Move CPU kernels out of experimental #2868
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/2868
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 79f1e3e with merge base 23f8a22 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
metascroy
added
the
topic: not user facing
| @@ -19,10 +19,6 @@ void torchao::parallel_1d(const int64_t begin, const int64_t end, const F& f) { | |||
| }); | |||
| } | |||
|
|
|||
| inline void torchao::set_num_threads(int num_threads) { | |||
There was a problem hiding this comment.
Removing these because code in csrc should have a path to being torch-agnostic, and there are not plans to support set_num_threads. Note this function is not actually used by the kernels, so we can just delete it.
|
@metascroy has imported this pull request. If you are a Meta employee, you can view this in D80958790. |
metascroy
force-pushed
the
migrate-cpp-code
branch
from
bef7c2f to
ccc9f93
Compare
|
@metascroy has imported this pull request. If you are a Meta employee, you can view this in D80958790. |
metascroy
force-pushed
the
migrate-cpp-code
branch
from
ccc9f93 to
79f1e3e
Compare
|
@metascroy has imported this pull request. If you are a Meta employee, you can view this in D80958790. |
This PR:
Moves torchao/experimental/kernels -> torchao/csrc/cpu/torch_free_kernels. This directory houses CPU kernels that do not depend on torch and are written with raw pointers.
Moves torchao/experimental/ops -> torchao/csrc/cpu/shared_kernels. This directory houses CPU kernels that are shared between ATen and ExecuTorch, and can be built for either runtime based on compile-time flags.
In addition, it:
The PR is on the larger side, but most changes are not meaningful. For review, I suggest focusing on the changes to setup.py.
Test plan: CI tests