Commits

Branch selector

 main

User selector

All users

Datepicker

All time

Commit History

Commits on Apr 20, 2025

• [AMD] Fix packed f32 to fp16 cast for single value (#6545)

  Show description for c5fed8e
  antiagainst

  authored
  c5fed8e

  Copy full SHA for c5fed8e

• [TEST] Check ttng.tc_gen5_mma only for CUDA (#6543)

  Show description for a8e5788
  antiagainst

  authored
  a8e5788

  Copy full SHA for a8e5788

• [AMD] Enable larger kBase mfma instructions for smaller kWidth (#6540)

  Show description for bceed1f
  zhanglx13

  authored
  bceed1f

  Copy full SHA for bceed1f

Commits on Apr 19, 2025

• [BENCH] move code around; added tests; improved expert-parallelism simulation (#6538)

  ptillet

  authored
  1294776

  Copy full SHA for 1294776

• [AMD] [gfx950] [FA-f16] use single rated mfma for 2nd dot (#6532)

  Show description for d066e15
  zhanglx13

  authored
  d066e15

  Copy full SHA for d066e15

• [BENCH][AMD] Enable gfx942 for bench_mlp.py (#6513)

  Show description for d1b0801
  

  jtang10

  and

  Jingning Tang

  authored
  d1b0801

  Copy full SHA for d1b0801

Commits on Apr 18, 2025

• [AMD] Update pingpong scheduler to ignore loads/stores in prologues and epilogues for persistent GEMMs (#6198)

  Show description for 1780734
  

  njriasan

  and

  Nick Riasanovsky

  authored
  1780734

  Copy full SHA for 1780734

• [AMD] Used packed f32->(b)f16 conversion for gfx950 (#6511)

  Show description for 7ba141e
  yiqian1

  authored
  7ba141e

  Copy full SHA for 7ba141e

Commits on Apr 17, 2025

• [DIALECT] Improve granularity of Triton(NVIDIA)GPU dialect ops (#6517)

  Show description for 6e3b1ff
  Mogball

  authored
  6e3b1ff

  Copy full SHA for 6e3b1ff

• [AMD] fix addptr canonicalization for i16 offsets (#6526)

  Show description for e341f7f
  makslevental

  authored
  e341f7f

  Copy full SHA for e341f7f

• [DIALECT] Improve granularity of Triton dialect op side effects (#6516)

  Show description for c6be5d4
  Mogball

  authored
  c6be5d4

  Copy full SHA for c6be5d4

• Remove import-time dependency on PyTorch in nvidia driver.py (#6525)

  Show description for 2118b80
  anmyachev

  authored
  2118b80

  Copy full SHA for 2118b80

• [DIALECT] Clean up memory effect definitions (NFC) (#6515)

  Show description for 56f8556
  Mogball

  authored
  56f8556

  Copy full SHA for 56f8556

• [RELAND][Blackwell] Optimize MMA warp specialization to allow multiple consumers of MMAv5 result (#6514)

  Show description for 65f646c
  Mogball

  authored
  65f646c

  Copy full SHA for 65f646c

• [INTERPRETER] Fix the semantic of `tl.reduce` and `tl.scan` handling tuples (#6510)

  Show description for 9332fa1
  Jokeren

  authored
  9332fa1

  Copy full SHA for 9332fa1

• [BENCH] Improvements to topk routing (#6521)

  Show description for 5d0fc1e
  

  3 peopleauthored
  5d0fc1e

  Copy full SHA for 5d0fc1e

• Revert #6487, #6485, #6484, #6476 (#6512)

  Show description for 0c16f82
  Mogball

  authored
  0c16f82

  Copy full SHA for 0c16f82

Commits on Apr 16, 2025

• [PIPELINER] Use AttrHelper (mostly NFC) (#6437)

  Show description for 3a91f81
  sjw36

  authored
  3a91f81

  Copy full SHA for 3a91f81

• [BENCH] fast top-k routing and matmul metadata preparation (#6507)

  Show description for beacdb0
  ptillet

  authored
  beacdb0

  Copy full SHA for beacdb0

• [FRONTEND] Fix argument annotation after host side TensorDescriptor (#6509)

  Show description for 5fdff50
  peterbell10

  authored
  5fdff50

  Copy full SHA for 5fdff50

• [FRONTEND] Cleanup nv_tma_desc_type (#6508)

  Show description for de1aad2
  peterbell10

  authored
  de1aad2

  Copy full SHA for de1aad2

• [NFC] Fix signed/unsigned comparison. (#6494)

  Show description for 863bf6b
  BabakkGraphcore

  authored
  863bf6b

  Copy full SHA for 863bf6b

Commits on Apr 15, 2025

• [Blackwell] Optimize MMA warp specialization to allow multiple consumers of MMAv5 result (#6487)

  Show description for 96fc40d
  Mogball

  authored
  96fc40d

  Copy full SHA for 96fc40d

• [LINEINFO] Relocate IR when compile from IR source (#6502)

  Jokeren

  authored
  e8b88e5

  Copy full SHA for e8b88e5

• [CI] Switch to use gfx942 runner set (#6470)

  Show description for 298306f
  antiagainst

  authored
  298306f

  Copy full SHA for 298306f

• [python][compiler] Memoize device max shared memory per device (#6503)

  Show description for 5ee013c
  danzimm

  authored
  5ee013c

  Copy full SHA for 5ee013c

• [NVWS] Add initial version of Aref Lowering (#6410)

  Show description for 18ef770
  mbrookhart

  authored
  18ef770

  Copy full SHA for 18ef770

• [Backends] Use normal import method instead of file import (#6500)

  Show description for a54399e
  peterbell10

  authored
  a54399e

  Copy full SHA for a54399e

• [DEV] Fix .gitignore (#6501)

  Show description for 08a7ca7
  peterbell10

  authored
  08a7ca7

  Copy full SHA for 08a7ca7

• Restore faster tl.sort implementation (#6492)

  Show description for b65304e
  apgoucher

  authored
  b65304e

  Copy full SHA for b65304e

• Allow tl.make_tensor_descriptor to produce 1D tensor descriptors (#6498)

  Show description for 86dd442
  adamjsanders277

  authored
  86dd442

  Copy full SHA for 86dd442

• [AMD] Clamp vmcnt instead of erroring out in AsyncWait lowering (#6493)

  Show description for 19a16ef
  AlexAUT

  authored
  19a16ef

  Copy full SHA for 19a16ef

• [BC-Breaking] Kill experimental descriptor APIs (#6488)

  Show description for 053f98b
  peterbell10

  authored
  053f98b

  Copy full SHA for 053f98b

• Revert recent changes to `tl.sort` (#6489)

  Show description for 7efc694
  peterbell10

  authored
  7efc694

  Copy full SHA for 7efc694

Commits on Apr 14, 2025

• [python][backends] Setup sys.modules with any loaded backend (#6468)

  Show description for 9487527
  danzimm

  authored
  9487527

  Copy full SHA for 9487527

Pagination

Previous

Next