Add unboxedCallRedispatch#35476
Closed
ezyang wants to merge 9 commits intogh/ezyang/694/basefrom
Closed
Conversation
A few things: - Add a new BackendGeneric dispatch key, intended to be a place where we can put code which is generic across backends. No operators are modified to use it but there is a simple use case test in op_registration_test.cpp. It's enabled by default (like Variable) but fallthrough by default (like BackendSelect) - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> [ghstack-poisoned]
This was referenced Mar 26, 2020
ezyang
added a commit
that referenced
this pull request
Mar 26, 2020
A few things: - Add a new BackendGeneric dispatch key, intended to be a place where we can put code which is generic across backends. No operators are modified to use it but there is a simple use case test in op_registration_test.cpp. It's enabled by default (like Variable) but fallthrough by default (like BackendSelect) - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> ghstack-source-id: 250a4ed Pull Request resolved: #35476
💊 CircleCI build failures summary and remediationsAs of commit ae3547c (more details on the Dr. CI page): 💚 💚 Looks good so far! There are no CircleCI failures yet. 💚 💚 This comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions on the GitHub issue tracker. This comment has been revised 62 times. |
smessmer
requested changes
Mar 26, 2020
Contributor
smessmer
left a comment
There was a problem hiding this comment.
Why do we introduce a new BackendGeneric instead of just reusing BackendSelect and potentially renaming it?
A few things: - Add a new BackendGeneric dispatch key, intended to be a place where we can put code which is generic across backends. No operators are modified to use it but there is a simple use case test in op_registration_test.cpp. It's enabled by default (like Variable) but fallthrough by default (like BackendSelect) - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> Differential Revision: [D20680518](https://our.internmc.facebook.com/intern/diff/D20680518) [ghstack-poisoned]
Contributor
Author
Because... they have different semantic meanings? |
A few things: - Add a new BackendGeneric dispatch key, intended to be a place where we can put code which is generic across backends. No operators are modified to use it but there is a simple use case test in op_registration_test.cpp. It's enabled by default (like Variable) but fallthrough by default (like BackendSelect) - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> Differential Revision: [D20680518](https://our.internmc.facebook.com/intern/diff/D20680518) [ghstack-poisoned]
A few things: - Add a new BackendGeneric dispatch key, intended to be a place where we can put code which is generic across backends. No operators are modified to use it but there is a simple use case test in op_registration_test.cpp. It's enabled by default (like Variable) but fallthrough by default (like BackendSelect) - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> Differential Revision: [D20680518](https://our.internmc.facebook.com/intern/diff/D20680518) [ghstack-poisoned]
Contributor
They're different use cases but they're the same mechanism of "do some work before calling into the actual kernel". I was thinking that maybe we can unify them. |
A few things: - Add a new BackendGeneric dispatch key, intended to be a place where we can put code which is generic across backends. No operators are modified to use it but there is a simple use case test in op_registration_test.cpp. It's enabled by default (like Variable) but fallthrough by default (like BackendSelect) - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> Differential Revision: [D20680518](https://our.internmc.facebook.com/intern/diff/D20680518) [ghstack-poisoned]
A few things: - Add a new BackendGeneric dispatch key, intended to be a place where we can put code which is generic across backends. No operators are modified to use it but there is a simple use case test in op_registration_test.cpp. It's enabled by default (like Variable) but fallthrough by default (like BackendSelect) - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> Differential Revision: [D20680518](https://our.internmc.facebook.com/intern/diff/D20680518) [ghstack-poisoned]
A few things: - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> Differential Revision: [D20680518](https://our.internmc.facebook.com/intern/diff/D20680518) [ghstack-poisoned]
Contributor
Author
|
@smessmer I updated this diff to not include BackendGeneric anymore. |
A few things: - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> Differential Revision: [D20680518](https://our.internmc.facebook.com/intern/diff/D20680518) [ghstack-poisoned]
smessmer
approved these changes
Mar 31, 2020
| template<class Return, class... Args> | ||
| Return callUnboxedWithDispatchKey(const OperatorHandle& op, DispatchKey dispatchKey, Args... args) const; | ||
|
|
||
| // Like callUnboxed, but intended for use in a redispatch: you currently |
| @@ -34,6 +34,7 @@ namespace c10 { | |||
| class DispatchKeySet final { | |||
| public: | |||
| enum Full { FULL }; | |||
Contributor
There was a problem hiding this comment.
Thinking about it, I probably would have put those into static factory functions instead of constructors with an enum argument, i.e.
static DispatchKeySet DispatchKeySet::full();
static DispatchKeySet DispatchKeySet::raw();
static DispatchKeySet DispatchKeySet::full_after(...);
Seems more standard. But not a strong opinion if you wanna keep it as is.
Contributor
Author
There was a problem hiding this comment.
I'm in the habit of doing constructors because it makes field initialization syntax better (doesn't matter here as we have copy/move constructors but when you don't have those it makes a ton of difference.)
A few things: - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> Differential Revision: [D20680518](https://our.internmc.facebook.com/intern/diff/D20680518) [ghstack-poisoned]
Contributor
karansachdev-1012
pushed a commit
to karansachdev-1012/pytorch
that referenced
this pull request
Feb 17, 2026
A few things: - Add a new BackendGeneric dispatch key, intended to be a place where we can put code which is generic across backends. No operators are modified to use it but there is a simple use case test in op_registration_test.cpp. It's enabled by default (like Variable) but fallthrough by default (like BackendSelect) - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> ghstack-source-id: 53143d0 Pull Request resolved: pytorch/pytorch#35476
laurentdupin
pushed a commit
to laurentdupin/pytorch
that referenced
this pull request
Apr 24, 2026
Summary: Pull Request resolved: pytorch#35476 A few things: - Add new callUnboxedRedispatch function which can be used to do a redispatch when you don't want to add a type id to the excluded set. This will recompute the dispatch key but ignore everything including and before the currentDispatchKey - Add FULL_AFTER constructor to DispatchKeySet; used to implement redispatch. Signed-off-by: Edward Z. Yang <ezyang@fb.com> Differential Revision: D20680518 Test Plan: Imported from OSS Pulled By: ezyang fbshipit-source-id: ecd7fbdfa916d0d2550a5b19dd3ee4a9f2272457
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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:
A few things:
redispatch when you don't want to add a type id to the excluded
set. This will recompute the dispatch key but ignore everything
including and before the currentDispatchKey
redispatch.
Signed-off-by: Edward Z. Yang ezyang@fb.com
Differential Revision: D20680518