1 // Test host codegen. 2 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 3 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s 4 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 5 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 6 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s 7 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 8 9 // RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s 10 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s 11 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s 12 // RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s 13 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s 14 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s 15 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}} 16 17 // Test target codegen - host bc file has to be created first. 18 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc 19 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 20 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s 21 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 22 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc 23 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 24 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s 25 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 26 27 // RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc 28 // RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s 29 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s 30 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s 31 // RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc 32 // RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s 33 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s 34 // RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s 35 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}} 36 37 // Test host codegen. 38 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 39 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s 40 // 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 -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 41 // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 42 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s 43 // 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 -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 44 45 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s 46 // 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 47 // 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 -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s 48 // RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s 49 // 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 50 // 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 -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s 51 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}} 52 53 // Test target codegen - host bc file has to be created first. 54 // 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 55 // 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 -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 56 // 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 57 // 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 -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 58 // 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 59 // 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 -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 60 // 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 61 // 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 -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 62 63 // 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 64 // 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 -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s 65 // 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 66 // 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 -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s 67 // 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 68 // 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 -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s 69 // 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 70 // 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 -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s 71 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}} 72 73 // expected-no-diagnostics 74 #ifndef HEADER 75 #define HEADER 76 77 // CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } 78 // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" 79 // CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } 80 81 // CHECK-DAG: [[TT:%.+]] = type { i64, i8 } 82 // CHECK-DAG: [[S1:%.+]] = type { double } 83 // CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } 84 85 // TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } 86 87 // We have 8 target regions, but only 7 that actually will generate offloading 88 // code, only 6 will have mapped arguments, and only 4 have all-constant map 89 // sizes. 90 91 // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [3 x i64] [i64 2, i64 4, i64 4] 92 // CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 800, i64 800] 93 // CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [1 x i64] [i64 2] 94 // CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 800] 95 // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i64] [i64 4, i64 2] 96 // CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 800] 97 // CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [10 x i64] [i64 800, i64 547, i64 800, i64 547, i64 547, i64 800, i64 800, i64 547, i64 547, i64 800] 98 // CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 800, i64 800, i64 547] 99 // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [5 x i64] [i64 4, i64 4, i64 2, i64 1, i64 40] 100 // CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [5 x i64] [i64 800, i64 800, i64 800, i64 800, i64 547] 101 // CHECK-DAG: [[SIZET7:@.+]] = private unnamed_addr constant [3 x i64] [i64 4, i64 2, i64 40] 102 // CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 547] 103 // CHECK-DAG: @{{.*}} = weak constant i8 0 104 // CHECK-DAG: @{{.*}} = weak constant i8 0 105 // CHECK-DAG: @{{.*}} = weak constant i8 0 106 // CHECK-DAG: @{{.*}} = weak constant i8 0 107 // CHECK-DAG: @{{.*}} = weak constant i8 0 108 // CHECK-DAG: @{{.*}} = weak constant i8 0 109 // CHECK-DAG: @{{.*}} = weak constant i8 0 110 111 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 112 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 113 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 114 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 115 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 116 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 117 // TCHECK: @{{.+}} = weak constant [[ENTTY]] 118 // TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] 119 120 // Check target registration is registered as a Ctor. 121 // CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @.omp_offloading.requires_reg, i8* null }] 122 123 124 template<typename tx, typename ty> 125 struct TT{ 126 tx X; 127 ty Y; 128 }; 129 130 int global; 131 132 // CHECK: define {{.*}}[[FOO:@.+]]( 133 int foo(int n) { 134 int a = 0; 135 short aa = 0; 136 float b[10]; 137 float bn[n]; 138 double c[5][10]; 139 double cn[5][n]; 140 TT<long long, char> d; 141 142 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams_nowait_mapper(i64 -1, i8* @{{[^,]+}}, i32 3, i8** [[BP:%[^,]+]], i8** [[P:%[^,]+]], i64* getelementptr inbounds ([3 x i64], [3 x i64]* [[SIZET]], i32 0, i32 0), i64* getelementptr inbounds ([3 x i64], [3 x i64]* [[MAPT]], i32 0, i32 0), i8** null, i32 {{[^,]+}}, i32 {{[^)]+}}) 143 // CHECK-DAG: [[BP]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR:%[^,]+]], i32 0, i32 0 144 // CHECK-DAG: [[P]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR:%[^,]+]], i32 0, i32 0 145 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR]], i32 0, i32 [[IDX0:[0-9]+]] 146 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR]], i32 0, i32 [[IDX0]] 147 // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]* 148 // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]* 149 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR0]] 150 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR0]] 151 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR]], i32 0, i32 [[IDX1:[0-9]+]] 152 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR]], i32 0, i32 [[IDX1]] 153 // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* 154 // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* 155 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR1]] 156 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR1]] 157 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BPR]], i32 0, i32 [[IDX1:[0-9]+]] 158 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PR]], i32 0, i32 [[IDX1]] 159 // CHECK-DAG: [[CBPADDR2:%.+]] = bitcast i8** [[BPADDR2]] to i[[SZ]]* 160 // CHECK-DAG: [[CPADDR2:%.+]] = bitcast i8** [[PADDR2]] to i[[SZ]]* 161 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR2]] 162 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR2]] 163 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 164 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 165 // CHECK: [[FAIL]] 166 // CHECK: call void [[HVT0:@.+]](i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^)]+}}) 167 // CHECK-NEXT: br label %[[END]] 168 // CHECK: [[END]] 169 #pragma omp target teams distribute num_teams(a) thread_limit(a) firstprivate(aa) nowait 170 for (int i = 0; i < 10; ++i) { 171 } 172 173 // CHECK: call void [[HVT1:@.+]](i[[SZ]] {{[^,]+}}) 174 #pragma omp target teams distribute if(target: 0) 175 for (int i = 0; i < 10; ++i) { 176 a += 1; 177 } 178 179 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams_mapper(i64 -1, i8* @{{[^,]+}}, i32 1, i8** [[BP:%[^,]+]], i8** [[P:%[^,]+]], i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[SIZET2]], i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[MAPT2]], i32 0, i32 0), i8** null, i32 0, i32 0) 180 // CHECK-DAG: [[BP]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BPR:%[^,]+]], i32 0, i32 0 181 // CHECK-DAG: [[P]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PR:%[^,]+]], i32 0, i32 0 182 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BPR]], i32 0, i32 [[IDX0:[0-9]+]] 183 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PR]], i32 0, i32 [[IDX0]] 184 // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]* 185 // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]* 186 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR0]] 187 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR0]] 188 189 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 190 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 191 // CHECK: [[FAIL]] 192 // CHECK: call void [[HVT2:@.+]](i[[SZ]] {{[^,]+}}) 193 // CHECK-NEXT: br label %[[END]] 194 // CHECK: [[END]] 195 #pragma omp target teams distribute if(target: 1) 196 for (int i = 0; i < 10; ++i) { 197 aa += 1; 198 } 199 200 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 10 201 // CHECK: br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] 202 // CHECK: [[IFTHEN]] 203 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams_mapper(i64 -1, i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[SIZET3]], i32 0, i32 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT3]], i32 0, i32 0), i8** null, i32 0, i32 0) 204 // CHECK-DAG: [[BPR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 205 // CHECK-DAG: [[PR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%[^,]+]], i32 0, i32 0 206 207 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 208 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 209 // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]* 210 // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]* 211 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR0]] 212 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR0]] 213 214 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 215 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 216 // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* 217 // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* 218 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR1]] 219 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR1]] 220 // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 221 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] 222 // CHECK: [[FAIL]] 223 // CHECK: call void [[HVT3:@.+]]({{[^,]+}}, {{[^,]+}}) 224 // CHECK-NEXT: br label %[[END]] 225 // CHECK: [[END]] 226 // CHECK-NEXT: br label %[[IFEND:.+]] 227 // CHECK: [[IFELSE]] 228 // CHECK: call void [[HVT3]]({{[^,]+}}, {{[^,]+}}) 229 // CHECK-NEXT: br label %[[IFEND]] 230 // CHECK: [[IFEND]] 231 #pragma omp target teams distribute if(target: n>10) 232 for (int i = 0; i < 10; ++i) { 233 a += 1; 234 aa += 1; 235 } 236 237 // We capture 3 VLA sizes in this target region 238 // CHECK: load i32, i32* % 239 // CHECK-64: [[A_VAL:%.+]] = load i32, i32* %{{.+}}, 240 // CHECK-64: [[A_ADDR:%.+]] = bitcast i[[SZ]]* [[A_CADDR:%.+]] to i32* 241 // CHECK-64: store i32 [[A_VAL]], i32* [[A_ADDR]], 242 // CHECK-64: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], 243 244 // CHECK-32: [[A_VAL:%.+]] = load i32, i32* %{{.+}}, 245 // CHECK-32: store i32 [[A_VAL]], i32* [[A_CADDR:%.+]], 246 // CHECK-32: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], 247 248 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 249 // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] 250 // CHECK: [[TRY]] 251 // CHECK-64: [[BNSIZE:%.+]] = mul nuw i64 [[VLA0:%.+]], 4 252 // CHECK-32: [[BNSZSIZE:%.+]] = mul nuw i32 [[VLA0:%.+]], 4 253 // CHECK-32: [[BNSIZE:%.+]] = sext i32 [[BNSZSIZE]] to i64 254 // CHECK: [[CNELEMSIZE2:%.+]] = mul nuw i[[SZ]] 5, [[VLA1:%.+]] 255 // CHECK-64: [[CNSIZE:%.+]] = mul nuw i64 [[CNELEMSIZE2]], 8 256 // CHECK-32: [[CNSZSIZE:%.+]] = mul nuw i32 [[CNELEMSIZE2]], 8 257 // CHECK-32: [[CNSIZE:%.+]] = sext i32 [[CNSZSIZE]] to i64 258 259 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams_mapper(i64 -1, i8* @{{[^,]+}}, i32 10, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i64* [[SR:%[^,]+]], i64* getelementptr inbounds ([10 x i64], [10 x i64]* [[MAPT4]], i32 0, i32 0), i8** null, i32 0, i32 0) 260 // CHECK-DAG: [[BPR]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 261 // CHECK-DAG: [[PR]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P:%[^,]+]], i32 0, i32 0 262 // CHECK-DAG: [[SR]] = getelementptr inbounds [10 x i64], [10 x i64]* [[S:%[^,]+]], i32 0, i32 0 263 264 // CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [10 x i64], [10 x i64]* [[S]], i32 0, i32 [[IDX0:[0-9]+]] 265 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX0]] 266 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX0]] 267 // CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [10 x i64], [10 x i64]* [[S]], i32 0, i32 [[IDX1:[0-9]+]] 268 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX1]] 269 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX1]] 270 // CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [10 x i64], [10 x i64]* [[S]], i32 0, i32 [[IDX2:[0-9]+]] 271 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX2]] 272 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX2]] 273 // CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [10 x i64], [10 x i64]* [[S]], i32 0, i32 [[IDX3:[0-9]+]] 274 // CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX3]] 275 // CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX3]] 276 // CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [10 x i64], [10 x i64]* [[S]], i32 0, i32 [[IDX4:[0-9]+]] 277 // CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX4]] 278 // CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX4]] 279 // CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [10 x i64], [10 x i64]* [[S]], i32 0, i32 [[IDX5:[0-9]+]] 280 // CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX5]] 281 // CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX5]] 282 // CHECK-DAG: [[SADDR6:%.+]] = getelementptr inbounds [10 x i64], [10 x i64]* [[S]], i32 0, i32 [[IDX6:[0-9]+]] 283 // CHECK-DAG: [[BPADDR6:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX6]] 284 // CHECK-DAG: [[PADDR6:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX6]] 285 // CHECK-DAG: [[SADDR7:%.+]] = getelementptr inbounds [10 x i64], [10 x i64]* [[S]], i32 0, i32 [[IDX7:[0-9]+]] 286 // CHECK-DAG: [[BPADDR7:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX7]] 287 // CHECK-DAG: [[PADDR7:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX7]] 288 // CHECK-DAG: [[SADDR8:%.+]] = getelementptr inbounds [10 x i64], [10 x i64]* [[S]], i32 0, i32 [[IDX8:[0-9]+]] 289 // CHECK-DAG: [[BPADDR8:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX8]] 290 // CHECK-DAG: [[PADDR8:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX8]] 291 // CHECK-DAG: [[SADDR9:%.+]] = getelementptr inbounds [10 x i64], [10 x i64]* [[S]], i32 0, i32 [[IDX9:[0-9]+]] 292 // CHECK-DAG: [[BPADDR9:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX9]] 293 // CHECK-DAG: [[PADDR9:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX9]] 294 295 // The names below are not necessarily consistent with the names used for the 296 // addresses above as some are repeated. 297 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CBPADDR0:%.+]], 298 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CPADDR0:%.+]], 299 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 300 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 301 // CHECK-DAG: store i64 {{4|8}}, i64* {{%[^,]+}} 302 303 // CHECK-DAG: store i[[SZ]] [[VLA1]], i[[SZ]]* [[CBPADDR1:%.+]], 304 // CHECK-DAG: store i[[SZ]] [[VLA1]], i[[SZ]]* [[CPADDR1:%.+]], 305 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 306 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 307 // CHECK-DAG: store i64 {{4|8}}, i64* {{%[^,]+}} 308 309 // CHECK-DAG: store i[[SZ]] 5, i[[SZ]]* [[CBPADDR2:%.+]], 310 // CHECK-DAG: store i[[SZ]] 5, i[[SZ]]* [[CPADDR2:%.+]], 311 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 312 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 313 // CHECK-DAG: store i64 {{4|8}}, i64* {{%[^,]+}} 314 315 // CHECK-DAG: store i[[SZ]] [[A_CVAL]], i[[SZ]]* [[CBPADDR3:%.+]], 316 // CHECK-DAG: store i[[SZ]] [[A_CVAL]], i[[SZ]]* [[CPADDR3:%.+]], 317 // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 318 // CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 319 // CHECK-DAG: store i64 4, i64* {{%[^,]+}} 320 321 // CHECK-DAG: store [10 x float]* %{{.+}}, [10 x float]** [[CBPADDR4:%.+]], 322 // CHECK-DAG: store [10 x float]* %{{.+}}, [10 x float]** [[CPADDR4:%.+]], 323 // CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [10 x float]** 324 // CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to [10 x float]** 325 // CHECK-DAG: store i64 40, i64* {{%[^,]+}} 326 327 // CHECK-DAG: store float* %{{.+}}, float** [[CBPADDR5:%.+]], 328 // CHECK-DAG: store float* %{{.+}}, float** [[CPADDR5:%.+]], 329 // CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to float** 330 // CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to float** 331 // CHECK-DAG: store i64 [[BNSIZE]], i64* {{%[^,]+}} 332 333 // CHECK-DAG: store [5 x [10 x double]]* %{{.+}}, [5 x [10 x double]]** [[CBPADDR6:%.+]], 334 // CHECK-DAG: store [5 x [10 x double]]* %{{.+}}, [5 x [10 x double]]** [[CPADDR6:%.+]], 335 // CHECK-DAG: [[CBPADDR6]] = bitcast i8** {{%[^,]+}} to [5 x [10 x double]]** 336 // CHECK-DAG: [[CPADDR6]] = bitcast i8** {{%[^,]+}} to [5 x [10 x double]]** 337 // CHECK-DAG: store i64 400, i64* {{%[^,]+}} 338 339 // CHECK-DAG: store double* %{{.+}}, double** [[CBPADDR7:%.+]], 340 // CHECK-DAG: store double* %{{.+}}, double** [[CPADDR7:%.+]], 341 // CHECK-DAG: [[CBPADDR7]] = bitcast i8** {{%[^,]+}} to double** 342 // CHECK-DAG: [[CPADDR7]] = bitcast i8** {{%[^,]+}} to double** 343 // CHECK-DAG: store i64 [[CNSIZE]], i64* {{%[^,]+}} 344 345 // CHECK-DAG: store [[TT]]* %{{.+}}, [[TT]]** [[CBPADDR8:%.+]], 346 // CHECK-DAG: store [[TT]]* %{{.+}}, [[TT]]** [[CPADDR8:%.+]], 347 // CHECK-DAG: [[CBPADDR8]] = bitcast i8** {{%[^,]+}} to [[TT]]** 348 // CHECK-DAG: [[CPADDR8]] = bitcast i8** {{%[^,]+}} to [[TT]]** 349 // CHECK-DAG: store i64 {{12|16}}, i64* {{%[^,]+}} 350 351 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR9:%.+]], 352 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR9:%.+]], 353 // CHECK-DAG: [[CBPADDR9]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 354 // CHECK-DAG: [[CPADDR9]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 355 // CHECK-DAG: store i64 {{4|8}}, i64* {{%[^,]+}} 356 357 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 358 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 359 360 // CHECK: [[FAIL]] 361 // CHECK: call void [[HVT4:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 362 // CHECK-NEXT: br label %[[END]] 363 // CHECK: [[END]] 364 #pragma omp target teams distribute if(target: n>20) dist_schedule(static, n) 365 for (int i = 0; i < 10; ++i) { 366 a += 1; 367 b[2] += 1.0; 368 bn[3] += 1.0; 369 c[1][2] += 1.0; 370 cn[1][3] += 1.0; 371 d.X += 1; 372 d.Y += 1; 373 } 374 375 return a; 376 } 377 378 // Check that the offloading functions are emitted and that the arguments are 379 // correct and loaded correctly for the target regions in foo(). 380 381 // CHECK: define internal void [[HVT0]](i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^)]+}}) 382 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] {{[^)]+}}) 383 // 384 // 385 // CHECK: define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] {{[^)]+}}) 386 // CHECK: alloca i[[SZ]], 387 // CHECK: bitcast i[[SZ]]* {{.+}} to i16* 388 // CHECK: ret void 389 // CHECK-NEXT: } 390 391 392 // CHECK: define internal void [[HVT1]](i[[SZ]] %{{.+}}) 393 // Create stack storage and store argument in there. 394 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 395 // CHECK: [[AA_CASTED:%.+]] = alloca i[[SZ]], align 396 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 397 // CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* 398 // CHECK-64: [[AA:%.+]] = load i32, i32* [[AA_CADDR]], align 399 // CHECK-32: [[AA:%.+]] = load i32, i32* [[AA_ADDR]], align 400 // CHECK-64: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i32* 401 // CHECK-64: store i32 [[AA]], i32* [[AA_C]], align 402 // CHECK-32: store i32 [[AA]], i32* [[AA_CASTED]], align 403 // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align 404 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) 405 // 406 // 407 // CHECK: define internal {{.*}}void [[OMP_OUTLINED1]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) 408 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 409 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 410 // CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* 411 // CHECK-64: [[AA:%.+]] = load i32, i32* [[AA_CADDR]], align 412 // CHECK-32: [[AA:%.+]] = load i32, i32* [[AA_ADDR]], align 413 // CHECK: ret void 414 // CHECK-NEXT: } 415 416 // CHECK: define internal void [[HVT2]](i[[SZ]] %{{.+}}) 417 // Create stack storage and store argument in there. 418 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 419 // CHECK: [[AA_CASTED:%.+]] = alloca i[[SZ]], align 420 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 421 // CHECK: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 422 // CHECK: [[AA:%.+]] = load i16, i16* [[AA_CADDR]], align 423 // CHECK: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* 424 // CHECK: store i16 [[AA]], i16* [[AA_C]], align 425 // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align 426 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) 427 // 428 // 429 // CHECK: define internal {{.*}}void [[OMP_OUTLINED2]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) 430 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 431 // CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 432 // CHECK: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 433 // CHECK: [[AA:%.+]] = load i16, i16* [[AA_CADDR]], align 434 // CHECK: ret void 435 // CHECK-NEXT: } 436 437 // CHECK: define internal void [[HVT3]] 438 // Create stack storage and store argument in there. 439 // CHECK: [[A_ADDR:%.+]] = alloca i[[SZ]], align 440 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 441 // CHECK: [[A_CASTED:%.+]] = alloca i[[SZ]], align 442 // CHECK: [[AA_CASTED:%.+]] = alloca i[[SZ]], align 443 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[A_ADDR]], align 444 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 445 // CHECK-64-DAG:[[A_CADDR:%.+]] = bitcast i[[SZ]]* [[A_ADDR]] to i32* 446 // CHECK-DAG: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 447 // CHECK-64-DAG:[[A:%.+]] = load i32, i32* [[A_CADDR]], align 448 // CHECK-32-DAG:[[A:%.+]] = load i32, i32* [[A_ADDR]], align 449 // CHECK-64-DAG:[[A_C:%.+]] = bitcast i[[SZ]]* [[A_CASTED]] to i32* 450 // CHECK-64-DAG:store i32 [[A]], i32* [[A_C]], align 451 // CHECK-32-DAG:store i32 [[A]], i32* [[A_CASTED]], align 452 // CHECK-DAG: [[AA:%.+]] = load i16, i16* [[AA_CADDR]], align 453 // CHECK-DAG: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* 454 // CHECK-DAG: store i16 [[AA]], i16* [[AA_C]], align 455 // CHECK-DAG: [[PARAM1:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CASTED]], align 456 // CHECK-DAG: [[PARAM2:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align 457 // CHECK-DAG: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) 458 // 459 // 460 // CHECK: define internal {{.*}}void [[OMP_OUTLINED3]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}) 461 // CHECK: [[A_ADDR:%.+]] = alloca i[[SZ]], align 462 // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align 463 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[A_ADDR]], align 464 // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align 465 // CHECK-64-DAG:[[A_CADDR:%.+]] = bitcast i[[SZ]]* [[A_ADDR]] to i32* 466 // CHECK-DAG: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* 467 // CHECK: ret void 468 // CHECK-NEXT: } 469 470 // CHECK: define internal void [[HVT4]] 471 // Create local storage for each capture. 472 // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] 473 // CHECK: [[LOCAL_B:%.+]] = alloca [10 x float]* 474 // CHECK: [[LOCAL_VLA1:%.+]] = alloca i[[SZ]] 475 // CHECK: [[LOCAL_BN:%.+]] = alloca float* 476 // CHECK: [[LOCAL_C:%.+]] = alloca [5 x [10 x double]]* 477 // CHECK: [[LOCAL_VLA2:%.+]] = alloca i[[SZ]] 478 // CHECK: [[LOCAL_VLA3:%.+]] = alloca i[[SZ]] 479 // CHECK: [[LOCAL_CN:%.+]] = alloca double* 480 // CHECK: [[LOCAL_D:%.+]] = alloca [[TT]]* 481 // CHECK: alloca i[[SZ]], 482 // CHECK: [[LOCAL_A_CASTED:%.+]] = alloca i[[SZ]] 483 // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] 484 // CHECK-DAG: store [10 x float]* [[ARG_B:%.+]], [10 x float]** [[LOCAL_B]] 485 // CHECK-DAG: store i[[SZ]] [[ARG_VLA1:%.+]], i[[SZ]]* [[LOCAL_VLA1]] 486 // CHECK-DAG: store float* [[ARG_BN:%.+]], float** [[LOCAL_BN]] 487 // CHECK-DAG: store [5 x [10 x double]]* [[ARG_C:%.+]], [5 x [10 x double]]** [[LOCAL_C]] 488 // CHECK-DAG: store i[[SZ]] [[ARG_VLA2:%.+]], i[[SZ]]* [[LOCAL_VLA2]] 489 // CHECK-DAG: store i[[SZ]] [[ARG_VLA3:%.+]], i[[SZ]]* [[LOCAL_VLA3]] 490 // CHECK-DAG: store double* [[ARG_CN:%.+]], double** [[LOCAL_CN]] 491 // CHECK-DAG: store [[TT]]* [[ARG_D:%.+]], [[TT]]** [[LOCAL_D]] 492 493 // CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* 494 // CHECK-DAG: [[REF_B:%.+]] = load [10 x float]*, [10 x float]** [[LOCAL_B]], 495 // CHECK-DAG: [[VAL_VLA1:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA1]], 496 // CHECK-DAG: [[REF_BN:%.+]] = load float*, float** [[LOCAL_BN]], 497 // CHECK-DAG: [[REF_C:%.+]] = load [5 x [10 x double]]*, [5 x [10 x double]]** [[LOCAL_C]], 498 // CHECK-DAG: [[VAL_VLA2:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA2]], 499 // CHECK-DAG: [[VAL_VLA3:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA3]], 500 // CHECK-DAG: [[REF_CN:%.+]] = load double*, double** [[LOCAL_CN]], 501 // CHECK-DAG: [[REF_D:%.+]] = load [[TT]]*, [[TT]]** [[LOCAL_D]], 502 503 // CHECK-64-DAG:[[CONV_A:%.+]] = load i32, i32* [[CONV_AP]] 504 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_A_CASTED]] to i32* 505 // CHECK-64-DAG:store i32 [[CONV_A]], i32* [[CONV]], align 506 // CHECK-32-DAG:[[LOCAL_AV:%.+]] = load i32, i32* [[LOCAL_A]] 507 // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align 508 // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], 509 510 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 10, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*, i[[SZ]])* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]], i[[SZ]] %{{.+}}) 511 // 512 // 513 // CHECK: define internal {{.*}}void [[OMP_OUTLINED4]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, [10 x float]* {{.+}}, i[[SZ]] %{{.+}}, float* {{.+}}, [5 x [10 x double]]* {{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, double* {{.+}}, [[TT]]* {{.+}}) 514 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 515 516 template<typename tx> 517 tx ftemplate(int n) { 518 tx a = 0; 519 short aa = 0; 520 tx b[10]; 521 522 #pragma omp target teams distribute if(target: n>40) 523 for (int i = 0; i < 10; ++i) { 524 a += 1; 525 aa += 1; 526 b[2] += 1; 527 } 528 529 return a; 530 } 531 532 static 533 int fstatic(int n) { 534 int a = 0; 535 short aa = 0; 536 char aaa = 0; 537 int b[10]; 538 539 #pragma omp target teams distribute if(target: n>50) 540 for (int i = a; i < n; ++i) { 541 a += 1; 542 aa += 1; 543 aaa += 1; 544 b[2] += 1; 545 } 546 547 return a; 548 } 549 550 struct S1 { 551 double a; 552 553 int r1(int n){ 554 int b = n+1; 555 short int c[2][n]; 556 557 #pragma omp target teams distribute if(target: n>60) 558 for (int i = 0; i < 10; ++i) { 559 this->a = (double)b + 1.5; 560 c[1][1] = ++a; 561 } 562 563 return c[1][1] + (int)b; 564 } 565 }; 566 567 // CHECK: define {{.*}}@{{.*}}bar{{.*}} 568 int bar(int n){ 569 int a = 0; 570 571 // CHECK: call {{.*}}i32 [[FOO]](i32 {{.*}}) 572 a += foo(n); 573 574 S1 S; 575 // CHECK: call {{.*}}i32 [[FS1:@.+]]([[S1]]* {{.*}}, i32 {{.*}}) 576 a += S.r1(n); 577 578 // CHECK: call {{.*}}i32 [[FSTATIC:@.+]](i32 {{.*}}) 579 a += fstatic(n); 580 581 // CHECK: call {{.*}}i32 [[FTEMPLATE:@.+]](i32 {{.*}}) 582 a += ftemplate<int>(n); 583 584 return a; 585 } 586 587 // 588 // CHECK: define {{.*}}[[FS1]] 589 // 590 // CHECK: i8* @llvm.stacksave() 591 // CHECK-64: [[B_ADDR:%.+]] = bitcast i[[SZ]]* [[B_CADDR:%.+]] to i32* 592 // CHECK-64: store i32 %{{.+}}, i32* [[B_ADDR]], 593 // CHECK-64: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]], 594 595 // CHECK-32: store i32 %{{.+}}, i32* %__vla_expr 596 // CHECK-32: store i32 %{{.+}}, i32* [[B_ADDR:%.+]], 597 // CHECK-32: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]], 598 599 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 600 // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] 601 // CHECK: [[TRY]] 602 // We capture 2 VLA sizes in this target region 603 // CHECK: [[CELEMSIZE2:%.+]] = mul nuw i[[SZ]] 2, [[VLA0:%.+]] 604 // CHECK-64: [[CSIZE:%.+]] = mul nuw i[[SZ]] [[CELEMSIZE2]], 2 605 // CHECK-32: [[CSSZSIZE:%.+]] = mul nuw i[[SZ]] [[CELEMSIZE2]], 2 606 // CHECK-32: [[CSIZE:%.+]] = sext i32 [[CSSZSIZE]] to i64 607 608 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams_mapper(i64 -1, i8* @{{[^,]+}}, i32 6, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i64* [[SR:%[^,]+]], i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT5]], i32 0, i32 0), i8** null, i32 0, i32 0) 609 // CHECK-DAG: [[BPR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP:%.+]], i32 0, i32 0 610 // CHECK-DAG: [[PR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P:%.+]], i32 0, i32 0 611 // CHECK-DAG: [[SR]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S:%.+]], i32 0, i32 0 612 // CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S]], i32 [[IDX0:[0-9]+]] 613 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX0]] 614 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX0]] 615 // CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S]], i32 [[IDX1:[0-9]+]] 616 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX1]] 617 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX1]] 618 // CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S]], i32 [[IDX2:[0-9]+]] 619 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX2]] 620 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX2]] 621 // CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S]], i32 [[IDX3:[0-9]+]] 622 // CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] 623 // CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] 624 // CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S]], i32 [[IDX3:[0-9]+]] 625 // CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] 626 // CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] 627 // CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [6 x i64], [6 x i64]* [[S]], i32 [[IDX3:[0-9]+]] 628 // CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] 629 // CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] 630 631 // The names below are not necessarily consistent with the names used for the 632 // addresses above as some are repeated. 633 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CBPADDR0:%.+]], 634 // CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CPADDR0:%.+]], 635 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 636 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 637 // CHECK-DAG: store i64 {{4|8}}, i64* {{%[^,]+}} 638 639 // CHECK-DAG: store i[[SZ]] 2, i[[SZ]]* [[CBPADDR1:%.+]], 640 // CHECK-DAG: store i[[SZ]] 2, i[[SZ]]* [[CPADDR1:%.+]], 641 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 642 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 643 // CHECK-DAG: store i64 {{4|8}}, i64* {{%[^,]+}} 644 645 // CHECK-DAG: store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CBPADDR2:%.+]], 646 // CHECK-DAG: store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CPADDR2:%.+]], 647 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 648 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 649 // CHECK-DAG: store i64 4, i64* {{%[^,]+}} 650 651 // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR3:%.+]], 652 // CHECK-DAG: store double* %{{.+}}, double** [[CPADDR3:%.+]], 653 // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** 654 // CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to double** 655 // CHECK-DAG: store i64 8, i64* {{%[^,]+}} 656 657 // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR4:%.+]], 658 // CHECK-DAG: store double* %{{.+}}, double** [[CPADDR4:%.+]], 659 // CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** 660 // CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to double** 661 // CHECK-DAG: store i64 8, i64* {{%[^,]+}} 662 663 // CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR5:%.+]], 664 // CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR5:%.+]], 665 // CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to i16** 666 // CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to i16** 667 // CHECK-DAG: store i64 [[CSIZE]], i64* {{%[^,]+}} 668 669 // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 670 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] 671 672 // CHECK: [[FAIL]] 673 // CHECK: call void [[HVT7:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 674 // CHECK-NEXT: br label %[[END]] 675 // CHECK: [[END]] 676 677 // 678 // CHECK: define {{.*}}[[FSTATIC]] 679 // 680 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 50 681 // CHECK: br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] 682 // CHECK: [[IFTHEN]] 683 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams_mapper(i64 -1, i8* @{{[^,]+}}, i32 5, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i64* getelementptr inbounds ([5 x i64], [5 x i64]* [[SIZET6]], i32 0, i32 0), i64* getelementptr inbounds ([5 x i64], [5 x i64]* [[MAPT6]], i32 0, i32 0), i8** null, i32 0, i32 0) 684 // CHECK-DAG: [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0 685 // CHECK-DAG: [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0 686 687 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 0 688 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 0 689 // CHECK-DAG: store i[[SZ]] [[VAL0:%.+]], i[[SZ]]* [[CBPADDR0:%.+]], 690 // CHECK-DAG: store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0:%.+]], 691 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 692 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 693 694 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 1 695 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 1 696 // CHECK-DAG: store i[[SZ]] [[VAL1:%.+]], i[[SZ]]* [[CBPADDR1:%.+]], 697 // CHECK-DAG: store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1:%.+]], 698 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 699 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 700 701 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 2 702 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 2 703 // CHECK-DAG: store i[[SZ]] [[VAL2:%.+]], i[[SZ]]* [[CBPADDR2:%.+]], 704 // CHECK-DAG: store i[[SZ]] [[VAL2]], i[[SZ]]* [[CPADDR2:%.+]], 705 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 706 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 707 708 // CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 3 709 // CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 3 710 // CHECK-DAG: store i[[SZ]] [[VAL3:%.+]], i[[SZ]]* [[CBPADDR3:%.+]], 711 // CHECK-DAG: store i[[SZ]] [[VAL3]], i[[SZ]]* [[CPADDR3:%.+]], 712 // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 713 // CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 714 715 // CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 4 716 // CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 4 717 // CHECK-DAG: store [10 x i32]* %{{.+}}, [10 x i32]** [[CBPADDR4:%.+]], 718 // CHECK-DAG: store [10 x i32]* %{{.+}}, [10 x i32]** [[CPADDR4:%.+]], 719 // CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [10 x i32]** 720 // CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to [10 x i32]** 721 722 // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 723 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] 724 // CHECK: [[FAIL]] 725 // CHECK: call void [[HVT6:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 726 // CHECK-NEXT: br label %[[END]] 727 // CHECK: [[END]] 728 // CHECK-NEXT: br label %[[IFEND:.+]] 729 // CHECK: [[IFELSE]] 730 // CHECK: call void [[HVT6]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 731 // CHECK-NEXT: br label %[[IFEND]] 732 // CHECK: [[IFEND]] 733 734 // 735 // CHECK: define {{.*}}[[FTEMPLATE]] 736 // 737 // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 40 738 // CHECK: br i1 [[IF]], label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] 739 // CHECK: [[IFTHEN]] 740 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams_mapper(i64 -1, i8* @{{[^,]+}}, i32 3, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i64* getelementptr inbounds ([3 x i64], [3 x i64]* [[SIZET7]], i32 0, i32 0), i64* getelementptr inbounds ([3 x i64], [3 x i64]* [[MAPT7]], i32 0, i32 0), i8** null, i32 0, i32 0) 741 // CHECK-DAG: [[BPR]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP:%.+]], i32 0, i32 0 742 // CHECK-DAG: [[PR]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P:%.+]], i32 0, i32 0 743 744 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 0 745 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 0 746 // CHECK-DAG: store i[[SZ]] [[VAL0:%.+]], i[[SZ]]* [[CBPADDR0:%.+]], 747 // CHECK-DAG: store i[[SZ]] [[VAL0]], i[[SZ]]* [[CPADDR0:%.+]], 748 // CHECK-DAG: [[CBPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 749 // CHECK-DAG: [[CPADDR0]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 750 751 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 1 752 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 1 753 // CHECK-DAG: store i[[SZ]] [[VAL1:%.+]], i[[SZ]]* [[CBPADDR1:%.+]], 754 // CHECK-DAG: store i[[SZ]] [[VAL1]], i[[SZ]]* [[CPADDR1:%.+]], 755 // CHECK-DAG: [[CBPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 756 // CHECK-DAG: [[CPADDR1]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 757 758 // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BP]], i32 0, i32 2 759 // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[P]], i32 0, i32 2 760 // CHECK-DAG: store i[[SZ]] [[VAL2:%.+]], i[[SZ]]* [[CBPADDR2:%.+]], 761 // CHECK-DAG: store i[[SZ]] [[VAL2]], i[[SZ]]* [[CPADDR2:%.+]], 762 // CHECK-DAG: [[CBPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 763 // CHECK-DAG: [[CPADDR2]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* 764 765 // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 766 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] 767 // CHECK: [[FAIL]] 768 // CHECK: call void [[HVT5:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 769 // CHECK-NEXT: br label %[[END]] 770 // CHECK: [[END]] 771 // CHECK-NEXT: br label %[[IFEND:.+]] 772 // CHECK: [[IFELSE]] 773 // CHECK: call void [[HVT5]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}) 774 // CHECK-NEXT: br label %[[IFEND]] 775 // CHECK: [[IFEND]] 776 777 778 779 // Check that the offloading functions are emitted and that the arguments are 780 // correct and loaded correctly for the target regions of the callees of bar(). 781 782 // CHECK: define internal void [[HVT7]] 783 // Create local storage for each capture. 784 // CHECK: [[LOCAL_THIS:%.+]] = alloca [[S1]]* 785 // CHECK: [[LOCAL_B:%.+]] = alloca i[[SZ]] 786 // CHECK: [[LOCAL_VLA1:%.+]] = alloca i[[SZ]] 787 // CHECK: [[LOCAL_VLA2:%.+]] = alloca i[[SZ]] 788 // CHECK: [[LOCAL_C:%.+]] = alloca i16* 789 // CHECK: [[LOCAL_B_CASTED:%.+]] = alloca i[[SZ]] 790 // CHECK-DAG: store [[S1]]* [[ARG_THIS:%.+]], [[S1]]** [[LOCAL_THIS]] 791 // CHECK-DAG: store i[[SZ]] [[ARG_B:%.+]], i[[SZ]]* [[LOCAL_B]] 792 // CHECK-DAG: store i[[SZ]] [[ARG_VLA1:%.+]], i[[SZ]]* [[LOCAL_VLA1]] 793 // CHECK-DAG: store i[[SZ]] [[ARG_VLA2:%.+]], i[[SZ]]* [[LOCAL_VLA2]] 794 // CHECK-DAG: store i16* [[ARG_C:%.+]], i16** [[LOCAL_C]] 795 // Store captures in the context. 796 // CHECK-DAG: [[REF_THIS:%.+]] = load [[S1]]*, [[S1]]** [[LOCAL_THIS]], 797 // CHECK-64-DAG:[[CONV_BP:%.+]] = bitcast i[[SZ]]* [[LOCAL_B]] to i32* 798 // CHECK-DAG: [[VAL_VLA1:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA1]], 799 // CHECK-DAG: [[VAL_VLA2:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_VLA2]], 800 // CHECK-DAG: [[REF_C:%.+]] = load i16*, i16** [[LOCAL_C]], 801 802 // CHECK-64-DAG:[[CONV_B:%.+]] = load i32, i32* [[CONV_BP]] 803 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_B_CASTED]] to i32* 804 // CHECK-64-DAG:store i32 [[CONV_B]], i32* [[CONV]], align 805 // CHECK-32-DAG:[[LOCAL_BV:%.+]] = load i32, i32* [[LOCAL_B]] 806 // CHECK-32-DAG:store i32 [[LOCAL_BV]], i32* [[LOCAL_B_CASTED]], align 807 // CHECK-DAG: [[REF_B:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_B_CASTED]], 808 809 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) 810 // 811 // 812 // CHECK: define internal {{.*}}void [[OMP_OUTLINED5]](i32* noalias %.global_tid., i32* noalias %.bound_tid., [[S1]]* %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i16* {{.+}}) 813 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 814 815 816 // CHECK: define internal void [[HVT6]] 817 // Create local storage for each capture. 818 // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] 819 // CHECK: alloca i[[SZ]], 820 // CHECK: [[LOCAL_AA:%.+]] = alloca i[[SZ]] 821 // CHECK: [[LOCAL_AAA:%.+]] = alloca i[[SZ]] 822 // CHECK: [[LOCAL_B:%.+]] = alloca [10 x i32]* 823 // CHECK: [[LOCAL_A_CASTED:%.+]] = alloca i[[SZ]] 824 // CHECK: alloca i[[SZ]], 825 // CHECK: [[LOCAL_AA_CASTED:%.+]] = alloca i[[SZ]] 826 // CHECK: [[LOCAL_AAA_CASTED:%.+]] = alloca i[[SZ]] 827 // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] 828 // CHECK-DAG: store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]] 829 // CHECK-DAG: store i[[SZ]] [[ARG_AAA:%.+]], i[[SZ]]* [[LOCAL_AAA]] 830 // CHECK-DAG: store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]] 831 // Store captures in the context. 832 // CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* 833 // CHECK-DAG: [[CONV_AAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16* 834 // CHECK-DAG: [[CONV_AAAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AAA]] to i8* 835 // CHECK-DAG: [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]], 836 837 // CHECK-64-DAG:[[CONV_A:%.+]] = load i32, i32* [[CONV_AP]] 838 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_A_CASTED]] to i32* 839 // CHECK-64-DAG:store i32 [[CONV_A]], i32* [[CONV]], align 840 // CHECK-32-DAG:[[LOCAL_AV:%.+]] = load i32, i32* [[LOCAL_A]] 841 // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align 842 // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], 843 844 // CHECK-DAG: [[CONV_AA:%.+]] = load i16, i16* [[CONV_AAP]] 845 // CHECK-DAG: [[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA_CASTED]] to i16* 846 // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align 847 // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], 848 849 // CHECK-DAG: [[CONV_AAA:%.+]] = load i8, i8* [[CONV_AAAP]] 850 // CHECK-DAG: [[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_AAA_CASTED]] to i8* 851 // CHECK-DAG: store i8 [[CONV_AAA]], i8* [[CONV]], align 852 // CHECK-DAG: [[REF_AAA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AAA_CASTED]], 853 854 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] {{.+}}, i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) 855 // 856 // 857 // CHECK: define internal {{.*}}void [[OMP_OUTLINED6]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) 858 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 859 860 // CHECK: define internal void [[HVT5]] 861 // Create local storage for each capture. 862 // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] 863 // CHECK: [[LOCAL_AA:%.+]] = alloca i[[SZ]] 864 // CHECK: [[LOCAL_B:%.+]] = alloca [10 x i32]* 865 // CHECK: [[LOCAL_A_CASTED:%.+]] = alloca i[[SZ]] 866 // CHECK: [[LOCAL_AA_CASTED:%.+]] = alloca i[[SZ]] 867 // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] 868 // CHECK-DAG: store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]] 869 // CHECK-DAG: store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]] 870 // Store captures in the context. 871 // CHECK-64-DAG:[[CONV_AP:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* 872 // CHECK-DAG: [[CONV_AAP:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16* 873 // CHECK-DAG: [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]], 874 875 // CHECK-64-DAG:[[CONV_A:%.+]] = load i32, i32* [[CONV_AP]] 876 // CHECK-64-DAG:[[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_A_CASTED]] to i32* 877 // CHECK-64-DAG:store i32 [[CONV_A]], i32* [[CONV]], align 878 // CHECK-32-DAG:[[LOCAL_AV:%.+]] = load i32, i32* [[LOCAL_A]] 879 // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align 880 // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], 881 882 // CHECK-DAG: [[CONV_AA:%.+]] = load i16, i16* [[CONV_AAP]] 883 // CHECK-DAG: [[CONV:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA_CASTED]] to i16* 884 // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align 885 // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], 886 887 // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) 888 // 889 // 890 // CHECK: define internal {{.*}}void [[OMP_OUTLINED7]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) 891 // To reduce complexity, we're only going as far as validating the signature of the outlined parallel function. 892 893 #endif 894