Skip to content

[AMD] Implement RepOrder for AMD MMA layouts #5126

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
lezcano merged 1 commit into from
Nov 18, 2024
Merged

[AMD] Implement RepOrder for AMD MMA layouts #5126

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
5 changes: 5 additions & 0 deletions include/triton/Dialect/TritonGPU/IR/TritonGPUAttrDefs.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -777,6 +777,11 @@ def MmaEncodingTrait : AttrInterface<"MmaEncodingTrait"> {
"getSizePerThreadForOperand",
(ins "int":$opIdx,
"int":$kWidth)>,

InterfaceMethod<"Get the order of reps (tiles of this layout that tile the whole tensor). The fastest-changing axis first",
"SmallVector<unsigned>",
"getRepOrderForOperand",
(ins "int":$opIdx)>,
];
}

Expand Down
21 changes: 18 additions & 3 deletions lib/Dialect/TritonGPU/IR/Dialect.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1658,7 +1658,14 @@ AMDMfmaEncodingAttr::getInstrShapeForOperand(int kWidth, int opIdx) const {
}

SmallVector<unsigned> AMDMfmaEncodingAttr::getRepOrder() const {
llvm::report_fatal_error("NYI. AMDMfmaEncodingAttr::getRepOrder");
auto rank = getWarpsPerCTA().size();
return getMatrixOrder(rank, /*rowMajor*/ true);
}

SmallVector<unsigned>
AMDMfmaEncodingAttr::getRepOrderForOperand(int opIdx) const {
auto rank = getWarpsPerCTA().size();
return getOrderForDotOperand(opIdx, rank, /*kMajor*/ true);
}

SmallVector<int64_t>
Expand Down Expand Up @@ -1745,8 +1752,16 @@ AMDWmmaEncodingAttr::getShapePerCTATile(ArrayRef<int64_t> tensorShape) const {
return shapePerCTATile;
}
SmallVector<unsigned> AMDWmmaEncodingAttr::getRepOrder() const {
llvm::report_fatal_error("NYI. AMDWmmaEncodingAttr::getRepOrder");
auto rank = getWarpsPerCTA().size();
return getMatrixOrder(rank, /*rowMajor*/ true);
}

SmallVector<unsigned>
AMDWmmaEncodingAttr::getRepOrderForOperand(int opIdx) const {
auto rank = getWarpsPerCTA().size();
return getOrderForDotOperand(opIdx, rank, /*kMajor*/ true);
}

SmallVector<unsigned> AMDWmmaEncodingAttr::getCTAsPerCGA() const {
return SmallVector<unsigned>(getCTALayout().getCTAsPerCGA());
}
Expand Down Expand Up @@ -2016,7 +2031,7 @@ NvidiaMmaEncodingAttr::getSizePerThreadForOperand(int kWidth, int opIdx) const {
// DotOperand Encoding
//===----------------------------------------------------------------------===//
SmallVector<unsigned> DotOperandEncodingAttr::getRepOrder() const {
if (auto mma = mlir::dyn_cast<NvidiaMmaEncodingAttr>(getParent())) {
if (auto mma = mlir::dyn_cast<MmaEncodingTrait>(getParent())) {
return mma.getRepOrderForOperand(getOpIdx());
}
llvm::report_fatal_error(
Expand Down