SIMD-less render optimizations

Daniel Amelang daniel.amelang at
Wed Apr 18 16:31:09 PDT 2007

On 4/18/07, Keith Packard <keithp at> wrote:
> On Tue, 2007-04-17 at 22:45 -0700, Daniel Amelang wrote:
> >  Yes,
> > you may be 1 bit off (AFAIK) in the result here in some cases, but
> > we're going to chop off the last 2-3 bits anyway when we convert to
> > 565,
> We've really tried to avoid generating different compositing results in
> different rendering paths...

Which is think is quite laudable. Really. I didn't expect this to go
into mainline, honestly. One thing I was investigating during this
exercise was the value of introducing a quality/performance switch in
the code generator to allow people to specify what they're most
interested in. A neo1974 running an interactive program has very
different priorities than a Dell workstation that's generating a PDF
document (think cairo pixman fallback here).

That said, I don't expect to incorporate this switch anytime soon, so
don't hold your breath :) I still need to get _something_ generated
for all the render operations and parameters. Right now I've only got
a few covered.


More information about the xorg mailing list