1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -basicaa -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; double foo(double * restrict b, double * restrict a, int n, int m) { 8; double r=a[1]; 9; double g=a[0]; 10; double x; 11; for (int i=0; i < 100; i++) { 12; r += 10; 13; g += 10; 14; r *= 4; 15; g *= 4; 16; x = g; <----- external user! 17; r += 4; 18; g += 4; 19; } 20; b[0] = g; 21; b[1] = r; 22; 23; return x; <-- must extract here! 24; } 25 26define double @ext_user(double* noalias nocapture %B, double* noalias nocapture %A, i32 %n, i32 %m) { 27; CHECK-LABEL: @ext_user( 28; CHECK-NEXT: entry: 29; CHECK-NEXT: [[TMP0:%.*]] = bitcast double* [[A:%.*]] to <2 x double>* 30; CHECK-NEXT: [[TMP1:%.*]] = load <2 x double>, <2 x double>* [[TMP0]], align 8 31; CHECK-NEXT: br label [[FOR_BODY:%.*]] 32; CHECK: for.body: 33; CHECK-NEXT: [[I_020:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[FOR_BODY]] ] 34; CHECK-NEXT: [[TMP2:%.*]] = phi <2 x double> [ [[TMP1]], [[ENTRY]] ], [ [[TMP5:%.*]], [[FOR_BODY]] ] 35; CHECK-NEXT: [[TMP3:%.*]] = fadd <2 x double> [[TMP2]], <double 1.000000e+01, double 1.000000e+01> 36; CHECK-NEXT: [[TMP4:%.*]] = fmul <2 x double> [[TMP3]], <double 4.000000e+00, double 4.000000e+00> 37; CHECK-NEXT: [[TMP5]] = fadd <2 x double> [[TMP4]], <double 4.000000e+00, double 4.000000e+00> 38; CHECK-NEXT: [[INC]] = add nsw i32 [[I_020]], 1 39; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[INC]], 100 40; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]] 41; CHECK: for.end: 42; CHECK-NEXT: [[TMP6:%.*]] = bitcast double* [[B:%.*]] to <2 x double>* 43; CHECK-NEXT: store <2 x double> [[TMP5]], <2 x double>* [[TMP6]], align 8 44; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x double> [[TMP4]], i32 0 45; CHECK-NEXT: ret double [[TMP7]] 46; 47entry: 48 %arrayidx = getelementptr inbounds double, double* %A, i64 1 49 %0 = load double, double* %arrayidx, align 8 50 %1 = load double, double* %A, align 8 51 br label %for.body 52 53for.body: ; preds = %for.body, %entry 54 %i.020 = phi i32 [ 0, %entry ], [ %inc, %for.body ] 55 %G.019 = phi double [ %1, %entry ], [ %add5, %for.body ] 56 %R.018 = phi double [ %0, %entry ], [ %add4, %for.body ] 57 %add = fadd double %R.018, 1.000000e+01 58 %add2 = fadd double %G.019, 1.000000e+01 59 %mul = fmul double %add, 4.000000e+00 60 %mul3 = fmul double %add2, 4.000000e+00 61 %add4 = fadd double %mul, 4.000000e+00 62 %add5 = fadd double %mul3, 4.000000e+00 63 %inc = add nsw i32 %i.020, 1 64 %exitcond = icmp eq i32 %inc, 100 65 br i1 %exitcond, label %for.end, label %for.body 66 67for.end: ; preds = %for.body 68 store double %add5, double* %B, align 8 69 %arrayidx7 = getelementptr inbounds double, double* %B, i64 1 70 store double %add4, double* %arrayidx7, align 8 71 ret double %mul3 72} 73 74; A need-to-gather entry cannot be an external use of the scalar element. 75; Instead the insertelement instructions of the need-to-gather entry are the 76; external users. 77; This test would assert because we would keep the scalar fpext and fadd alive. 78; PR18129 79 80define i32 @needtogather(double *noalias %a, i32 *noalias %b, float * noalias %c, 81; CHECK-LABEL: @needtogather( 82; CHECK-NEXT: entry: 83; CHECK-NEXT: [[TMP0:%.*]] = load i32, i32* [[D:%.*]], align 4 84; CHECK-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP0]] to float 85; CHECK-NEXT: [[TMP1:%.*]] = load float, float* [[C:%.*]] 86; CHECK-NEXT: [[SUB:%.*]] = fsub float 0.000000e+00, [[TMP1]] 87; CHECK-NEXT: [[MUL:%.*]] = fmul float [[SUB]], 0.000000e+00 88; CHECK-NEXT: [[ADD:%.*]] = fadd float [[CONV]], [[MUL]] 89; CHECK-NEXT: [[CONV1:%.*]] = fpext float [[ADD]] to double 90; CHECK-NEXT: [[SUB3:%.*]] = fsub float 1.000000e+00, [[TMP1]] 91; CHECK-NEXT: [[MUL4:%.*]] = fmul float [[SUB3]], 0.000000e+00 92; CHECK-NEXT: [[ADD5:%.*]] = fadd float [[CONV]], [[MUL4]] 93; CHECK-NEXT: [[CONV6:%.*]] = fpext float [[ADD5]] to double 94; CHECK-NEXT: [[TOBOOL:%.*]] = fcmp une float [[ADD]], 0.000000e+00 95; CHECK-NEXT: br i1 [[TOBOOL]], label [[IF_THEN:%.*]], label [[IF_END:%.*]] 96; CHECK: if.then: 97; CHECK-NEXT: br label [[IF_END]] 98; CHECK: if.end: 99; CHECK-NEXT: [[STOREMERGE:%.*]] = phi double [ [[CONV6]], [[IF_THEN]] ], [ [[CONV1]], [[ENTRY:%.*]] ] 100; CHECK-NEXT: [[E_0:%.*]] = phi double [ [[CONV1]], [[IF_THEN]] ], [ [[CONV6]], [[ENTRY]] ] 101; CHECK-NEXT: store double [[STOREMERGE]], double* [[A:%.*]], align 8 102; CHECK-NEXT: [[CONV7:%.*]] = fptosi double [[E_0]] to i32 103; CHECK-NEXT: store i32 [[CONV7]], i32* [[B:%.*]], align 4 104; CHECK-NEXT: ret i32 undef 105; 106 i32 * noalias %d) { 107entry: 108 %0 = load i32, i32* %d, align 4 109 %conv = sitofp i32 %0 to float 110 %1 = load float, float* %c 111 %sub = fsub float 0.000000e+00, %1 112 %mul = fmul float %sub, 0.000000e+00 113 %add = fadd float %conv, %mul 114 %conv1 = fpext float %add to double 115 %sub3 = fsub float 1.000000e+00, %1 116 %mul4 = fmul float %sub3, 0.000000e+00 117 %add5 = fadd float %conv, %mul4 118 %conv6 = fpext float %add5 to double 119 %tobool = fcmp une float %add, 0.000000e+00 120 br i1 %tobool, label %if.then, label %if.end 121 122if.then: 123 br label %if.end 124 125if.end: 126 %storemerge = phi double [ %conv6, %if.then ], [ %conv1, %entry ] 127 %e.0 = phi double [ %conv1, %if.then ], [ %conv6, %entry ] 128 store double %storemerge, double* %a, align 8 129 %conv7 = fptosi double %e.0 to i32 130 store i32 %conv7, i32* %b, align 4 131 ret i32 undef 132} 133