[Intel-gfx] [PATCH 6/6] drm/i915: Stop caching the "golden" renderstate

[Rodrigo Vivi]

On Fri, Nov 03, 2017 at 08:39:21AM +0000, Joonas Lahtinen wrote:
> On Thu, 2017-11-02 at 16:34 -0700, Rodrigo Vivi wrote:
> > On Thu, Nov 02, 2017 at 10:36:26PM +0000, Oscar Mateo wrote:
> > > 
> > > 
> > > On 11/02/2017 07:56 AM, Joonas Lahtinen wrote:
> > > > On Thu, 2017-11-02 at 07:54 -0700, Rodrigo Vivi wrote:
> > > > > On Thu, Nov 02, 2017 at 02:43:16PM +0000, Joonas Lahtinen wrote:
> > > > > > On Thu, 2017-11-02 at 12:42 +0000, Chris Wilson wrote:
> > > > > > > As we now record the default HW state and so only emit the "golden"
> > > > > > > renderstate once to prepare the HW, there is no advantage in keeping the
> > > > > > > renderstate batch around as it will never be used again.
> > > > > 
> > > > > So, with this in place we really don't need that null context for CNL.
> > > > > to fullfill all Mesa needs, right?!
> > > > 
> > > > Separate issue, this only fixes isolation. This patch just releases it
> > > > from memory after it's been applied to the default context states to be
> > > > stored.
> > > > 
> > > > But yes, we also decided not to have null context for new platforms.
> > > 
> > > At last until, two years from now, we find out that there is a very subtle
> > > reason why we need it :)
> > 
> > :( - Yeap, for me just to be on the safe side and start with a clean context
> > would be a good reason already...
> 
> Indeed, we are starting with a clean context now. If the HW division
> messes a new design so badly that it won't be able to execute without
> null render state, we shall reiterate :) But I have high hopes that we
> don't need to.

Well, yeah!
That makes sense and also it makes the platform enabling easier.
Thanks for the work and explanations.

> 
> Regards, Joonas
> -- 
> Joonas Lahtinen
> Open Source Technology Center
> Intel Corporation