-
Notifications
You must be signed in to change notification settings - Fork 26.3k
[PyTorch] Avoid registering ops into dispatcher in lightweight dispatch #74664
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.
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
Conversation
🔗 Helpful links
💊 CI failures summary and remediationsAs of commit 1e72c89 (more details on the Dr. CI page):
🕵️ 2 new failures recognized by patternsThe following CI failures do not appear to be due to upstream breakages:
|
| Job | Step | Action |
|---|---|---|
| Unknown | 🔁 rerun |
🚧 1 fixed upstream failure:
These were probably caused by upstream breakages that were already fixed.
Please rebase on the viable/strict branch (expand for instructions)
If your commit is older than viable/strict, run these commands:
git fetch https://github.com/pytorch/pytorch viable/strict
git rebase FETCH_HEAD
- pull / linux-xenial-py3.7-gcc5.4 / test (docs_test, 1, 1, linux.2xlarge) on Mar 24 from 7:29am to 3:17pm (a1e284d - 7f996b8)
This comment was automatically generated by Dr. CI (expand for details).
Please report bugs/suggestions to the (internal) Dr. CI Users group.
|
This pull request was exported from Phabricator. Differential Revision: D34634300 |
6136b35 to
9ab340d
Compare
|
@larryliu0820 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
…ch (pytorch#74664) Summary: This change adds the following logic: If lightweight dispatch is enabled, do not generate `TORCH_LIBARAY` API calls for operator schema and implementations, since these operators will be registered into JIT op registry. `skip_dispatcher_op_registration` is an existing argument to `gen.py`. With that set, `RegisterDispatchKey.cpp` will not generate `m.def` and `m.impl` for each native function. This logic will be removed once we find a better way to skip op registration into dispatcher. Pull Request resolved: pytorch#74664 Test Plan: Rely on unit tests for lightweight dispatch. Reviewed By: priyaramani Differential Revision: D34634300 Pulled By: larryliu0820 fbshipit-source-id: 9bf9c44a019bfc7651a357640bec9e5f29ed3bb9
9ab340d to
1e72c89
Compare
|
This pull request was exported from Phabricator. Differential Revision: D34634300 |
…ch (pytorch#74664) Summary: This change adds the following logic: If lightweight dispatch is enabled, do not generate `TORCH_LIBARAY` API calls for operator schema and implementations, since these operators will be registered into JIT op registry. `skip_dispatcher_op_registration` is an existing argument to `gen.py`. With that set, `RegisterDispatchKey.cpp` will not generate `m.def` and `m.impl` for each native function. This logic will be removed once we find a better way to skip op registration into dispatcher. Pull Request resolved: pytorch#74664 Test Plan: Rely on unit tests for lightweight dispatch. Reviewed By: priyaramani Differential Revision: D34634300 Pulled By: larryliu0820 fbshipit-source-id: 2882b96b30c8e6edadb6efa065c6310747cc08dc
…ch (#74664) Summary: This change adds the following logic: If lightweight dispatch is enabled, do not generate `TORCH_LIBARAY` API calls for operator schema and implementations, since these operators will be registered into JIT op registry. `skip_dispatcher_op_registration` is an existing argument to `gen.py`. With that set, `RegisterDispatchKey.cpp` will not generate `m.def` and `m.impl` for each native function. This logic will be removed once we find a better way to skip op registration into dispatcher. Pull Request resolved: #74664 Test Plan: Rely on unit tests for lightweight dispatch. Reviewed By: priyaramani Differential Revision: D34634300 Pulled By: larryliu0820 fbshipit-source-id: d87828f2c6c62f15024ce9e98823b09ee5a81336
…ch (#74664) Summary: This change adds the following logic: If lightweight dispatch is enabled, do not generate `TORCH_LIBARAY` API calls for operator schema and implementations, since these operators will be registered into JIT op registry. `skip_dispatcher_op_registration` is an existing argument to `gen.py`. With that set, `RegisterDispatchKey.cpp` will not generate `m.def` and `m.impl` for each native function. This logic will be removed once we find a better way to skip op registration into dispatcher. Pull Request resolved: #74664 Test Plan: Rely on unit tests for lightweight dispatch. Reviewed By: priyaramani Differential Revision: D34634300 Pulled By: larryliu0820 fbshipit-source-id: d87828f2c6c62f15024ce9e98823b09ee5a81336 (cherry picked from commit 3eb1c27)
Summary:
This change adds the following logic:
If lightweight dispatch is enabled, do not generate
TORCH_LIBARAYAPI calls for operator schema and implementations, since these operators will be registered into JIT op registry.skip_dispatcher_op_registrationis an existing argument togen.py. With that set,RegisterDispatchKey.cppwill not generatem.defandm.implfor each native function. This logic will be removed once we find a better way to skip op registration into dispatcher.Test Plan: Rely on unit tests for lightweight dispatch.
Differential Revision: D34634300