Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Deduplicate obligations in opt_normalize_projection_type #91186

Closed

Conversation

the8472
Copy link
Member

@the8472 the8472 commented Nov 24, 2021

This adds a second deduplication site (in addition to the original one at impl_or_trait_obligations) in opt_normalize_projection_type since that's where the OOMing allocations occured in #74456.

Fixes an OOM and compile time blowup in the reduced test-case on that issue.

Since it only touches one place where obligations are processed it might not fix all of these blowups. If more places need deduplication then either they also need to have the vec locally wrapped in ObligationsDedup or we need to start plumbing the deduplication or a sorted set all the way through the APIs that currently use Vec.

perf results from original PR: #90913
The deep-vector debug incr-unchanged regression is most likely spurious due to incremental verification noise, so the perf impact is mostly positive.

@the8472 the8472 added the T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. label Nov 24, 2021
@rust-highfive
Copy link
Collaborator

r? @estebank

(rust-highfive has picked a reviewer for you, use r? to override)

@rust-highfive rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label Nov 24, 2021
@Aaron1011
Copy link
Member

@bors try @rust-timer queue

@rust-timer
Copy link
Collaborator

Awaiting bors try build completion.

@rustbot label: +S-waiting-on-perf

@rustbot rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Nov 26, 2021
@bors
Copy link
Contributor

bors commented Nov 26, 2021

⌛ Trying commit 34812ba124bf254ef86abf1b61c5801d58b2375e with merge 1937cf62d0cb08000a49346c146830a7de817de3...

@bors
Copy link
Contributor

bors commented Nov 26, 2021

☀️ Try build successful - checks-actions
Build commit: 1937cf62d0cb08000a49346c146830a7de817de3 (1937cf62d0cb08000a49346c146830a7de817de3)

@rust-timer
Copy link
Collaborator

Queued 1937cf62d0cb08000a49346c146830a7de817de3 with parent dd549dc, future comparison URL.

@rust-timer
Copy link
Collaborator

Finished benchmarking commit (1937cf62d0cb08000a49346c146830a7de817de3): comparison url.

Summary: This change led to very large relevant mixed results 🤷 in compiler performance.

  • Very large improvement in instruction counts (up to -6.7% on full builds of deeply-nested)
  • Very large regression in instruction counts (up to 5.4% on incr-unchanged builds of inflate)

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

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 led 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-review -S-waiting-on-perf +perf-regression

@rustbot rustbot added perf-regression Performance regression. and removed S-waiting-on-perf Status: Waiting on a perf run to be completed. labels Nov 26, 2021
@the8472
Copy link
Member Author

the8472 commented Nov 26, 2021

Mostly the same results, except a different incr-unchanged one showing up, just making it more likely that it's incremental verification noise.

@the8472
Copy link
Member Author

the8472 commented Dec 7, 2021

ping @estebank

@bors
Copy link
Contributor

bors commented Dec 15, 2021

☔ The latest upstream changes (presumably #91945) made this pull request unmergeable. Please resolve the merge conflicts.

@the8472 the8472 force-pushed the simplify-obligation-cause-hash branch from 441d34f to df24a7f Compare December 15, 2021 19:00
@the8472
Copy link
Member Author

the8472 commented Dec 15, 2021

Rerunning perf since #90423 landed.

@bors try @rust-timer queue

@rust-timer
Copy link
Collaborator

Awaiting bors try build completion.

@rustbot label: +S-waiting-on-perf

@rustbot rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Dec 15, 2021
@bors
Copy link
Contributor

bors commented Dec 15, 2021

⌛ Trying commit df24a7fe0c8de8459b11d2022276c9222f8c1191 with merge a00e9c1413f84459311a0b7c905264d28c199c08...

@rust-timer
Copy link
Collaborator

Queued a00e9c1413f84459311a0b7c905264d28c199c08 with parent c5ecc15, future comparison URL.

@rust-timer
Copy link
Collaborator

Finished benchmarking commit (a00e9c1413f84459311a0b7c905264d28c199c08): comparison url.

Summary: This change led to moderate relevant regressions 😿 in compiler performance.

  • Moderate regression in instruction counts (up to 1.5% on full builds of wg-grammar)

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

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 led 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-review -S-waiting-on-perf +perf-regression

@rustbot rustbot removed the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Dec 16, 2021
compiler/rustc_infer/src/traits/mod.rs Outdated Show resolved Hide resolved
let expected_new = iter.len();
let combined_size = initial_size + expected_new;

if combined_size <= 16 || combined_size <= current_capacity {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could using SmallVec be an alternative to making these checks yourself?

Copy link
Member Author

@the8472 the8472 Dec 21, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The checks here are to determine when to skip deduplication work, not directly to reduce memory footprint (although it does indirectly by preventing blowup with duplicate items)

compiler/rustc_infer/src/traits/mod.rs Outdated Show resolved Hide resolved
};
};

self.obligations.retain(|obligation| !is_duplicate(obligation));
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I know that we don't dedup earlier because of cache effects (a perf optimization), but doesn't that mean that you always incur the O(n) cost of deduplicating the obligations caught here? Can we check what the perf implications would be of having state in the struct for "self.obligations is already deduped"?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

but doesn't that mean that you always incur the O(n) cost of deduplicating the obligations caught here?

Well, not always the vec is grown, no, that would result in O(n²) time. It's amortized O(n) because we only do it when the vec would grow.

Can we check what the perf implications would be of having state in the struct for "self.obligations is already deduped"?

Currently the dedup struct is only constructed temporarily in method scope, it lives much shorter than the vec, so it can't remember whether the vec has already been deduped. It would require larger changes to plumb the dedup state through all method calls that pass the &mut Vec<Obligations> around. That's what I meant with the following in the PR description

or we need to start plumbing the deduplication or a sorted set all the way through the APIs that currently use Vec.

ObligationsDedup { obligations: vec }
}

pub fn extend<'b>(&mut self, iter: impl ExactSizeIterator<Item = Cow<'b, Obligation<'tcx, T>>>)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm somewhat confused. Why does Item need to be a Cow? I'm not seeing where borrows make sense, but I might be missing something ^_^

Wouldn't passing a &[Obligation<'tcx, T>] give us maybe better perf? Because doing obligations.into_iter().map(...) in the callers is incurring new allocations in the hot path.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not seeing where borrows make sense, but I might be missing something ^_^

Items from the projection cache are borrowed, items from a fresh project_type + normalization are owned.

Because doing obligations.into_iter().map(...) in the callers is incurring new allocations in the hot path.

Do you mean obligations.extend(result.obligations.into_iter().map(Cow::Owned));? That doesn't allocate new obligations, it just stuffs already owned ones into a Cow::Owned enum variant.

@apiraino
Copy link
Contributor

(reminder: after this patch will land on nightly, it's a good candidate for a beta-backport, mentioned on Zulip)

@the8472
Copy link
Member Author

the8472 commented Dec 23, 2021

@apiraino I'm unsure whether the PR should be taken as-is or refactored to replace the &mut Vec<Obligation> that is passed through various functions with a &mut ObligationSet or something like it which could be a bit more efficient.

If it's wanted for beta-inclusion I guess the smaller scope makes more sense.

@the8472 the8472 force-pushed the simplify-obligation-cause-hash branch from df24a7f to a83720c Compare December 26, 2021 19:28
@the8472 the8472 force-pushed the simplify-obligation-cause-hash branch from a83720c to f29ef49 Compare December 27, 2021 02:22
@jackh726
Copy link
Member

r? @jackh726

I'll take over the review here, but it might take me a couple days due to holidays

@rust-highfive rust-highfive assigned jackh726 and unassigned estebank Dec 30, 2021
@JohnCSimon JohnCSimon added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Jan 30, 2022
@jackh726
Copy link
Member

Let's check perf with new benchmark

@bors try @rust-timer queue

@rust-timer
Copy link
Collaborator

Awaiting bors try build completion.

@rustbot label: +S-waiting-on-perf

@rustbot rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jan 30, 2022
@bors
Copy link
Contributor

bors commented Jan 30, 2022

⌛ Trying commit f29ef49 with merge d644ece99d13e11fc68b6183ddb5b79b4ca2b14b...

@bors
Copy link
Contributor

bors commented Jan 30, 2022

☀️ Try build successful - checks-actions
Build commit: d644ece99d13e11fc68b6183ddb5b79b4ca2b14b (d644ece99d13e11fc68b6183ddb5b79b4ca2b14b)

@rust-timer
Copy link
Collaborator

Queued d644ece99d13e11fc68b6183ddb5b79b4ca2b14b with parent 8c7f2bf, future comparison URL.

@rust-timer
Copy link
Collaborator

Finished benchmarking commit (d644ece99d13e11fc68b6183ddb5b79b4ca2b14b): comparison url.

Summary: This benchmark run shows 4 relevant improvements 🎉 but 14 relevant regressions 😿 to instruction counts.

  • Average relevant regression: 1.5%
  • Average relevant improvement: -0.5%
  • Largest improvement in instruction counts: -0.5% on full builds of keccak check
  • Largest regression in instruction counts: 2.1% on full builds of projection-caching check

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

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 led 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-review -S-waiting-on-perf +perf-regression

@rustbot rustbot removed the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jan 31, 2022
@jackh726
Copy link
Member

Does this still fix a perf regression for something not covered by the benchmarks? @the8472

@jackh726 jackh726 added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Feb 11, 2022
@the8472
Copy link
Member Author

the8472 commented Feb 11, 2022

@jackh726 not that I am aware of. There still is some duplication happening in obligations that needs to be deduplicated but there already is the deduplication happening later in impl_or_trait_obligations, so this change would mostly move it to happen earlier so that the vecs reach a smaller peak size. But there's no reported exponential blowup this would fix.

It does show up as a few potential max-rss reductions in the latest benchmarks. But considering that this isn't consistent with previous perf runs that's possibly noise.

I think there's still some optimization potential here, but it'll require more changes than just this. I'll close it for now.

@the8472 the8472 closed this Feb 11, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
perf-regression Performance regression. perf-regression-triaged The performance regression has been triaged. 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.
Projects
None yet
Development

Successfully merging this pull request may close these issues.