[Intel-gfx] [PATCH 0/4] Attempt to re-enable FBC on gen3

[Alexander Lam]

On Fri, Apr 23, 2010 at 11:17 AM, Adam Jackson <ajax at redhat.com> wrote:
> Disclaimer: I haven't tested this extensively, it just seems logical, and I
> really hate to see us lose FBC on gen3 since that's the family being used in
> low-wattage devices.  Wider testing would be greatly appreciated.

Seems to work here on a 945GSE (Acer Aspire One) and save a bit of
power (like .2W)

> The only thing I really don't like about this is how we hit every CRTC on
> resume trying to enable FBC.  But that's sort of a problem in the normal case
> anyway.  On pre-gen4, we can only compress plane A, so we hardwire that to
> pipe B since LVDS is limited to pipe B.  Ideally, you'd like to compress
> whichever CRTC has more pixels, assuming they're both mostly static. It's
> really awkward to do that in the current one-at-a-time CRTC setup kind of
> world though.
>
> On gen4 and later we can compress either plane, so it's a little messier;
> we'll compress whichever CRTC gets set up last, before suspend, but then after
> suspend we'll compress the highest-numbered CRTC.  Either way, when multiple
> CRTCs are active, we're already sometimes compressing a suboptimal plane, so
> making more ways for that to happen isn't a huge deal.
>
> (Then, of course, you'd like to switch which pipe you compress to the one with
> the more static image, if they have different update rates.  But that's way
> into diminishing returns territory.)
>
> - ajax


-- 
Alexander Lam