Skip to content

[BugFix][DP/EP] Fix CUTLASS MLA hang under load #26026

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 2 commits into from
Oct 1, 2025
+32 −32
Merged

[BugFix][DP/EP] Fix CUTLASS MLA hang under load #26026

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
3a8a634
arrive and wait
LucasWilkinson Oct 1, 2025
3c496cb
remove tabs
LucasWilkinson Oct 1, 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
64 changes: 32 additions & 32 deletions csrc/attention/mla/cutlass_sm100_mla/kernel/sm100_fmha_mla_tma_warpspecialized.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -580,22 +580,22 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
for (; tile_scheduler.is_valid(); ++tile_scheduler) {
auto blk_coord = tile_scheduler.get_block_coord();
auto problem_shape = params.problem_shape;
auto local_split_kv = params.split_kv;
auto local_split_kv = params.split_kv;
if (params.mainloop.ptr_seq != nullptr) {
get<1>(problem_shape) = params.mainloop.ptr_seq[get<2>(blk_coord)];
if (params.ptr_split_kv != nullptr) {
if (params.ptr_split_kv != nullptr) {
local_split_kv = params.ptr_split_kv[get<2>(blk_coord)];
}
}
if (local_split_kv <= get<3>(blk_coord))
continue;
if (local_split_kv <= get<3>(blk_coord))
continue;
load_page_table(
blk_coord,
problem_shape,
params.mainloop,
shared_storage.tensors,
pipeline_page_table, pipeline_pt_producer_state,
local_split_kv
local_split_kv
);
}
}
Expand All @@ -604,15 +604,15 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
CUTLASS_PRAGMA_NO_UNROLL
for (; tile_scheduler.is_valid(); ++tile_scheduler) {
auto blk_coord = tile_scheduler.get_block_coord();
auto problem_shape = params.problem_shape;
auto local_split_kv = params.split_kv;
auto problem_shape = params.problem_shape;
auto local_split_kv = params.split_kv;
if (params.mainloop.ptr_seq != nullptr) {
get<1>(problem_shape) = params.mainloop.ptr_seq[get<2>(blk_coord)];
if (params.ptr_split_kv != nullptr) {
if (params.ptr_split_kv != nullptr) {
local_split_kv = params.ptr_split_kv[get<2>(blk_coord)];
}
}
if (local_split_kv <= get<3>(blk_coord))
if (local_split_kv <= get<3>(blk_coord))
continue;
load_cpasync(
blk_coord,
Expand All @@ -621,7 +621,7 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
params.mainloop_params,
shared_storage.tensors,
pipeline_load_qk, pipeline_load_qk_producer_state,
local_split_kv,
local_split_kv,
/* must be shared pipe */
pipeline_page_table, pipeline_pt_consumer_state
);
Expand All @@ -633,15 +633,15 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
CUTLASS_PRAGMA_NO_UNROLL
for (; tile_scheduler.is_valid(); ++tile_scheduler) {
auto blk_coord = tile_scheduler.get_block_coord();
auto problem_shape = params.problem_shape;
auto local_split_kv = params.split_kv;
auto problem_shape = params.problem_shape;
auto local_split_kv = params.split_kv;
if (params.mainloop.ptr_seq != nullptr) {
get<1>(problem_shape) = params.mainloop.ptr_seq[get<2>(blk_coord)];
if (params.ptr_split_kv != nullptr) {
local_split_kv = params.ptr_split_kv[get<2>(blk_coord)];
}
if (params.ptr_split_kv != nullptr) {
local_split_kv = params.ptr_split_kv[get<2>(blk_coord)];
}
}
if (local_split_kv <= get<3>(blk_coord))
if (local_split_kv <= get<3>(blk_coord))
continue;
load_tma</* paged= */ true>(
blk_coord,
Expand All @@ -651,7 +651,7 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
shared_storage.tensors,
pipeline_load_qk, pipeline_load_qk_producer_state,
pipeline_load_qk, pipeline_load_qk_producer_state,
local_split_kv
local_split_kv
);
cutlass::arch::NamedBarrier((kNumComputeWarps + kNumLoadWarps) * NumThreadsPerWarp, kNamedBarrierEpilogue).arrive_and_wait();
}
Expand All @@ -660,15 +660,15 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
CUTLASS_PRAGMA_NO_UNROLL
for (; tile_scheduler.is_valid(); ++tile_scheduler) {
auto blk_coord = tile_scheduler.get_block_coord();
auto problem_shape = params.problem_shape;
auto local_split_kv = params.split_kv;
auto problem_shape = params.problem_shape;
auto local_split_kv = params.split_kv;
if (params.mainloop.ptr_seq != nullptr) {
get<1>(problem_shape) = params.mainloop.ptr_seq[get<2>(blk_coord)];
if (params.ptr_split_kv != nullptr) {
if (params.ptr_split_kv != nullptr) {
local_split_kv = params.ptr_split_kv[get<2>(blk_coord)];
}
}
}
if (local_split_kv <= get<3>(blk_coord))
if (local_split_kv <= get<3>(blk_coord))
continue;
load_tma<false>(
blk_coord,
Expand All @@ -678,7 +678,7 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
shared_storage.tensors,
pipeline_load_qk, pipeline_load_qk_producer_state,
pipeline_load_qk, pipeline_load_qk_producer_state,
local_split_kv
local_split_kv
);
cutlass::arch::NamedBarrier((kNumComputeWarps + kNumLoadWarps) * NumThreadsPerWarp, kNamedBarrierEpilogue).arrive_and_wait();
}
Expand All @@ -694,14 +694,14 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
for (; tile_scheduler.is_valid(); ++tile_scheduler) {
auto blk_coord = tile_scheduler.get_block_coord();
auto problem_shape = params.problem_shape;
auto local_split_kv = params.split_kv;
auto local_split_kv = params.split_kv;
if (params.mainloop.ptr_seq != nullptr) {
get<1>(problem_shape) = params.mainloop.ptr_seq[get<2>(blk_coord)];
if (params.ptr_split_kv != nullptr) {
local_split_kv = params.ptr_split_kv[get<2>(blk_coord)];
}
}
if (local_split_kv <= get<3>(blk_coord))
if (local_split_kv <= get<3>(blk_coord))
continue;
mma(blk_coord,
problem_shape,
Expand All @@ -711,7 +711,7 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
pipeline_mma_s, pipeline_mma_s_producer_state,
pipeline_p_mma, pipeline_p_mma_consumer_state,
pipeline_mma_o, pipeline_mma_o_producer_state,
local_split_kv
local_split_kv
);
}
}
Expand All @@ -726,15 +726,15 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
for (; tile_scheduler.is_valid(); ++tile_scheduler) {
auto blk_coord = tile_scheduler.get_block_coord();
auto problem_shape = params.problem_shape;
auto split_kv = params.split_kv;
auto local_split_kv = split_kv;
auto split_kv = params.split_kv;
auto local_split_kv = split_kv;
if (params.mainloop.ptr_seq != nullptr) {
get<1>(problem_shape) = params.mainloop.ptr_seq[get<2>(blk_coord)];
if (params.ptr_split_kv != nullptr) {
if (params.ptr_split_kv != nullptr) {
local_split_kv = params.ptr_split_kv[get<2>(blk_coord)];
}
}
if (local_split_kv <= get<3>(blk_coord))
if (local_split_kv <= get<3>(blk_coord))
continue;
compute(
blk_coord,
Expand All @@ -745,7 +745,7 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
pipeline_mma_s, pipeline_mma_s_consumer_state,
pipeline_p_mma, pipeline_p_mma_producer_state,
pipeline_mma_o, pipeline_mma_o_consumer_state,
local_split_kv
local_split_kv
);
}

Expand Down Expand Up @@ -1900,7 +1900,7 @@ struct Sm100FmhaMlaKernelTmaWarpspecialized {
cutlass::arch::NamedBarrier(
(kNumComputeWarps + kNumLoadWarps) * NumThreadsPerWarp,
kNamedBarrierEpilogue
).arrive();
).arrive_and_wait();
Copy link
Collaborator

Choose a reason for hiding this comment

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

note for reviewer --- this is the line that fixes the deadlock


return;
}
Expand Down