forward c10 registrations to aten #25969
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
pytorchbot
added
the
module: internals
smessmer
changed the title
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
smessmer
added a commit
that referenced
this pull request
Sep 11, 2019
Pull Request resolved: #25969 When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. ghstack-source-id: 89860596 Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
smessmer
added a commit
that referenced
this pull request
Sep 11, 2019
Pull Request resolved: #25969 When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. ghstack-source-id: 89873100 Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
This was referenced Sep 11, 2019
Closed
Closed
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch(). This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API. Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there. Differential Revision: [D17301917](https://our.internmc.facebook.com/intern/diff/D17301917/)
ezyang
reviewed
Sep 12, 2019
Contributor
Author
|
merged into #23667 |
ezyang
reviewed
Sep 12, 2019
| .op("aten::empty.memory_format(int[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None, MemoryFormat? memory_format=None) -> Tensor", torch::RegisterOperators::options() | ||
| .impl_unboxedOnlyKernel<decltype(empty_override), &empty_override>( | ||
| "aten::empty.memory_format(int[] size, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None, MemoryFormat? memory_format=None) -> Tensor", | ||
| TensorTypeId::MSNPUTensorId) |
Contributor
There was a problem hiding this comment.
This registration looks WAY worse than the old one. A big problem is that the schema string shows up twice. Why does the schema string show up twice?
Contributor
Author
There was a problem hiding this comment.
I don't like this API either but it's the only way to implement the forwarding to ATen without serious refactorings and probably bc breaking changes to the custom op API. This is only needed for the forwarding of registrations to ATen, it won't be needed anymore once everything is on c10. I'm planning to fix the API then and deprecate the one taking an additional string.
ezyang
approved these changes
Sep 12, 2019
Contributor
ezyang
left a comment
ezyang
left a comment
There was a problem hiding this comment.
But I think the registration API is really bad and needs to be improved.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack:
When an op is registered with the c10 dispatcher but the op is actually one of the ops that still lives on globalATenDispatch(), the c10 dispatcher will forward the registration call to globalATenDispatch().
This will allow backend extensions like XLA to immediately move to the new c10 registration API, while we are moving ops from globalATenDispatch to the c10 dispatcher behind the scenes without breaking their API.
Note that for consistency, this also changes all op registrations in TypeDefault/CPUType/... They now also directly go to the c10 dispatcher, and will be rerouted to globalATenDispatch from there.
Differential Revision: D17301917