[Intel-gfx] [PATCH] Revert "drm/i915: Kill GTT mappings when moving from GTT domain"

[Chris Wilson]

On Fri, 17 Jun 2011 12:06:54 -0700, Eric Anholt <eric at anholt.net> wrote:
> On Wed, 15 Jun 2011 17:03:58 +0100, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > Moving back to LLC+semaphores (2.6.39-rc2+), firefox-talos-gfx:
> > xlib: 4.473
> > gl:  20.753
> > 
> > applying the patch:
> > xlib: 4.472
> > gl:  20.824
> > 
> > I'm just not reproducing the same issue you are seeing. Are you using a
> > standard distro Kconfig, or if not, can you send me yours?
> > -Chris
> 
> http://people.freedesktop.org/~anholt/dotconfig
> 
> Pushed my kernel tree to "gtt-revert" branch.

Thanks, I'm testing with those on my SNB desktop, still only see around
a 5% hit for firefox-talos-gfx.

I still think this is optimising for bad behaviour of the client, and
papering over two slow linear lookups in the kernel. [One for find_vma
which is exacerbated by the cached vma on the bo and the other checking
whether the GTT aperture is RAM.]

Looking at what cairo-gl is doing, the root cause of why we are
repeatedly rewriting textures is:

commit 3b1c0a4bd66660780095e6016e3db451f34503a3
Author: Benjamin Otte <otte at redhat.com>
Date:   Fri May 14 15:56:17 2010 +0200

    fallback: Remove span renderer paths
    
    Those paths were broken, as they didn't properly translate the polygon
    to the destination size. And rather than adding lots of code that allows
    translation, it's easier to just delete this code.
    
    Note that the only user of the code was the GL backend anyway.

i.e. cairo-gl is no longer using spans for firefox-talos-gfx where
unaligned clipping dominates.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre