From: Craig Topper <craig.topper@intel.com>
Date: Sun, 8 Sep 2019 19:24:42 +0000 (+0000)
Subject: [X86] Use DAG.getConstant instead of getZeroVector in combinePMULDQ.
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=9b414d577f2e8254ce6776e9afc18525ff08dda7;p=llvm

[X86] Use DAG.getConstant instead of getZeroVector in combinePMULDQ.

getZeroVector canonicalizes the type to vXi32, but that's a
legalization action. We should use the most correct type if
possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@371345 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index c7844220dfc..3950b292ab3 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -44768,7 +44768,7 @@ static SDValue combinePMULDQ(SDNode *N, SelectionDAG &DAG,
   // Multiply by zero.
   // Don't return RHS as it may contain UNDEFs.
   if (ISD::isBuildVectorAllZeros(RHS.getNode()))
-    return getZeroVector(N->getSimpleValueType(0), Subtarget, DAG, SDLoc(N));
+    return DAG.getConstant(0, SDLoc(N), N->getValueType(0));
 
   // PMULDQ/PMULUDQ only uses lower 32 bits from each vector element.
   const TargetLowering &TLI = DAG.getTargetLoweringInfo();