2020-10-21 Triage Log

A variety of changes, nothing particularly notable from a performance perspective. #77792 is an interesting win on migrating to tracing values rather than the older log formatting. #76859 is also interesting, starting out as a functional fix but appears to have been a performance win for incremental benchmarks in some cases as well.

Triage done by @simulacrum. Revision range: 06a079c43efb062e335e6e6c9dabd3c750619980..22e6b9c68941996daa45786b4145e6196e51f0f4

4 Regressions, 7 Improvements, 0 Mixed 2 of them in rollups

Regressions

#77755

#77873

  • Moderate regression in instruction counts (up to 2.7% on full builds of wg-grammar-check)
  • Functional change. Regression not significantly major to warrant a revert.

#78060

  • Moderate regression in instruction counts (up to 1.2% on incr-unchanged builds of clap-rs-check)
  • Rollup; likely due to BTreeMap or meta-prelude changes. Not worth deep investigation: regression is minor and interesting changes are likely not readily changeable.

#77250

  • Moderate regression in instruction counts (up to 3.1% on incr-unchanged builds of packed-simd-check)
  • Functional change that we definitely want, performance regression is not too significant.

Improvements

#76859

  • Very large improvement in instruction counts (up to -58.0% on incr-patched: Compiler new builds of regex-opt)
  • Correctness is the key here, but it seems like our hand-rolled incremental tracking was (at least on the perf benchmarks) worse anyway.

#77792

  • Moderate improvement in instruction counts (up to -1.7% on incr-unchanged builds of deeply-nested-async-opt)
  • Switching to tracing spans and tracing‘s “values” over string-formatting seems to have been a improvement here. Definitely an interesting result -- presumably tracing’s handling is somehow more visible to LLVM?

#77796

  • Moderate improvement in instruction counts (up to -1.4% on full builds of deeply-nested-async-check)
  • Seems like potentially a slight regression on wall times, but ultimately the change is good on its merits, so no action is taken at this time.

#77947

  • Large improvement in instruction counts (up to -7.4% on incr-unchanged builds of cranelift-codegen-debug)
  • Seems to be a mixed result on wall times though.

#77373

  • Moderate improvement in instruction counts (up to -1.6% on full builds of deeply-nested-async-check)
  • Wall times are less positive, but mostly on incremental. Not deemed a concern.

#77908

#78151

  • Large improvement in instruction counts (up to -7.6% on incr-patched: println builds of regression-31157-opt)

Nags requiring follow up