97 lines · plain
1// RUN: mlir-translate --mlir-to-llvmir %s | FileCheck %s2 3omp.private {type = private} @_QFsimd_reductionEi_private_i32 : i324omp.declare_reduction @add_reduction_f32 : f32 init {5^bb0(%arg0: f32):6 %0 = llvm.mlir.constant(0.000000e+00 : f32) : f327 omp.yield(%0 : f32)8} combiner {9^bb0(%arg0: f32, %arg1: f32):10 %0 = llvm.fadd %arg0, %arg1 {fastmathFlags = #llvm.fastmath<contract>} : f3211 omp.yield(%0 : f32)12}13llvm.func @_QPsimd_reduction(%arg0: !llvm.ptr {fir.bindc_name = "a", llvm.nocapture}, %arg1: !llvm.ptr {fir.bindc_name = "sum", llvm.nocapture}) {14 %0 = llvm.mlir.constant(0.000000e+00 : f32) : f3215 %1 = llvm.mlir.constant(1 : i32) : i3216 %2 = llvm.mlir.constant(1024 : i32) : i3217 %3 = llvm.mlir.constant(1 : i64) : i6418 %4 = llvm.alloca %3 x i32 {bindc_name = "i"} : (i64) -> !llvm.ptr19 llvm.store %0, %arg1 : f32, !llvm.ptr20 omp.simd private(@_QFsimd_reductionEi_private_i32 %4 -> %arg2 : !llvm.ptr) reduction(@add_reduction_f32 %arg1 -> %arg3 : !llvm.ptr) {21 omp.loop_nest (%arg4) : i32 = (%1) to (%2) inclusive step (%1) {22 llvm.store %arg4, %arg2 : i32, !llvm.ptr23 %5 = llvm.load %arg3 : !llvm.ptr -> f3224 %6 = llvm.load %arg2 : !llvm.ptr -> i3225 %7 = llvm.sext %6 : i32 to i6426 %8 = llvm.sub %7, %3 overflow<nsw> : i6427 %9 = llvm.getelementptr %arg0[%8] : (!llvm.ptr, i64) -> !llvm.ptr, f3228 %10 = llvm.load %9 : !llvm.ptr -> f3229 %11 = llvm.fadd %5, %10 {fastmathFlags = #llvm.fastmath<contract>} : f3230 llvm.store %11, %arg3 : f32, !llvm.ptr31 omp.yield32 }33 }34 llvm.return35}36 37// CHECK-LABEL: define void @_QPsimd_reduction(38// CHECK: %[[ORIG_I:.*]] = alloca i32, i64 1, align 439// CHECK: store float 0.000000e+00, ptr %[[ORIG_SUM:.*]], align 440// CHECK: %[[PRIV_I:.*]] = alloca i32, align 441// CHECK: %[[RED_VAR:.*]] = alloca float, align 442// CHECK: br label %[[VAL_4:.*]]43// CHECK: omp.region.after_alloca: ; preds = %[[VAL_5:.*]]44// CHECK: br label %[[VAL_6:.*]]45// CHECK: entry: ; preds = %[[VAL_4]]46// CHECK: br label %[[VAL_7:.*]]47// CHECK: omp.private.init: ; preds = %[[VAL_6]]48// CHECK: br label %[[VAL_8:.*]]49// CHECK: omp.reduction.init: ; preds = %[[VAL_7]]50// CHECK: store float 0.000000e+00, ptr %[[RED_VAR]], align 451// CHECK: br label %[[VAL_9:.*]]52// CHECK: omp.simd.region: ; preds = %[[VAL_8]]53// CHECK: br label %[[VAL_10:.*]]54// CHECK: omp_loop.preheader: ; preds = %[[VAL_9]]55// CHECK: br label %[[VAL_11:.*]]56// CHECK: omp_loop.header: ; preds = %[[VAL_12:.*]], %[[VAL_10]]57// CHECK: %[[VAL_13:.*]] = phi i32 [ 0, %[[VAL_10]] ], [ %[[VAL_14:.*]], %[[VAL_12]] ]58// CHECK: br label %[[VAL_15:.*]]59// CHECK: omp_loop.cond: ; preds = %[[VAL_11]]60// CHECK: %[[VAL_16:.*]] = icmp ult i32 %[[VAL_13]], 102461// CHECK: br i1 %[[VAL_16]], label %[[VAL_17:.*]], label %[[VAL_18:.*]]62// CHECK: omp_loop.body: ; preds = %[[VAL_15]]63// CHECK: %[[VAL_19:.*]] = mul i32 %[[VAL_13]], 164// CHECK: %[[VAL_20:.*]] = add i32 %[[VAL_19]], 165// CHECK: br label %[[VAL_21:.*]]66// CHECK: omp.loop_nest.region: ; preds = %[[VAL_17]]67// CHECK: store i32 %[[VAL_20]], ptr %[[PRIV_I]], align 4, !llvm.access.group ![[ACCESS_GROUP:.*]]68// CHECK: %[[RED_VAL:.*]] = load float, ptr %[[RED_VAR]], align 4, !llvm.access.group ![[ACCESS_GROUP]]69// CHECK: %[[VAL_23:.*]] = load i32, ptr %[[PRIV_I]], align 4, !llvm.access.group ![[ACCESS_GROUP]]70// CHECK: %[[VAL_24:.*]] = sext i32 %[[VAL_23]] to i6471// CHECK: %[[VAL_25:.*]] = sub nsw i64 %[[VAL_24]], 172// CHECK: %[[VAL_26:.*]] = getelementptr float, ptr %[[VAL_27:.*]], i64 %[[VAL_25]]73// CHECK: %[[VAL_28:.*]] = load float, ptr %[[VAL_26]], align 4, !llvm.access.group ![[ACCESS_GROUP]]74// CHECK: %[[VAL_29:.*]] = fadd contract float %[[RED_VAL]], %[[VAL_28]]75// CHECK: store float %[[VAL_29]], ptr %[[RED_VAR]], align 4, !llvm.access.group ![[ACCESS_GROUP]]76// CHECK: br label %[[VAL_30:.*]]77// CHECK: omp.region.cont1: ; preds = %[[VAL_21]]78// CHECK: br label %[[VAL_12]]79// CHECK: omp_loop.inc: ; preds = %[[VAL_30]]80// CHECK: %[[VAL_14]] = add nuw i32 %[[VAL_13]], 181// CHECK: br label %[[VAL_11]], !llvm.loop ![[LOOP:.*]]82// CHECK: omp_loop.exit: ; preds = %[[VAL_15]]83// CHECK: br label %[[VAL_31:.*]]84// CHECK: omp_loop.after: ; preds = %[[VAL_18]]85// CHECK: br label %[[VAL_32:.*]]86// CHECK: omp.region.cont: ; preds = %[[VAL_31]]87// CHECK: %[[SUM_VAL:.*]] = load float, ptr %[[ORIG_SUM]], align 488// CHECK: %[[RED_VAL:.*]] = load float, ptr %[[RED_VAR]], align 489// CHECK: %[[COMBINED_VAL:.*]] = fadd contract float %[[SUM_VAL]], %[[RED_VAL]]90// CHECK: store float %[[COMBINED_VAL]], ptr %[[ORIG_SUM]], align 491// CHECK: ret void92 93// CHECK: ![[ACCESS_GROUP]] = distinct !{}94// CHECK: ![[LOOP]] = distinct !{![[LOOP]], ![[PARALLEL_ACCESS:.*]], ![[VECTORIZE:.*]]}95// CHECK: ![[PARALLEL_ACCESS]] = !{!"llvm.loop.parallel_accesses", ![[ACCESS_GROUP]]}96// CHECK: ![[VECTORIZE]] = !{!"llvm.loop.vectorize.enable", i1 true}97