T Compiler Meeting Agenda 2024 08 01

T-compiler Meeting Agenda 2024-08-01

Announcements

  • Reminder: if you see a PR/issue that seems like there might be legal implications due to copyright/IP/etc, please let us know (or at least message @davidtwco or @Wesley Wiser so we can pass it along).

Other WG meetings

MCPs/FCPs

WG checkins

None

Backport nominations

T-compiler beta / T-compiler stable

  • :beta: “Disable jump threading of float equality” rust#128271
    • Authored by Noratrieb
    • fixes #128243, critical unsoundness related to the Jump threading MIR opt
  • :beta: “Revert recent changes to dead code analysis” rust#128404
    • Authored by compiler-errors
    • reverts a bunch of changes causing false diags on linting, fixes #128272, #126169
    • Not yet merged but r’ed by @pnkfelix
  • :stable: “Disable jump threading of float equality” rust#128271
    • Authored by Noratrieb
    • see above. stable nominated by @Ben Kimock (Saethlin) (inferred intent: “if a dot release happens”)

T-types stable / T-types beta

  • No beta nominations for T-types this time.
  • No stable nominations for T-types this time.

PRs S-waiting-on-team

T-compiler

Issues of Note

Short Summary

P-critical

T-compiler

  • “Jump threading MIR opt unsoundly uses bitpattern equality for floats” rust#128243
    • fixed by #128271, PR backport nominated

T-types

  • No P-critical issues for T-types this time.

P-high regressions

P-high beta regressions

  • No P-high beta regressions this time.

Unassigned P-high nightly regressions

  • “Generated WebAssembly unexpectedly requires reference types” rust#128475
    • rustc basically produces code not compliant with wasm 1.0 spec (but this has been going on for a while, see comment on Zulip)
    • possibly this can downgraded (waiting for some WASM experts to chime in)

Performance logs

triage logs 2024-07-30

There were some notable regressions this week. Some of them are being addressed via follow-up PRs (such as the change to whitespace diagnostic reporting), and some via reverts (such as the dead code analysis that tried to flag pub structs without pub constructors). A few regressions have not yet been addressed. See report for details.

Triage done by @pnkfelix. Revision range: 9629b90b..7e3a9718

Summary:

(instructions:u) mean range count
Regressions (primary) 1.3% [0.2%, 6.1%] 43
Regressions (secondary) 1.9% [0.1%, 10.4%] 46
Improvements (primary) -1.0% [-3.9%, -0.2%] 27
Improvements (secondary) -1.6% [-6.8%, -0.2%] 43
All (primary) 0.4% [-3.9%, 6.1%] 70

5 Regressions, 6 Improvements, 6 Mixed; 8 of them in rollups 65 artifact comparisons made in total

Regressions

Do not use global caches if opaque types can be defined #126024 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) 3.4% [1.6%, 5.5%] 6
Regressions (secondary) 3.1% [0.4%, 5.4%] 11
Improvements (primary) - - 0
Improvements (secondary) - - 0
All (primary) 3.4% [1.6%, 5.5%] 6
  • This PR says it is fixing a soundness problem. (Its not clear to me if the wrong issue was linked; the linked one is an ICE that was not actually resolved.)
  • All six of the regressions are to hyper: {check,debug,opt} x {incr-full, full}.
  • we probably should just accept this cost

Rollup of 5 pull requests #128169 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) 0.9% [0.2%, 3.0%] 26
Regressions (secondary) 0.5% [0.3%, 2.2%] 13
Improvements (primary) - - 0
Improvements (secondary) - - 0
All (primary) 0.9% [0.2%, 3.0%] 26
  • the bulk of the regressions are to syn (i.e. 8 out of the 9 that are > 1%).
  • this was due to a change in how diagnostics handle certain “whitespace” characters (PR #127528); there is a revert proposed in PR #128179, but there is also a PR to address the issue itself as a followup in PR #128200
  • not marking as triaged until either PR #128179 or PR #128200 is landed.

Rollup of 7 pull requests #128186 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) 0.3% [0.2%, 0.5%] 11
Regressions (secondary) - - 0
Improvements (primary) - - 0
Improvements (secondary) - - 0
All (primary) 0.3% [0.2%, 0.5%] 11
  • already marked as triaged

Rollup of 9 pull requests #128253 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) 0.5% [0.4%, 0.5%] 3
Regressions (secondary) - - 0
Improvements (primary) - - 0
Improvements (secondary) - - 0
All (primary) 0.5% [0.4%, 0.5%] 3
  • regressed incr-full for bitmaps-{check,opt} and typenum-check
  • seems like noise from the graph over time; marking as triaged.

Document 0x10.checked_shl(BITS - 1) does not overflow #128255 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) 0.5% [0.5%, 0.6%] 4
Regressions (secondary) 2.2% [2.2%, 2.2%] 1
Improvements (primary) - - 0
Improvements (secondary) - - 0
All (primary) 0.5% [0.5%, 0.6%] 4
  • noise, already marked as triaged

Improvements

Remove unnecessary impl sorting in queries and metadata #120812 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) - - 0
Regressions (secondary) - - 0
Improvements (primary) -1.2% [-2.1%, -0.4%] 2
Improvements (secondary) -0.3% [-0.4%, -0.3%] 2
All (primary) -1.2% [-2.1%, -0.4%] 2

rustdoc: clean up and fix ord violations in item sorting #128146 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) - - 0
Regressions (secondary) - - 0
Improvements (primary) -0.7% [-1.6%, -0.2%] 4
Improvements (secondary) - - 0
All (primary) -0.7% [-1.6%, -0.2%] 4

Rollup of 6 pull requests #128195 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) - - 0
Regressions (secondary) - - 0
Improvements (primary) -0.4% [-0.5%, -0.4%] 5
Improvements (secondary) - - 0
All (primary) -0.4% [-0.5%, -0.4%] 5
  • (just noise I think)

Switch from derivative to derive-where #127042 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) - - 0
Regressions (secondary) - - 0
Improvements (primary) -0.2% [-0.3%, -0.2%] 16
Improvements (secondary) -0.5% [-0.6%, -0.4%] 8
All (primary) -0.2% [-0.3%, -0.2%] 16

Always set result during finish() in debug builders #127946 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) - - 0
Regressions (secondary) - - 0
Improvements (primary) -0.5% [-0.6%, -0.5%] 6
Improvements (secondary) - - 0
All (primary) -0.5% [-0.6%, -0.5%] 6
  • (just noise I think)

Rollup of 6 pull requests #128313 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) - - 0
Regressions (secondary) - - 0
Improvements (primary) -1.0% [-1.1%, -1.0%] 2
Improvements (secondary) -0.9% [-1.9%, -0.2%] 10
All (primary) -1.0% [-1.1%, -1.0%] 2

Mixed

Try to fix ICE from re-interning an AllocId with different allocation contents #127442 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) - - 0
Regressions (secondary) 0.8% [0.2%, 2.5%] 4
Improvements (primary) - - 0
Improvements (secondary) -0.7% [-1.4%, -0.3%] 7
All (primary) - - 0
  • the regressions are to secondary benchmarks and this is fixing a subtle ICE that arises from a race condition (and may actually represent a chance of miscompilation, maybe?)
  • marked as triaged

Rollup of 8 pull requests #128155 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) 0.5% [0.2%, 0.8%] 6
Regressions (secondary) 0.9% [0.7%, 1.0%] 7
Improvements (primary) -0.5% [-0.6%, -0.4%] 4
Improvements (secondary) - - 0
All (primary) 0.1% [-0.6%, 0.8%] 10
  • regressions are to hyper and exa. Mostly in hyper check-full, check-incr-full, and debug-incr-full.
  • bulk of time might be from spike in time spent in mir_const_qualif query ?
  • not marking as triaged, (though it is, to be clear, a relatively minor regression).

Allow optimizing u32::from::<char>. #124905 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) 0.2% [0.2%, 0.3%] 4
Regressions (secondary) - - 0
Improvements (primary) -0.2% [-0.2%, -0.2%] 1
Improvements (secondary) - - 0
All (primary) 0.1% [-0.2%, 0.3%] 5
  • regressions are to image opt {full, incr-full}, cargo opt {full, incr-full}, and syn opt incr-unchanged
  • It appears that its due to extra time spent in LLVM opt, especially lto optimize, which makes sense given that this is meant to be enabling LLVM to attempt more such optimizations?
  • marked as triaged.

Rollup of 3 pull requests #128301 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) - - 0
Regressions (secondary) 2.1% [2.1%, 2.1%] 1
Improvements (primary) -0.2% [-0.3%, -0.2%] 2
Improvements (secondary) -1.6% [-3.0%, -0.2%] 2
All (primary) -0.2% [-0.3%, -0.2%] 2
  • sole regression is to secondary benchmark coercions debug-full.
  • seems like noise.
  • marked as triaged

Perform instsimplify before inline to eliminate some trivial calls #128265 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) 1.2% [0.2%, 2.6%] 4
Regressions (secondary) 0.5% [0.5%, 0.5%] 1
Improvements (primary) -0.5% [-0.8%, -0.2%] 12
Improvements (secondary) -0.3% [-0.4%, -0.3%] 2
All (primary) -0.0% [-0.8%, 2.6%] 16
  • main primary regressions are to ripgrep opt full and image opt-full
  • these changes were anticipated during review, seems likely result of changes to inlining decisions
  • marked as triaged

Rollup of 6 pull requests #128360 (Comparison Link)

(instructions:u) mean range count
Regressions (primary) 0.6% [0.4%, 0.7%] 4
Regressions (secondary) 4.4% [0.3%, 12.0%] 10
Improvements (primary) -0.3% [-0.3%, -0.3%] 4
Improvements (secondary) - - 0
All (primary) 0.1% [-0.3%, 0.7%] 8
  • primary regressions are to doc-full for html5ever, stm32f4, libc, and typenum
  • those are presumably due to PR #126247; pnkfelix thinks the above not worth further investigation
  • however, Kobzol has pointed out that the secondary regressions are significant, and has identified the root cause as PR #128104
  • we are in any case planning to revert the changes to dead code analysis (see PR #128404) which should address those regressions.
  • marked as triaged.

Nominated topics for discussions

  • Bjorn asks about \x01 in symbol names (on Zulip)
  • @pnkfelix asks Compiler team members to have a look and share an opinion about removing/muting some lints from the rustc benchmarks runs (see Zulip topic)

    Jakub says: Recently, our rustc-perf benchmarks started emitting many lints rustc-perf#1819. I don’t think that it’s good to disable these lints (or their output) during benchmarking, but over time if we accrue too many lints in the benchmarks, they will probably stop being realistic (most crates probably don’t generate thousands of lints/warnings). It hasn’t been three years yet (doc link), but I wonder if we should start thinking about doing another round of mass benchmark update, to make sure that they don’t generate so many warnings.

Nominated Issues

T-compiler

  • “Async drop codegen” rust#123948
    • Review was assigned last week to Vadim who commented (on github) not having enough context.
    • So, any taker for this review? 😅

RFC

  • No I-compiler-nominated RFCs this time.

Oldest PRs waiting for review

T-compiler

  • “Consistently use the highest bit of vector masks when converting to i1 vectors” rust#104693 (last review activity: 13 months ago)
    • cc @Wesley Wiser (this is a very old PR, unsure who was actually reviewing this, there are comments from many people)
  • “sess: stabilize -C stack-protector=all” rust#121742
    • Unsure if this blocked on this comment
    • cc @davidtwco @Nikita Popov
  • “Simplify match based on the cast result of IntToInt” rust#127324 (last review activity: about 24 days ago)
    • cc @cjgillot

Next week’s WG checkins

None

Next meetings' agenda draft: hackmd link