[NFC] Avoid repeated work in Precompute of GC allocations #7763
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
tlively
approved these changes
Jul 28, 2025
Member
tlively
left a comment
tlively
left a comment
There was a problem hiding this comment.
Let's add a test that would trigger the previous slowdown. Of course the test would pass without this change, but we would still notice if a future regression made a test start taking a much longer time to finish.
Member
Author
|
Heh, good idea. Added, with an extra |
tlively
approved these changes
Jul 28, 2025
kripken
added a commit
that referenced
this pull request
Jul 29, 2025
Fixes a regression from #7763 This is pretty subtle, unfortunately. We cache the results of a struct.new etc as we go, so that identity is preserved for precomputing RefEq etc. The old code always computed structs in full before checking the cache. #7763 removed that code as an optimization. But, it turns out, that was load-bearing in that it, by pure luck, happened to avoid a bug. The bug that was avoided, and which this PR fixes, is that when we partially precompute, we modify a struct.new and see if we can precompute that version. We were still using the same heap value cache, so we could cache the value, then later end up using it for the original (before, we'd still compute the struct.new again and use those values). To fix that, do such speculative precomputations on a side (throwaway) cache. Fixes #7765
kripken
added a commit
that referenced
this pull request
Aug 8, 2025
Precompute has two modes, one where we ignore effects. We were caching the effectless results for where the normal mode (which cares about effects) would reuse them, leading to bugs. Fixes a regression from #7763, another after #7766. We were not using the cache fully, before, and caches are bug-prone, sadly.
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.
When we precompute
struct.new, we have a cache of allocations, as weneed the ID of allocations to remain the same. The logic for managing that
cache was actually first creating a new instance, then throwing it away if
we used the cached value... so we ended up visiting the children and
computing them etc., with exponential (wasted) work in the worst case.
To fix that, just check the cache first.
Fixes #7760