1; RUN: opt < %s -passes=loop-vectorize,dce,instcombine -force-vector-interleave=1 -force-vector-width=4 -S | FileCheck %s 2 3target 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" 4 5;CHECK-LABEL: @flags1( 6;CHECK: load <4 x i32> 7;CHECK: mul nsw <4 x i32> 8;CHECK: store <4 x i32> 9;CHECK: ret i32 10define i32 @flags1(i32 %n, ptr nocapture %A) nounwind uwtable ssp { 11 %1 = icmp sgt i32 %n, 9 12 br i1 %1, label %.lr.ph, label %._crit_edge 13 14.lr.ph: ; preds = %0, %.lr.ph 15 %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 9, %0 ] 16 %2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv 17 %3 = load i32, ptr %2, align 4 18 %4 = mul nsw i32 %3, 3 19 store i32 %4, ptr %2, align 4 20 %indvars.iv.next = add i64 %indvars.iv, 1 21 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 22 %exitcond = icmp eq i32 %lftr.wideiv, %n 23 br i1 %exitcond, label %._crit_edge, label %.lr.ph 24 25._crit_edge: ; preds = %.lr.ph, %0 26 ret i32 undef 27} 28 29 30;CHECK-LABEL: @flags2( 31;CHECK: load <4 x i32> 32;CHECK: mul <4 x i32> 33;CHECK: store <4 x i32> 34;CHECK: ret i32 35define i32 @flags2(i32 %n, ptr nocapture %A) nounwind uwtable ssp { 36 %1 = icmp sgt i32 %n, 9 37 br i1 %1, label %.lr.ph, label %._crit_edge 38 39.lr.ph: ; preds = %0, %.lr.ph 40 %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 9, %0 ] 41 %2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv 42 %3 = load i32, ptr %2, align 4 43 %4 = mul i32 %3, 3 44 store i32 %4, ptr %2, align 4 45 %indvars.iv.next = add i64 %indvars.iv, 1 46 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 47 %exitcond = icmp eq i32 %lftr.wideiv, %n 48 br i1 %exitcond, label %._crit_edge, label %.lr.ph 49 50._crit_edge: ; preds = %.lr.ph, %0 51 ret i32 undef 52} 53 54; Make sure we copy fast math flags and use them for the final reduction. 55; CHECK-LABEL: fast_math 56; CHECK: load <4 x float> 57; CHECK: fadd fast <4 x float> 58; CHECK: br 59; CHECK: call fast float @llvm.vector.reduce.fadd.v4f32 60define float @fast_math(ptr noalias %s) { 61entry: 62 br label %for.body 63 64for.body: 65 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 66 %q.04 = phi float [ 0.000000e+00, %entry ], [ %add, %for.body ] 67 %arrayidx = getelementptr inbounds float, ptr %s, i64 %indvars.iv 68 %0 = load float, ptr %arrayidx, align 4 69 %add = fadd fast float %q.04, %0 70 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 71 %exitcond = icmp eq i64 %indvars.iv.next, 256 72 br i1 %exitcond, label %for.end, label %for.body 73 74for.end: 75 %add.lcssa = phi float [ %add, %for.body ] 76 ret float %add.lcssa 77} 78