From: Craig Topper <craig.topper@intel.com>
Date: Mon, 23 Sep 2019 05:35:23 +0000 (+0000)
Subject: [X86] Canonicalize all zeroes vector to RHS in X86DAGToDAGISel::tryVPTESTM.
X-Git-Tag: llvmorg-11-init~8620
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=03b5a13ee32e577346dae13df86f29dd0e75580f;p=platform%2Fupstream%2Fllvm.git

[X86] Canonicalize all zeroes vector to RHS in X86DAGToDAGISel::tryVPTESTM.

llvm-svn: 372544
---

diff --git a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
index 1cff706..83c1251 100644
--- a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -4098,12 +4098,18 @@ bool X86DAGToDAGISel::tryVPTESTM(SDNode *Root, SDValue Setcc,
   if (CC != ISD::SETEQ && CC != ISD::SETNE)
     return false;
 
+  SDValue SetccOp0 = Setcc.getOperand(0);
+  SDValue SetccOp1 = Setcc.getOperand(1);
+
+  // Canonicalize the all zero vector to the RHS.
+  if (ISD::isBuildVectorAllZeros(SetccOp0.getNode()))
+    std::swap(SetccOp0, SetccOp1);
+
   // See if we're comparing against zero.
-  // FIXME: Handle all zeros on LHS.
-  if (!ISD::isBuildVectorAllZeros(Setcc.getOperand(1).getNode()))
+  if (!ISD::isBuildVectorAllZeros(SetccOp1.getNode()))
     return false;
 
-  SDValue N0 = Setcc.getOperand(0);
+  SDValue N0 = SetccOp0;
 
   MVT CmpVT = N0.getSimpleValueType();
   MVT CmpSVT = CmpVT.getVectorElementType();
diff --git a/llvm/test/CodeGen/X86/avx512f-vec-test-testn.ll b/llvm/test/CodeGen/X86/avx512f-vec-test-testn.ll
index 58278df..b76f991 100644
--- a/llvm/test/CodeGen/X86/avx512f-vec-test-testn.ll
+++ b/llvm/test/CodeGen/X86/avx512f-vec-test-testn.ll
@@ -215,27 +215,16 @@ entry:
 }
 
 define <2 x i64> @setcc_commute(<2 x i64> %a) {
-; X64-LABEL: setcc_commute:
-; X64:       # %bb.0:
-; X64-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; X64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; X64-NEXT:    vpsubq %xmm0, %xmm1, %xmm1
-; X64-NEXT:    vptestnmq %zmm0, %zmm0, %k1
-; X64-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; X64-NEXT:    vmovdqa %xmm1, %xmm0
-; X64-NEXT:    vzeroupper
-; X64-NEXT:    retq
-;
-; X86-LABEL: setcc_commute:
-; X86:       # %bb.0:
-; X86-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; X86-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X86-NEXT:    vpsubq %xmm0, %xmm2, %xmm1
-; X86-NEXT:    vpcmpeqq %zmm0, %zmm2, %k1
-; X86-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
-; X86-NEXT:    vmovdqa %xmm1, %xmm0
-; X86-NEXT:    vzeroupper
-; X86-NEXT:    retl
+; CHECK-LABEL: setcc_commute:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; CHECK-NEXT:    vpsubq %xmm0, %xmm1, %xmm1
+; CHECK-NEXT:    vptestnmq %zmm0, %zmm0, %k1
+; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vmovdqa %xmm1, %xmm0
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    ret{{[l|q]}}
   %1 = sub <2 x i64> zeroinitializer, %a
   %2 = icmp eq <2 x i64> %a, zeroinitializer
   %3 = select <2 x i1> %2, <2 x i64> %a, <2 x i64> %1