1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -basic-aa -slp-vectorizer -dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s 3 4target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128" 5target triple = "x86_64-apple-macosx10.8.0" 6 7; Simple 3-pair chain with loads and stores 8define void @test1(double* %a, double* %b, double* %c) { 9; CHECK-LABEL: @test1( 10; CHECK-NEXT: [[TMP1:%.*]] = bitcast double* [[A:%.*]] to <2 x double>* 11; CHECK-NEXT: [[TMP2:%.*]] = load <2 x double>, <2 x double>* [[TMP1]], align 8 12; CHECK-NEXT: [[TMP3:%.*]] = bitcast double* [[B:%.*]] to <2 x double>* 13; CHECK-NEXT: [[TMP4:%.*]] = load <2 x double>, <2 x double>* [[TMP3]], align 8 14; CHECK-NEXT: [[TMP5:%.*]] = fmul <2 x double> [[TMP2]], [[TMP4]] 15; CHECK-NEXT: [[TMP6:%.*]] = bitcast double* [[C:%.*]] to <2 x double>* 16; CHECK-NEXT: store <2 x double> [[TMP5]], <2 x double>* [[TMP6]], align 8 17; CHECK-NEXT: ret void 18; 19 %i0 = load double, double* %a, align 8 20 %i1 = load double, double* %b, align 8 21 %mul = fmul double %i0, %i1 22 %arrayidx3 = getelementptr inbounds double, double* %a, i64 1 23 %i3 = load double, double* %arrayidx3, align 8 24 %arrayidx4 = getelementptr inbounds double, double* %b, i64 1 25 %i4 = load double, double* %arrayidx4, align 8 26 %mul5 = fmul double %i3, %i4 27 store double %mul, double* %c, align 8 28 %arrayidx5 = getelementptr inbounds double, double* %c, i64 1 29 store double %mul5, double* %arrayidx5, align 8 30 ret void 31} 32 33; Simple 3-pair chain with loads and stores, obfuscated with bitcasts 34define void @test2(double* %a, double* %b, i8* %e) { 35; CHECK-LABEL: @test2( 36; CHECK-NEXT: [[C:%.*]] = bitcast i8* [[E:%.*]] to double* 37; CHECK-NEXT: [[TMP1:%.*]] = bitcast double* [[A:%.*]] to <2 x double>* 38; CHECK-NEXT: [[TMP2:%.*]] = load <2 x double>, <2 x double>* [[TMP1]], align 8 39; CHECK-NEXT: [[TMP3:%.*]] = bitcast double* [[B:%.*]] to <2 x double>* 40; CHECK-NEXT: [[TMP4:%.*]] = load <2 x double>, <2 x double>* [[TMP3]], align 8 41; CHECK-NEXT: [[TMP5:%.*]] = fmul <2 x double> [[TMP2]], [[TMP4]] 42; CHECK-NEXT: [[TMP6:%.*]] = bitcast double* [[C]] to <2 x double>* 43; CHECK-NEXT: store <2 x double> [[TMP5]], <2 x double>* [[TMP6]], align 8 44; CHECK-NEXT: ret void 45; 46 %i0 = load double, double* %a, align 8 47 %i1 = load double, double* %b, align 8 48 %mul = fmul double %i0, %i1 49 %arrayidx3 = getelementptr inbounds double, double* %a, i64 1 50 %i3 = load double, double* %arrayidx3, align 8 51 %arrayidx4 = getelementptr inbounds double, double* %b, i64 1 52 %i4 = load double, double* %arrayidx4, align 8 53 %mul5 = fmul double %i3, %i4 54 %c = bitcast i8* %e to double* 55 store double %mul, double* %c, align 8 56 %carrayidx5 = getelementptr inbounds i8, i8* %e, i64 8 57 %arrayidx5 = bitcast i8* %carrayidx5 to double* 58 store double %mul5, double* %arrayidx5, align 8 59 ret void 60} 61 62; Don't vectorize volatile loads. 63define void @test_volatile_load(double* %a, double* %b, double* %c) { 64; CHECK-LABEL: @test_volatile_load( 65; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds double, double* [[A:%.*]], i64 1 66; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds double, double* [[B:%.*]], i64 1 67; CHECK-NEXT: [[I0:%.*]] = load volatile double, double* [[A]], align 8 68; CHECK-NEXT: [[I1:%.*]] = load volatile double, double* [[B]], align 8 69; CHECK-NEXT: [[I3:%.*]] = load double, double* [[ARRAYIDX3]], align 8 70; CHECK-NEXT: [[I4:%.*]] = load double, double* [[ARRAYIDX4]], align 8 71; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x double> poison, double [[I0]], i32 0 72; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double [[I3]], i32 1 73; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x double> poison, double [[I1]], i32 0 74; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x double> [[TMP3]], double [[I4]], i32 1 75; CHECK-NEXT: [[TMP5:%.*]] = fmul <2 x double> [[TMP2]], [[TMP4]] 76; CHECK-NEXT: [[TMP6:%.*]] = bitcast double* [[C:%.*]] to <2 x double>* 77; CHECK-NEXT: store <2 x double> [[TMP5]], <2 x double>* [[TMP6]], align 8 78; CHECK-NEXT: ret void 79; 80 %i0 = load volatile double, double* %a, align 8 81 %i1 = load volatile double, double* %b, align 8 82 %mul = fmul double %i0, %i1 83 %arrayidx3 = getelementptr inbounds double, double* %a, i64 1 84 %i3 = load double, double* %arrayidx3, align 8 85 %arrayidx4 = getelementptr inbounds double, double* %b, i64 1 86 %i4 = load double, double* %arrayidx4, align 8 87 %mul5 = fmul double %i3, %i4 88 store double %mul, double* %c, align 8 89 %arrayidx5 = getelementptr inbounds double, double* %c, i64 1 90 store double %mul5, double* %arrayidx5, align 8 91 ret void 92} 93 94; Don't vectorize volatile stores. 95define void @test_volatile_store(double* %a, double* %b, double* %c) { 96; CHECK-LABEL: @test_volatile_store( 97; CHECK-NEXT: [[I0:%.*]] = load double, double* [[A:%.*]], align 8 98; CHECK-NEXT: [[I1:%.*]] = load double, double* [[B:%.*]], align 8 99; CHECK-NEXT: [[MUL:%.*]] = fmul double [[I0]], [[I1]] 100; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds double, double* [[A]], i64 1 101; CHECK-NEXT: [[I3:%.*]] = load double, double* [[ARRAYIDX3]], align 8 102; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds double, double* [[B]], i64 1 103; CHECK-NEXT: [[I4:%.*]] = load double, double* [[ARRAYIDX4]], align 8 104; CHECK-NEXT: [[MUL5:%.*]] = fmul double [[I3]], [[I4]] 105; CHECK-NEXT: store volatile double [[MUL]], double* [[C:%.*]], align 8 106; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds double, double* [[C]], i64 1 107; CHECK-NEXT: store volatile double [[MUL5]], double* [[ARRAYIDX5]], align 8 108; CHECK-NEXT: ret void 109; 110 %i0 = load double, double* %a, align 8 111 %i1 = load double, double* %b, align 8 112 %mul = fmul double %i0, %i1 113 %arrayidx3 = getelementptr inbounds double, double* %a, i64 1 114 %i3 = load double, double* %arrayidx3, align 8 115 %arrayidx4 = getelementptr inbounds double, double* %b, i64 1 116 %i4 = load double, double* %arrayidx4, align 8 117 %mul5 = fmul double %i3, %i4 118 store volatile double %mul, double* %c, align 8 119 %arrayidx5 = getelementptr inbounds double, double* %c, i64 1 120 store volatile double %mul5, double* %arrayidx5, align 8 121 ret void 122} 123 124