2020-11-03 Triage Log
A number of improvements on various benchmarks. The most notable news this week in compiler performance is the progress on instruction metric collection on a per-query level; see measureme#143 for the latest.
Otherwise, this week was an excellent one for performance (though mostly on stress tests rather than commonly seen code).
Triage done by @simulacrum. Revision range: 824f900a96d752da2d882863c65f9736e5f2b347..5cdf5b882da9e8b7c73b5cadeb7745cb68f6ff63
0 Regressions, 5 Improvements, 0 Mixed
Improvements
#78323
- Slight improvement in instruction counts (up to -1.3% on
incr-unchanged builds of packed-simd-check) - Possibly within noise; unclear.
#78508
- Moderate improvement in instruction counts (up to -2.0% on
incr-unchanged builds of packed-simd-check)
#78432
- Large improvement in instruction counts (up to -5.7% on
full builds of match-stress-enum-check) - An unexpected improvement for a seemingly bugfix PR; would be good to verify this is not an unintentional behavior change (nag left).
#78553
- Very large improvement in instruction counts (up to -10.1% on
full builds of match-stress-enum-check)
#78448
- Very large improvement in instruction counts (up to -95.4% on
full builds of externs-debug) - Notable case of adding a new benchmark to perf; this is much appreciated and illustrates that perf does not yet have full coverage of Rust code (though this is not really expected either, though is always a goal).
#78430
- Very large improvement in instruction counts (up to -23.6% on
incr-patched: println builds of unicode_normalization-check) - Fairly large refactor to the match checking infrastructure, with a correspondigly large performance improvement. There does appear to be a slight regression on #58319, but this is in the “Improvements” category since it seem categorically a win.
Nags requiring follow up
Compiler team attention requested: