Add additional default decompositions for upsample operators #7402
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/7402
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit f16afd7 with merge base 5190106 ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D67443180 |
GregoryComer
force-pushed
the
export-D67443180
branch
from
8fe514f to
e61c9b5
Compare
…#7402) Summary: There are several core ATen ops that are not yet supported on ExecuTorch, including upsample_bilinear2d.vec and upsample_nearest2d.vec. These ops are currently not decomposed by default with PyTorch export default decompositions, but should be. Existing ET consumers rely on this behavior, so we need to preserve it until we have upsample kernels ready. This change allows ET to opt-into decomposing these ops, regardless of the PyTorch default export decomposition table. This will unblock updating PyTorch with the correct behavior (see pytorch/pytorch#116684). Once the upsample kernels land in ET, we can remove these decompositions. This is currently blocked by pin bumps, which may take a while to resolve. Differential Revision: D67443180
GregoryComer
force-pushed
the
export-D67443180
branch
from
e61c9b5 to
f16afd7
Compare
|
This pull request was exported from Phabricator. Differential Revision: D67443180 |
|
Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as |
|
Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as |
2 participants
Summary:
There are several core ATen ops that are not yet supported on ExecuTorch, including upsample_bilinear2d.vec and upsample_nearest2d.vec. These ops are currently not decomposed by default with PyTorch export default decompositions, but should be. Existing ET consumers rely on this behavior, so we need to preserve it until we have upsample kernels ready.
This change allows ET to opt-into decomposing these ops, regardless of the PyTorch default export decomposition table. This will unblock updating PyTorch with the correct behavior (see pytorch/pytorch#116684).
Once the upsample kernels land in ET, we can remove these decompositions. This is currently blocked by pin bumps, which may take a while to resolve.
Differential Revision: D67443180