Am 23.07.2013 00:42, schrieb Grigori Goronzy: > On 23.07.2013 00:30, Dave Airlie wrote: >> On Mon, Jul 22, 2013 at 2:06 PM, Grigori Goronzy <[email protected]> >> wrote: >>> Subpixel text rendering is typically done with a solid src and a >>> pixmap mask. Traditionally, this cannot be accelerated in a single >>> pass and requires two passes [1]. However, we can cheat a little >>> with a constant blend color. >> >> The hw should be able to do things in one pass with dual source >> blending afaik. >> > > Yes it can do that, and this was actually my first idea. However, dual > source blending requires additional render target state setup, > modifications to the shaders, and so on. It's likely also a bit slower > than single source blending, after all the shader needs to export twice > as much data. > > It might be worthwhile to get back to this to accelerate some of the > other PictOps with component alpha, but using the constant blend color > was the easier and faster solution for the common PictOpOver case. I > think it's a nice trick, I wonder if any other driver implements it? > > Grigori > _______________________________________________ > xorg-driver-ati mailing list > [email protected] > http://lists.x.org/mailman/listinfo/xorg-driver-ati >
I guess it's a pity that it can't be done at some higher level as this optimization is really generic neither tied to evergreen/ni nor radeons (the same is true for the other patches as well), just about any card can do it (as long as it has constant blend color). Dual-source, OTOH is limited to d3d10-class hardware. Looks good to me though. Roland _______________________________________________ xorg-driver-ati mailing list [email protected] http://lists.x.org/mailman/listinfo/xorg-driver-ati
