Hi, thanks a lot for the PR!

Regarding the LLVM installation approach, have you tried that BOLT indeed works, both for instrumentation and then for generating samples? I haven't tried with LLVM 18, but with previous versions, the installed bolt-XY package from Ubuntu haven't been functioning properly (#31). It was detected, but failed during instrumentation/sample gathering.

Yeah the BOLT also works well. You can test it locally using cargo-pgo itself in these steps:

Firstly, build the image and get into the container:

git clone https://github.com/bioinformatist/cargo-pgo.git && cd cargo-pgo
docker build -t pgo .
docker run --entrypoint bash -v $(pwd):/workdir --rm -it pgo

Then in the container:

cargo pgo build
/workdir/target/x86_64-unknown-linux-gnu/release/cargo-pgo pgo info
cargo pgo bolt build --with-pgo
/workdir/target/x86_64-unknown-linux-gnu/release/cargo-pgo-bolt-instrumented pgo info
cargo pgo bolt optimize --with-pgo

Below shows the result:

I believe the issue lies in the suffix of the binaries (like the -18 in merge-fdata-18) and the incorrect location of the linking library (.a), which caused the bolt-XY package from Ubuntu to not work. The soft linking (ln -s) in Dockerfile can fix it. While compiling LLVM from the source code is a more elegant way, it spends more time, which maybe not ideal for CI.

Got that! I'll finish it ASAP.

No need to hurry ;)