1 // Test host codegen. 2 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fomptargets=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 -fomptargets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s 4 // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fomptargets=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 -fomptargets=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 -fomptargets=i386-pc-linux-gnu -emit-pch -o %t %s 7 // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fomptargets=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 -x c++ -triple powerpc64le-unknown-unknown -fomptargets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc 11 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fomptargets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fomp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 12 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fomptargets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fomp-host-ir-file-path %t-ppc-host.bc -o %t %s 13 // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fomptargets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fomp-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 -x c++ -triple i386-unknown-unknown -fomptargets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc 15 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fomptargets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fomp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 16 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fomptargets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fomp-host-ir-file-path %t-x86-host.bc -o %t %s 17 // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fomptargets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fomp-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 template<typename tx, typename ty> 24 struct TT{ 25 tx X; 26 ty Y; 27 }; 28 29 // CHECK: [[TT:%.+]] = type { i64, i8 } 30 // CHECK: [[S1:%.+]] = type { double } 31 32 // TCHECK: [[TT:%.+]] = type { i64, i8 } 33 // TCHECK: [[S1:%.+]] = type { double } 34 35 // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] [i[[SZ:32|64]] 4] 36 // CHECK: [[MAPT:@.+]] = private unnamed_addr constant [1 x i32] [i32 128] 37 // CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [9 x i32] [i32 128, i32 3, i32 128, i32 3, i32 3, i32 128, i32 128, i32 3, i32 3] 38 // CHECK-64-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] [i[[SZ]] 8] 39 // CHECK-32-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [1 x i32] [i[[SZ]] 4] 40 // CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [1 x i32] [i32 128] 41 // CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [5 x i32] [i32 3, i32 128, i32 128, i32 128, i32 3] 42 // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 1, i[[SZ]] 40] 43 // CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i32] [i32 128, i32 128, i32 3] 44 // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [2 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 40] 45 // CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [2 x i32] [i32 128, i32 3] 46 47 48 // CHECK: define {{.*}}[[FOO:@.+]]( 49 int foo(int n, double *ptr) { 50 int a = 0; 51 short aa = 0; 52 float b[10]; 53 float bn[n]; 54 double c[5][10]; 55 double cn[5][n]; 56 TT<long long, char> d; 57 58 #pragma omp target firstprivate(a) 59 { 60 } 61 62 // a is passed by value to tgt_target 63 // CHECK: [[N_ADDR:%.+]] = alloca i{{[0-9]+}}, 64 // CHECK: [[PTR_ADDR:%.+]] = alloca double*, 65 // CHECK: [[A:%.+]] = alloca i{{[0-9]+}}, 66 // CHECK: [[A2:%.+]] = alloca i{{[0-9]+}}, 67 // CHECK: [[B:%.+]] = alloca [10 x float], 68 // CHECK: [[SSTACK:%.+]] = alloca i8*, 69 // CHECK: [[C:%.+]] = alloca [5 x [10 x double]], 70 // CHECK: [[D:%.+]] = alloca [[TT]], 71 // CHECK: [[ACAST:%.+]] = alloca i{{[0-9]+}}, 72 // CHECK: {{.+}} = alloca i{{[0-9]+}}, 73 // CHECK: [[BASE_PTR_ARR:%.+]] = alloca [1 x i8*], 74 // CHECK: [[PTR_ARR:%.+]] = alloca [1 x i8*], 75 // CHECK: [[A2CAST:%.+]] = alloca i{{[0-9]+}}, 76 // CHECK: [[BASE_PTR_ARR2:%.+]] = alloca [9 x i8*], 77 // CHECK: [[PTR_ARR2:%.+]] = alloca [9 x i8*], 78 // CHECK: [[SIZET2:%.+]] = alloca [9 x i{{[0-9]+}}], 79 // CHECK: [[BASE_PTR_ARR3:%.+]] = alloca [1 x i8*], 80 // CHECK: [[PTR_ARR3:%.+]] = alloca [1 x i8*], 81 // CHECK: [[N_ADDR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[N_ADDR]], 82 // CHECK-64: [[N_EXT:%.+]] = zext i{{[0-9]+}} [[N_ADDR_VAL]] to i{{[0-9]+}} 83 // CHECK: [[SSAVE_RET:%.+]] = call i8* @llvm.stacksave() 84 // CHECK: store i8* [[SSAVE_RET]], i8** [[SSTACK]], 85 // CHECK-64: [[BN_VLA:%.+]] = alloca float, i{{[0-9]+}} [[N_EXT]], 86 // CHECK-32: [[BN_VLA:%.+]] = alloca float, i{{[0-9]+}} [[N_ADDR_VAL]], 87 // CHECK: [[N_ADDR_VAL2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[N_ADDR]], 88 // CHECK-64: [[N_EXT2:%.+]] = zext i{{[0-9]+}} [[N_ADDR_VAL2]] to i{{[0-9]+}} 89 // CHECK-64: [[CN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_EXT2]] 90 // CHECK-32: [[CN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_ADDR_VAL2]] 91 // CHECK: [[CN_VLA:%.+]] = alloca double, i{{[0-9]+}} [[CN_SIZE]], 92 // CHECK: [[AVAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A]], 93 // CHECK-64: [[CONV:%.+]] = bitcast i{{[0-9]+}}* [[ACAST]] to i{{[0-9]+}}* 94 // CHECK-64: store i{{[0-9]+}} [[AVAL]], i{{[0-9]+}}* [[CONV]], 95 // CHECK-32: store i{{[0-9]+}} [[AVAL]], i{{[0-9]+}}* [[ACAST]], 96 // CHECK: [[ACAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[ACAST]], 97 // CHECK: [[ACAST_TOPTR:%.+]] = inttoptr i{{[0-9]+}} [[ACAST_VAL]] to i8* 98 // CHECK: [[BASE_PTR_GEP:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 99 // CHECK: store i8* [[ACAST_TOPTR]], i8** [[BASE_PTR_GEP]], 100 // CHECK: [[ACAST_TOPTR2:%.+]] = inttoptr i{{[0-9]+}} [[ACAST_VAL]] to i8* 101 // CHECK: [[PTR_GEP:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 102 // CHECK: store i8* [[ACAST_TOPTR2]], i8** [[PTR_GEP]], 103 // CHECK: [[BASE_PTR_GEP_ARG:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 104 // CHECK: [[PTR_GEP_ARG:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 105 // CHECK: {{.+}} = call i32 @__tgt_target(i32 -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), i32* getelementptr inbounds ([1 x i32], [1 x i32]* [[MAPT]], i32 0, i32 0)) 106 107 // TCHECK: define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]]) 108 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, 109 // TCHECK-NOT: alloca i{{[0-9]+}}, 110 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]], 111 // TCHECK-NOT: store i{{[0-9]+}} % 112 // TCHECK: ret void 113 114 #pragma omp target firstprivate(aa,b,bn,c,cn,d) 115 { 116 aa += 1; 117 b[2] = 1.0; 118 bn[3] = 1.0; 119 c[1][2] = 1.0; 120 cn[1][3] = 1.0; 121 d.X = 1; 122 d.Y = 1; 123 } 124 125 // CHECK: [[A2VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A2]], 126 // CHECK: [[A2CASTCONV:%.+]] = bitcast i{{[0-9]+}}* [[A2CAST]] to i{{[0-9]+}}* 127 // CHECK: store i{{[0-9]+}} [[A2VAL]], i{{[0-9]+}}* [[A2CASTCONV]], 128 // CHECK: [[A2CAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A2CAST]], 129 // CHECK-64: [[BN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} [[N_EXT]], 4 130 // CHECK-32: [[BN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} [[N_ADDR_VAL]], 4 131 // CHECK-64: [[CN_SIZE_1:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_EXT2]] 132 // CHECK-32: [[CN_SIZE_1:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_ADDR_VAL2]] 133 // CHECK: [[CN_SIZE_2:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SIZE_1]], 8 134 135 // firstprivate(aa) --> base_ptr = aa, ptr = aa, size = 2 (short) 136 // CHECK: [[A2CAST_TO_INT:%.+]] = inttoptr i{{[0-9]+}} [[A2CAST_VAL]] to i8* 137 // CHECK: [[BASE_PTR_GEP2_0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 138 // CHECK: store i8* [[A2CAST_TO_INT]], i8** [[BASE_PTR_GEP2_0]], 139 // CHECK: [[A2CAST_TO_INT_2:%.+]] = inttoptr i{{[0-9]+}} [[A2CAST_VAL]] to i8* 140 // CHECK: [[PTR_GEP2_0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 141 // CHECK: store i8* [[A2CAST_TO_INT_2]], i8** [[PTR_GEP2_0]], 142 // CHECK: [[SIZE_GEPA2:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 143 // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[SIZE_GEPA2]], 144 145 // firstprivate(b): base_ptr = &b[0], ptr = &b[0], size = 40 (sizeof(float)*10) 146 // CHECK: [[BCAST:%.+]] = bitcast [10 x float]* [[B]] to i8* 147 // CHECK: [[BASE_PTR_GEP2_1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 148 // CHECK: store i8* [[BCAST]], i8** [[BASE_PTR_GEP2_1]], 149 // CHECK: [[BCAST2:%.+]] = bitcast [10 x float]* [[B]] to i8* 150 // CHECK: [[PTR_GEP2_1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 151 // CHECK: store i8* [[BCAST2]], i8** [[PTR_GEP2_1]], 152 // CHECK: [[SIZE_GEPB:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 153 // CHECK: store i{{[0-9]+}} 40, i{{[0-9]+}}* [[SIZE_GEPB]], 154 155 // 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) 156 // CHECK-64: [[N_EXT3_1:%.+]] = inttoptr i{{[0-9]+}} [[N_EXT]] to i8* 157 // CHECK-32: [[N_EXT3_1:%.+]] = inttoptr i{{[0-9]+}} [[N_ADDR_VAL]] to i8* 158 // CHECK: [[BASE_PTR_GEP2_2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 159 // CHECK: store i8* [[N_EXT3_1]], i8** [[BASE_PTR_GEP2_2]], 160 // CHECK-64: [[N_EXT3_2:%.+]] = inttoptr i{{[0-9]+}} [[N_EXT]] to i8* 161 // CHECK-32: [[N_EXT3_2:%.+]] = inttoptr i{{[0-9]+}} [[N_ADDR_VAL]] to i8* 162 // CHECK: [[PTR_GEP2_2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 163 // CHECK: store i8* [[N_EXT3_2]], i8** [[PTR_GEP2_2]], 164 // CHECK: [[SIZE_GEPBN_1:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 165 // CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPBN_1]], 166 // CHECK: [[VLABN_BCAST:%.+]] = bitcast float* [[BN_VLA]] to i8* 167 // CHECK: [[BASE_PTR_GEP2_3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 168 // CHECK: store i8* [[VLABN_BCAST]], i8** [[BASE_PTR_GEP2_3]], 169 // CHECK: [[SIZE_GEPBN_3:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 170 // CHECK: store i{{[0-9]+}} [[BN_SIZE]], i{{[0-9]+}}* [[SIZE_GEPBN_3]] 171 172 // firstprivate(c): base_ptr = &c[0], ptr = &c[0], size = 400 (5*10*sizeof(double)) 173 // CHECK: [[C_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C]] to i8* 174 // CHECK: [[BASE_PTR_GEP2_4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 175 // CHECK: store i8* [[C_BCAST]], i8** [[BASE_PTR_GEP2_4]], 176 // CHECK: [[C_BCAST2:%.+]] = bitcast [5 x [10 x double]]* [[C]] to i8* 177 // CHECK: [[PTR_GEP2_4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 178 // CHECK: store i8* [[C_BCAST2]], i8** [[PTR_GEP2_4]], 179 // CHECK: [[SIZE_GEPC_4:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 180 // CHECK: store i{{[0-9]+}} 400, i{{[0-9]+}}* [[SIZE_GEPC_4]], 181 182 // 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) 183 // CHECK: [[BASE_PTR_GEP2_5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 184 // CHECK: store i8* inttoptr (i{{[0-9]+}} 5 to i8*), i8** [[BASE_PTR_GEP2_5]], 185 // CHECK: [[PTR_GEP2_5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 186 // CHECK: store i8* inttoptr (i{{[0-9]+}} 5 to i8*), i8** [[PTR_GEP2_5]], 187 // CHECK: [[SIZE_GEPCN_5:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 188 // CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPCN_5]], 189 // CHECK-64: [[CN_SZ_2_1:%.+]] = inttoptr i{{[0-9]+}} [[N_EXT2]] to i8* 190 // CHECK-32: [[CN_SZ_2_1:%.+]] = inttoptr i{{[0-9]+}} [[N_ADDR_VAL2]] to i8* 191 // CHECK: [[BASE_PTR_GEP2_6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6 192 // CHECK: store i8* [[CN_SZ_2_1]], i8** [[BASE_PTR_GEP2_6]], 193 // CHECK-64: [[CN_SZ_2_2:%.+]] = inttoptr i{{[0-9]+}} [[N_EXT2]] to i8* 194 // CHECK-32: [[CN_SZ_2_2:%.+]] = inttoptr i{{[0-9]+}} [[N_ADDR_VAL2]] to i8* 195 // CHECK: [[PTR_GEP2_6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6 196 // CHECK: store i8* [[CN_SZ_2_2]], i8** [[PTR_GEP2_6]], 197 // CHECK: [[SIZE_GEPCN_6:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6 198 // CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPCN_6]], 199 // CHECK: [[VLA_CN_BCAST:%.+]] = bitcast double* [[CN_VLA]] to i8* 200 // CHECK: [[BASE_PTR_GEP2_7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7 201 // CHECK: store i8* [[VLA_CN_BCAST]], i8** [[BASE_PTR_GEP2_7]], 202 // CHECK: [[VLA_CN_BCAST2:%.+]] = bitcast double* [[CN_VLA]] to i8* 203 // CHECK: [[PTR_GEP2_7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7 204 // CHECK: store i8* [[VLA_CN_BCAST2]], i8** [[PTR_GEP2_7]], 205 // CHECK: [[SIZE_GEPCN_7:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7 206 // CHECK: store i{{[0-9]+}} [[CN_SIZE_2]], i{{[0-9]+}}* [[SIZE_GEPCN_7]], 207 208 // firstprivate(d): base_ptr = &d, ptr = &d, size = 16 209 // CHECK: [[D_REF:%.+]] = bitcast [[TT]]* [[D]] to i8* 210 // CHECK: [[BASE_PTR_GEP2_8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8 211 // CHECK: store i8* [[D_REF]], i8** [[BASE_PTR_GEP2_8]], 212 // CHECK: [[D_REF2:%.+]] = bitcast [[TT]]* [[D]] to i8* 213 // CHECK: [[PTR_GEP2_8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8 214 // CHECK: store i8* [[D_REF2]], i8** [[PTR_GEP2_8]], 215 // CHECK: [[SIZE_GEPCN_8:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8 216 // CHECK: store i{{[0-9]+}} {{[0-9]+}}, i{{[0-9]+}}* [[SIZE_GEPCN_8]], 217 218 219 // CHECK: [[BASE_PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 220 // CHECK: [[PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 221 // CHECK: [[SIZES_ARG2:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 222 // CHECK: {{.+}} = call i32 @__tgt_target(i32 -1, {{.+}}, i32 9, i8** [[BASE_PTR_GEP_ARG2]], i8** [[PTR_GEP_ARG2]], i[[SZ]]* [[SIZES_ARG2]], i32* getelementptr inbounds ([9 x i32], [9 x i32]* [[MAPT2]], i32 0, i32 0)) 223 224 // make sure that firstprivate variables are generated in all cases and that we use those instances for operations inside the 225 // target region 226 // 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:%.+]]) 227 // TCHECK: [[A2_ADDR:%.+]] = alloca i{{[0-9]+}}, 228 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x float]*, 229 // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}}, 230 // TCHECK: [[BN_ADDR:%.+]] = alloca float*, 231 // TCHECK: [[C_ADDR:%.+]] = alloca [5 x [10 x double]]*, 232 // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}}, 233 // TCHECK: [[VLA_ADDR4:%.+]] = alloca i{{[0-9]+}}, 234 // TCHECK: [[CN_ADDR:%.+]] = alloca double*, 235 // TCHECK: [[D_ADDR:%.+]] = alloca [[TT]]*, 236 // TCHECK-NOT: alloca i{{[0-9]+}}, 237 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x float], 238 // TCHECK: [[SSTACK:%.+]] = alloca i8*, 239 // TCHECK: [[C_PRIV:%.+]] = alloca [5 x [10 x double]], 240 // TCHECK: [[D_PRIV:%.+]] = alloca [[TT]], 241 // TCHECK: store i{{[0-9]+}} [[A2_IN]], i{{[0-9]+}}* [[A2_ADDR]], 242 // TCHECK: store [10 x float]* [[B_IN]], [10 x float]** [[B_ADDR]], 243 // TCHECK: store i{{[0-9]+}} [[BN_SZ]], i{{[0-9]+}}* [[VLA_ADDR]], 244 // TCHECK: store float* [[BN_IN]], float** [[BN_ADDR]], 245 // TCHECK: store [5 x [10 x double]]* [[C_IN]], [5 x [10 x double]]** [[C_ADDR]], 246 // TCHECK: store i{{[0-9]+}} [[CN_SZ1]], i{{[0-9]+}}* [[VLA_ADDR2]], 247 // TCHECK: store i{{[0-9]+}} [[CN_SZ2]], i{{[0-9]+}}* [[VLA_ADDR4]], 248 // TCHECK: store double* [[CN_IN]], double** [[CN_ADDR]], 249 // TCHECK: store [[TT]]* [[D_IN]], [[TT]]** [[D_ADDR]], 250 // TCHECK: [[CONV_A2ADDR:%.+]] = bitcast i{{[0-9]+}}* [[A2_ADDR]] to i{{[0-9]+}}* 251 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x float]*, [10 x float]** [[B_ADDR]], 252 // TCHECK: [[BN_SZ_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]], 253 // TCHECK: [[BN_ADDR_REF:%.+]] = load float*, float** [[BN_ADDR]], 254 // TCHECK: [[C_ADDR_REF:%.+]] = load [5 x [10 x double]]*, [5 x [10 x double]]** [[C_ADDR]], 255 // TCHECK: [[CN_SZ1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]], 256 // TCHECK: [[CN_SZ2_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR4]], 257 // TCHECK: [[CN_ADDR_REF:%.+]] = load double*, double** [[CN_ADDR]], 258 // TCHECK: [[D_ADDR_REF:%.+]] = load [[TT]]*, [[TT]]** [[D_ADDR]], 259 260 // firstprivate(aa): a_priv = a_in 261 // TCHECK-NOT: store i{{[0-9]+}} % 262 263 // firstprivate(b): memcpy(b_priv,b_in) 264 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x float]* [[B_PRIV]] to i8* 265 // TCHECK: [[B_ADDR_REF_BCAST:%.+]] = bitcast [10 x float]* [[B_ADDR_REF]] to i8* 266 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[B_PRIV_BCAST]], i8* [[B_ADDR_REF_BCAST]], {{.+}}) 267 268 // TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave() 269 // TCHECK: store i8* [[RET_STACK]], i8** [[SSTACK]], 270 271 // firstprivate(bn) 272 // TCHECK: [[BN_PRIV:%.+]] = alloca float, i{{[0-9]+}} [[BN_SZ_VAL]], 273 // TCHECK: [[BN_COPY_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[BN_SZ_VAL]], 4 274 // TCHECK: [[BN_PRIV__BCAST:%.+]] = bitcast float* [[BN_PRIV]] to i8* 275 // TCHECK: [[BN_REF_IN_BCAST:%.+]] = bitcast float* [[BN_ADDR_REF]] to i8* 276 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[BN_PRIV__BCAST]], i8* [[BN_REF_IN_BCAST]], i{{[0-9]+}} [[BN_COPY_SZ]],{{.+}}) 277 278 // firstprivate(c) 279 // TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_PRIV]] to i8* 280 // TCHECK: [[C_IN_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_ADDR_REF]] to i8* 281 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[C_PRIV_BCAST]], i8* [[C_IN_BCAST]],{{.+}}) 282 283 // firstprivate(cn) 284 // TCHECK: [[CN_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ1_VAL]], [[CN_SZ2_VAL]] 285 // TCHECK: [[CN_PRIV:%.+]] = alloca double, i{{[0-9]+}} [[CN_SZ]], 286 // TCHECK: [[CN_SZ2:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ1_VAL]], [[CN_SZ2_VAL]] 287 // TCHECK: [[CN_SZ2_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ2]], 8 288 // TCHECK: [[CN_PRIV_BCAST:%.+]] = bitcast double* [[CN_PRIV]] to i8* 289 // TCHECK: [[CN_IN_BCAST:%.+]] = bitcast double* [[CN_ADDR_REF]] to i8* 290 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[CN_PRIV_BCAST]], i8* [[CN_IN_BCAST]], i{{[0-9]+}} [[CN_SZ2_CPY]],{{.+}}) 291 292 // firstprivate(d) 293 // TCHECK: [[D_PRIV_BCAST:%.+]] = bitcast [[TT]]* [[D_PRIV]] to i8* 294 // TCHECK: [[D_IN_BCAST:%.+]] = bitcast [[TT]]* [[D_ADDR_REF]] to i8* 295 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[D_PRIV_BCAST]], i8* [[D_IN_BCAST]],{{.+}}) 296 297 298 #pragma omp target firstprivate(ptr) 299 { 300 ptr[0]++; 301 } 302 // CHECK: [[PTR_ADDR_REF:%.+]] = load double*, double** [[PTR_ADDR]], 303 // CHECK: [[PTR_ADDR_BCAST:%.+]] = bitcast double* [[PTR_ADDR_REF]] to i8* 304 305 // CHECK: [[BASE_PTR_GEP3_0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 306 // CHECK: store i8* [[PTR_ADDR_BCAST]], i8** [[BASE_PTR_GEP3_0]], 307 // CHECK: [[PTR_ADDR_BCAST2:%.+]] = bitcast double* [[PTR_ADDR_REF]] to i8* 308 // CHECK: [[PTR_GEP3_0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 309 // CHECK: store i8* [[PTR_ADDR_BCAST2]], i8** [[PTR_GEP3_0]], 310 311 // CHECK: [[BASE_PTR_GEP_ARG3:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 312 // CHECK: [[PTR_GEP_ARG3:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 313 // CHECK: {{.+}} = call i32 @__tgt_target(i32 -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), i32* getelementptr inbounds ([1 x i32], [1 x i32]* [[MAPT3]], i32 0, i32 0)) 314 315 // TCHECK: define void @__omp_offloading_{{.+}}(double* [[PTR_IN:%.+]]) 316 // TCHECK: [[PTR_ADDR:%.+]] = alloca double*, 317 // TCHECK-NOT: alloca double*, 318 // TCHECK: store double* [[PTR_IN]], double** [[PTR_ADDR]], 319 // TCHECK-NOT: store double* % 320 321 return a; 322 } 323 324 325 template<typename tx> 326 tx ftemplate(int n) { 327 tx a = 0; 328 tx b[10]; 329 330 #pragma omp target firstprivate(a,b) 331 { 332 a += 1; 333 b[2] += 1; 334 } 335 336 return a; 337 } 338 339 static 340 int fstatic(int n) { 341 int a = 0; 342 char aaa = 0; 343 int b[10]; 344 345 #pragma omp target firstprivate(a,aaa,b) 346 { 347 a += 1; 348 aaa += 1; 349 b[2] += 1; 350 } 351 352 return a; 353 } 354 355 // TCHECK: define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], i{{[0-9]+}} [[A3_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]]) 356 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, 357 // TCHECK: [[A3_ADDR:%.+]] = alloca i{{[0-9]+}}, 358 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*, 359 // TCHECK-NOT: alloca i{{[0-9]+}}, 360 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}], 361 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]], 362 // TCHECK: store i{{[0-9]+}} [[A3_IN]], i{{[0-9]+}}* [[A3_ADDR]], 363 // TCHECK: store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]], 364 // TCHECK-64: [[A_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}* 365 // TCHECK: [[A3_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A3_ADDR]] to i8* 366 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]], 367 368 // firstprivate(a): a_priv = a_in 369 370 // firstprivate(aaa) 371 // TCHECK-NOT: store i{{[0-9]+}} % 372 373 // firstprivate(b) 374 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8* 375 // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8* 376 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[B_PRIV_BCAST]], i8* [[B_IN_BCAST]],{{.+}}) 377 378 // TCHECK: ret void 379 380 struct S1 { 381 double a; 382 383 int r1(int n){ 384 int b = n+1; 385 short int c[2][n]; 386 387 #pragma omp target firstprivate(b,c) 388 { 389 this->a = (double)b + 1.5; 390 c[1][1] = ++a; 391 } 392 393 return c[1][1] + (int)b; 394 } 395 396 // on the host side, we first generate r1, then the static function and the template above 397 // CHECK: define{{.+}} i32 {{.+}}([[S1]]* {{.+}}, i{{[0-9]+}} {{.+}}) 398 // CHECK: [[BASE_PTRS4:%.+]] = alloca [5 x i8*], 399 // CHECK: [[PTRS4:%.+]] = alloca [5 x i8*], 400 // CHECK: [[SIZET4:%.+]] = alloca [5 x i{{[0-9]+}}], 401 402 // map(this: this ptr is implicitly captured (not firstprivate matter) 403 // CHECK: {{.+}} = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 404 // CHECK: store {{.+}}, {{.+}}, 405 // CHECK: {{.+}} = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 406 // CHECK: store {{.+}}, {{.+}}, 407 // CHECK: {{.+}} getelementptr inbounds [5 x i{{[0-9]+}}], [5 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 408 // CHECK: store {{.+}}, {{.+}} 409 410 // firstprivate(b): base_ptr = b, ptr = b, size = 4 (pass by-value) 411 // CHECK: [[B_CAST_PTR:%.+]] = inttoptr i{{[0-9]+}} [[B_CAST:%.+]] to i8* 412 // CHECK: [[BASE_PTRS_GEP4_1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 413 // CHECK: store i8* [[B_CAST_PTR]], i8** [[BASE_PTRS_GEP4_1]], 414 // CHECK: [[B_CAST_PTR2:%.+]] = inttoptr i{{[0-9]+}} [[B_CAST:%.+]] to i8* 415 // CHECK: [[PTRS_GEP4_1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 416 // CHECK: store i8* [[B_CAST_PTR2]], i8** [[PTRS_GEP4_1]], 417 // CHECK: [[SIZES_GEP4_1:%.+]] = getelementptr inbounds [5 x i{{[0-9]+}}], [5 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 418 // CHECK: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_1]], 419 420 // firstprivate(c), 3 entries: 2, n, c 421 // CHECK: [[BASE_PTRS_GEP4_2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 422 // CHECK: store i8* inttoptr (i{{[0-9]+}} 2 to i8*), i8** [[BASE_PTRS_GEP4_2]], 423 // CHECK: [[PTRS_GEP4_2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 424 // CHECK: store i8* inttoptr (i{{[0-9]+}} 2 to i8*), i8** [[PTRS_GEP4_2]], 425 // CHECK: [[SIZES_GEP4_2:%.+]] = getelementptr inbounds [5 x i{{[0-9]+}}], [5 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 426 // CHECK-64: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SIZES_GEP4_2]], 427 // CHECK-32: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_2]], 428 // CHECK: [[N_PTR:%.+]] = inttoptr i{{[0-9]+}} [[N:%.+]] to i8* 429 // CHECK: [[BASE_PTRS_GEP4_3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 430 // CHECK: store i8* [[N_PTR]], i8** [[BASE_PTRS_GEP4_3]], 431 // CHECK: [[N_PTR2:%.+]] = inttoptr i{{[0-9]+}} [[N:%.+]] to i8* 432 // CHECK: [[PTRS_GEP4_3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 433 // CHECK: store i8* [[N_PTR2]], i8** [[PTRS_GEP4_3]], 434 // CHECK: [[SIZES_GEP4_3:%.+]] = getelementptr inbounds [5 x i{{[0-9]+}}], [5 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 435 // CHECK-64: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SIZES_GEP4_3]], 436 // CHECK-32: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_3]], 437 // CHECK: [[B_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[B:%.+]] to i8* 438 // CHECK: [[BASE_PTRS_GEP4_4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 439 // CHECK: store i8* [[B_BCAST]], i8** [[BASE_PTRS_GEP4_4]], 440 // CHECK: [[B_BCAST2:%.+]] = bitcast i{{[0-9]+}}* [[B:%.+]] to i8* 441 // CHECK: [[PTRS_GEP4_4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 442 // CHECK: store i8* [[B_BCAST2]], i8** [[PTRS_GEP4_4]], 443 // CHECK: [[SIZES_GEP4_4:%.+]] = getelementptr inbounds [5 x i{{[0-9]+}}], [5 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 444 // CHECK: store i{{[0-9]+}} [[B_SIZE:%.+]], i{{[0-9]+}}* [[SIZES_GEP4_4]], 445 446 // only check that we use the map types stored in the global variable 447 // CHECK: call i32 @__tgt_target(i32 -1, {{.+}}, i32 5, i8** {{.+}}, i8** {{.+}}, i{{[0-9]+}}* {{.+}}, i32* getelementptr inbounds ([5 x i32], [5 x i32]* [[MAPT4]], i32 0, i32 0)) 448 449 // TCHECK: define void @__omp_offloading_{{.+}}([[S1]]* [[TH:%.+]], i{{[0-9]+}} [[B_IN:%.+]], i{{[0-9]+}} [[VLA:%.+]], i{{[0-9]+}} [[VLA1:%.+]], i{{[0-9]+}}{{.+}} [[C_IN:%.+]]) 450 // TCHECK: [[TH_ADDR:%.+]] = alloca [[S1]]*, 451 // TCHECK: [[B_ADDR:%.+]] = alloca i{{[0-9]+}}, 452 // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}}, 453 // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}}, 454 // TCHECK: [[C_ADDR:%.+]] = alloca i{{[0-9]+}}*, 455 // TCHECK-NOT: alloca i{{[0-9]+}}, 456 // TCHECK: [[SSTACK:%.+]] = alloca i8*, 457 458 // TCHECK: store [[S1]]* [[TH]], [[S1]]** [[TH_ADDR]], 459 // TCHECK: store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[B_ADDR]], 460 // TCHECK: store i{{[0-9]+}} [[VLA]], i{{[0-9]+}}* [[VLA_ADDR]], 461 // TCHECK: store i{{[0-9]+}} [[VLA1]], i{{[0-9]+}}* [[VLA_ADDR2]], 462 // TCHECK: store i{{[0-9]+}}* [[C_IN]], i{{[0-9]+}}** [[C_ADDR]], 463 // TCHECK: [[TH_ADDR_REF:%.+]] = load [[S1]]*, [[S1]]** [[TH_ADDR]], 464 // TCHECK-64: [[B_ADDR_CONV:%.+]] = bitcast i{{[0-9]+}}* [[B_ADDR]] to i{{[0-9]+}}* 465 // TCHECK: [[VLA_ADDR_REF:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]], 466 // TCHECK: [[VLA_ADDR_REF2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]], 467 // TCHECK: [[C_ADDR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[C_ADDR]], 468 469 // firstprivate(b) 470 // TCHECK-NOT: store i{{[0-9]+}} % 471 472 // TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave() 473 // TCHECK: store i8* [[RET_STACK:%.+]], i8** [[SSTACK]], 474 475 // firstprivate(c) 476 // TCHECK: [[C_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF]], [[VLA_ADDR_REF2]] 477 // TCHECK: [[C_PRIV:%.+]] = alloca i{{[0-9]+}}, i{{[0-9]+}} [[C_SZ]], 478 // TCHECK: [[C_SZ2:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF]], [[VLA_ADDR_REF2]] 479 // TCHECK: [[C_SZ_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[C_SZ2]], 2 480 // TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_PRIV]] to i8* 481 // TCHECK: [[C_IN_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_ADDR_REF]] to i8* 482 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[C_PRIV_BCAST]], i8* [[C_IN_BCAST]],{{.+}}) 483 484 // finish 485 // TCHECK: [[RELOAD_SSTACK:%.+]] = load i8*, i8** [[SSTACK]], 486 // TCHECK: call void @llvm.stackrestore(i8* [[RELOAD_SSTACK]]) 487 // TCHECK: ret void 488 489 490 // static host function 491 // CHECK: define{{.+}} i32 {{.+}}(i{{[0-9]+}} {{.+}}) 492 // CHECK: [[BASE_PTRS5:%.+]] = alloca [3 x i8*], 493 // CHECK: [[PTRS5:%.+]] = alloca [3 x i8*], 494 495 // firstprivate(a): by value 496 // CHECK: [[A_CAST_PTR:%.+]] = inttoptr i{{[0-9]+}} [[A_CAST:%.+]] to i8* 497 // CHECK: [[BASE_PTRS_GEP5_0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 498 // CHECK: store i8* [[A_CAST_PTR]], i8** [[BASE_PTRS_GEP5_0]], 499 // CHECK: [[A_CAST_PTR2:%.+]] = inttoptr i{{[0-9]+}} [[A_CAST:%.+]] to i8* 500 // CHECK: [[PTRS_GEP5_0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 501 // CHECK: store i8* [[A_CAST_PTR2]], i8** [[PTRS_GEP5_0]], 502 503 // firstprivate(aaa): by value 504 // CHECK: [[A3_CAST_PTR:%.+]] = inttoptr i{{[0-9]+}} [[A3_CAST:%.+]] to i8* 505 // CHECK: [[BASE_PTRS_GEP5_1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 506 // CHECK: store i8* [[A3_CAST_PTR]], i8** [[BASE_PTRS_GEP5_1]], 507 // CHECK: [[A3_CAST_PTR2:%.+]] = inttoptr i{{[0-9]+}} [[A3_CAST:%.+]] to i8* 508 // CHECK: [[PTRS_GEP5_1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 509 // CHECK: store i8* [[A3_CAST_PTR2]], i8** [[PTRS_GEP5_1]], 510 511 // firstprivate(b): base_ptr = &b[0], ptr= &b[0] 512 // CHECK: [[B_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B:%.+]] to i8* 513 // CHECK: [[BASE_PTRS_GEP5_2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 514 // CHECK: store i8* [[B_BCAST]], i8** [[BASE_PTRS_GEP5_2]], 515 // CHECK: [[B_BCAST2:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B:%.+]] to i8* 516 // CHECK: [[PTRS_GEP5_2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 517 // CHECK: store i8* [[B_BCAST2]], i8** [[PTRS_GEP5_2]], 518 519 // only check that the right sizes and map types are used 520 // CHECK: call i32 @__tgt_target(i32 -1, {{.+}}, i32 3, i8** {{.+}}, i8** {{.+}}, 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)) 521 }; 522 523 524 525 int bar(int n, double *ptr){ 526 int a = 0; 527 a += foo(n, ptr); 528 S1 S; 529 a += S.r1(n); 530 a += fstatic(n); 531 a += ftemplate<int>(n); 532 533 return a; 534 } 535 536 // template host and device 537 538 // CHECK: define{{.+}} i32 {{.+}}(i{{[0-9]+}} {{.+}}) 539 // CHECK: [[BASE_PTRS6:%.+]] = alloca [2 x i8*], 540 // CHECK: [[PTRS6:%.+]] = alloca [2 x i8*], 541 542 // firstprivate(a): by value 543 // CHECK: [[AT_CAST_PTR:%.+]] = inttoptr i{{[0-9]+}} [[AT_CAST:%.+]] to i8* 544 // CHECK: [[BASE_PTRS_GEP6_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 545 // CHECK: store i8* [[AT_CAST_PTR]], i8** [[BASE_PTRS_GEP6_0]], 546 // CHECK: [[AT_CAST_PTR2:%.+]] = inttoptr i{{[0-9]+}} [[AT_CAST:%.+]] to i8* 547 // CHECK: [[PTRS_GEP6_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 548 // CHECK: store i8* [[AT_CAST_PTR2]], i8** [[PTRS_GEP6_0]], 549 550 // firstprivate(b): pointer 551 // CHECK: [[B_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B:%.+]] to i8* 552 // CHECK: [[BASE_PTRS_GEP6_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 553 // CHECK: store i8* [[B_BCAST]], i8** [[BASE_PTRS_GEP6_1]], 554 // CHECK: [[B_BCAST2:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B:%.+]] to i8* 555 // CHECK: [[PTRS_GEP6_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 556 // CHECK: store i8* [[B_BCAST2]], i8** [[PTRS_GEP6_1]], 557 558 // CHECK: call i32 @__tgt_target(i32 -1, {{.+}}, i32 2, i8** {{.+}}, i8** {{.+}}, i[[SZ]]* getelementptr inbounds ([2 x i[[SZ]]], [2 x i[[SZ]]]* [[SIZET6]], i32 0, i32 0), i32* getelementptr inbounds ([2 x i32], [2 x i32]* [[MAPT6]], i32 0, i32 0)) 559 560 561 // TCHECK: define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]]) 562 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, 563 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*, 564 // TCHECK-NOT: alloca i{{[0-9]+}}, 565 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}], 566 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]], 567 // TCHECK: store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]], 568 // TCHECK-64: [[A_ADDR_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}* 569 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]], 570 571 // firstprivate(a) 572 // TCHECK-NOT: store i{{[0-9]+}} % 573 574 // firstprivate(b) 575 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8* 576 // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8* 577 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[B_PRIV_BCAST]], i8* [[B_IN_BCAST]],{{.+}}) 578 579 // TCHECK: ret void 580 581 #endif 582