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, 9 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/X86/extractelement-load.ll b/llvm/test/CodeGen/X86/extractelement-load.ll
new file mode 100644
index 00000000000..4850eba6092
--- /dev/null
+++ b/llvm/test/CodeGen/X86/extractelement-load.ll
@@ -0,0 +1,9 @@
+; RUN: llvm-as %s -o - | llc -march=x86 -mattr=+sse2 -mcpu=yonah | not grep movd
+; RUN: llvm-as %s -o - | llc -march=x86-64 -mattr=+sse2 -mcpu=yonah | not grep movd
+
+define i32 @t(<2 x i64>* %val) nounwind  {
+	%tmp2 = load <2 x i64>* %val, align 16		; <<2 x i64>> [#uses=1]
+	%tmp3 = bitcast <2 x i64> %tmp2 to <4 x i32>		; <<4 x i32>> [#uses=1]
+	%tmp4 = extractelement <4 x i32> %tmp3, i32 2		; <i32> [#uses=1]
+	ret i32 %tmp4
+}