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 %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 %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 %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 %s --check-prefix CHECK --check-prefix CHECK-32 8 9 // Test target codegen - host bc file has to be created first. 10 // 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 11 // 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 %s --check-prefix TCHECK --check-prefix TCHECK-64 12 // 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 13 // 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 %s --check-prefix TCHECK --check-prefix TCHECK-64 14 // 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 15 // 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 %s --check-prefix TCHECK --check-prefix TCHECK-32 16 // 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 17 // 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 %s --check-prefix TCHECK --check-prefix TCHECK-32 18 19 // expected-no-diagnostics 20 #ifndef HEADER 21 #define HEADER 22 23 // CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } 24 // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" 25 // CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } 26 27 // CHECK-DAG: [[TT:%.+]] = type { i64, i8 } 28 // CHECK-DAG: [[S1:%.+]] = type { double } 29 // CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } 30 // CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } 31 // CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } 32 33 // TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } 34 35 // CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat 36 37 // We have 8 target regions, but only 7 that actually will generate offloading 38 // code, only 6 will have mapped arguments, and only 4 have all-constant map 39 // sizes. 40 41 // CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] [i[[SZ:32|64]] 2] 42 // CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i32] [i32 288] 43 // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2] 44 // CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i32] [i32 288, i32 288] 45 // CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i32] [i32 288, i32 547, i32 288, i32 547, i32 547, i32 288, i32 288, i32 547, i32 547] 46 // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 40] 47 // CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i32] [i32 288, i32 288, i32 547] 48 // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [4 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 1, i[[SZ]] 40] 49 // CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i32] [i32 288, i32 288, i32 288, i32 547] 50 // CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [5 x i32] [i32 547, i32 288, i32 288, i32 288, i32 547] 51 // CHECK-DAG: @{{.*}} = private constant i8 0 52 // CHECK-DAG: @{{.*}} = private constant i8 0 53 // CHECK-DAG: @{{.*}} = private constant i8 0 54 // CHECK-DAG: @{{.*}} = private constant i8 0 55 // CHECK-DAG: @{{.*}} = private constant i8 0 56 // CHECK-DAG: @{{.*}} = private constant i8 0 57 // CHECK-DAG: @{{.*}} = private constant i8 0 58 59 // TCHECK: @{{.+}} = constant [[ENTTY]] 60 // TCHECK: @{{.+}} = constant [[ENTTY]] 61 // TCHECK: @{{.+}} = constant [[ENTTY]] 62 // TCHECK: @{{.+}} = constant [[ENTTY]] 63 // TCHECK: @{{.+}} = constant [[ENTTY]] 64 // TCHECK: @{{.+}} = constant [[ENTTY]] 65 // TCHECK: @{{.+}} = constant [[ENTTY]] 66 // TCHECK-NOT: @{{.+}} = constant [[ENTTY]] 67 68 // Check if offloading descriptor is created. 69 // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] 70 // CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] 71 // CHECK: [[DEVBEGIN:@.+]] = external constant i8 72 // CHECK: [[DEVEND:@.+]] = external constant i8 73 // CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) 74 // 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]]) 75 76 // Check target registration is registered as a Ctor. 77 // CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] 78 79 80 template<typename tx, typename ty> 81 struct TT{ 82 tx X; 83 ty Y; 84 }; 85 86 int global; 87 88 // CHECK: define {{.*}}[[FOO:@.+]]( 89 int foo(int n) { 90 int a = 0; 91 short aa = 0; 92 float b[10]; 93 float bn[n]; 94 double c[5][10]; 95 double cn[5][n]; 96 TT<long long, char> d; 97 98 // CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams(i32 -1, i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i32* null, i32 0, i32 0) 99 // CHECK: store i32 [[RET]], i32* [[RHV:%.+]], align 4 100 // CHECK: [[RET2:%.+]] = load i32, i32* [[RHV]], align 4 101 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET2]], 0 102 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 103 // CHECK: [[FAIL]] 104 // CHECK: call void [[HVT0:@.+]]() 105 // CHECK-NEXT: br label %[[END]] 106 // CHECK: [[END]] 107 #pragma omp target teams 108 { 109 } 110 111 // CHECK: store i32 0, i32* [[RHV:%.+]], align 4 112 // CHECK: store i32 -1, i32* [[RHV]], align 4 113 // CHECK: [[RET2:%.+]] = load i32, i32* [[RHV]], align 4 114 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET2]], 0 115 // CHECK: call void [[HVT1:@.+]](i[[SZ]] {{[^,]+}}) 116 #pragma omp target teams if(target: 0) 117 { 118 a += 1; 119 } 120 121 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i32 -1, i8* @{{[^,]+}}, i32 1, i8** [[BP:%[^,]+]], i8** [[P:%[^,]+]], i[[SZ]]* getelementptr inbounds ([1 x i[[SZ]]], [1 x i[[SZ]]]* [[SIZET2]], i32 0, i32 0), i32* getelementptr inbounds ([1 x i32], [1 x i32]* [[MAPT2]], i32 0, i32 0), i32 0, i32 0) 122 // CHECK-DAG: [[BP]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BPR:%[^,]+]], i32 0, i32 0 123 // CHECK-DAG: [[P]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PR:%[^,]+]], i32 0, i32 0 124 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BPR]], i32 0, i32 [[IDX0:[0-9]+]] 125 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [1 x i8*], [1 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]] 130 131 // CHECK: store i32 [[RET]], i32* [[RHV:%.+]], align 4 132 // CHECK: [[RET2:%.+]] = load i32, i32* [[RHV]], align 4 133 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET2]], 0 134 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 135 // CHECK: [[FAIL]] 136 // CHECK: call void [[HVT2:@.+]](i[[SZ]] {{[^,]+}}) 137 // CHECK-NEXT: br label %[[END]] 138 // CHECK: [[END]] 139 #pragma omp target teams if(target: 1) 140 { 141 aa += 1; 142 } 143 144 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 10 145 // CHECK: br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] 146 // CHECK: [[IFTHEN]] 147 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i32 -1, i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* getelementptr inbounds ([2 x i[[SZ]]], [2 x i[[SZ]]]* [[SIZET3]], i32 0, i32 0), i32* getelementptr inbounds ([2 x i32], [2 x i32]* [[MAPT3]], i32 0, i32 0), i32 0, i32 0) 148 // CHECK-DAG: [[BPR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 149 // CHECK-DAG: [[PR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%[^,]+]], i32 0, i32 0 150 151 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 152 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 153 // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]* 154 // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]* 155 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR0]] 156 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR0]] 157 158 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 159 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 160 // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* 161 // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* 162 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR1]] 163 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR1]] 164 // CHECK: store i32 [[RET]], i32* [[RHV:%.+]], align 4 165 // CHECK-NEXT: br label %[[IFEND:.+]] 166 167 // CHECK: [[IFELSE]] 168 // CHECK: store i32 -1, i32* [[RHV]], align 4 169 // CHECK-NEXT: br label %[[IFEND:.+]] 170 171 // CHECK: [[IFEND]] 172 // CHECK: [[RET2:%.+]] = load i32, i32* [[RHV]], align 4 173 // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET2]], 0 174 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] 175 // CHECK: [[FAIL]] 176 // CHECK: call void [[HVT3:@.+]]({{[^,]+}}, {{[^,]+}}) 177 // CHECK-NEXT: br label %[[END]] 178 // CHECK: [[END]] 179 #pragma omp target teams if(target: n>10) 180 { 181 a += 1; 182 aa += 1; 183 global += 1; 184 } 185 186 // We capture 3 VLA sizes in this target region 187 // CHECK-64: [[A_VAL:%.+]] = load i32, i32* %{{.+}}, 188 // CHECK-64: [[A_ADDR:%.+]] = bitcast i[[SZ]]* [[A_CADDR:%.+]] to i32* 189 // CHECK-64: store i32 [[A_VAL]], i32* [[A_ADDR]], 190 // CHECK-64: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], 191 192 // CHECK-32: [[A_VAL:%.+]] = load i32, i32* %{{.+}}, 193 // CHECK-32: store i32 [[A_VAL]], i32* [[A_CADDR:%.+]], 194 // CHECK-32: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], 195 196 // CHECK: [[BNSIZE:%.+]] = mul nuw i[[SZ]] [[VLA0:%.+]], 4 197 // CHECK: [[CNELEMSIZE2:%.+]] = mul nuw i[[SZ]] 5, [[VLA1:%.+]] 198 // CHECK: [[CNSIZE:%.+]] = mul nuw i[[SZ]] [[CNELEMSIZE2]], 8 199 200 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 201 // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] 202 // CHECK: [[TRY]] 203 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i32 -1, i8* @{{[^,]+}}, i32 9, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i32* getelementptr inbounds ([9 x i32], [9 x i32]* [[MAPT4]], i32 0, i32 0), i32 0, i32 0) 204 // CHECK-DAG: [[BPR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 205 // CHECK-DAG: [[PR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P:%[^,]+]], i32 0, i32 0 206 // CHECK-DAG: [[SR]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S:%[^,]+]], i32 0, i32 0 207 208 // CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX0:[0-9]+]] 209 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX0]] 210 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX0]] 211 // CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX1:[0-9]+]] 212 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX1]] 213 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX1]] 214 // CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX2:[0-9]+]] 215 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX2]] 216 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX2]] 217 // CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX3:[0-9]+]] 218 // CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX3]] 219 // CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX3]] 220 // CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX4:[0-9]+]] 221 // CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX4]] 222 // CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX4]] 223 // CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX5:[0-9]+]] 224 // CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX5]] 225 // CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX5]] 226 // CHECK-DAG: [[SADDR6:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX6:[0-9]+]] 227 // CHECK-DAG: [[BPADDR6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX6]] 228 // CHECK-DAG: [[PADDR6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX6]] 229 // CHECK-DAG: [[SADDR7:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX7:[0-9]+]] 230 // CHECK-DAG: [[BPADDR7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX7]] 231 // CHECK-DAG: [[PADDR7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX7]] 232 // CHECK-DAG: [[SADDR8:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX8:[0-9]+]] 233 // CHECK-DAG: [[BPADDR8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX8]] 234 // CHECK-DAG: [[PADDR8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX8]] 235 236 // The names below are not necessarily consistent with the names used for the 237 // addresses above as some are repeated. 238 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CBPADDR0:%.+]], 239 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CPADDR0:%.+]], 240 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 241 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 242 // CHECK-DAG: store i[[SZ]] {{4|8}}, i[[SZ]]* {{%[^,]+}} 243 244 // CHECK-DAG: store i[[SZ]] [[VLA1]], i[[SZ]]* [[CBPADDR1:%.+]], 245 // CHECK-DAG: store i[[SZ]] [[VLA1]], i[[SZ]]* [[CPADDR1:%.+]], 246 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 247 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 248 // CHECK-DAG: store i[[SZ]] {{4|8}}, i[[SZ]]* {{%[^,]+}} 249 250 // CHECK-DAG: store i[[SZ]] 5, i[[SZ]]* [[CBPADDR2:%.+]], 251 // CHECK-DAG: store i[[SZ]] 5, i[[SZ]]* [[CPADDR2:%.+]], 252 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 253 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 254 // CHECK-DAG: store i[[SZ]] {{4|8}}, i[[SZ]]* {{%[^,]+}} 255 256 // CHECK-DAG: store i[[SZ]] [[A_CVAL]], i[[SZ]]* [[CBPADDR3:%.+]], 257 // CHECK-DAG: store i[[SZ]] [[A_CVAL]], i[[SZ]]* [[CPADDR3:%.+]], 258 // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 259 // CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 260 // CHECK-DAG: store i[[SZ]] 4, i[[SZ]]* {{%[^,]+}} 261 262 // CHECK-DAG: store [10 x float]* %{{.+}}, [10 x float]** [[CBPADDR4:%.+]], 263 // CHECK-DAG: store [10 x float]* %{{.+}}, [10 x float]** [[CPADDR4:%.+]], 264 // CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [10 x float]** 265 // CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to [10 x float]** 266 // CHECK-DAG: store i[[SZ]] 40, i[[SZ]]* {{%[^,]+}} 267 268 // CHECK-DAG: store float* %{{.+}}, float** [[CBPADDR5:%.+]], 269 // CHECK-DAG: store float* %{{.+}}, float** [[CPADDR5:%.+]], 270 // CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to float** 271 // CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to float** 272 // CHECK-DAG: store i[[SZ]] [[BNSIZE]], i[[SZ]]* {{%[^,]+}} 273 274 // CHECK-DAG: store [5 x [10 x double]]* %{{.+}}, [5 x [10 x double]]** [[CBPADDR6:%.+]], 275 // CHECK-DAG: store [5 x [10 x double]]* %{{.+}}, [5 x [10 x double]]** [[CPADDR6:%.+]], 276 // CHECK-DAG: [[CBPADDR6]] = bitcast i8** {{%[^,]+}} to [5 x [10 x double]]** 277 // CHECK-DAG: [[CPADDR6]] = bitcast i8** {{%[^,]+}} to [5 x [10 x double]]** 278 // CHECK-DAG: store i[[SZ]] 400, i[[SZ]]* {{%[^,]+}} 279 280 // CHECK-DAG: store double* %{{.+}}, double** [[CBPADDR7:%.+]], 281 // CHECK-DAG: store double* %{{.+}}, double** [[CPADDR7:%.+]], 282 // CHECK-DAG: [[CBPADDR7]] = bitcast i8** {{%[^,]+}} to double** 283 // CHECK-DAG: [[CPADDR7]] = bitcast i8** {{%[^,]+}} to double** 284 // CHECK-DAG: store i[[SZ]] [[CNSIZE]], i[[SZ]]* {{%[^,]+}} 285 286 // CHECK-DAG: store [[TT]]* %{{.+}}, [[TT]]** [[CBPADDR8:%.+]], 287 // CHECK-DAG: store [[TT]]* %{{.+}}, [[TT]]** [[CPADDR8:%.+]], 288 // CHECK-DAG: [[CBPADDR8]] = bitcast i8** {{%[^,]+}} to [[TT]]** 289 // CHECK-DAG: [[CPADDR8]] = bitcast i8** {{%[^,]+}} to [[TT]]** 290 // CHECK-DAG: store i[[SZ]] {{12|16}}, i[[SZ]]* {{%[^,]+}} 291 292 // CHECK: store i32 [[RET]], i32* [[RHV:%.+]], align 4 293 // CHECK: [[RET2:%.+]] = load i32, i32* [[RHV]], align 4 294 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET2]], 0 295 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 296 297 // CHECK: [[FAIL]] 298 // CHECK: call void [[HVT4:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 299 // CHECK-NEXT: br label %[[END]] 300 // CHECK: [[END]] 301 #pragma omp target teams if(target: n>20) 302 { 303 a += 1; 304 b[2] += 1.0; 305 bn[3] += 1.0; 306 c[1][2] += 1.0; 307 cn[1][3] += 1.0; 308 d.X += 1; 309 d.Y += 1; 310 } 311 312 return a; 313 } 314 315 // Check that the offloading functions are emitted and that the arguments are 316 // correct and loaded correctly for the target regions in foo(). 317 318 // CHECK: define internal void [[HVT0]]() 319 // CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*)) 320 // 321 // 322 // CHECK: define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid.) 323 // CHECK: ret void 324 // CHECK-NEXT: } 325 326 327 // CHECK: define internal void [[HVT1]](i[[SZ]] %{{.+}}) 328 // Create stack storage and store argument in there. 329 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 330 // CHECK: [[AA_CASTED:%.+]] = alloca i[[SZ]], align 331 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 332 // CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* 333 // CHECK-64: [[AA:%.+]] = load i32, i32* [[AA_CADDR]], align 334 // CHECK-32: [[AA:%.+]] = load i32, i32* [[AA_ADDR]], align 335 // CHECK-64: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i32* 336 // CHECK-64: store i32 [[AA]], i32* [[AA_C]], align 337 // CHECK-32: store i32 [[AA]], i32* [[AA_CASTED]], align 338 // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align 339 // CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) 340 // 341 // 342 // CHECK: define internal {{.*}}void [[OMP_OUTLINED1]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) 343 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 344 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 345 // CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* 346 // CHECK-64: [[AA:%.+]] = load i32, i32* [[AA_CADDR]], align 347 // CHECK-32: [[AA:%.+]] = load i32, i32* [[AA_ADDR]], align 348 // CHECK: ret void 349 // CHECK-NEXT: } 350 351 // CHECK: define internal void [[HVT2]](i[[SZ]] %{{.+}}) 352 // Create stack storage and store argument in there. 353 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 354 // CHECK: [[AA_CASTED:%.+]] = alloca i[[SZ]], align 355 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 356 // CHECK: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 357 // CHECK: [[AA:%.+]] = load i16, i16* [[AA_CADDR]], align 358 // CHECK: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* 359 // CHECK: store i16 [[AA]], i16* [[AA_C]], align 360 // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align 361 // CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) 362 // 363 // 364 // CHECK: define internal {{.*}}void [[OMP_OUTLINED2]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) 365 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 366 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 367 // CHECK: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 368 // CHECK: [[AA:%.+]] = load i16, i16* [[AA_CADDR]], align 369 // CHECK: ret void 370 // CHECK-NEXT: } 371 372 // CHECK: define internal void [[HVT3]] 373 // Create stack storage and store argument in there. 374 // CHECK: [[A_ADDR:%.+]] = alloca i[[SZ]], align 375 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 376 // CHECK: [[A_CASTED:%.+]] = alloca i[[SZ]], align 377 // CHECK: [[AA_CASTED:%.+]] = alloca i[[SZ]], align 378 // CHECK: [[GLOBAL_CASTED:%.+]] = alloca i[[SZ]], align 379 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[A_ADDR]], align 380 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 381 // CHECK-64-DAG:[[A_CADDR:%.+]] = bitcast i[[SZ]]* [[A_ADDR]] to i32* 382 // CHECK-DAG: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 383 // CHECK-64-DAG:[[A:%.+]] = load i32, i32* [[A_CADDR]], align 384 // CHECK-32-DAG:[[A:%.+]] = load i32, i32* [[A_ADDR]], align 385 // CHECK-64-DAG:[[A_C:%.+]] = bitcast i[[SZ]]* [[A_CASTED]] to i32* 386 // CHECK-64-DAG:store i32 [[A]], i32* [[A_C]], align 387 // CHECK-32-DAG:store i32 [[A]], i32* [[A_CASTED]], align 388 // CHECK-DAG: [[AA:%.+]] = load i16, i16* [[AA_CADDR]], align 389 // CHECK-DAG: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* 390 // CHECK-DAG: store i16 [[AA]], i16* [[AA_C]], align 391 // CHECK-DAG: [[GLOBAL:%.+]] = load i32, i32* @global, align 392 // CHECK-64-DAG:[[GLOBAL_C:%.+]] = bitcast i[[SZ]]* [[GLOBAL_CASTED]] to i32* 393 // CHECK-64-DAG:store i32 [[GLOBAL]], i32* [[GLOBAL_C]], align 394 // CHECK-32-DAG:store i32 [[GLOBAL]], i32* [[GLOBAL_CASTED]], align 395 // CHECK-DAG: [[PARAM1:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CASTED]], align 396 // CHECK-DAG: [[PARAM2:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align 397 // CHECK-DAG: [[PARAM3:%.+]] = load i[[SZ]], i[[SZ]]* [[GLOBAL_CASTED]], align 398 // CHECK-DAG: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]], i[[SZ]] [[PARAM3]]) 399 // 400 // 401 // CHECK: define internal {{.*}}void [[OMP_OUTLINED3]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}) 402 // CHECK: [[A_ADDR:%.+]] = alloca i[[SZ]], align 403 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 404 // CHECK: [[GLOBAL_ADDR:%.+]] = alloca i[[SZ]], align 405 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[A_ADDR]], align 406 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 407 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[GLOBAL_ADDR]], align 408 // CHECK-64-DAG:[[A_CADDR:%.+]] = bitcast i[[SZ]]* [[A_ADDR]] to i32* 409 // CHECK-DAG: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 410 // CHECK-64-DAG:[[GLOBAL_CADDR:%.+]] = bitcast i[[SZ]]* [[GLOBAL_ADDR]] to i32* 411 // CHECK-64: store i32 {{.+}}, i32* [[GLOBAL_CADDR]], 412 // CHECK-32: store i32 {{.+}}, i32* [[GLOBAL_ADDR]], 413 // CHECK: ret void 414 // CHECK-NEXT: } 415 416 // CHECK: define internal void [[HVT4]] 417 // Create local storage for each capture. 418 // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] 419 // CHECK: [[LOCAL_B:%.+]] = alloca [10 x float]* 420 // CHECK: [[LOCAL_VLA1:%.+]] = alloca i[[SZ]] 421 // CHECK: [[LOCAL_BN:%.+]] = alloca float* 422 // CHECK: [[LOCAL_C:%.+]] = alloca [5 x [10 x double]]* 423 // CHECK: [[LOCAL_VLA2:%.+]] = alloca i[[SZ]] 424 // CHECK: [[LOCAL_VLA3:%.+]] = alloca i[[SZ]] 425 // CHECK: [[LOCAL_CN:%.+]] = alloca double* 426 // CHECK: [[LOCAL_D:%.+]] = alloca [[TT]]* 427 // CHECK: [[LOCAL_A_CASTED:%.+]] = alloca i[[SZ]] 428 // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] 429 // CHECK-DAG: store [10 x float]* [[ARG_B:%.+]], [10 x float]** [[LOCAL_B]] 430 // CHECK-DAG: store i[[SZ]] [[ARG_VLA1:%.+]], i[[SZ]]* [[LOCAL_VLA1]] 431 // CHECK-DAG: store float* [[ARG_BN:%.+]], float** [[LOCAL_BN]] 432 // CHECK-DAG: store [5 x [10 x double]]* [[ARG_C:%.+]], [5 x [10 x double]]** [[LOCAL_C]] 433 // CHECK-DAG: store i[[SZ]] [[ARG_VLA2:%.+]], i[[SZ]]* [[LOCAL_VLA2]] 434 // CHECK-DAG: store i[[SZ]] [[ARG_VLA3:%.+]], i[[SZ]]* [[LOCAL_VLA3]] 435 // CHECK-DAG: store double* [[ARG_CN:%.+]], double** [[LOCAL_CN]] 436 // CHECK-DAG: store [[TT]]* [[ARG_D:%.+]], [[TT]]** [[LOCAL_D]] 437 438 // CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* 439 // CHECK-DAG: [[REF_B:%.+]] = load [10 x float]*, [10 x float]** [[LOCAL_B]], 440 // CHECK-DAG: [[VAL_VLA1:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA1]], 441 // CHECK-DAG: [[REF_BN:%.+]] = load float*, float** [[LOCAL_BN]], 442 // CHECK-DAG: [[REF_C:%.+]] = load [5 x [10 x double]]*, [5 x [10 x double]]** [[LOCAL_C]], 443 // CHECK-DAG: [[VAL_VLA2:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA2]], 444 // CHECK-DAG: [[VAL_VLA3:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA3]], 445 // CHECK-DAG: [[REF_CN:%.+]] = load double*, double** [[LOCAL_CN]], 446 // CHECK-DAG: [[REF_D:%.+]] = load [[TT]]*, [[TT]]** [[LOCAL_D]], 447 448 // CHECK-64-DAG:[[CONV_A:%.+]] = load i32, i32* [[CONV_AP]] 449 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_A_CASTED]] to i32* 450 // CHECK-64-DAG:store i32 [[CONV_A]], i32* [[CONV]], align 451 // CHECK-32-DAG:[[LOCAL_AV:%.+]] = load i32, i32* [[LOCAL_A]] 452 // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align 453 // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], 454 455 // CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%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]]) 456 // 457 // 458 // 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]]* {{.+}}) 459 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 460 461 template<typename tx> 462 tx ftemplate(int n) { 463 tx a = 0; 464 short aa = 0; 465 tx b[10]; 466 467 #pragma omp target teams if(target: n>40) 468 { 469 a += 1; 470 aa += 1; 471 b[2] += 1; 472 } 473 474 return a; 475 } 476 477 static 478 int fstatic(int n) { 479 int a = 0; 480 short aa = 0; 481 char aaa = 0; 482 int b[10]; 483 484 #pragma omp target teams if(target: n>50) 485 { 486 a += 1; 487 aa += 1; 488 aaa += 1; 489 b[2] += 1; 490 } 491 492 return a; 493 } 494 495 struct S1 { 496 double a; 497 498 int r1(int n){ 499 int b = n+1; 500 short int c[2][n]; 501 502 #pragma omp target teams if(target: n>60) 503 { 504 this->a = (double)b + 1.5; 505 c[1][1] = ++a; 506 } 507 508 return c[1][1] + (int)b; 509 } 510 }; 511 512 // CHECK: define {{.*}}@{{.*}}bar{{.*}} 513 int bar(int n){ 514 int a = 0; 515 516 // CHECK: call {{.*}}i32 [[FOO]](i32 {{.*}}) 517 a += foo(n); 518 519 S1 S; 520 // CHECK: call {{.*}}i32 [[FS1:@.+]]([[S1]]* {{.*}}, i32 {{.*}}) 521 a += S.r1(n); 522 523 // CHECK: call {{.*}}i32 [[FSTATIC:@.+]](i32 {{.*}}) 524 a += fstatic(n); 525 526 // CHECK: call {{.*}}i32 [[FTEMPLATE:@.+]](i32 {{.*}}) 527 a += ftemplate<int>(n); 528 529 return a; 530 } 531 532 // 533 // CHECK: define {{.*}}[[FS1]] 534 // 535 // CHECK: i8* @llvm.stacksave() 536 // CHECK-64: [[B_ADDR:%.+]] = bitcast i[[SZ]]* [[B_CADDR:%.+]] to i32* 537 // CHECK-64: store i32 %{{.+}}, i32* [[B_ADDR]], 538 // CHECK-64: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]], 539 540 // CHECK-32: store i32 %{{.+}}, i32* [[B_ADDR:%.+]], 541 // CHECK-32: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]], 542 543 // We capture 2 VLA sizes in this target region 544 // CHECK: [[CELEMSIZE2:%.+]] = mul nuw i[[SZ]] 2, [[VLA0:%.+]] 545 // CHECK: [[CSIZE:%.+]] = mul nuw i[[SZ]] [[CELEMSIZE2]], 2 546 547 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 548 // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] 549 // CHECK: [[TRY]] 550 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i32 -1, i8* @{{[^,]+}}, i32 5, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i32* getelementptr inbounds ([5 x i32], [5 x i32]* [[MAPT7]], i32 0, i32 0), i32 0, i32 0) 551 // CHECK-DAG: [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0 552 // CHECK-DAG: [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0 553 // CHECK-DAG: [[SR]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 554 // CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] 555 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX0]] 556 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX0]] 557 // CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] 558 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX1]] 559 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX1]] 560 // CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] 561 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX2]] 562 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX2]] 563 // CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] 564 // CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX3]] 565 // CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX3]] 566 567 // The names below are not necessarily consistent with the names used for the 568 // addresses above as some are repeated. 569 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CBPADDR0:%.+]], 570 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CPADDR0:%.+]], 571 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 572 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 573 // CHECK-DAG: store i[[SZ]] {{4|8}}, i[[SZ]]* {{%[^,]+}} 574 575 // CHECK-DAG: store i[[SZ]] 2, i[[SZ]]* [[CBPADDR1:%.+]], 576 // CHECK-DAG: store i[[SZ]] 2, i[[SZ]]* [[CPADDR1:%.+]], 577 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 578 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 579 // CHECK-DAG: store i[[SZ]] {{4|8}}, i[[SZ]]* {{%[^,]+}} 580 581 // CHECK-DAG: store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CBPADDR2:%.+]], 582 // CHECK-DAG: store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CPADDR2:%.+]], 583 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 584 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 585 // CHECK-DAG: store i[[SZ]] 4, i[[SZ]]* {{%[^,]+}} 586 587 // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR3:%.+]], 588 // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR3:%.+]], 589 // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** 590 // CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** 591 // CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} 592 593 // CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR4:%.+]], 594 // CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR4:%.+]], 595 // CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to i16** 596 // CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to i16** 597 // CHECK-DAG: store i[[SZ]] [[CSIZE]], i[[SZ]]* {{%[^,]+}} 598 599 // CHECK: store i32 [[RET]], i32* [[RHV:%.+]], align 4 600 // CHECK: [[RET2:%.+]] = load i32, i32* [[RHV]], align 4 601 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET2]], 0 602 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 603 604 // CHECK: [[FAIL]] 605 // CHECK: call void [[HVT7:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 606 // CHECK-NEXT: br label %[[END]] 607 // CHECK: [[END]] 608 609 // 610 // CHECK: define {{.*}}[[FSTATIC]] 611 // 612 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 50 613 // CHECK: br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] 614 // CHECK: [[IFTHEN]] 615 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i32 -1, i8* @{{[^,]+}}, i32 4, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* getelementptr inbounds ([4 x i[[SZ]]], [4 x i[[SZ]]]* [[SIZET6]], i32 0, i32 0), i32* getelementptr inbounds ([4 x i32], [4 x i32]* [[MAPT6]], i32 0, i32 0), i32 0, i32 0) 616 // CHECK-DAG: [[BPR]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP:%.+]], i32 0, i32 0 617 // CHECK-DAG: [[PR]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P:%.+]], i32 0, i32 0 618 619 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 0 620 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 0 621 // CHECK-DAG: store i[[SZ]] [[VAL0:%.+]], i[[SZ]]* [[CBPADDR0:%.+]], 622 // CHECK-DAG: store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0:%.+]], 623 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 624 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 625 626 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 1 627 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 1 628 // CHECK-DAG: store i[[SZ]] [[VAL1:%.+]], i[[SZ]]* [[CBPADDR1:%.+]], 629 // CHECK-DAG: store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1:%.+]], 630 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 631 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 632 633 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 2 634 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 2 635 // CHECK-DAG: store i[[SZ]] [[VAL2:%.+]], i[[SZ]]* [[CBPADDR2:%.+]], 636 // CHECK-DAG: store i[[SZ]] [[VAL2]], i[[SZ]]* [[CPADDR2:%.+]], 637 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 638 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 639 640 // CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[BP]], i32 0, i32 3 641 // CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* [[P]], i32 0, i32 3 642 // CHECK-DAG: store [10 x i32]* %{{.+}}, [10 x i32]** [[CBPADDR3:%.+]], 643 // CHECK-DAG: store [10 x i32]* %{{.+}}, [10 x i32]** [[CPADDR3:%.+]], 644 // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to [10 x i32]** 645 // CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to [10 x i32]** 646 647 // CHECK: store i32 [[RET]], i32* [[RHV:%.+]], align 4 648 // CHECK-NEXT: br label %[[IFEND:.+]] 649 650 // CHECK: [[IFELSE]] 651 // CHECK: store i32 -1, i32* [[RHV]], align 4 652 // CHECK-NEXT: br label %[[IFEND:.+]] 653 654 // CHECK: [[IFEND]] 655 // CHECK: [[RET2:%.+]] = load i32, i32* [[RHV]], align 4 656 // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET2]], 0 657 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] 658 // CHECK: [[FAIL]] 659 // CHECK: call void [[HVT6:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 660 // CHECK-NEXT: br label %[[END]] 661 // CHECK: [[END]] 662 663 // 664 // CHECK: define {{.*}}[[FTEMPLATE]] 665 // 666 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 40 667 // CHECK: br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] 668 // CHECK: [[IFTHEN]] 669 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i32 -1, i8* @{{[^,]+}}, i32 3, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* getelementptr inbounds ([3 x i[[SZ]]], [3 x i[[SZ]]]* [[SIZET5]], i32 0, i32 0), i32* getelementptr inbounds ([3 x i32], [3 x i32]* [[MAPT5]], i32 0, i32 0), i32 0, i32 0) 670 // CHECK-DAG: [[BPR]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP:%.+]], i32 0, i32 0 671 // CHECK-DAG: [[PR]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P:%.+]], i32 0, i32 0 672 673 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 0 674 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 0 675 // CHECK-DAG: store i[[SZ]] [[VAL0:%.+]], i[[SZ]]* [[CBPADDR0:%.+]], 676 // CHECK-DAG: store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0:%.+]], 677 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 678 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 679 680 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 1 681 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 1 682 // CHECK-DAG: store i[[SZ]] [[VAL1:%.+]], i[[SZ]]* [[CBPADDR1:%.+]], 683 // CHECK-DAG: store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1:%.+]], 684 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 685 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 686 687 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 2 688 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 2 689 // CHECK-DAG: store [10 x i32]* [[VAL2:%.+]], [10 x i32]** [[CBPADDR2:%.+]], 690 // CHECK-DAG: store [10 x i32]* [[VAL2]], [10 x i32]** [[CPADDR2:%.+]], 691 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to [10 x i32]** 692 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to [10 x i32]** 693 694 // CHECK: store i32 [[RET]], i32* [[RHV:%.+]], align 4 695 // CHECK-NEXT: br label %[[IFEND:.+]] 696 697 // CHECK: [[IFELSE]] 698 // CHECK: store i32 -1, i32* [[RHV]], align 4 699 // CHECK-NEXT: br label %[[IFEND:.+]] 700 701 // CHECK: [[IFEND]] 702 // CHECK: [[RET2:%.+]] = load i32, i32* [[RHV]], align 4 703 // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET2]], 0 704 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] 705 // CHECK: [[FAIL]] 706 // CHECK: call void [[HVT5:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 707 // CHECK-NEXT: br label %[[END]] 708 // CHECK: [[END]] 709 710 711 712 // Check that the offloading functions are emitted and that the arguments are 713 // correct and loaded correctly for the target regions of the callees of bar(). 714 715 // CHECK: define internal void [[HVT7]] 716 // Create local storage for each capture. 717 // CHECK: [[LOCAL_THIS:%.+]] = alloca [[S1]]* 718 // CHECK: [[LOCAL_B:%.+]] = alloca i[[SZ]] 719 // CHECK: [[LOCAL_VLA1:%.+]] = alloca i[[SZ]] 720 // CHECK: [[LOCAL_VLA2:%.+]] = alloca i[[SZ]] 721 // CHECK: [[LOCAL_C:%.+]] = alloca i16* 722 // CHECK: [[LOCAL_B_CASTED:%.+]] = alloca i[[SZ]] 723 // CHECK-DAG: store [[S1]]* [[ARG_THIS:%.+]], [[S1]]** [[LOCAL_THIS]] 724 // CHECK-DAG: store i[[SZ]] [[ARG_B:%.+]], i[[SZ]]* [[LOCAL_B]] 725 // CHECK-DAG: store i[[SZ]] [[ARG_VLA1:%.+]], i[[SZ]]* [[LOCAL_VLA1]] 726 // CHECK-DAG: store i[[SZ]] [[ARG_VLA2:%.+]], i[[SZ]]* [[LOCAL_VLA2]] 727 // CHECK-DAG: store i16* [[ARG_C:%.+]], i16** [[LOCAL_C]] 728 // Store captures in the context. 729 // CHECK-DAG: [[REF_THIS:%.+]] = load [[S1]]*, [[S1]]** [[LOCAL_THIS]], 730 // CHECK-64-DAG:[[CONV_BP:%.+]] = bitcast i[[SZ]]* [[LOCAL_B]] to i32* 731 // CHECK-DAG: [[VAL_VLA1:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA1]], 732 // CHECK-DAG: [[VAL_VLA2:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA2]], 733 // CHECK-DAG: [[REF_C:%.+]] = load i16*, i16** [[LOCAL_C]], 734 735 // CHECK-64-DAG:[[CONV_B:%.+]] = load i32, i32* [[CONV_BP]] 736 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_B_CASTED]] to i32* 737 // CHECK-64-DAG:store i32 [[CONV_B]], i32* [[CONV]], align 738 // CHECK-32-DAG:[[LOCAL_BV:%.+]] = load i32, i32* [[LOCAL_B]] 739 // CHECK-32-DAG:store i32 [[LOCAL_BV]], i32* [[LOCAL_B_CASTED]], align 740 // CHECK-DAG: [[REF_B:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_B_CASTED]], 741 742 // CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%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]]) 743 // 744 // 745 // CHECK: define internal {{.*}}void [[OMP_OUTLINED5]](i32* noalias %.global_tid., i32* noalias %.bound_tid., [[S1]]* %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i16* %{{.+}}) 746 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 747 748 749 // CHECK: define internal void [[HVT6]] 750 // Create local storage for each capture. 751 // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] 752 // CHECK: [[LOCAL_AA:%.+]] = alloca i[[SZ]] 753 // CHECK: [[LOCAL_AAA:%.+]] = alloca i[[SZ]] 754 // CHECK: [[LOCAL_B:%.+]] = alloca [10 x i32]* 755 // CHECK: [[LOCAL_A_CASTED:%.+]] = alloca i[[SZ]] 756 // CHECK: [[LOCAL_AA_CASTED:%.+]] = alloca i[[SZ]] 757 // CHECK: [[LOCAL_AAA_CASTED:%.+]] = alloca i[[SZ]] 758 // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] 759 // CHECK-DAG: store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]] 760 // CHECK-DAG: store i[[SZ]] [[ARG_AAA:%.+]], i[[SZ]]* [[LOCAL_AAA]] 761 // CHECK-DAG: store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]] 762 // Store captures in the context. 763 // CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* 764 // CHECK-DAG: [[CONV_AAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16* 765 // CHECK-DAG: [[CONV_AAAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AAA]] to i8* 766 // CHECK-DAG: [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]], 767 768 // CHECK-64-DAG:[[CONV_A:%.+]] = load i32, i32* [[CONV_AP]] 769 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_A_CASTED]] to i32* 770 // CHECK-64-DAG:store i32 [[CONV_A]], i32* [[CONV]], align 771 // CHECK-32-DAG:[[LOCAL_AV:%.+]] = load i32, i32* [[LOCAL_A]] 772 // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align 773 // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], 774 775 // CHECK-DAG: [[CONV_AA:%.+]] = load i16, i16* [[CONV_AAP]] 776 // CHECK-DAG: [[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA_CASTED]] to i16* 777 // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align 778 // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], 779 780 // CHECK-DAG: [[CONV_AAA:%.+]] = load i8, i8* [[CONV_AAAP]] 781 // CHECK-DAG: [[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_AAA_CASTED]] to i8* 782 // CHECK-DAG: store i8 [[CONV_AAA]], i8* [[CONV]], align 783 // CHECK-DAG: [[REF_AAA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AAA_CASTED]], 784 785 // CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%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]]) 786 // 787 // 788 // CHECK: define internal {{.*}}void [[OMP_OUTLINED6]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) 789 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 790 791 // CHECK: define internal void [[HVT5]] 792 // Create local storage for each capture. 793 // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] 794 // CHECK: [[LOCAL_AA:%.+]] = alloca i[[SZ]] 795 // CHECK: [[LOCAL_B:%.+]] = alloca [10 x i32]* 796 // CHECK: [[LOCAL_A_CASTED:%.+]] = alloca i[[SZ]] 797 // CHECK: [[LOCAL_AA_CASTED:%.+]] = alloca i[[SZ]] 798 // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] 799 // CHECK-DAG: store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]] 800 // CHECK-DAG: store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]] 801 // Store captures in the context. 802 // CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* 803 // CHECK-DAG: [[CONV_AAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16* 804 // CHECK-DAG: [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]], 805 806 // CHECK-64-DAG:[[CONV_A:%.+]] = load i32, i32* [[CONV_AP]] 807 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_A_CASTED]] to i32* 808 // CHECK-64-DAG:store i32 [[CONV_A]], i32* [[CONV]], align 809 // CHECK-32-DAG:[[LOCAL_AV:%.+]] = load i32, i32* [[LOCAL_A]] 810 // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align 811 // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], 812 813 // CHECK-DAG: [[CONV_AA:%.+]] = load i16, i16* [[CONV_AAP]] 814 // CHECK-DAG: [[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA_CASTED]] to i16* 815 // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align 816 // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], 817 818 // CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%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]]) 819 // 820 // 821 // CHECK: define internal {{.*}}void [[OMP_OUTLINED7]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) 822 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 823 824 #endif 825