81 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 32; RUN: opt < %s -passes=loop-vectorize -force-vector-width=4 -S | FileCheck %s3 4target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"5 6define void @fn(ptr %hbuf, ptr %ref, i32 %height) {7; CHECK-LABEL: define void @fn(8; CHECK-SAME: ptr [[HBUF:%.*]], ptr [[REF:%.*]], i32 [[HEIGHT:%.*]]) {9; CHECK-NEXT: entry:10; CHECK-NEXT: store i16 0, ptr [[HBUF]], align 111; CHECK-NEXT: [[CMP1:%.*]] = icmp sgt i32 [[HEIGHT]], 012; CHECK-NEXT: br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]13; CHECK: for.body.preheader:14; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[HEIGHT]], 415; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]16; CHECK: vector.memcheck:17; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[HBUF]], i64 218; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[REF]], i64 219; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[HBUF]], [[SCEVGEP1]]20; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[REF]], [[SCEVGEP]]21; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]22; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]23; CHECK: vector.ph:24; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[HEIGHT]], 425; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[HEIGHT]], [[N_MOD_VF]]26; CHECK-NEXT: [[TMP0:%.*]] = load i16, ptr [[REF]], align 1, !alias.scope [[META0:![0-9]+]]27; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i16> poison, i16 [[TMP0]], i64 028; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT]], <4 x i16> poison, <4 x i32> zeroinitializer29; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]30; CHECK: vector.body:31; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]32; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i16> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP1:%.*]], [[VECTOR_BODY]] ]33; CHECK-NEXT: [[TMP1]] = add <4 x i16> [[BROADCAST_SPLAT]], [[VEC_PHI]]34; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 435; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]36; CHECK-NEXT: br i1 [[TMP2]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]37; CHECK: middle.block:38; CHECK-NEXT: [[TMP3:%.*]] = call i16 @llvm.vector.reduce.add.v4i16(<4 x i16> [[TMP1]])39; CHECK-NEXT: store i16 [[TMP3]], ptr [[HBUF]], align 1, !alias.scope [[META6:![0-9]+]], !noalias [[META0]]40; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[HEIGHT]], [[N_VEC]]41; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]]42; CHECK: scalar.ph:43; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i16 [ [[TMP3]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ]44; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ]45; CHECK-NEXT: br label [[FOR_BODY:%.*]]46; CHECK: for.body:47; CHECK-NEXT: [[TMP4:%.*]] = phi i16 [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]48; CHECK-NEXT: [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]49; CHECK-NEXT: [[TMP5:%.*]] = load i16, ptr [[REF]], align 150; CHECK-NEXT: [[ADD]] = add i16 [[TMP5]], [[TMP4]]51; CHECK-NEXT: store i16 [[ADD]], ptr [[HBUF]], align 152; CHECK-NEXT: [[INC]] = add i32 [[I]], 153; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[HEIGHT]]54; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]55; CHECK: for.end.loopexit:56; CHECK-NEXT: br label [[FOR_END]]57; CHECK: for.end:58; CHECK-NEXT: ret void59;60entry:61 store i16 0, ptr %hbuf, align 162 %cmp1 = icmp sgt i32 %height, 063 br i1 %cmp1, label %for.body.preheader, label %for.end64 65for.body.preheader:66 br label %for.body67 68for.body:69 %0 = phi i16 [ %add, %for.body ], [ 0, %for.body.preheader ]70 %i = phi i32 [ 0, %for.body.preheader ], [ %inc, %for.body ]71 %1 = load i16, ptr %ref, align 172 %add = add i16 %1, %073 store i16 %add, ptr %hbuf, align 174 %inc = add i32 %i, 175 %exitcond.not = icmp eq i32 %inc, %height76 br i1 %exitcond.not, label %for.end, label %for.body77 78for.end:79 ret void80}81