I am studying the effectiveness of Profile Feedback Optimizations in LLVM. In particular I am interested on how effective it is on Spec2006.

I did a comparison between LLVM, Open64 and GCC compilers using Spec2006 as a benchmark. LLVM came last in the performance. I did not use PFO for LLVM. I was wondering if any of you have used PFO in Spec2006 and can give me some idea about its effectiveness. I also like to know how and with what flags PFO works correctly and produces the optimal result.

Thanks in advance,
Reza Yazdani