Skip to content

[KVCache] Fix attention prefill kernel for Metal and Android #17539

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
Nov 21, 2024
Merged

[KVCache] Fix attention prefill kernel for Metal and Android #17539

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: 6 additions & 1 deletion python/tvm/relax/frontend/nn/llm/kv_cache.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1579,6 +1579,12 @@ def _attention_prefill_ragged(h_kv, h_q, d, dtype, rope_scaling: Dict[str, Any],
d,
64 // ((DataType(dtype).bits + 7) // 8) // max(d // 128, 1),
)
original_tile_y = tile_y
original_tile_z = tile_z
while (tile_x * tile_z) % (bdx * num_warps) != 0:
tile_z += original_tile_z
while (tile_x * tile_y) % (bdx * num_warps) != 0:
tile_y += original_tile_y

# Otherwise we would exceed maxComputeWorkgroupStorageSize
if (
Expand Down Expand Up @@ -1907,7 +1913,6 @@ def apply_to_gemm( # pylint: disable=unused-argument
sch.unroll(yio)
sch.vectorize(yiv)
sch.unroll(xi)
sch.unroll(ki)
sch.decompose_reduction(block, ty)

def apply_to_md(sch, block):
Expand Down
Loading