Amir Ghamarian 583b017321 Add decode-tuned unified attention kernel (4 warps, kBlockM=128) ...

Generalize the unified attention pipeline to support NumWarpGroups=1
(single warp group) with a serial K/V loop, in addition to the existing
NumWarpGroups=2 interleaved pipeline.

New decode kernel traits use 4 warps (sequence<4,1,1>) with kBlockM=128
and kBlockQ=16 for GQA-8, reducing Q tile padding waste from 31/32 to
15/16 for decode workloads (max_seqlen_q=1).

Host-side dispatch (is_decode_shape) routes low-token workloads to the
decode kernel automatically.

Benchmark results on d64 GQA-8 (via aiter):
- 64-seq decode:  2.2x slower -> 1.27x slower (1.73x speedup)
- 512-seq decode: 3.5x slower -> 1.6x slower  (2.2x speedup)
- 1-seq decode:   0.83x (CK wins) -> 0.81x    (no regression)
- Prefill:        unchanged (uses original 8-warp kernel)

Made-with: Cursor

2026-03-28 10:41:27 +00:00

..

ck

Implement batched gemm bias permute for RDNA4 (#3534)

2026-01-17 08:30:27 +01:00

ck_tile

Add decode-tuned unified attention kernel (4 warps, kBlockM=128)

2026-03-28 10:41:27 +00:00

rapidjson

Update pre-commit to fixed versions, run remod for ck_tile (#2895)

2025-10-16 15:29:17 -07:00