[llvm-commits] [llvm] r56627 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/vec_set-9.ll

[Evan Cheng]

Author: evancheng
Date: Thu Sep 25 18:35:16 2008
New Revision: 56627

URL: http://llvm.org/viewvc/llvm-project?rev=56627&view=rev
Log:
Prefer movlhps over punpcklqdq, etc. in more cases.

Modified:
    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
    llvm/trunk/test/CodeGen/X86/vec_set-9.ll

Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=56627&r1=56626&r2=56627&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Thu Sep 25 18:35:16 2008
@@ -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) {

Modified: llvm/trunk/test/CodeGen/X86/vec_set-9.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vec_set-9.ll?rev=56627&r1=56626&r2=56627&view=diff

==============================================================================
--- llvm/trunk/test/CodeGen/X86/vec_set-9.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vec_set-9.ll Thu Sep 25 18:35:16 2008
@@ -1,5 +1,5 @@
 ; RUN: llvm-as < %s | llc -march=x86-64 | grep movd | count 1
-; RUN: llvm-as < %s | llc -march=x86-64 | grep {punpcklqdq.*%xmm0, %xmm0}
+; RUN: llvm-as < %s | llc -march=x86-64 | grep {movlhps.*%xmm0, %xmm0}
 
 define <2 x i64> @test3(i64 %A) nounwind {
 entry: