This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Suggested addition to documentation
- To: jbuck at synopsys dot COM (Joe Buck)
- Subject: Re: Suggested addition to documentation
- From: Brad Lucier <lucier at math dot purdue dot edu>
- Date: Tue, 22 May 2001 14:48:44 -0500 (EST)
- Cc: lucier at math dot purdue dot edu (Brad Lucier), gcc at gcc dot gnu dot org
Joe Buch writes:
>
> Brad Lucier writes:
>
> > I couldn't find the best place in the documentation to put this note.
> >
> > This is true for 3.0; I believe there have been three questions
> > about this performance issue since last summer posted to the
> > gcc mail list.
> >
> > Note: If you write a program using computed gotos, a GCC extension,
> > and you compile the program with optimization level @samp{-O2} or higher,
> > you may get better runtime performance if you disable the global common
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> > subexpression elmination (gcse) pass by adding the @samp{-fno-gcse} option'
> > to the command line.
>
> Do we have enough data to be confident that this is true in general?
> Clearly it is true for your applications, and thanks for all the benchmark
> studies you ran. But will it be true for (almost) anyone who uses
> computed gotos?
>
In my note I suggested saying "you may get better runtime performance",
which means also that "you may not get better runtime performance".
I believe that the statement is true as I presented it, i.e., this is
something to try if you have a performance problem.
I suppose that people who don't have a problem don't write to the
gcc mail list :-), but there have been several people writing
codes using computed gotos have noticed this performance problem
with 3.0 versus 2.95.*. See, e.g.:
http://gcc.gnu.org/ml/gcc/2001-03/msg00296.html
and other messages with this subject line in March,
http://gcc.gnu.org/ml/gcc/2001-01/msg00990.html
and other messages with this subject line in January; and, of
course, there are all of my codes ;-).
Brad