refactor(mir-transform): Calculate optimization status inside `run_passes_inner` by clubby789 · Pull Request #160057 · rust-lang/rust

clubby789 · 2026-07-28T09:07:01Z

The Optimizations enum represents whether #[optimize(none)] has been applied to a function. Currently, run_passes takes an Optimizations which is confusing (#160015 (comment)).
Calculate this in run_passes_inner instead (which also means #[optimize(none)] is applied more consistently.

r? @RalfJung

rustbot · 2026-07-28T09:07:04Z

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

rustbot · 2026-07-28T09:07:07Z

RalfJung is not on the review rotation at the moment.
They may take a while to respond.

RalfJung · 2026-07-28T09:11:52Z

This changes behavior, right? I am not sure how to figure out whether this has any bad fallout.

rustbot · 2026-07-28T09:27:03Z

Some changes occurred in coverage instrumentation.

cc @Zalathar

clubby789 · 2026-07-28T09:28:52Z

This did introduce a regression in that InstrumentCoverage is currently is_required=false. That will be fixed by the other PR, but can be fixed here by making it required.

RalfJung · 2026-07-28T09:33:01Z

Yeah the point of splitting it was to separate the functional change from the larger refactor. :)

…sses_inner`

scottmcm · 2026-07-28T16:48:19Z

meta-pondering: Does optimize(none) really want to be different from opt-level=0?

Should maybe we stop checking mir-opt-level from the session in passes and instead check something passed-in only, with that passed-in thing merging the various knobs into one?

I would be sad if people started cargo-culting #[optimize(none)] onto things in debug because they heard that made it do less than opt-level=0...

RalfJung

I think I found a few cases where this changes behavior -- previously we'd run SimplifyCfg, now we don't.

Those do seem like good changes, though I wonder how far we want to go with promising "no" optimizations...

View changes since this review

RalfJung · 2026-07-28T16:58:30Z

            &add_call_guards::CriticalCallEdges,
        ],
        None,
-        pm::Optimizations::Allowed,


This is a behavior change I think? Previously SimplifyCfg::MakeShim was always run, now it is controlled by #[optimize].

RalfJung · 2026-07-28T17:00:01Z

@@ -479,7 +478,6 @@
        &mut body,
        &[&promote_pass, &simplify::SimplifyCfg::PromoteConsts, &coverage::InstrumentCoverage],


Another SimplifyCfg that's newly controlled by the attribute.

RalfJung · 2026-07-28T17:01:14Z

            body,
            &[
                &remove_uninit_drops::RemoveUninitDrops,
                &simplify::SimplifyCfg::RemoveFalseEdges,


And another one.

RalfJung · 2026-07-28T17:01:49Z

        &impossible_predicates::ImpossiblePredicates,
        &cleanup_post_borrowck::CleanupPostBorrowck,
        &remove_noop_landing_pads::RemoveNoopLandingPads,
        &simplify::SimplifyCfg::PostAnalysis,


And another one

RalfJung · 2026-07-28T17:04:40Z

@@ -682,13 +673,7 @@
        &simplify::SimplifyCfg::PreOptimizations,


and another one

RalfJung · 2026-07-28T17:11:48Z

meta-pondering: Does optimize(none) really want to be different from opt-level=0?

You mean -Copt-level=0?

Yeah TBH I tend to agree... I feel like optimize would ideally be implemented by overwriting the mir_opt_level that passes use as the basis for their decision. (I don't know if any MIR opts do anything with -Copt-level=s/z.)

Should maybe we stop checking mir-opt-level from the session in passes and instead check something passed-in only, with that passed-in thing merging the various knobs into one?

👍

And level 0 should then IMO mean "preserve UB" so we don't need a separate knob for that either.

... though that makes it tricky to have a pass that optionally can preserve UB, since that pass will check the mir-opt-level at pass execution time rather than in is_enabled/policy.

RalfJung

Independent of that I think this PR is an improvement, so r=me if what I wrote about the SimplifyCfg changes sounds right.

View changes since this review

clubby789 · 2026-07-28T17:43:31Z

I agree that all the SimplifyCfg passes should be controlled by the attribute
@bors=RalfJung

rust-bors · 2026-07-28T17:43:34Z

Unknown command "". Run @bors help or go to https://bors.rust-lang.org/help to see available commands.

RalfJung · 2026-07-28T18:26:16Z

@bors r=RalfJung

rust-bors · 2026-07-28T18:26:19Z

📌 Commit a6b2d2b has been approved by RalfJung

It is now in the queue for this repository.

…uwer Rollup of 21 pull requests Successful merges: - #159990 (Many "predicate"-to-"clause" renamings) - #159665 (Replace most `Ty::new_fn_def` calls with `type_of` queries directly) - #159687 (rustdoc: Set tracing max_level_info when debug-logging is false) - #160057 (refactor(mir-transform): Calculate optimization status inside `run_passes_inner`) - #160060 (codegen: skip stores for entirely-uninit constant aggregate fields, attempt #2) - #160063 (Fix ICE when dumping the dep graph with the parallel frontend) - #160065 (Distinguish the dep-graph index space from the live node count) - #158038 (Split register_tool into register_attribute_tool and register_lint_tool) - #159776 (remove const hack in alloc) - #159978 (run intrinsic-test by default on x86_64-gnu) - #160008 (Avoid stale closure recovery state across statements) - #160027 (Add regression test for #132767) - #160030 (Update `browser-ui-test` version to `0.25.0`) - #160046 (Improve consistency of attribute error messages (part 2)) - #160056 (Fix associated function suggestion for generic ADTs) - #160069 (Update Rust crate tracing-subscriber to v0.3.23 [SECURITY]) - #160071 (sanitize_standard_fds: clarify macos comment) - #160076 (use unstable features when updating dependencies) - #160092 (miri ui tests: don't run native tests on stage 0) - #160093 (Switch cargo assignments to weihanglo) - #160094 (Update assignment for docs)

@RalfJung

Rollup merge of #160057 - clubby789:local-optimizations, r=RalfJung refactor(mir-transform): Calculate optimization status inside `run_passes_inner` The `Optimizations` enum represents whether `#[optimize(none)]` has been applied to a function. Currently, `run_passes` takes an `Optimizations` which is confusing (#160015 (comment)). Calculate this in `run_passes_inner` instead (which also means `#[optimize(none)]` is applied more consistently. r? @RalfJung

…uwer Rollup of 21 pull requests Successful merges: - rust-lang/rust#159990 (Many "predicate"-to-"clause" renamings) - rust-lang/rust#159665 (Replace most `Ty::new_fn_def` calls with `type_of` queries directly) - rust-lang/rust#159687 (rustdoc: Set tracing max_level_info when debug-logging is false) - rust-lang/rust#160057 (refactor(mir-transform): Calculate optimization status inside `run_passes_inner`) - rust-lang/rust#160060 (codegen: skip stores for entirely-uninit constant aggregate fields, attempt rust-lang/rust#2) - rust-lang/rust#160063 (Fix ICE when dumping the dep graph with the parallel frontend) - rust-lang/rust#160065 (Distinguish the dep-graph index space from the live node count) - rust-lang/rust#158038 (Split register_tool into register_attribute_tool and register_lint_tool) - rust-lang/rust#159776 (remove const hack in alloc) - rust-lang/rust#159978 (run intrinsic-test by default on x86_64-gnu) - rust-lang/rust#160008 (Avoid stale closure recovery state across statements) - rust-lang/rust#160027 (Add regression test for rust-lang/rust#132767) - rust-lang/rust#160030 (Update `browser-ui-test` version to `0.25.0`) - rust-lang/rust#160046 (Improve consistency of attribute error messages (part 2)) - rust-lang/rust#160056 (Fix associated function suggestion for generic ADTs) - rust-lang/rust#160069 (Update Rust crate tracing-subscriber to v0.3.23 [SECURITY]) - rust-lang/rust#160071 (sanitize_standard_fds: clarify macos comment) - rust-lang/rust#160076 (use unstable features when updating dependencies) - rust-lang/rust#160092 (miri ui tests: don't run native tests on stage 0) - rust-lang/rust#160093 (Switch cargo assignments to weihanglo) - rust-lang/rust#160094 (Update assignment for docs)

panstromek · 2026-08-03T16:53:05Z

for #160102

@rust-timer build f98c410

rust-timer · 2026-08-03T18:20:28Z

Finished benchmarking commit (f98c410): comparison URL.

Overall result: ❌✅ regressions and improvements - no action needed

Benchmarking means the PR may be perf-sensitive. Consider adding rollup=never if this change is not fit for rolling up.

@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)	0.2%	[0.2%, 0.3%]	5
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.1%	[-0.1%, -0.1%]	3
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

Results (primary -1.1%, secondary -1.3%)

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)	0.7%	[0.7%, 0.7%]	1
Improvements ✅ (primary)	-1.1%	[-2.5%, -0.5%]	19
Improvements ✅ (secondary)	-1.3%	[-2.4%, -0.7%]	26
All ❌✅ (primary)	-1.1%	[-2.5%, -0.5%]	19

Cycles

Results (primary 0.5%, 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)	0.8%	[0.4%, 1.2%]	6
Regressions ❌ (secondary)	5.1%	[0.4%, 22.7%]	7
Improvements ✅ (primary)	-0.7%	[-0.9%, -0.4%]	2
Improvements ✅ (secondary)	-3.2%	[-9.4%, -0.4%]	10
All ❌✅ (primary)	0.5%	[-0.9%, 1.2%]	8

Binary size

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

Bootstrap: 490.165s -> 497.007s (1.40%)
Artifact size: 388.34 MiB -> 392.39 MiB (1.04%)

panstromek · 2026-08-03T18:59:06Z

perf triage:

Looks like this caused a small regression in #160102. As I understand it, this PR should be mostly a cleanup, so this feels somewhat unexpected. Is there a possible problem we have missed or is it actually expected? Can we do something about the regression?

The main regression is in many-assoc-items which is a MIR-relevant stress test. include-blob is noisy and coercions debug incr-unchanged had a positive result in the rollup so it's probably also noise.

clubby789 · 2026-08-03T19:29:56Z

Perf changes beyond inlining fluctuations would be surprising, given that this should only change optimizations in the presence of #[optimize(none)]

panstromek · 2026-08-03T19:40:38Z

Dep graph size has increased, so there's definitely something more going on. From detailed results, it looks like there's 59934 more query calls to def_kind, so it looks like this call might be culrpit:

https://github.com/rust-lang/rust/pull/160057/changes#diff-deb9513f9d3c0f87f924c627700124bed199d659cb223b0796a6509198139cd0R291

clubby789 · 2026-08-03T19:50:04Z

Hm, makes sense. I'll look into this

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 Jul 28, 2026

rustbot assigned RalfJung Jul 28, 2026

clubby789 mentioned this pull request Jul 28, 2026

refactor(mir-transform): Merge can_be_overridden, is_required and is_enabled into one #160015

Merged

RalfJung reviewed Jul 28, 2026

View reviewed changes

Comment thread compiler/rustc_mir_transform/src/pass_manager.rs Outdated

clubby789 force-pushed the local-optimizations branch 2 times, most recently from 053bad2 to c325709 Compare July 28, 2026 09:27

RalfJung reviewed Jul 28, 2026

View reviewed changes

Comment thread compiler/rustc_mir_transform/src/pass_manager.rs Outdated

refactor(mir-transform): Calculate optimization status inside `run_pa…

a6b2d2b

…sses_inner`

clubby789 force-pushed the local-optimizations branch from c325709 to a6b2d2b Compare July 28, 2026 09:34

clubby789 mentioned this pull request Jul 28, 2026

Stabilize optimize attribute #157273

Open

RalfJung reviewed Jul 28, 2026

View reviewed changes

RalfJung approved these changes Jul 28, 2026

View reviewed changes

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

JonathanBrouwer mentioned this pull request Jul 28, 2026

Rollup of 21 pull requests #160102

Merged

rust-bors Bot merged commit 3048347 into rust-lang:main Jul 28, 2026
13 checks passed

rustbot added this to the 1.99.0 milestone Jul 28, 2026

This comment has been minimized.

Sign in to view

		@@ -479,7 +478,6 @@
		&mut body,
		&[&promote_pass, &simplify::SimplifyCfg::PromoteConsts, &coverage::InstrumentCoverage],

		@@ -682,13 +673,7 @@
		&simplify::SimplifyCfg::PreOptimizations,

Uh oh!

Uh oh!

Conversation

clubby789 commented Jul 28, 2026

Uh oh!

rustbot commented Jul 28, 2026

Uh oh!

rustbot commented Jul 28, 2026

Uh oh!

Uh oh!

RalfJung commented Jul 28, 2026

Uh oh!

rustbot commented Jul 28, 2026

Uh oh!

clubby789 commented Jul 28, 2026

Uh oh!

RalfJung commented Jul 28, 2026

Uh oh!

Uh oh!

scottmcm commented Jul 28, 2026

Uh oh!

RalfJung 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

RalfJung commented Jul 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

RalfJung 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!

clubby789 commented Jul 28, 2026

Uh oh!

rust-bors Bot commented Jul 28, 2026

Uh oh!

RalfJung commented Jul 28, 2026

Uh oh!

rust-bors Bot commented Jul 28, 2026

Uh oh!

Uh oh!

panstromek commented Aug 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment has been minimized.

rust-timer commented Aug 3, 2026

Overall result: ❌✅ regressions and improvements - no action needed

Uh oh!

panstromek commented Aug 3, 2026

Uh oh!

clubby789 commented Aug 3, 2026

Uh oh!

panstromek commented Aug 3, 2026

Uh oh!

clubby789 commented Aug 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

RalfJung left a comment •

edited by rustbot

Loading

RalfJung commented Jul 28, 2026 •

edited

Loading

RalfJung left a comment •

edited by rustbot

Loading

panstromek commented Aug 3, 2026 •

edited

Loading