Skip to content

[Metax][MCTLE] Add NVMMAShared encoding builder binding #1007

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

Open
cc-c122 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[Metax][MCTLE] Add NVMMAShared encoding builder binding #1007

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
31 changes: 31 additions & 0 deletions third_party/metax/plugin/mctle/triton_mctle.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -95,6 +95,37 @@ void init_triton_mctle_ir(py::module &&m) {
return mlir::cast<Attribute>(ttg::SwizzledSharedEncodingAttr::get(
context, vectorSize, perPhase, maxPhase, order, CTALayout));
})
.def("make_nv_mma_shared_encoding_attr",
[](TritonOpBuilder &self, std::vector<int64_t> shape,
std::vector<unsigned> order, Type &elemType,
std::vector<unsigned> CTAsPerCGA,
std::vector<unsigned> CTASplitNum,
std::vector<unsigned> CTAOrder, bool fp4Padded,
bool swizzled) {
assert(shape.size() == order.size());
assert(order.size() == CTAsPerCGA.size());
assert(CTAsPerCGA.size() == CTASplitNum.size());
assert(CTASplitNum.size() == CTAOrder.size());

auto context = self.getBuilder().getContext();
auto CTALayout = ttg::CTAEncodingAttr::fromSplitParams(
context, CTAsPerCGA, CTASplitNum, CTAOrder);

if (swizzled) {
return mlir::cast<Attribute>(
ttg::NVMMASharedEncodingAttr::get(
context, shape, order, CTALayout, elemType,
fp4Padded));
}

return mlir::cast<Attribute>(
ttg::NVMMASharedEncodingAttr::get(
context,
/*swizzlingByteWidth=*/0,
/*transposed=*/order[0] == 0,
elemType.getIntOrFloatBitWidth(),
fp4Padded, CTALayout));
})
.def("create_local_alloc",
[](TritonOpBuilder &self, std::vector<int64_t> shape,
Type &elementType, Attribute &encoding) -> mlir::Value {
Expand Down