Using profile-guided optimization with shared libraries and multiple executables

[Alan Lehotsky]

I tried building the entire library with -ipo (it SEGVs on execution). I also selectively built the 3 sources that were most heavily represented in caliper scgprof, and that ran perhaps 1% faster on one of my benchmarks and 2.5% SLOWER on the other one.

As for the documentation, I was using the a.06.15 online users guide. I can't find ANY mention of unique or per-process in there, or of fdm, for that matter.

The 22% improvement exceeds my goal of 20%, so I'm satisfied at this point.