[PyTorch][2/4] Support static dispatch with multiple backends #75605
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
Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs. This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels Differential Revision: [D35510644](https://our.internmc.facebook.com/intern/diff/D35510644/) [ghstack-poisoned]
🔗 Helpful links
💊 CI failures summary and remediationsAs of commit 5f2af30 (more details on the Dr. CI page): Expand to see more
🕵️ 1 new failure recognized by patternsThe following CI failures do not appear to be due to upstream breakages
|
This was referenced Apr 11, 2022
…nds" Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs. This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels Differential Revision: [D35510644](https://our.internmc.facebook.com/intern/diff/D35510644/) [ghstack-poisoned]
|
@priyaramani has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
bdhirsh
reviewed
Apr 11, 2022
bdhirsh
reviewed
Apr 11, 2022
bdhirsh
reviewed
Apr 11, 2022
tools/codegen/gen.py
Outdated
| elif len(keys) == 0: | ||
| return generate_static_dispatch(f, cpp_sig, method=method, backend_index=backend_indices[0]) | ||
| else: | ||
| return f'TORCH_CHECK(false, "Static dispatch does not support backends with multiple kernels");' |
There was a problem hiding this comment.
Hmm... if you pass in two backends to the codegen (say CPU and QuantizedCPU), then won't the codegen emit this torch check for every f.structured_delegate NativeFunction?
There was a problem hiding this comment.
The else part is implemented later in the stack #75607. Split them up to keep diff sizes manageable.
Also if we pass in multiple backends for a native function, and if they have a structured_delegate and one or more kernels for the same, then this torch check will be emitted as of this diff.
bdhirsh
reviewed
Apr 11, 2022
| if len(keys) == 1: | ||
| return generate_static_dispatch(f, cpp_sig, method=method, backend_index=keys[0]) | ||
| elif len(keys) == 0: | ||
| return generate_static_dispatch(f, cpp_sig, method=method, backend_index=backend_indices[0]) |
There was a problem hiding this comment.
What's the idea behind passing backend_indices[0] here?
There was a problem hiding this comment.
@larryliu0820
If none of the static dispatch backends have kernels for the native function, then we just pass in the first one in the list, could be any one of them, as in generate_static_dispatch(f, ..) if the backend doesn't have a kernel or if the native function doesn't have a structured delegate we only see if the native function has compositeImplicitAugradKernel or compositeImplicitAugradKernel and use that if not emit a TORCH_CHECK https://github.com/pytorch/pytorch/pull/75605/files#diff-6382c847dac2d011f822c492e61f9fa0eceb47c2d6fc5fb4f0c42af726c1a3aaR283-R288
…nds" Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs. This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels Differential Revision: [D35510644](https://our.internmc.facebook.com/intern/diff/D35510644/) [ghstack-poisoned]
…nds" Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs. This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels Differential Revision: [D35510644](https://our.internmc.facebook.com/intern/diff/D35510644/) [ghstack-poisoned]
…nds" Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs. This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels Differential Revision: [D35510644](https://our.internmc.facebook.com/intern/diff/D35510644/) [ghstack-poisoned]
…nds" Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs. This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels Differential Revision: [D35510644](https://our.internmc.facebook.com/intern/diff/D35510644/) [ghstack-poisoned]
larryliu0820
approved these changes
Apr 16, 2022
…nds" Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs. This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels Differential Revision: [D35510644](https://our.internmc.facebook.com/intern/diff/D35510644/) [ghstack-poisoned]
This was referenced Apr 19, 2022
…nds" Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs. This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels Differential Revision: [D35510644](https://our.internmc.facebook.com/intern/diff/D35510644/) [ghstack-poisoned]
…nds" Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs. This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels Differential Revision: [D35510644](https://our.internmc.facebook.com/intern/diff/D35510644/) [ghstack-poisoned]
priyaramani
added a commit
that referenced
this pull request
Apr 22, 2022
Pull Request resolved: #75605 Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs. This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels ghstack-source-id: 154525738 Differential Revision: [D35510644](https://our.internmc.facebook.com/intern/diff/D35510644/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D35510644/)!
facebook-github-bot
pushed a commit
that referenced
this pull request
Apr 22, 2022
Summary: Pull Request resolved: #75605 Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs. This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels ghstack-source-id: 154525738 (Note: this ignores all push blocking failures!) Test Plan: Builds lite_predictor_flatbuffer with multiple backends ``` buck build --config pt.enable_lightweight_dispatch=1 --config pt.static_dispatch_backend=CPU,QuantizedCPU,CompositeExplicitAutograd //xplat/caffe2/fb/lite_predictor:lite_predictor_flatbuffer ``` Reviewed By: larryliu0820 Differential Revision: D35510644 fbshipit-source-id: f985718ad066f8578b006b4759c4a3bd6caac176
|
Hey @priyaramani. |
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 (oldest at bottom):
Usecase: Milan models have multiple backends and need to use static dispatch to save on static initialization time and to hit native functions directly from the unboxed APIs.
This change passes in List[BackendIndex] and adds ability to generate code for multiple static backends with 1 or 0 kernels
Differential Revision: D35510644