juuso-oskari a4d3ff34fb CK-UA: widen FP8 K/V async loads to dwordx4 (per-WG load-thread count) ...

The K/V async-load width selector (GetKVAlignmentBytes via GetAlignmentK/V)
computed its dwordx4 budget from the full block (kBlockSize=512 threads), so
the 4 KB FP8 prefill tile never tiled cleanly and fell back to dword. With the
FA4 per-warp-group decoupling a single 4-warp group (256 thr) fills the tile by
itself -> 4 KB / 256 = exactly 16 B/thr = dwordx4.

Thread the load-thread count into GetAlignmentK/V as a NumWarps template param
(default = shape NumWarps, so all sizing/paged/decode instantiations are
byte-identical). Only the load-path callers (MakeK/VDramTileDistribution, the
LDS store/load descriptors, and the kAlignmentK/V DRAM-view vector) pass the
decoupled GetK/VLoadNumWarps count to unlock the wide load.

Effect: global_load_lds_dword 36->9 dwordx4 and buffer_load_dword 36->9 dwordx4
(both runtime branches); VGPR 181->173; LDS/SGPR unchanged. Accuracy PASS 0%
(non-causal + causal). Latency-neutral on sq8192 (kernel is memory-latency
bound, not load-issue bound) but a strictly-better instruction/VGPR footprint.

Co-authored-by: Cursor <cursoragent@cursor.com>

2026-06-09 16:40:21 +00:00

..

ck

[rocm-libraries] ROCm/rocm-libraries#6022 (commit 54b284a)

2026-03-31 15:19:43 +00:00

ck_tile

CK-UA: widen FP8 K/V async loads to dwordx4 (per-WG load-thread count)

2026-06-09 16:40:21 +00:00

rapidjson

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

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