gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
768 Commits 14 Branches 9 Tags
keren/v100-perf-regression
Commit Graph

25 Commits

Author SHA1 Message Date
Keren Zhou
3ed36dcb4d [BACKEND] MMA->DotOperand conversion for chain dot of float32 tensors (#962) 
Co-authored-by: Philippe Tillet <phil@openai.com>
 2022-12-08 20:11:51 +00:00
Keren Zhou
83f3b9165b [FRONTEND][BACKEND] Fix bool and int8 load when the other operand is given (#968) 2022-12-08 11:52:18 -08:00
Keren Zhou
71c35bcf9c [Triton-MLIR][BACKEND] Mark float to integer in Arithmetic Dialect as legal (#963) 2022-12-08 09:07:01 -08:00
Philippe Tillet
532e10cf87 [FRONTEND][BACKEND] Clean-up transpositions (#953) 2022-12-06 09:32:13 -08:00
Crutcher Dunnavant
9490252261 [FRONTEND] Support alternative install locations of system libdevice.10.bc (#951) 2022-12-06 03:41:44 +00:00
Philippe Tillet
8edfe813a5 [FRONTEND][BACKEND] Added trans instruction; made flash attention bwd pass work (#943) 2022-12-03 09:58:24 -08:00
donproc
9def1bcebf [TRITON-MLIR][FRONTEND]minor fix to run through atomic_cas test (#925) 
Co-authored-by: dongdongl <dongdongl@nvidia.com>
 2022-12-01 13:43:26 +00:00
Keren Zhou
35c9ec1103 [Triton-MLIR][Backend] Fix number of warps and threads per warp when matrices are small (#917) 2022-11-26 12:30:38 -08:00
donproc
f63be0e9b5 [TRITON-MLIR][BACKEND]support atomic_cas (#914) 
1. support atomics-cas
2. add xchg support in atomic_rmw

Co-authored-by: dongdongl <dongdongl@nvidia.com>
 2022-11-25 12:02:08 +08:00
donproc
8925c2cd11 [TRITON-MLIR][BACKEND]AtomicRMWOp supports scalar (#903) 
AtomicRMWOp supports scalar

Co-authored-by: dongdongl <dongdongl@nvidia.com>
 2022-11-23 07:59:09 +00:00
donproc
afaf59b0c9 [TRITON-MLIR][BACKEND] Atomic support mask (#889) 
Co-authored-by: dongdongl <dongdongl@nvidia.com>
 2022-11-19 19:57:19 +08:00
donproc
5eee738df7 [Triton-MLIR][FRONTEND] [BACKEND] fix atomics (#879) 
minor fix to backend and frontend of atomics, we can pass 1 test without
mask and the shape aligned with CTA size now

Co-authored-by: dongdongl <dongdongl@nvidia.com>
 2022-11-16 12:25:15 +08:00
Chenggang Zhao
57fd1864a7 [Triton-MLIR] Support FP8 (#864) 
Co-authored-by: Superjomn <yanchunwei@outlook.com>
 2022-11-10 15:53:06 +08:00
Philippe Tillet
976cf12af1 [OPTIMIZER] Fixed memory coalescing (#847) 2022-11-07 06:22:18 -08:00
ben-zhang-609
84ad215268 [Triton-MLIR] Enable libdevice for ptx backend when has external functions. (#848) 
At the phase from ptx to cubin, check whether llvm::Module has external
functions. if has, link with libdevice at:
https://github.com/openai/triton/blob/triton-mlir/python/triton/language/libdevice.10.bc
 2022-11-07 08:01:50 +00:00
Keren Zhou
4218e68d74 [Triton-MLIR] [Frontend] Return a scalar if all input args are scalar (#839) 2022-11-03 20:27:47 -07:00
ben-zhang-609
5feb6e24f9 [Triton-MLIR]Add ptx vprintf support (#825) 
Not know how to write unit test for this feature.

Co-authored-by: Yan Chunwei <yanchunwei@outlook.com>
 2022-11-02 16:39:09 +08:00
Philippe Tillet
12d60cb4a3 [BACKEND] Added support for 1D conversion blocked -> slice (#831) 2022-11-01 13:19:58 -07:00
Qingyi Liu
cdc0ec5077 [Triton-MLIR][Backend] Fix reduce conversion and unit tests for int dtypes (#826) 2022-11-01 17:42:59 +08:00
Philippe Tillet
cb1b87a688 [FRONTEND] Made test_if/test_default pass (#823) 2022-10-30 15:32:55 -07:00
Philippe Tillet
e61dc75942 [FRONTEND] Fixed inliner and got more tests to pass (#822) 
This adds a `DialectInlinerInterface` to the Triton dialect. This, along
with a few other minor semantic changes, fixes our tests on call
instructions. Also added the option to provide use an "LLVM_SYSPATH"
environment variable to link against locally build of LLVM; this was
useful for debugging this issue.
 2022-10-30 14:10:02 -07:00
Philippe Tillet
7dfab26a39 [FRONTEND][BACKEND] Fixed various bugs (#819) 
- Fixed bugs on layout conversions for int1 data (we should use int8
internally for int1 data to prevent llvm from using vec<i1> which has
different semantics)
- Fixed semantics of some casts to bool in the frontend
 2022-10-29 06:34:14 +00:00
Philippe Tillet
ac0f6793cc [BACKEND] Added support for scalars in LoadOp / StoreOp / ElementwiseOp (#814) 
Also fixed various errors that showed up in `test_core.py`, and added more TODOs for open (hopefully relatively minor) issues
 2022-10-28 16:17:55 +08:00
Philippe Tillet
3e6cc6d66c [FRONTEND] Made more tests pass (#805) 2022-10-26 17:47:33 -07:00
Philippe Tillet
a2cbe7af91 [FRONTEND] Enhanced support for binary operators (#801) 
Disabled modulo test (due to change in behavior for `frem` in nvptx
between llvm-11 and llvm-14) and bfloat16 (will require some work to
emulate in software similar to how it's done in `master`)
 2022-10-24 19:47:01 -07:00