[X86][AVX512] Fixed VPERMT2* shuffle mask decoding and enabled target shuffle combining.

Patch to add support for target shuffle combining of X86ISD::VPERMV3 nodes, including support for detecting unary shuffles.

This uncovered several issues with the X86ISD::VPERMV3 shuffle mask decoding of non-64 bit shuffle mask elements - the bit masking wasn't being correctly computed.

Removed non-constant pool mask decode path as we have no way of testing it right now.

Differential Revision: http://reviews.llvm.org/D17916

llvm-svn: 262809

3 files changed, 9 insertions, 11 deletions

diff --git a/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp b/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp
index b7b444a80e4..f007349d184 100644
--- a/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp
+++ b/llvm/lib/Target/X86/Utils/X86ShuffleDecode.cpp
@@ -482,8 +482,9 @@ void DecodeVPERMVMask(ArrayRef<uint64_t> RawMask,
 
 void DecodeVPERMV3Mask(ArrayRef<uint64_t> RawMask,
                       SmallVectorImpl<int> &ShuffleMask) {
-  for (int i = 0, e = RawMask.size(); i < e; ++i) {
-    uint64_t M = RawMask[i];
+  uint64_t EltMaskSize = (RawMask.size() * 2) - 1;
+  for (auto M : RawMask) {
+    M &= EltMaskSize;
     ShuffleMask.push_back((int)M);
   }
 }
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 4608df1281d..c3c6cf331ac 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -5138,16 +5138,11 @@ static bool getTargetShuffleMask(SDNode *N, MVT VT, bool AllowSentinelZero,
     return false;
   }
   case X86ISD::VPERMV3: {
+    IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(2);
     // Unlike most shuffle nodes, VPERMV3's mask operand is the middle one.
     Ops.push_back(N->getOperand(0));
     Ops.push_back(N->getOperand(2));
     SDValue MaskNode = N->getOperand(1);
-    SmallVector<uint64_t, 32> RawMask;
-    unsigned MaskLoBits = Log2_64(VT.getVectorNumElements() * 2);
-    if (getTargetShuffleMaskIndices(MaskNode, MaskLoBits, RawMask)) {
-      DecodeVPERMV3Mask(RawMask, Mask);
-      break;
-    }
     if (auto *C = getTargetShuffleMaskConstant(MaskNode)) {
       DecodeVPERMV3Mask(C, VT, Mask);
       break;
@@ -29202,6 +29197,7 @@ SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
   case X86ISD::MOVDDUP:
   case X86ISD::MOVSS:
   case X86ISD::MOVSD:
+  case X86ISD::VPERMV3:
   case X86ISD::VPERMILPI:
   case X86ISD::VPERMILPV:
   case X86ISD::VPERM2X128:
diff --git a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
index e84d76adff4..4e7714eeb2e 100644
--- a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
+++ b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
@@ -191,6 +191,7 @@ void DecodeVPERMV3Mask(const Constant *C, MVT VT,
   Type *MaskTy = C->getType();
   unsigned NumElements = MaskTy->getVectorNumElements();
   if (NumElements == VT.getVectorNumElements()) {
+    unsigned EltMaskSize = Log2_64(NumElements * 2);
     for (unsigned i = 0; i < NumElements; ++i) {
       Constant *COp = C->getAggregateElement(i);
       if (!COp) {
@@ -200,9 +201,9 @@ void DecodeVPERMV3Mask(const Constant *C, MVT VT,
       if (isa<UndefValue>(COp))
         ShuffleMask.push_back(SM_SentinelUndef);
       else {
-        uint64_t Element = cast<ConstantInt>(COp)->getZExtValue();
-        Element &= (1 << NumElements*2) - 1;
-        ShuffleMask.push_back(Element);
+        APInt Element = cast<ConstantInt>(COp)->getValue();
+        Element = Element.getLoBits(EltMaskSize);
+        ShuffleMask.push_back(Element.getZExtValue());
       }
     }
   }