[Intel-gfx] [PATCH] Fall back to fb pixmaps for backing glyph pictures

[Kristian Høgsberg]

On Tue, Jun 9, 2009 at 4:48 PM, Eric Anholt<eric at anholt.net> wrote:
> On Tue, 2009-06-09 at 13:14 -0700, Keith Packard wrote:
>> On Tue, 2009-06-09 at 11:01 -0700, Eric Anholt wrote:
>>
>> > The firefox trace should have involved moving things through the glyph
>> > cache, yet didn't show any performance difference.  I'd love to see a
>> > cairo-trace showing that this patch breaks things, though.
>>
>> Are you concerned that this patch will cause mis-rendering? Or just that
>> we need a better benchmark to see the effect in a more 'real'
>> environment?
>
> You seemed to be concerned that there would be a performance problem for
>>32x32 due to hitting the upload path.  The firefox trace should hit the
> upload path AFAIK.  So let's get a trace showing a problem before trying
> to fix a problem.

If glyphs bigger than 32x32 turns out to trigger a performance cliff,
we can change the check to

    if (usage == CREATE_PIXMAP_USAGE_GLYPH_PICTURE &&
        width <= 32 && height <= 32)
       return fbCreatePixmap (screen, w, h, depth, usage);

So that glyphs that we can't cache at least get their own uxa pixmap
so we don't hit the sw compositing path.  Not ideal, but a simple fix.

Kristian