Evaluate Profile-Guided Optimization (PGO) and LLVM BOLT

[zamazan4ik]

Hi!

Recently I checked Profile-Guided Optimization (PGO) improvements on multiple projects. The results are here. LLVM-related results are here. According to the tests, PGO usually helps a lot with the compiler and compiler-like workloads (like static analysis). That's why I think trying to optimize Phasar with PGO can be a good idea.

I can suggest the following action points:

• Perform PGO benchmarks on Phasar. And if it shows improvements - add a note about possible improvements in Phasar's performance with PGO.
• Providing an easier way (e.g. a build option) to build scripts with PGO can be useful for the end-users and maintainers since they will be able to optimize Phasar according to their own workloads.
• Optimize pre-built Phasar binaries with PGO.

Maybe testing Post-Link Optimization techniques (like LLVM BOLT) would be interesting too but I recommend starting from the usual PGO.

[fabianbs96]

Hi @zamazan4ik, thank you for this suggestion. I aggree that we should make use of PGO at some time -- recently, we enabled LTO which already helps.

To automate the process of PGO-optimized builds, we need representative (real-world) analysis targets for all kinds of analyses that PhASAR supports. This seems very doable for our helper-analyses, such as alias-sets and ICFG. However, for our most performance-critical part, the IDESolver, this is more problematic, as it is implemented as template that gets instantiated for each analysis problem to solve.

We will add PGO experiments to our roadmap.