[FMHA] Support Vectorized KV Cache Layout and vLLM/SGLang block table in Batch Prefill kernel#1754

Merged

valarLip merged 34 commits into

mainfrom

batch_prefill_page_size_16_rebase

Jan 13, 2026

Commits on Dec 30, 2025

add page size 16 to test and op

ltqin
authored and
Jeff-Huang
committed
add num_total_pages to kernel parameter

ltqin
authored and
Jeff-Huang
committed
add is_sglang parameter

ltqin
authored and
Jeff-Huang
committed
chang is_sglang to is_sglang_layout

ltqin
authored and
Jeff-Huang
committed
kv last page size=16 pass

ltqin
authored and
Jeff-Huang
committed
pass kv_last_page_lens to kernel

ltqin
authored and
Jeff-Huang
committed
add parameters check before calling kernel

ltqin
authored and
Jeff-Huang
committed
change kv layout to [page_num, page_size, nhead, hdim]

ltqin
authored and
Jeff-Huang
committed
adopt the changes of struct fmha_fwd_batch_prefill_traits
Jeff-Huang
committed
change kv cache memory layout to [num_blocks, num_kv_heads, head_size/8, block_size, 8], [num_blocks, num_kv_heads, block_size/8, head_size, 8]
Jeff-Huang
committed
[FMHA] Integrate vLLM block table support and enforce vectorized KV layout
Jeff-Huang
committed
update CK
Jeff-Huang
committed
Merge branch 'main' into batch_prefill_page_size_16_rebase
Jeff-Huang
committed
update ck
Jeff-Huang
committed
adopt api changes from fmha_batch_prefill_traits
Jeff-Huang
committed

Commits on Jan 5, 2026

Commits on Jan 6, 2026

Add profile measurements for batch prefill function
Jeff-Huang
committed

Commits on Jan 8, 2026

Merge branch 'main' into batch_prefill_page_size_16_rebase
Jeff-Huang
committed

Commits on Jan 9, 2026

Merge branch 'main' into batch_prefill_page_size_16_rebase
Jeff-Huang
committed

Commits on Jan 12, 2026

Commits on Jan 13, 2026