1 ; RUN: opt -mtriple=x86_64-apple-darwin -mcpu=core2 -loop-vectorize -dce -instcombine -S < %s | FileCheck %s
3 @B = common global [1024 x i32] zeroinitializer, align 16
4 @A = common global [1024 x i32] zeroinitializer, align 16
6 ; We use to not vectorize this loop because the shift was deemed to expensive.
7 ; Now that we differentiate shift cost base on the operand value kind, we will
9 ; CHECK: ashr <4 x i32>
15 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
16 %arrayidx = getelementptr inbounds [1024 x i32]* @B, i64 0, i64 %indvars.iv
17 %0 = load i32* %arrayidx, align 4
19 %arrayidx2 = getelementptr inbounds [1024 x i32]* @A, i64 0, i64 %indvars.iv
20 store i32 %shl, i32* %arrayidx2, align 4
21 %indvars.iv.next = add i64 %indvars.iv, 1
22 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
23 %exitcond = icmp eq i32 %lftr.wideiv, 1024
24 br i1 %exitcond, label %for.end, label %for.body