[Intel-gfx] [PATCH 02/26] drm/i915: don't stop+start FBC at every flip

[Zanoni, Paulo R]

Em Ter, 2015-10-27 às 19:50 +0000, Chris Wilson escreveu:
> On Tue, Oct 27, 2015 at 02:50:04PM -0200, Paulo Zanoni wrote:
> > @@ -1021,13 +1078,48 @@ void intel_fbc_flush(struct
> > drm_i915_private *dev_priv,
> >  	if (origin == ORIGIN_GTT)
> >  		return;
> >  
> > +	/* Hardware tracking already recompresses the CFB (nuke)
> > for us if FBC
> > +	 * is enabled and we do a page flip, so we can safely
> > ignore it here.
> > +	 * FBC may be disabled in case we got an invalidate()
> > before the
> > +	 * flush(), so we'll still have to check that case below.
> > */
> 
> I feel like I understand what is going on a bit better now, thanks!
> 
> > +	if (origin == ORIGIN_FLIP && dev_priv->fbc.enabled)
> > +		return;
> > +
> >  	mutex_lock(&dev_priv->fbc.lock);
> >  
> >  	dev_priv->fbc.busy_bits &= ~frontbuffer_bits;
> >  
> >  	if (!dev_priv->fbc.busy_bits) {
> > -		__intel_fbc_disable(dev_priv);
> > -		__intel_fbc_update(dev_priv);
> > +		if (origin == ORIGIN_FLIP) {
> 
> Note this test here is redundant.
> 
> We know that for an origin == FLIP to be here, FBC is disabled and
> calling intel_fbc_disable() is then a no-op.
> 
> So it turns out that the origin two lines of disable(); update()
> works
> just as well. Or is there some later patch that tweaks this branch
> further?

Right, you commented about this on the patch that is currently 04/26 an
I only fixed it there.

> -Chris
>