1 ; RUN: opt < %s -basicaa -slp-vectorizer -dce -S -mtriple=i386-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s 2 3 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32-S128" 4 target triple = "i386-apple-macosx10.9.0" 5 6 ; We disable the vectorization of <3 x float> for now 7 8 ; float foo(float *A) { 9 ; 10 ; float R = A[0]; 11 ; float G = A[1]; 12 ; float B = A[2]; 13 ; for (int i=0; i < 121; i+=3) { 14 ; R+=A[i+0]*7; 15 ; G+=A[i+1]*8; 16 ; B+=A[i+2]*9; 17 ; } 18 ; 19 ; return R+G+B; 20 ; } 21 22 ;CHECK-LABEL: @foo( 23 ;CHECK: br 24 ;CHECK-NOT: phi <3 x float> 25 ;CHECK-NOT: fmul <3 x float> 26 ;CHECK-NOT: fadd <3 x float> 27 ; At the moment we don't sink extractelements. 28 ;CHECK: br 29 ;CHECK-NOT: extractelement 30 ;CHECK-NOT: extractelement 31 ;CHECK-NOT: extractelement 32 ;CHECK: ret 33 34 define float @foo(float* nocapture readonly %A) { 35 entry: 36 %0 = load float* %A, align 4 37 %arrayidx1 = getelementptr inbounds float* %A, i64 1 38 %1 = load float* %arrayidx1, align 4 39 %arrayidx2 = getelementptr inbounds float* %A, i64 2 40 %2 = load float* %arrayidx2, align 4 41 br label %for.body 42 43 for.body: ; preds = %for.body.for.body_crit_edge, %entry 44 %3 = phi float [ %0, %entry ], [ %.pre, %for.body.for.body_crit_edge ] 45 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body.for.body_crit_edge ] 46 %B.032 = phi float [ %2, %entry ], [ %add14, %for.body.for.body_crit_edge ] 47 %G.031 = phi float [ %1, %entry ], [ %add9, %for.body.for.body_crit_edge ] 48 %R.030 = phi float [ %0, %entry ], [ %add4, %for.body.for.body_crit_edge ] 49 %mul = fmul float %3, 7.000000e+00 50 %add4 = fadd float %R.030, %mul 51 %4 = add nsw i64 %indvars.iv, 1 52 %arrayidx7 = getelementptr inbounds float* %A, i64 %4 53 %5 = load float* %arrayidx7, align 4 54 %mul8 = fmul float %5, 8.000000e+00 55 %add9 = fadd float %G.031, %mul8 56 %6 = add nsw i64 %indvars.iv, 2 57 %arrayidx12 = getelementptr inbounds float* %A, i64 %6 58 %7 = load float* %arrayidx12, align 4 59 %mul13 = fmul float %7, 9.000000e+00 60 %add14 = fadd float %B.032, %mul13 61 %indvars.iv.next = add i64 %indvars.iv, 3 62 %8 = trunc i64 %indvars.iv.next to i32 63 %cmp = icmp slt i32 %8, 121 64 br i1 %cmp, label %for.body.for.body_crit_edge, label %for.end 65 66 for.body.for.body_crit_edge: ; preds = %for.body 67 %arrayidx3.phi.trans.insert = getelementptr inbounds float* %A, i64 %indvars.iv.next 68 %.pre = load float* %arrayidx3.phi.trans.insert, align 4 69 br label %for.body 70 71 for.end: ; preds = %for.body 72 %add16 = fadd float %add4, %add9 73 %add17 = fadd float %add16, %add14 74 ret float %add17 75 } 76 77
