2021-08-11 Triage Log

A quiet week for performance. Changes, both positive and negative, tended to be isolated to few benchmarks.

2 Regressions, 1 Improvements, 0 Mixed; 0 of them in rollups

Use zeroed allocations in the mir interpreter instead eagerly touching the memory #87777

Moderate regression in instruction counts (up to 1.7% on incr-unchanged builds of ctfe-stress-4-check)
The motivation for this PR was to reduce the number of page faults, for the same family of ctfe-stress-4 benchmarks.
So, perf runs were done on the PR itself, to illustrate the motivated change.
The number of page faults did decrease (by up to 25%).
The regression with respect to instruction counts appears isolated to just the ctfe-stress-4 family; the page fault reduction outweighs instruction count hit.
(Unfortunately the page fault decrease did not yield a corresponding improvement to reported wall-clock or task-clock times.)

Hide allocator details from TryReserveError #87408

Moderate regression in instruction counts (up to 4.2% on full builds of html5ever-opt)
The regression with respect to instruction counts appears isolated to just the html5ever-opt benchmark; all other reported changes pale in significance.
Potentially significant: the max-rss for html5ever-opt also regressed here, by 8%.
(Skimming over the PR itself, I do not see any obvious reason for this significant of a regression to max-rss.)
Some other benchmarks also regressed with respect to max-rss, but none so significantly.
Left a comment on the PR as a heads up.

#[inline] slice::Iter::advance_by #87736

Moderate improvement in instruction counts (up to -1.7% on full builds of regex-opt)
This was put in to address the regression previously injected by PR #87387.