118 lines · plain
1// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s2 3// Test that the block argument to the initialization region of4// omp.declare_reduction gets mapped properly when translating to LLVMIR.5 6module {7 omp.declare_reduction @add_reduction_byref_box_Uxf64 : !llvm.ptr init {8 ^bb0(%arg0: !llvm.ptr):9// test usage of %arg0:10 %11 = llvm.load %arg0 : !llvm.ptr -> !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)>11 omp.yield(%arg0 : !llvm.ptr)12 } combiner {13 ^bb0(%arg0: !llvm.ptr, %arg1: !llvm.ptr):14 omp.yield(%arg0 : !llvm.ptr)15 }16 17 llvm.func internal @_QFPreduce(%arg0: !llvm.ptr {fir.bindc_name = "r"}, %arg1: !llvm.ptr {fir.bindc_name = "r2"}) attributes {sym_visibility = "private"} {18 %8 = llvm.mlir.constant(1 : i32) : i3219 %9 = llvm.mlir.constant(10 : i32) : i3220 %10 = llvm.mlir.constant(0 : i32) : i3221 %83 = llvm.mlir.constant(1 : i64) : i6422 %84 = llvm.alloca %83 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> : (i64) -> !llvm.ptr23 %86 = llvm.mlir.constant(1 : i64) : i6424 %87 = llvm.alloca %86 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> : (i64) -> !llvm.ptr25// test multiple reduction variables to ensure they don't intefere with each other26// when inlining the reduction init region multiple times27 omp.parallel reduction(byref @add_reduction_byref_box_Uxf64 %84 -> %arg3, byref @add_reduction_byref_box_Uxf64 %87 -> %arg4 : !llvm.ptr, !llvm.ptr) {28 omp.terminator29 }30 llvm.return31 }32}33 34// CHECK-LABEL: define internal void @_QFPreduce35// CHECK: %[[VAL_0:.*]] = alloca { ptr, ptr }, align 836// CHECK: %[[VAL_1:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 837// CHECK: %[[VAL_2:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 838// CHECK: br label %[[VAL_3:.*]]39// CHECK: entry: ; preds = %[[VAL_4:.*]]40// CHECK: %[[VAL_5:.*]] = call i32 @__kmpc_global_thread_num(ptr @1)41// CHECK: br label %[[VAL_6:.*]]42// CHECK: omp_parallel: ; preds = %[[VAL_3]]43// CHECK: %[[VAL_7:.*]] = getelementptr { ptr, ptr }, ptr %[[VAL_0]], i32 0, i32 044// CHECK: store ptr %[[VAL_1]], ptr %[[VAL_7]], align 845// CHECK: %[[VAL_8:.*]] = getelementptr { ptr, ptr }, ptr %[[VAL_0]], i32 0, i32 146// CHECK: store ptr %[[VAL_2]], ptr %[[VAL_8]], align 847// CHECK: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @1, i32 1, ptr @_QFPreduce..omp_par, ptr %[[VAL_0]])48// CHECK: br label %[[VAL_9:.*]]49// CHECK: omp.par.exit: ; preds = %[[VAL_6]]50// CHECK: ret void51// CHECK: [[PAR_ENTRY:omp.par.entry]]:52// CHECK: %[[VAL_11:.*]] = getelementptr { ptr, ptr }, ptr %[[VAL_12:.*]], i32 0, i32 053// CHECK: %[[VAL_13:.*]] = load ptr, ptr %[[VAL_11]], align 854// CHECK: %[[VAL_14:.*]] = getelementptr { ptr, ptr }, ptr %[[VAL_12]], i32 0, i32 155// CHECK: %[[VAL_15:.*]] = load ptr, ptr %[[VAL_14]], align 856// CHECK: %[[VAL_16:.*]] = alloca i32, align 457// CHECK: %[[VAL_17:.*]] = load i32, ptr %[[VAL_18:.*]], align 458// CHECK: store i32 %[[VAL_17]], ptr %[[VAL_16]], align 459// CHECK: %[[VAL_19:.*]] = load i32, ptr %[[VAL_16]], align 460// CHECK: %[[VAL_21:.*]] = alloca ptr, align 861// CHECK: %[[VAL_23:.*]] = alloca ptr, align 862// CHECK: %[[VAL_24:.*]] = alloca [2 x ptr], align 863// CHECK: br label %[[VAL_25:.*]]64 65// CHECK: [[VAL_25]]:66// CHECK: br label %[[PAR_REG:omp.par.region]]67 68// CHECK: [[PAR_REG]]: ; preds = %[[VAL_25]]69// CHECK: br label %[[INIT_LABEL:.*]]70// CHECK: [[INIT_LABEL]]:71// CHECK: %[[VAL_20:.*]] = load { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, ptr %[[VAL_13]], align 872// CHECK: store ptr %[[VAL_13]], ptr %[[VAL_21]], align 873// CHECK: %[[VAL_22:.*]] = load { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, ptr %[[VAL_15]], align 874// CHECK: store ptr %[[VAL_15]], ptr %[[VAL_23]], align 875// CHECK: br label %[[VAL_27:.*]]76// CHECK: omp.par.region1: ; preds = %[[INIT_LABEL]]77// CHECK: br label %[[VAL_28:.*]]78// CHECK: omp.region.cont: ; preds = %[[VAL_27]]79// CHECK: %[[VAL_29:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_24]], i64 0, i64 080// CHECK: store ptr %[[VAL_21]], ptr %[[VAL_29]], align 881// CHECK: %[[VAL_30:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_24]], i64 0, i64 182// CHECK: store ptr %[[VAL_23]], ptr %[[VAL_30]], align 883// CHECK: %[[VAL_31:.*]] = call i32 @__kmpc_global_thread_num(ptr @1)84// CHECK: %[[VAL_32:.*]] = call i32 @__kmpc_reduce(ptr @1, i32 %[[VAL_31]], i32 2, i64 16, ptr %[[VAL_24]], ptr @.omp.reduction.func, ptr @.gomp_critical_user_.reduction.var)85// CHECK: switch i32 %[[VAL_32]], label %[[VAL_33:.*]] [86// CHECK: i32 1, label %[[VAL_34:.*]]87// CHECK: i32 2, label %[[VAL_35:.*]]88// CHECK: ]89// CHECK: reduce.switch.atomic: ; preds = %[[VAL_28]]90// CHECK: unreachable91// CHECK: reduce.switch.nonatomic: ; preds = %[[VAL_28]]92// CHECK: %[[VAL_36:.*]] = load ptr, ptr %[[VAL_21]], align 893// CHECK: %[[VAL_37:.*]] = load ptr, ptr %[[VAL_23]], align 894// CHECK: call void @__kmpc_end_reduce(ptr @1, i32 %[[VAL_31]], ptr @.gomp_critical_user_.reduction.var)95// CHECK: br label %[[VAL_33]]96// CHECK: reduce.finalize: ; preds = %[[VAL_34]], %[[VAL_28]]97// CHECK: br label %[[VAL_38:.*]]98// CHECK: omp.par.pre_finalize: ; preds = %[[VAL_33]]99// CHECK: br label %[[FINI:.*]]100// CHECK: [[FINI]]:101// CHECK: br label %[[VAL_39:.*]]102// CHECK: omp.par.exit.exitStub: ; preds = %[[FINI]]103// CHECK: ret void104// CHECK: %[[VAL_40:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_41:.*]], i64 0, i64 0105// CHECK: %[[VAL_42:.*]] = load ptr, ptr %[[VAL_40]], align 8106// CHECK: %[[VAL_43:.*]] = load ptr, ptr %[[VAL_42]], align 8107// CHECK: %[[VAL_44:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_45:.*]], i64 0, i64 0108// CHECK: %[[VAL_46:.*]] = load ptr, ptr %[[VAL_44]], align 8109// CHECK: %[[VAL_47:.*]] = load ptr, ptr %[[VAL_46]], align 8110// CHECK: %[[VAL_48:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_41]], i64 0, i64 1111// CHECK: %[[VAL_49:.*]] = load ptr, ptr %[[VAL_48]], align 8112// CHECK: %[[VAL_50:.*]] = load ptr, ptr %[[VAL_49]], align 8113// CHECK: %[[VAL_51:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_45]], i64 0, i64 1114// CHECK: %[[VAL_52:.*]] = load ptr, ptr %[[VAL_51]], align 8115// CHECK: %[[VAL_53:.*]] = load ptr, ptr %[[VAL_52]], align 8116// CHECK: ret void117 118