clang/test/OpenMP/target_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
   3 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
   4 // 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
   5 // 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
   6 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
   7 // 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
   8
   9 // 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
  10 // 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
  11 // 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
  12 // 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
  13 // 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
  14 // 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
  15 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
  16
  17 // Test target codegen - host bc file has to be created first.
  18 // 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
  19 // 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
  20 // 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
  21 // 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
  22 // 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
  23 // 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
  24 // 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
  25 // 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
  26
  27 // 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
  28 // 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
  29 // 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
  30 // 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
  31 // 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
  32 // 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
  33 // 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
  34 // 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
  35 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
  36
  37 // expected-no-diagnostics
  38 #ifndef HEADER
  39 #define HEADER
  40
  41 // CHECK-DAG: [[TT:%.+]] = type { i64, i8 }
  42 // CHECK-DAG: [[ENTTY:%.+]] = type { ptr, ptr, i[[SZ:32|64]], i32, i32 }
  43
  44 // TCHECK: [[ENTTY:%.+]] = type { ptr, ptr, i{{32|64}}, i32, i32 }
  45
  46 // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [3 x i64] [i64 0, i64 4, i64 {{16|12}}]
  47 // CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [3 x i64] [i64 544, i64 800, i64 3]
  48 // CHECK-DAG: @{{.*}} = weak constant i8 0
  49
  50 // TCHECK: @{{.+}} = weak constant [[ENTTY]]
  51 // TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]]
  52 // TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]]
  53
  54 template<typename tx, typename ty>
  55 struct TT{
  56   tx X;
  57   ty Y;
  58 };
  59
  60 #pragma omp declare mapper(id                     \
  61                            : TT <long long, char> \
  62                                s) map(s.X, s.Y)
  63 int global;
  64 extern int global;
  65
  66 // CHECK: define {{.*}}[[FOO:@.+]](
  67 int foo(int n) {
  68   int a = 0;
  69   short aa = 0;
  70   float b[10];
  71   float bn[n];
  72   double c[5][10];
  73   double cn[5][n];
  74   TT<long long, char> d;
  75   static long *plocal;
  76
  77 // CHECK:       [[ADD:%.+]] = add nsw i32
  78 // CHECK:       store i32 [[ADD]], ptr [[DEVICE_CAP:%.+]],
  79 // CHECK:       [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 0
  80 // CHECK:       [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],
  81 // CHECK:       store i32 [[DEV]], ptr [[GEP]],
  82 // CHECK:       [[TASK:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr @1, i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, ptr [[TASK_ENTRY0:@.+]])
  83 // CHECK:       getelementptr %struct.kmp_depend_info, ptr [[DEP_START:%.+]], i[[SZ]] 1
  84 // CHECK:       getelementptr %struct.kmp_depend_info, ptr [[DEP_START]], i[[SZ]] 2
  85 // CHECK:       getelementptr %struct.kmp_depend_info, ptr [[DEP_START]], i[[SZ]] 3
  86 // CHECK:       call void @__kmpc_omp_taskwait_deps_51(ptr @1, i32 [[GTID]], i32 4, ptr [[DEP_START]], i32 0, ptr null, i32 0)
  87 // CHECK:       call void @__kmpc_omp_task_begin_if0(ptr @1, i32 [[GTID]], ptr [[TASK]])
  88 // CHECK:       call i32 [[TASK_ENTRY0]](i32 [[GTID]], ptr [[TASK]])
  89 // CHECK:       call void @__kmpc_omp_task_complete_if0(ptr @1, i32 [[GTID]], ptr [[TASK]])
  90 #pragma omp target device(global + a) depend(in                   \
  91                                              : global) depend(out \
  92                                                               : a, b, cn[4])
  93   {
  94   }
  95
  96   // CHECK:       [[ADD:%.+]] = add nsw i32
  97   // CHECK:       store i32 [[ADD]], ptr [[DEVICE_CAP:%.+]],
  98
  99   // CHECK:       [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0
 100   // CHECK:       br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]]
 101   // CHECK:       [[THEN]]:
 102   // CHECK-DAG:   [[BPADDR0:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP:%.+]], i32 0, i32 0
 103   // CHECK-DAG:   [[PADDR0:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P:%.+]], i32 0, i32 0
 104   // CHECK-DAG:   [[MADDR0:%.+]] = getelementptr inbounds [3 x ptr], ptr [[M:%.+]], i[[SZ]] 0, i[[SZ]] 0
 105   // CHECK-DAG:   store ptr [[BP0:%[^,]+]], ptr [[BPADDR0]]
 106   // CHECK-DAG:   store ptr [[BP0]], ptr [[PADDR0]]
 107   // CHECK-DAG:   store ptr null, ptr [[MADDR0]],
 108
 109   // CHECK-DAG:   [[BPADDR1:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 1
 110   // CHECK-DAG:   [[PADDR1:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 1
 111   // CHECK-DAG:   [[MADDR1:%.+]] = getelementptr inbounds [3 x ptr], ptr [[M]], i[[SZ]] 0, i[[SZ]] 1
 112   // CHECK-DAG:   store i[[SZ]] [[BP1:%[^,]+]], ptr [[BPADDR1]]
 113   // CHECK-DAG:   store i[[SZ]] [[BP1]], ptr [[PADDR1]]
 114   // CHECK-DAG:   store ptr null, ptr [[MADDR1]],
 115
 116   // CHECK-DAG:   [[BPADDR2:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 2
 117   // CHECK-DAG:   [[PADDR2:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 2
 118   // CHECK-DAG:   [[MADDR2:%.+]] = getelementptr inbounds [3 x ptr], ptr [[M]], i[[SZ]] 0, i[[SZ]] 2
 119   // CHECK-DAG:   store ptr [[D_ADDR:%.+]], ptr [[BPADDR2]]
 120   // CHECK-DAG:   store ptr [[D_ADDR]], ptr [[PADDR2]]
 121   // CHECK-DAG:   store ptr [[MAPPER_ID:@.+]], ptr [[MADDR2]],
 122
 123   // CHECK-DAG:   [[BP_START:%.+]] = getelementptr inbounds [3 x ptr], ptr [[BP]], i32 0, i32 0
 124   // CHECK-DAG:   [[P_START:%.+]] = getelementptr inbounds [3 x ptr], ptr [[P]], i32 0, i32 0
 125   // CHECK:       [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2
 126   // CHECK:       [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],
 127   // CHECK:       store i32 [[DEV]], ptr [[GEP]],
 128   // CHECK:       [[DEV1:%.+]] = load i32, ptr [[DEVICE_CAP]],
 129   // CHECK:       [[DEV2:%.+]] = sext i32 [[DEV1]] to i64
 130
 131   // CHECK:       [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr @{{.*}}, i32 [[GTID]], i32 1, i[[SZ]] {{152|88}}, i[[SZ]] {{16|12}}, ptr [[TASK_ENTRY1_:@.+]], i64 [[DEV2]])
 132   // CHECK:       [[BASE:%.+]] = getelementptr inbounds nuw [[TASK_TY1_:%.+]], ptr [[TASK]], i32 0, i32 1
 133   // CHECK-64:    [[BP_BASE:%.+]] = getelementptr inbounds nuw [[PRIVS_TY:%.+]], ptr [[BASE]], i32 0, i32 1
 134   // CHECK-64:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[BP_BASE]], ptr align 8 [[BP_START]], i64 24, i1 false)
 135   // CHECK-64:    [[P_BASE:%.+]] = getelementptr inbounds nuw [[PRIVS_TY:%.+]], ptr [[BASE]], i32 0, i32 2
 136   // CHECK-64:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[P_BASE]], ptr align 8 [[P_START]], i64 24, i1 false)
 137   // CHECK-64:    [[SZ_BASE:%.+]] = getelementptr inbounds nuw [[PRIVS_TY]], ptr [[BASE]], i32 0, i32 3
 138   // CHECK-64:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[SZ_BASE]], ptr align 8 [[SIZET]], i64 24, i1 false)
 139   // CHECK-64:    [[M_BASE:%.+]] = getelementptr inbounds nuw [[PRIVS_TY]], ptr [[BASE]], i32 0, i32 4
 140   // CHECK-64:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[M_BASE]], ptr align 8 [[M]], i64 24, i1 false)
 141   // CHECK-32:    [[SZ_BASE:%.+]] = getelementptr inbounds nuw [[PRIVS_TY:%.+]], ptr [[BASE]], i32 0, i32 0
 142   // CHECK-32:    call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[SZ_BASE]], ptr align 4 [[SIZET]], i32 24, i1 false)
 143   // CHECK-32:    [[BP_BASE:%.+]] = getelementptr inbounds nuw [[PRIVS_TY:%.+]], ptr [[BASE]], i32 0, i32 3
 144   // CHECK-32:    call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[BP_BASE]], ptr align 4 [[BP_START]], i32 12, i1 false)
 145   // CHECK-32:    [[P_BASE:%.+]] = getelementptr inbounds nuw [[PRIVS_TY]], ptr [[BASE]], i32 0, i32 4
 146   // CHECK-32:    call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[P_BASE]], ptr align 4 [[P_START]], i32 12, i1 false)
 147   // CHECK-32:    [[M_BASE:%.+]] = getelementptr inbounds nuw [[PRIVS_TY]], ptr [[BASE]], i32 0, i32 5
 148   // CHECK-32:    call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[M_BASE]], ptr align 4 [[M]], i32 12, i1 false)
 149   // CHECK:       getelementptr %struct.kmp_depend_info, ptr [[DEP_START:%.+]], i[[SZ]] 1
 150   // CHECK:       getelementptr %struct.kmp_depend_info, ptr [[DEP_START]], i[[SZ]] 2
 151   // CHECK:       call i32 @__kmpc_omp_task_with_deps(ptr @1, i32 [[GTID]], ptr [[TASK]], i32 3, ptr [[DEP_START]], i32 0, ptr null)
 152   // CHECK:       br label %[[EXIT:.+]]
 153
 154   // CHECK:       [[ELSE]]:
 155   // CHECK-NOT:   getelementptr inbounds [2 x ptr], ptr
 156   // CHECK:       [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2
 157   // CHECK:       [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],
 158   // CHECK:       store i32 [[DEV]], ptr [[GEP]],
 159   // CHECK:       [[DEV1:%.+]] = load i32, ptr [[DEVICE_CAP]],
 160   // CHECK:       [[DEV2:%.+]] = sext i32 [[DEV1]] to i64
 161
 162   // CHECK:       [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr @1, i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, ptr [[TASK_ENTRY1__:@.+]], i64 [[DEV2]])
 163   // CHECK:       getelementptr %struct.kmp_depend_info, ptr [[DEP_START:%.+]], i[[SZ]] 1
 164   // CHECK:       getelementptr %struct.kmp_depend_info, ptr [[DEP_START]], i[[SZ]] 2
 165   // CHECK:       call i32 @__kmpc_omp_task_with_deps(ptr @1, i32 [[GTID]], ptr [[TASK]], i32 3, ptr [[DEP_START]], i32 0, ptr null)
 166   // CHECK:       br label %[[EXIT:.+]]
 167   // CHECK:       [[EXIT]]:
 168
 169 #pragma omp target device(global + a) nowait depend(inout                                          \
 170                                                     : global, a, bn) if (a) map(mapper(id), tofrom \
 171                                                                                 : d)
 172   {
 173     static int local1;
 174     *plocal = global;
 175     local1 = global;
 176   }
 177
 178 // CHECK:       [[TASK:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr @1, i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, ptr [[TASK_ENTRY2:@.+]])
 179 // CHECK:       call void @__kmpc_omp_taskwait_deps_51(ptr @1, i32 [[GTID]], i32 1, ptr %{{.+}}, i32 0, ptr null, i32 0)
 180 // CHECK:       call void @__kmpc_omp_task_begin_if0(ptr @1, i32 [[GTID]], ptr [[TASK]])
 181 // CHECK:       call i32 [[TASK_ENTRY2]](i32 [[GTID]], ptr [[TASK]])
 182 // CHECK:       call void @__kmpc_omp_task_complete_if0(ptr @1, i32 [[GTID]], ptr [[TASK]])
 183 #pragma omp target if (0) firstprivate(global) depend(out \
 184                                                       : global)
 185   {
 186     global += 1;
 187   }
 188
 189   return a;
 190 }
 191
 192 // Check that the offloading functions are emitted and that the arguments are
 193 // correct and loaded correctly for the target regions in foo().
 194
 195 // CHECK:       define internal void [[HVT0:@.+]]()
 196
 197 // CHECK:       define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, ptr noalias noundef %1)
 198 // CHECK:       store ptr null, ptr %
 199 // CHECK:       [[DEVICE_CAP:%.+]] = load ptr,
 200 // CHECK:       [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],
 201 // CHECK:       [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
 202 // CHECK:       [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE]], i32 -1, i32 0, ptr @.[[TGT_REGION:.+]].region_id, ptr %[[KERNEL_ARGS:.+]])
 203 // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 204 // CHECK:       br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
 205 // CHECK:       [[FAIL]]
 206 // CHECK:       call void [[HVT0]]()
 207 // CHECK:       br label %[[END]]
 208 // CHECK:       [[END]]
 209 // CHECK:       ret i32 0
 210
 211 // CHECK:       define internal void [[HVT1:@.+]](ptr noundef %{{.+}}, i[[SZ]] noundef %{{.+}})
 212
 213 // CHECK:       define internal void [[MAPPER_ID]](ptr noundef %{{.+}}, ptr noundef %{{.+}}, ptr noundef %{{.+}}, i64 noundef %{{.+}}, i64 noundef %{{.+}})
 214
 215 // CHECK:       define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, ptr noalias noundef %{{.+}})
 216 // CHECK:       call void %{{.+}}(ptr %{{.+}}, ptr %{{.+}}, ptr %{{.+}}, ptr [[BPTR_ADDR:%.+]], ptr [[PTR_ADDR:%.+]], ptr [[SZ_ADDR:%.+]], ptr [[M_ADDR:%.+]])
 217 // CHECK:       [[BPTR_REF:%.+]] = load ptr, ptr [[BPTR_ADDR]],
 218 // CHECK:       [[PTR_REF:%.+]] = load ptr, ptr [[PTR_ADDR]],
 219 // CHECK:       [[SZ_REF:%.+]] = load ptr, ptr [[SZ_ADDR]],
 220 // CHECK:       [[M_REF:%.+]] = load ptr, ptr [[M_ADDR]],
 221 // CHECK:       [[DEVICE_CAP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2
 222 // CHECK:       [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],
 223 // CHECK:       [[DEVICE:%.+]] = sext i32 [[DEV]] to i64
 224 // CHECK:       [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE]], i32 -1, i32 0, ptr @.[[TGT_REGION:.+]].region_id, ptr %[[KERNEL_ARGS:.+]])
 225
 226 // CHECK:       [[ERROR:%.+]] = icmp ne i32 [[RET]], 0
 227 // CHECK:       br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]
 228 // CHECK:       [[FAIL]]
 229 // CHECK:       [[BP0:%.+]] = load ptr, ptr %
 230 // CHECK:       [[BP1_I32:%.+]] = load i32, ptr @
 231 // CHECK-64:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 232 // CHECK-32:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 233 // CHECK:       [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]],
 234 // CHECK:       call void [[HVT1]](ptr [[BP0]], i[[SZ]] [[BP1]])
 235 // CHECK:       br label %[[END]]
 236 // CHECK:       [[END]]
 237 // CHECK:       ret i32 0
 238
 239 // CHECK:       define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, ptr noalias noundef %1)
 240 // CHECK:       call void {{%.*}}(
 241 // CHECK:       [[DEVICE_CAP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2
 242 // CHECK:       [[BP0:%.+]] = load ptr, ptr %
 243 // CHECK:       [[BP1_I32:%.+]] = load i32, ptr @
 244 // CHECK-64:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 245 // CHECK-32:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 246 // CHECK:       [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]],
 247 // CHECK:       call void [[HVT1]](ptr [[BP0]], i[[SZ]] [[BP1]])
 248 // CHECK:       ret i32 0
 249
 250 // CHECK:       define internal void [[HVT2:@.+]](i[[SZ]] noundef %{{.+}})
 251 // Create stack storage and store argument in there.
 252 // CHECK:       [[AA_ADDR:%.+]] = alloca i[[SZ]], align
 253 // CHECK:       store i[[SZ]] %{{.+}}, ptr [[AA_ADDR]], align
 254 // CHECK-64:    load i32, ptr [[AA_ADDR]], align
 255 // CHECK-32:    load i32, ptr [[AA_ADDR]], align
 256
 257 // CHECK:       define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, ptr noalias noundef %1)
 258 // CHECK:       call void {{%.*}}(
 259 // CHECK:       [[BP1_I32:%.+]] = load i32, ptr %
 260 // CHECK-64:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 261 // CHECK-32:    store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],
 262 // CHECK:       [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]],
 263 // CHECK:       call void [[HVT2]](i[[SZ]] [[BP1]])
 264 // CHECK:       ret i32 0
 265
 266 #endif