gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
1,370 Commits 14 Branches 9 Tags
b10e9b89e9c3a1dbb80a930ee503e82c4ef30a5b
Commit Graph

164 Commits

Author SHA1 Message Date
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
Tom B Brown
cf5a1ee79e [PYTHON] Avoid dangerous global variables in kwarg default values (#68) 2021-02-18 17:56:54 -05:00
Philippe Tillet
f8846d95ff [PYTHON][OPS] Added compiler hints to improve performance of 
cross-entropy
 2021-02-10 16:47:50 -05:00
Philippe Tillet
0125ab1740 [CI] Added benchmarking to CI script (#65) 2021-02-10 14:01:47 -05:00
Philippe Tillet
53fd9631ef [PYTHON] Added automated benchmark script (#63) 
This adds a bench functionality to the setup.py that can be used to run the benchmark suite and generates a bunch of csv files (and optionally plots)

python setup.py bench
python setup.py bench --with-plots
python setup.py bench --filter=cross_entropy
 2021-02-08 15:16:41 -05:00
Philippe Tillet
ae3c6a1022 [PYTHON] Removed .softmax from ops/__init__.py following previous commit 2021-02-07 17:02:11 -05:00
Philippe Tillet
9ed392db9c [PYTHON] Removed support for dense softmax 
Interest seems limited now that it is fused in cross_entropy. Will
likely re-add once it's easier to share code between ops
 2021-02-07 16:46:47 -05:00
Jared Kaplan
db55ef4fa7 Added a Softmax Xent Op (#53) 
Also includes a bugfix in kernel.py to set the device before registering the c++ function object
 2021-02-07 12:53:42 -08:00
Philippe Tillet
c8b5cb4ad5 [PYTHON] Made codebase pep8 compliant 2021-02-07 15:06:57 -05:00
Philippe Tillet
14fee16886 [PYTHON] Some cleaning of the PyBind11 wrappers (#62) 2021-02-06 20:10:44 -05:00
Philippe Tillet
d1e39d7f98 [PYTHON][OPS][BLOCKSPARSE] Now rounding softmax tile sizes to next power 
of 2
 2021-02-04 16:13:45 -05:00
Philippe Tillet
8ab5498d26 [TESTS] test_matmul.py now plots benchmarks 2021-02-04 15:35:53 -05:00
Philippe Tillet
7116df3a32 [PYTHON] Added triton.__version__ string 2021-02-03 17:39:55 -05:00
Philippe Tillet
1726197bb4 Improvements w/ Auto-Tuning and standard benchmarks (#57) 
[PYTHON] Bug-fixes in the auto-tuning module and improvement of the existing API for it
 2021-02-03 16:37:21 -05:00
Philippe Tillet
8ab68f5424 [PYTHON] Added benchmark code for CUTLASS 2021-01-31 21:23:01 -05:00
Philippe Tillet
6e77538087 [RUNTIME] Auto-tuning now works as expected when the values of 
autotune_key change
 2021-01-31 19:23:51 -05:00
Philippe Tillet
d60b989bec [PYTHON] Fixed issue with IS_TK_DIV_K 2021-01-31 13:36:55 -05:00
Philippe Tillet
ffb1e14268 [TUTORIALS] Fixed TYPO in CMakeLists.txt 2021-01-31 13:36:55 -05:00
Philippe Tillet
7db9f56d61 [PYTHON] Added benchmarking code 2021-01-31 12:59:18 -05:00
Philippe Tillet
6ecc40e2be [PYTHON][OPS] Added block-sparse softmax 2021-01-30 20:27:20 -05:00
Philippe Tillet
112bca3b8f [PYTHON] Added utility to read single Triton kernel from provided file 
in triton.read
 2021-01-30 18:09:49 -05:00
Philippe Tillet
79d098450f [PYTHON][TESTS][DOC] Various improvement of the API and code quality: 
* Simplified `triton.kernel` API to achieve lower latency:
  > .data_ptr() must now be passed as kernel argument. No more implicit
conversion from torch.tensor
  > compilation options are now constant attributes, i.e., opt.d('VAR')
becomes opt.VAR
  > torch.device must now be passed explicitly to triton.kernel (no
longer inferred from torch.tensor arguments)
* C++ tests moved to `python/tests/`
* C++ tutorial created in `tutorials/`
* Python tutorial created in python/tutorials/
* Version changed to 1.0alpha
* No longer copying C++ headers into the Python package
* added python/triton/ops/ package for pre-written Triton ops
 2021-01-29 17:27:16 -05:00
Philippe Tillet
af080740f2 [GENERAL] Merged v1.0alpha into master. Added features are: 
- A100 support via mma.16816
- Thread swizzling for conflict-free shared memory accesses without
padding
- Complete overhaul of the LLVM code generation in
codegen/selection/generator.cc to remove overengineering
- Added debugging capabilities in the Python binding
- Compilation error for kernels that spill
 2021-01-11 19:23:24 -05:00
Philippe Tillet
0cf2d22ffc [PYTHON] Added TRITON_DEBUG_MODE which reallocates input tensors outside of the pytorch memory pool to spot out-of-bounds accesses more easily 2020-12-26 15:15:46 -05:00
Philippe Tillet
d913cbd916 [VERSION] 0.2.3 -> 0.3.0 2020-12-08 23:34:55 -05:00
Philippe Tillet
22fc1cef16 [PYTHON] Context switching logic moved to PyTorch 2020-11-26 22:46:27 -05:00
Philippe Tillet
7710e048f4 [DRIVER] Simplified Driver API by substantially removing reliance on driver::context 2020-11-26 00:38:25 -05:00
Philippe Tillet
fa066b531c [PYTHON] Compiling Triton in Release mode now... 2020-11-13 01:44:52 -05:00
Philippe Tillet
51025ca2ad [DRIVER] Improved performance of Host driver code 2020-11-12 02:11:45 -05:00
Philippe Tillet
6c5284ed3b [GENERAL] Various bugfixes 2020-11-11 14:44:56 -05:00
Philippe Tillet
a2d54b5ad3 [General] LLVM-9 -> LLVM-10 2020-11-07 22:46:18 -05:00
Philippe Tillet
81000db9e9 [PYTHON] Added option to show PTX source code in Python 2020-11-07 02:55:48 -05:00
Philippe Tillet
c1920cbabb [PYTHON] Fixed torch ABI issue 2020-11-05 12:37:00 -05:00
Philippe Tillet
ced0f5f944 [GENERAL] Fix compatibility issue with older Torch versions 2020-11-03 16:02:02 -05:00
Philippe Tillet
e2c1ac8d24 [LANG] Added log intrinsic 2020-11-03 15:50:11 -05:00
Philippe Tillet
37ee888d88 [PYTHON] Cleaning C++ bindings 2020-11-02 15:06:08 -05:00
Philippe Tillet
9be1d5afc2 [GENERAL] Various improvements: 
* Sparse einsum in triton.ops.einsum
* Hacky support for fixed-tile-size atomic-add
* Various bugfixes in parser
 2020-10-25 12:16:40 -07:00
Philippe Tillet
0cbee3ec56 [CODEGEN] More work on the CPU backend 2020-09-14 10:59:40 -04:00
Philippe Tillet
526a3468e0 [Version] Now version 0.2.3 2020-09-03 14:36:30 -04:00
Philippe Tillet
010f85e91d [VERSION] Now version 0.2.2 2020-08-31 15:16:09 -04:00
Philippe Tillet
3195aca452 [PYTHON] Added credits to Scott Gray for the idea used in launch.cc 2020-08-12 19:39:57 -04:00
Philippe Tillet
30ac1359b9 [RUNTIME] Lower-level interface for executing functions 2020-08-12 18:33:35 -04:00
Philippe Tillet
2d6484482f [CODEGEN][ANALYSIS] Fixed issue in layout inference 2020-08-10 11:53:11 -04:00
Philippe Tillet
bd2067606c [EXAMPLES] Improved mat_mul example 2020-08-06 17:29:52 -04:00
Philippe Tillet
f01bdd6207 [EXAMPLES] Added conv2d example 2020-08-06 17:29:52 -04:00
Philippe Tillet
ec923cfc5e [PACKAGING] Now version 0.2.1 2020-07-29 14:07:28 -04:00
Philippe Tillet
3683b38963 [PACKAGING] Now version 0.2.0 2020-07-14 19:50:39 -04:00
Philippe Tillet
cc7c77246b [EXAMPLES][TUTORIAL] Changed to new triton.kernel API 2020-07-08 13:39:19 -04:00
Philippe Tillet
baabaf0d58 [TRITON][PYTHON] Reverted back to distutils 2020-06-24 16:39:49 -04:00
Philippe Tillet
8e3ec0556f [TRITON][KERNEL] Fixed issue for concurrent compilation of torch 
extensions
 2020-06-24 15:47:43 -04:00