179 lines · cpp
1// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _2// Test target codegen - host bc file has to be created first.3// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc4// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -disable-llvm-optzns | FileCheck %s --check-prefix=CHECK5// expected-no-diagnostics6#ifndef HEADER7#define HEADER8 9template<typename tx>10tx ftemplate(int n) {11 tx b[10];12 13 #pragma omp target14 {15 tx d = n;16 #pragma omp parallel for17 for(int i=0; i<10; i++) {18 b[i] += d;19 }20 b[3] += 1;21 }22 23 return b[3];24}25 26int bar(int n){27 int a = 0;28 29 a += ftemplate<int>(n);30 31 return a;32}33 34#endif35// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l1336// CHECK-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], i64 noundef [[N:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0:[0-9]+]] {37// CHECK-NEXT: entry:38// CHECK-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 839// CHECK-NEXT: [[N_ADDR:%.*]] = alloca i64, align 840// CHECK-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 841// CHECK-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [2 x ptr], align 842// CHECK-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 843// CHECK-NEXT: store i64 [[N]], ptr [[N_ADDR]], align 844// CHECK-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 845// CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 846// CHECK-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l13_kernel_environment, ptr [[DYN_PTR]])47// CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -148// CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]49// CHECK: user_code.entry:50// CHECK-NEXT: [[D:%.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 4)51// CHECK-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])52// CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 453// CHECK-NEXT: store i32 [[TMP3]], ptr [[D]], align 454// CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 055// CHECK-NEXT: store ptr [[TMP0]], ptr [[TMP4]], align 856// CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 157// CHECK-NEXT: store ptr [[D]], ptr [[TMP5]], align 858// CHECK-NEXT: call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l13_omp_outlined, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l13_omp_outlined_wrapper, ptr [[CAPTURED_VARS_ADDRS]], i64 2)59// CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 360// CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr [[ARRAYIDX]], align 461// CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP6]], 162// CHECK-NEXT: store i32 [[ADD]], ptr [[ARRAYIDX]], align 463// CHECK-NEXT: call void @__kmpc_free_shared(ptr [[D]], i64 4)64// CHECK-NEXT: call void @__kmpc_target_deinit()65// CHECK-NEXT: ret void66// CHECK: worker.exit:67// CHECK-NEXT: ret void68//69//70// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l13_omp_outlined71// CHECK-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[D:%.*]]) #[[ATTR2:[0-9]+]] {72// CHECK-NEXT: entry:73// CHECK-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 874// CHECK-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 875// CHECK-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 876// CHECK-NEXT: [[D_ADDR:%.*]] = alloca ptr, align 877// CHECK-NEXT: [[DOTOMP_IV:%.*]] = alloca i32, align 478// CHECK-NEXT: [[TMP:%.*]] = alloca i32, align 479// CHECK-NEXT: [[DOTOMP_LB:%.*]] = alloca i32, align 480// CHECK-NEXT: [[DOTOMP_UB:%.*]] = alloca i32, align 481// CHECK-NEXT: [[DOTOMP_STRIDE:%.*]] = alloca i32, align 482// CHECK-NEXT: [[DOTOMP_IS_LAST:%.*]] = alloca i32, align 483// CHECK-NEXT: [[I:%.*]] = alloca i32, align 484// CHECK-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 885// CHECK-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 886// CHECK-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 887// CHECK-NEXT: store ptr [[D]], ptr [[D_ADDR]], align 888// CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 889// CHECK-NEXT: [[TMP1:%.*]] = load ptr, ptr [[D_ADDR]], align 890// CHECK-NEXT: store i32 0, ptr [[DOTOMP_LB]], align 491// CHECK-NEXT: store i32 9, ptr [[DOTOMP_UB]], align 492// CHECK-NEXT: store i32 1, ptr [[DOTOMP_STRIDE]], align 493// CHECK-NEXT: store i32 0, ptr [[DOTOMP_IS_LAST]], align 494// CHECK-NEXT: [[TMP2:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 895// CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[TMP2]], align 496// CHECK-NEXT: call void @__kmpc_for_static_init_4(ptr @[[GLOB2:[0-9]+]], i32 [[TMP3]], i32 33, ptr [[DOTOMP_IS_LAST]], ptr [[DOTOMP_LB]], ptr [[DOTOMP_UB]], ptr [[DOTOMP_STRIDE]], i32 1, i32 1)97// CHECK-NEXT: br label [[OMP_DISPATCH_COND:%.*]]98// CHECK: omp.dispatch.cond:99// CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4100// CHECK-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP4]], 9101// CHECK-NEXT: br i1 [[CMP]], label [[COND_TRUE:%.*]], label [[COND_FALSE:%.*]]102// CHECK: cond.true:103// CHECK-NEXT: br label [[COND_END:%.*]]104// CHECK: cond.false:105// CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4106// CHECK-NEXT: br label [[COND_END]]107// CHECK: cond.end:108// CHECK-NEXT: [[COND:%.*]] = phi i32 [ 9, [[COND_TRUE]] ], [ [[TMP5]], [[COND_FALSE]] ]109// CHECK-NEXT: store i32 [[COND]], ptr [[DOTOMP_UB]], align 4110// CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4111// CHECK-NEXT: store i32 [[TMP6]], ptr [[DOTOMP_IV]], align 4112// CHECK-NEXT: [[TMP7:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4113// CHECK-NEXT: [[TMP8:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4114// CHECK-NEXT: [[CMP1:%.*]] = icmp sle i32 [[TMP7]], [[TMP8]]115// CHECK-NEXT: br i1 [[CMP1]], label [[OMP_DISPATCH_BODY:%.*]], label [[OMP_DISPATCH_END:%.*]]116// CHECK: omp.dispatch.body:117// CHECK-NEXT: br label [[OMP_INNER_FOR_COND:%.*]]118// CHECK: omp.inner.for.cond:119// CHECK-NEXT: [[TMP9:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4120// CHECK-NEXT: [[TMP10:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4121// CHECK-NEXT: [[CMP2:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]]122// CHECK-NEXT: br i1 [[CMP2]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]123// CHECK: omp.inner.for.body:124// CHECK-NEXT: [[TMP11:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4125// CHECK-NEXT: [[MUL:%.*]] = mul nsw i32 [[TMP11]], 1126// CHECK-NEXT: [[ADD:%.*]] = add nsw i32 0, [[MUL]]127// CHECK-NEXT: store i32 [[ADD]], ptr [[I]], align 4128// CHECK-NEXT: [[TMP12:%.*]] = load i32, ptr [[TMP1]], align 4129// CHECK-NEXT: [[TMP13:%.*]] = load i32, ptr [[I]], align 4130// CHECK-NEXT: [[IDXPROM:%.*]] = sext i32 [[TMP13]] to i64131// CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 [[IDXPROM]]132// CHECK-NEXT: [[TMP14:%.*]] = load i32, ptr [[ARRAYIDX]], align 4133// CHECK-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP14]], [[TMP12]]134// CHECK-NEXT: store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4135// CHECK-NEXT: br label [[OMP_BODY_CONTINUE:%.*]]136// CHECK: omp.body.continue:137// CHECK-NEXT: br label [[OMP_INNER_FOR_INC:%.*]]138// CHECK: omp.inner.for.inc:139// CHECK-NEXT: [[TMP15:%.*]] = load i32, ptr [[DOTOMP_IV]], align 4140// CHECK-NEXT: [[ADD4:%.*]] = add nsw i32 [[TMP15]], 1141// CHECK-NEXT: store i32 [[ADD4]], ptr [[DOTOMP_IV]], align 4142// CHECK-NEXT: br label [[OMP_INNER_FOR_COND]]143// CHECK: omp.inner.for.end:144// CHECK-NEXT: br label [[OMP_DISPATCH_INC:%.*]]145// CHECK: omp.dispatch.inc:146// CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[DOTOMP_LB]], align 4147// CHECK-NEXT: [[TMP17:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4148// CHECK-NEXT: [[ADD5:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]149// CHECK-NEXT: store i32 [[ADD5]], ptr [[DOTOMP_LB]], align 4150// CHECK-NEXT: [[TMP18:%.*]] = load i32, ptr [[DOTOMP_UB]], align 4151// CHECK-NEXT: [[TMP19:%.*]] = load i32, ptr [[DOTOMP_STRIDE]], align 4152// CHECK-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP18]], [[TMP19]]153// CHECK-NEXT: store i32 [[ADD6]], ptr [[DOTOMP_UB]], align 4154// CHECK-NEXT: br label [[OMP_DISPATCH_COND]]155// CHECK: omp.dispatch.end:156// CHECK-NEXT: call void @__kmpc_for_static_fini(ptr @[[GLOB2]], i32 [[TMP3]])157// CHECK-NEXT: ret void158//159//160// CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l13_omp_outlined_wrapper161// CHECK-SAME: (i16 noundef zeroext [[TMP0:%.*]], i32 noundef [[TMP1:%.*]]) #[[ATTR4:[0-9]+]] {162// CHECK-NEXT: entry:163// CHECK-NEXT: [[DOTADDR:%.*]] = alloca i16, align 2164// CHECK-NEXT: [[DOTADDR1:%.*]] = alloca i32, align 4165// CHECK-NEXT: [[DOTZERO_ADDR:%.*]] = alloca i32, align 4166// CHECK-NEXT: [[GLOBAL_ARGS:%.*]] = alloca ptr, align 8167// CHECK-NEXT: store i16 [[TMP0]], ptr [[DOTADDR]], align 2168// CHECK-NEXT: store i32 [[TMP1]], ptr [[DOTADDR1]], align 4169// CHECK-NEXT: store i32 0, ptr [[DOTZERO_ADDR]], align 4170// CHECK-NEXT: call void @__kmpc_get_shared_variables(ptr [[GLOBAL_ARGS]])171// CHECK-NEXT: [[TMP2:%.*]] = load ptr, ptr [[GLOBAL_ARGS]], align 8172// CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds ptr, ptr [[TMP2]], i64 0173// CHECK-NEXT: [[TMP4:%.*]] = load ptr, ptr [[TMP3]], align 8174// CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds ptr, ptr [[TMP2]], i64 1175// CHECK-NEXT: [[TMP6:%.*]] = load ptr, ptr [[TMP5]], align 8176// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l13_omp_outlined(ptr [[DOTADDR1]], ptr [[DOTZERO_ADDR]], ptr [[TMP4]], ptr [[TMP6]]) #[[ATTR3:[0-9]+]]177// CHECK-NEXT: ret void178//179