Expand V_SET0 to xorps by default.

The xorps instruction is smaller than pxor, so prefer that encoding.

The ExecutionDepsFix pass will switch the encoding to pxor and xorpd
when appropriate.

llvm-svn: 143996

1 files changed, 12 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/X86/vec_zero.ll b/llvm/test/CodeGen/X86/vec_zero.ll
index 4d1f05629b4..682a0dfca80 100644
--- a/llvm/test/CodeGen/X86/vec_zero.ll
+++ b/llvm/test/CodeGen/X86/vec_zero.ll
@@ -1,5 +1,6 @@
 ; RUN: llc < %s -march=x86 -mattr=+sse2 | FileCheck %s
 
+; CHECK: foo
 ; CHECK: xorps
 define void @foo(<4 x float>* %P) {
         %T = load <4 x float>* %P               ; <<4 x float>> [#uses=1]
@@ -8,6 +9,7 @@ define void @foo(<4 x float>* %P) {
         ret void
 }
 
+; CHECK: bar
 ; CHECK: pxor
 define void @bar(<4 x i32>* %P) {
         %T = load <4 x i32>* %P         ; <<4 x i32>> [#uses=1]
@@ -16,3 +18,13 @@ define void @bar(<4 x i32>* %P) {
         ret void
 }
 
+; Without any type hints from operations, we fall back to the smaller xorps.
+; The IR type <4 x i32> is ignored.
+; CHECK: untyped_zero
+; CHECK: xorps
+; CHECK: movaps
+define void @untyped_zero(<4 x i32>* %p) {
+entry:
+  store <4 x i32> zeroinitializer, <4 x i32>* %p, align 16
+  ret void
+}