1
; RUN: opt < %s -loop-vectorize -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx -S | FileCheck %s
3
target 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
target triple = "x86_64-apple-macosx10.8.0"
6
@c = common global [2048 x i32] zeroinitializer, align 16
7
@b = common global [2048 x i32] zeroinitializer, align 16
8
@d = common global [2048 x i32] zeroinitializer, align 16
9
@a = common global [2048 x i32] zeroinitializer, align 16
11
; The program below gathers and scatters data. We better not vectorize it.
12
;CHECK-LABEL: @cost_model_1(
17
define void @cost_model_1() nounwind uwtable noinline ssp {
21
for.body: ; preds = %for.body, %entry
22
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
23
%0 = shl nsw i64 %indvars.iv, 1
24
%arrayidx = getelementptr inbounds [2048 x i32], [2048 x i32]* @c, i64 0, i64 %0
25
%1 = load i32, i32* %arrayidx, align 8
26
%idxprom1 = sext i32 %1 to i64
27
%arrayidx2 = getelementptr inbounds [2048 x i32], [2048 x i32]* @b, i64 0, i64 %idxprom1
28
%2 = load i32, i32* %arrayidx2, align 4
29
%arrayidx4 = getelementptr inbounds [2048 x i32], [2048 x i32]* @d, i64 0, i64 %indvars.iv
30
%3 = load i32, i32* %arrayidx4, align 4
31
%idxprom5 = sext i32 %3 to i64
32
%arrayidx6 = getelementptr inbounds [2048 x i32], [2048 x i32]* @a, i64 0, i64 %idxprom5
33
store i32 %2, i32* %arrayidx6, align 4
34
%indvars.iv.next = add i64 %indvars.iv, 1
35
%lftr.wideiv = trunc i64 %indvars.iv.next to i32
36
%exitcond = icmp eq i32 %lftr.wideiv, 256
37
br i1 %exitcond, label %for.end, label %for.body
39
for.end: ; preds = %for.body