Skip to content

Adds column stride support to tensor memory layouts #41

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
Jun 30, 2025
Merged

Adds column stride support to tensor memory layouts #41

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
8 changes: 4 additions & 4 deletions csrc/src/flash_fwd_kernel.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -180,19 +180,19 @@ inline __device__ void compute_attn_1rowblock(const Params &params, const int bi
make_stride(params.seqlen_k_rounded, _1{})
);
Tensor mZOH = make_tensor(
make_gmem_ptr(reinterpret_cast<Element*>(params.zoh_ptr) + binfo.zoh_offset(params.zoh_batch_stride, params.zoh_row_stride, bidb)),
make_gmem_ptr(reinterpret_cast<Element*>(params.zoh_ptr) + binfo.zoh_offset(params.zoh_batch_stride, params.zoh_row_stride, params.zoh_col_stride, bidb)),
make_shape(params.h_k, binfo.actual_seqlen_q, binfo.actual_seqlen_k),
make_stride(params.zoh_head_stride, params.zoh_row_stride, _1{})
make_stride(params.zoh_head_stride, params.zoh_row_stride, params.zoh_col_stride)
);
Tensor gZOH = local_tile(
mZOH(bidh / params.h_h_k_ratio, _, _),
Shape<Int<kBlockM>, Int<kBlockN>>{},
make_coord(m_block, _)
); // (kBlockM, kBlockN, nblocksN)
Tensor mActiveMask = make_tensor(
make_gmem_ptr(reinterpret_cast<Element*>(params.active_mask_ptr) + binfo.active_mask_offset(params.active_mask_batch_stride, params.active_mask_row_stride, bidb)),
make_gmem_ptr(reinterpret_cast<Element*>(params.active_mask_ptr) + binfo.active_mask_offset(params.active_mask_batch_stride, params.active_mask_row_stride, params.active_mask_col_stride, bidb)),
make_shape(params.h_k, binfo.actual_seqlen_q, binfo.actual_seqlen_k),
make_stride(params.active_mask_head_stride, params.active_mask_row_stride, _1{})
make_stride(params.active_mask_head_stride, params.active_mask_row_stride, params.active_mask_col_stride)
);
Tensor gActiveMask = local_tile(
mActiveMask(bidh / params.h_h_k_ratio, _, _),
Expand Down