A mixed week with improvements still outweighing regressions. Perhaps the biggest highlight was the move to compiling rustc crates with the initial-exec TLS model which results in fewer calls to _tls_get_addr
and thus faster compile times.
Triage done by @rylev. Revision range: 5cdf5b882da9e8b7c73b5cadeb7745cb68f6ff63..cf9cf7c923eb01146971429044f216a3ca905e06
1 Regressions, 2 Improvements, 2 Mixed
- Small regression in instruction counts (up to 1.2% on
full
builds ofdeeply-nested-async-check
) - This might be noise as this only affects one benchmark negatively, and that benchmark tends to be on the noisier side.
- Moderate improvement in instruction counts (up to -1.3% on
incr-patched: new row
builds oftuple-stress-opt
)
- Large improvement in instruction counts (up to -7.2% on
incr-full
builds ofwebrender-wrench-check
) - This change may produce similar performance gains in related tooling such as rustdoc and clippy.
- Large improvement in instruction counts (up to -5.0% on
incr-unchanged
builds ofdeeply-nested-async-check
) - Moderate regression in instruction counts (up to 1.3% on
full
builds ofctfe-stress-4-check
)
- Very large improvement in instruction counts (up to -26.0% on
incr-unchanged
builds ofdeeply-nested-async-opt
) - Moderate regression in instruction counts (up to 3.8% on
full
builds ofctfe-stress-4-check
) - This change is a revert of a previous change, and at least one user was reporting massive performance gains.
The compiler team is once again requested to look into: