Tweak `VecCache` to improve performance#138405

Zoxc · 2025-03-12T13:49:48Z

This has some tweaks to VecCache to improve performance.

It saves a compare_exchange in complete using the new put_unique function.
It removes bound checks on entries. These are instead checked in the slot_index_exhaustive test.
initialize_bucket is outlined and tuned for that.

rustbot · 2025-03-12T13:49:53Z

rustbot has assigned @Noratrieb.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Kobzol · 2025-03-12T14:40:26Z

@bors try @rust-timer queue

bors · 2025-03-12T14:41:37Z

⌛ Trying commit 00e7039 with merge 0b0612c...

Tweak `VecCache` to improve performance This has some tweaks to `VecCache` to improve performance. - It saves a `compare_exchange` in `complete` using the new `put_unique` function. - It removes bound checks on entries. These are instead checked in the `slot_index_exhaustive` test. - `initialize_bucket` is outlined and tuned for that. cc `@Mark-Simulacrum`

bors · 2025-03-12T16:45:51Z

☀️ Try build successful - checks-actions
Build commit: 0b0612c (0b0612c92faadc7268901ca7f18b6859d25f4eba)

Mark-Simulacrum · 2025-03-12T16:55:07Z

compiler/rustc_data_structures/src/vec_cache.rs

    #[cold]
-    fn initialize_bucket<V>(&self, bucket: &AtomicPtr<Slot<V>>) -> *mut Slot<V> {
+    #[inline(never)]
+    fn initialize_bucket<V>(bucket: &AtomicPtr<Slot<V>>, bucket_idx: usize) -> *mut Slot<V> {


Why this change?

It avoids Self needing it exist when not inlined, as the needed information can be passed in registers instead.

rust-timer · 2025-03-12T18:02:35Z

Finished benchmarking commit (0b0612c): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

This is the most reliable metric that we have; it was used to determine the overall result at the top of this comment. However, even this metric can sometimes exhibit noise.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	0.9%	[0.2%, 1.8%]	15
Improvements ✅ (primary)	-0.4%	[-0.9%, -0.2%]	140
Improvements ✅ (secondary)	-0.6%	[-1.2%, -0.2%]	42
All ❌✅ (primary)	-0.4%	[-0.9%, -0.2%]	140

Max RSS (memory usage)

Results (primary -2.0%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-2.0%	[-2.8%, -1.2%]	2
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-2.0%	[-2.8%, -1.2%]	2

Cycles

Results (primary 4.1%, secondary -3.1%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	6.8%	[1.6%, 13.7%]	13
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-1.7%	[-2.1%, -1.4%]	6
Improvements ✅ (secondary)	-3.1%	[-3.5%, -2.5%]	3
All ❌✅ (primary)	4.1%	[-2.1%, 13.7%]	19

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 779.667s -> 779.823s (0.02%)
Artifact size: 365.24 MiB -> 365.04 MiB (-0.06%)

Zoxc · 2025-03-12T18:23:19Z

Local results:

Benchmark	Before	After		Before	After		Before	After
Benchmark	Time	Time	%	Physical Memory	Physical Memory	%	Committed Memory	Committed Memory	%
🟣 clap:check	1.4028s	1.3962s	-0.47%	147.60 MiB	147.57 MiB	-0.02%	192.07 MiB	192.06 MiB	-0.01%
🟣 hyper:check	0.2335s	0.2321s	-0.58%	80.01 MiB	79.95 MiB	-0.07%	121.83 MiB	121.83 MiB	-0.01%
🟣 regex:check	0.7944s	0.7918s	-0.33%	108.43 MiB	108.50 MiB	0.07%	146.66 MiB	146.80 MiB	0.09%
🟣 syn:check	1.3258s	1.3183s	-0.57%	141.93 MiB	141.92 MiB	-0.01%	180.31 MiB	180.36 MiB	0.03%
Total	3.7565s	3.7385s	-0.48%	477.97 MiB	477.95 MiB	-0.00%	640.87 MiB	641.04 MiB	0.03%
Summary	1.0000s	0.9951s	-0.49%	1 byte	1.00 bytes	-0.01%	1 byte	1.00 bytes	0.03%

Looks like something might be up with html5ever.

Zoxc · 2025-03-12T18:43:07Z

I can't reproduce the html5ever regression locally:

Benchmark	Before	After		Before	After		Before	After
Benchmark	Time	Time	%	Physical Memory	Physical Memory	%	Committed Memory	Committed Memory	%
🟣 html5ever:check	0.6299s	0.6256s	-0.67%	104.10 MiB	103.98 MiB	-0.11%	147.54 MiB	147.58 MiB	0.02%
Total	0.6299s	0.6256s	-0.67%	104.10 MiB	103.98 MiB	-0.11%	147.54 MiB	147.58 MiB	0.02%
Summary	1.0000s	0.9933s	-0.67%	1 byte	1.00 bytes	-0.11%	1 byte	1.00 bytes	0.02%

Kobzol · 2025-03-12T18:56:26Z

Cycles are noisy, although this was high above the threshold. Let's try again, just in case it was a fluke.

@bors try @rust-timer queue

bors · 2025-03-12T18:57:36Z

⌛ Trying commit 00e7039 with merge d06bb14...

Tweak `VecCache` to improve performance This has some tweaks to `VecCache` to improve performance. - It saves a `compare_exchange` in `complete` using the new `put_unique` function. - It removes bound checks on entries. These are instead checked in the `slot_index_exhaustive` test. - `initialize_bucket` is outlined and tuned for that. cc `@Mark-Simulacrum`

bors · 2025-03-12T20:59:19Z

☀️ Try build successful - checks-actions
Build commit: d06bb14 (d06bb14bb9c4c2e005971e2f02d13bb79af6845e)

rust-timer · 2025-03-13T01:36:34Z

Finished benchmarking commit (d06bb14): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

This is the most reliable metric that we have; it was used to determine the overall result at the top of this comment. However, even this metric can sometimes exhibit noise.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	1.3%	[0.6%, 1.8%]	10
Improvements ✅ (primary)	-0.4%	[-0.9%, -0.2%]	143
Improvements ✅ (secondary)	-0.6%	[-1.3%, -0.2%]	43
All ❌✅ (primary)	-0.4%	[-0.9%, -0.2%]	143

Max RSS (memory usage)

This benchmark run did not return any relevant results for this metric.

Cycles

Results (primary -1.5%, secondary -2.1%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-1.5%	[-1.5%, -1.5%]	1
Improvements ✅ (secondary)	-2.1%	[-2.1%, -2.1%]	1
All ❌✅ (primary)	-1.5%	[-1.5%, -1.5%]	1

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 779.458s -> 779.323s (-0.02%)
Artifact size: 365.27 MiB -> 365.05 MiB (-0.06%)

Kobzol · 2025-03-13T06:41:42Z

Looks like it was a fluke after all.

apiraino · 2025-04-24T11:11:39Z

So, is this patch actually ... needed? Thanks

cc @Zoxc @Kobzol

lqd · 2025-04-24T12:16:25Z

yes

wesleywiser · 2025-06-12T14:24:11Z

r? rust-lang/compiler

rust-timer · 2026-01-27T04:00:29Z

Finished benchmarking commit (f020bce): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please do so in sufficient writing along with @rustbot label: +perf-regression-triaged. If not, please fix the regressions and do another perf run. If its results are neutral or positive, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	0.1%	[0.1%, 0.1%]	2
Improvements ✅ (primary)	-0.4%	[-1.0%, -0.1%]	145
Improvements ✅ (secondary)	-0.5%	[-1.3%, -0.0%]	98
All ❌✅ (primary)	-0.4%	[-1.0%, -0.1%]	145

Max RSS (memory usage)

Results (primary 1.7%, secondary 3.2%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	1.7%	[1.7%, 1.7%]	1
Regressions ❌ (secondary)	5.2%	[4.5%, 5.6%]	6
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-2.5%	[-3.2%, -1.9%]	2
All ❌✅ (primary)	1.7%	[1.7%, 1.7%]	1

Cycles

Results (secondary -0.3%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	3.4%	[2.3%, 4.6%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-4.0%	[-4.2%, -3.8%]	2
All ❌✅ (primary)	-	-	0

Binary size

Results (primary -0.0%, secondary -0.0%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.0%	[-0.0%, -0.0%]	4
Improvements ✅ (secondary)	-0.0%	[-0.0%, -0.0%]	1
All ❌✅ (primary)	-0.0%	[-0.0%, -0.0%]	4

Bootstrap: 472.506s -> 470.841s (-0.35%)
Artifact size: 383.60 MiB -> 383.49 MiB (-0.03%)

Zoxc · 2026-01-27T14:30:15Z

Are we actually sure that just because the indices are unique they won't cause a race?

If the indices are unique, the location they operate on are also unique and won't cause a race.

I'm worried about multiple threads calling VecCache::complete at the same time (if that could ever happen), at which point acquiring the lock on the slot seems necessary.

It still uses a lock for the main slots, but this changes the secondary list of inserted keys to just insert without atomics as the length (used as the index) is unique for each VecCache::complete operation.

Zoxc · 2026-01-27T14:30:59Z

@rustbot ready

fee1-dead · 2026-01-28T17:17:12Z

Not too comfortable reviewing this, so

@rustbot reroll

JonathanBrouwer · 2026-01-28T20:29:38Z

Also not comfortable,
r? @Mark-Simulacrum as you seem to have made changes here before

Mark-Simulacrum · 2026-02-01T16:34:05Z

Can you update the PR description to reflect the more limited change being made here? r=me with that done.

Zoxc · 2026-02-01T18:40:08Z

@Mark-Simulacrum Not sure what you mean, the description is still accurate?

Mark-Simulacrum · 2026-02-01T18:54:22Z

@bors r+

Ah, I misread the line about complete. Sorry about that.

rust-bors · 2026-02-01T18:54:25Z

📌 Commit cd18a8d has been approved by Mark-Simulacrum

It is now in the queue for this repository.

rust-bors · 2026-02-01T22:04:56Z

☀️ Test successful - CI
Approved by: Mark-Simulacrum
Duration: 3h 7m 38s
Pushing 57d2fb1 to main...

github-actions · 2026-02-01T22:07:37Z

What is this?

This is an experimental post-merge analysis report that shows differences in test outcomes between the merged PR and its parent PR.

Comparing 8340622 (parent) -> 57d2fb1 (this PR)

Test differences

Show 5 test diffs

Stage 0

vec_cache::tests::bucket_entries_matches: [missing] -> pass (J0)

Stage 1

vec_cache::tests::bucket_entries_matches: [missing] -> pass (J1)

Additionally, 3 doctest diffs were found. These are ignored, as they are noisy.

Job group index

Test dashboard

Run

cargo run --manifest-path src/ci/citool/Cargo.toml -- \
    test-dashboard 57d2fb136650d05efb3ed3ea33b330bfc85844d5 --output-dir test-dashboard

And then open test-dashboard/index.html in your browser to see an overview of all executed tests.

Job duration changes

dist-aarch64-apple: 6182.7s -> 7882.0s (+27.5%)
i686-gnu-2: 5092.9s -> 6297.8s (+23.7%)
aarch64-gnu-llvm-20-1: 3300.1s -> 3920.0s (+18.8%)
dist-aarch64-msvc: 6052.5s -> 7180.5s (+18.6%)
x86_64-rust-for-linux: 2573.5s -> 3044.8s (+18.3%)
pr-check-1: 1645.8s -> 1928.9s (+17.2%)
dist-aarch64-llvm-mingw: 5535.6s -> 6393.4s (+15.5%)
i686-gnu-nopt-1: 7174.1s -> 8255.3s (+15.1%)
aarch64-gnu-debug: 3869.3s -> 4448.6s (+15.0%)
x86_64-gnu-stable: 7542.5s -> 8600.9s (+14.0%)

How to interpret the job duration changes?

Job durations can vary a lot, based on the actual runner instance
that executed the job, system noise, invalidated caches, etc. The table above is provided
mostly for t-infra members, for simpler debugging of potential CI slow-downs.

rust-timer · 2026-02-01T22:45:51Z

Finished benchmarking commit (57d2fb1): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

Our benchmarks found a performance regression caused by this PR.
This might be an actual regression, but it can also be just noise.

Next Steps:

If the regression was expected or you think it can be justified,
please write a comment with sufficient written justification, and add
@rustbot label: +perf-regression-triaged to it, to mark the regression as triaged.
If you think that you know of a way to resolve the regression, try to create
a new PR with a fix for the regression.
If you do not understand the regression or you think that it is just noise,
you can ask the @rust-lang/wg-compiler-performance working group for help (members of this group
were already notified of this PR).

@rustbot label: +perf-regression
cc @rust-lang/wg-compiler-performance

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	0.1%	[0.0%, 0.2%]	3
Improvements ✅ (primary)	-0.4%	[-1.2%, -0.2%]	126
Improvements ✅ (secondary)	-0.5%	[-1.3%, -0.0%]	93
All ❌✅ (primary)	-0.4%	[-1.2%, -0.2%]	126

Max RSS (memory usage)

Results (primary -1.1%, secondary -3.6%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-1.1%	[-1.1%, -1.1%]	1
Improvements ✅ (secondary)	-3.6%	[-3.6%, -3.6%]	1
All ❌✅ (primary)	-1.1%	[-1.1%, -1.1%]	1

Cycles

Results (secondary 0.0%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	5.5%	[2.0%, 14.8%]	5
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-6.7%	[-8.1%, -5.2%]	4
All ❌✅ (primary)	-	-	0

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 472.577s -> 473.934s (0.29%)
Artifact size: 397.87 MiB -> 397.68 MiB (-0.05%)

lqd · 2026-02-02T07:54:59Z

Tiny regressions on a synthetic benchmark, and the overall benefits easily outweigh them.

@rustbot label: +perf-regression-triaged

rustbot assigned Noratrieb Mar 12, 2025

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Mar 12, 2025