Re: Thread-safety issues in QtHandles

[Søren Hauberg]

tir, 15 11 2011 kl. 11:26 +0000, skrev Michael Goffioul:
> On Tue, Nov 15, 2011 at 10:54 AM, Michael Goffioul
> <address@hidden> wrote:
> >> This is indeed a large slow-down, I'll test it on my platform as well,
> >> for comparison.
> >
> > For comparison, on my Linux platform, Intel E5300 2.6GHz 2GB RAM, I
> > ran the test above with the patch applied (on
> > http://hg.savannah.gnu.org/hgweb/octave/rev/69afe121f07e), once with
> > --enable-atomic-refcount, once with --disable-atomic-refcount, octave
> > compiled with "-g -O2", the execution time is as follows:
> > - ./configure --enable-atomic-refcount: ~13s
> > - ./configure --disable-atomic-refcount: ~10.7s
> 
> And when the patch is not applied at all (that is, a pristine
> http://hg.savannah.gnu.org/hgweb/octave/rev/69afe121f07e), then the
> execution time is ~10.3s. I'm not sure how you managed to get such a
> large difference on your platform. Just for completeness, I'm using
> Fedora Core 12 with gcc-4.4.4.

Hmm, I tried recompiling and explicitly setting setting CXXFLAGS="-g
-O2", which gave me

- ./configure --enable-atomic-refcount: ~12.5s
- ./configure --disable-atomic-refcount: ~9.7s

The previous number I gave for run-time when not using the atomic
refcount was from my installed Octave, which is a fairly recent Hg
checkout:

        changeset:   13706:5cd1193ac1c4
        tag:         tip
        user:        John W. Eaton <address@hidden>
        date:        Fri Oct 14 19:42:33 2011 -0400
        summary:     allow assignment of class objects to indexed
        structure arrays
        
So, it seems Octave has gotten a bit slower since that checkout, which
explains some of the slowdown I saw.

Søren