Prefer movlhps over punpcklqdq, etc. in more cases.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56627 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index fc97c4e..8b5104f 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -3942,17 +3942,17 @@
else if (isIdentityMask(PermMask.getNode(), true))
return V2;
+ // Canonicalize movddup shuffles.
+ if (V2IsUndef && Subtarget->hasSSE2() &&
+ X86::isMOVDDUPMask(PermMask.getNode()))
+ return CanonicalizeMovddup(Op, V1, PermMask, DAG, Subtarget->hasSSE3());
+
if (isSplatMask(PermMask.getNode())) {
if (isMMX || NumElems < 4) return Op;
// Promote it to a v4{if}32 splat.
return PromoteSplat(Op, DAG, Subtarget->hasSSE2());
}
- // Canonicalize movddup shuffles.
- if (V2IsUndef && Subtarget->hasSSE2() &&
- X86::isMOVDDUPMask(PermMask.getNode()))
- return CanonicalizeMovddup(Op, V1, PermMask, DAG, Subtarget->hasSSE3());
-
// If the shuffle can be profitably rewritten as a narrower shuffle, then
// do it!
if (VT == MVT::v8i16 || VT == MVT::v16i8) {