gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
1,419 Commits 14 Branches 9 Tags
b01d514815eca4929e2652e849149cdec0f117d0
Commit Graph

1419 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Philippe Tillet
b01d514815 [GH-PAGES] Updated website 2021-08-02 00:13:11 +00:00
Philippe Tillet
462b4d9bc9 [GH-PAGES] Updated website 2021-08-01 00:16:31 +00:00
Philippe Tillet
ec78e18faf [GH-PAGES] Updated website 2021-07-31 06:25:55 +00:00
Philippe Tillet
be5bc8a654 [GH-PAGES] Updated website 2021-07-31 05:49:14 +00:00
Philippe Tillet
df9dd891c9 [GH-PAGES] Updated website 2021-07-31 05:42:19 +00:00
Philippe Tillet
551f68c481 [GH-PAGES] Updated website 2021-07-31 05:36:03 +00:00
Philippe Tillet
9e4ac7e794 [GH-PAGES] Updated website 2021-07-31 05:27:59 +00:00
Philippe Tillet
a86020efbc [GH-PAGES] Updated website 2021-07-28 11:39:54 +00:00
Philippe Tillet
040ec5e252 [GH-PAGES] Updated website 2021-07-28 10:15:45 +00:00
Philippe Tillet
cfaad88904 [GH-PAGES] Updated website 2021-07-28 09:06:14 +00:00
Philippe Tillet
282fc17966 [GH-PAGES] Updated website 2021-07-28 07:36:00 +00:00
Philippe Tillet
3489e2e0d7 [GH-PAGES] Updated website 2021-07-28 06:19:22 +00:00
Philippe Tillet
3d015bafaf [GH-PAGES] Updated website 2021-07-28 06:03:23 +00:00
Philippe Tillet
1da7ca38c7 [GH-PAGES] Updated website 2021-07-24 06:04:37 +00:00
Philippe Tillet
496b786cfc [GH-PAGES] Updated website 2021-07-23 07:43:04 +00:00
Philippe Tillet
c0a449409d [GH-PAGES] Updated website 2021-07-23 05:18:00 +00:00
Philippe Tillet
98967714bd [GH-PAGES] Updated website 2021-07-23 04:39:46 +00:00
Philippe Tillet
e581b452bd [GH-PAGES] Updated website 2021-04-23 16:42:55 -04:00
Philippe Tillet
817a174c88 [GH-PAGES] Updated website 2021-04-21 16:45:37 -04:00
Philippe Tillet
4f471eaf18 [GH-PAGES] Updated website 2021-04-21 02:08:14 -04:00
Philippe Tillet
4d0ba90bc7 [GH-PAGES] Updated website 2021-04-21 01:58:48 -04:00
Philippe Tillet
92126eb098 [GH-PAGES] Updated website 2021-04-21 01:40:29 -04:00
Philippe Tillet
5cefc81fce [GH-PAGES] Updated website 2021-03-29 11:59:18 -04:00
Philippe Tillet
64141f0fca [GH-PAGES] Updated website 2021-03-23 17:10:07 -04:00
Philippe Tillet
3db1455cda [GH-PAGES] Updated website 2021-03-19 16:19:50 -04:00
Philippe Tillet
4d1c282eb2 Update CNAME 2021-03-15 21:31:25 -04:00
Philippe Tillet
746b15ee0a [GH-PAGES] Updated website 2021-03-15 13:58:20 -04:00
Philippe Tillet
b4495e0ddc [GH-PAGES] Updated website 2021-03-11 12:09:07 -05:00
Philippe Tillet
8316c4bbb1 [GH-PAGES] Updated website 2021-03-11 11:58:42 -05:00
Philippe Tillet
7a7ed5da3b [GH-PAGES] Added small script to automatically update website (assuming 
it's been compiled in master)
 2021-03-06 22:09:04 -05:00
Philippe Tillet
32aaf8b469 [GH-PAGES] Updated website 2021-03-06 22:06:32 -05:00
Philippe Tillet
6f789b29ab [GH-PAGES] Updated website 2021-03-06 17:35:11 -05:00
Philippe Tillet
9d95317e80 [GH-PAGES] Updated documentation 2021-03-06 17:29:46 -05:00
Philippe Tillet
449387489c switching to py file for tutorials 2021-03-06 13:17:43 -05:00
Philippe Tillet
61cf3ddd96 [GH-PAGES] Added missing files 2021-03-06 03:09:03 -05:00
Philippe Tillet
8b18c19875 [GH-PAGES] Updated documentation to include ipynb from tutorials 2021-03-06 03:02:48 -05:00
Philippe Tillet
7771e9e27d Create CNAME 2021-03-05 22:58:05 -05:00
Philippe Tillet
1989726fa8 [GH-PAGES] Added website 2021-03-05 22:49:39 -05:00
Philippe Tillet
a4dc56ee57 [GH-PAGES] Added .nojekyll file 2021-03-05 22:48:33 -05:00
Philippe Tillet
28d897c7d4 [DOCS] First build to gh-pages 2021-03-05 22:38:55 -05:00
Philippe Tillet
90f953931e [DOCS] Updated and improved docs (#73) 2021-03-05 22:33:50 -05:00
Philippe Tillet
2b9b284026 [PYTHON] Deleted 01-vector-add.py: it is an unnecessary duplicate of 
01-vector-add.ipynb
 2021-03-04 02:06:57 -05:00
Philippe Tillet
a7437e14c5 [RUNTIME] Added auto-alignment mechanism (#71) 
This PR adds an automatic memory alignment mechanism in the Triton runtime. Specifically, the JIT compiler detects the alignment (in bytes) of each pointer argument as well as the largest power of two divisor (between 1 and 16) of each integer argument. Proper .aligned and .multipleof attributes are then added to the Triton-IR on-the-fly for all auto-tunable kernels. There is a cache that remembers all the kernels compiled for each possible configuration.

This PR also includes substantial cleaning of the Python API. This adds 2-3us overhead, mostly due to accessing integer #defines from the auto-tuned compilation options. The previous solution was slightly faster but hacky and potentially unsafe, so this is preferred for now.
 2021-03-04 01:51:11 -05:00
Philippe Tillet
4e6fe6329f [PYTHON] bugfix in bench_cross_entropy 2021-02-26 02:37:46 -05:00
Philippe Tillet
8e8e65023b [DOCS] Added Python 02-fused-softmax.ipynb tutorial 2021-02-25 14:49:47 -05:00
Philippe Tillet
b0f37346b0 [PYTHON] Bugfix on FP32 blocksparse matmul 2021-02-24 14:44:23 -05:00
Philippe Tillet
db6bf71564 [CODEGEN] Bugfixes with FP32 async copy 2021-02-24 14:44:09 -05:00
Philippe Tillet
bfe92a5d91 [CODEGEN] Now initializing cp.async to zero when predicate is false 
WARNING: case for non-zero initialization is still not handled. Will
require manual copy to shared
 2021-02-22 17:14:33 -05:00
Philippe Tillet
15f8e8c3b7 [CODEGEN] Major performance improvements on A100 (#70) 
Improved handling of asynchronous copy, scheduling and synchronization for A100. Now achieving CUTLASS-like performance on large square dense matrix multiplication tasks
 2021-02-21 18:19:39 -05:00
Jared Kaplan
b10e9b89e9 [PYTHON] Add Blocksparse Attention Fwd/Bwd Test (#69) 
Also includes small bugfix for block-sparse softmax
 2021-02-19 17:46:05 -05:00