Runtime statistics

[Crazy Diamond]

Does GCC have any facilities to collect runtime information 
(such as branching statistics) that can be applied to future compilations?  
If not, has anyone expressed any interest in adding it?  This is something 
that could allow us to squeeze a few % more speed from executables.  The 
best thing to do would be to make this part of the debug/symbol information, so 
that it can be stripped from final compiles (once the info has been used).  
We don't want .exe's being larger than they have to 
be.