Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[TRITON-MLIR] Merge master #995

Merged
merged 19 commits into from
Dec 20, 2022
Merged

[TRITON-MLIR] Merge master #995

merged 19 commits into from
Dec 20, 2022

Conversation

ptillet
Copy link
Collaborator

@ptillet ptillet commented Dec 20, 2022

No description provided.

ptillet and others added 19 commits December 19, 2022 12:26
@ptillet
[BACKEND] Fixed missing barrier in basic reduction codegen
2eeee0c
@ptillet
.
8803c84
@ptillet
[TESTING] allclose fixup (#724)
a65cb5a
@ptillet
[TUTORIALS] Attention tutorial fixup
ee481ad
@ptillet
.
7bb72cd
@ptillet
[RUNTIME] Fixed JIT bug that leg some constexpr values to be override…
0373f0a 
…n by specialization parameters (#742)
@Jokeren @ptillet
[RUNTIME] Add callback functions for external tools (#738)
873d708
@ChrisCarucci @ptillet
[DOCS] Fixed typos in 01-vector-add.py (#751)
5fac643
@fps7806 @ptillet
[RUNTIME] support multiple devices in the same process (#757)
8f38047
@ptillet
[RUNTIME] Make entry point cache key depend on triton version hash (#765
f3a1976 
)
@yanboliang @ptillet
Add bf16/fp16/fp64 support for ty_to_cpp (#800)
16a0286 
In ```torch._inductor```, we [convert 0d CPU tensor to scalar during
triton codegen](pytorch/pytorch#87329), so need
add missing triton support for bf16/fp16/fp64.
@ptillet
[BUILD] Now using cibuildwheel default
e0d25a2
@ptillet
Revert "[BUILD] Now using cibuildwheel default"
715c960 
This reverts commit 584086f.
@msaroufim @ptillet
[DOCS] Add install from source instructions to README (#821)
454d906
@ptillet
Better NVIDIA Pascal GPU Support (#827)
d43dfea
@ptillet
[TESTING] use torch.int for autotuning cache (#840)
9cde543 
For stupid reasons, ops on int8 are 3 times slower than on int, and for
another set of stupid reasons we are not using cudaMemset for `zero_`,
so using `int8` buffer in `do_bench` makes it slow.

Co-authored-by: Philippe Tillet <phil@openai.com>
@crutcher @ptillet
[FRONTEND] Fix ExternLibrary(format=) bug; type annotate build_extern…
fc8e5ca 
….py (#883)

Ran mypy over `build_extern.py`, cleaned up type annotations.

Found a fixed a bug where `ExternLibrary(format=)` was being ignored.
@crutcher @ptillet
Fix format double substitution bug: {i} => {{i}} (#886)
443c91e 
The previous `{i}` was silently expanding to the `i` from the
enumeration loop on `regular_args` (when it wasn't empty).
@ptillet
Merge branch 'triton-mlir' into phil/triton-mlir-merge-master
4283266
@ptillet ptillet merged commit 8d496ed into triton-mlir Dec 20, 2022
@ptillet ptillet deleted the phil/triton-mlir-merge-master branch December 20, 2022 01:03
ZzEeKkAa pushed a commit to ZzEeKkAa/triton that referenced this pull request Aug 5, 2024
@whitneywhtsang
[intel] Remove ClusterOpstoLLVM pass (triton-lang#995)
9d6d945 
The pass lowers nvidia_gpu operations, which is not needed for intel
backend.

Signed-off-by: Whitney Tsang <whitney.tsang@intel.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
7 participants
@ptillet @Jokeren @ChrisCarucci @fps7806 @yanboliang @msaroufim @crutcher