diff options
| author | Michael Liao <michael.liao@intel.com> | 2012-08-31 20:12:31 +0000 |
|---|---|---|
| committer | Michael Liao <michael.liao@intel.com> | 2012-08-31 20:12:31 +0000 |
| commit | 265bcb1e5b106a7c5db2bfcfb13cceffe0c413be (patch) | |
| tree | 4b71ac71eae05fb78c0c2225f0dc2e843c84bd53 /lib | |
| parent | 3185f9a2ea80afec30064b7cd095f82c31dc154e (diff) | |
| download | external_llvm-265bcb1e5b106a7c5db2bfcfb13cceffe0c413be.tar.gz external_llvm-265bcb1e5b106a7c5db2bfcfb13cceffe0c413be.tar.bz2 external_llvm-265bcb1e5b106a7c5db2bfcfb13cceffe0c413be.zip | |
Fix PR12359
- In addition to undefined, if V2 is zero vector, skip 2nd PSHUFB and POR as
well as PSHUFB will zero elements with negative indices.
Patch by Sriram Murali <sriram.murali@intel.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@163018 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib')
| -rw-r--r-- | lib/Target/X86/X86ISelLowering.cpp | 8 |
1 files changed, 5 insertions, 3 deletions
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp index f9184f693d..f4329d3bf9 100644 --- a/lib/Target/X86/X86ISelLowering.cpp +++ b/lib/Target/X86/X86ISelLowering.cpp @@ -5881,8 +5881,6 @@ SDValue LowerVECTOR_SHUFFLEv16i8(ShuffleVectorSDNode *SVOp, DebugLoc dl = SVOp->getDebugLoc(); ArrayRef<int> MaskVals = SVOp->getMask(); - bool V2IsUndef = V2.getOpcode() == ISD::UNDEF; - // If we have SSSE3, case 1 is generated when all result bytes come from // one of the inputs. Otherwise, case 2 is generated. If no SSSE3 is // present, fall back to case 3. @@ -5906,7 +5904,11 @@ SDValue LowerVECTOR_SHUFFLEv16i8(ShuffleVectorSDNode *SVOp, V1 = DAG.getNode(X86ISD::PSHUFB, dl, MVT::v16i8, V1, DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v16i8, &pshufbMask[0], 16)); - if (V2IsUndef) + + // As PSHUFB will zero elements with negative indices, it's safe to ignore + // the 2nd operand if it's undefined or zero. + if (V2.getOpcode() == ISD::UNDEF || + ISD::isBuildVectorAllZeros(V2.getNode())) return V1; // Calculate the shuffle mask for the second input, shuffle it, and |