Repository navigation

[Common] Look up table based activation calculation in MXFP8 quantization fused kernels - #3761

Draft

tdophung wants to merge 62 commits into

NVIDIA:mainfrom

tdophung:teddy/bf16-activation-lut

Draft

[Common] Look up table based activation calculation in MXFP8 quantization fused kernels#3761
tdophung wants to merge 62 commits into
NVIDIA:mainfrom
tdophung:teddy/bf16-activation-lut

Commits on Oct 7, 2026
- [Common] More CuTeDSL MXFP8 special kernels
  kainzhong
  committed
  7c3abf6
- Merge branch 'main' into more_cutedsl_mxfp8_specialized
  kainzhong
  committed
  6e854d5
Commits on Oct 8, 2026
- wip
  kainzhong
  committed
  1be6447
- wip
  kainzhong
  committed
  1995fea
Commits on Oct 9, 2026
- refactor
  kainzhong
  committed
  3e7a653
- refactor
  kainzhong
  committed
  79c5d53
- refactor
  kainzhong
  committed
  c51a3b8
- [pre-commit.ci] auto fixes from pre-commit.com hooks
  pre-commit-ci[bot]
  committed
  f14e433
- fix
  kainzhong
  committed
  5f9ece2
- disable invariant temporarily
  kainzhong
  committed
  870b23c
Commits on Oct 10, 2026
- fix
  kainzhong
  committed
  4ef3404
Commits on Oct 12, 2026
- refactor
  kainzhong
  committed
  ef9877d
- [pre-commit.ci] auto fixes from pre-commit.com hooks
  pre-commit-ci[bot]
  committed
  d6bb258
- fix
  kainzhong
  committed
  fffc874
- fix
  kainzhong
  committed
  2fd0aca
- also support swizzled scale for g2r bidim
  kainzhong
  committed
  8afd230
- lint
  kainzhong
  committed
  87ab24e
- add tests
  kainzhong
  committed
  cae94f5
- fix
  kainzhong
  committed
  f67b1e5
- nit
  kainzhong
  committed
  177647e
Commits on Oct 14, 2026
- refactor
  kainzhong
  committed
  8f4c54d
- refactor
  kainzhong
  committed
  fddd076
- more refactor
  kainzhong
  committed
  3a8a882
- g2r bidim can also support colwise only
  kainzhong
  committed
  30c92ab
Commits on Oct 15, 2026
- make g2r support activations
  kainzhong
  committed
  5c3465a
- fix
  kainzhong
  committed
  b1db277
- fix
  kainzhong
  committed
  dcd8b08
- +0 and -0 should both legal
  kainzhong
  committed
  86233c8
- nit
  kainzhong
  committed
  dba8dbf
Commits on Oct 17, 2026
- fix
  kainzhong
  committed
  287505f
Commits on Oct 20, 2026
- fix
  kainzhong
  committed
  2c6f6ec
Commits on Oct 21, 2026
- close fp32 perf gap
  kainzhong
  committed
  53bc647
- can always use static tile num
  kainzhong
  committed
  bfdb240
Commits on Oct 26, 2026
Commits on Oct 27, 2026
Commits on Oct 28, 2026
Commits on Nov 3, 2026
- [Common] Add CuTeDSL gated MXFP8 kernels, including a register-resident one
  
  tdophung
  and
  kainzhong
  committed
  6d7e361

Commits

Commits on Oct 7, 2026

Commits on Oct 8, 2026

Commits on Oct 9, 2026

Commits on Oct 10, 2026

Commits on Oct 12, 2026

Commits on Oct 14, 2026

Commits on Oct 15, 2026

Commits on Oct 17, 2026

Commits on Oct 20, 2026

Commits on Oct 21, 2026

Commits on Oct 26, 2026

Commits on Oct 27, 2026

Commits on Oct 28, 2026

Commits on Nov 3, 2026