zahimoud
fd34b20fba
[BACKEND] Fixed bug in reduce; add tests
2023-04-11 18:09:18 -07:00
Keren Zhou
032509384a
[ANALYSIS] Fine-tune comments for shared memory allocation ( #1492 )
...
And add a new test to check multiple color cases which have never be
tested before
2023-04-10 09:00:36 -07:00
long.chen
f7ad8ae022
[Refine] remove const ref of mlir::Attribute ( #1486 )
...
https://mlir.llvm.org/docs/DefiningDialects/AttributesAndTypes/
https://github.com/isocpp/CppCoreGuidelines/blob/master/CppCoreGuidelines.md#f16-for-in-parameters-pass-cheaply-copied-types-by-value-and-others-by-reference-to-const
```
The C++ Attribute and Type classes in MLIR (like Ops, and many other things) are value-typed.
This means that instances of Attribute or Type are passed around by-value,
as opposed to by-pointer or by-reference.
The Attribute and Type classes act as wrappers around internal storage objects that are uniqued within an instance of an MLIRContext.
```
2023-04-08 10:38:59 -07:00
Berke Kocaoğlu
ba91f39dbf
[DOC] Fix syntax errors, typos, formatting; increase consistency ( #1357 )
...
This PR;
- Fixes syntax errors like `.type values: dict[str,
Callable[[list[Any]], Any]]` to `:type values: dict[str,
Callable[[list[Any]], Any]]`,
- Fixes typos,
- Fixes formatting like `k ++` to ` k++`,
- Increases consistency (e.g. by transforming the minority `cd dir/` to
the majority `cd dir`).
2023-03-16 15:32:02 -07:00
Christian Sigg
9ef4b5d773
Rebase to LLVM-head. ( #1200 )
...
Rebase to
37b7a60cd7
2023-02-17 13:16:11 -08:00
Christian Sigg
fc7a8e3581
Rebase Triton to LLVM-15. ( #1070 )
...
This PR rebases Triton from LLVM-14 to LLVM-15. Most changes are
mechanical, except for the analysis framework changes.
2023-02-16 06:40:53 -08:00
Yan Chunwei
88498d104a
[BACKEND] DotOp enable ld.v4 in MMAv1 ( #1020 )
...
The existing convert distributed to distributed layouts logic is based
on processing each MMA-block, this requires each MMA-block to share
exactly the same fixed pattern(such as the one described in the [NV PTX
doc](https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#warp-level-matrix-fragment-mma-16816-float )).
While for MMAv1, things are different, the MMA-block has variant
patterns for different shapes and data layouts as below
<img width="200" alt="image"
src="https://user-images.githubusercontent.com/328693/213354941-731d7856-ad24-4f48-be0e-3cf41532cfa4.png ">
This requires all the cell coordinates in DotOp output to be computed.
2023-01-19 09:42:33 -08:00
Philippe Tillet
408d1d7e87
[OPTIMIZER] Improved flash attention forward pass performance ( #1075 )
...
- Fixed typo in instruction reordering pass
- Minor additional optimizations for shared memory allocator
- Optimized flash attention tutorial forward pass kernel
2023-01-19 06:46:01 +00:00
Keren Zhou
678b9f53a2
[Backend] Use post-order traversal for liveness numbering ( #1027 )
...
Also add tests for `tt.trans`.
2023-01-03 15:11:54 -08:00
Philippe Tillet
20100a7254
Merge triton-mlir branch - Complete rewrite of the backend from scratch ( #1004 )
...
This PR merges the `triton-mlir` branch, in which we have been quietly
rewriting the Triton backend from scratch to increase maintainability,
stability and ultimately performance. Changes to the runtime are
minimal, and this new version aims to remain backward-compatible with
the previous commit. The legacy backend is now officially deprecated,
but can still be accessed via the `legacy-backend` tag.
Co-authored-by: Keren Zhou <kerenzhou@openai.com >
Co-authored-by: Yan Chunwei <yanchunwei@outlook.com >
Co-authored-by: goostavz <109190422+goostavz@users.noreply.github.com >
Co-authored-by: Shintaro Iwasaki <siwasaki@fb.com >
Co-authored-by: Yan Da <dyanab@connect.ust.hk >
Co-authored-by: Jun Yang <yangjunpro@gmail.com >
Co-authored-by: Ian Bearman <ianb@microsoft.com >
Co-authored-by: Jason Ansel <jansel@jansel.net >
Co-authored-by: Qingyi Liu <qingyil@nvidia.com >
Co-authored-by: ben-zhang-609 <110140741+ben-zhang-609@users.noreply.github.com >
Co-authored-by: Chenggang Zhao <lyricz@yeah.net >
Co-authored-by: ben-zhang-609 <benzh609@gmail.com >
Co-authored-by: dongdongl <dongdongl@nvidia.com >
2022-12-21 01:30:50 -08:00
