Improve waterline bootstrapping - #3248
Closed
copybara-service[bot] wants to merge 1 commit into
Closed
Conversation
This change replaces the single greedy waterline-bootstrapping pattern with an analysis pass that marks all the places bootstraps should be inserted before a secondary pass that actually mutates the IR. The original pattern was failing to compile a number of larger programs involving loops: if the op-result of a loop that exhausts all levels was later used as the input to a ct-ct mul, the pattern would not fire and it would try to modreduce after (or before) a level 0 ciphetext. The new analysis pass properly inserts a bootstrap in situations like this. After this change, the hotword convolutional model successfully compiles in ~70 seconds on my dev machine. update: this change also required a small change to the scale analysis code, so that it initializes plaintexts to the default scale in the forward pass for multiplications (which should always happen at the default scale), and updated backward exit states to respect function return annotations when present. This hardening step was needed because the change to how bootstraps are inserted necessitated adding additional adjust_scale ops (e.g., a bootstrap just before a modreduce) which don't have forward propagation in scale analysis. PiperOrigin-RevId: 953661136
|
Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA). View this failed invocation of the CLA check for more information. For the most up to date status, view the checks section at the bottom of the pull request. |
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
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.
Improve waterline bootstrapping
This change replaces the single greedy waterline-bootstrapping pattern with an analysis pass that marks all the places bootstraps should be inserted before a secondary pass that actually mutates the IR.
The original pattern was failing to compile a number of larger programs involving loops: if the op-result of a loop that exhausts all levels was later used as the input to a ct-ct mul, the pattern would not fire and it would try to modreduce after (or before) a level 0 ciphetext. The new analysis pass properly inserts a bootstrap in situations like this.
After this change, the hotword convolutional model successfully compiles in ~70 seconds on my dev machine.
update: this change also required a small change to the scale analysis code, so that it initializes plaintexts to the default scale in the forward pass for multiplications (which should always happen at the default scale), and updated backward exit states to respect function return annotations when present. This hardening step was needed because the change to how bootstraps are inserted necessitated adding additional adjust_scale ops (e.g., a bootstrap just before a modreduce) which don't have forward propagation in scale analysis.