1 // Test host codegen. 2 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 3 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s 4 // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 5 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 6 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s 7 // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 8 9 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s 10 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s 11 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s 12 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s 13 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s 14 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s 15 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}} 16 17 // Test target codegen - host bc file has to be created first. 18 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc 19 // RUN: %clang_cc1 -verify -fopenmp -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 20 // RUN: %clang_cc1 -fopenmp -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 -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 22 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc 23 // RUN: %clang_cc1 -verify -fopenmp -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 24 // RUN: %clang_cc1 -fopenmp -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 -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 26 27 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc 28 // RUN: %clang_cc1 -verify -fopenmp-simd -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 --check-prefix SIMD-ONLY1 %s 29 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s 30 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s 31 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc 32 // RUN: %clang_cc1 -verify -fopenmp-simd -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 --check-prefix SIMD-ONLY1 %s 33 // RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s 34 // RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s 35 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}} 36 37 // expected-no-diagnostics 38 #ifndef HEADER 39 #define HEADER 40 41 template<typename tx, typename ty> 42 struct TT{ 43 tx X; 44 ty Y; 45 }; 46 47 // CHECK: [[TT:%.+]] = type { i64, i8 } 48 // CHECK: [[S1:%.+]] = type { double } 49 50 // TCHECK: [[TT:%.+]] = type { i64, i8 } 51 // TCHECK: [[S1:%.+]] = type { double } 52 53 // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] [i[[SZ:32|64]] 4] 54 // CHECK: [[MAPT:@.+]] = private unnamed_addr constant [1 x i64] [i64 800] 55 // CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [9 x i64] [i64 800, i64 673, i64 288, i64 673, i64 673, i64 288, i64 288, i64 673, i64 673] 56 // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] zeroinitializer 57 // CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [1 x i64] [i64 544] 58 // CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 673] 59 // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 1, i[[SZ]] 40] 60 // CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 673] 61 // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [2 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 40] 62 // CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 673] 63 64 65 // CHECK: define {{.*}}[[FOO:@.+]]( 66 int foo(int n, double *ptr) { 67 int a = 0; 68 short aa = 0; 69 float b[10]; 70 float bn[n]; 71 double c[5][10]; 72 double cn[5][n]; 73 TT<long long, char> d; 74 75 #pragma omp target firstprivate(a) 76 { 77 } 78 79 // a is passed by value to tgt_target 80 // CHECK: [[N_ADDR:%.+]] = alloca i{{[0-9]+}}, 81 // CHECK: [[PTR_ADDR:%.+]] = alloca double*, 82 // CHECK: [[A:%.+]] = alloca i{{[0-9]+}}, 83 // CHECK: [[A2:%.+]] = alloca i{{[0-9]+}}, 84 // CHECK: [[B:%.+]] = alloca [10 x float], 85 // CHECK: [[SSTACK:%.+]] = alloca i8*, 86 // CHECK: [[C:%.+]] = alloca [5 x [10 x double]], 87 // CHECK: [[D:%.+]] = alloca [[TT]], 88 // CHECK: [[ACAST:%.+]] = alloca i{{[0-9]+}}, 89 // CHECK: [[BASE_PTR_ARR:%.+]] = alloca [1 x i8*], 90 // CHECK: [[PTR_ARR:%.+]] = alloca [1 x i8*], 91 // CHECK: [[A2CAST:%.+]] = alloca i{{[0-9]+}}, 92 // CHECK: [[BASE_PTR_ARR2:%.+]] = alloca [9 x i8*], 93 // CHECK: [[PTR_ARR2:%.+]] = alloca [9 x i8*], 94 // CHECK: [[SIZET2:%.+]] = alloca [9 x i{{[0-9]+}}], 95 // CHECK: [[BASE_PTR_ARR3:%.+]] = alloca [1 x i8*], 96 // CHECK: [[PTR_ARR3:%.+]] = alloca [1 x i8*], 97 // CHECK: [[N_ADDR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[N_ADDR]], 98 // CHECK-64: [[N_EXT:%.+]] = zext i{{[0-9]+}} [[N_ADDR_VAL]] to i{{[0-9]+}} 99 // CHECK: [[SSAVE_RET:%.+]] = call i8* @llvm.stacksave() 100 // CHECK: store i8* [[SSAVE_RET]], i8** [[SSTACK]], 101 // CHECK-64: [[BN_VLA:%.+]] = alloca float, i{{[0-9]+}} [[N_EXT]], 102 // CHECK-32: [[BN_VLA:%.+]] = alloca float, i{{[0-9]+}} [[N_ADDR_VAL]], 103 // CHECK: [[N_ADDR_VAL2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[N_ADDR]], 104 // CHECK-64: [[N_EXT2:%.+]] = zext i{{[0-9]+}} [[N_ADDR_VAL2]] to i{{[0-9]+}} 105 // CHECK-64: [[CN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_EXT2]] 106 // CHECK-32: [[CN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_ADDR_VAL2]] 107 // CHECK: [[CN_VLA:%.+]] = alloca double, i{{[0-9]+}} [[CN_SIZE]], 108 // CHECK: [[AVAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A]], 109 // CHECK-64: [[CONV:%.+]] = bitcast i{{[0-9]+}}* [[ACAST]] to i{{[0-9]+}}* 110 // CHECK-64: store i{{[0-9]+}} [[AVAL]], i{{[0-9]+}}* [[CONV]], 111 // CHECK-32: store i{{[0-9]+}} [[AVAL]], i{{[0-9]+}}* [[ACAST]], 112 // CHECK: [[ACAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[ACAST]], 113 // CHECK: [[BASE_PTR_GEP:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 114 // CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP]] to i{{[0-9]+}}* 115 // CHECK: store i{{[0-9]+}} [[ACAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]], 116 // CHECK: [[PTR_GEP:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 117 // CHECK: [[ACAST_TOPTR2:%.+]] = bitcast i8** [[PTR_GEP]] to i{{[0-9]+}}* 118 // CHECK: store i{{[0-9]+}} [[ACAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR2]], 119 // CHECK: [[BASE_PTR_GEP_ARG:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 120 // CHECK: [[PTR_GEP_ARG:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 121 // CHECK: {{.+}} = call i32 @__tgt_target(i64 -1, {{.+}}, i32 1, i8** [[BASE_PTR_GEP_ARG]], i8** [[PTR_GEP_ARG]], i[[SZ]]* getelementptr inbounds ([1 x i[[SZ]]], [1 x i[[SZ]]]* [[SIZET]], i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[MAPT]], i32 0, i32 0)) 122 123 // TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]]) 124 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, 125 // TCHECK-NOT: alloca i{{[0-9]+}}, 126 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]], 127 // TCHECK-NOT: store i{{[0-9]+}} % 128 // TCHECK: ret void 129 130 #pragma omp target firstprivate(aa,b,bn,c,cn,d) 131 { 132 aa += 1; 133 b[2] = 1.0; 134 bn[3] = 1.0; 135 c[1][2] = 1.0; 136 cn[1][3] = 1.0; 137 d.X = 1; 138 d.Y = 1; 139 } 140 141 // CHECK: [[A2VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A2]], 142 // CHECK: [[A2CASTCONV:%.+]] = bitcast i{{[0-9]+}}* [[A2CAST]] to i{{[0-9]+}}* 143 // CHECK: store i{{[0-9]+}} [[A2VAL]], i{{[0-9]+}}* [[A2CASTCONV]], 144 // CHECK: [[A2CAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A2CAST]], 145 // CHECK-64: [[BN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} [[N_EXT]], 4 146 // CHECK-32: [[BN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} [[N_ADDR_VAL]], 4 147 // CHECK-64: [[CN_SIZE_1:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_EXT2]] 148 // CHECK-32: [[CN_SIZE_1:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_ADDR_VAL2]] 149 // CHECK: [[CN_SIZE_2:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SIZE_1]], 8 150 151 // firstprivate(aa) --> base_ptr = aa, ptr = aa, size = 2 (short) 152 // CHECK: [[BASE_PTR_GEP2_0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 153 // CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_0]] to i{{[0-9]+}}* 154 // CHECK: store i{{[0-9]+}} [[A2CAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]], 155 // CHECK: [[PTR_GEP2_0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 156 // CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_0]] to i{{[0-9]+}}* 157 // CHECK: store i{{[0-9]+}} [[A2CAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]], 158 // CHECK: [[SIZE_GEPA2:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 159 // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[SIZE_GEPA2]], 160 161 // firstprivate(b): base_ptr = &b[0], ptr = &b[0], size = 40 (sizeof(float)*10) 162 // CHECK: [[BASE_PTR_GEP2_1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 163 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_1]] to [10 x float]** 164 // CHECK: store [10 x float]* [[B]], [10 x float]** [[BCAST_TOPTR]], 165 // CHECK: [[PTR_GEP2_1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 166 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_1]] to [10 x float]** 167 // CHECK: store [10 x float]* [[B]], [10 x float]** [[BCAST_TOPTR]], 168 // CHECK: [[SIZE_GEPB:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 169 // CHECK: store i{{[0-9]+}} 40, i{{[0-9]+}}* [[SIZE_GEPB]], 170 171 // firstprivate(bn), 2 entries, n and bn: (1) base_ptr = n, ptr = n, size = 8 ; (2) base_ptr = &c[0], ptr = &c[0], size = n*sizeof(float) 172 // CHECK: [[BASE_PTR_GEP2_2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 173 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_2]] to i{{[0-9]+}}* 174 // CHECK-64: store i{{[0-9]+}} [[N_EXT]], i{{[0-9]+}}* [[BCAST_TOPTR]], 175 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL]], i{{[0-9]+}}* [[BCAST_TOPTR]], 176 // CHECK: [[PTR_GEP2_2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 177 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_2]] to i{{[0-9]+}}* 178 // CHECK-64: store i{{[0-9]+}} [[N_EXT]], i{{[0-9]+}}* [[BCAST_TOPTR]], 179 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL]], i{{[0-9]+}}* [[BCAST_TOPTR]], 180 // CHECK: [[SIZE_GEPBN_1:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 181 // CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPBN_1]], 182 // CHECK: [[BASE_PTR_GEP2_3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 183 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_3]] to float** 184 // CHECK: store float* [[BN_VLA]], float** [[BCAST_TOPTR]], 185 // CHECK: [[SIZE_GEPBN_3:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 186 // CHECK: store i{{[0-9]+}} [[BN_SIZE]], i{{[0-9]+}}* [[SIZE_GEPBN_3]] 187 188 // firstprivate(c): base_ptr = &c[0], ptr = &c[0], size = 400 (5*10*sizeof(double)) 189 // CHECK: [[BASE_PTR_GEP2_4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 190 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_4]] to [5 x [10 x double]]** 191 // CHECK: store [5 x [10 x double]]* [[C]], [5 x [10 x double]]** [[BCAST_TOPTR]], 192 // CHECK: [[PTR_GEP2_4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 193 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_4]] to [5 x [10 x double]]** 194 // CHECK: store [5 x [10 x double]]* [[C]], [5 x [10 x double]]** [[BCAST_TOPTR]], 195 // CHECK: [[SIZE_GEPC_4:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 196 // CHECK: store i{{[0-9]+}} 400, i{{[0-9]+}}* [[SIZE_GEPC_4]], 197 198 // firstprivate(cn), 3 entries, 5, n, cn: (1) base_ptr = 5, ptr = 5, size = 8; (2) (1) base_ptr = n, ptr = n, size = 8; (3) base_ptr = &cn[0], ptr = &cn[0], size = 5*n*sizeof(double) 199 // CHECK: [[BASE_PTR_GEP2_5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 200 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_5]] to i{{[0-9]+}}* 201 // CHECK: store i{{[0-9]+}} 5, i{{[0-9]+}}* [[BCAST_TOPTR]], 202 // CHECK: [[PTR_GEP2_5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 203 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_5]] to i{{[0-9]+}}* 204 // CHECK: store i{{[0-9]+}} 5, i{{[0-9]+}}* [[BCAST_TOPTR]], 205 // CHECK: [[SIZE_GEPCN_5:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 206 // CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPCN_5]], 207 // CHECK: [[BASE_PTR_GEP2_6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6 208 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_6]] to i{{[0-9]+}}* 209 // CHECK-64: store i{{[0-9]+}} [[N_EXT2]], i{{[0-9]+}}* [[BCAST_TOPTR]], 210 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL2]], i{{[0-9]+}}* [[BCAST_TOPTR]], 211 // CHECK: [[PTR_GEP2_6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6 212 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_6]] to i{{[0-9]+}}* 213 // CHECK-64: store i{{[0-9]+}} [[N_EXT2]], i{{[0-9]+}}* [[BCAST_TOPTR]], 214 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL2]], i{{[0-9]+}}* [[BCAST_TOPTR]], 215 // CHECK: [[SIZE_GEPCN_6:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6 216 // CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPCN_6]], 217 // CHECK: [[BASE_PTR_GEP2_7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7 218 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_7]] to double** 219 // CHECK: store double* [[CN_VLA]], double** [[BCAST_TOPTR]], 220 // CHECK: [[PTR_GEP2_7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7 221 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_7]] to double** 222 // CHECK: store double* [[CN_VLA]], double** [[BCAST_TOPTR]], 223 // CHECK: [[SIZE_GEPCN_7:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7 224 // CHECK: store i{{[0-9]+}} [[CN_SIZE_2]], i{{[0-9]+}}* [[SIZE_GEPCN_7]], 225 226 // firstprivate(d): base_ptr = &d, ptr = &d, size = 16 227 // CHECK: [[BASE_PTR_GEP2_8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8 228 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_8]] to [[TT]]** 229 // CHECK: store [[TT]]* [[D]], [[TT]]** [[BCAST_TOPTR]], 230 // CHECK: [[PTR_GEP2_8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8 231 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_8]] to [[TT]]** 232 // CHECK: store [[TT]]* [[D]], [[TT]]** [[BCAST_TOPTR]], 233 // CHECK: [[SIZE_GEPCN_8:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8 234 // CHECK: store i{{[0-9]+}} {{[0-9]+}}, i{{[0-9]+}}* [[SIZE_GEPCN_8]], 235 236 237 // CHECK: [[BASE_PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 238 // CHECK: [[PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 239 // CHECK: [[SIZES_ARG2:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 240 // CHECK: {{.+}} = call i32 @__tgt_target(i64 -1, {{.+}}, i32 9, i8** [[BASE_PTR_GEP_ARG2]], i8** [[PTR_GEP_ARG2]], i[[SZ]]* [[SIZES_ARG2]], i64* getelementptr inbounds ([9 x i64], [9 x i64]* [[MAPT2]], i32 0, i32 0)) 241 242 // make sure that firstprivate variables are generated in all cases and that we use those instances for operations inside the 243 // target region 244 // TCHECK: define {{.*}}void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A2_IN:%.+]], [10 x float]* {{.+}} [[B_IN:%.+]], i{{[0-9]+}} [[BN_SZ:%.+]], float* {{.+}} [[BN_IN:%.+]], [5 x [10 x double]]* {{.+}} [[C_IN:%.+]], i{{[0-9]+}} [[CN_SZ1:%.+]], i{{[0-9]+}} [[CN_SZ2:%.+]], double* {{.+}} [[CN_IN:%.+]], [[TT]]* {{.+}} [[D_IN:%.+]]) 245 // TCHECK: [[A2_ADDR:%.+]] = alloca i{{[0-9]+}}, 246 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x float]*, 247 // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}}, 248 // TCHECK: [[BN_ADDR:%.+]] = alloca float*, 249 // TCHECK: [[C_ADDR:%.+]] = alloca [5 x [10 x double]]*, 250 // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}}, 251 // TCHECK: [[VLA_ADDR4:%.+]] = alloca i{{[0-9]+}}, 252 // TCHECK: [[CN_ADDR:%.+]] = alloca double*, 253 // TCHECK: [[D_ADDR:%.+]] = alloca [[TT]]*, 254 // TCHECK-NOT: alloca i{{[0-9]+}}, 255 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x float], 256 // TCHECK: [[SSTACK:%.+]] = alloca i8*, 257 // TCHECK: [[C_PRIV:%.+]] = alloca [5 x [10 x double]], 258 // TCHECK: [[D_PRIV:%.+]] = alloca [[TT]], 259 // TCHECK: store i{{[0-9]+}} [[A2_IN]], i{{[0-9]+}}* [[A2_ADDR]], 260 // TCHECK: store [10 x float]* [[B_IN]], [10 x float]** [[B_ADDR]], 261 // TCHECK: store i{{[0-9]+}} [[BN_SZ]], i{{[0-9]+}}* [[VLA_ADDR]], 262 // TCHECK: store float* [[BN_IN]], float** [[BN_ADDR]], 263 // TCHECK: store [5 x [10 x double]]* [[C_IN]], [5 x [10 x double]]** [[C_ADDR]], 264 // TCHECK: store i{{[0-9]+}} [[CN_SZ1]], i{{[0-9]+}}* [[VLA_ADDR2]], 265 // TCHECK: store i{{[0-9]+}} [[CN_SZ2]], i{{[0-9]+}}* [[VLA_ADDR4]], 266 // TCHECK: store double* [[CN_IN]], double** [[CN_ADDR]], 267 // TCHECK: store [[TT]]* [[D_IN]], [[TT]]** [[D_ADDR]], 268 // TCHECK: [[CONV_A2ADDR:%.+]] = bitcast i{{[0-9]+}}* [[A2_ADDR]] to i{{[0-9]+}}* 269 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x float]*, [10 x float]** [[B_ADDR]], 270 // TCHECK: [[BN_SZ_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]], 271 // TCHECK: [[BN_ADDR_REF:%.+]] = load float*, float** [[BN_ADDR]], 272 // TCHECK: [[C_ADDR_REF:%.+]] = load [5 x [10 x double]]*, [5 x [10 x double]]** [[C_ADDR]], 273 // TCHECK: [[CN_SZ1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]], 274 // TCHECK: [[CN_SZ2_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR4]], 275 // TCHECK: [[CN_ADDR_REF:%.+]] = load double*, double** [[CN_ADDR]], 276 // TCHECK: [[D_ADDR_REF:%.+]] = load [[TT]]*, [[TT]]** [[D_ADDR]], 277 278 // firstprivate(aa): a_priv = a_in 279 // TCHECK-NOT: store i{{[0-9]+}} % 280 281 // firstprivate(b): memcpy(b_priv,b_in) 282 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x float]* [[B_PRIV]] to i8* 283 // TCHECK: [[B_ADDR_REF_BCAST:%.+]] = bitcast [10 x float]* [[B_ADDR_REF]] to i8* 284 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_ADDR_REF_BCAST]], {{.+}}) 285 286 // TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave() 287 // TCHECK: store i8* [[RET_STACK]], i8** [[SSTACK]], 288 289 // firstprivate(bn) 290 // TCHECK: [[BN_PRIV:%.+]] = alloca float, i{{[0-9]+}} [[BN_SZ_VAL]], 291 // TCHECK: [[BN_COPY_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[BN_SZ_VAL]], 4 292 // TCHECK: [[BN_PRIV__BCAST:%.+]] = bitcast float* [[BN_PRIV]] to i8* 293 // TCHECK: [[BN_REF_IN_BCAST:%.+]] = bitcast float* [[BN_ADDR_REF]] to i8* 294 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BN_PRIV__BCAST]], i8* align {{[0-9]+}} [[BN_REF_IN_BCAST]], i{{[0-9]+}} [[BN_COPY_SZ]],{{.+}}) 295 296 // firstprivate(c) 297 // TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_PRIV]] to i8* 298 // TCHECK: [[C_IN_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_ADDR_REF]] to i8* 299 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[C_PRIV_BCAST]], i8* align {{[0-9]+}} [[C_IN_BCAST]],{{.+}}) 300 301 // firstprivate(cn) 302 // TCHECK: [[CN_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ1_VAL]], [[CN_SZ2_VAL]] 303 // TCHECK: [[CN_PRIV:%.+]] = alloca double, i{{[0-9]+}} [[CN_SZ]], 304 // TCHECK: [[CN_SZ2:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ1_VAL]], [[CN_SZ2_VAL]] 305 // TCHECK: [[CN_SZ2_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ2]], 8 306 // TCHECK: [[CN_PRIV_BCAST:%.+]] = bitcast double* [[CN_PRIV]] to i8* 307 // TCHECK: [[CN_IN_BCAST:%.+]] = bitcast double* [[CN_ADDR_REF]] to i8* 308 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[CN_PRIV_BCAST]], i8* align {{[0-9]+}} [[CN_IN_BCAST]], i{{[0-9]+}} [[CN_SZ2_CPY]],{{.+}}) 309 310 // firstprivate(d) 311 // TCHECK: [[D_PRIV_BCAST:%.+]] = bitcast [[TT]]* [[D_PRIV]] to i8* 312 // TCHECK: [[D_IN_BCAST:%.+]] = bitcast [[TT]]* [[D_ADDR_REF]] to i8* 313 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[D_PRIV_BCAST]], i8* align {{[0-9]+}} [[D_IN_BCAST]],{{.+}}) 314 315 316 #pragma omp target firstprivate(ptr) 317 { 318 ptr[0]++; 319 } 320 // CHECK: [[PTR_ADDR_REF:%.+]] = load double*, double** [[PTR_ADDR]], 321 322 // CHECK: [[BASE_PTR_GEP3_0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 323 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP3_0]] to double** 324 // CHECK: store double* [[PTR_ADDR_REF]], double** [[BCAST_TOPTR]], 325 // CHECK: [[PTR_GEP3_0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 326 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP3_0]] to double** 327 // CHECK: store double* [[PTR_ADDR_REF]], double** [[BCAST_TOPTR]], 328 329 // CHECK: [[BASE_PTR_GEP_ARG3:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 330 // CHECK: [[PTR_GEP_ARG3:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 331 // CHECK: {{.+}} = call i32 @__tgt_target(i64 -1, {{.+}}, i32 1, i8** [[BASE_PTR_GEP_ARG3]], i8** [[PTR_GEP_ARG3]], i[[SZ]]* getelementptr inbounds ([1 x i[[SZ]]], [1 x i[[SZ]]]* [[SIZET3]], i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[MAPT3]], i32 0, i32 0)) 332 333 // TCHECK: define weak void @__omp_offloading_{{.+}}(double* [[PTR_IN:%.+]]) 334 // TCHECK: [[PTR_ADDR:%.+]] = alloca double*, 335 // TCHECK-NOT: alloca double*, 336 // TCHECK: store double* [[PTR_IN]], double** [[PTR_ADDR]], 337 // TCHECK-NOT: store double* % 338 339 return a; 340 } 341 342 343 template<typename tx> 344 tx ftemplate(int n) { 345 tx a = 0; 346 tx b[10]; 347 348 #pragma omp target firstprivate(a,b) 349 { 350 a += 1; 351 b[2] += 1; 352 } 353 354 return a; 355 } 356 357 static 358 int fstatic(int n) { 359 int a = 0; 360 char aaa = 0; 361 int b[10]; 362 363 #pragma omp target firstprivate(a,aaa,b) 364 { 365 a += 1; 366 aaa += 1; 367 b[2] += 1; 368 } 369 370 return a; 371 } 372 373 // TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], i{{[0-9]+}} [[A3_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]]) 374 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, 375 // TCHECK: [[A3_ADDR:%.+]] = alloca i{{[0-9]+}}, 376 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*, 377 // TCHECK-NOT: alloca i{{[0-9]+}}, 378 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}], 379 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]], 380 // TCHECK: store i{{[0-9]+}} [[A3_IN]], i{{[0-9]+}}* [[A3_ADDR]], 381 // TCHECK: store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]], 382 // TCHECK-64: [[A_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}* 383 // TCHECK: [[A3_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A3_ADDR]] to i8* 384 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]], 385 386 // firstprivate(a): a_priv = a_in 387 388 // firstprivate(aaa) 389 // TCHECK-NOT: store i{{[0-9]+}} % 390 391 // firstprivate(b) 392 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8* 393 // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8* 394 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}}) 395 396 // TCHECK: ret void 397 398 struct S1 { 399 double a; 400 401 int r1(int n){ 402 int b = n+1; 403 short int c[2][n]; 404 405 #pragma omp target firstprivate(b,c) 406 { 407 this->a = (double)b + 1.5; 408 c[1][1] = ++a; 409 } 410 411 return c[1][1] + (int)b; 412 } 413 414 // on the host side, we first generate r1, then the static function and the template above 415 // CHECK: define{{.+}} i32 {{.+}}([[S1]]* {{.+}}, i{{[0-9]+}} {{.+}}) 416 // CHECK: [[BASE_PTRS4:%.+]] = alloca [6 x i8*], 417 // CHECK: [[PTRS4:%.+]] = alloca [6 x i8*], 418 // CHECK: [[SIZET4:%.+]] = alloca [6 x i{{[0-9]+}}], 419 420 // map(this: this ptr is implicitly captured (not firstprivate matter) 421 // CHECK: [[BP0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 422 // CHECK: [[CBP0:%.+]] = bitcast i8** [[BP0]] to %struct.S1** 423 // CHECK: store %struct.S1* [[THIS:%.+]], %struct.S1** [[CBP0]], 424 // CHECK: [[P0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 425 // CHECK: [[CP0:%.+]] = bitcast i8** [[P0]] to double** 426 // CHECK: store double* [[A:%.+]], double** [[CP0]], 427 // CHECK: [[SZ0:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 428 // CHECK: store i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}* [[SZ0]], 429 430 // CHECK: [[BP1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 431 // CHECK: [[CBP1:%.+]] = bitcast i8** [[BP1]] to %struct.S1** 432 // CHECK: store %struct.S1* [[THIS]], %struct.S1** [[CBP1]], 433 // CHECK: [[P1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 434 // CHECK: [[CP1:%.+]] = bitcast i8** [[P1]] to double** 435 // CHECK: store double* [[A]], double** [[CP1]], 436 // CHECK: [[SZ1:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 437 // CHECK: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SZ1]], 438 439 // firstprivate(b): base_ptr = b, ptr = b, size = 4 (pass by-value) 440 // CHECK: [[BASE_PTRS_GEP4_1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 441 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_1]] to i{{[0-9]+}}* 442 // CHECK: store i{{[0-9]+}} [[B_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], 443 // CHECK: [[PTRS_GEP4_1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 444 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_1]] to i{{[0-9]+}}* 445 // CHECK: store i{{[0-9]+}} [[B_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]], 446 // CHECK: [[SIZES_GEP4_1:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 447 // CHECK: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_1]], 448 449 // firstprivate(c), 3 entries: 2, n, c 450 // CHECK: [[BASE_PTRS_GEP4_2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 451 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_2]] to i{{[0-9]+}}* 452 // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[BCAST_TOPTR]], 453 // CHECK: [[PTRS_GEP4_2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 454 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_2]] to i{{[0-9]+}}* 455 // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[BCAST_TOPTR]], 456 // CHECK: [[SIZES_GEP4_2:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 457 // CHECK-64: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SIZES_GEP4_2]], 458 // CHECK-32: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_2]], 459 // CHECK: [[BASE_PTRS_GEP4_3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 460 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_3]] to i{{[0-9]+}}* 461 // CHECK: store i{{[0-9]+}} [[N:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], 462 // CHECK: [[PTRS_GEP4_3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 463 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_3]] to i{{[0-9]+}}* 464 // CHECK: store i{{[0-9]+}} [[N]], i{{[0-9]+}}* [[BCAST_TOPTR]], 465 // CHECK: [[SIZES_GEP4_3:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 466 // CHECK-64: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SIZES_GEP4_3]], 467 // CHECK-32: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_3]], 468 // CHECK: [[BASE_PTRS_GEP4_4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 469 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_4]] to i{{[0-9]+}}** 470 // CHECK: store i{{[0-9]+}}* [[B:%.+]], i{{[0-9]+}}** [[BCAST_TOPTR]], 471 // CHECK: [[PTRS_GEP4_4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 472 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_4]] to i{{[0-9]+}}** 473 // CHECK: store i{{[0-9]+}}* [[B]], i{{[0-9]+}}** [[BCAST_TOPTR]], 474 // CHECK: [[SIZES_GEP4_4:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 475 // CHECK: store i{{[0-9]+}} [[B_SIZE:%.+]], i{{[0-9]+}}* [[SIZES_GEP4_4]], 476 477 // only check that we use the map types stored in the global variable 478 // CHECK: call i32 @__tgt_target(i64 -1, {{.+}}, i32 6, i8** {{.+}}, i8** {{.+}}, i{{[0-9]+}}* {{.+}}, i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT4]], i32 0, i32 0)) 479 480 // TCHECK: define weak void @__omp_offloading_{{.+}}([[S1]]* [[TH:%.+]], i{{[0-9]+}} [[B_IN:%.+]], i{{[0-9]+}} [[VLA:%.+]], i{{[0-9]+}} [[VLA1:%.+]], i{{[0-9]+}}{{.+}} [[C_IN:%.+]]) 481 // TCHECK: [[TH_ADDR:%.+]] = alloca [[S1]]*, 482 // TCHECK: [[B_ADDR:%.+]] = alloca i{{[0-9]+}}, 483 // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}}, 484 // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}}, 485 // TCHECK: [[C_ADDR:%.+]] = alloca i{{[0-9]+}}*, 486 // TCHECK-NOT: alloca i{{[0-9]+}}, 487 // TCHECK: [[SSTACK:%.+]] = alloca i8*, 488 489 // TCHECK: store [[S1]]* [[TH]], [[S1]]** [[TH_ADDR]], 490 // TCHECK: store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[B_ADDR]], 491 // TCHECK: store i{{[0-9]+}} [[VLA]], i{{[0-9]+}}* [[VLA_ADDR]], 492 // TCHECK: store i{{[0-9]+}} [[VLA1]], i{{[0-9]+}}* [[VLA_ADDR2]], 493 // TCHECK: store i{{[0-9]+}}* [[C_IN]], i{{[0-9]+}}** [[C_ADDR]], 494 // TCHECK: [[TH_ADDR_REF:%.+]] = load [[S1]]*, [[S1]]** [[TH_ADDR]], 495 // TCHECK-64: [[B_ADDR_CONV:%.+]] = bitcast i{{[0-9]+}}* [[B_ADDR]] to i{{[0-9]+}}* 496 // TCHECK: [[VLA_ADDR_REF:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]], 497 // TCHECK: [[VLA_ADDR_REF2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]], 498 // TCHECK: [[C_ADDR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[C_ADDR]], 499 500 // firstprivate(b) 501 // TCHECK-NOT: store i{{[0-9]+}} % 502 503 // TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave() 504 // TCHECK: store i8* [[RET_STACK:%.+]], i8** [[SSTACK]], 505 506 // firstprivate(c) 507 // TCHECK: [[C_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF]], [[VLA_ADDR_REF2]] 508 // TCHECK: [[C_PRIV:%.+]] = alloca i{{[0-9]+}}, i{{[0-9]+}} [[C_SZ]], 509 // TCHECK: [[C_SZ2:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF]], [[VLA_ADDR_REF2]] 510 // TCHECK: [[C_SZ_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[C_SZ2]], 2 511 // TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_PRIV]] to i8* 512 // TCHECK: [[C_IN_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_ADDR_REF]] to i8* 513 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[C_PRIV_BCAST]], i8* align {{[0-9]+}} [[C_IN_BCAST]],{{.+}}) 514 515 // finish 516 // TCHECK: [[RELOAD_SSTACK:%.+]] = load i8*, i8** [[SSTACK]], 517 // TCHECK: call void @llvm.stackrestore(i8* [[RELOAD_SSTACK]]) 518 // TCHECK: ret void 519 520 521 // static host function 522 // CHECK: define{{.+}} i32 {{.+}}(i{{[0-9]+}} {{.+}}) 523 // CHECK: [[BASE_PTRS5:%.+]] = alloca [3 x i8*], 524 // CHECK: [[PTRS5:%.+]] = alloca [3 x i8*], 525 526 // firstprivate(a): by value 527 // CHECK: [[BASE_PTRS_GEP5_0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 528 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_0]] to i{{[0-9]+}}* 529 // CHECK: store i{{[0-9]+}} [[A_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], 530 // CHECK: [[PTRS_GEP5_0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 531 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_0]] to i{{[0-9]+}}* 532 // CHECK: store i{{[0-9]+}} [[A_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]], 533 534 // firstprivate(aaa): by value 535 // CHECK: [[BASE_PTRS_GEP5_1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 536 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_1]] to i{{[0-9]+}}* 537 // CHECK: store i{{[0-9]+}} [[A3_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], 538 // CHECK: [[PTRS_GEP5_1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 539 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_1]] to i{{[0-9]+}}* 540 // CHECK: store i{{[0-9]+}} [[A3_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]], 541 542 // firstprivate(b): base_ptr = &b[0], ptr= &b[0] 543 // CHECK: [[BASE_PTRS_GEP5_2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 544 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_2]] to [10 x i{{[0-9]+}}]** 545 // CHECK: store [10 x i{{[0-9]+}}]* [[B:%.+]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]], 546 // CHECK: [[PTRS_GEP5_2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 547 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_2]] to [10 x i{{[0-9]+}}]** 548 // CHECK: store [10 x i{{[0-9]+}}]* [[B]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]], 549 550 // only check that the right sizes and map types are used 551 // CHECK: call i32 @__tgt_target(i64 -1, {{.+}}, i32 3, i8** {{.+}}, i8** {{.+}}, i[[SZ]]* getelementptr inbounds ([3 x i[[SZ]]], [3 x i[[SZ]]]* [[SIZET5]], i32 0, i32 0), i64* getelementptr inbounds ([3 x i64], [3 x i64]* [[MAPT5]], i32 0, i32 0)) 552 }; 553 554 555 556 int bar(int n, double *ptr){ 557 int a = 0; 558 a += foo(n, ptr); 559 S1 S; 560 a += S.r1(n); 561 a += fstatic(n); 562 a += ftemplate<int>(n); 563 564 return a; 565 } 566 567 // template host and device 568 569 // CHECK: define{{.+}} i32 {{.+}}(i{{[0-9]+}} {{.+}}) 570 // CHECK: [[BASE_PTRS6:%.+]] = alloca [2 x i8*], 571 // CHECK: [[PTRS6:%.+]] = alloca [2 x i8*], 572 573 // firstprivate(a): by value 574 // CHECK: [[BASE_PTRS_GEP6_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 575 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP6_0]] to i{{[0-9]+}}* 576 // CHECK: store i{{[0-9]+}} [[AT_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], 577 // CHECK: [[PTRS_GEP6_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 578 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP6_0]] to i{{[0-9]+}}* 579 // CHECK: store i{{[0-9]+}} [[AT_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]], 580 581 // firstprivate(b): pointer 582 // CHECK: [[BASE_PTRS_GEP6_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 583 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP6_1]] to [10 x i{{[0-9]+}}]** 584 // CHECK: store [10 x i{{[0-9]+}}]* [[B:%.+]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]], 585 // CHECK: [[PTRS_GEP6_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 586 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP6_1]] to [10 x i{{[0-9]+}}]** 587 // CHECK: store [10 x i{{[0-9]+}}]* [[B]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]], 588 589 // CHECK: call i32 @__tgt_target(i64 -1, {{.+}}, i32 2, i8** {{.+}}, i8** {{.+}}, i[[SZ]]* getelementptr inbounds ([2 x i[[SZ]]], [2 x i[[SZ]]]* [[SIZET6]], i32 0, i32 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT6]], i32 0, i32 0)) 590 591 // TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]]) 592 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, 593 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*, 594 // TCHECK-NOT: alloca i{{[0-9]+}}, 595 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}], 596 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]], 597 // TCHECK: store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]], 598 // TCHECK-64: [[A_ADDR_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}* 599 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]], 600 601 // firstprivate(a) 602 // TCHECK-NOT: store i{{[0-9]+}} % 603 604 // firstprivate(b) 605 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8* 606 // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8* 607 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}}) 608 609 // TCHECK: ret void 610 611 #endif 612