gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
486 Commits 14 Branches 9 Tags
58c8889235e343066d48570e6b59c5383bbe7e6e
Commit Graph

486 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Philippe Tillet
a714b6b856 [PYTHON] re-activated auto-tuner configurations for triton.ops.matmul (#212) 2021-08-16 22:56:21 -07:00
Philippe Tillet
bb1eebb4b4 [CODEGEN] Fixed bug for visit_reduce1d with 64-bit data-types (#207) 2021-08-14 21:07:01 -07:00
Philippe Tillet
6e7593b446 added reset_to_zero in vector addition (#205) 2021-08-14 10:58:38 -07:00
Philippe Tillet
c45c2e9684 [DOCS] Added docs for cos/sin/sqrt (#204) 2021-08-14 10:34:07 -07:00
Philippe Tillet
c7a272cb91 [FRONTEND] Added default arguments for range (#203) 2021-08-14 10:11:18 -07:00
Philippe Tillet
b120d70a0a [CI] Moved from assert_allclose to assert_almost_equal (#200) 2021-08-12 12:00:30 -07:00
Philippe Tillet
70e28ff380 [DOCS] Minor modifications of the matmul tutorial (#199) 
Making the code more compact and fixing inconsistencies between text variable names and final python program.
 2021-08-11 18:59:15 -07:00
Philippe Tillet
398d4b4aeb [DOCS] softmax tutorial fixup (#198) 2021-08-11 17:35:00 -07:00
Philippe Tillet
83da7065da [DRIVER] Portability fixup (#195) 2021-08-07 18:53:11 -07:00
Philippe Tillet
298da78058 [CODEGEN/DRIVER] Tweaks for performance optimization (#193) 2021-08-07 16:41:44 -07:00
Nicholas Joseph
6cd1ec3955 [DOCS] Fix formatting mistakes (#192) 2021-08-06 12:58:43 -07:00
Nicholas Joseph
68f7eeba92 [DOCS] Improve matmul tutorial readability (#188) 2021-08-05 16:05:56 -07:00
Nicholas Joseph
4e6f667c2f [DOCS] Improve readability of 02-fused-softmax.py (#186) 2021-08-05 09:39:07 -07:00
Nicholas Joseph
23c71538fc [DOCS] Improve tutorial readability (#185) 2021-08-05 09:27:06 -07:00
Philippe Tillet
3cb77aa126 [README] Added "we're hiring!" with link to some of our blog posts (#180) 2021-08-02 16:46:26 -07:00
Xiangru Lian
9967e9d4b4 [DOCS] Fix fused softmax example script naive softmax implementation (#178) 2021-08-02 09:37:31 -07:00
Philippe Tillet
e8031fe61f [DRIVER] More robust support of unsupported CUDA version (#179) 2021-08-02 09:06:55 -07:00
milesial
b7cdf670c3 [DOCS] Fix related work (#172) 2021-08-01 11:06:37 -07:00
daadaada
c7060eadb2 [CODEGEN] Fix bug in auto-pipeline pass when a value depends on multiple phis (#164) 2021-07-31 23:40:36 -07:00
Philippe Tillet
c0bb895d9d [BUILD] More portable detection of terminfo (#173) 2021-07-31 17:09:49 -07:00
Philippe Tillet
a34c57402f [PYTHON] Improved error message for CPU (#167) 2021-07-30 09:47:27 -07:00
Ikko Ashimine
2293afece7 [README] GitHub format (#165) 
Github -> GitHub
 2021-07-30 09:47:08 -07:00
Philippe Tillet
cb5c280691 [DOCS] Added contributions section to README.md 2021-07-29 11:40:34 -07:00
Reid Draper
2322d6df2a [CI] Update ptillet to openai (#152) 2021-07-29 11:39:50 -07:00
Philippe Tillet
2f0f51be50 [DRIVER] No longer crashing when encountering CUDA version >11.4 2021-07-29 11:27:55 -07:00
Philippe Tillet
41ecd96300 [DOCS] minor grammar improvements 2021-07-28 14:18:31 -07:00
Avi Radinsky
d3851d8989 [DOCS] Typo fix (#151) 2021-07-28 12:07:12 -07:00
Philippe Tillet
4b9df06568 [CI] Bumped dev version to 1.0.1 and fixed permissions in documentation.yml (#149) 2021-07-28 04:35:14 -07:00
Philippe Tillet
046160b7f4 [README] Update Wheels badge URL v1.0 2021-07-28 02:04:41 -07:00
Philippe Tillet
acd5e44611 [GENERAL] Some minor improvements here and there to build systems and docs (#148) 2021-07-28 01:51:17 -07:00
Philippe Tillet
57c1fd3366 [BUILD] Now downloading LLVM from web if system does not have llvm-config-11 (#142) 2021-07-28 01:02:31 -07:00
Philippe Tillet
1365e96330 [CI] Fixup website build (#147) 2021-07-28 00:29:07 -07:00
Justin Jay Wang
8ddf909093 Add logo to README (#146) 2021-07-27 23:38:04 -07:00
Philippe Tillet
b736fdc740 [CI] More fixups (#145) 2021-07-27 22:14:51 -07:00
Philippe Tillet
1c48bd623e [CI] More bugfixes (#144) 2021-07-27 18:35:22 -07:00
Philippe Tillet
84521a5c82 [CI] Switch to Github Actions (#143) 2021-07-27 17:57:02 -07:00
Philippe Tillet
52d311f302 [CI] Updated build-website.yml (#141) 2021-07-27 12:38:49 -07:00
Philippe Tillet
bd70f10668 [CI] Added name to "Build Website" pipeline (#140) 2021-07-27 12:38:49 -07:00
Philippe Tillet
b253b77c71 [DOCS] Improved documentation and integration in CI (#139) 2021-07-27 12:38:49 -07:00
Philippe Tillet
76c6f24fb6 [CI] Made build-wheels compatible with system LLVM setup (#138) 
This speeds up wheelhouse build time by ~10x
 2021-07-27 12:38:49 -07:00
Philippe Tillet
8eb63bcb01 [CI] Various improvements to CI (#137) 
Add clean-up before CI runs. Now using static LLVM-11 libraries from system rather than recompilation. Still no run-time LLVM dependencies
 2021-07-27 12:38:49 -07:00
Philippe Tillet
298aead378 [FRONTEND] Fixed bugs in global symbols resolution of @triton.jit'd functions (#136) 2021-07-27 12:38:49 -07:00
Philippe Tillet
94ce6aa80f [DRIVER] Added support for CUDA 11.4 (#135) 2021-07-27 12:38:49 -07:00
Philippe Tillet
01276b5153 [FRONTEND] Added compilation flag to force use of .nc cache modifier (#134) 
in DRAM loads. /!\ USE CAREFULLY - THIS CAN BREAK CORRECTNESS IF MISUSED
/!\
 2021-07-27 12:38:49 -07:00
Philippe Tillet
2824345065 [LANGUAGE] Added cos/sin (#132) 2021-07-27 12:38:49 -07:00
Philippe Tillet
3169e4355c [PYTHON] Bugfix in trans_c for dsd in blocksparse matmul (#131) 2021-07-27 12:38:49 -07:00
Philippe Tillet
8cea583109 [IR] Preliminary support for BF16 (#129) 
This PR adds a BF16 data-type, along with FP32 <-> BF16 conversion instructions in the LLVM codegen. Other kinds of ops on bfloat16 are not yet supported.
 2021-07-27 12:38:49 -07:00
Philippe Tillet
9b4e2cae2d [PYTHON] Added isinstance to list of builtin symbols exposed (#128) 2021-07-27 12:38:49 -07:00
daadaada
0b05e06c0d cu_device::max_shared_memory() now returns max dynamic shared memory size (#127) 2021-07-27 12:38:49 -07:00
daadaada
d8d6b715c8 [CODEGEN] Performance improvement on A100 (#125) 
Improved codegen for the Ampere GPUs.

    * Make the layout pass recognize the multistage pipelined pattern.
    * Now the pipeline pass can automate the multistage pipelining transformation.
    * Remove extra barriers (from the prefetch pass & WAR) on Ampere.
    * Update the code generator (generator.cc) to make Triton generate n-buffered shared memory loads/stores.
 2021-07-27 12:38:49 -07:00