wangtao
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[X86][SSE] Reuse zeroable element mask in lowerVectorShuffleAsBitMask. NFCI 

Don't regenerate a zeroable element mask with computeZeroableShuffleElements when its already available.

llvm-svn: 286039
This commit is contained in:
Simon Pilgrim 2016-11-05 17:12:19 +00:00
parent 54b0be048e
commit 009befbd88
1 changed files with 11 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
llvm/lib/Target/X86/X86ISelLowering.cpp
View File

@ -7484,6 +7484,7 @@ static SDValue lowerVectorShuffleWithUNPCK(const SDLoc &DL, MVT VT,
/// one of the inputs being zeroable.
static SDValue lowerVectorShuffleAsBitMask(const SDLoc &DL, MVT VT, SDValue V1,
SDValue V2, ArrayRef<int> Mask,
const SmallBitVector &Zeroable,
SelectionDAG &DAG) {
MVT EltVT = VT.getVectorElementType();
int NumEltBits = EltVT.getSizeInBits();
@ -7496,7 +7497,6 @@ static SDValue lowerVectorShuffleAsBitMask(const SDLoc &DL, MVT VT, SDValue V1,
AllOnes = DAG.getBitcast(EltVT, AllOnes);
}
SmallVector<SDValue, 16> VMaskOps(Mask.size(), Zero);
SmallBitVector Zeroable = computeZeroableShuffleElements(Mask, V1, V2);
SDValue V;
for (int i = 0, Size = Mask.size(); i < Size; ++i) {
if (Zeroable[i])
@ -7673,7 +7673,8 @@ static SDValue lowerVectorShuffleAsBlend(const SDLoc &DL, MVT VT, SDValue V1,
"256-bit byte-blends require AVX2 support!");
// Attempt to lower to a bitmask if we can. VPAND is faster than VPBLENDVB.
if (SDValue Masked = lowerVectorShuffleAsBitMask(DL, VT, V1, V2, Mask, DAG))
if (SDValue Masked =
lowerVectorShuffleAsBitMask(DL, VT, V1, V2, Mask, Zeroable, DAG))
return Masked;
// Scale the blend by the number of bytes per element.
@ -9488,8 +9489,8 @@ static SDValue lowerV4I32VectorShuffle(const SDLoc &DL, ArrayRef<int> Mask,
Subtarget, DAG))
return Blend;
if (SDValue Masked =
lowerVectorShuffleAsBitMask(DL, MVT::v4i32, V1, V2, Mask, DAG))
if (SDValue Masked = lowerVectorShuffleAsBitMask(DL, MVT::v4i32, V1, V2, Mask,
Zeroable, DAG))
return Masked;
// Use dedicated unpack instructions for masks that match their pattern.
@ -10127,8 +10128,8 @@ static SDValue lowerV8I16VectorShuffle(const SDLoc &DL, ArrayRef<int> Mask,
Subtarget, DAG))
return Blend;
if (SDValue Masked =
lowerVectorShuffleAsBitMask(DL, MVT::v8i16, V1, V2, Mask, DAG))
if (SDValue Masked = lowerVectorShuffleAsBitMask(DL, MVT::v8i16, V1, V2, Mask,
Zeroable, DAG))
return Masked;
// Use dedicated unpack instructions for masks that match their pattern.
@ -10366,8 +10367,8 @@ static SDValue lowerV16I8VectorShuffle(const SDLoc &DL, ArrayRef<int> Mask,
return V;
}
if (SDValue Masked =
lowerVectorShuffleAsBitMask(DL, MVT::v16i8, V1, V2, Mask, DAG))
if (SDValue Masked = lowerVectorShuffleAsBitMask(DL, MVT::v16i8, V1, V2, Mask,
Zeroable, DAG))
return Masked;
// Use dedicated unpack instructions for masks that match their pattern.
@ -11881,7 +11882,8 @@ static SDValue lower256BitVectorShuffle(const SDLoc &DL, ArrayRef<int> Mask,
if (ElementBits < 32) {
// No floating point type available, if we can't use the bit operations
// for masking/blending then decompose into 128-bit vectors.
if (SDValue V = lowerVectorShuffleAsBitMask(DL, VT, V1, V2, Mask, DAG))
if (SDValue V =
lowerVectorShuffleAsBitMask(DL, VT, V1, V2, Mask, Zeroable, DAG))
return V;
if (SDValue V = lowerVectorShuffleAsBitBlend(DL, VT, V1, V2, Mask, DAG))
return V;