99 lines · plain
1// RUN: mlir-translate --mlir-to-llvmir %s | FileCheck %s2 3llvm.func @init(%arg0: !llvm.ptr {llvm.nocapture}, %arg1: !llvm.ptr {llvm.nocapture}) {4 llvm.return5}6llvm.func @combine(%arg0: !llvm.ptr {llvm.nocapture}, %arg1: !llvm.ptr {llvm.nocapture}) {7 llvm.return8}9llvm.func @cleanup(%arg0: !llvm.ptr {llvm.nocapture}) {10 llvm.return11}12omp.private {type = private} @_QFsimd_reductionEi_private_i32 : i3213omp.declare_reduction @add_reduction_byref_box_2xf32 : !llvm.ptr alloc {14 %0 = llvm.mlir.constant(1 : i64) : i6415 %1 = llvm.alloca %0 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> : (i64) -> !llvm.ptr16 omp.yield(%1 : !llvm.ptr)17} init {18^bb0(%arg0: !llvm.ptr, %arg1: !llvm.ptr):19 llvm.call @init(%arg0, %arg1) : (!llvm.ptr, !llvm.ptr) -> ()20 omp.yield(%arg1 : !llvm.ptr)21} combiner {22^bb0(%arg0: !llvm.ptr, %arg1: !llvm.ptr):23 llvm.call @combine(%arg0, %arg1) : (!llvm.ptr, !llvm.ptr) -> ()24 omp.yield(%arg0 : !llvm.ptr)25} cleanup {26^bb0(%arg0: !llvm.ptr):27 llvm.call @cleanup(%arg0) : (!llvm.ptr) -> ()28 omp.yield29}30llvm.func @_QPsimd_reduction(%arg0: !llvm.ptr {fir.bindc_name = "a", llvm.nocapture}, %arg1: !llvm.ptr {fir.bindc_name = "sum", llvm.nocapture}) {31 %0 = llvm.mlir.constant(1024 : i32) : i3232 %1 = llvm.mlir.constant(1 : i32) : i3233 %2 = llvm.mlir.constant(1 : i64) : i6434 %3 = llvm.alloca %2 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> : (i64) -> !llvm.ptr35 %4 = llvm.alloca %2 x i32 {bindc_name = "i"} : (i64) -> !llvm.ptr36 omp.simd private(@_QFsimd_reductionEi_private_i32 %4 -> %arg2 : !llvm.ptr) reduction(byref @add_reduction_byref_box_2xf32 %3 -> %arg3 : !llvm.ptr) {37 omp.loop_nest (%arg4) : i32 = (%1) to (%0) inclusive step (%1) {38 llvm.store %arg4, %arg2 : i32, !llvm.ptr39 omp.yield40 }41 }42 llvm.return43}44 45// CHECK-LABEL: define void @_QPsimd_reduction46// CHECK: %[[MOLD:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 847// CHECK: %[[ORIG_I:.*]] = alloca i32, i64 1, align 448// CHECK: %[[PRIV_I:.*]] = alloca i32, align 449// CHECK: %[[RED_VAR:.*]] = alloca { ptr, i64, i32, i8, i8, i8, i8, [1 x [3 x i64]] }, i64 1, align 850// CHECK: %[[PTR_RED_VAR:.*]] = alloca ptr, align 851// CHECK: br label %[[VAL_5:.*]]52// CHECK: omp.region.after_alloca: ; preds = %[[VAL_6:.*]]53// CHECK: br label %[[VAL_7:.*]]54// CHECK: entry: ; preds = %[[VAL_5]]55// CHECK: br label %[[VAL_8:.*]]56// CHECK: omp.private.init: ; preds = %[[VAL_7]]57// CHECK: br label %[[VAL_9:.*]]58// CHECK: omp.reduction.init: ; preds = %[[VAL_8]]59// CHECK: store ptr %[[RED_VAR]], ptr %[[PTR_RED_VAR]], align 860// CHECK: call void @init(ptr %[[MOLD]], ptr %[[RED_VAR]])61// CHECK: br label %[[VAL_10:.*]]62// CHECK: omp.simd.region: ; preds = %[[VAL_9]]63// CHECK: br label %[[VAL_11:.*]]64// CHECK: omp_loop.preheader: ; preds = %[[VAL_10]]65// CHECK: br label %[[VAL_12:.*]]66// CHECK: omp_loop.header: ; preds = %[[VAL_13:.*]], %[[VAL_11]]67// CHECK: %[[VAL_14:.*]] = phi i32 [ 0, %[[VAL_11]] ], [ %[[VAL_15:.*]], %[[VAL_13]] ]68// CHECK: br label %[[VAL_16:.*]]69// CHECK: omp_loop.cond: ; preds = %[[VAL_12]]70// CHECK: %[[VAL_17:.*]] = icmp ult i32 %[[VAL_14]], 102471// CHECK: br i1 %[[VAL_17]], label %[[VAL_18:.*]], label %[[VAL_19:.*]]72// CHECK: omp_loop.body: ; preds = %[[VAL_16]]73// CHECK: %[[VAL_20:.*]] = mul i32 %[[VAL_14]], 174// CHECK: %[[VAL_21:.*]] = add i32 %[[VAL_20]], 175// CHECK: br label %[[VAL_22:.*]]76// CHECK: omp.loop_nest.region: ; preds = %[[VAL_18]]77// CHECK: store i32 %[[VAL_21]], ptr %[[PRIV_I]], align 4, !llvm.access.group ![[ACCESS_GROUP:.*]]78// CHECK: br label %[[VAL_23:.*]]79// CHECK: omp.region.cont1: ; preds = %[[VAL_22]]80// CHECK: br label %[[VAL_13]]81// CHECK: omp_loop.inc: ; preds = %[[VAL_23]]82// CHECK: %[[VAL_15]] = add nuw i32 %[[VAL_14]], 183// CHECK: br label %[[VAL_12]], !llvm.loop ![[LOOP:.*]]84// CHECK: omp_loop.exit: ; preds = %[[VAL_16]]85// CHECK: br label %[[VAL_24:.*]]86// CHECK: omp_loop.after: ; preds = %[[VAL_19]]87// CHECK: br label %[[VAL_25:.*]]88// CHECK: omp.region.cont: ; preds = %[[VAL_24]]89// CHECK: %[[RED_VAR2:.*]] = load ptr, ptr %[[PTR_RED_VAR]], align 890// CHECK: call void @combine(ptr %[[MOLD]], ptr %[[RED_VAR2]])91// CHECK: %[[RED_VAR3:.*]] = load ptr, ptr %[[PTR_RED_VAR]], align 892// CHECK: call void @cleanup(ptr %[[RED_VAR3]])93// CHECK: ret void94 95// CHECK: ![[ACCESS_GROUP]] = distinct !{}96// CHECK: ![[LOOP]] = distinct !{![[LOOP]], ![[PARALLEL_ACCESS:.*]], ![[VECTORIZE:.*]]}97// CHECK: ![[PARALLEL_ACCESS]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP]]}98// CHECK: ![[VECTORIZE]] = !{!"llvm.loop.vectorize.enable", i1 true}99