2020 Triage Logs

See the README for details on how these logs are collated. They are in reverse chronological order.

2020-07-07

Triage done by njn. Latest revision: 0c03aee8b81185d65b5821518661c30ecdb42de5. One unimportant regression, on a rollup; six improvements, two on rollups.

Regressions

Rollup of 12 pull requests #74073 (instructions): Up to 1.9% losses, but only on the synthetic wf-projection-stress-65510 benchmark, which doesn't matter that much.

Improvements

Serialize all foreign SourceFiles into proc-macro crate metadata #73706 (instructions): Up to 7.5% wins across numerous benchmarks, mostly the shorter-running ones.
Handle inactive enum variants in MaybeUninitializedPlaces #73879 (instructions): Up to 7.1% wins across numerous benchmarks, mostly on opt builds. (Plus one 21% improvement, but that benchmark has high variability.)
Rollup of 16 pull requests #73950 (instructions): Up to 1.9% wins across numerous benchmarks.
Avoid unwrap_or_else in RawVec::allocate_in. #73882 (instructions): Up to 1.2% wins across a few benchmarks, mostly on debug builds. Maybe #73345 or #73569 is responsible?
Remove TypeckTables::empty(None) and make hir_owner non-optional. #73751 (instructions): Up to 3.8% wins, mostly on unused-warnings.
Rollup of 17 pull requests #73924 (instructions): Up to 1.9% wins, mostly on clap-rs. Unclear which PR is responsible.

2020-06-30

Triage done by njn. Latest revision: 0ca7f74dbd23a3e8ec491cd3438f490a3ac22741. Three regressions, two of them on rollups; two improvements, one on a rollup.

Regressions

Rollup of 13 pull requests (instructions): Up to 1.7% losses on many benchmarks, affecting incr-unchanged runs the most. Might be due to #73102 or #73597.
Rollup of 11 pull requests (instructions): Up to 2.1% losses on a few benchmarks, mostly on incr-unchanged runs.
Always capture tokens for macro_rules! arguments (instructions, max-rss): Up to 6.1% losses on the artificial deep-vector benchmark, and memory increases on the same benchmark.

Improvements

Rollup of 9 pull requests (instructions): Up to 13.6% wins on wg-grammar, tiny wins on a couple of other benchmarks.
Revert PR #72389 - “Explain move errors that occur due to method calls involving self” (instructions): Up to 3.1% wins on a few benchmarks.

2020-06-23

Triage done by njn. Latest revision: 6bb3dbfc6c6d8992d08431f320ba296a0c2f7498. Lots of improvements this week!

Having done this for a few weeks now, I see that close to half of the PRs with significant performance effects are rollups.

Regressions

None.

Improvements

Rollup of 9 pull requests (instructions, max-rss): Up to 33.6% instruction wins on a stress test, and up to 10.3% instructions wins on several real-world benchmarks. Also a max-rss win. #72788 may be the cause?
Cache flags and escaping vars for predicates (instructions): Up to 4.9% wins across numerous benchmarks.
store ObligationCause on the heap (instructions): Up to 3.5% wins across numerous benchmarks.
Upgrade Chalk (instructions): Up to 1.5% wins across numerous benchmarks.
Rollup of 13 pull requests (instructions): Up to 1.5% wins across numerous benchmarks.
Rollup of 16 pull requests (instructions): Up to 3.4% wins on wg-grammar, little change elsewhere.

2020-06-16

Triage done by njn.

Some bustage involving Cargo meant that some merges in this period didn't get measured.

Regressions

Rollup of 10 pull requests (instructions): Regressed up to 3.5% on multiple benchmarks. Maybe #71824 is the cause?
Disable the SimplifyArmIdentity pass (instructions): A small regression on a couple of benchmarks from disabling a buggy pass, which may be re-enabled in the future?
A small regression occurred for syn-opt, but I can‘t tell which PR was responsible because the improvement occurred in one of the merges that wasn’t measured due to the Cargo bustage.

Improvements

Some huge improvements occurred for clap-rs-opt, but I can‘t tell which PR was responsible because the improvement occurred in one of the merges that wasn’t measured due to the Cargo bustage.
Rollup of 8 pull requests (instructions): A small improvement on many-assoc-items, and a tiny regression on a couple of other benchmarks.

2020-06-09

Triage done by njn.

Regressions

Update LLVM submodule to include lld NOLOAD fix (instructions): A small regression in debug builds due to LLVM. Pulling in this LLVM change may fix it.
Rollup of 5 pull requests (instructions): Possibly due to #72989.

Improvements

None.

2020-06-02

Triage done by njn.

Regressions

Nothing of note!

Improvements

perf: Revert accidental inclusion of a part of #69218 (instructions)
Pass more Copy types by value. (instructions): This fixes part of the regression from #72055 from last week.
Rollup of 9 pull requests (instructions)

2020-05-26

Triage done by njn.

Regressions

Update to LLVM 10 (instructions): Lots of grumbling about this. wg-prioritization will consider at their next meeting?
Intern predicates (instructions): @lcnr has promised a follow-up to at least partly fix it, currently in #72494.

Improvements