[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

1 files changed, 4 insertions, 3 deletions

diff --git a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
index e84d76a..4e7714e 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());
       }
     }
   }