1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -S -passes=openmp-opt-cgscc -aa-pipeline=basic-aa -openmp-hide-memory-transfer-latency < %s | FileCheck %s 3 4target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8:9" 5 6@.__omp_offloading_heavyComputation.region_id = weak constant i8 0 7@.offload_maptypes. = private unnamed_addr constant [2 x i64] [i64 35, i64 35] 8 9%struct.ident_t = type { i32, i32, i32, i32, ptr } 10 11@.str = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00", align 1 12@0 = private unnamed_addr global %struct.ident_t { i32 0, i32 2, i32 0, i32 0, ptr @.str }, align 8 13 14;int heavyComputation(ptr a, unsigned size) { 15; int random = rand() % 7; 16; 17; //#pragma omp target data map(a[0:size], size) 18; ptr args[2]; 19; args[0] = &a; 20; args[1] = &size; 21; __tgt_target_data_begin(..., args, ...) 22; 23; #pragma omp target teams 24; for (int i = 0; i < size; ++i) { 25; a[i] = ++aptr 3.141624; 26; } 27; 28; return random; 29;} 30define dso_local i32 @heavyComputation(ptr %a, i32 %size) { 31; CHECK-LABEL: @heavyComputation( 32; CHECK-NEXT: entry: 33; CHECK-NEXT: [[SIZE_ADDR:%.*]] = alloca i32, align 4, addrspace(5) 34; CHECK-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [2 x ptr], align 8, addrspace(5) 35; CHECK-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [2 x ptr], align 8, addrspace(5) 36; CHECK-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [2 x i64], align 8, addrspace(5) 37; CHECK-NEXT: [[HANDLE:%.*]] = alloca [[STRUCT___TGT_ASYNC_INFO:%.*]], align 8, addrspace(5) 38; CHECK-NEXT: [[TMP0:%.*]] = addrspacecast ptr addrspace(5) [[HANDLE]] to ptr 39; CHECK-NEXT: store i32 [[SIZE:%.*]], ptr addrspace(5) [[SIZE_ADDR]], align 4 40; CHECK-NEXT: [[CALL:%.*]] = tail call i32 (...) @rand() 41; CHECK-NEXT: [[CONV:%.*]] = zext i32 [[SIZE]] to i64 42; CHECK-NEXT: [[SHL:%.*]] = shl nuw nsw i64 [[CONV]], 3 43; CHECK-NEXT: store ptr [[A:%.*]], ptr addrspace(5) [[DOTOFFLOAD_BASEPTRS]], align 8 44; CHECK-NEXT: store ptr [[A]], ptr addrspace(5) [[DOTOFFLOAD_PTRS]], align 8 45; CHECK-NEXT: store i64 [[SHL]], ptr addrspace(5) [[DOTOFFLOAD_SIZES]], align 8 46; CHECK-NEXT: [[GEP0:%.*]] = getelementptr inbounds [2 x ptr], ptr addrspace(5) [[DOTOFFLOAD_BASEPTRS]], i64 0, i64 1 47; CHECK-NEXT: store ptr addrspace(5) [[SIZE_ADDR]], ptr addrspace(5) [[GEP0]], align 8 48; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [2 x ptr], ptr addrspace(5) [[DOTOFFLOAD_PTRS]], i64 0, i64 1 49; CHECK-NEXT: store ptr addrspace(5) [[SIZE_ADDR]], ptr addrspace(5) [[GEP1]], align 8 50; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds [2 x i64], ptr addrspace(5) [[DOTOFFLOAD_SIZES]], i64 0, i64 1 51; CHECK-NEXT: store i64 4, ptr addrspace(5) [[GEP2]], align 8 52; CHECK-NEXT: [[DOTOFFLOAD_BASEPTRS_FLAT:%.*]] = addrspacecast ptr addrspace(5) [[DOTOFFLOAD_BASEPTRS]] to ptr 53; CHECK-NEXT: [[DOTOFFLOAD_PTRS_FLAT:%.*]] = addrspacecast ptr addrspace(5) [[DOTOFFLOAD_PTRS]] to ptr 54; CHECK-NEXT: [[DOTOFFLOAD_SIZES_FLAT:%.*]] = addrspacecast ptr addrspace(5) [[DOTOFFLOAD_SIZES]] to ptr 55; CHECK-NEXT: call void @__tgt_target_data_begin_mapper_issue(ptr @[[GLOB0:[0-9]+]], i64 -1, i32 2, ptr [[DOTOFFLOAD_BASEPTRS_FLAT]], ptr [[DOTOFFLOAD_PTRS_FLAT]], ptr [[DOTOFFLOAD_SIZES_FLAT]], ptr @.offload_maptypes., ptr null, ptr null, ptr [[TMP0]]) 56; CHECK-NEXT: [[REM:%.*]] = srem i32 [[CALL]], 7 57; CHECK-NEXT: call void @__tgt_target_data_begin_mapper_wait(i64 -1, ptr [[TMP0]]) 58; CHECK-NEXT: call void @__tgt_target_data_end_mapper(ptr @[[GLOB0]], i64 -1, i32 2, ptr nonnull [[DOTOFFLOAD_BASEPTRS_FLAT]], ptr nonnull [[DOTOFFLOAD_PTRS_FLAT]], ptr nonnull [[DOTOFFLOAD_SIZES_FLAT]], ptr @.offload_maptypes., ptr null, ptr null) 59; CHECK-NEXT: ret i32 [[REM]] 60; 61entry: 62 %size.addr = alloca i32, align 4, addrspace(5) 63 %.offload_baseptrs = alloca [2 x ptr], align 8, addrspace(5) 64 %.offload_ptrs = alloca [2 x ptr], align 8, addrspace(5) 65 %.offload_sizes = alloca [2 x i64], align 8, addrspace(5) 66 store i32 %size, ptr addrspace(5) %size.addr, align 4 67 %call = tail call i32 (...) @rand() 68 %conv = zext i32 %size to i64 69 %shl = shl nuw nsw i64 %conv, 3 70 store ptr %a, ptr addrspace(5) %.offload_baseptrs, align 8 71 store ptr %a, ptr addrspace(5) %.offload_ptrs, align 8 72 store i64 %shl, ptr addrspace(5) %.offload_sizes, align 8 73 %gep0 = getelementptr inbounds [2 x ptr], ptr addrspace(5) %.offload_baseptrs, i64 0, i64 1 74 store ptr addrspace(5) %size.addr, ptr addrspace(5) %gep0, align 8 75 %gep1 = getelementptr inbounds [2 x ptr], ptr addrspace(5) %.offload_ptrs, i64 0, i64 1 76 store ptr addrspace(5) %size.addr, ptr addrspace(5) %gep1, align 8 77 %gep2 = getelementptr inbounds [2 x i64], ptr addrspace(5) %.offload_sizes, i64 0, i64 1 78 store i64 4, ptr addrspace(5) %gep2, align 8 79 %.offload_baseptrs.flat = addrspacecast ptr addrspace(5) %.offload_baseptrs to ptr 80 %.offload_ptrs.flat = addrspacecast ptr addrspace(5) %.offload_ptrs to ptr 81 %.offload_sizes.flat = addrspacecast ptr addrspace(5) %.offload_sizes to ptr 82 call void @__tgt_target_data_begin_mapper(ptr @0, i64 -1, i32 2, ptr nonnull %.offload_baseptrs.flat, ptr nonnull %.offload_ptrs.flat, ptr nonnull %.offload_sizes.flat, ptr @.offload_maptypes., ptr null, ptr null) 83 %rem = srem i32 %call, 7 84 call void @__tgt_target_data_end_mapper(ptr @0, i64 -1, i32 2, ptr nonnull %.offload_baseptrs.flat, ptr nonnull %.offload_ptrs.flat, ptr nonnull %.offload_sizes.flat, ptr @.offload_maptypes., ptr null, ptr null) 85 ret i32 %rem 86} 87 88declare void @__tgt_target_data_begin_mapper(ptr, i64, i32, ptr, ptr, ptr, ptr, ptr, ptr) 89declare void @__tgt_target_data_end_mapper(ptr, i64, i32, ptr, ptr, ptr, ptr, ptr, ptr) 90 91declare dso_local i32 @rand(...) 92 93!llvm.module.flags = !{!0} 94 95!0 = !{i32 7, !"openmp", i32 50} 96