We have been testing profile-guided builds for a database project and found that it was significantly improving the performance of particular parts of the systems (in our case, the aggregation).
As we would like to remove the required overheads for PGO builds, we are curious if there are principled ways to analyze the changes the PGO build did. Principled in a sense that it does not mean just dumping the compiler internals (we are aware of -fdump-ipa-cgraph).
Does anything helpful in that direction exist?