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-DAG: [[TT:%.+]] = type { i64, i8 } 48 // CHECK-DAG: [[TTII:%.+]] = type { i32, i32 } 49 // CHECK-DAG: [[S1:%.+]] = type { double } 50 51 // TCHECK-DAG: [[TT:%.+]] = type { i64, i8 } 52 // TCHECK-DAG: [[TTII:%.+]] = type { i32, i32 } 53 // TCHECK-DAG: [[S1:%.+]] = type { double } 54 55 // CHECK-DAG: [[FP_E:@__omp_offloading_firstprivate_.+_e_l76]] = internal global [[TTII]] zeroinitializer 56 // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i{{32|64}}] [i[[SZ:32|64]] 4, i{{64|32}} {{8|4}}] 57 // CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 561] 58 // 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] 59 // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i{{32|64}}] [i{{32|64}} 0, i{{32|64}} 8] 60 // CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 549] 61 // CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 673] 62 // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 1, i[[SZ]] 40] 63 // CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 673] 64 // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [2 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 40] 65 // CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 673] 66 67 // CHECK: define {{.*}}[[FOO:@.+]]( 68 int foo(int n, double *ptr) { 69 int a = 0; 70 short aa = 0; 71 float b[10]; 72 float bn[n]; 73 double c[5][10]; 74 double cn[5][n]; 75 TT<long long, char> d; 76 const TT<int, int> e = {n, n}; 77 int *p __attribute__ ((aligned (64))) = &a; 78 79 #pragma omp target firstprivate(a, p) 80 { 81 } 82 83 // a is passed by value to tgt_target 84 // CHECK: [[N_ADDR:%.+]] = alloca i{{[0-9]+}}, 85 // CHECK: [[PTR_ADDR:%.+]] = alloca double*, 86 // CHECK: [[A:%.+]] = alloca i{{[0-9]+}}, 87 // CHECK: [[A2:%.+]] = alloca i{{[0-9]+}}, 88 // CHECK: [[B:%.+]] = alloca [10 x float], 89 // CHECK: [[SSTACK:%.+]] = alloca i8*, 90 // CHECK: [[C:%.+]] = alloca [5 x [10 x double]], 91 // CHECK: [[D:%.+]] = alloca [[TT]], 92 // CHECK: [[P:%.+]] = alloca i32*, align 64 93 // CHECK: [[ACAST:%.+]] = alloca i{{[0-9]+}}, 94 // CHECK: [[BASE_PTR_ARR:%.+]] = alloca [2 x i8*], 95 // CHECK: [[PTR_ARR:%.+]] = alloca [2 x i8*], 96 // CHECK: [[A2CAST:%.+]] = alloca i{{[0-9]+}}, 97 // CHECK: [[BASE_PTR_ARR2:%.+]] = alloca [9 x i8*], 98 // CHECK: [[PTR_ARR2:%.+]] = alloca [9 x i8*], 99 // CHECK: [[SIZET2:%.+]] = alloca [9 x i{{[0-9]+}}], 100 // CHECK: [[BASE_PTR_ARR3:%.+]] = alloca [2 x i8*], 101 // CHECK: [[PTR_ARR3:%.+]] = alloca [2 x i8*], 102 // CHECK: [[N_ADDR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[N_ADDR]], 103 // CHECK-64: [[N_EXT:%.+]] = zext i{{[0-9]+}} [[N_ADDR_VAL]] to i{{[0-9]+}} 104 // CHECK: [[SSAVE_RET:%.+]] = call i8* @llvm.stacksave() 105 // CHECK: store i8* [[SSAVE_RET]], i8** [[SSTACK]], 106 // CHECK-64: [[BN_VLA:%.+]] = alloca float, i{{[0-9]+}} [[N_EXT]], 107 // CHECK-32: [[BN_VLA:%.+]] = alloca float, i{{[0-9]+}} [[N_ADDR_VAL]], 108 // CHECK: [[N_ADDR_VAL2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[N_ADDR]], 109 // CHECK-64: [[N_EXT2:%.+]] = zext i{{[0-9]+}} [[N_ADDR_VAL2]] to i{{[0-9]+}} 110 // CHECK-64: [[CN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_EXT2]] 111 // CHECK-32: [[CN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_ADDR_VAL2]] 112 // CHECK: [[CN_VLA:%.+]] = alloca double, i{{[0-9]+}} [[CN_SIZE]], 113 // CHECK: [[AVAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A]], 114 // CHECK-64: [[CONV:%.+]] = bitcast i{{[0-9]+}}* [[ACAST]] to i{{[0-9]+}}* 115 // CHECK-64: store i{{[0-9]+}} [[AVAL]], i{{[0-9]+}}* [[CONV]], 116 // CHECK-32: store i{{[0-9]+}} [[AVAL]], i{{[0-9]+}}* [[ACAST]], 117 // CHECK: [[ACAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[ACAST]], 118 // CHECK: [[P_PTR:%.+]] = load i32*, i32** [[P]], align 64 119 // CHECK: [[BASE_PTR_GEP:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 120 // CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP]] to i{{[0-9]+}}* 121 // CHECK: store i{{[0-9]+}} [[ACAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]], 122 // CHECK: [[PTR_GEP:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 123 // CHECK: [[ACAST_TOPTR2:%.+]] = bitcast i8** [[PTR_GEP]] to i{{[0-9]+}}* 124 // CHECK: store i{{[0-9]+}} [[ACAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR2]], 125 // CHECK: [[BASE_PTR_GEP:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 126 // CHECK: [[PCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP]] to i32*** 127 // CHECK: store i32** [[P]], i32*** [[PCAST_TOPTR]], 128 // CHECK: [[PTR_GEP:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 129 // CHECK: [[PCAST_TOPTR2:%.+]] = bitcast i8** [[PTR_GEP]] to i32** 130 // CHECK: store i32* [[P_PTR]], i32** [[PCAST_TOPTR2]], 131 // CHECK: [[BASE_PTR_GEP_ARG:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 132 // CHECK: [[PTR_GEP_ARG:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 133 // CHECK: {{.+}} = call i32 @__tgt_target(i64 -1, {{.+}}, i32 2, i8** [[BASE_PTR_GEP_ARG]], i8** [[PTR_GEP_ARG]], i[[SZ]]* getelementptr inbounds ([2 x i[[SZ]]], [2 x i[[SZ]]]* [[SIZET]], i32 0, i32 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0)) 134 135 // TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], i32** dereferenceable{{.+}} [[P_IN:%.+]]) 136 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, 137 // TCHECK: [[P_ADDR:%.+]] = alloca i32**, 138 // TCHECK: [[P_PRIV:%.+]] = alloca i32*, 139 // TCHECK-NOT: alloca i{{[0-9]+}} 140 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]], 141 // TCHECK: store i32** [[P_IN]], i32*** [[P_ADDR]], 142 // TCHECK-NOT: store i{{[0-9]+}} % 143 // TCHECK: ret void 144 145 #pragma omp target firstprivate(aa, b, bn, c, cn, d) 146 { 147 aa += 1; 148 b[2] = 1.0; 149 bn[3] = 1.0; 150 c[1][2] = 1.0; 151 cn[1][3] = 1.0; 152 d.X = 1; 153 d.Y = 1; 154 } 155 156 // CHECK: [[A2VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A2]], 157 // CHECK: [[A2CASTCONV:%.+]] = bitcast i{{[0-9]+}}* [[A2CAST]] to i{{[0-9]+}}* 158 // CHECK: store i{{[0-9]+}} [[A2VAL]], i{{[0-9]+}}* [[A2CASTCONV]], 159 // CHECK: [[A2CAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A2CAST]], 160 // CHECK-64: [[BN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} [[N_EXT]], 4 161 // CHECK-32: [[BN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} [[N_ADDR_VAL]], 4 162 // CHECK-64: [[CN_SIZE_1:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_EXT2]] 163 // CHECK-32: [[CN_SIZE_1:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_ADDR_VAL2]] 164 // CHECK: [[CN_SIZE_2:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SIZE_1]], 8 165 166 // firstprivate(aa) --> base_ptr = aa, ptr = aa, size = 2 (short) 167 // CHECK: [[BASE_PTR_GEP2_0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 168 // CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_0]] to i{{[0-9]+}}* 169 // CHECK: store i{{[0-9]+}} [[A2CAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]], 170 // CHECK: [[PTR_GEP2_0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 171 // CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_0]] to i{{[0-9]+}}* 172 // CHECK: store i{{[0-9]+}} [[A2CAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]], 173 // CHECK: [[SIZE_GEPA2:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 174 // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[SIZE_GEPA2]], 175 176 // firstprivate(b): base_ptr = &b[0], ptr = &b[0], size = 40 (sizeof(float)*10) 177 // CHECK: [[BASE_PTR_GEP2_1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 178 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_1]] to [10 x float]** 179 // CHECK: store [10 x float]* [[B]], [10 x float]** [[BCAST_TOPTR]], 180 // CHECK: [[PTR_GEP2_1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 181 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_1]] to [10 x float]** 182 // CHECK: store [10 x float]* [[B]], [10 x float]** [[BCAST_TOPTR]], 183 // CHECK: [[SIZE_GEPB:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 184 // CHECK: store i{{[0-9]+}} 40, i{{[0-9]+}}* [[SIZE_GEPB]], 185 186 // 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) 187 // CHECK: [[BASE_PTR_GEP2_2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 188 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_2]] to i{{[0-9]+}}* 189 // CHECK-64: store i{{[0-9]+}} [[N_EXT]], i{{[0-9]+}}* [[BCAST_TOPTR]], 190 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL]], i{{[0-9]+}}* [[BCAST_TOPTR]], 191 // CHECK: [[PTR_GEP2_2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 192 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_2]] to i{{[0-9]+}}* 193 // CHECK-64: store i{{[0-9]+}} [[N_EXT]], i{{[0-9]+}}* [[BCAST_TOPTR]], 194 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL]], i{{[0-9]+}}* [[BCAST_TOPTR]], 195 // CHECK: [[SIZE_GEPBN_1:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 196 // CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPBN_1]], 197 // CHECK: [[BASE_PTR_GEP2_3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 198 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_3]] to float** 199 // CHECK: store float* [[BN_VLA]], float** [[BCAST_TOPTR]], 200 // CHECK: [[SIZE_GEPBN_3:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 201 // CHECK: store i{{[0-9]+}} [[BN_SIZE]], i{{[0-9]+}}* [[SIZE_GEPBN_3]] 202 203 // firstprivate(c): base_ptr = &c[0], ptr = &c[0], size = 400 (5*10*sizeof(double)) 204 // CHECK: [[BASE_PTR_GEP2_4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 205 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_4]] to [5 x [10 x double]]** 206 // CHECK: store [5 x [10 x double]]* [[C]], [5 x [10 x double]]** [[BCAST_TOPTR]], 207 // CHECK: [[PTR_GEP2_4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 208 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_4]] to [5 x [10 x double]]** 209 // CHECK: store [5 x [10 x double]]* [[C]], [5 x [10 x double]]** [[BCAST_TOPTR]], 210 // CHECK: [[SIZE_GEPC_4:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 211 // CHECK: store i{{[0-9]+}} 400, i{{[0-9]+}}* [[SIZE_GEPC_4]], 212 213 // 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) 214 // CHECK: [[BASE_PTR_GEP2_5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 215 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_5]] to i{{[0-9]+}}* 216 // CHECK: store i{{[0-9]+}} 5, i{{[0-9]+}}* [[BCAST_TOPTR]], 217 // CHECK: [[PTR_GEP2_5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 218 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_5]] to i{{[0-9]+}}* 219 // CHECK: store i{{[0-9]+}} 5, i{{[0-9]+}}* [[BCAST_TOPTR]], 220 // CHECK: [[SIZE_GEPCN_5:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 221 // CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPCN_5]], 222 // CHECK: [[BASE_PTR_GEP2_6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6 223 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_6]] to i{{[0-9]+}}* 224 // CHECK-64: store i{{[0-9]+}} [[N_EXT2]], i{{[0-9]+}}* [[BCAST_TOPTR]], 225 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL2]], i{{[0-9]+}}* [[BCAST_TOPTR]], 226 // CHECK: [[PTR_GEP2_6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6 227 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_6]] to i{{[0-9]+}}* 228 // CHECK-64: store i{{[0-9]+}} [[N_EXT2]], i{{[0-9]+}}* [[BCAST_TOPTR]], 229 // CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL2]], i{{[0-9]+}}* [[BCAST_TOPTR]], 230 // CHECK: [[SIZE_GEPCN_6:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6 231 // CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPCN_6]], 232 // CHECK: [[BASE_PTR_GEP2_7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7 233 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_7]] to double** 234 // CHECK: store double* [[CN_VLA]], double** [[BCAST_TOPTR]], 235 // CHECK: [[PTR_GEP2_7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7 236 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_7]] to double** 237 // CHECK: store double* [[CN_VLA]], double** [[BCAST_TOPTR]], 238 // CHECK: [[SIZE_GEPCN_7:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7 239 // CHECK: store i{{[0-9]+}} [[CN_SIZE_2]], i{{[0-9]+}}* [[SIZE_GEPCN_7]], 240 241 // firstprivate(d): base_ptr = &d, ptr = &d, size = 16 242 // CHECK: [[BASE_PTR_GEP2_8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8 243 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_8]] to [[TT]]** 244 // CHECK: store [[TT]]* [[D]], [[TT]]** [[BCAST_TOPTR]], 245 // CHECK: [[PTR_GEP2_8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8 246 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_8]] to [[TT]]** 247 // CHECK: store [[TT]]* [[D]], [[TT]]** [[BCAST_TOPTR]], 248 // CHECK: [[SIZE_GEPCN_8:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8 249 // CHECK: store i{{[0-9]+}} {{[0-9]+}}, i{{[0-9]+}}* [[SIZE_GEPCN_8]], 250 251 // CHECK: [[BASE_PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 252 // CHECK: [[PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 253 // CHECK: [[SIZES_ARG2:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 254 // 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)) 255 256 // make sure that firstprivate variables are generated in all cases and that we use those instances for operations inside the 257 // target region 258 // 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:%.+]]) 259 // TCHECK: [[A2_ADDR:%.+]] = alloca i{{[0-9]+}}, 260 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x float]*, 261 // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}}, 262 // TCHECK: [[BN_ADDR:%.+]] = alloca float*, 263 // TCHECK: [[C_ADDR:%.+]] = alloca [5 x [10 x double]]*, 264 // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}}, 265 // TCHECK: [[VLA_ADDR4:%.+]] = alloca i{{[0-9]+}}, 266 // TCHECK: [[CN_ADDR:%.+]] = alloca double*, 267 // TCHECK: [[D_ADDR:%.+]] = alloca [[TT]]*, 268 // TCHECK-NOT: alloca i{{[0-9]+}}, 269 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x float], 270 // TCHECK: [[SSTACK:%.+]] = alloca i8*, 271 // TCHECK: [[C_PRIV:%.+]] = alloca [5 x [10 x double]], 272 // TCHECK: [[D_PRIV:%.+]] = alloca [[TT]], 273 // TCHECK: store i{{[0-9]+}} [[A2_IN]], i{{[0-9]+}}* [[A2_ADDR]], 274 // TCHECK: store [10 x float]* [[B_IN]], [10 x float]** [[B_ADDR]], 275 // TCHECK: store i{{[0-9]+}} [[BN_SZ]], i{{[0-9]+}}* [[VLA_ADDR]], 276 // TCHECK: store float* [[BN_IN]], float** [[BN_ADDR]], 277 // TCHECK: store [5 x [10 x double]]* [[C_IN]], [5 x [10 x double]]** [[C_ADDR]], 278 // TCHECK: store i{{[0-9]+}} [[CN_SZ1]], i{{[0-9]+}}* [[VLA_ADDR2]], 279 // TCHECK: store i{{[0-9]+}} [[CN_SZ2]], i{{[0-9]+}}* [[VLA_ADDR4]], 280 // TCHECK: store double* [[CN_IN]], double** [[CN_ADDR]], 281 // TCHECK: store [[TT]]* [[D_IN]], [[TT]]** [[D_ADDR]], 282 // TCHECK: [[CONV_A2ADDR:%.+]] = bitcast i{{[0-9]+}}* [[A2_ADDR]] to i{{[0-9]+}}* 283 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x float]*, [10 x float]** [[B_ADDR]], 284 // TCHECK: [[BN_SZ_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]], 285 // TCHECK: [[BN_ADDR_REF:%.+]] = load float*, float** [[BN_ADDR]], 286 // TCHECK: [[C_ADDR_REF:%.+]] = load [5 x [10 x double]]*, [5 x [10 x double]]** [[C_ADDR]], 287 // TCHECK: [[CN_SZ1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]], 288 // TCHECK: [[CN_SZ2_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR4]], 289 // TCHECK: [[CN_ADDR_REF:%.+]] = load double*, double** [[CN_ADDR]], 290 // TCHECK: [[D_ADDR_REF:%.+]] = load [[TT]]*, [[TT]]** [[D_ADDR]], 291 292 // firstprivate(aa): a_priv = a_in 293 // TCHECK-NOT: store i{{[0-9]+}} % 294 295 // firstprivate(b): memcpy(b_priv,b_in) 296 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x float]* [[B_PRIV]] to i8* 297 // TCHECK: [[B_ADDR_REF_BCAST:%.+]] = bitcast [10 x float]* [[B_ADDR_REF]] to i8* 298 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_ADDR_REF_BCAST]], {{.+}}) 299 300 // TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave() 301 // TCHECK: store i8* [[RET_STACK]], i8** [[SSTACK]], 302 303 // firstprivate(bn) 304 // TCHECK: [[BN_PRIV:%.+]] = alloca float, i{{[0-9]+}} [[BN_SZ_VAL]], 305 // TCHECK: [[BN_COPY_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[BN_SZ_VAL]], 4 306 // TCHECK: [[BN_PRIV__BCAST:%.+]] = bitcast float* [[BN_PRIV]] to i8* 307 // TCHECK: [[BN_REF_IN_BCAST:%.+]] = bitcast float* [[BN_ADDR_REF]] to i8* 308 // 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]],{{.+}}) 309 310 // firstprivate(c) 311 // TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_PRIV]] to i8* 312 // TCHECK: [[C_IN_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_ADDR_REF]] to i8* 313 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[C_PRIV_BCAST]], i8* align {{[0-9]+}} [[C_IN_BCAST]],{{.+}}) 314 315 // firstprivate(cn) 316 // TCHECK: [[CN_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ1_VAL]], [[CN_SZ2_VAL]] 317 // TCHECK: [[CN_PRIV:%.+]] = alloca double, i{{[0-9]+}} [[CN_SZ]], 318 // TCHECK: [[CN_SZ2:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ1_VAL]], [[CN_SZ2_VAL]] 319 // TCHECK: [[CN_SZ2_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ2]], 8 320 // TCHECK: [[CN_PRIV_BCAST:%.+]] = bitcast double* [[CN_PRIV]] to i8* 321 // TCHECK: [[CN_IN_BCAST:%.+]] = bitcast double* [[CN_ADDR_REF]] to i8* 322 // 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]],{{.+}}) 323 324 // firstprivate(d) 325 // TCHECK: [[D_PRIV_BCAST:%.+]] = bitcast [[TT]]* [[D_PRIV]] to i8* 326 // TCHECK: [[D_IN_BCAST:%.+]] = bitcast [[TT]]* [[D_ADDR_REF]] to i8* 327 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[D_PRIV_BCAST]], i8* align {{[0-9]+}} [[D_IN_BCAST]],{{.+}}) 328 329 #pragma omp target firstprivate(ptr, e) 330 { 331 ptr[0] = e.X; 332 ptr[0]++; 333 } 334 // CHECK: [[PTR_ADDR_REF:%.+]] = load double*, double** [[PTR_ADDR]], 335 336 // CHECK: [[FP_E_BC:%.+]] = bitcast [[TTII]]* [[FP_E]] to i8* 337 // CHECK: [[E_BC:%.+]] = bitcast [[TTII]]* [[E:%.+]] to i8* 338 // CHECK: call void @llvm.memcpy.p0i8.p0i8.i{{64|32}}(i8* {{.*}} [[FP_E_BC]], i8* {{.*}} [[E_BC]], i{{64|32}} 8, i1 false) 339 // CHECK: [[BASE_PTR_GEP3_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 340 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP3_0]] to double** 341 // CHECK: store double* [[PTR_ADDR_REF]], double** [[BCAST_TOPTR]], 342 // CHECK: [[PTR_GEP3_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 343 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP3_0]] to double** 344 // CHECK: store double* [[PTR_ADDR_REF]], double** [[BCAST_TOPTR]], 345 // CHECK: [[BASE_PTR_GEP3_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 346 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP3_1]] to [[TTII]]** 347 // CHECK: store [[TTII]]* [[FP_E]], [[TTII]]** [[BCAST_TOPTR]], 348 // CHECK: [[PTR_GEP3_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 349 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP3_1]] to [[TTII]]** 350 // CHECK: store [[TTII]]* [[FP_E]], [[TTII]]** [[BCAST_TOPTR]], 351 352 // CHECK: [[BASE_PTR_GEP_ARG3:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 353 // CHECK: [[PTR_GEP_ARG3:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 354 // CHECK: {{.+}} = call i32 @__tgt_target(i64 -1, {{.+}}, i32 2, i8** [[BASE_PTR_GEP_ARG3]], i8** [[PTR_GEP_ARG3]], i[[SZ]]* getelementptr inbounds ([2 x i[[SZ]]], [2 x i[[SZ]]]* [[SIZET3]], i32 0, i32 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT3]], i32 0, i32 0)) 355 356 // TCHECK: define weak void @__omp_offloading_{{.+}}(double* [[PTR_IN:%.+]], [[TTII]]* dereferenceable{{.+}} [[E:%.+]]) 357 // TCHECK-NOT: alloca [[TTII]], 358 // TCHECK: [[PTR_ADDR:%.+]] = alloca double*, 359 // TCHECK-NOT: alloca [[TTII]], 360 // TCHECK-NOT: alloca double*, 361 // TCHECK: store double* [[PTR_IN]], double** [[PTR_ADDR]], 362 // TCHECK-NOT: store double* % 363 364 return a; 365 } 366 367 template <typename tx> 368 tx ftemplate(int n) { 369 tx a = 0; 370 tx b[10]; 371 372 #pragma omp target firstprivate(a, b) 373 { 374 a += 1; 375 b[2] += 1; 376 } 377 378 return a; 379 } 380 381 static int fstatic(int n) { 382 int a = 0; 383 char aaa = 0; 384 int b[10]; 385 386 #pragma omp target firstprivate(a, aaa, b) 387 { 388 a += 1; 389 aaa += 1; 390 b[2] += 1; 391 } 392 393 return a; 394 } 395 396 // TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], i{{[0-9]+}} [[A3_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]]) 397 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, 398 // TCHECK: [[A3_ADDR:%.+]] = alloca i{{[0-9]+}}, 399 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*, 400 // TCHECK-NOT: alloca i{{[0-9]+}}, 401 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}], 402 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]], 403 // TCHECK: store i{{[0-9]+}} [[A3_IN]], i{{[0-9]+}}* [[A3_ADDR]], 404 // TCHECK: store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]], 405 // TCHECK-64: [[A_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}* 406 // TCHECK: [[A3_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A3_ADDR]] to i8* 407 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]], 408 409 // firstprivate(a): a_priv = a_in 410 411 // firstprivate(aaa) 412 // TCHECK-NOT: store i{{[0-9]+}} % 413 414 // firstprivate(b) 415 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8* 416 // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8* 417 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}}) 418 419 // TCHECK: ret void 420 421 struct S1 { 422 double a; 423 424 int r1(int n) { 425 int b = n + 1; 426 short int c[2][n]; 427 428 #pragma omp target firstprivate(b, c) 429 { 430 this->a = (double)b + 1.5; 431 c[1][1] = ++a; 432 } 433 434 return c[1][1] + (int)b; 435 } 436 437 // on the host side, we first generate r1, then the static function and the template above 438 // CHECK: define{{.+}} i32 {{.+}}([[S1]]* {{.+}}, i{{[0-9]+}} {{.+}}) 439 // CHECK: [[BASE_PTRS4:%.+]] = alloca [6 x i8*], 440 // CHECK: [[PTRS4:%.+]] = alloca [6 x i8*], 441 // CHECK: [[SIZET4:%.+]] = alloca [6 x i{{[0-9]+}}], 442 443 // map(this: this ptr is implicitly captured (not firstprivate matter) 444 // CHECK: [[BP0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 445 // CHECK: [[CBP0:%.+]] = bitcast i8** [[BP0]] to %struct.S1** 446 // CHECK: store %struct.S1* [[THIS:%.+]], %struct.S1** [[CBP0]], 447 // CHECK: [[P0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 448 // CHECK: [[CP0:%.+]] = bitcast i8** [[P0]] to double** 449 // CHECK: store double* [[A:%.+]], double** [[CP0]], 450 // CHECK: [[SZ0:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 451 // CHECK: store i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}* [[SZ0]], 452 453 // CHECK: [[BP1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 454 // CHECK: [[CBP1:%.+]] = bitcast i8** [[BP1]] to %struct.S1** 455 // CHECK: store %struct.S1* [[THIS]], %struct.S1** [[CBP1]], 456 // CHECK: [[P1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 457 // CHECK: [[CP1:%.+]] = bitcast i8** [[P1]] to double** 458 // CHECK: store double* [[A]], double** [[CP1]], 459 // CHECK: [[SZ1:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 460 // CHECK: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SZ1]], 461 462 // firstprivate(b): base_ptr = b, ptr = b, size = 4 (pass by-value) 463 // CHECK: [[BASE_PTRS_GEP4_1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 464 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_1]] to i{{[0-9]+}}* 465 // CHECK: store i{{[0-9]+}} [[B_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], 466 // CHECK: [[PTRS_GEP4_1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 467 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_1]] to i{{[0-9]+}}* 468 // CHECK: store i{{[0-9]+}} [[B_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]], 469 // CHECK: [[SIZES_GEP4_1:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 470 // CHECK: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_1]], 471 472 // firstprivate(c), 3 entries: 2, n, c 473 // CHECK: [[BASE_PTRS_GEP4_2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 474 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_2]] to i{{[0-9]+}}* 475 // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[BCAST_TOPTR]], 476 // CHECK: [[PTRS_GEP4_2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 477 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_2]] to i{{[0-9]+}}* 478 // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[BCAST_TOPTR]], 479 // CHECK: [[SIZES_GEP4_2:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 480 // CHECK-64: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SIZES_GEP4_2]], 481 // CHECK-32: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_2]], 482 // CHECK: [[BASE_PTRS_GEP4_3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 483 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_3]] to i{{[0-9]+}}* 484 // CHECK: store i{{[0-9]+}} [[N:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], 485 // CHECK: [[PTRS_GEP4_3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 486 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_3]] to i{{[0-9]+}}* 487 // CHECK: store i{{[0-9]+}} [[N]], i{{[0-9]+}}* [[BCAST_TOPTR]], 488 // CHECK: [[SIZES_GEP4_3:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 489 // CHECK-64: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SIZES_GEP4_3]], 490 // CHECK-32: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_3]], 491 // CHECK: [[BASE_PTRS_GEP4_4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 492 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_4]] to i{{[0-9]+}}** 493 // CHECK: store i{{[0-9]+}}* [[B:%.+]], i{{[0-9]+}}** [[BCAST_TOPTR]], 494 // CHECK: [[PTRS_GEP4_4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 495 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_4]] to i{{[0-9]+}}** 496 // CHECK: store i{{[0-9]+}}* [[B]], i{{[0-9]+}}** [[BCAST_TOPTR]], 497 // CHECK: [[SIZES_GEP4_4:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 498 // CHECK: store i{{[0-9]+}} [[B_SIZE:%.+]], i{{[0-9]+}}* [[SIZES_GEP4_4]], 499 500 // only check that we use the map types stored in the global variable 501 // 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)) 502 503 // 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:%.+]]) 504 // TCHECK: [[TH_ADDR:%.+]] = alloca [[S1]]*, 505 // TCHECK: [[B_ADDR:%.+]] = alloca i{{[0-9]+}}, 506 // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}}, 507 // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}}, 508 // TCHECK: [[C_ADDR:%.+]] = alloca i{{[0-9]+}}*, 509 // TCHECK-NOT: alloca i{{[0-9]+}}, 510 // TCHECK: [[SSTACK:%.+]] = alloca i8*, 511 512 // TCHECK: store [[S1]]* [[TH]], [[S1]]** [[TH_ADDR]], 513 // TCHECK: store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[B_ADDR]], 514 // TCHECK: store i{{[0-9]+}} [[VLA]], i{{[0-9]+}}* [[VLA_ADDR]], 515 // TCHECK: store i{{[0-9]+}} [[VLA1]], i{{[0-9]+}}* [[VLA_ADDR2]], 516 // TCHECK: store i{{[0-9]+}}* [[C_IN]], i{{[0-9]+}}** [[C_ADDR]], 517 // TCHECK: [[TH_ADDR_REF:%.+]] = load [[S1]]*, [[S1]]** [[TH_ADDR]], 518 // TCHECK-64: [[B_ADDR_CONV:%.+]] = bitcast i{{[0-9]+}}* [[B_ADDR]] to i{{[0-9]+}}* 519 // TCHECK: [[VLA_ADDR_REF:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]], 520 // TCHECK: [[VLA_ADDR_REF2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]], 521 // TCHECK: [[C_ADDR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[C_ADDR]], 522 523 // firstprivate(b) 524 // TCHECK-NOT: store i{{[0-9]+}} % 525 526 // TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave() 527 // TCHECK: store i8* [[RET_STACK:%.+]], i8** [[SSTACK]], 528 529 // firstprivate(c) 530 // TCHECK: [[C_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF]], [[VLA_ADDR_REF2]] 531 // TCHECK: [[C_PRIV:%.+]] = alloca i{{[0-9]+}}, i{{[0-9]+}} [[C_SZ]], 532 // TCHECK: [[C_SZ2:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF]], [[VLA_ADDR_REF2]] 533 // TCHECK: [[C_SZ_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[C_SZ2]], 2 534 // TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_PRIV]] to i8* 535 // TCHECK: [[C_IN_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_ADDR_REF]] to i8* 536 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[C_PRIV_BCAST]], i8* align {{[0-9]+}} [[C_IN_BCAST]],{{.+}}) 537 538 // finish 539 // TCHECK: [[RELOAD_SSTACK:%.+]] = load i8*, i8** [[SSTACK]], 540 // TCHECK: call void @llvm.stackrestore(i8* [[RELOAD_SSTACK]]) 541 // TCHECK: ret void 542 543 // static host function 544 // CHECK: define{{.+}} i32 {{.+}}(i{{[0-9]+}} {{.+}}) 545 // CHECK: [[BASE_PTRS5:%.+]] = alloca [3 x i8*], 546 // CHECK: [[PTRS5:%.+]] = alloca [3 x i8*], 547 548 // firstprivate(a): by value 549 // CHECK: [[BASE_PTRS_GEP5_0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 550 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_0]] to i{{[0-9]+}}* 551 // CHECK: store i{{[0-9]+}} [[A_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], 552 // CHECK: [[PTRS_GEP5_0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 553 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_0]] to i{{[0-9]+}}* 554 // CHECK: store i{{[0-9]+}} [[A_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]], 555 556 // firstprivate(aaa): by value 557 // CHECK: [[BASE_PTRS_GEP5_1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 558 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_1]] to i{{[0-9]+}}* 559 // CHECK: store i{{[0-9]+}} [[A3_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], 560 // CHECK: [[PTRS_GEP5_1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 561 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_1]] to i{{[0-9]+}}* 562 // CHECK: store i{{[0-9]+}} [[A3_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]], 563 564 // firstprivate(b): base_ptr = &b[0], ptr= &b[0] 565 // CHECK: [[BASE_PTRS_GEP5_2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 566 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_2]] to [10 x i{{[0-9]+}}]** 567 // CHECK: store [10 x i{{[0-9]+}}]* [[B:%.+]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]], 568 // CHECK: [[PTRS_GEP5_2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 569 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_2]] to [10 x i{{[0-9]+}}]** 570 // CHECK: store [10 x i{{[0-9]+}}]* [[B]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]], 571 572 // only check that the right sizes and map types are used 573 // 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)) 574 }; 575 576 int bar(int n, double *ptr) { 577 int a = 0; 578 a += foo(n, ptr); 579 S1 S; 580 a += S.r1(n); 581 a += fstatic(n); 582 a += ftemplate<int>(n); 583 584 return a; 585 } 586 587 // template host and device 588 589 // CHECK: define{{.+}} i32 {{.+}}(i{{[0-9]+}} {{.+}}) 590 // CHECK: [[BASE_PTRS6:%.+]] = alloca [2 x i8*], 591 // CHECK: [[PTRS6:%.+]] = alloca [2 x i8*], 592 593 // firstprivate(a): by value 594 // CHECK: [[BASE_PTRS_GEP6_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 595 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP6_0]] to i{{[0-9]+}}* 596 // CHECK: store i{{[0-9]+}} [[AT_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], 597 // CHECK: [[PTRS_GEP6_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 598 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP6_0]] to i{{[0-9]+}}* 599 // CHECK: store i{{[0-9]+}} [[AT_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]], 600 601 // firstprivate(b): pointer 602 // CHECK: [[BASE_PTRS_GEP6_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 603 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP6_1]] to [10 x i{{[0-9]+}}]** 604 // CHECK: store [10 x i{{[0-9]+}}]* [[B:%.+]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]], 605 // CHECK: [[PTRS_GEP6_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 606 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP6_1]] to [10 x i{{[0-9]+}}]** 607 // CHECK: store [10 x i{{[0-9]+}}]* [[B]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]], 608 609 // 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)) 610 611 // TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]]) 612 // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, 613 // TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*, 614 // TCHECK-NOT: alloca i{{[0-9]+}}, 615 // TCHECK: [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}], 616 // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]], 617 // TCHECK: store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]], 618 // TCHECK-64: [[A_ADDR_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}* 619 // TCHECK: [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]], 620 621 // firstprivate(a) 622 // TCHECK-NOT: store i{{[0-9]+}} % 623 624 // firstprivate(b) 625 // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8* 626 // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8* 627 // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}}) 628 629 // TCHECK: ret void 630 631 #endif 632