Skip to content

Reuse tile prefix in recurrence kernel #29

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

Open
0z5a wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Reuse tile prefix in recurrence kernel #29

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
7 changes: 2 additions & 5 deletions csrc/smxx/fwd_kernel2.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -147,6 +147,7 @@ __global__ void __launch_bounds__(NumThreads) _flash_kda_fwd_recurrence(
int H,
int N,
int64_t const* cu_seqlens,
int const* tile_prefix,
int total_tiles
) {
using BF16 = cutlass::bfloat16_t;
Expand Down Expand Up @@ -221,11 +222,7 @@ __global__ void __launch_bounds__(NumThreads) _flash_kda_fwd_recurrence(
if constexpr (IsVarlen) {
bos = cu_seqlens[seq_idx];
eos = cu_seqlens[seq_idx + 1];
// Compute tile_base via linear scan (no host-precomputed table)
tile_base = 0;
for (int i = 0; i < seq_idx; i++) {
tile_base += (int(cu_seqlens[i + 1] - cu_seqlens[i]) + CHUNK - 1) / CHUNK;
}
tile_base = tile_prefix[seq_idx];
} else {
int T_seq = T_total / N;
bos = seq_idx * T_seq;
Expand Down
2 changes: 1 addition & 1 deletion csrc/smxx/fwd_launch.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -210,7 +210,7 @@ void launch_fwd(
tma_load_initial_state,
tma_store_final_state,
tma_store_out,
out_ptr, T_total, H, N, cu_seqlens_ptr, total_tiles
out_ptr, T_total, H, N, cu_seqlens_ptr, ws_tile_prefix, total_tiles
);
}
#endif
Expand Down