[cairo] [PATCH] Altivec optimizations

Baz brian.ewins at gmail.com
Wed Mar 7 07:55:01 PST 2007


On 07/03/07, Baz <brian.ewins at gmail.com> wrote:
> On 07/03/07, Luca Barbato <lu_zero at gentoo.org> wrote:
> > Luca Barbato wrote:
> > > I completed the fbcompose part the regtest are consistent, hopefully
> > > I'll start adding the pict ops later in the next week.
> > >
> >
> > Patch ping, anybody could test&commit/report back ?
> >
>
> Yes, I'll do this unless anyone else gets in first - it'll be tomorrow
> though, big footy match tonight. I'll be testing on OSX 10.4/ppc.
> Anyone with 10.3.9 ppc, or some PPC Linux, jump in too.

Actually, if I can dig out the disks, I'll set that box up as
gentoo/10.3.9 dual boot. I only use  the laptop for testing these
days. Reading the patch:

+#if defined(__GNUC__) && (__GNUC__ < 3 || (__GNUC__ == 3 &&
__GNUC_MINOR__ < 4))
+#error "Need GCC >= 3.4 for sane altivec support"
+#endif

This jumps out at me because Mozilla use gcc 3.3 for Mac PPC builds,
but only because they want to support releases earlier than 10.3.9. So
this is going to mean no altivec in firefox on older macs. What goes
wrong with 3.3?

Also, anyone else trying this out - cairo 1.4 has been crashing
Xquartz fairly regularly in the extend-reflect test, in
fbCompositeSrc_8888x8888. Behdad's seen similar (but rarer) crashes in
fbCompositeSrc_8888x8888mmx in the same test, so a crash there in
fbCompositeSrc_8888x8888vmx may not be a problem with the patch. Code
seems to work reliably when EXTEND_REFLECT is changed to EXTEND_REPEAT
in that test.

Cheers,
Baz


More information about the cairo mailing list