ViperEkura d565d44c43 fix: harden attention kernel boundaries ...

- fix scalar prefill head_dim=32 out-of-bounds via G=4 dispatch
- fix MMA decode 4D mask head indexing and invalid-row mask access
- add q_head/kv_head divisibility and head-dim contiguity checks
- validate split-KV scratch and decode out_buf layout in bindings
- set max dynamic shared memory for scalar decode D=256
- cover scalar prefill D=32 in pure C test

2026-08-09 14:53:24 +08:00

..

attn_common.h

refactor: unify paged and contiguous attention kernels via KVSource policy

2026-08-05 14:06:13 +08:00

attn_decode_split_kv_mma.cuh

fix: harden attention kernel boundaries

2026-08-09 14:53:24 +08:00

attn_decode_split_kv.cuh

perf: preload V in decode split-kv shared mem and cache partial tensors

2026-08-06 18:27:00 +08:00

attn_decode.cu

fix: harden attention kernel boundaries

2026-08-09 14:53:24 +08:00

attn_dispatchers.cuh

fix: harden attention kernel boundaries

2026-08-09 14:53:24 +08:00

attn_entry_utils.cuh

fix: harden attention kernel boundaries

2026-08-09 14:53:24 +08:00

attn_kv_source.cuh

fix: use max_context_len for stable num_splits in paged decode

2026-08-07 14:42:53 +08:00

attn_mma_utils.cuh

fix: harden attention kernel boundaries

2026-08-09 14:53:24 +08:00

attn_paged_decode.cu

fix: harden attention kernel boundaries

2026-08-09 14:53:24 +08:00

attn_paged_prefill.cu

refactor: unify paged and contiguous attention kernels via KVSource policy

2026-08-05 14:06:13 +08:00

attn_prefill_split_q_mma.cuh

fix: harden attention kernel boundaries

2026-08-09 14:53:24 +08:00

attn_prefill_split_q.cuh

refactor: unify paged and contiguous attention kernels via KVSource policy

2026-08-05 14:06:13 +08:00

attn_prefill.cu

perf: launch CUDA kernels on torch's current stream

2026-08-02 13:20:14 +08:00

attn_warp_utils.cuh

perf: extend MMA decode to arbitrary GQA ratio, add launch bounds, vectorize combine

2026-07-27 00:35:34 +08:00

rotary_emb.cu

perf: launch CUDA kernels on torch's current stream

2026-08-02 13:20:14 +08:00