perf: speed up iter, range, scan methods with Rc<Node> #356
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
|
|
|
|
|
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
maksymar
changed the title
|
FYI @frankdavid |
mraszyk
reviewed
Jun 30, 2025
maksymar
changed the title
maksymar
changed the title
you're right -- at first glance using but my guess is the improvement comes from avoiding repeated moves of large value. as benchmarks show this change provides a measurable impact:
updated github PR description with this explanation. does it make sense? |
There was a problem hiding this comment.
Pull Request Overview
This PR improves iteration performance for BTreeMap (and BTreeSet benchmarks) by switching from moving entire Node structs to storing nodes behind reference‐counted pointers (Rc<Node>). Key changes include modifying the Cursor enum in src/btreemap/iter.rs to wrap nodes in Rc, and updating numerous iteration methods to create Rc instances, along with adjusting benchmark values to reflect performance improvements.
Reviewed Changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
| src/btreemap/iter.rs | Updated Cursor and related iteration logic to wrap nodes in Rc. |
| benchmarks/btreeset/canbench_results.yml | Benchmark instruction counts adjusted to reflect performance changes. |
| benchmarks/btreemap/canbench_results.yml | Benchmark instruction counts adjusted to reflect performance changes. |
mraszyk
approved these changes
Jul 1, 2025
maksymar
added a commit
that referenced
this pull request
Jul 7, 2025
This PR replaces the use of reference-counted `Rc<Node>` pointers with
unique `Box<Node>` allocations in the `BTreeMap` iterator to reduce
overhead.
btreemap
```
instructions:
status: Improvements detected 🟢
counts: [total 285 | regressed 0 | improved 9 | new 0 | unchanged 276]
change: [max 0 | p75 -97.33K | median -147.22K | p25 -481.26K | min -5.34M]
change %: [max 0.00% | p75 -0.02% | median -0.03% | p25 -0.06% | min -3.25%]
```
btreeset
```
instructions:
status: Improvements detected 🟢
counts: [total 100 | regressed 0 | improved 2 | new 0 | unchanged 98]
change: [max +625.36K | p75 -288 | median -27.44K | p25 -80.94K | min -377.91K]
change %: [max +0.05% | p75 -0.07% | median -0.30% | p25 -1.17% | min -2.35%]
```
Previous change #356
was adding `Rc<Node>` (vs simple moving) which added some improvements
as well as regressions.
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 PR improves iteration performance by storing nodes as Rc<Node>.
The change is inspired by the Entry API PoC in #251, but applied separately to avoid breaking changes.
btreeset:
btreemap:
Most of the improved benchmarks related to
iter,range,scanmethods.Some regressions are small and limited to benchmarks not involving traversal.
UPD(from comment):
At first glance using
Rc<Node<K>>should not help since we don't explicitly cloneNodeanywhere.The improvement comes from avoiding repeated moves of large value.
Nodeis fairly big (multiple fields and vectors) comparing toRc, so moving it around during iteration adds overhead.Switching to
Rcmakes those moves much cheaper, it's just a pointer copy.