66 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -S -passes=slp-vectorizer -mtriple=x86_64-pc-windows-msvc19.16.0 < %s | FileCheck %s3 4define void @test(ptr %a, ptr %b) {5; CHECK-LABEL: @test(6; CHECK-NEXT: entry:7; CHECK-NEXT: [[RESULT:%.*]] = alloca [4 x [4 x float]], i32 0, align 48; CHECK-NEXT: [[TMP0:%.*]] = load float, ptr null, align 49; CHECK-NEXT: [[ARRAYIDX120:%.*]] = getelementptr [4 x float], ptr [[B:%.*]], i64 0, i64 310; CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[ARRAYIDX120]], align 411; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <4 x i32> <i32 1, i32 poison, i32 poison, i32 0>12; CHECK-NEXT: [[TMP6:%.*]] = insertelement <4 x float> [[TMP2]], float [[TMP0]], i32 213; CHECK-NEXT: br label [[FOR_BODY:%.*]]14; CHECK: for.body:15; CHECK-NEXT: [[TMP3:%.*]] = load float, ptr null, align 416; CHECK-NEXT: [[TMP4:%.*]] = load <2 x float>, ptr [[A:%.*]], align 417; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <4 x i32> <i32 1, i32 0, i32 1, i32 0>18; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x float> [[TMP6]], float [[TMP3]], i32 119; CHECK-NEXT: [[TMP8:%.*]] = fmul <4 x float> [[TMP5]], [[TMP7]]20; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>21; CHECK-NEXT: [[TMP10:%.*]] = fmul <4 x float> [[TMP9]], zeroinitializer22; CHECK-NEXT: [[TMP11:%.*]] = fadd <4 x float> [[TMP8]], [[TMP10]]23; CHECK-NEXT: [[TMP12:%.*]] = fadd <4 x float> [[TMP11]], zeroinitializer24; CHECK-NEXT: store <4 x float> [[TMP12]], ptr [[RESULT]], align 425; CHECK-NEXT: br label [[FOR_BODY]]26;27entry:28 %result = alloca [4 x [4 x float]], i32 0, align 429 %arrayidx11 = getelementptr [4 x [4 x float]], ptr %b, i64 0, i64 130 %0 = load float, ptr %arrayidx11, align 431 %1 = load float, ptr null, align 432 %arrayidx120 = getelementptr [4 x float], ptr %b, i64 0, i64 333 %2 = load float, ptr %arrayidx120, align 434 br label %for.body35 36for.body:37 %3 = load float, ptr %a, align 438 %mul = fmul float %3, 0.000000e+0039 %arrayidx9 = getelementptr [4 x [4 x float]], ptr %a, i64 0, i64 0, i64 140 %4 = load float, ptr %arrayidx9, align 441 %mul13 = fmul float %4, %042 %add = fadd float %mul, %mul1343 %add22 = fadd float %add, 0.000000e+0044 store float %add22, ptr %result, align 445 %5 = load float, ptr null, align 446 %mul43 = fmul float %3, %547 %mul51 = fmul float %4, 0.000000e+0048 %add52 = fadd float %mul43, %mul5149 %add61 = fadd float %add52, 0.000000e+0050 %arrayidx74 = getelementptr [4 x [4 x float]], ptr %result, i64 0, i64 0, i64 151 store float %add61, ptr %arrayidx74, align 452 %mul82 = fmul float %3, 0.000000e+0053 %mul90 = fmul float %4, %154 %add91 = fadd float %mul82, %mul9055 %add100 = fadd float %add91, 0.000000e+0056 %arrayidx113 = getelementptr [4 x [4 x float]], ptr %result, i64 0, i64 0, i64 257 store float %add100, ptr %arrayidx113, align 458 %mul121 = fmul float %3, %259 %mul129 = fmul float %4, 0.000000e+0060 %add130 = fadd float %mul121, %mul12961 %add139 = fadd float %add130, 0.000000e+0062 %arrayidx152 = getelementptr [4 x [4 x float]], ptr %result, i64 0, i64 0, i64 363 store float %add139, ptr %arrayidx152, align 464 br label %for.body65}66