[PATCH] exa: Unset and restore per alpha in exaTryMagicTwoPassCompositeHelper

[Michel Dänzer]

On Mon, 2009-10-19 at 15:02 -0400, Zack Rusin wrote: 
> On Monday 19 October 2009 08:58:22 Michel Dänzer wrote:
> > On Fri, 2009-10-16 at 00:29 +0100, jakob at vmware.com wrote:
> > > From: Zack Rusin <zackr at vmware.com>
> > >
> > > This fixes the component alpha fallback in exa. I'm not sure which
> > > branches this should go into. Also before committing this patch make sure
> > > that we get a Tested-by by somebody with a radeon card as this turns on
> > > new and wonderful paths not tested in the drivers.
> > 
> > This shouldn't make any difference for sub-pixel AA text with the radeon
> > driver, as it doesn't simply check for pMask->componentAlpha but:
> > 
> > 	if (pMaskPicture->componentAlpha) {
> > 	    /* Check if it's component alpha that relies on a source alpha and
> > 	     * on the source value.  We can only get one of those into the
> > 	     * single source value that we get to blend with.
> > 	     */
> > 	    if (RadeonBlendOp[op].src_alpha &&
> > 		(RadeonBlendOp[op].blend_cntl & RADEON_SRC_BLEND_MASK) !=
> > 		RADEON_SRC_BLEND_GL_ZERO) {
> > 		RADEON_FALLBACK(("Component alpha not supported with source "
> > 				 "alpha and source value blending.\n"));
> > 	    }
> > 	}
> > 
> > I'm not sure offhand if the proposed change is necessary or even
> > sensible given this, or if the Gallium Xorg state tracker shouldn't just
> > use similar checks, and I may not find the time to build a firmer
> > opinion before I'm back from vacation next month.
> 
> Well, the question we're asking the driver is: do you support the following 
> operation, we expect an answer based on its own capabilities not on the 
> capabilities of the acceleration architecture.
> 
> So the question: do you support component alpha when the driver doesn't should 
> always be a no, not a conditional maybe that depends on code it has no control 
> over. 
> IMO if Exa decomposes PictOpOver component alpha into two non-compoment-alpha 
> passes, it should actually do that and not depend on driver being able to 
> figure out what it's trying to do.

Makes sense - assuming the fact that the mask picture has component
alpha actually makes a difference for the decomposed operations. Does
it? I'm not sure offhand, and I'm not inclined to find out before the
end of the month for reasons stated earlier. :) Though I'm inclined to
assume it doesn't, or I'm not sure how the decomposition would help for
hardware that doesn't support component alpha.



-- 
Earthling Michel Dänzer           |                http://www.vmware.com
Libre software enthusiast         |          Debian, X and DRI developer