1// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s 2 3// Test that the block argument to the initialization region of 4// 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) : i32 19 %9 = llvm.mlir.constant(10 : i32) : i32 20 %10 = llvm.mlir.constant(0 : i32) : i32 21 %83 = llvm.mlir.constant(1 : i64) : i64 22 %84 = llvm.alloca %83 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> : (i64) -> !llvm.ptr 23 %86 = llvm.mlir.constant(1 : i64) : i64 24 %87 = llvm.alloca %86 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> : (i64) -> !llvm.ptr 25// test multiple reduction variables to ensure they don't intefere with each other 26// when inlining the reduction init region multiple times 27 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.terminator 29 } 30 llvm.return 31 } 32} 33 34// CHECK-LABEL: define internal void @_QFPreduce 35// CHECK: %[[VAL_0:.*]] = alloca { ptr, ptr }, align 8 36// CHECK: %[[VAL_1:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 8 37// CHECK: %[[VAL_2:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 8 38// 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 0 44// CHECK: store ptr %[[VAL_1]], ptr %[[VAL_7]], align 8 45// CHECK: %[[VAL_8:.*]] = getelementptr { ptr, ptr }, ptr %[[VAL_0]], i32 0, i32 1 46// CHECK: store ptr %[[VAL_2]], ptr %[[VAL_8]], align 8 47// 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.outlined.exit: ; preds = %[[VAL_6]] 50// CHECK: br label %[[VAL_10:.*]] 51// CHECK: omp.par.exit.split: ; preds = %[[VAL_9]] 52// CHECK: ret void 53// CHECK: [[PAR_ENTRY:omp.par.entry]]: 54// CHECK: %[[VAL_11:.*]] = getelementptr { ptr, ptr }, ptr %[[VAL_12:.*]], i32 0, i32 0 55// CHECK: %[[VAL_13:.*]] = load ptr, ptr %[[VAL_11]], align 8 56// CHECK: %[[VAL_14:.*]] = getelementptr { ptr, ptr }, ptr %[[VAL_12]], i32 0, i32 1 57// CHECK: %[[VAL_15:.*]] = load ptr, ptr %[[VAL_14]], align 8 58// CHECK: %[[VAL_16:.*]] = alloca i32, align 4 59// CHECK: %[[VAL_17:.*]] = load i32, ptr %[[VAL_18:.*]], align 4 60// CHECK: store i32 %[[VAL_17]], ptr %[[VAL_16]], align 4 61// CHECK: %[[VAL_19:.*]] = load i32, ptr %[[VAL_16]], align 4 62// CHECK: %[[VAL_21:.*]] = alloca ptr, align 8 63// CHECK: %[[VAL_23:.*]] = alloca ptr, align 8 64// CHECK: %[[VAL_24:.*]] = alloca [2 x ptr], align 8 65// CHECK: br label %[[VAL_25:.*]] 66 67// CHECK: [[VAL_25]]: 68// CHECK: br label %[[PAR_REG:omp.par.region]] 69 70// CHECK: [[PAR_REG]]: ; preds = %[[VAL_25]] 71// CHECK: br label %[[INIT_LABEL:.*]] 72// CHECK: [[INIT_LABEL]]: 73// CHECK: %[[VAL_20:.*]] = load { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, ptr %[[VAL_13]], align 8 74// CHECK: store ptr %[[VAL_13]], ptr %[[VAL_21]], align 8 75// CHECK: %[[VAL_22:.*]] = load { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, ptr %[[VAL_15]], align 8 76// CHECK: store ptr %[[VAL_15]], ptr %[[VAL_23]], align 8 77// CHECK: br label %[[VAL_27:.*]] 78// CHECK: omp.par.region1: ; preds = %[[INIT_LABEL]] 79// CHECK: br label %[[VAL_28:.*]] 80// CHECK: omp.region.cont: ; preds = %[[VAL_27]] 81// CHECK: %[[VAL_29:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_24]], i64 0, i64 0 82// CHECK: store ptr %[[VAL_21]], ptr %[[VAL_29]], align 8 83// CHECK: %[[VAL_30:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_24]], i64 0, i64 1 84// CHECK: store ptr %[[VAL_23]], ptr %[[VAL_30]], align 8 85// CHECK: %[[VAL_31:.*]] = call i32 @__kmpc_global_thread_num(ptr @1) 86// 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) 87// CHECK: switch i32 %[[VAL_32]], label %[[VAL_33:.*]] [ 88// CHECK: i32 1, label %[[VAL_34:.*]] 89// CHECK: i32 2, label %[[VAL_35:.*]] 90// CHECK: ] 91// CHECK: reduce.switch.atomic: ; preds = %[[VAL_28]] 92// CHECK: unreachable 93// CHECK: reduce.switch.nonatomic: ; preds = %[[VAL_28]] 94// CHECK: %[[VAL_36:.*]] = load ptr, ptr %[[VAL_21]], align 8 95// CHECK: %[[VAL_37:.*]] = load ptr, ptr %[[VAL_23]], align 8 96// CHECK: call void @__kmpc_end_reduce(ptr @1, i32 %[[VAL_31]], ptr @.gomp_critical_user_.reduction.var) 97// CHECK: br label %[[VAL_33]] 98// CHECK: reduce.finalize: ; preds = %[[VAL_34]], %[[VAL_28]] 99// CHECK: br label %[[VAL_38:.*]] 100// CHECK: omp.par.pre_finalize: ; preds = %[[VAL_33]] 101// CHECK: br label %[[VAL_39:.*]] 102// CHECK: omp.par.outlined.exit.exitStub: ; preds = %[[VAL_38]] 103// CHECK: ret void 104// CHECK: %[[VAL_40:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_41:.*]], i64 0, i64 0 105// CHECK: %[[VAL_42:.*]] = load ptr, ptr %[[VAL_40]], align 8 106// CHECK: %[[VAL_43:.*]] = load ptr, ptr %[[VAL_42]], align 8 107// CHECK: %[[VAL_44:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_45:.*]], i64 0, i64 0 108// CHECK: %[[VAL_46:.*]] = load ptr, ptr %[[VAL_44]], align 8 109// CHECK: %[[VAL_47:.*]] = load ptr, ptr %[[VAL_46]], align 8 110// CHECK: %[[VAL_48:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_41]], i64 0, i64 1 111// CHECK: %[[VAL_49:.*]] = load ptr, ptr %[[VAL_48]], align 8 112// CHECK: %[[VAL_50:.*]] = load ptr, ptr %[[VAL_49]], align 8 113// CHECK: %[[VAL_51:.*]] = getelementptr inbounds [2 x ptr], ptr %[[VAL_45]], i64 0, i64 1 114// CHECK: %[[VAL_52:.*]] = load ptr, ptr %[[VAL_51]], align 8 115// CHECK: %[[VAL_53:.*]] = load ptr, ptr %[[VAL_52]], align 8 116// CHECK: ret void 117 118