Absorb ciphertext layout changes into plaintext matrix for convolutions - #3335
Draft
mdgrs wants to merge 1 commit into
Draft
Absorb ciphertext layout changes into plaintext matrix for convolutions#3335mdgrs wants to merge 1 commit into
mdgrs wants to merge 1 commit into
Conversation
Collaborator
|
I will be on vacation for two weeks. If Alex can review the code I can give it a brief scan and then we can merge it |
Collaborator
Author
|
@AlexanderViand tells me that this is quite large to review and I agree. I want to try to simplify this further, so I'll mark it as draft in the meantime. In all cases, I expect to be busy with other unrelated things for the next week or so |
mdgrs
marked this pull request as draft
August 17, 2026 08:31
AlexanderViand
added a commit
to belfortlabs/heir
that referenced
this pull request
Aug 19, 2026
…nvolutions" This reverts commit 8b68a2e, the cherry-pick of google#3335. That PR is an unreviewed draft and, as its author notes, a runtime optimization rather than something we need: it avoids rotating the ciphertext after a convolution by pushing the permutation into the plaintext. The compact convolution transform used its layoutMatrixShape helper, which the PR introduces; that call reverts to the matrix shape the kernel was expanded to, which is what the code used before the PR.
AlexanderViand
added a commit
to belfortlabs/heir
that referenced
this pull request
Aug 19, 2026
The compact convolution transform took the matrix shape from layoutMatrixShape, a helper that only exists with google#3335 (an unreviewed draft we are not carrying). Take the shape the kernel was expanded to instead, which is what the surrounding code uses.
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
2 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.
We should be able to avoid rotating the ciphertext after convolutions and push the permutation into the plaintext. For our convolutional networks, this is particularly beneficial because they often look like:
Activation -> Convolution -> Activation' -> Convolution'->...
This feels a bit too complex, but I have not been able to pull it apart into simpler pieces.