This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: changes.html to mention Pavel's contribution
> On Thu, 9 May 2002, Jan Hubicka wrote:
> > + <li>Pavel Nejedly, Charles University Prague, has contributed new arc
> > + profiler file format.
>
> "...a new..."
>
> Should "arc" be replaced by something more descriptive?
It is called arc profiling in the documentation. Perhaps we can call it
edge coverage profiler...
>
> > The new format is robust, diagnoze common
> > + mistakes resulting in nonsential profiles and allows overall statistics
> > + to be generated.
>
> "...diagnoses..."
>
> What kinds of mistakes? Specifically, *whose* mistakes? (GCC developer,
> GCC user,...?)
>
> What do you mean by "nonsential"?
When you compile your program with instrumentation, run it, modify, recompile,
run it and use profile, the profile is bogus, since it gets combined the
two runs of different program. New format stores the checksum of CFG so this
is reported to user.
This is very common mistake to do and it is not diagnozed at all currently.
GCC simply fetches wrong profile, produces slower code and user is disappointed.
>
> > This helps optimizers to identify hot spots of a
> > + program globally replacing the old intraprocedural scheme resulting in
>
> "intra-procedural", I believe.
>
> > + smaller code generated. Note that <tt>gcov</tt> tool from the older GCC
>
> Omit "generated" here and use <code> instead of <tt>.
>
> "Note that gcov from..." or "the gcov tool from".
>
> "from older versions of GCC..."
>
> > + versions will not be able to parse the profiles generated by the new GCC
> > + and vice versa.</li>
>
> "...by GCC 3.1".
GCC 3.2 I guess.
Thanks!
Honza
>
> Gerald
> --
> Gerald "Jerry" pfeifer@dbai.tuwien.ac.at http://www.dbai.tuwien.ac.at/~pfeifer/