[Intel-gfx] [ANNOUNCE] xf86-video-intel 2.11.901

[Clemens Eisserer]

Hi Chris,

>  $ git checkout 2.11.901
>  $ git revert f74b3f82
Ah thanks, didn't know the checkout thing.

Hmm, if I revert that commit it no longer builds:
  CC     i915_render.lo
i915_render.c: In function ‘i915_prepare_composite’:
i915_render.c:836: error: ‘intel_screen_private’ has no member named
‘render_current_dest’

- Clemens

> Give-or-take that it is dubious that we can do a simple coverage based
> accumulation of subpixels trapezoids within the specification of
> PolyModeImprecise, benchmarking shows a small performance win for most
> cairo traces [the improvements matching the elimination of
> pixman_image_rasterize_trapezoid() and co], but in a few cases it can be
> twice as slow.

Whatever the result will look like, I am really curious - as some
workloads show a lot of time spent in rasterize_edge().

> So I think it is back to a span renderer -- even if I just
> trade GPU time for CPU time, that should be a win for application
> performance.

just reminds me at the commit "OMG! spans!" ;)