[WIP] [AutoDiff] Initial support for class-values differentiation #67545
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
Contributor
Author
|
Tagging @rxwei @BradLarson @dan-zheng Currently there are some known issues mostly around derivative thunks. Tests were not updated as well. This PR is on top of #67230 |
asl
changed the title
Handle differentiation of inouts via semantic result parameter abstraction.
…t be non-wrt and not differentiable at all
…ters there, not just semantic results. The rationale for this change is as follows: the interface type of linear maps (pullback in particular) depends on whether there are semantic result parameters. E.g. the pullback type for (T1, T2) -> R might be (R.Tan) -> (T1.Tan, T2.Tan) or (R.Tan, T2.Tan) -> (T1.Tan) depending on whether T2 is a semantic result parameter (inout or class-bound) or not. There was no way to deduce this just from the generic function signature and param / result indices as previously result indices included number of semantic result parameters only (first semantic result parameter, second semantic result parameter and so on). E.g. in the example above we might have result indices to be (0, 1), however, this will not give any information if T1 or T2 is a semantic result parameter type given a generic signature. This change extends result indices. Now their capacity is assumed to be number of results plus number of all function parameters and we encode indices of all parameters that are assumed to be semantic result parameter. So, in the example above we will have result indices to be (0, 1) if T1 is a semantic result parameter or (0, 2) is T2 is a semantic result parameter. This allows one to correctly deduce the generic linear map types for reabstraction thunks and also keep a variety of other invariants.
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.
No description provided.