Structure query engine caches around DefId being a CrateNum & intra-crate DefIndex. #45275
Labels
A-query-system
Area: The rustc query system (https://rustc-dev-guide.rust-lang.org/query.html)
C-enhancement
Category: An issue proposing an enhancement or a PR with one.
T-compiler
Relevant to the compiler team, which will review and decide on the PR/issue.
Comments
|
Sounds like a good idea if the lookup array ends up being densely populated enough. In general, we should provide a way to implement customized in-memory caching for specific queries. |
bors
added a commit
to rust-lang-ci/rust
that referenced
this issue
Jan 14, 2024
Cache DefId-keyed queries without hashing Not yet ready for review: * My guess is that this will be a significant memory footprint hit for sparser queries and require some more logic. * Likely merits some further consideration for parallel rustc, though as noted in a separate comment the existing IndexVec sharding looks useless to me (likely always selecting the same shard today in 99% of cases). cc rust-lang#45275 r? `@ghost`
bors
added a commit
to rust-lang-ci/rust
that referenced
this issue
Jan 15, 2024
Cache DefId-keyed queries without hashing Not yet ready for review: * My guess is that this will be a significant memory footprint hit for sparser queries and require some more logic. * Likely merits some further consideration for parallel rustc, though as noted in a separate comment the existing IndexVec sharding looks useless to me (likely always selecting the same shard today in 99% of cases). Perf notes: * rust-lang#119977 (comment) evaluated a `IndexVec<CrateNum, IndexVec<DefIndex, Option<(V, DepNodeIndex)>>` scheme. This showed poor performance on incremental scenarios as the `iter()` callbacks are slower when walking the sparse vecs. In `full` scenarios this was a win for many primary benchmarks (~1-6% instructions, ~1-10% cycles), but did show significant memory overhead (+50% on many benchmarks). Next attempt will (a) skip hashing for local storage (expected to be denser) and retains the hashing for foreign storage (expected to be sparse) and (b) keep a present Vec to speed up `iter()` callbacks. cc rust-lang#45275 r? `@ghost`
lnicola
pushed a commit
to lnicola/rust-analyzer
that referenced
this issue
Apr 7, 2024
Cache local DefId-keyed queries without hashing This caches local DefId-keyed queries using just an IndexVec. This costs ~5% extra max-rss at most but brings significant runtime improvement, up to 13% cycle counts (mean: 4%) on primary benchmarks. It's possible that further tweaks could reduce the memory overhead further but this win seems worth landing despite the increased memory, particularly with regards to eliminating the present set in non-incr or storing it inline (skip list?) with the main data. We tried applying this scheme to all keys in the [first perf run] but found that it carried a significant memory hit (50%). instructions/cycle counts were also much more mixed, though that may have been due to the lack of the present set optimization (needed for fast iter() calls in incremental scenarios). Closes rust-lang/rust#45275 [first perf run]: https://perf.rust-lang.org/compare.html?start=30dfb9e046aeb878db04332c74de76e52fb7db10&end=6235575300d8e6e2cc6f449cb9048722ef43f9c7&stat=instructions:u
RalfJung
pushed a commit
to RalfJung/rust-analyzer
that referenced
this issue
Apr 27, 2024
Cache local DefId-keyed queries without hashing This caches local DefId-keyed queries using just an IndexVec. This costs ~5% extra max-rss at most but brings significant runtime improvement, up to 13% cycle counts (mean: 4%) on primary benchmarks. It's possible that further tweaks could reduce the memory overhead further but this win seems worth landing despite the increased memory, particularly with regards to eliminating the present set in non-incr or storing it inline (skip list?) with the main data. We tried applying this scheme to all keys in the [first perf run] but found that it carried a significant memory hit (50%). instructions/cycle counts were also much more mixed, though that may have been due to the lack of the present set optimization (needed for fast iter() calls in incremental scenarios). Closes rust-lang/rust#45275 [first perf run]: https://perf.rust-lang.org/compare.html?start=30dfb9e046aeb878db04332c74de76e52fb7db10&end=6235575300d8e6e2cc6f449cb9048722ef43f9c7&stat=instructions:u
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Per-crate cache maps keyed on
DefIndexmay be more efficient and/or allow us to useVecfor queries keyed onDefIdinstead ofHashMap.We can use specialization (e.g. for splitting
(DefId, T)intoCrateNumand(DefIndex, T)), I think.cc @nikomatsakis @michaelwoerister
The text was updated successfully, but these errors were encountered: