This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH, middle-end] Switch initializations conversion
On Tue, Aug 28, 2007 at 11:23:57AM +0200, Richard Guenther wrote:
> On 8/28/07, Martin Jambor <jamborm@matfyz.cz> wrote:
> > For further information, please see the previous posting (at
> > http://gcc.gnu.org/ml/gcc-patches/2007-08/msg01638.html)
> >
> > As promised, I have collected at least some the information about how
> > much conversions are performed by the pass. Perhaps not surprisingly I
> > have counted the number of converted switches when bootstrapping gcc
> > with respect to the MAX_RANGE_BRANCH_RATIO constant. The results are
> > summarised in the following table:
> >
> > MAX_RANGE_BRANCH_RATIO Converted switches
> > --------------------------------------------------
> > 4 96
> > 8 116
> > 16 124
> >
> > I have then decided to leave the constant as it was (i.e. 8). The
> > time overhead was not really measurable (i.e. All compilations took
> > about 247 minutes of user time and the unpatched version did so too).
>
> Did you look at what kind of switches that are? The numbers above are
> certainly higher than I expected ;)
Frankly, I was a bit surprised too. But most of the switches got
converted (and reported) twice, in stage two and again in stage three.
At this point I was really interested in what effect the value of
MAX_RANGE_BRANCH_RATIO had. The bootstraps take a lot of time and so
these numbers are all I've got now. However, you are right, I should
be more curious. I'll be away for four days and will look into it this
weekend. I also intend to (at least statistically) analyse reasons for
rejection of candidates so I can answer Ken's email.
Martin