Tweak compiler options for the Release profile

[zamazan4ik]

Hi!

I noticed that current Cargo Release profile consists of enabling debug information. As far as I can understand, it was introduced in this commit 3 years ago. I guess that settings can be changed nowadays to something more optimal like building by default without debug info. This script also doesn't enable more optimizations.

Additionally, I suggest enabling optimizations like Link-Time Optimization (LTO) and codegen-units = 1 (CG1).

I've made quick local tests. My env: AMD Ryzen 9 5900x, Fedora 42, Rust 1.86, I used this build instruction.

Results:

• Current Release (with enabled debug info): 139 Mib, clean build time: 1m 03s
• Release (without debug): 16 Mib, clean build time: 59s
• Release (without debug) + Fat LTO + CG1: 13 Mib, clean build time: 1m 25s

Thank you.

[unexge]

Hey @zamazan4ik, thanks for the suggestions!

Regarding optimizations, we've experimented with similar options before, but it didn't improve the benchmark performance, and we didn't commit the change: #1078. We can take another look to see if we missed anything or if you saw any improvements with your experiments.

Regarding artifact size, we've an open PR that splits debug info from the Mountpoint binary, and it reduces the size of final artifacts similar to what you shared: #1370. Our current packaging script overrides the debug info configuration to only keep line numbers. With the proposed change, that info will be also stripped away and the debugging experience with bare release artifacts will decrease a bit unless the full debug info is present. That's why we still haven't made that change as well. Would like to hear more if you're trying to solve a specific problem.

[zamazan4ik]

Regarding optimizations, we've experimented with similar options before, but it didn't improve the benchmark performance, and we didn't commit the change: #1078. We can take another look to see if we missed anything or if you saw any improvements with your experiments.

That's strange. Enabled by you optimizations definitely should improve binary size. Regarding runtime performance - yep, it's possible that LTO in this case doesn't improve performance significantly but improving the binary size is also a good thing.

Regarding artifact size, we've an open PR that splits debug info from the Mountpoint binary, and it reduces the size of final artifacts similar to what you shared: #1370. Our current packaging script overrides the debug info configuration to only keep line numbers. With the proposed change, that info will be also stripped away and the debugging experience with bare release artifacts will decrease a bit unless the full debug info is present. That's why we still haven't made that change as well. Would like to hear more if you're trying to solve a specific problem.

Got it. We can try to add LTO and CG1 optimizations to this PR - it should improve the binary size even more.

[dannycjones]

Hey @zamazan4ik

Just to be clear: are you looking for only binary size improvements or is runtime performance also one of the goals of this proposed change?

[zamazan4ik]

Hi!

Just to be clear: are you looking for only binary size improvements or is runtime performance also one of the goals of this proposed change?

I am interested in both goals.