Another quiet week with very small changes to compiler performance.
Triage done by @rylev. Revision range: 5258a74c887f8ae14717e1f98b652b470877ce4e..6df26f897cffb2d86880544bb451c6b5f8509b2d
1 Regressions, 0 Improvements, 1 Mixed 0 of them in rollups
Fix lookahead with None-delimited group #84130
- Moderate regression in instruction counts (up to 1.4% on
incr-unchanged
builds ofdeep-vector-check
) - While this change fixes a bug, it does introduce moderate regressions in runtime performance on some benchmarks and a fairly large regression on the compilation of the rustc_parse crate itself.
- We will follow up on this particularly in the matter of
rustc_parse
's compilation times regressing.
Add some #[inline(always)] to arithmetic methods of integers #84061
- Moderate regression in instruction counts (up to 1.7% on
full
builds ofregex-debug
) - Moderate improvement in instruction counts (up to -1.3% on
incr-unchanged
builds ofcargo-check
) - This change inlines a bunch of code that might not have been previously. It's not really a surprise that codegen now is taking longer to churn through all those inline calls.
- While this tradeoff on compile times for runtime performance is likely worth it, we are checking in with the PR author and reviewer to make sure this is discussed.
- This week's follow ups: