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

[AVX-512] Simplify code for creating 512-bit SHUF128 operations. 

We don't need two loops and we can safely assume assume and hardcode the size of the widened mask.

llvm-svn: 290942
This commit is contained in:
Craig Topper 2017-01-04 07:31:51 +00:00
parent 6039656441
commit 83115a809f
1 changed files with 11 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
llvm/lib/Target/X86/X86ISelLowering.cpp
View File

@ -12616,33 +12616,26 @@ static SDValue lowerV4X128VectorShuffle(const SDLoc &DL, MVT VT,
return DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, LoV, HiV); return DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, LoV, HiV);
} }
// Try to lower to to vshuf64x2/vshuf32x4.
assert(WidenedMask.size() == 4);
SDValue Ops[2] = {DAG.getUNDEF(VT), DAG.getUNDEF(VT)}; SDValue Ops[2] = {DAG.getUNDEF(VT), DAG.getUNDEF(VT)};
unsigned PermMask = 0;
// Insure elements came from the same Op. // Insure elements came from the same Op.
int MaxOp1Index = VT.getVectorNumElements()/2 - 1; for (int i = 0; i < 4; ++i) {
for (int i = 0, Size = WidenedMask.size(); i < Size; ++i) { assert(WidenedMask[i] >= -1);
if (WidenedMask[i] == SM_SentinelZero) if (WidenedMask[i] < 0)
return SDValue();
if (WidenedMask[i] == SM_SentinelUndef)
continue; continue;
SDValue Op = WidenedMask[i] > MaxOp1Index ? V2 : V1; SDValue Op = WidenedMask[i] >= 4 ? V2 : V1;
unsigned OpIndex = (i < Size/2) ? 0 : 1; unsigned OpIndex = i / 2;
if (Ops[OpIndex].isUndef()) if (Ops[OpIndex].isUndef())
Ops[OpIndex] = Op; Ops[OpIndex] = Op;
else if (Ops[OpIndex] != Op) else if (Ops[OpIndex] != Op)
return SDValue(); return SDValue();
}
// Form a 128-bit permutation. // Convert the 128-bit shuffle mask selection values into 128-bit selection
// Convert the 64-bit shuffle mask selection values into 128-bit selection // bits defined by a vshuf64x2 instruction's immediate control byte.
// bits defined by a vshuf64x2 instruction's immediate control byte. PermMask |= (WidenedMask[i] % 4) << (i * 2);
unsigned PermMask = 0, Imm = 0;
unsigned ControlBitsNum = WidenedMask.size() / 2;
for (int i = 0, Size = WidenedMask.size(); i < Size; ++i) {
// Use first element in place of undef mask.
Imm = (WidenedMask[i] == SM_SentinelUndef) ? 0 : WidenedMask[i];
PermMask |= (Imm % WidenedMask.size()) << (i * ControlBitsNum);
} }
return DAG.getNode(X86ISD::SHUF128, DL, VT, Ops[0], Ops[1], return DAG.getNode(X86ISD::SHUF128, DL, VT, Ops[0], Ops[1],