Skip to content

[X86] Use GFNI for LZCNT vXi8 ops #141888

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Jun 4, 2025
Merged

[X86] Use GFNI for LZCNT vXi8 ops #141888

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
157a532
[X86][GFNI] Add lowering support for CTLZ vXi8 using GF2P8AFFINEQB
houngkoungting May 31, 2025
e998750
Merge branch 'main' into main
houngkoungting May 31, 2025
e789fab
[X86][GFNI] Lower vXi8 ctlz using GF2P8AFFINEQB
houngkoungting Jun 1, 2025
c087674
[X86][GFNI] Fix style and logic for CTLZ vXi8 lowering
houngkoungting Jun 2, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 32 additions & 0 deletions llvm/lib/Target/X86/X86ISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28998,6 +28998,35 @@ static SDValue LowerVectorCTLZ(SDValue Op, const SDLoc &DL,
assert(Subtarget.hasSSSE3() && "Expected SSSE3 support for PSHUFB");
return LowerVectorCTLZInRegLUT(Op, DL, Subtarget, DAG);
}
static SDValue LowerVectorCTLZ_GFNI(SDValue Op, SelectionDAG &DAG,
const X86Subtarget &Subtarget) {
SDLoc dl(Op);
MVT VT = Op.getSimpleValueType();
SDValue Input = Op.getOperand(0);

if (!VT.isVector() || VT.getVectorElementType() != MVT::i8)
return SDValue();
SmallVector<SDValue, 16> MatrixVals;
for (unsigned i = 0; i < VT.getVectorNumElements(); ++i) {
uint8_t mask = 1 << (7 - (i % 8));
MatrixVals.push_back(DAG.getConstant(mask, dl, MVT::i8));
}

SDValue Matrix = DAG.getBuildVector(VT, dl, MatrixVals);
SDValue Reversed = DAG.getNode(X86ISD::GF2P8AFFINEQB, dl, VT, Input, Matrix,
DAG.getTargetConstant(0, dl, MVT::i8));
SDValue AddMask = DAG.getConstant(0xFF, dl, MVT::i8);

SDValue AddVec = DAG.getSplatBuildVector(VT, dl, AddMask);
SDValue Summed = DAG.getNode(ISD::ADD, dl, VT, Reversed, AddVec);
SDValue NotSummed = DAG.getNode(ISD::XOR, dl, VT, Summed, AddVec);
SDValue Filtered = DAG.getNode(ISD::AND, dl, VT, NotSummed, Reversed);
SDValue FinalMatrix = DAG.getBuildVector(VT, dl, MatrixVals);
SDValue LZCNT =
DAG.getNode(X86ISD::GF2P8AFFINEQB, dl, VT, Filtered, FinalMatrix,
DAG.getTargetConstant(8, dl, MVT::i8));
return LZCNT;
}

static SDValue LowerCTLZ(SDValue Op, const X86Subtarget &Subtarget,
SelectionDAG &DAG) {
Expand All @@ -29007,6 +29036,9 @@ static SDValue LowerCTLZ(SDValue Op, const X86Subtarget &Subtarget,
SDLoc dl(Op);
unsigned Opc = Op.getOpcode();

if (VT.isVector() && VT.getScalarType() == MVT::i8 && Subtarget.hasGFNI())
return LowerVectorCTLZ_GFNI(Op, DAG, Subtarget);

if (VT.isVector())
return LowerVectorCTLZ(Op, dl, Subtarget, DAG);

Expand Down
Loading
Loading