233 lines · cpp
1// Test host codegen.2// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-643// 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 %s4// 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 -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-645// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-326// 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 %s7// 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 -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-328 9// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s10// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s11// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s12// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s13// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s14// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s15// SIMD-ONLY0-NOT: {{__kmpc|__tgt}}16 17// Test target codegen - host bc file has to be created first.18// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc19// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-6420// 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-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s21// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-6422// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc23// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-3224// 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-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s25// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-3226 27// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc28// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s29// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s30// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s31// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc32// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s33// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s34// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s35// SIMD-ONLY1-NOT: {{__kmpc|__tgt}}36 37// expected-no-diagnostics38#ifndef HEADER39#define HEADER40 41// CHECK-DAG: [[TT:%.+]] = type { i64, i8 }42// CHECK-DAG: [[ENTTY:%.+]] = type { i64, i16, i16, i32, ptr, ptr, i64, i64, ptr }43 44// TCHECK: [[ENTTY:%.+]] = type { i64, i16, i16, i32, ptr, ptr, i64, i64, ptr }45 46// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i64] [i64 0, i64 4]47// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800]48// CHECK-DAG: @{{.*}} = weak constant i8 049 50// TCHECK: @{{.+}} = weak constant [[ENTTY]]51// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]]52// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]]53 54template<typename tx, typename ty>55struct TT{56 tx X;57 ty Y;58};59 60int global;61extern int global;62 63// CHECK: define {{.*}}[[FOO:@.+]](64int foo(int n) {65 int a = 0;66 short aa = 0;67 float b[10];68 float bn[n];69 double c[5][10];70 double cn[5][n];71 TT<long long, char> d;72 static long *plocal;73 74// CHECK: [[ADD:%.+]] = add nsw i3275// CHECK: store i32 [[ADD]], ptr [[DEVICE_CAP:%.+]],76// CHECK: [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 077// CHECK: [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],78// CHECK: store i32 [[DEV]], ptr [[GEP]],79// CHECK: [[TASK:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr @1, i32 [[GTID:%.+]], i32 1, i[[SZ:32|64]] {{20|40}}, i[[SZ]] 4, ptr [[TASK_ENTRY0:@.+]])80// CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 081// CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 182// CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 283// CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 384// CHECK: call void @__kmpc_omp_taskwait_deps_51(ptr @1, i32 [[GTID]], i32 4, ptr %{{.+}}, i32 0, ptr null, i32 0)85// CHECK: call void @__kmpc_omp_task_begin_if0(ptr @1, i32 [[GTID]], ptr [[TASK]])86// CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], ptr [[TASK]])87// CHECK: call void @__kmpc_omp_task_complete_if0(ptr @1, i32 [[GTID]], ptr [[TASK]])88#pragma omp target parallel device(global + a) depend(in \89 : global) depend(out \90 : a, b, cn[4])91 {92 }93 94 // CHECK: [[ADD:%.+]] = add nsw i3295 // CHECK: store i32 [[ADD]], ptr [[DEVICE_CAP:%.+]],96 97 // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 098 // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]]99 // CHECK: [[THEN]]:100 // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP:%.+]], i32 0, i32 0101 // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P:%.+]], i32 0, i32 0102 // CHECK-DAG: store ptr [[BP0:%[^,]+]], ptr [[BPADDR0]]103 // CHECK-DAG: store ptr [[BP0]], ptr [[PADDR0]]104 105 // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 1106 // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 1107 // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], ptr [[BPADDR1]]108 // CHECK-DAG: store i[[SZ]] [[BP1]], ptr [[PADDR1]]109 // CHECK-DAG: getelementptr inbounds [2 x ptr], ptr [[BP]], i32 0, i32 0110 // CHECK-DAG: getelementptr inbounds [2 x ptr], ptr [[P]], i32 0, i32 0111 // CHECK: [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2112 // CHECK: [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],113 // CHECK: store i32 [[DEV]], ptr [[GEP]],114 // CHECK: [[DEV1:%.+]] = load i32, ptr [[DEVICE_CAP]],115 // CHECK: [[DEV2:%.+]] = sext i32 [[DEV1]] to i64116 117 // CHECK: [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr @{{.*}}, i32 [[GTID]], i32 1, i[[SZ]] {{104|60}}, i[[SZ]] {{16|12}}, ptr [[TASK_ENTRY1_:@.+]], i64 [[DEV2]])118 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 0119 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 1120 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 2121 // CHECK: call i32 @__kmpc_omp_task_with_deps(ptr @1, i32 [[GTID]], ptr [[TASK]], i32 3, ptr %{{.+}}, i32 0, ptr null)122 // CHECK: br label %[[EXIT:.+]]123 124 // CHECK: [[ELSE]]:125 // CHECK-NOT: getelementptr inbounds [2 x ptr], ptr126 // CHECK: [[GEP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2127 // CHECK: [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],128 // CHECK: store i32 [[DEV]], ptr [[GEP]],129 // CHECK: [[DEV1:%.+]] = load i32, ptr [[DEVICE_CAP]],130 // CHECK: [[DEV2:%.+]] = sext i32 [[DEV1]] to i64131 132 // CHECK: [[TASK:%.+]] = call ptr @__kmpc_omp_target_task_alloc(ptr @1, i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, ptr [[TASK_ENTRY1__:@.+]], i64 [[DEV2]])133 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 0134 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 1135 // CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 2136 // CHECK: call i32 @__kmpc_omp_task_with_deps(ptr @1, i32 [[GTID]], ptr [[TASK]], i32 3, ptr %{{.+}}, i32 0, ptr null)137 // CHECK: br label %[[EXIT:.+]]138 // CHECK: [[EXIT]]:139 140#pragma omp target parallel device(global + a) nowait depend(inout \141 : global, a, bn) if (target \142 : a)143 {144 static int local1;145 *plocal = global;146 local1 = global;147 }148 149// CHECK: [[TASK:%.+]] = call ptr @__kmpc_omp_task_alloc(ptr @1, i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, ptr [[TASK_ENTRY2:@.+]])150// CHECK: getelementptr %struct.kmp_depend_info, ptr %{{.+}}, i[[SZ]] 0151// CHECK: call void @__kmpc_omp_taskwait_deps_51(ptr @1, i32 [[GTID]], i32 1, ptr %{{.+}}, i32 0, ptr null, i32 0)152// CHECK: call void @__kmpc_omp_task_begin_if0(ptr @1, i32 [[GTID]], ptr [[TASK]])153// CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], ptr [[TASK]])154// CHECK: call void @__kmpc_omp_task_complete_if0(ptr @1, i32 [[GTID]], ptr [[TASK]])155#pragma omp target parallel if (0) firstprivate(global) depend(out \156 : global)157 {158 global += 1;159 }160 161 return a;162}163 164// Check that the offloading functions are emitted and that the arguments are165// correct and loaded correctly for the target regions in foo().166 167// CHECK: define internal void [[HVT0:@.+]]()168 169// CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, ptr noalias noundef %1)170// CHECK: store ptr null, ptr %171// CHECK: [[DEVICE_CAP:%.+]] = load ptr,172// CHECK: [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],173// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64174// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE]], i32 1, i32 0, ptr @.{{.+}}.region_id, ptr %{{.+}})175// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0176// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]177// CHECK: [[FAIL]]178// CHECK: call void [[HVT0]]()179// CHECK-NEXT: br label %[[END]]180// CHECK: [[END]]181// CHECK: ret i32 0182 183// CHECK: define internal void [[HVT1:@.+]](ptr noundef %{{.+}}, i[[SZ]] noundef %{{.+}})184 185// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, ptr noalias noundef %1)186// CHECK: call void {{%.*}}(187// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2188// CHECK: [[DEV:%.+]] = load i32, ptr [[DEVICE_CAP]],189// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64190// CHECK: [[RET:%.+]] = call i32 @__tgt_target_kernel(ptr @{{.+}}, i64 [[DEVICE]], i32 1, i32 0, ptr @.{{.+}}.region_id, ptr %{{.+}})191 192// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0193// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]]194// CHECK: [[FAIL]]195// CHECK: [[BP0:%.+]] = load ptr, ptr %196// CHECK: [[BP1_I32:%.+]] = load i32, ptr @197// CHECK-64: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],198// CHECK-32: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],199// CHECK: [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]],200// CHECK: call void [[HVT1]](ptr [[BP0]], i[[SZ]] [[BP1]])201// CHECK-NEXT: br label %[[END]]202// CHECK: [[END]]203// CHECK: ret i32 0204 205// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, ptr noalias noundef %1)206// CHECK: call void {{%.*}}(207// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds nuw %{{.+}}, ptr %{{.+}}, i32 0, i32 2208// CHECK: [[BP0:%.+]] = load ptr, ptr %209// CHECK: [[BP1_I32:%.+]] = load i32, ptr @210// CHECK-64: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],211// CHECK-32: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],212// CHECK: [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]],213// CHECK: call void [[HVT1]](ptr [[BP0]], i[[SZ]] [[BP1]])214// CHECK: ret i32 0215 216// CHECK: define internal void [[HVT2:@.+]](i[[SZ]] noundef %{{.+}})217// Create stack storage and store argument in there.218// CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align219// CHECK: store i[[SZ]] %{{.+}}, ptr [[AA_ADDR]], align220// CHECK-64: load i32, ptr [[AA_ADDR]], align221// CHECK-32: load i32, ptr [[AA_ADDR]], align222 223// CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, ptr noalias noundef %1)224// CHECK: call void {{%.*}}(225// CHECK: [[BP1_I32:%.+]] = load i32, ptr %226// CHECK-64: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],227// CHECK-32: store i32 [[BP1_I32]], ptr [[BP1_PTR:%[^,]+]],228// CHECK: [[BP1:%.+]] = load i[[SZ]], ptr [[BP1_PTR]],229// CHECK: call void [[HVT2]](i[[SZ]] [[BP1]])230// CHECK: ret i32 0231 232#endif233