fix perf regression from abby canonical form by khyperia · Pull Request #162531 · rust-lang/rust

khyperia · 2026-09-09T12:29:05Z

View all comments

#161306 regressed performance on stable

I believe all the perf impact is due to this

rust/compiler/rustc_infer/src/infer/mod.rs

Line 186 in d8df826

solver_region_constraint_storage: SolverRegionConstraintStorage::new(),

before, SolverRegionConstraintStorage was a dummy simple thing, no allocations.

now, it's a RegionConstraint { and: Box([]), or: Box([Box([])]) }

the majority of the perf impact (70%ish I think) is because the compiler is not sufficiently smart to optimize And::new([]) into And(Box::new([])) (the former does a bunch of IndexSet allocations and stuff, the latter is a no-op, just a nullptr plus zero length metadata)

the the rest of the perf impact (30%ish) is due to the or case allocating the Box([Box([])]), it's not just a nullptr+zero

options to fix:

option A: just fix the And::new([]) being terrible
option B: option A, and also, SolverRegionConstraintStorage stores an Option that is lazily init on first access, to prevent the perf hit from the or case
option C: option A, and also, use some kind of SmallVec something or other to make the or case be zero-alloc. I have not profiled this due to it being an invasive change and effort, this might not actually fix the perf.

This PR is out option A to see if it actually works with the full perf machinery with PGO and whatnot instead of just on my machine (I am very inexperienced with perf testing!). It might be the case that full PGO blah blah is actually sufficiently smart to optimize And::new([]), and the actual perf diff is due to the or case (which I'm calling the 30%ish impact, might be actually 100%), in which case this PR should produce a no-op perf diff.

r? @BoxyUwU

khyperia · 2026-09-09T13:57:06Z

@bors try @rust-timer queue

fix perf regression from abby canonical form

rust-bors · 2026-09-09T14:56:59Z

☀️ Try build successful (CI)
Build commit: 6045edf (6045edfc980d86d93057e102a0447621562f9c8b)
Base parent: 55c4dfe (55c4dfed758e741620e3320ed472ff5c4140856a)

rust-timer · 2026-09-09T18:53:06Z

Finished benchmarking commit (6045edf): comparison URL.

Overall result: ✅ improvements - no action needed

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.

@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
Improvements ✅ (primary)	-0.2%	[-0.8%, -0.1%]	21
Improvements ✅ (secondary)	-0.2%	[-0.3%, -0.1%]	11
All ❌✅ (primary)	-0.2%	[-0.8%, -0.1%]	21

Max RSS (memory usage)

Results (primary 0.0%)

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

	mean	range	count
Regressions ❌ (primary)	2.1%	[0.7%, 5.0%]	3
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-2.1%	[-5.2%, -0.5%]	3
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.0%	[-5.2%, 5.0%]	6

Cycles

Results (primary -0.0%)

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

	mean	range	count
Regressions ❌ (primary)	0.7%	[0.5%, 1.0%]	9
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-1.1%	[-2.3%, -0.7%]	6
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.0%	[-2.3%, 1.0%]	15

Binary size

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

Bootstrap: 480.775s -> 481.555s (0.16%)
Artifact size: 403.36 MiB -> 403.42 MiB (0.02%)

khyperia · 2026-09-10T06:56:18Z

looks like when I said

70%ish of the perf impact is because of an insufficiently smart compiler to optimize And::new([]) into And(Box::new([]))
30%ish of the perf impact is because of allocating in the default path for an empty true region constraint due to Or

when under PGO and whatnot instead of my local machine, seeeems the first point is more like 15% rather than 70%. (this PR is fixing just the first point, and seems to come back as recovering 15% of the original regression)

I don't know how to read the online web report, but running these locally:

cargo build --release -p collector && ./target/release/collector profile_local cachegrind +4fcf39725a9c99bd495d8c73af83628a256ff9a9 --rustc2 +d8df82673d5911b6112a85bf91d9adefb2c66a1a --exact-match libc-0.2.172 --profiles Check --scenarios Full # original regression
cargo build --release -p collector && ./target/release/collector profile_local cachegrind +55c4dfed758e741620e3320ed472ff5c4140856a --rustc2 +6045edfc980d86d93057e102a0447621562f9c8b --exact-match libc-0.2.172 --profiles Check --scenarios Full # this PR

shows the first one as +8,686,194 PROGRAM TOTALS, this PR as -1,324,876 PROGRAM TOTALS

still, this PR came back fully green, so I'm down to merge this as-is, I'll undraft this. I'll try to poke around and investigate further to remove the alloc on the Or path, if this PR isn't r+ by then I'll probably schlorp it into this PR, otherwise I'll just make a new PR.

rustbot · 2026-09-10T06:57:45Z

BoxyUwU is currently at their maximum review capacity.
They may take a while to respond.

rustbot · 2026-09-10T09:03:59Z

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.

khyperia · 2026-09-10T13:10:51Z

pushed option B in the PR description, will do a perf run on that too. since yeah, basically what panstromek said in zulip:

I feel like in your situation (working on important but unstable feature) you kinda want to quickly put the perf aside and focus on the important bits first.

@bors try @rust-timer queue

fix perf regression from abby canonical form

rust-bors · 2026-09-10T14:10:33Z

☀️ Try build successful (CI)
Build commit: d851dfd (d851dfd72cbfb660ee4f04b3de37881d03833cae)
Base parent: c4c4a57 (c4c4a576936e9e67717d0deb8e74e02dd5dd10de)

rust-log-analyzer · 2026-09-10T16:43:29Z

A job failed! Check out the build log: (web) (plain enhanced) (plain)

Click to see the possible cause of the failure (guessed by this bot)

rust-log-analyzer · 2026-09-10T17:15:23Z

A job failed! Check out the build log: (web) (plain enhanced) (plain)

Click to see the possible cause of the failure (guessed by this bot)

JonathanBrouwer · 2026-09-10T17:28:48Z

@bors p=10

@BoxyUwU

fix perf regression from abby canonical form #161306 regressed performance on stable I believe all the perf impact is due to this https://github.com/rust-lang/rust/blob/d8df82673d5911b6112a85bf91d9adefb2c66a1a/compiler/rustc_infer/src/infer/mod.rs#L186 before, `SolverRegionConstraintStorage` was a dummy simple thing, no allocations. now, it's a `RegionConstraint { and: Box([]), or: Box([Box([])]) }` the majority of the perf impact (70%ish I think) is because the compiler is not sufficiently smart to optimize `And::new([])` into `And(Box::new([]))` (the former does a bunch of `IndexSet` allocations and stuff, the latter is a no-op, just a nullptr plus zero length metadata) the the rest of the perf impact (30%ish) is due to the `or` case allocating the `Box([Box([])])`, it's not just a nullptr+zero options to fix: - option A: just fix the `And::new([])` being terrible - option B: option A, *and also*, `SolverRegionConstraintStorage` stores an `Option` that is lazily init on first access, to prevent the perf hit from the `or` case - option C: option A, *and also*, use some kind of `SmallVec` something or other to make the `or` case be zero-alloc. I have not profiled this due to it being an invasive change and effort, this might not actually fix the perf. This PR is out option A to see if it actually works with the full perf machinery with PGO and whatnot instead of just on my machine (I am very inexperienced with perf testing!). It might be the case that full PGO blah blah *is* actually sufficiently smart to optimize `And::new([])`, and the actual perf diff is due to the `or` case (which I'm calling the 30%ish impact, might be actually 100%), in which case this PR should produce a no-op perf diff. r? @BoxyUwU

rust-bors · 2026-09-10T17:31:52Z

💔 Test for 5261ebd failed: CI. Failed job:

auto - dist-x86_64-linux-alt (web logs, enhanced plaintext logs)

JonathanBrouwer · 2026-09-10T17:34:51Z

@bors treeclosed=10 Runner problem git not found on EC2 runners

rust-bors · 2026-09-10T17:34:54Z

Tree closed for PRs with priority less than 10.

rust-log-analyzer · 2026-09-10T17:47:22Z

A job failed! Check out the build log: (web) (plain enhanced) (plain)

Click to see the possible cause of the failure (guessed by this bot)

JonathanBrouwer · 2026-09-10T18:07:36Z

@bors retry

JonathanBrouwer · 2026-09-10T19:37:57Z

@bors treeopen
sorry for all the tree commands on this pr :3

rust-bors · 2026-09-10T19:38:01Z

Tree is now open for merging.

rust-bors · 2026-09-10T21:16:55Z

☀️ Test successful - CI
Approved by: BoxyUwU
Duration: 3h 8m 58s
Pushing 67eda61 to main...

github-actions · 2026-09-10T21:20:22Z

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 018018e (parent) -> 67eda61 (this PR)

Test differences

Show 10 test diffs

10 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 67eda617e6a8f8ecec01e1ba7fafe2072a64adcc --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-x86_64-solaris: 1h 15m -> 1h 47m (+43.6%)
dist-i686-mingw: 55m 31s -> 34m 20s (-38.2%)
test-x86_64-gnu-next-trait-solver-polonius: 44m 55s -> 1h 1m (+36.5%)
test-x86_64-gnu-miri: 1h 3m -> 1h 24m (+33.0%)
test-pr-check-1: 28m 44s -> 38m 9s (+32.8%)
dist-apple-various: 1h 53m -> 2h 25m (+27.8%)
dist-sparcv9-solaris: 1h 14m -> 1h 34m (+27.4%)
dist-s390x-linux: 1h 31m -> 1h 6m (-27.3%)
dist-powerpc64-linux-musl: 1h 15m -> 1h 35m (+27.0%)
dist-arm-linux-gnueabi: 1h 33m -> 1h 8m (-26.0%)

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 · 2026-09-10T22:01:26Z

Finished benchmarking commit (67eda61): comparison URL.

Overall result: ✅ improvements - no action needed

@rustbot label: -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%]	1
Improvements ✅ (primary)	-0.3%	[-0.6%, -0.1%]	75
Improvements ✅ (secondary)	-0.3%	[-0.8%, -0.0%]	57
All ❌✅ (primary)	-0.3%	[-0.6%, -0.1%]	75

Max RSS (memory usage)

Results (primary 0.5%, secondary -2.7%)

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

	mean	range	count
Regressions ❌ (primary)	0.9%	[0.4%, 2.2%]	4
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.8%	[-0.8%, -0.8%]	1
Improvements ✅ (secondary)	-2.7%	[-2.7%, -2.7%]	1
All ❌✅ (primary)	0.5%	[-0.8%, 2.2%]	5

Cycles

Results (primary -0.2%)

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

	mean	range	count
Regressions ❌ (primary)	0.9%	[0.5%, 1.5%]	3
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.7%	[-1.0%, -0.4%]	7
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.2%	[-1.0%, 1.5%]	10

Binary size

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

Bootstrap: 479.823s -> 476.479s (-0.70%)
Artifact size: 403.70 MiB -> 403.80 MiB (0.02%)

rustbot added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Sep 9, 2026

This comment has been minimized.

Sign in to view

khyperia force-pushed the abby-perf branch from 6a69dda to 598fa63 Compare September 9, 2026 12:39

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 Sep 9, 2026

This comment has been minimized.

Sign in to view

rust-bors Bot pushed a commit that referenced this pull request Sep 9, 2026

Auto merge of #162531 - khyperia:abby-perf, r=<try>

6045edf

fix perf regression from abby canonical form

This comment has been minimized.

Sign in to view

rustbot removed the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Sep 9, 2026

rustbot assigned BoxyUwU Sep 10, 2026

khyperia marked this pull request as ready for review September 10, 2026 06:57

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels Sep 10, 2026

khyperia added 2 commits September 10, 2026 10:53

fix perf regression from abby canonical form

83d63c0

lazily init SolverRegionConstraintStorage on access

a8f5602

khyperia force-pushed the abby-perf branch from f591f53 to a8f5602 Compare September 10, 2026 09:03

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 Sep 10, 2026

This comment has been minimized.

Sign in to view

rust-bors Bot pushed a commit that referenced this pull request Sep 10, 2026

Auto merge of #162531 - khyperia:abby-perf, r=<try>

d851dfd

fix perf regression from abby canonical form

This comment has been minimized.

Sign in to view

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 Sep 10, 2026

rust-bors Bot mentioned this pull request Sep 10, 2026

doc: replace exceeding_bitshifts with arithmetic_overflow #162273

Merged

This comment has been minimized.

Sign in to view

rust-bors Bot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. labels Sep 10, 2026

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 Sep 10, 2026

This comment has been minimized.

Sign in to view

This was referenced Sep 10, 2026

rustdoc: add missing CCI union logic #162346

Merged

Rollup of 4 pull requests #162609

Closed

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 Sep 10, 2026

rust-bors Bot merged commit 67eda61 into rust-lang:main Sep 10, 2026
15 checks passed

rustbot added this to the 1.100.0 milestone Sep 10, 2026

khyperia deleted the abby-perf branch September 11, 2026 07:32

Uh oh!

Uh oh!

Conversation

khyperia commented Sep 9, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment has been minimized.

khyperia commented Sep 9, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

rust-bors Bot commented Sep 9, 2026

Uh oh!

This comment has been minimized.

rust-timer commented Sep 9, 2026

Overall result: ✅ improvements - no action needed

Uh oh!

khyperia commented Sep 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rustbot commented Sep 10, 2026

Uh oh!

rustbot commented Sep 10, 2026

Uh oh!

khyperia commented Sep 10, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

rust-bors Bot commented Sep 10, 2026

Uh oh!

This comment has been minimized.

rust-log-analyzer commented Sep 10, 2026

Uh oh!

rust-log-analyzer commented Sep 10, 2026

Uh oh!

JonathanBrouwer commented Sep 10, 2026

Uh oh!

This comment has been minimized.

rust-bors Bot commented Sep 10, 2026

Uh oh!

JonathanBrouwer commented Sep 10, 2026

Uh oh!

rust-bors Bot commented Sep 10, 2026

Uh oh!

rust-log-analyzer commented Sep 10, 2026

Uh oh!

JonathanBrouwer commented Sep 10, 2026

Uh oh!

This comment has been minimized.

JonathanBrouwer commented Sep 10, 2026

Uh oh!

rust-bors Bot commented Sep 10, 2026

Uh oh!

rust-bors Bot commented Sep 10, 2026

Uh oh!

Uh oh!

github-actions Bot commented Sep 10, 2026

Test differences

Job duration changes

Uh oh!

rust-timer commented Sep 10, 2026

Overall result: ✅ improvements - no action needed

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

khyperia commented Sep 9, 2026 •

edited by rustbot

Loading

khyperia commented Sep 10, 2026 •

edited

Loading