Skip to content

[BENCHMARKS] Update Cutlass to v0.5 #5510

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
Nov 21, 2025
+14 −13
Merged

[BENCHMARKS] Update Cutlass to v0.5 #5510

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
23 changes: 12 additions & 11 deletions benchmarks/cutlass_kernel/attention/attention.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,11 +20,11 @@ template <typename FMHA> static auto run(typename FMHA::Params params) -> void {

int smem_size = FMHA::SharedStorageSize;

const auto sycl_block = syclcompat::dim3(block.x, block.y, block.z);
const auto sycl_grid = syclcompat::dim3(grid.x, grid.y, grid.z);
const auto sycl_block = compat::dim3(block.x, block.y, block.z);
const auto sycl_grid = compat::dim3(grid.x, grid.y, grid.z);

#if !defined(SYCL_EXT_ONEAPI_WORK_GROUP_SCRATCH_MEMORY)
using namespace syclcompat::experimental;
using namespace compat::experimental;
auto event = launch<cutlass::device_kernel<FMHA>>(
launch_policy{
sycl_grid, sycl_block,
Expand All @@ -33,15 +33,15 @@ template <typename FMHA> static auto run(typename FMHA::Params params) -> void {
sycl_exp::sub_group_size<FMHA::DispatchPolicy::SubgroupSize>}},
params);
#else
syclcompat::experimental::launch_properties launch_props{
compat::experimental::launch_properties launch_props{
sycl::ext::oneapi::experimental::work_group_scratch_size(smem_size),
};
syclcompat::experimental::kernel_properties kernel_props{
compat::experimental::kernel_properties kernel_props{
sycl::ext::oneapi::experimental::sub_group_size<
FMHA::DispatchPolicy::SubgroupSize>};
syclcompat::experimental::launch_policy policy{sycl_grid, sycl_block,
launch_props, kernel_props};
auto event = syclcompat::experimental::launch<cutlass::device_kernel<FMHA>>(
compat::experimental::launch_policy policy{sycl_grid, sycl_block,
launch_props, kernel_props};
auto event = compat::experimental::launch<cutlass::device_kernel<FMHA>>(
policy, params);
#endif

Expand Down Expand Up @@ -102,8 +102,9 @@ static auto attention_run(const at::Tensor &Q, const at::Tensor &K,
using CollectiveEpilogue =
cutlass::flash_attention::collective::FlashPrefillEpilogue<
EpilogueDispatchPolicy, MMAOperation, TileShapeOutput, SubgroupLayout,
ElementAccumulator, cutlass::gemm::TagToStrideC_t<LayoutO>,
ElementOutput, GmemTiledCopyStore>;
ElementAccumulator, ElementOutput,
Copy link
Contributor Author

@anmyachev anmyachev Nov 20, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

intel/sycl-tla@d5f1886

cutlass::gemm::TagToStrideC_t<LayoutO>, ElementOutput,
GmemTiledCopyStore>;

/// FA ///

Expand Down Expand Up @@ -181,7 +182,7 @@ static auto attention_run(const at::Tensor &Q, const at::Tensor &K,
FMHAPrefillKernel::to_underlying_arguments(arguments, workspace_ptr);
run<FMHAPrefillKernel>(params);

syclcompat::wait();
compat::wait();

} catch (std::exception &e) {
std::cerr << "Runtime error: " << e.what() << std::endl;
Expand Down
2 changes: 1 addition & 1 deletion benchmarks/cutlass_kernel/cutlass-library.conf
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1 +1 @@
dd43242ea2f3e08e73a73153f00a5dbe5a31c41c
b0cb10e655d8f9b1d0474e9538a82d218f74c694
2 changes: 1 addition & 1 deletion benchmarks/cutlass_kernel/gemm/gemm.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,7 +86,7 @@ static auto gemm_run(const at::Tensor &A, const at::Tensor &B, at::Tensor &C,
CUTLASS_CHECK(gemm_op.initialize(arguments, workspace.get()));
CUTLASS_CHECK(gemm_op.run());

syclcompat::wait();
compat::wait();

} catch (std::exception &e) {
std::cerr << "Runtime error: " << e.what() << std::endl;
Expand Down