2021-09-14 Triage Log

Fairly busy week, with some large improvements on several benchmarks. Several larger rollups landed, in part due to recovery from a temporary CI outage, which has complicated some of the performance monitoring work. These should, however, now be resolved.

Triage done by @simulacrum. Revision range: 69c4aa2901ffadf69deaf91b2f90604bcbc2eb36..9f85cd6f2ab2769c16e89dcdddb3e11d9736b351

2 Regressions, 2 Improvements, 5 Mixed; 2 of them in rollups

31 comparisons made in total

Regressions

Encode spans relative to the enclosing item #84373

  • Large regression in instruction counts (up to 2.1% on incr-unchanged builds of tuple-stress)
  • Regressions are much smaller on “full” benchmarks (<0.5%). This support has also landed gated behind a -Z flag, so the incremental cost is being paid without the possible wins.

Rollup of 7 pull requests #88881

  • Large regression in instruction counts (up to 2.1% on full builds of inflate)
  • No clear cause. Investigation is partially ongoing, but may warrant an assignee -- there's several possible candidates.

Improvements

Rollup of 10 pull requests #88857

  • Very large improvement in instruction counts (up to -45.0% on full builds of coercions)
  • Unexpected improvement, though certainly nice to see. Suspected cause is #88147, but this has not been verified.

Use FxHashSet instead of Vec for well formed tys #88771

Mixed

Split rustc_mir #80522

  • Large improvement in instruction counts (up to -2.0% on full builds of deeply-nested-async)
  • Moderate regression in instruction counts (up to 0.8% on full builds of await-call-tree)
  • Mixed results. Noted in a comment that this also was a slight regression in bootstrap time as measured by perf.rust-lang.org, which seems unfortunate and may merit some investigation.

Rollup of 15 pull requests #88824

  • Moderate improvement in instruction counts (up to -0.7% on incr-patched: println builds of html5ever)
  • Moderate regression in instruction counts (up to 1.0% on incr-unchanged builds of derive)
  • Report seems a little bit close to noise, and is definitely mixed. No clear cause, but also a fairly large rollup.

Refactor query forcing #78780

  • Moderate improvement in instruction counts (up to -0.9% on incr-full builds of unused-warnings)
  • Small regression in instruction counts (up to 1.3% on incr-patched: println builds of coercions)
  • Solid improvement in rustc_query_impl compile times (8%). Overall looks like results are overall more of an improvement than a regression.

Update LLVM submodule #88765

  • Moderate improvement in instruction counts (up to -0.7% on full builds of ctfe-stress-4)
  • Moderate regression in instruction counts (up to 0.7% on incr-unchanged builds of ripgrep)
  • Fairly mixed results, no clear picture either way.

Untriaged Pull Requests