }
  54
  55 // OMP45-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i64] [i64 0, i64 4]
  56 // OMP45-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800]
  57 // OMP45-DAG: @{{.*}} = weak constant i8 0
  58
  59 // OMP50-DAG: [[SIZET:@.+]] = private unnamed_addr constant [3 x i64] [i64 0, i64 4, i64 1]
  60 // OMP50-DAG: [[MAPT:@.+]] = private unnamed_addr constant [3 x i64] [i64 544, i64 800, i64 800]
  61 // OMP50-DAG: @{{.*}} = weak constant i8 0
  62
  63
  64 // TCHECK: @{{.+}} = weak constant [[ENTTY]]
  65 // TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]]
  66 // TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]]
  67
  68 template<typename tx, typename ty>
  69 struct TT{
  70   tx X;
  71   ty Y;
  72 };
  73
  74 int global;
  75 extern int global;
  76
  77 // CHECK: define {{.*}}[[FOO:@.+]](
  78 int foo(int n) {
  79   int a = 0;
  80   short aa = 0;
  81   float b[10];
  82   float bn[n];
  83   double c[5][10];
  84   double cn[5][n];
  85   TT<long long, char> d;
  86   static long *plocal;
  87
  88   // CHECK:       [[ADD:%.+]] = add nsw i32
  89   // CHECK:       store i32 [[ADD]], ptr [[DEVICE_CAP:%.+]],
  90   // CHECK:       [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 0
  91   // CHECK:       [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],
  92   // CHECK:       store i32 [[DEV]], ptr [[GEP]],
  93   // CHECK:       [[TASK:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr [[ID:@.+]], i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, ptr [[TASK_ENTRY0:@.+]])
  94   // CHECK:       getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 0
  95   // CHECK:       getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 1
  96   // CHECK:       getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 2
  97   // CHECK:       getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 3
  98   // CHECK:       call void @__kmpc_omp_taskwait_deps_51(ptr [[ID]], i32 [[GTID]], i32 4, ptr %{{.+}}, i32 0, ptr null, i32 0)
  99   // CHECK:       call void @__kmpc_omp_task_begin_if0(ptr [[ID]], i32 [[GTID]], ptr [[TASK]])
 100   // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], ptr [[TASK]])
 101   // CHECK:       call void @__kmpc_omp_task_complete_if0(ptr [[ID]], i32 [[GTID]], ptr [[TASK]])
 102   #pragma omp target teams distribute simd device(global + a) depend(in: global) depend(out: a, b, cn[4])
 103   for (int i = 0; i < 10; ++i) {
 104   }
 105
 106   // CHECK:       [[ADD:%.+]] = add nsw i32
 107   // CHECK:       store i32 [[ADD]], ptr [[DEVICE_CAP:%.+]],
 108
 109   // OMP45:       [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0
 110   // OMP45:       br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]]
 111   // OMP50:       br i1 {{.+}}, label %[[THEN:.+]], label %[[ELSE:.+]]
 112   // CHECK:       [[THEN]]:
 113   // OMP45-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP:%.+]], i32 0, i32 0
 114   // OMP45-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P:%.+]], i32 0, i32 0
 115   // OMP50-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP:%.+]], i32 0, i32 0
 116   // OMP50-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P:%.+]], i32 0, i32 0
 117   // CHECK-DAG:   store ptr [[BP0:%[^,]+]], ptr [[BPADDR0]]
 118   // CHECK-DAG:   store ptr [[BP0]], ptr [[PADDR0]]
 119
 120   // OMP45-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 1
 121   // OMP45-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 1
 122
 123   // OMP50-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 1
 124   // OMP50-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 1
 125   // CHECK-DAG:   store i[[SZ]] [[BP1:%[^,]+]], ptr [[BPADDR1]]
 126   // CHECK-DAG:   store i[[SZ]] [[BP1]], ptr [[PADDR1]]
 127
 128   // OMP50-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 2
 129   // OMP50-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 2
 130   // OMP50-DAG:   store i[[SZ]] [[BP2:%[^,]+]], ptr [[BPADDR2]]
 131   // OMP50-DAG:   store i[[SZ]] [[BP2]], ptr [[PADDR2]]
 132
 133
 134   // OMP45-DAG:   getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 0
 135   // OMP45-DAG:   getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 0
 136   // OMP50-DAG:   getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 0
 137   // OMP50-DAG:   getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 0
 138
 139   // OMP45:       [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2
 140   // OMP50-64:       [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 4
 141   // OMP50-32:    [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 3
 142   // CHECK:       [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],
 143   // CHECK:       store i32 [[DEV]], ptr [[GEP]],
 144   // CHECK:       [[DEV1:%.+]] = load i32, ptr [[DEVICE_CAP]],
 145   // CHECK:       [[DEV2:%.+]] = sext i32 [[DEV1]] to i64
 146
 147   // OMP45:       [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{104|60}}, i[[SZ]] {{16|12}}, ptr [[TASK_ENTRY1_:@.+]], i64 [[DEV2]])
 148   // OMP50:       [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{28|128|76}}, i[[SZ]] {{16|12|24}}, ptr [[TASK_ENTRY1_:@.+]], i64 [[DEV2]])
 149   // CHECK:       getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 0
 150   // CHECK:       getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 1
 151   // CHECK:       getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 2
 152   // CHECK:       call i32 @__kmpc_omp_task_with_deps(ptr [[ID]], i32 [[GTID]], ptr [[TASK]], i32 3, ptr %{{.+}}, i32 0, ptr null)
 153   // CHECK:       br label %[[EXIT:.+]]
 154
 155   // CHECK:       [[ELSE]]:
 156   // OMP45-NOT:   getelementptr inbounds [2 x ptr], ptr
 157   // OMP50-NOT:   getelementptr inbounds [3 x ptr], ptr
 158   // OMP45:       [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2
 159   // OMP50-64:       [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 4
 160   // OMP50-32:       [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 3
 161   // CHECK:       [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],
 162   // CHECK:       store i32 [[DEV]], ptr [[GEP]],
 163   // CHECK:       [[DEV1:%.+]] = load i32, ptr [[DEVICE_CAP]],
 164   // CHECK:       [[DEV2:%.+]] = sext i32 [[DEV1]] to i64
 165
 166   // OMP45:       [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, ptr [[TASK_ENTRY1__:@.+]], i64 [[DEV2]])
 167   // OMP50:       [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12|24}}, ptr [[TASK_ENTRY1__:@.+]], i64 [[DEV2]])
 168   // CHECK:       getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 0
 169   // CHECK:       getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 1
 170   // CHECK:       getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 2
 171   // CHECK:       call i32 @__kmpc_omp_task_with_deps(ptr [[ID]], i32 [[GTID]], ptr [[TASK]], i32 3, ptr %{{.+}}, i32 0, ptr null)
 172   // CHECK:       br label %[[EXIT:.+]]
 173   // CHECK:       [[EXIT]]:
 174
 175   #pragma omp target teams distribute simd device(global + a) nowait depend(inout: global, a, bn) if(a)
 176   for (int i = 0; i < *plocal; ++i) {
 177     static int local1;
 178     *plocal = global;
 179     local1 = global;
 180   }
 181
 182   // CHECK:       [[TASK:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, ptr [[TASK_ENTRY2:@.+]])
 183   // CHECK:       getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 0
 184   // CHECK:       call void @__kmpc_omp_taskwait_deps_51(ptr [[ID]], i32 [[GTID]], i32 1, ptr %{{.+}}, i32 0, ptr null, i32 0)
 185   // CHECK:       call void @__kmpc_omp_task_begin_if0(ptr [[ID]], i32 [[GTID]], ptr [[TASK]])
 186   // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], ptr [[TASK]])
 187   // CHECK:       call void @__kmpc_omp_task_complete_if0(ptr [[ID]], i32 [[GTID]], ptr [[TASK]])
 188   #pragma omp target teams distribute simd if(0) firstprivate(global) depend(out:global)
 189   for (int i = 0; i < global; ++i) {
 190     global += 1;
 191   }
 192
 193   return a;
 194 }
 195
 196 // Check that the offloading functions are emitted and that the arguments are
 197 // correct and loaded correctly for the target regions in foo().
 198
 199 // CHECK:       define internal void [[HVT0:@.+]]()
 200
 201 // CHECK:       define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, ptr noalias noundef %1)
 202 // CHECK:       store ptr null, ptr %
 203 // CHECK:       [[DEVICE_CAP:%.+]] = load ptr,
 204 // CHECK:       [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],
 205 // CHECK:       [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
 206 // CHECK:       [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE]], i32 0, i32 1, ptr @.{{.+}}.region_id, ptr %{{.+}})
 207 // CHECK-NEXT:  [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 208 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
 209 // CHECK:       [[FAIL]]
 210 // CHECK:       call void [[HVT0]]()
 211 // CHECK-NEXT:  br label %[[END]]
 212 // CHECK:       [[END]]
 213 // CHECK:       ret i32 0
 214
 215 // CHECK:       define internal void [[HVT1:@.+]](ptr noundef %{{.+}}, i[[SZ]] noundef %{{.+}})
 216
 217 // CHECK:       define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, ptr noalias noundef %1)
 218 // CHECK:       call void {{%.*}}(
 219 // OMP45:       [[DEVICE_CAP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2
 220 // OMP50-64:       [[DEVICE_CAP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 4
 221 // OMP50-32:       [[DEVICE_CAP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 3
 222 // CHECK:       [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],
 223 // CHECK:       [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
 224 // OMP45:       [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE]], i32 0, i32 1, ptr @.{{.+}}.region_id, ptr %{{.+}})
 225 // OMP50:       [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE]], i32 0, i32 1, ptr @.{{.+}}.region_id, ptr %{{.+}})
 226
 227 // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 228 // CHECK-NEXT:  br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
 229 // CHECK:       [[FAIL]]
 230 // CHECK:       [[BP0:%.+]] = load ptr, ptr %
 231 // CHECK:       [[BP1_I32:%.+]] = load i32, ptr @
 232 // CHECK-64:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 233 // CHECK-32:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 234 // CHECK:       [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]],
 235 // OMP45:       call void [[HVT1]](ptr [[BP0]], i[[SZ]] [[BP1]])
 236
 237 // OMP50:       [[BP2:%.+]] = load i[[SZ]], ptr
 238 // OMP50:       call void [[HVT1]](ptr [[BP0]], i[[SZ]] [[BP1]], i[[SZ]] [[BP2]])
 239 // CHECK-NEXT:  br label %[[END]]
 240 // CHECK:       [[END]]
 241 // CHECK:       ret i32 0
 242
 243 // CHECK:       define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, ptr noalias noundef %1)
 244 // CHECK:       call void {{%.*}}(
 245 // CHECK:       [[DEVICE_CAP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2
 246 // CHECK:       [[BP0:%.+]] = load ptr, ptr %
 247 // CHECK:       [[BP1_I32:%.+]] = load i32, ptr @
 248 // CHECK-64:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 249 // CHECK-32:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 250 // CHECK:       [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]],
 251 // OMP50:       [[BP2:%.+]] = load i[[SZ]], ptr
 252 // OMP45:       call void [[HVT1]](ptr [[BP0]], i[[SZ]] [[BP1]])
 253 // OMP50:       call void [[HVT1]](ptr [[BP0]], i[[SZ]] [[BP1]], i[[SZ]] [[BP2]])
 254
 255 // CHECK:       ret i32 0
 256
 257 // CHECK:       define internal void [[HVT2:@.+]](i[[SZ]] noundef %{{.+}})
 258 // Create stack storage and store argument in there.
 259 // CHECK:       [[AA_ADDR:%.+]] = alloca i[[SZ]], align
 260 // CHECK:       store i[[SZ]] %{{.+}}, ptr [[AA_ADDR]], align
 261 // CHECK-64:    load i32, ptr [[AA_ADDR]], align
 262 // CHECK-32:    load i32, ptr [[AA_ADDR]], align
 263
 264 // CHECK:       define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, ptr noalias noundef %1)
 265 // CHECK:       call void {{%.*}}(
 266 // CHECK:       [[BP1_I32:%.+]] = load i32, ptr %
 267 // CHECK-64:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 268 // CHECK-32:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 269 // CHECK:       [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]],
 270 // CHECK:       call void [[HVT2]](i[[SZ]] [[BP1]])
 271 // CHECK:       ret i32 0
 272
 273
 274 #endif