Skip to content

[Bugfix][Kernel] Only set PDL launch attr for MoE top-k gating on SM90+ #65059

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
gau-nernst merged 4 commits into from
Nov 2, 2026
+3 −1
Merged

[Bugfix][Kernel] Only set PDL launch attr for MoE top-k gating on SM90+ #65059

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
dcaf590
[Bugfix][Kernel] Only set PDL launch attr for MoE top-k gating on SM90+
srkim-business Nov 2, 2026
303fe03
Merge branch 'main' into fix/topk-gating-pdl-pre-sm90
vllm-agent Nov 2, 2026
0390140
Merge branch 'main' into fix/topk-gating-pdl-pre-sm90
srkim-business Nov 2, 2026
220fa30
Merge branch 'main' into fix/topk-gating-pdl-pre-sm90
gau-nernst Nov 2, 2026
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
4 changes: 3 additions & 1 deletion csrc/moe/topk_softmax_kernels.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -627,6 +627,8 @@ void topkGatingLauncherHelper(const InputType* input, const bool* finished, floa

dim3 block_dim(WARP_SIZE_PARAM, WARPS_PER_TB);
#ifndef USE_ROCM
// Pre-SM90 cannot use PDL, but the attribute still adds a programmatic edge to captured CUDA graphs.
static const bool use_pdl = get_device_prop()->major >= 9;
cudaLaunchConfig_t config{};
config.gridDim = dim3(num_blocks);
config.blockDim = block_dim;
Expand All @@ -635,7 +637,7 @@ void topkGatingLauncherHelper(const InputType* input, const bool* finished, floa
attr.id = cudaLaunchAttributeProgrammaticStreamSerialization;
attr.val.programmaticStreamSerializationAllowed = 1;
config.attrs = &attr;
config.numAttrs = 1;
config.numAttrs = use_pdl ? 1 : 0;
auto err = cudaLaunchKernelEx(
&config,
topkGating<VPT, EXPERTS, WARPS_PER_TB, BYTES_PER_LDG, WARP_SIZE_PARAM, IndType, InputType, SF>,
Expand Down