Change IR generation to minimize the depth of the allocation nesting #543
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
dsharlet
reviewed
Jan 9, 2025
dsharlet
reviewed
Jan 10, 2025
vksnk
pushed a commit
that referenced
this pull request
Jan 13, 2025
This changes `alias_buffers` to only alias copies, rather than copies and in-place calls. Doing in-place calls correctly is tricky, and doesn't really need all the complexity of `alias_buffers`. I think it will be better done as a separate pass. We also don't have much that actually relies on this functionality, and in fact a few tests were accidentally relying on it (fixed in this PR), and this should unblock #543.
dsharlet
approved these changes
Jan 16, 2025
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Outdated suggestions cannot be applied.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.
This attempts to lay out allocation nodes such that the nesting is minimized. The general idea is to iteratively build up a tree of allocations starting from the allocations with the allocations with the shortest life time as the lowest level of the tree. This is not always possible in general to do and there are corner cases where nesting of allocations will have depth N regardless of the approach, but in most practical situations this will produce a structure close to the tree (for example, for the linear pipeline it should build a perfect tree of depth ~log(N)).
There are a couple of tests in the copy_pipeline which don't alias anymore, so the checks there are commented out until simplifier is able to use buffer bounds better.