On Tue, 2007-03-13 at 13:24 -0400, Adam Jackson wrote: > On Tue, 2007-03-13 at 13:00 -0400, Owen Williams wrote: > > This is a total non-expert opinion, but aren't we scaling all of the > > images to correct for the high-dpi screen? That sounds like it would > > call for bilinear filtering. > > If so, ow. Ow ow ow. > > But I didn't think so. And even if we were, we could probably get away > with doing nearest scaling.
Ow, apparently. Blizzard tells me we really are scaling to doublesize to compensate for the increased DPI. That's... less than optimal? We're not going to be able to do scaling on the GPU, regardless of what magnification filter you might want. I'm not sure what the image quality would be like just doing nearest-neighbor upscaling. It might be okay, since we're sampling color channels anyway, but there might be weird color banding effects. Tough to know without testing. Even then though, we don't have stretch blits in hardware, and I can't think of any way to fake them up in hardware. So it looks like the only path remaining is to make fbFetchTransformed go as fast as possible. For the case of integer scale factors it might be possible to cheat, but the general path is likely to just hurt. There's a reason we don't do 3D on the CPU. - ajax _______________________________________________ Devel mailing list [email protected] http://mailman.laptop.org/mailman/listinfo/devel
