Fix conv with groups when falling in direct backend #468
Conversation
This file contains 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
mcabbott
reviewed
Feb 1, 2023
ToucheSir
reviewed
Feb 3, 2023
gabrielpreviato
commented
Feb 5, 2023
There was a problem hiding this comment.
@ToucheSir @mcabbott not the most elegant solution, but this is the idea that I had for merging both backends with the type declaration differences. Open for suggestions.
gabrielpreviato
requested review from
ToucheSir,
mcabbott and
CarloLucibello
and removed request for
ToucheSir,
mcabbott and
CarloLucibello
ToucheSir
approved these changes
Feb 10, 2023
|
This was a nice enhancement, thanks @gabrielpreviato ! |
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Outdated suggestions cannot be applied.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.
Fixing #369.
When the convolution filter type is not the same as the input array, conv uses the direct backend
Probably this backed was not updated when grouped convolution was added to conv, which led to some false warnings about dimension mismatches given that the check_dimension function changed as well.
This PR adds the same logic used in im2col backend to the direct backed, so grouped convolutions also works for the direct backend.
What has changed
Now for both conv and depthwiseconv (and their forwarding definitions), we have the functions definitions of both im2col and direct backend in a for-loop.
The idea is that future features or corrections will always be implemented for both backends avoiding the problem that happened in #369.
For this, the loop needs a new element, where we had the frontend and backend, we now have the frontend, backend, and signature. This signature is used for the parametric types of the AbstractArray that are used for the functions since there are some type restrictions for the im2col backend.
PR Checklist