1
; RUN: llc < %s -march=x86 -mattr=+sse2 -o %t
2
; RUN: grep unpcklps %t | count 1
3
; RUN: grep pshufd %t | count 1
4
; RUN: not grep {sub.*esp} %t
6
define void @test(<4 x float>* %res, <4 x float>* %A, <4 x float>* %B) {
7
%tmp = load <4 x float>* %B ; <<4 x float>> [#uses=2]
8
%tmp3 = load <4 x float>* %A ; <<4 x float>> [#uses=2]
9
%tmp.upgrd.1 = extractelement <4 x float> %tmp3, i32 0 ; <float> [#uses=1]
10
%tmp7 = extractelement <4 x float> %tmp, i32 0 ; <float> [#uses=1]
11
%tmp8 = extractelement <4 x float> %tmp3, i32 1 ; <float> [#uses=1]
12
%tmp9 = extractelement <4 x float> %tmp, i32 1 ; <float> [#uses=1]
13
%tmp10 = insertelement <4 x float> undef, float %tmp.upgrd.1, i32 0 ; <<4 x float>> [#uses=1]
14
%tmp11 = insertelement <4 x float> %tmp10, float %tmp7, i32 1 ; <<4 x float>> [#uses=1]
15
%tmp12 = insertelement <4 x float> %tmp11, float %tmp8, i32 2 ; <<4 x float>> [#uses=1]
16
%tmp13 = insertelement <4 x float> %tmp12, float %tmp9, i32 3 ; <<4 x float>> [#uses=1]
17
store <4 x float> %tmp13, <4 x float>* %res
21
define void @test2(<4 x float> %X, <4 x float>* %res) {
22
%tmp5 = shufflevector <4 x float> %X, <4 x float> undef, <4 x i32> < i32 2, i32 6, i32 3, i32 7 > ; <<4 x float>> [#uses=1]
23
store <4 x float> %tmp5, <4 x float>* %res