Skip to content

feat: use global residue_mnk for oob handling #476

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
guocuimi merged 3 commits into from
Jun 25, 2025
Merged

feat: use global residue_mnk for oob handling #476

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
68af476
feat: use global residue_mnk for oob handling
guocuimi Jun 25, 2025
147084d
Merge branch 'main' into attn_residue
guocuimi Jun 25, 2025
82c5ca5
move tile handling into kernel
guocuimi Jun 25, 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
157 changes: 0 additions & 157 deletions src/kernels/attention/mha_tile.h
View file
Open in desktop

This file was deleted.

147 changes: 0 additions & 147 deletions src/kernels/attention/mla_tile.h
View file
Open in desktop

This file was deleted.

31 changes: 13 additions & 18 deletions src/kernels/attention/sm80_collective_epilogue.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,21 +71,19 @@ struct Sm80CollectiveEpilogue {
template <class FrgTensor,
class TiledMma,
class TensorO,
class BlockCoordMNK,
class ProblemShapeMNK>
CUTE_DEVICE void operator()(const Params& /*params*/,
const FrgTensor& tOrAccO, // (MMA, MMA_M, MMA_N)
TiledMma tiled_mma,
TensorO& gO, // (BLK_M, HEAD_DIM)
int tidx,
const BlockCoordMNK& block_coord_mnk,
const ProblemShapeMNK& problem_shape_mnk,
char* smem) {
class TensorCO,
class ResidueMNK>
CUTE_DEVICE void operator()(
const Params& /*params*/,
const FrgTensor& tOrAccO, // (MMA, MMA_M, MMA_N)
TiledMma tiled_mma,
TensorO& gO, // (BLK_M, HEAD_DIM)
const TensorCO& cO, // (BLK_M, HEAD_DIM) => (M, K)
int tidx,
const ResidueMNK& residue_mnk,
char* smem) {
static constexpr int kBlockM = get<0>(TileShape{});

const auto [batch_idx, m_block_idx, kv_head_idx] = block_coord_mnk;
const auto [q_packed_len, kv_len, head_dim] = problem_shape_mnk;

// Smem
auto& ss = *reinterpret_cast<SharedStorage*>(smem);
// (BLK_M, HEAD_DIM)
Expand All @@ -106,9 +104,6 @@ struct Sm80CollectiveEpilogue {
GmemTiledCopyO gmem_tiled_copy_O;
auto gmem_thr_copy_O = gmem_tiled_copy_O.get_thread_slice(tidx);

// (BLK_M, HEAD_DIM) -> (blk_m, head_dim)
auto cO = make_identity_tensor(Shape<BLK_M, HEAD_DIM>{});

auto tOsO = gmem_thr_copy_O.partition_S(sO); // (CPY,CPY_M,CPY_K)
auto tOgO = gmem_thr_copy_O.partition_D(gO); // (CPY,CPY_M,CPY_K)
// (CPY,CPY_M,CPY_K) -> (blk_m, head_dim)
Expand All @@ -117,9 +112,9 @@ struct Sm80CollectiveEpilogue {
// wait for smem copy done before gmem copy
__syncthreads();

auto max_coord = make_coord(q_packed_len - m_block_idx * kBlockM, head_dim);
const auto residue_mk = select<0, 2>(residue_mnk);
safe_copy</*EVEN_M=*/false, EVEN_K, /*ZFILL_M=*/false, /*ZFILL_K=*/false>(
gmem_tiled_copy_O, tOsO, tOgO, tOcO, max_coord);
gmem_tiled_copy_O, tOsO, tOgO, tOcO, residue_mk);
}
};
} // namespace llm
Loading