ViperEkura 41cd40363a perf: post-multiply attention scale in float instead of pre-scaling Q in bf16 ... - Replace bf16 pre-scale Q loading with direct 32-bit aligned bf16x2 reads - Apply scale in float32 after Q@K^T, before online softmax - Reduces causal max error from 2^-6 to 2^-8 with zero perf cost		2026-07-11 00:13:32 +08:00
..
attn_common.cuh	refactor: rename gqa_* to attn_*, split-KV for all decode paths	2026-07-10 23:35:14 +08:00
attn_decode.cu	refactor: rename gqa_* to attn_*, split-KV for all decode paths	2026-07-10 23:35:14 +08:00
attn_decode_split_kv.cuh	refactor: rename gqa_* to attn_*, split-KV for all decode paths	2026-07-10 23:35:14 +08:00
attn_decode_split_kv_mma.cuh	perf: post-multiply attention scale in float instead of pre-scaling Q in bf16	2026-07-11 00:13:32 +08:00
attn_entry_utils.cuh	refactor: rename gqa_* to attn_*, split-KV for all decode paths	2026-07-10 23:35:14 +08:00
attn_mma_utils.cuh	refactor: rename gqa_* to attn_*, split-KV for all decode paths	2026-07-10 23:35:14 +08:00
attn_prefill.cu	refactor: rename gqa_* to attn_*, split-KV for all decode paths	2026-07-10 23:35:14 +08:00
attn_prefill_split_q.cuh	refactor: rename gqa_* to attn_*, split-KV for all decode paths	2026-07-10 23:35:14 +08:00
attn_prefill_split_q_mma.cuh	perf: post-multiply attention scale in float instead of pre-scaling Q in bf16	2026-07-11 00:13:32 +08:00