[PATCH] profile feedback: -fprofile-use= and -fprofile-correction, correctness fixes and option semantic changes.

[Jan Hubicka]

> 
> We alread have -Wcoverage-mismatch which will instead of aborting on
> old profile data with new source just ignore those cases it cannot match.
> So this new code would probably fit in the same place.

There are two things: ignoring profile instead of erroring when the
profile is obviously not matching the file (ie function checksums does
not match, that is what -Wcoverage-mismatch is good for, i.e. when you
changed some part of program but you don't want to worry) or ignoring
cases where the read back profile does not lead to consistent solution
while computing the optimized out edge counts (that is the functions
appear to be same, but the data measured are not 100% valid due to races).

Honza
> 
> Richard.