Code

pr#12

Code

Merged

1proprogrammerchant merged 118 commits into1proprogrammerchant:main1proprogrammerchant/triton:mainfrom triton-lang:maintriton-lang/triton:mainCopy head branch name to clipboardSep 20, 2023

+7,959-5,542

Commits

Commits on Aug 20, 2023

• [TESTS] Fix tl.dot test on sm75 (#2140)

  Show description for a7b40a1
  alexander-zinoviev

  authored
  a7b40a1

• [FRONTEND] Disable IfExp on dynamic conditions (#2100)

  Show description for 584d5c2
  Jokeren

  authored
  584d5c2

Commits on Aug 21, 2023

• [BACKEND] Remove dead code related to old libhopper_helpers.bc (#2145)

  ThomasRaoux

  authored
  ad3e363

• [FRONTEND] Use inline asm for global timer and smid functions (#2143)

  Show description for 54ca7fc
  ThomasRaoux

  authored
  54ca7fc

• [OPTIMIZER] Fix Warp Specialized kernel launch failure (#2146)

  Show description for 7e5cd95
  bealwang

  authored
  7e5cd95

• [FRONTEND] name mangling fixup (#2148)

  ptillet

  authored
  ea84161

• [BACKEND] Optimize performance for f16 epilogue with TMA store (#2135)

  Show description for ec801ce
  ivanyinwz

  authored
  ec801ce

Commits on Aug 22, 2023

• [DOCS] update meetup/08-22-2023.md (#2149)

  kshama-msft

  authored
  0410652

• [FRONTEND] fix a typo (#2152)

  daemyung

  authored
  c4a9006

Commits on Aug 23, 2023

• [FRONTEND] Emit warning if the result of tl.advance is unused (#2155)

  Show description for 5fa1fa1
  Jokeren

  authored
  5fa1fa1

• [TUTORIALS] Skip running TMA tutorials on non-hopper architectures (#2153)

  Jokeren

  authored
  6a65c89

• [CI] Add back pre-commit to nvidia CI job (#2159)

  zahimoud

  authored
  5282ed8

• [TOOLS] Add support for autotuning AOT kernel (#2123)

  Show description for dad83f9
  BinFan

  authored
  dad83f9

• [DOCS] Add Intel XPU status update doc to Meetup Doc (#2160)

  EikanWang

  authored
  5d47054

• [DOCS] Add AMD GPU 08/23 status update doc to /docs/meetups (#2167)

  Show description for 9ad6327
  jayfurmanek

  authored
  9ad6327

• [BACKEND] Don't do dead code elimination on volatile load (#2165)

  ThomasRaoux

  authored
  3116933

• [DOCS] Update README.md (#2168)

  Show description for 2231403
  kshama-msft

  authored
  2231403

• [DOCS] update 08-22-2023.md (#2164)

  Show description for a1c5ef3
  kshama-msft

  authored
  a1c5ef3

Commits on Aug 24, 2023

• [DOCS] fix typo in readme

  ptillet

  committed
  b43c28f

• [FRONTEND]To use proper default num_warps and num_stages based on the device backend in JITFucntion (#2130)

  Show description for 6cb6718
  

  chengjunlu

  and

  EikanWang

  authored
  6cb6718

• [DOCS] update README.md (#2174)

  Show description for 387c8d9
  kshama-msft

  authored
  387c8d9

• [DOCS] Fixing docs (#2175)

  zahimoud

  authored
  120ce0a

Commits on Aug 25, 2023

• [Optimizer][Hopper] change mmaV3InstrN for flash attention (#2169)

  Show description for 22a2fe3
  ben-zhang-609

  authored
  22a2fe3

• [FRONTEND] drop the GIL around more CUDA ops (#2173)

  hauntsaninja

  authored
  7083dae

• [FRONTEND] handle errors from launch_enter_hook (#2178)

  gdb

  authored
  64d8df4

• [BACKEND] Fix BF16 dot operand type mismatch (#2162)

  Show description for f6cdcf1
  Jokeren

  authored
  f6cdcf1

• [FRONTEND] Fix benchmark plotting (#2177)

  EPronovost

  authored
  56fee37

• [FRONTEND] fix for undefined dtypes in jit during loading defaults (#2114)

  Show description for ebfe0ff
  

  Moerafaat

  and

  Jokeren

  authored
  ebfe0ff

• [DOCS] Fixing docs by ignoring examples 09/10 in sphinx-gallery (#2189)

  zahimoud

  authored
  4ea94a8

Commits on Aug 26, 2023

• [FRONTEND] fix handling of do_not_specialize with interior constantexprs (#2188)

  gdb

  authored
  ab3e8b0

• [OPTIMIZER] Add folder for MakeRangeOp (#2187)

  Show description for fa03b92
  

  peterbell10

  and

  ptillet

  authored
  fa03b92

• [BACKEND] Fix fma mixed-precision (#2184)

  Show description for 6e4932c
  Jokeren

  authored
  6e4932c

• [FRONTEND] drop GIL for launch, and set value=false upon pointer error (#2185)

  gdb

  authored
  a9b8c8c

• [FRONTEND] remove dead libhopper_helpers.bc file (#2190)

  ptillet

  authored
  5f448b2

Commits on Aug 28, 2023

• [TESTS][HOPPER] Prune hopper tests to speedup CI (#2193)

  Show description for 1465b57
  

  goostavz

  and

  Goostav Zhu

  authored
  1465b57

Commits on Aug 29, 2023

• [BACKEND] Refactor RemoveLayoutConversion pass (#2181)

  Show description for d4644d6
  ThomasRaoux

  authored
  d4644d6

• [BACKEND] Fix crash when propagating layout and slice axis doesn't ma… (#2205)

  Show description for 17d633a
  ThomasRaoux

  authored
  17d633a

• [README] Added hiring plug on top of README.md

  ptillet

  committed
  c8536d6

• [README] Removed old hiring plug

  ptillet

  committed
  9f3b631

Commits on Aug 30, 2023

• [BACKEND] Extend hoisting of convert op above ext ops (#2206)

  Show description for 2ff88c1
  ThomasRaoux

  authored
  2ff88c1

• [RUNTIME] Fix cache dir (#2196)

  Show description for 9af76e7
  

  jon-chuang

  and

  Jokeren

  authored
  9af76e7

• [BACKEND] Handle more cases of folding convert into reduce op (#2209)

  Show description for 3175ee4
  ThomasRaoux

  authored
  3175ee4

• [BACKEND] Lift restriction for float8e4b15 to only support row-col layout (#2212)

  ptillet

  authored
  ec51552

Commits on Aug 31, 2023

• [BACKEND] Added float8e4b15 codegen for SM < 80 (#2216)

  ptillet

  authored
  c4b27d0

• [BACKEND] Add support for propagating convert op through while loops (#2213)

  Show description for fff97d8
  ThomasRaoux

  authored
  fff97d8

• [DOCS] Fixing docs by using sphinx-build instead of multiversion (#2217)

  zahimoud

  authored
  cf31f4d

• [DOCS] Fixing dependency (#2219)

  zahimoud

  authored
  f922ecb

• [DOCS] Fixing docs (#2221)

  zahimoud

  authored
  bffb76e

• [BACKEND] cleaner V100 tensor core packing (#2222)

  ptillet

  authored
  36fc54b

• [ROCM] Core Functionality for AMD (#1983)

  Show description for c6d33dc
  

  micmelesse

  and

  ptillet

  authored
  c6d33dc

Commits on Sep 1, 2023

• [BACKEND] More minor backend fixups (#2223)

  Show description for 9b8c48f
  ptillet

  authored
  9b8c48f

• [FRONTEND/OPS] wap stride_vn and stride_vk in flash attention (#2208)

  Show description for 1367f3a
  EPronovost

  authored
  1367f3a

• [BACKEND] Fix several bugs related to multiCTA (#2170)

  Show description for 691c963
  qliu93

  authored
  691c963

• [FRONTEND] Fix GIL handling in error conditions (#2225)

  Show description for a4df60e
  hauntsaninja

  authored
  a4df60e

• [BACKEND] Refactoring NVGPUToLLVMPass (#2158)

  zahimoud

  authored
  acbf716

Commits on Sep 2, 2023

• Coalesce pass - group values with same shape/order but different element type (#2199)

  Show description for 13189bf
  davidberard98

  authored
  13189bf

Commits on Sep 4, 2023

• Fix predicate for store tiled op (#2215)

  Show description for a539836
  ivanyinwz

  authored
  a539836

• [FRONTEND] Fix specialization on triton integer types (#2236)

  Show description for 9e9fbe0
  Jokeren

  authored
  9e9fbe0

Commits on Sep 5, 2023

• [OPS] Remove unnecessary perf bug workaround (#2240)

  Show description for 99f8f91
  jon-chuang

  authored
  99f8f91

• [FRONTEND] clean build directly when executing python setup.py clean (#2238)

  Show description for e721911
  EikanWang

  authored
  e721911

Commits on Sep 6, 2023

• [BACKEND][NFC] Simplify coalescing pass (#2230)

  Show description for 60643f2
  ThomasRaoux

  authored
  60643f2

• [BACKEND] Rebased Intel GPU Backend to be comptiable with latest Triton version (#2245)

  EikanWang

  authored
  92e2c32

• [DOCS] Add MLIR Autogenerated Docs to Sphinx Docs (#2234)

  Show description for 36859ae
  jon-chuang

  authored
  36859ae

• [CI] Fix submodule issue (#2253)

  Show description for 3dec616
  zahimoud

  authored
  3dec616

• [CLEANUP] Delete binaries that went in by mistake (#2256)

  zahimoud

  authored
  f21b36c

Commits on Sep 7, 2023

• [OPTIMIZER] async launch dots for hopper warp-specialized kernel (#2251)

  bealwang

  authored
  7aae350

• Revert unintentional change (#2257)

  Show description for 7d01c18
  IzzyPutterman

  authored
  7d01c18

Commits on Sep 8, 2023

• [BACKEND] Remove dead code (#2263)

  ThomasRaoux

  authored
  52aa663

• [RUNTIME] Get the correct end idx for regular arguments of GPU kernels (#2262)

  Show description for 10f59d8
  Jokeren

  authored
  10f59d8

Commits on Sep 10, 2023

• [OPTIMIZER] improvements to layout conversion removal (#2268)

  Show description for 3747843
  ptillet

  authored
  3747843

Commits on Sep 11, 2023

• [Backend] Make ConvertTritonGPUToLLVMPass's tmaMetadata a member (#2271)

  Show description for f6828e1
  chsigg

  authored
  f6828e1

• [BACKEND] Make sure getAxisBlockStride does not return 0 (#2273)

  Show description for 28d4c3b
  zhanglx13

  authored
  28d4c3b

• [TESTS] replace deprecated torch.testing.assert_allclose (#2250)

  Show description for 5231d57
  jon-chuang

  authored
  5231d57

• [TESTS] Enhance benchmark flexibility (#2239)

  Show description for ec4a968
  

  daemyung

  and

  Jokeren

  authored
  ec4a968

• Remove wrong dependency between TritonGPU and NVGPU dialect (#2276)

  ThomasRaoux

  authored
  a9db6b9

Commits on Sep 12, 2023

• [Build] Fix very minor compilation problems (#2277)

  Show description for 8da27c1
  shintaro-iwasaki

  authored
  8da27c1

• [NFC] Remove hard-coded warpSize=32 in scanOp lowering (#2272)

  Show description for a5e4836
  zhanglx13

  authored
  a5e4836

• [FRONTEND] allow mixed precision FP8 matmul on pre-H100 hardware (#2281)

  ptillet

  authored
  bf4f937

• [FRONTEND] Fix expand_dims and tl.full to handle scalar tensors (#2275)

  Show description for ab9da3b
  peterbell10

  authored
  ab9da3b

• [FRONTEND] Improve grid calculation for persistent kernels to hoist pe… (#2283)

  Show description for fc5d7e6
  jsh-20

  authored
  fc5d7e6

• [BACKEND] Unify slow/fast reduce codegen (#2220)

  zahimoud

  authored
  a47f1f5

• [FRONTEND] Add PyTorch fp8 dtypes to Triton (#2279)

  Show description for 37f1249
  ipiszy

  authored
  37f1249

• [BACKEND] Remove dependency between NVGPU and TritonNvidiaGPU (#2282)

  ThomasRaoux

  authored
  994f7e4

Commits on Sep 13, 2023

• [BACKEND] Convert layout illegal mem access fix (#2287)

  zahimoud

  authored
  e95e1f1

• [BACKEND] Add LLVM pre-processing pass to break struct types (#2285)

  Show description for d3956a2
  ThomasRaoux

  authored
  d3956a2

• [NFC] Create explicit conversion pattern for ExternElementwiseOp in TT->TTGPU pass (#2284)

  Show description for 896ee61
  sergeykozub

  authored
  896ee61

• [BACKEND] Optimization to sink broadcast ops (#2274)

  Show description for cf7f8c5
  ThomasRaoux

  authored
  cf7f8c5

• [FRONTEND] Override prototype (#2214)

  Show description for b63e8f8
  ThomasRaoux

  authored
  b63e8f8

• [DOCS] add missing docs (#2154)

  khushi-411

  authored
  c61d772

• [OPTIMIZER] Fix Shared layout in OptimizeDotOperands pass to generate correct swizzling code (#2180)

  Show description for 38a2ecd
  

  BinFan

  and

  ptillet

  authored
  38a2ecd

• [BACKEND] Fixing assert in shared encoding swizzling addresses calculation (#2292)

  zahimoud

  authored
  a301502

• [FRONTEND] Added SASS to asm dict (#2280)

  zahimoud

  authored
  36087a1

Commits on Sep 14, 2023

• [FRONTEND] Accommodate new triton IR format (#2294)

  Show description for 08c1658
  Jokeren

  authored
  08c1658

Commits on Sep 15, 2023

• [BUILD] Fix few dependencies and layering issues to make lld work (#2307)

  Show description for 976aabd
  ThomasRaoux

  authored
  976aabd

• [DOCS] update README.md (#2311)

  Show description for ac1c216
  kshama-msft

  authored
  ac1c216

• [FRONTEND] Add sass to asm dict with lazy evaluation (#2309)

  zahimoud

  authored
  db5c793

Commits on Sep 16, 2023

• [CI] update integration-tests.yml (#2310)

  micmelesse

  authored
  78a0b5d

• [FRONTEND][BACKEND] Add flag to control accumulation for fp8 (#2300)

  Show description for 31b0c52
  ThomasRaoux

  authored
  31b0c52

• [BACKEND] Move struct optimization down the LLVM pipeline (#2312)

  Show description for bb949d1
  ThomasRaoux

  authored
  bb949d1

Commits on Sep 17, 2023

• Add cuobjdump and nvsisasm to gitignore. (#2319)

  Show description for 41584c7
  jlebar

  authored
  41584c7

• [DOCS] Add build instrs for running in a virtualenv. (#2320)

  Show description for 0015611
  jlebar

  authored
  0015611

• Revert "Update integration-tests.yml" (#2323)

  Show description for c98671c
  ptillet

  authored
  c98671c

• [BUILD] use ninja (#2318)

  jlebar

  authored
  073aa16

• [FRONTEND] Explicitly forbid dot(.., out_dtype=bfloat16) (#2308)

  Show description for 4f2d995
  jon-chuang

  authored
  4f2d995

• [FRONTEND] fix xpu stages logic (#2305)

  Stonepia

  authored
  68e1bd1

• [FRONTEND] fix matmul int8 overflow issue (#2297)

  Show description for 2b06600
  lightb0x

  authored
  2b06600

• [FRONTEND] interpreter rewrite (#2321)

  Show description for e686b4d
  ptillet

  authored
  e686b4d

• [RUNTIME][INTERPRETER] now also override __str__ method for tensors (#2325)

  ptillet

  authored
  894fa9e

Commits on Sep 18, 2023

• Integration fixes for llvm/llvm-project#66512 (#2328)

  Show description for a9ae988
  matthias-springer

  authored
  a9ae988

• [BACKEND] Fix scan issues on repetitive warps and improve perf when there's a single warp on the axis (#2330)

  Show description for 307b5ca
  Jokeren

  authored
  307b5ca

• [DOCS] improved fused attention tutorial (bwd pass) (#2332)

  ptillet

  authored
  73dae77

• [BACKEND] Relax patterns to move sink broadcast and hoist convert (#2331)

  Show description for 3a848e2
  ThomasRaoux

  authored
  3a848e2

Commits on Sep 19, 2023

• [FRONTEND] Fix Python error handling in launch (#2334)

  Show description for 8e75e39
  hauntsaninja

  authored
  8e75e39

• Integration fixes for llvm/llvm-project#66754 (#2338)

  Show description for ae07b7b
  matthias-springer

  authored
  ae07b7b

• Switch pre-commit clang-format to v16.0.6. (#2342)

  Show description for 5491707
  jlebar

  authored
  5491707

Commits on Sep 20, 2023

• Add instructions for building with custom LLVM (#2344)

  Show description for 3631829
  jlebar

  authored
  3631829

• [TESTS] fix flash attention (#2086)

  Show description for e5eda09
  

  donproc

  and

  dongdongl

  authored
  e5eda09

• [BACKEND] Handle repetitive threads in scan op when the tensor dim is small (#2345)

  Show description for ed5a530
  Jokeren

  authored
  ed5a530