Instead of a vector load, shuffle and then extract an element. Load the element from address with an offset.

        pshufd $1, (%rdi), %xmm0
        movd %xmm0, %eax
=>
        movl 4(%rdi), %eax

llvm-svn: 51026

1 files changed, 8 insertions, 7 deletions

diff --git a/llvm/test/CodeGen/X86/vec_extract-sse4.ll b/llvm/test/CodeGen/X86/vec_extract-sse4.ll
index 1ef5e8803ef..d6726be1db6 100644
--- a/llvm/test/CodeGen/X86/vec_extract-sse4.ll
+++ b/llvm/test/CodeGen/X86/vec_extract-sse4.ll
@@ -1,29 +1,30 @@
 ; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse41 -o %t -f
-; RUN: grep extractps %t | count 1
-; RUN: grep pextrd    %t | count 2
-; RUN: grep pshufd    %t | count 1
+; RUN: grep extractps   %t | count 1
+; RUN: grep pextrd      %t | count 1
+; RUN: not grep pshufd  %t
+; RUN: not grep movss   %t
 
-define void @t1(float* %R, <4 x float>* %P1) {
+define void @t1(float* %R, <4 x float>* %P1) nounwind {
 	%X = load <4 x float>* %P1
 	%tmp = extractelement <4 x float> %X, i32 3
 	store float %tmp, float* %R
 	ret void
 }
 
-define float @t2(<4 x float>* %P1) {
+define float @t2(<4 x float>* %P1) nounwind {
 	%X = load <4 x float>* %P1
 	%tmp = extractelement <4 x float> %X, i32 2
 	ret float %tmp
 }
 
-define void @t3(i32* %R, <4 x i32>* %P1) {
+define void @t3(i32* %R, <4 x i32>* %P1) nounwind {
 	%X = load <4 x i32>* %P1
 	%tmp = extractelement <4 x i32> %X, i32 3
 	store i32 %tmp, i32* %R
 	ret void
 }
 
-define i32 @t4(<4 x i32>* %P1) {
+define i32 @t4(<4 x i32>* %P1) nounwind {
 	%X = load <4 x i32>* %P1
 	%tmp = extractelement <4 x i32> %X, i32 3
 	ret i32 %tmp