[Intel-gfx] [PATCH] drm/i915: Android sync points for i915 v2

Jesse Barnes jbarnes at virtuousgeek.org
Tue Aug 5 18:32:04 CEST 2014


On Tue, 5 Aug 2014 18:08:16 +0200
Daniel Vetter <daniel at ffwll.ch> wrote:

> On Tue, Aug 5, 2014 at 6:05 PM, Jesse Barnes <jbarnes at virtuousgeek.org> wrote:
> > But yes, I want the Android guys to try this out too.  I've already
> > pinged them internally to check things out.  Probably the biggest
> > remaining opens there would be having a timeline for the display side
> > of things too that covers vblank and page flip events as fences in a
> > separate namespace.
> >
> > Which makes me think going back to just using the Android structs
> > directly might be easier...
> 
> Why that? They should really just provide the userspace interface on top ...

Because the internal implementation has the nice timeline/sync pt/fence
abstraction.  It just fits nicely, conceptually, with the various
timelines we have in gfx.

For fences, I guess we'll need separate fence contexts for display and
the rings.  And we can still combine them with the sync_fence api, so
it should be equivalent.  I'm just still thinking in terms of the
Android bits and not the struct fence bits; I'll probably come around
eventually.

-- 
Jesse Barnes, Intel Open Source Technology Center



More information about the Intel-gfx mailing list