Improve pgo builds

[Ddystopia]

• feat: allow local absolute paths
• feat: allow measuring against multiple crates

---

I'm not as familiar with PGO, do I understand correctly that by running more commands, more measurements do accumulate and then get merged?

[Ddystopia]

One possible improvement would be to report error in case the string passed to --pgo is an absolute path but doesn't exists.

[ChayimFriedman2]

Do you actually need this, or just thought this is nice to have?

[Ddystopia]

Well I used it to compile RA to better suit my projects. Does this count as "I actually need this"?

[ChayimFriedman2]

Was there a measurable benefit?

[Ddystopia]

I didn't benchmark it, I'll try to do this in a span of two weeks

[ChayimFriedman2]

The reason I ask is that I suspect there won't be any perf benefit from customizing to a project (all rust projects look the same, when looking at the level of branch and cache predictions), therefore I don't want to accept this just because.

[Ddystopia]

Yeah, I get the point, but my objective is that my projects are quite different as it is either wasm front-ends or embedded software with lots of automatically generated FFI bindings and ra often just dies. I would try to benchmark it against stock ra if there would be some significant benefit.

[lnicola]

See #19585 (comment), or better, run your own numbers.

[ChayimFriedman2]

@Ddystopia Did you check that?

[Ddystopia]

@ChayimFriedman2

Hi, I did not, I am quite busy lately 😅 . I'll probably get around to this during this weekend. Note my previous comment from a week ago:

I didn't benchmark it, I'll try to do this in a span of two weeks

[Ddystopia]

I ran tests, there seem to be small gains, not too huge though. It also has some uncaught type error. I'll be using it locally, together with target-cpu=native (I didn't look into how to enable it during compilation with xtastk and will do it later, local-only). Those are some seconds so I'll take them, if you think it's not worth it I'm fine if you close the PR.

simple-fw.txt
complex-fw.txt
simple-backend.txt
complex-backend.txt

[lnicola]

Is simple our release binary (PGOed on clap) and complex PGOed on specific project, plus target-cpu=native? I see a difference on fw, but not on backend.

[Ddystopia]

Both are without target-cpu=native, simple is master with pgo on rust-analyzer (I thought you used it). And yeah I also only see the difference on fw.