Repository navigation

[ASM] [MLA v4 nm] Swap gfx950 asm decode kernel - #5195

Merged

amd-ruitang3 merged 6 commits into

mainfrom

liyjiang/mla_v4_nm_perf27

Sep 24, 2026

Merged

[ASM] [MLA v4 nm] Swap gfx950 asm decode kernel#5195
amd-ruitang3 merged 6 commits into
mainfrom
liyjiang/mla_v4_nm_perf27

Commits on Sep 22, 2026
- [MLA v4 nm] Swap gfx950 asm decode kernel
  
  liyjiang
  and
  cursoragent
  committed
  2b55568
- [MLA v4 nm] Fix Q scale over-read past the packed padding
  
  liyjiang
  and
  cursoragent
  committed
  3a58541
- [MLA v4 nm] Add a packed-row padding regression test
  
  liyjiang
  and
  cursoragent
  committed
  79e4b94
- [MLA v4 nm] Swap gfx950 asm decode kernel, add repeatability, byte-range and kv-tail regression tests
  
  liyjiang
  and
  cursoragent
  committed
  41ecb13
- [MLA v4 nm] Hoist the poison helper out of its loop (ruff B023)
  
  liyjiang
  and
  cursoragent
  committed
  f340f57
- [MLA v4 nm] Swap .co, minor update for gqa=16/32
  
  liyjiang
  and
  cursoragent
  committed
  0a3ed7f