Re: [PATCH 2/3] glamor: turn on glamor.

[Chris Wilson <chris@chris-wilson.co.uk>]

On Fri, 11 Nov 2011 18:52:11 +0800, "Zhigang Gong" <zhigang.gong@linux.intel.com> wrote:
> 
> 
> > -----Original Message-----
> > From: Chris Wilson [mailto:chris@chris-wilson.co.uk]
> > Sent: Friday, November 11, 2011 5:12 PM
> > To: Zhigang Gong; intel-gfx@lists.freedesktop.org
> > Subject: Re: [Intel-gfx] [PATCH 2/3] glamor: turn on glamor.
> > 
> > On Fri, 11 Nov 2011 16:31:20 +0800, Zhigang Gong
> > <zhigang.gong@linux.intel.com> wrote:
> > > @@ -965,6 +969,9 @@ void
> > intel_uxa_block_handler(intel_screen_private *intel)
> > >  	 * framebuffer until significantly later.
> > >  	 */
> > >  	intel_flush_rendering(intel);
> > > +#ifdef GLAMOR
> > > +	intel_glamor_block_handler(intel);
> > > +#endif
> > >  }
> > 
> > I suspect this is the wrong way around as we are not flushing the render
> > cache of glamor's rendering to the scanout until the next block handler.
> I don't understand here. Would you please explain more detail? Thanks.

Whenever we render, the data ends up in the Render Cache and needs to be
flushed out to memory before it is coherent with the CPU or in this case
the Display Engine (i.e. scanout).

intel_flush_rendering() does two tasks. The first is to submit any
pending batch, and the second is to flush the Render Cache so that the
modifications land on the scanout in a timely manner. It is probably
best if those two tasks were separated so that we do:

  intel_uxa_block_handler(intel); // flush the UXA batch
  intel_glamor_block_handler(intel); // flush the GL batch
  intel_flush_rendering(intel); // flush the RenderCache to scanout

However, you can simply rearrange the code and achieve it with the
existing functions:

  intel_glamor_block_handler(intel); // mark the front bo as dirty as needbe
  intel_flush_rendering(intel); // flush UXA batch along with RenderCache

-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre