[x86] Adjust the patterns for lowering X86vzmovl nodes which don't

perform a load to use blendps rather than movss when it is available.

For non-loads, blendps is *much* faster. It can execute on two ports in
Sandy Bridge and Ivy Bridge, and *three* ports on Haswell. This fixes
one of the "regressions" from aggressively taking the "insertion" path
in the new vector shuffle lowering.

This does highlight one problem with blendps -- it isn't commuted as
heavily as it should be. That's future work though.

llvm-svn: 219022

1 files changed, 1 insertions, 1 deletions

diff --git a/llvm/test/CodeGen/X86/vec_set-3.ll b/llvm/test/CodeGen/X86/vec_set-3.ll
index 043cf96a671..b38b8bfb81f 100644
--- a/llvm/test/CodeGen/X86/vec_set-3.ll
+++ b/llvm/test/CodeGen/X86/vec_set-3.ll
@@ -39,7 +39,7 @@ entry:
 define <4 x float> @test3(<4 x float> %A) {
 ; CHECK-LABEL: test3:
 ; CHECK:         xorps %[[X1:xmm[0-9]+]], %[[X1]]
-; CHECK-NEXT:    movss %xmm0, %[[X1]]
+; CHECK-NEXT:    blendps $1, %xmm0, %[[X1]]
 ; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = [[X1]][1,0,1,1]
 ; CHECK-NEXT:    retl
 ;