Avoid excessive `memcpy`s with the new solver by nnethercote · Pull Request #160005 · rust-lang/rust

nnethercote · 2026-07-27T11:03:28Z

Cachegrind says that rustc with the new solver spends an enormous amount of time running memcpy when compiling nacl-0.5.3, ijson-0.1.6, and nvml-wrapper-sys-0.9.1. DHAT's copy mode says there are two places responsible for these excessive memcpys. This PR fixes them, for big performance wins on those crates. Details in the individual commits.

r? @lcnr

rustbot · 2026-07-27T11:03:31Z

Some changes occurred to the core trait solver

cc @rust-lang/initiative-trait-system-refactor

These commits modify the Cargo.lock file. Unintentional changes to Cargo.lock can be introduced when switching branches and rebasing PRs.

If this was unintentional then you should revert the changes before this PR is merged.
Otherwise, you can ignore this comment.

Some changes occurred in compiler/rustc_attr_parsing

cc @jdonszelmann, @JonathanBrouwer

The parser was modified, potentially altering the grammar of (stable) Rust
which would be a breaking change.

cc @fmease

nnethercote · 2026-07-27T11:03:43Z

@bors try @rust-timer queue

Avoid excessive `memcpy`s with the new solver

lcnr

can you add comments to the places using ThinVec to state that we do this for perf reasons for X crates and link to this PR?

View changes since this review

nnethercote · 2026-07-27T13:02:53Z

can you add comments to the places using ThinVec to state that we do this for perf reasons for X crates and link to this PR?

Done.

The perf run seems to be stuck. I re-pushed; let's try again.

@bors try @rust-timer queue

Avoid excessive `memcpy`s with the new solver

rust-bors · 2026-07-27T15:11:43Z

☀️ Try build successful (CI)
Build commit: 3f08afa (3f08afab6dff94d9ad4eb9c29e0d71bcf620b626)
Base parent: dfbea5b (dfbea5bc945a9d7ff017d3d4bfbd1af9295fe5b8)

rust-timer · 2026-07-27T15:54:09Z

Finished benchmarking commit (3f08afa): comparison URL.

Overall result: ❌✅ regressions and improvements - please read:

Benchmarking means the PR may be perf-sensitive. It's automatically marked not fit for rolling up. Overriding is possible but disadvised: it risks changing compiler perf.

Next, please: If you can, justify the regressions found in this try perf run in writing along with @rustbot label: +perf-regression-triaged. If not, fix the regressions and do another perf run. Neutral or positive results will clear the label automatically.

@bors rollup=never rustc-perf
@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)	0.5%	[0.3%, 0.8%]	6
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-2.2%	[-4.8%, -0.2%]	13
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

This perf run didn't have relevant results for this metric.

Cycles

Results (secondary -0.2%)

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)	3.3%	[2.3%, 4.4%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-2.5%	[-3.1%, -2.1%]	3
All ❌✅ (primary)	-	-	0

Binary size

This perf run didn't have relevant results for this metric.

Bootstrap: 486.74s -> 486.287s (-0.09%)
Artifact size: 387.74 MiB -> 387.63 MiB (-0.03%)

nnethercote · 2026-07-27T23:11:52Z

Here are the local check full icount results for the new-solver benchmarks, including nacl-0.5.3, ijson-0.1.6, and nvml-wrapper-sys-0.9.1. cycles and wall-time results were similar (but noiser), e.g. around -40% for nacl and ijson.

Looks like the CI results aren't as good as the local results (e.g. -4.81% vs -7.31% for wg-grammar, -4.07% vs -5.77% for deeply-nested-multi, -2.03% vs -3.27% for serde). Probably a PGO thing. Still a very good result.

In `try_evaluate_obligations` we call `drain_pending` with a `|_, _| true` closure. This closure is then used in a call to `partition`, which is a very complicated way of moving all the elements into a new vector that is identical to the old one. Each of those moves involves a `memcpy` because the element type is large. This commit changes the code to avoid the useless partition and just take the vector directly, for a ~40% instruction count reduction when doing `cargo check`.

`PendingObligations` is a `ThinVec` containing a pair `(PredicateObligation<'tcx>, Option<GoalStalledOn<TyCtxt<'tcx>>>)`. This pair is currently 136 bytes. This is more than 128 bytes, which means that LLVM uses (slow) `memcpy` to move the elements. This commit shrinks `GoalStalledOn` by changing its two `Vec` fields to `ThinVec`. This shrinks the pair from 136 bytes to 104 bytes, under the `memcpy` threshold. The use of `ThinVec` in `GoalStalledOn` forces us to use `ThinVec` in a number of other places. Some of these required the thin-vec `unstable` feature for `may_dangle` support. The commit also adds some size assertions, to prevent backsliding.

rustbot · 2026-07-27T23:36:31Z

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

nnethercote · 2026-07-27T23:37:40Z

I rebased over #159974, so this is now ready to land. Improvements greatly outweigh regressions.

@rustbot label: +perf-regression-triaged

lcnr · 2026-07-28T07:43:10Z

@bors r+

rust-bors · 2026-07-28T07:43:14Z

📌 Commit 94a8720 has been approved by lcnr

It is now in the queue for this repository.

rust-bors · 2026-07-28T12:45:02Z

☀️ Test successful - CI
Approved by: lcnr
Duration: 3h 10m 35s
Pushing e19d321 to main...

github-actions · 2026-07-28T12:48:32Z

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 e7b5955 (parent) -> e19d321 (this PR)

Test differences

Show 25 test diffs

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

Test dashboard

Run

cargo run --manifest-path src/ci/citool/Cargo.toml -- \
    test-dashboard e19d321c06479c6fd77533582b0d5a86651f1be3 --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

x86_64-msvc-ext1: 1h 18m -> 2h 23m (+83.1%)
x86_64-msvc-ext2: 1h 11m -> 1h 55m (+60.1%)
x86_64-rust-for-linux: 36m 25s -> 49m 22s (+35.6%)
x86_64-gnu-miri: 1h 31m -> 1h 2m (-31.8%)
test-various: 1h 41m -> 2h 13m (+31.5%)
dist-sparcv9-solaris: 1h 14m -> 1h 37m (+30.2%)
i686-msvc-2: 2h 3m -> 1h 29m (-27.7%)
dist-aarch64-apple: 1h 42m -> 2h 7m (+24.3%)
x86_64-gnu-llvm-22-3: 1h 49m -> 1h 25m (-22.3%)
dist-i686-mingw: 2h 48m -> 2h 10m (-22.2%)

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 · 2026-07-28T13:50:51Z

Finished benchmarking commit (e19d321): comparison URL.

Overall result: ❌✅ regressions and improvements - please read:

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)	0.4%	[0.1%, 0.6%]	7
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-1.9%	[-4.9%, -0.1%]	15
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

Results (primary -0.8%, secondary 4.1%)

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

	mean	range	count
Regressions ❌ (primary)	1.2%	[1.2%, 1.2%]	1
Regressions ❌ (secondary)	4.1%	[2.2%, 6.0%]	2
Improvements ✅ (primary)	-2.7%	[-2.7%, -2.7%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.8%	[-2.7%, 1.2%]	2

Cycles

Results (primary -2.0%, secondary 1.2%)

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)	4.1%	[2.1%, 6.3%]	7
Improvements ✅ (primary)	-2.0%	[-2.0%, -2.0%]	1
Improvements ✅ (secondary)	-2.9%	[-3.5%, -2.4%]	5
All ❌✅ (primary)	-2.0%	[-2.0%, -2.0%]	1

Binary size

This perf run didn't have relevant results for this metric.

Bootstrap: 490.22s -> 490.165s (-0.01%)
Artifact size: 390.35 MiB -> 388.34 MiB (-0.51%)

rustbot assigned lcnr Jul 27, 2026

This comment has been minimized.

Sign in to view

rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jul 27, 2026

This comment has been minimized.

Sign in to view

rust-bors Bot pushed a commit that referenced this pull request Jul 27, 2026

Auto merge of #160005 - nnethercote:avoid-memcpy, r=<try>

7c7fb10

Avoid excessive `memcpy`s with the new solver

lcnr reviewed Jul 27, 2026

View reviewed changes

This comment has been minimized.

Sign in to view

panstromek mentioned this pull request Jul 27, 2026

Update thinvec to 0.2.19 #159974

Merged

nnethercote force-pushed the avoid-memcpy branch from 06dbf32 to bc91096 Compare July 27, 2026 13:02

This comment has been minimized.

Sign in to view

rust-bors Bot pushed a commit that referenced this pull request Jul 27, 2026

Auto merge of #160005 - nnethercote:avoid-memcpy, r=<try>

3f08afa

Avoid excessive `memcpy`s with the new solver

This comment has been minimized.

Sign in to view

rustbot added perf-regression Performance regression. and removed S-waiting-on-perf Status: Waiting on a perf run to be completed. labels Jul 27, 2026

nnethercote added 2 commits July 28, 2026 09:12

nnethercote force-pushed the avoid-memcpy branch from bc91096 to 94a8720 Compare July 27, 2026 23:36

rustbot added the perf-regression-triaged The performance regression has been triaged. label Jul 27, 2026

panstromek mentioned this pull request Jul 28, 2026

Vec::from_iter specialization is slower on TrustedLen iterators from thin-vec #160048

Open

rust-bors Bot added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Jul 28, 2026

This comment has been minimized.

Sign in to view

rust-bors Bot added merged-by-bors This PR was explicitly merged by bors. and removed S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. labels Jul 28, 2026

rust-bors Bot merged commit e19d321 into rust-lang:main Jul 28, 2026
14 checks passed

rustbot added this to the 1.99.0 milestone Jul 28, 2026

This was referenced Jul 28, 2026

[experiment] try to enable TrustedLen for (old) thin vec #160010

Draft

Diagnostics ICE when replaying proof trees with next-solver #154329

Open

panstromek mentioned this pull request Jul 28, 2026

Avoid TrustedLen specialization that optimize poorly in collect_remaining_errors #160073

Open

nnethercote deleted the avoid-memcpy branch July 28, 2026 21:05

Uh oh!

Uh oh!

Conversation

nnethercote commented Jul 27, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rustbot commented Jul 27, 2026

Uh oh!

nnethercote commented Jul 27, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

lcnr left a comment • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

nnethercote commented Jul 27, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

rust-bors Bot commented Jul 27, 2026

Uh oh!

This comment has been minimized.

rust-timer commented Jul 27, 2026

Overall result: ❌✅ regressions and improvements - please read:

Uh oh!

nnethercote commented Jul 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rustbot commented Jul 27, 2026

Uh oh!

nnethercote commented Jul 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lcnr commented Jul 28, 2026

Uh oh!

rust-bors Bot commented Jul 28, 2026

Uh oh!

This comment has been minimized.

rust-bors Bot commented Jul 28, 2026

Uh oh!

Uh oh!

github-actions Bot commented Jul 28, 2026

Test differences

Job duration changes

Uh oh!

rust-timer commented Jul 28, 2026

Overall result: ❌✅ regressions and improvements - please read:

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

nnethercote commented Jul 27, 2026 •

edited by rustbot

Loading

lcnr left a comment •

edited by rustbot

Loading

nnethercote commented Jul 27, 2026 •

edited

Loading

nnethercote commented Jul 27, 2026 •

edited

Loading