Skip to content

Fix KV chunking for POD. #1054

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
May 13, 2025
Merged

Fix KV chunking for POD. #1054

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
86b6fc7
Fix KV chunking for POD. For some reason cudaOccupancyMaxActiveBlocks…
AKKamath May 13, 2025
dfc9af1
Run pre-commit and fix styling
AKKamath May 13, 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
7 changes: 5 additions & 2 deletions include/flashinfer/attention/pod.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -346,8 +346,11 @@ cudaError_t PODWithKVCacheTensorDispatched(PrefillParams prefill_params,
int num_sm = 0;
FLASHINFER_CUDA_CALL(
cudaDeviceGetAttribute(&num_sm, cudaDevAttrMultiProcessorCount, dev_id));
FLASHINFER_CUDA_CALL(cudaOccupancyMaxActiveBlocksPerMultiprocessor(
&num_blocks_per_sm, kernel, num_threads_p, smem_size_p));
// FLASHINFER_CUDA_CALL(cudaOccupancyMaxActiveBlocksPerMultiprocessor(
Copy link
Collaborator

@yzh119 yzh119 May 13, 2025

Choose a reason for hiding this comment

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

It's interesting to me, and likely a bug of cudaOccupancyMaxActiveBlocksPerMultiprocessor.
Let's merge this first, thanks for the contribution!

Copy link
Contributor

@Edenzzzz Edenzzzz May 13, 2025

Choose a reason for hiding this comment

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

cudaOccupancyMaxActiveBlocksPerMultiprocessor is buggy on both A100 40G and H100 for me

// &num_blocks_per_sm, kernel, num_threads_p, smem_size_p));
// Above function returns 0 for some reason, so we use a workaround
num_blocks_per_sm = std::max(
1, std::min((int)(max_smem_per_sm / smem_size_p), (int)(256 / num_threads_p)));
uint32_t max_num_kv_chunks =
(num_blocks_per_sm * num_sm) /
(num_kv_heads * ceil_div(qo_len * group_size, KTraits_P::CTA_TILE_Q));
Expand Down