Skip to content

Fixing trD compute type in the Xe Epilogue #580

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 1 commit into from
Oct 28, 2025
Merged

Fixing trD compute type in the Xe Epilogue #580

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
7 changes: 5 additions & 2 deletions include/cutlass/epilogue/collective/xe_epilogue.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -389,7 +389,10 @@ class CollectiveEpilogue<
cst_callbacks.begin();

auto acc_frag = recast<Array<ElementCompute, FragmentSize>>(accumulators);
auto trD_compute_frag = recast<Array<ElementCompute, FragmentSize>>(trD_compute);
using FragmentVisit = decltype(cst_callbacks.visit(acc_frag(0), 0, 0, 0));
constexpr bool IsDirectR2S = cute::is_same_v<FragmentVisit, Array<ElementD, FragmentSize>>;
using RegisterElementD = cute::conditional_t<!IsDirectR2S, ElementCompute, ElementD>;
auto trD_compute_frag = recast<Array<RegisterElementD, FragmentSize>>(trD_compute);

Tensor trD = make_tensor<ElementOutput>(Shape<Int<FragmentSize>>{});
auto trD_frag = recast<Array<ElementOutput, FragmentSize>>(trD);
Expand Down Expand Up @@ -423,7 +426,7 @@ class CollectiveEpilogue<
if constexpr (is_destination_supported) {
CUTLASS_PRAGMA_UNROLL
for (int i = 0; i < size(trD_compute_frag); ++i) {
trD_frag(i) = cutlass::NumericArrayConverter<ElementOutput, ElementCompute, FragmentSize>{}(trD_compute_frag(i));
trD_frag(i) = cutlass::NumericArrayConverter<ElementOutput, RegisterElementD, FragmentSize>{}(trD_compute_frag(i));
}
copy(params.xe_store_d, trD, tCgD(_, epi_m, epi_n));
}
Expand Down
Loading