Add unstable loop unrolling hint attributes

[saethlin]

View all comments

Tracking issue: #156874

This adds as new attribute #[unroll]/#[unroll(full)]/#[unroll(never)]/#[unroll(16)] (or any u32).

#[unroll] is behind a new feature gate #![feature(loop_hints)] because I intend to add an attribute for loop vectorization as well. If a user wants to turn off loop unrolling to locally minimize code size, LLVM may vectorize the loop even though it isn't unrolled which can produce a similar code size explosion.

[rustbot]

Some changes occurred in compiler/rustc_passes/src/check_attr.rs

cc @jdonszelmann, @JonathanBrouwer

Some changes occurred in compiler/rustc_attr_parsing

cc @jdonszelmann, @JonathanBrouwer

Some changes occurred in compiler/rustc_hir/src/attrs

cc @jdonszelmann, @JonathanBrouwer

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

Some changes occurred in match lowering

cc @Nadrieril

Some changes occurred in coverage instrumentation.

cc @Zalathar

[rustbot]

r? @folkertdev

rustbot has assigned @folkertdev.
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

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: compiler
• compiler expanded to 73 candidates
• Random selection from 17 candidates

[JonathanBrouwer]

(would like to take a look at this as well, should have time in the next few days)

[folkertdev]

Looks reasonable, I'll let Jonathan take a closer look at the attribute stuff.

Maybe I'm missing something or am just used to different parts of the code base, but some style things seemed off to me. Feel free to disregard that though, I guess.

View changes since this review

[rustbot]

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.

[saethlin]

The regressions in #156816 (comment) are primarily if not entirely due to the increase in the size of Terminator, but I don't see an easy way to improve that. The size of TerminatorKind is limited by the size of the Call variant, and I do not see any opportunities to make that smaller.

I could improve the size by stuffing the attributes into TerminatorKind::Goto, but I'm not sure if that is good design, and we would be right back in this situation again if anyone (like me) wants to add call site inlining attributes.

[folkertdev]

LGTM, assuming we want to take the perf hit (which sounds reasonable). I think @JonathanBrouwer wanted to have another look?

View changes since this review

[JonathanBrouwer]

I'd like to take a look, will do so in the next few days :)

[folkertdev]

Then for the administration, let's

r? JonathanBrouwer

[JonathanBrouwer]

Yeah from taking a further look I think any way to gain back the perf would seriously regress code maintainablity, so

@bors r=JonathanBrouwer,folkertdev rollup=never

View changes since this review

[rust-bors]

📌 Commit 8ff48f5 has been approved by JonathanBrouwer,folkertdev

It is now in the queue for this repository.

[rust-bors]

☀️ Test successful - CI
Approved by: JonathanBrouwer,folkertdev
Duration: 3h 34m 7s
Pushing b998449 to main...

[github-actions]

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 9d862dd (parent) -> b998449 (this PR)

Test differences

Show 46 test diffs

Stage 1

• [codegen] tests/codegen-llvm/loop-attrs/unroll-for-metadata.rs: [missing] -> pass (J0)
• [codegen] tests/codegen-llvm/loop-attrs/unroll-for-works.rs: [missing] -> pass (J0)
• [codegen] tests/codegen-llvm/loop-attrs/unroll-loop-metadata.rs: [missing] -> pass (J0)
• [codegen] tests/codegen-llvm/loop-attrs/unroll-while-metadata.rs: [missing] -> pass (J0)
• [ui] tests/ui/attributes/unroll/invalid-unroll.rs: [missing] -> pass (J3)
• [ui] tests/ui/feature-gates/feature-gate-loop-hints.rs: [missing] -> pass (J3)
• [ui (polonius)] tests/ui/attributes/unroll/invalid-unroll.rs: [missing] -> pass (J4)
• [ui (polonius)] tests/ui/feature-gates/feature-gate-loop-hints.rs: [missing] -> pass (J4)

Stage 2

• [ui] tests/ui/attributes/unroll/invalid-unroll.rs: [missing] -> pass (J1)
• [ui] tests/ui/feature-gates/feature-gate-loop-hints.rs: [missing] -> pass (J1)
• [codegen] tests/codegen-llvm/loop-attrs/unroll-for-metadata.rs: [missing] -> pass (J2)
• [codegen] tests/codegen-llvm/loop-attrs/unroll-for-works.rs: [missing] -> pass (J2)
• [codegen] tests/codegen-llvm/loop-attrs/unroll-loop-metadata.rs: [missing] -> pass (J2)
• [codegen] tests/codegen-llvm/loop-attrs/unroll-while-metadata.rs: [missing] -> pass (J2)

Additionally, 32 doctest diffs were found. These are ignored, as they are noisy.

Job group index

• J0: pr-check-2, x86_64-gnu-llvm-21-3, x86_64-gnu-llvm-22-3
• J1: aarch64-apple, aarch64-apple-macos-26, aarch64-gnu, aarch64-gnu-llvm-21-1, aarch64-msvc-1, arm-android, armhf-gnu, dist-i586-gnu-i586-i686-musl, i686-gnu-1, i686-gnu-nopt-1, i686-msvc-1, optional-x86_64-gnu-parallel-frontend, test-various, x86_64-gnu, x86_64-gnu-debug, x86_64-gnu-gcc, x86_64-gnu-llvm-21, x86_64-gnu-llvm-21-2, x86_64-gnu-llvm-22-2, x86_64-gnu-nopt, x86_64-gnu-stable, x86_64-mingw-1, x86_64-msvc-1
• J2: aarch64-apple, aarch64-apple-macos-26, aarch64-gnu, aarch64-gnu-llvm-21-1, aarch64-msvc-1, arm-android, armhf-gnu, dist-i586-gnu-i586-i686-musl, i686-gnu-1, i686-gnu-nopt-1, i686-msvc-1, optional-x86_64-gnu-parallel-frontend, test-various, x86_64-gnu, x86_64-gnu-llvm-21-2, x86_64-gnu-llvm-22-2, x86_64-gnu-nopt, x86_64-gnu-stable, x86_64-mingw-1, x86_64-msvc-1
• J3: x86_64-gnu-llvm-21-3, x86_64-gnu-llvm-22-3, x86_64-gnu-pre-stabilization
• J4: x86_64-gnu-pre-stabilization

Test dashboard

Run

cargo run --manifest-path src/ci/citool/Cargo.toml -- \
    test-dashboard b998449636a48e2c4a362809085b600a0174e1f2 --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

1. i686-gnu-1: 1h 34m -> 2h 16m (+44.3%)
2. i686-gnu-nopt-2: 2h 27m -> 1h 25m (-42.5%)
3. x86_64-msvc-ext1: 1h 43m -> 2h 23m (+39.2%)
4. dist-powerpc64le-linux-gnu: 1h 47m -> 1h 6m (-38.3%)
5. i686-msvc-1: 2h 26m -> 3h 11m (+30.2%)
6. aarch64-apple: 2h 42m -> 3h 27m (+27.6%)
7. test-various: 2h 21m -> 1h 46m (-24.6%)
8. x86_64-mingw-1: 3h -> 2h 17m (-24.2%)
9. dist-x86_64-solaris: 1h 39m -> 1h 16m (-22.4%)
10. dist-x86_64-illumos: 1h 49m -> 1h 25m (-21.6%)

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]

Finished benchmarking commit (b998449): comparison URL.

Overall result: ❌ regressions - 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)	0.2%	[0.1%, 0.3%]	17
Regressions ❌ (secondary)	0.2%	[0.0%, 0.5%]	22
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.2%	[0.1%, 0.3%]	17

Max RSS (memory usage)

Results (primary 0.0%, secondary 1.0%)

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

	mean	range	count
Regressions ❌ (primary)	2.6%	[2.6%, 2.6%]	1
Regressions ❌ (secondary)	2.0%	[1.1%, 2.9%]	6
Improvements ✅ (primary)	-1.3%	[-1.6%, -0.9%]	2
Improvements ✅ (secondary)	-2.0%	[-2.5%, -1.5%]	2
All ❌✅ (primary)	0.0%	[-1.6%, 2.6%]	3

Cycles

Results (primary 2.3%, secondary 3.1%)

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

	mean	range	count
Regressions ❌ (primary)	2.3%	[1.9%, 2.6%]	2
Regressions ❌ (secondary)	3.7%	[2.2%, 5.8%]	10
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-2.5%	[-2.5%, -2.5%]	1
All ❌✅ (primary)	2.3%	[1.9%, 2.6%]	2

Binary size

Results (primary 0.2%, secondary 0.4%)

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

	mean	range	count
Regressions ❌ (primary)	0.2%	[0.0%, 0.7%]	103
Regressions ❌ (secondary)	0.4%	[0.0%, 1.3%]	70
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.2%	[0.0%, 0.7%]	103

Bootstrap: 519.36s -> 524.623s (1.01%)
Artifact size: 400.90 MiB -> 401.77 MiB (0.22%)