Use PGO on macOS x64 and arm64 builds

[zamazan4ik]

• related to Apply Profile-guided optimization to improve performance #9412 (comment)

[lnicola]

You can test this by adding on: pull_request and making it fail like in https://github.com/rust-lang/rust-analyzer/pull/19602/files.

[lnicola]

I guess it's easier to merge and revert if necessary.

[zamazan4ik]

So, it doesn't work: https://github.com/rust-lang/rust-analyzer/actions/runs/14510906031/job/40709149737

As far as I understand, this due to that we use x86-64 macOS image instead of ARM image: https://github.com/actions/runner-images/blob/main/images/macos/macos-14-arm64-Readme.md . Are we allowed to use ARM macOS machines for RA? If yes, we can try to change the image and try once again.

[lnicola]

#19614

You can already test it in the nightly.

[lnicola]

Ha! I just saw someone link to https://www.youtube.com/watch?v=_EpALMNXM24.

[zamazan4ik]

PGO world is not so big nowadays, yeah :D

[lnicola]

And look:

[zamazan4ik]

As I did previously in #19585 (comment) for Windows - now we test PGO vs non-PGO RA for macOS!

Environment:

• OS: macOS 15.4 with the latest stable updates
• Hardware: Macbook M1 Pro (ARM CPU)
• Rust-Analyzer versions: PGO version from https://github.com/rust-lang/rust-analyzer/actions/runs/14511602298 , non-PGO version from https://github.com/rust-lang/rust-analyzer/actions/runs/14505288557 (the latest build before PGO enable for macOS, no other performance-related changes were committed since this so we can use this artifact)
• Latest at the moment Rust Analyzer sources

For benchmark purposes, I used rust-analyzer analysis-stats --run-all-ide-things <rust-analyzer-checkout-directory> command. Measurements are done multiple times, in different orders, etc. First runs are skipped to avoid any first-time-run fluctuations (like "XProtectService" or other "magic" stuff). The results are pretty stable across runs.

Results (the Total number at the end of the execution. Providing the full log will be too lengthy):

• No PGO: 140s
• PGO: 110s

So we can conclude that PGO for RA works well for macOS platform too.

I didn't test x86 macOS version (I am a bit lazy to run it under Rosetta2) but I expect exactly the same performance wins for that platform too. As far as I understand, RA in CI uses x86 macOS images to build x86 macOS RA version, so no PGO issues due to cross-compilation should appear. If anyone wants to test it and report the results back - it would be nice :)