1 // Test host codegen. 2 // 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 -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 3 // 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 4 // 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 -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 5 // 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 -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 6 // 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 7 // 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 -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 8 9 // 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 -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s 10 // 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 11 // 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 -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s 12 // 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 -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s 13 // 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 14 // 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 -allow-deprecated-dag-overlap --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 -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 19 // 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-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 20 // 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-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s 21 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 22 // 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 23 // 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-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 24 // 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-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s 25 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 26 27 // 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 28 // 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-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s 29 // 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-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s 30 // 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-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s 31 // 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 32 // 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-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s 33 // 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-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s 34 // 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-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --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: %struct.ident_t = type { i32, i32, i32, i32, i8* } 42 // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" 43 // CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr global %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } 44 45 // CHECK-DAG: [[TT:%.+]] = type { i64, i8 } 46 // CHECK-DAG: [[S1:%.+]] = type { double } 47 // CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } 48 // CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } 49 // CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } 50 51 // TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } 52 53 // CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat 54 55 // We have 8 target regions, but only 6 that actually will generate offloading 56 // code and have mapped arguments, and only 4 have all-constant map sizes. 57 58 // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [3 x i64] [i64 2, i64 4, i64 4] 59 // CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 800, i64 800] 60 // CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [1 x i64] [i64 2] 61 // CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 800] 62 // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i64] [i64 4, i64 2] 63 // CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 800] 64 // CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 800, i64 547, i64 800, i64 547, i64 547, i64 800, i64 800, i64 547, i64 547] 65 // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i64] [i64 4, i64 2, i64 40] 66 // CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 547] 67 // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [4 x i64] [i64 4, i64 2, i64 1, i64 40] 68 // CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 800, i64 800, i64 800, i64 547] 69 // CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 800, i64 800, i64 547] 70 // CHECK-DAG: @{{.*}} = weak constant i8 0 71 // CHECK-DAG: @{{.*}} = weak constant i8 0 72 // CHECK-DAG: @{{.*}} = weak constant i8 0 73 // CHECK-DAG: @{{.*}} = weak constant i8 0 74 // CHECK-DAG: @{{.*}} = weak constant i8 0 75 // CHECK-DAG: @{{.*}} = weak constant i8 0 76 // CHECK-DAG: @{{.*}} = weak constant i8 0 77 // CHECK-DAG: @{{.*}} = weak constant i8 0 78 // CHECK-DAG: @{{.*}} = weak constant i8 0 79 80 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 81 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 82 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 83 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 84 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 85 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 86 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 87 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 88 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 89 // TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] 90 91 // Check if offloading descriptor is created. 92 // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] 93 // CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] 94 // CHECK: [[DEVBEGIN:@.+]] = extern_weak constant i8 95 // CHECK: [[DEVEND:@.+]] = extern_weak constant i8 96 // CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) 97 // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) 98 99 // Check target registration is registered as a Ctor. 100 // CHECK: appending global [2 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @.omp_offloading.requires_reg, i8* null }, { i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] 101 102 103 template<typename tx, typename ty> 104 struct TT{ 105 tx X; 106 ty Y; 107 }; 108 109 int global; 110 111 // CHECK: define {{.*}}[[FOO:@.+]]( 112 int foo(int n) { 113 int a = 0; 114 short aa = 0; 115 float b[10]; 116 float bn[n]; 117 double c[5][10]; 118 double cn[5][n]; 119 TT<long long, char> d; 120 121 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams_nowait(i64 -1, i8* @{{[^,]+}}, i32 3, i8** [[BP:%[^,]+]], i8** [[P:%[^,]+]], i64* getelementptr inbounds ([3 x i64], [3 x i64]* [[SIZET]], i32 0, i32 0), i64* getelementptr inbounds ([3 x i64], [3 x i64]* [[MAPT]], i32 0, i32 0), i32 {{[^,]+}}, i32 {{[^)]+}}) 122 // CHECK-DAG: [[BP]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR:%[^,]+]], i32 0, i32 0 123 // CHECK-DAG: [[P]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR:%[^,]+]], i32 0, i32 0 124 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR]], i32 0, i32 [[IDX0:[0-9]+]] 125 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR]], i32 0, i32 [[IDX0]] 126 // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]* 127 // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]* 128 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR0]] 129 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR0]] 130 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR]], i32 0, i32 [[IDX1:[0-9]+]] 131 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR]], i32 0, i32 [[IDX1]] 132 // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* 133 // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* 134 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR1]] 135 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR1]] 136 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR]], i32 0, i32 [[IDX1:[0-9]+]] 137 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR]], i32 0, i32 [[IDX1]] 138 // CHECK-DAG: [[CBPADDR2:%.+]] = bitcast i8** [[BPADDR2]] to i[[SZ]]* 139 // CHECK-DAG: [[CPADDR2:%.+]] = bitcast i8** [[PADDR2]] to i[[SZ]]* 140 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR2]] 141 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR2]] 142 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 143 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 144 // CHECK: [[FAIL]] 145 // CHECK: call void [[HVT0:@.+]](i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^)]+}}) 146 // CHECK-NEXT: br label %[[END]] 147 // CHECK: [[END]] 148 #pragma omp target teams num_teams(a) thread_limit(a) firstprivate(aa) nowait 149 { 150 } 151 152 // CHECK: call void [[HVT1:@.+]](i[[SZ]] {{[^,]+}}) 153 #pragma omp target teams if(target: 0) 154 { 155 a += 1; 156 } 157 158 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, i8** [[BP:%[^,]+]], i8** [[P:%[^,]+]], i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[SIZET2]], i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[MAPT2]], i32 0, i32 0), i32 0, i32 0) 159 // CHECK-DAG: [[BP]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BPR:%[^,]+]], i32 0, i32 0 160 // CHECK-DAG: [[P]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PR:%[^,]+]], i32 0, i32 0 161 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BPR]], i32 0, i32 [[IDX0:[0-9]+]] 162 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PR]], i32 0, i32 [[IDX0]] 163 // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]* 164 // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]* 165 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR0]] 166 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR0]] 167 168 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 169 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 170 // CHECK: [[FAIL]] 171 // CHECK: call void [[HVT2:@.+]](i[[SZ]] {{[^,]+}}) 172 // CHECK-NEXT: br label %[[END]] 173 // CHECK: [[END]] 174 #pragma omp target teams if(target: 1) 175 { 176 aa += 1; 177 } 178 179 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 10 180 // CHECK: br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] 181 // CHECK: [[IFTHEN]] 182 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[SIZET3]], i32 0, i32 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT3]], i32 0, i32 0), i32 0, i32 0) 183 // CHECK-DAG: [[BPR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 184 // CHECK-DAG: [[PR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%[^,]+]], i32 0, i32 0 185 186 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 187 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 188 // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]* 189 // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]* 190 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR0]] 191 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR0]] 192 193 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 194 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 195 // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* 196 // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* 197 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR1]] 198 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR1]] 199 // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 200 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] 201 // CHECK: [[FAIL]] 202 // CHECK: call void [[HVT3:@.+]]({{[^,]+}}, {{[^,]+}}) 203 // CHECK-NEXT: br label %[[END]] 204 // CHECK: [[END]] 205 // CHECK-NEXT: br label %[[IFEND:.+]] 206 // CHECK: [[IFELSE]] 207 // CHECK: call void [[HVT3]]({{[^,]+}}, {{[^,]+}}) 208 // CHECK-NEXT: br label %[[IFEND]] 209 // CHECK: [[IFEND]] 210 #pragma omp target teams if(target: n>10) 211 { 212 a += 1; 213 aa += 1; 214 } 215 216 // We capture 3 VLA sizes in this target region 217 // CHECK-64: [[A_VAL:%.+]] = load i32, i32* %{{.+}}, 218 // CHECK-64: [[A_ADDR:%.+]] = bitcast i[[SZ]]* [[A_CADDR:%.+]] to i32* 219 // CHECK-64: store i32 [[A_VAL]], i32* [[A_ADDR]], 220 // CHECK-64: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], 221 222 // CHECK-32: [[A_VAL:%.+]] = load i32, i32* %{{.+}}, 223 // CHECK-32: store i32 [[A_VAL]], i32* [[A_CADDR:%.+]], 224 // CHECK-32: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], 225 226 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 227 // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] 228 // CHECK: [[TRY]] 229 // CHECK-64: [[BNSIZE:%.+]] = mul nuw i64 [[VLA0:%.+]], 4 230 // CHECK-32: [[BNSZSIZE:%.+]] = mul nuw i32 [[VLA0:%.+]], 4 231 // CHECK-32: [[BNSIZE:%.+]] = sext i32 [[BNSZSIZE]] to i64 232 // CHECK: [[CNELEMSIZE2:%.+]] = mul nuw i[[SZ]] 5, [[VLA1:%.+]] 233 // CHECK-64: [[CNSIZE:%.+]] = mul nuw i64 [[CNELEMSIZE2]], 8 234 // CHECK-32: [[CNSZSIZE:%.+]] = mul nuw i32 [[CNELEMSIZE2]], 8 235 // CHECK-32: [[CNSIZE:%.+]] = sext i32 [[CNSZSIZE]] to i64 236 237 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 9, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i64* [[SR:%[^,]+]], i64* getelementptr inbounds ([9 x i64], [9 x i64]* [[MAPT4]], i32 0, i32 0), i32 0, i32 0) 238 // CHECK-DAG: [[BPR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 239 // CHECK-DAG: [[PR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P:%[^,]+]], i32 0, i32 0 240 // CHECK-DAG: [[SR]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S:%[^,]+]], i32 0, i32 0 241 242 // CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S]], i32 0, i32 [[IDX0:[0-9]+]] 243 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX0]] 244 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX0]] 245 // CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S]], i32 0, i32 [[IDX1:[0-9]+]] 246 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX1]] 247 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX1]] 248 // CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S]], i32 0, i32 [[IDX2:[0-9]+]] 249 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX2]] 250 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX2]] 251 // CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S]], i32 0, i32 [[IDX3:[0-9]+]] 252 // CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX3]] 253 // CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX3]] 254 // CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S]], i32 0, i32 [[IDX4:[0-9]+]] 255 // CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX4]] 256 // CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX4]] 257 // CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S]], i32 0, i32 [[IDX5:[0-9]+]] 258 // CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX5]] 259 // CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX5]] 260 // CHECK-DAG: [[SADDR6:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S]], i32 0, i32 [[IDX6:[0-9]+]] 261 // CHECK-DAG: [[BPADDR6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX6]] 262 // CHECK-DAG: [[PADDR6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX6]] 263 // CHECK-DAG: [[SADDR7:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S]], i32 0, i32 [[IDX7:[0-9]+]] 264 // CHECK-DAG: [[BPADDR7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX7]] 265 // CHECK-DAG: [[PADDR7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX7]] 266 // CHECK-DAG: [[SADDR8:%.+]] = getelementptr inbounds [9 x i64], [9 x i64]* [[S]], i32 0, i32 [[IDX8:[0-9]+]] 267 // CHECK-DAG: [[BPADDR8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX8]] 268 // CHECK-DAG: [[PADDR8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX8]] 269 270 // The names below are not necessarily consistent with the names used for the 271 // addresses above as some are repeated. 272 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CBPADDR0:%.+]], 273 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CPADDR0:%.+]], 274 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 275 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 276 // CHECK-DAG: store i64 {{4|8}}, i64* {{%[^,]+}} 277 278 // CHECK-DAG: store i[[SZ]] [[VLA1]], i[[SZ]]* [[CBPADDR1:%.+]], 279 // CHECK-DAG: store i[[SZ]] [[VLA1]], i[[SZ]]* [[CPADDR1:%.+]], 280 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 281 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 282 // CHECK-DAG: store i64 {{4|8}}, i64* {{%[^,]+}} 283 284 // CHECK-DAG: store i[[SZ]] 5, i[[SZ]]* [[CBPADDR2:%.+]], 285 // CHECK-DAG: store i[[SZ]] 5, i[[SZ]]* [[CPADDR2:%.+]], 286 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 287 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 288 // CHECK-DAG: store i64 {{4|8}}, i64* {{%[^,]+}} 289 290 // CHECK-DAG: store i[[SZ]] [[A_CVAL]], i[[SZ]]* [[CBPADDR3:%.+]], 291 // CHECK-DAG: store i[[SZ]] [[A_CVAL]], i[[SZ]]* [[CPADDR3:%.+]], 292 // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 293 // CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 294 // CHECK-DAG: store i64 4, i64* {{%[^,]+}} 295 296 // CHECK-DAG: store [10 x float]* %{{.+}}, [10 x float]** [[CBPADDR4:%.+]], 297 // CHECK-DAG: store [10 x float]* %{{.+}}, [10 x float]** [[CPADDR4:%.+]], 298 // CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [10 x float]** 299 // CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to [10 x float]** 300 // CHECK-DAG: store i64 40, i64* {{%[^,]+}} 301 302 // CHECK-DAG: store float* %{{.+}}, float** [[CBPADDR5:%.+]], 303 // CHECK-DAG: store float* %{{.+}}, float** [[CPADDR5:%.+]], 304 // CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to float** 305 // CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to float** 306 // CHECK-DAG: store i64 [[BNSIZE]], i64* {{%[^,]+}} 307 308 // CHECK-DAG: store [5 x [10 x double]]* %{{.+}}, [5 x [10 x double]]** [[CBPADDR6:%.+]], 309 // CHECK-DAG: store [5 x [10 x double]]* %{{.+}}, [5 x [10 x double]]** [[CPADDR6:%.+]], 310 // CHECK-DAG: [[CBPADDR6]] = bitcast i8** {{%[^,]+}} to [5 x [10 x double]]** 311 // CHECK-DAG: [[CPADDR6]] = bitcast i8** {{%[^,]+}} to [5 x [10 x double]]** 312 // CHECK-DAG: store i64 400, i64* {{%[^,]+}} 313 314 // CHECK-DAG: store double* %{{.+}}, double** [[CBPADDR7:%.+]], 315 // CHECK-DAG: store double* %{{.+}}, double** [[CPADDR7:%.+]], 316 // CHECK-DAG: [[CBPADDR7]] = bitcast i8** {{%[^,]+}} to double** 317 // CHECK-DAG: [[CPADDR7]] = bitcast i8** {{%[^,]+}} to double** 318 // CHECK-DAG: store i64 [[CNSIZE]], i64* {{%[^,]+}} 319 320 // CHECK-DAG: store [[TT]]* %{{.+}}, [[TT]]** [[CBPADDR8:%.+]], 321 // CHECK-DAG: store [[TT]]* %{{.+}}, [[TT]]** [[CPADDR8:%.+]], 322 // CHECK-DAG: [[CBPADDR8]] = bitcast i8** {{%[^,]+}} to [[TT]]** 323 // CHECK-DAG: [[CPADDR8]] = bitcast i8** {{%[^,]+}} to [[TT]]** 324 // CHECK-DAG: store i64 {{12|16}}, i64* {{%[^,]+}} 325 326 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 327 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 328 329 // CHECK: [[FAIL]] 330 // CHECK: call void [[HVT4:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 331 // CHECK-NEXT: br label %[[END]] 332 // CHECK: [[END]] 333 #pragma omp target teams if(target: n>20) 334 { 335 a += 1; 336 b[2] += 1.0; 337 bn[3] += 1.0; 338 c[1][2] += 1.0; 339 cn[1][3] += 1.0; 340 d.X += 1; 341 d.Y += 1; 342 } 343 344 return a; 345 } 346 347 // Check that the offloading functions are emitted and that the arguments are 348 // correct and loaded correctly for the target regions in foo(). 349 350 // CHECK: define internal void [[HVT0]](i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^)]+}}) 351 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] {{[^)]+}}) 352 // 353 // 354 // CHECK: define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] {{[^)]+}}) 355 // CHECK: alloca i[[SZ]], 356 // CHECK: bitcast i[[SZ]]* {{.+}} to i16* 357 // CHECK: ret void 358 // CHECK-NEXT: } 359 360 361 // CHECK: define internal void [[HVT1]](i[[SZ]] %{{.+}}) 362 // Create stack storage and store argument in there. 363 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 364 // CHECK: [[AA_CASTED:%.+]] = alloca i[[SZ]], align 365 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 366 // CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* 367 // CHECK-64: [[AA:%.+]] = load i32, i32* [[AA_CADDR]], align 368 // CHECK-32: [[AA:%.+]] = load i32, i32* [[AA_ADDR]], align 369 // CHECK-64: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i32* 370 // CHECK-64: store i32 [[AA]], i32* [[AA_C]], align 371 // CHECK-32: store i32 [[AA]], i32* [[AA_CASTED]], align 372 // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align 373 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) 374 // 375 // 376 // CHECK: define internal {{.*}}void [[OMP_OUTLINED1]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) 377 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 378 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 379 // CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* 380 // CHECK-64: [[AA:%.+]] = load i32, i32* [[AA_CADDR]], align 381 // CHECK-32: [[AA:%.+]] = load i32, i32* [[AA_ADDR]], align 382 // CHECK: ret void 383 // CHECK-NEXT: } 384 385 // CHECK: define internal void [[HVT2]](i[[SZ]] %{{.+}}) 386 // Create stack storage and store argument in there. 387 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 388 // CHECK: [[AA_CASTED:%.+]] = alloca i[[SZ]], align 389 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 390 // CHECK: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 391 // CHECK: [[AA:%.+]] = load i16, i16* [[AA_CADDR]], align 392 // CHECK: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* 393 // CHECK: store i16 [[AA]], i16* [[AA_C]], align 394 // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align 395 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) 396 // 397 // 398 // CHECK: define internal {{.*}}void [[OMP_OUTLINED2]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) 399 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 400 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 401 // CHECK: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 402 // CHECK: [[AA:%.+]] = load i16, i16* [[AA_CADDR]], align 403 // CHECK: ret void 404 // CHECK-NEXT: } 405 406 // CHECK: define internal void [[HVT3]] 407 // Create stack storage and store argument in there. 408 // CHECK: [[A_ADDR:%.+]] = alloca i[[SZ]], align 409 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 410 // CHECK: [[A_CASTED:%.+]] = alloca i[[SZ]], align 411 // CHECK: [[AA_CASTED:%.+]] = alloca i[[SZ]], align 412 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[A_ADDR]], align 413 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 414 // CHECK-64-DAG:[[A_CADDR:%.+]] = bitcast i[[SZ]]* [[A_ADDR]] to i32* 415 // CHECK-DAG: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 416 // CHECK-64-DAG:[[A:%.+]] = load i32, i32* [[A_CADDR]], align 417 // CHECK-32-DAG:[[A:%.+]] = load i32, i32* [[A_ADDR]], align 418 // CHECK-64-DAG:[[A_C:%.+]] = bitcast i[[SZ]]* [[A_CASTED]] to i32* 419 // CHECK-64-DAG:store i32 [[A]], i32* [[A_C]], align 420 // CHECK-32-DAG:store i32 [[A]], i32* [[A_CASTED]], align 421 // CHECK-DAG: [[AA:%.+]] = load i16, i16* [[AA_CADDR]], align 422 // CHECK-DAG: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* 423 // CHECK-DAG: store i16 [[AA]], i16* [[AA_C]], align 424 // CHECK-DAG: [[PARAM1:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CASTED]], align 425 // CHECK-DAG: [[PARAM2:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align 426 // CHECK-DAG: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) 427 // 428 // 429 // CHECK: define internal {{.*}}void [[OMP_OUTLINED3]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}) 430 // CHECK: [[A_ADDR:%.+]] = alloca i[[SZ]], align 431 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 432 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[A_ADDR]], align 433 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 434 // CHECK-64-DAG:[[A_CADDR:%.+]] = bitcast i[[SZ]]* [[A_ADDR]] to i32* 435 // CHECK-DAG: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 436 // CHECK: ret void 437 // CHECK-NEXT: } 438 439 // CHECK: define internal void [[HVT4]] 440 // Create local storage for each capture. 441 // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] 442 // CHECK: [[LOCAL_B:%.+]] = alloca [10 x float]* 443 // CHECK: [[LOCAL_VLA1:%.+]] = alloca i[[SZ]] 444 // CHECK: [[LOCAL_BN:%.+]] = alloca float* 445 // CHECK: [[LOCAL_C:%.+]] = alloca [5 x [10 x double]]* 446 // CHECK: [[LOCAL_VLA2:%.+]] = alloca i[[SZ]] 447 // CHECK: [[LOCAL_VLA3:%.+]] = alloca i[[SZ]] 448 // CHECK: [[LOCAL_CN:%.+]] = alloca double* 449 // CHECK: [[LOCAL_D:%.+]] = alloca [[TT]]* 450 // CHECK: [[LOCAL_A_CASTED:%.+]] = alloca i[[SZ]] 451 // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] 452 // CHECK-DAG: store [10 x float]* [[ARG_B:%.+]], [10 x float]** [[LOCAL_B]] 453 // CHECK-DAG: store i[[SZ]] [[ARG_VLA1:%.+]], i[[SZ]]* [[LOCAL_VLA1]] 454 // CHECK-DAG: store float* [[ARG_BN:%.+]], float** [[LOCAL_BN]] 455 // CHECK-DAG: store [5 x [10 x double]]* [[ARG_C:%.+]], [5 x [10 x double]]** [[LOCAL_C]] 456 // CHECK-DAG: store i[[SZ]] [[ARG_VLA2:%.+]], i[[SZ]]* [[LOCAL_VLA2]] 457 // CHECK-DAG: store i[[SZ]] [[ARG_VLA3:%.+]], i[[SZ]]* [[LOCAL_VLA3]] 458 // CHECK-DAG: store double* [[ARG_CN:%.+]], double** [[LOCAL_CN]] 459 // CHECK-DAG: store [[TT]]* [[ARG_D:%.+]], [[TT]]** [[LOCAL_D]] 460 461 // CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* 462 // CHECK-DAG: [[REF_B:%.+]] = load [10 x float]*, [10 x float]** [[LOCAL_B]], 463 // CHECK-DAG: [[VAL_VLA1:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA1]], 464 // CHECK-DAG: [[REF_BN:%.+]] = load float*, float** [[LOCAL_BN]], 465 // CHECK-DAG: [[REF_C:%.+]] = load [5 x [10 x double]]*, [5 x [10 x double]]** [[LOCAL_C]], 466 // CHECK-DAG: [[VAL_VLA2:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA2]], 467 // CHECK-DAG: [[VAL_VLA3:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA3]], 468 // CHECK-DAG: [[REF_CN:%.+]] = load double*, double** [[LOCAL_CN]], 469 // CHECK-DAG: [[REF_D:%.+]] = load [[TT]]*, [[TT]]** [[LOCAL_D]], 470 471 // CHECK-64-DAG:[[CONV_A:%.+]] = load i32, i32* [[CONV_AP]] 472 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_A_CASTED]] to i32* 473 // CHECK-64-DAG:store i32 [[CONV_A]], i32* [[CONV]], align 474 // CHECK-32-DAG:[[LOCAL_AV:%.+]] = load i32, i32* [[LOCAL_A]] 475 // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align 476 // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], 477 478 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 9, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*)* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]]) 479 // 480 // 481 // CHECK: define internal {{.*}}void [[OMP_OUTLINED4]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, [10 x float]* {{.+}}, i[[SZ]] %{{.+}}, float* {{.+}}, [5 x [10 x double]]* {{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, double* {{.+}}, [[TT]]* {{.+}}) 482 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 483 484 template<typename tx> 485 tx ftemplate(int n) { 486 tx a = 0; 487 short aa = 0; 488 tx b[10]; 489 490 #pragma omp target teams if(target: n>40) 491 { 492 a += 1; 493 aa += 1; 494 b[2] += 1; 495 } 496 497 return a; 498 } 499 500 static 501 int fstatic(int n) { 502 int a = 0; 503 short aa = 0; 504 char aaa = 0; 505 int b[10]; 506 507 #pragma omp target teams if(target: n>50) 508 { 509 a += 1; 510 aa += 1; 511 aaa += 1; 512 b[2] += 1; 513 } 514 515 return a; 516 } 517 518 struct S1 { 519 double a; 520 521 int r1(int n){ 522 int b = n+1; 523 short int c[2][n]; 524 525 #pragma omp target teams if(target: n>60) 526 { 527 this->a = (double)b + 1.5; 528 c[1][1] = ++a; 529 } 530 531 return c[1][1] + (int)b; 532 } 533 }; 534 535 // CHECK: define {{.*}}@{{.*}}bar{{.*}} 536 int bar(int n){ 537 int a = 0; 538 539 // CHECK: call {{.*}}i32 [[FOO]](i32 {{.*}}) 540 a += foo(n); 541 542 S1 S; 543 // CHECK: call {{.*}}i32 [[FS1:@.+]]([[S1]]* {{.*}}, i32 {{.*}}) 544 a += S.r1(n); 545 546 // CHECK: call {{.*}}i32 [[FSTATIC:@.+]](i32 {{.*}}) 547 a += fstatic(n); 548 549 // CHECK: call {{.*}}i32 [[FTEMPLATE:@.+]](i32 {{.*}}) 550 a += ftemplate<int>(n); 551 552 return a; 553 } 554 555 // 556 // CHECK: define {{.*}}[[FS1]] 557 // 558 // CHECK: i8* @llvm.stacksave() 559 // CHECK-64: [[B_ADDR:%.+]] = bitcast i[[SZ]]* [[B_CADDR:%.+]] to i32* 560 // CHECK-64: store i32 %{{.+}}, i32* [[B_ADDR]], 561 // CHECK-64: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]], 562 563 // CHECK-32: store i32 %{{.+}}, i32* %__vla_expr 564 // CHECK-32: store i32 %{{.+}}, i32* [[B_ADDR:%.+]], 565 // CHECK-32: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]], 566 567 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 568 // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] 569 // CHECK: [[TRY]] 570 // We capture 2 VLA sizes in this target region 571 // CHECK: [[CELEMSIZE2:%.+]] = mul nuw i[[SZ]] 2, [[VLA0:%.+]] 572 // CHECK-64: [[CSIZE:%.+]] = mul nuw i64 [[CELEMSIZE2]], 2 573 // CHECK-32: [[CSZSIZE:%.+]] = mul nuw i32 [[CELEMSIZE2]], 2 574 // CHECK-32: [[CSIZE:%.+]] = sext i32 [[CSZSIZE]] to i64 575 576 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 6, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i64* [[SR:%[^,]+]], i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT7]], i32 0, i32 0), i32 0, i32 0) 577 // CHECK-DAG: [[BPR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP:%.+]], i32 0, i32 0 578 // CHECK-DAG: [[PR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P:%.+]], i32 0, i32 0 579 // CHECK-DAG: [[SR]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S:%.+]], i32 0, i32 0 580 // CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S]], i32 [[IDX0:[0-9]+]] 581 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX0]] 582 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX0]] 583 // CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S]], i32 [[IDX1:[0-9]+]] 584 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX1]] 585 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX1]] 586 // CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S]], i32 [[IDX2:[0-9]+]] 587 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX2]] 588 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX2]] 589 // CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S]], i32 [[IDX3:[0-9]+]] 590 // CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] 591 // CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] 592 // CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S]], i32 [[IDX4:[0-9]+]] 593 // CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX4]] 594 // CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX4]] 595 596 // The names below are not necessarily consistent with the names used for the 597 // addresses above as some are repeated. 598 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CBPADDR0:%.+]], 599 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CPADDR0:%.+]], 600 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 601 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 602 // CHECK-DAG: store i64 {{4|8}}, i64* {{%[^,]+}} 603 604 // CHECK-DAG: store i[[SZ]] 2, i[[SZ]]* [[CBPADDR1:%.+]], 605 // CHECK-DAG: store i[[SZ]] 2, i[[SZ]]* [[CPADDR1:%.+]], 606 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 607 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 608 // CHECK-DAG: store i64 {{4|8}}, i64* {{%[^,]+}} 609 610 // CHECK-DAG: store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CBPADDR2:%.+]], 611 // CHECK-DAG: store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CPADDR2:%.+]], 612 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 613 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 614 // CHECK-DAG: store i64 4, i64* {{%[^,]+}} 615 616 // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR3:%.+]], 617 // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR3:%.+]], 618 // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** 619 // CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** 620 // CHECK-DAG: store i64 8, i64* {{%[^,]+}} 621 622 // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR4:%.+]], 623 // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR4:%.+]], 624 // CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** 625 // CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** 626 // CHECK-DAG: store i64 8, i64* {{%[^,]+}} 627 628 // CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR5:%.+]], 629 // CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR5:%.+]], 630 // CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to i16** 631 // CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to i16** 632 // CHECK-DAG: store i64 [[CSIZE]], i64* {{%[^,]+}} 633 634 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 635 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 636 637 // CHECK: [[FAIL]] 638 // CHECK: call void [[HVT7:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 639 // CHECK-NEXT: br label %[[END]] 640 // CHECK: [[END]] 641 642 // 643 // CHECK: define {{.*}}[[FSTATIC]] 644 // 645 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 50 646 // CHECK: br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] 647 // CHECK: [[IFTHEN]] 648 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 4, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i64* getelementptr inbounds ([4 x i64], [4 x i64]* [[SIZET6]], i32 0, i32 0), i64* getelementptr inbounds ([4 x i64], [4 x i64]* [[MAPT6]], i32 0, i32 0), i32 0, i32 0) 649 // CHECK-DAG: [[BPR]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP:%.+]], i32 0, i32 0 650 // CHECK-DAG: [[PR]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P:%.+]], i32 0, i32 0 651 652 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 0 653 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 0 654 // CHECK-DAG: store i[[SZ]] [[VAL0:%.+]], i[[SZ]]* [[CBPADDR0:%.+]], 655 // CHECK-DAG: store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0:%.+]], 656 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 657 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 658 659 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 1 660 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 1 661 // CHECK-DAG: store i[[SZ]] [[VAL1:%.+]], i[[SZ]]* [[CBPADDR1:%.+]], 662 // CHECK-DAG: store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1:%.+]], 663 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 664 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 665 666 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 2 667 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 2 668 // CHECK-DAG: store i[[SZ]] [[VAL2:%.+]], i[[SZ]]* [[CBPADDR2:%.+]], 669 // CHECK-DAG: store i[[SZ]] [[VAL2]], i[[SZ]]* [[CPADDR2:%.+]], 670 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 671 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 672 673 // CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 3 674 // CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 3 675 // CHECK-DAG: store [10 x i32]* %{{.+}}, [10 x i32]** [[CBPADDR3:%.+]], 676 // CHECK-DAG: store [10 x i32]* %{{.+}}, [10 x i32]** [[CPADDR3:%.+]], 677 // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to [10 x i32]** 678 // CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to [10 x i32]** 679 680 // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 681 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] 682 // CHECK: [[FAIL]] 683 // CHECK: call void [[HVT6:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 684 // CHECK-NEXT: br label %[[END]] 685 // CHECK: [[END]] 686 // CHECK-NEXT: br label %[[IFEND:.+]] 687 // CHECK: [[IFELSE]] 688 // CHECK: call void [[HVT6]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 689 // CHECK-NEXT: br label %[[IFEND]] 690 // CHECK: [[IFEND]] 691 692 // 693 // CHECK: define {{.*}}[[FTEMPLATE]] 694 // 695 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 40 696 // CHECK: br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] 697 // CHECK: [[IFTHEN]] 698 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i64* getelementptr inbounds ([3 x i64], [3 x i64]* [[SIZET5]], i32 0, i32 0), i64* getelementptr inbounds ([3 x i64], [3 x i64]* [[MAPT5]], i32 0, i32 0), i32 0, i32 0) 699 // CHECK-DAG: [[BPR]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP:%.+]], i32 0, i32 0 700 // CHECK-DAG: [[PR]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P:%.+]], i32 0, i32 0 701 702 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 0 703 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 0 704 // CHECK-DAG: store i[[SZ]] [[VAL0:%.+]], i[[SZ]]* [[CBPADDR0:%.+]], 705 // CHECK-DAG: store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0:%.+]], 706 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 707 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 708 709 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 1 710 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 1 711 // CHECK-DAG: store i[[SZ]] [[VAL1:%.+]], i[[SZ]]* [[CBPADDR1:%.+]], 712 // CHECK-DAG: store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1:%.+]], 713 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 714 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 715 716 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 2 717 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 2 718 // CHECK-DAG: store [10 x i32]* [[VAL2:%.+]], [10 x i32]** [[CBPADDR2:%.+]], 719 // CHECK-DAG: store [10 x i32]* [[VAL2]], [10 x i32]** [[CPADDR2:%.+]], 720 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to [10 x i32]** 721 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to [10 x i32]** 722 723 // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 724 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] 725 // CHECK: [[FAIL]] 726 // CHECK: call void [[HVT5:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 727 // CHECK-NEXT: br label %[[END]] 728 // CHECK: [[END]] 729 // CHECK-NEXT: br label %[[IFEND:.+]] 730 // CHECK: [[IFELSE]] 731 // CHECK: call void [[HVT5]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 732 // CHECK-NEXT: br label %[[IFEND]] 733 // CHECK: [[IFEND]] 734 735 736 737 // Check that the offloading functions are emitted and that the arguments are 738 // correct and loaded correctly for the target regions of the callees of bar(). 739 740 // CHECK: define internal void [[HVT7]] 741 // Create local storage for each capture. 742 // CHECK: [[LOCAL_THIS:%.+]] = alloca [[S1]]* 743 // CHECK: [[LOCAL_B:%.+]] = alloca i[[SZ]] 744 // CHECK: [[LOCAL_VLA1:%.+]] = alloca i[[SZ]] 745 // CHECK: [[LOCAL_VLA2:%.+]] = alloca i[[SZ]] 746 // CHECK: [[LOCAL_C:%.+]] = alloca i16* 747 // CHECK: [[LOCAL_B_CASTED:%.+]] = alloca i[[SZ]] 748 // CHECK-DAG: store [[S1]]* [[ARG_THIS:%.+]], [[S1]]** [[LOCAL_THIS]] 749 // CHECK-DAG: store i[[SZ]] [[ARG_B:%.+]], i[[SZ]]* [[LOCAL_B]] 750 // CHECK-DAG: store i[[SZ]] [[ARG_VLA1:%.+]], i[[SZ]]* [[LOCAL_VLA1]] 751 // CHECK-DAG: store i[[SZ]] [[ARG_VLA2:%.+]], i[[SZ]]* [[LOCAL_VLA2]] 752 // CHECK-DAG: store i16* [[ARG_C:%.+]], i16** [[LOCAL_C]] 753 // Store captures in the context. 754 // CHECK-DAG: [[REF_THIS:%.+]] = load [[S1]]*, [[S1]]** [[LOCAL_THIS]], 755 // CHECK-64-DAG:[[CONV_BP:%.+]] = bitcast i[[SZ]]* [[LOCAL_B]] to i32* 756 // CHECK-DAG: [[VAL_VLA1:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA1]], 757 // CHECK-DAG: [[VAL_VLA2:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA2]], 758 // CHECK-DAG: [[REF_C:%.+]] = load i16*, i16** [[LOCAL_C]], 759 760 // CHECK-64-DAG:[[CONV_B:%.+]] = load i32, i32* [[CONV_BP]] 761 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_B_CASTED]] to i32* 762 // CHECK-64-DAG:store i32 [[CONV_B]], i32* [[CONV]], align 763 // CHECK-32-DAG:[[LOCAL_BV:%.+]] = load i32, i32* [[LOCAL_B]] 764 // CHECK-32-DAG:store i32 [[LOCAL_BV]], i32* [[LOCAL_B_CASTED]], align 765 // CHECK-DAG: [[REF_B:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_B_CASTED]], 766 767 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) 768 // 769 // 770 // CHECK: define internal {{.*}}void [[OMP_OUTLINED5]](i32* noalias %.global_tid., i32* noalias %.bound_tid., [[S1]]* %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i16* {{.+}}) 771 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 772 773 774 // CHECK: define internal void [[HVT6]] 775 // Create local storage for each capture. 776 // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] 777 // CHECK: [[LOCAL_AA:%.+]] = alloca i[[SZ]] 778 // CHECK: [[LOCAL_AAA:%.+]] = alloca i[[SZ]] 779 // CHECK: [[LOCAL_B:%.+]] = alloca [10 x i32]* 780 // CHECK: [[LOCAL_A_CASTED:%.+]] = alloca i[[SZ]] 781 // CHECK: [[LOCAL_AA_CASTED:%.+]] = alloca i[[SZ]] 782 // CHECK: [[LOCAL_AAA_CASTED:%.+]] = alloca i[[SZ]] 783 // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] 784 // CHECK-DAG: store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]] 785 // CHECK-DAG: store i[[SZ]] [[ARG_AAA:%.+]], i[[SZ]]* [[LOCAL_AAA]] 786 // CHECK-DAG: store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]] 787 // Store captures in the context. 788 // CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* 789 // CHECK-DAG: [[CONV_AAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16* 790 // CHECK-DAG: [[CONV_AAAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AAA]] to i8* 791 // CHECK-DAG: [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]], 792 793 // CHECK-64-DAG:[[CONV_A:%.+]] = load i32, i32* [[CONV_AP]] 794 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_A_CASTED]] to i32* 795 // CHECK-64-DAG:store i32 [[CONV_A]], i32* [[CONV]], align 796 // CHECK-32-DAG:[[LOCAL_AV:%.+]] = load i32, i32* [[LOCAL_A]] 797 // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align 798 // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], 799 800 // CHECK-DAG: [[CONV_AA:%.+]] = load i16, i16* [[CONV_AAP]] 801 // CHECK-DAG: [[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA_CASTED]] to i16* 802 // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align 803 // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], 804 805 // CHECK-DAG: [[CONV_AAA:%.+]] = load i8, i8* [[CONV_AAAP]] 806 // CHECK-DAG: [[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_AAA_CASTED]] to i8* 807 // CHECK-DAG: store i8 [[CONV_AAA]], i8* [[CONV]], align 808 // CHECK-DAG: [[REF_AAA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AAA_CASTED]], 809 810 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) 811 // 812 // 813 // CHECK: define internal {{.*}}void [[OMP_OUTLINED6]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) 814 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 815 816 // CHECK: define internal void [[HVT5]] 817 // Create local storage for each capture. 818 // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] 819 // CHECK: [[LOCAL_AA:%.+]] = alloca i[[SZ]] 820 // CHECK: [[LOCAL_B:%.+]] = alloca [10 x i32]* 821 // CHECK: [[LOCAL_A_CASTED:%.+]] = alloca i[[SZ]] 822 // CHECK: [[LOCAL_AA_CASTED:%.+]] = alloca i[[SZ]] 823 // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] 824 // CHECK-DAG: store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]] 825 // CHECK-DAG: store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]] 826 // Store captures in the context. 827 // CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* 828 // CHECK-DAG: [[CONV_AAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16* 829 // CHECK-DAG: [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]], 830 831 // CHECK-64-DAG:[[CONV_A:%.+]] = load i32, i32* [[CONV_AP]] 832 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_A_CASTED]] to i32* 833 // CHECK-64-DAG:store i32 [[CONV_A]], i32* [[CONV]], align 834 // CHECK-32-DAG:[[LOCAL_AV:%.+]] = load i32, i32* [[LOCAL_A]] 835 // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align 836 // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], 837 838 // CHECK-DAG: [[CONV_AA:%.+]] = load i16, i16* [[CONV_AAP]] 839 // CHECK-DAG: [[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA_CASTED]] to i16* 840 // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align 841 // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], 842 843 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) 844 // 845 // 846 // CHECK: define internal {{.*}}void [[OMP_OUTLINED7]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) 847 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 848 849 void foo1() { 850 const int n = 0; 851 #pragma omp target teams shared(n) 852 #pragma omp parallel firstprivate(n) 853 (void)n; 854 } 855 void foo() { 856 const int n = 0; 857 #pragma omp target teams firstprivate(n) 858 #pragma omp parallel shared(n) 859 (void)n; 860 } 861 862 // define {{.*}}void @__omp_offloading_{{.*}}foo1{{.*}}_l841(i[[SZ]] %{{.+}}) 863 // define internal void {{@.+}}(i32* {{.+}}, i32* {{.+}}, i[[SZ]] %{{.+}}) 864 // define {{.*}}void @__omp_offloading_{{.*}}foo1{{.*}}_l847(i[[SZ]] %{{.+}}) 865 // define internal void {{@.+}}(i32* {{.+}}, i32* {{.+}}, i32* dereferenceable{{.+}}) 866 #endif 867