On Saturday 12 March 2005 03:50, Timothy Miller wrote: > On Wed, 9 Mar 2005 09:47:42 +0100, Martijn Sipkema <[EMAIL PROTECTED]> wrote: > > From: "Timothy Miller" <[EMAIL PROTECTED]> > > Sent: Wednesday, March 09, 2005 07:31 > > > > > For implementing FSAA, I can't remember whether it's called > > > multisampling or supersampling, but the easy technique is to have a > > > stage in the pipeline that divides the geometry down and manipulates > > > the alpha channel so as to convert 2x2 pixels into one pixel in the > > > framebuffer. This is trivial to implement, and I can add it in. > > > > That won't work I think as information on subpixels is lost on converting > > to a single fragment with alpha---that conversion is to be done after all > > drawing is completed, I think... > > All 3D drawing, yes. I was going to insert it just before the extra > stuff I inserted to do 2D stuff that OpenGL doesn't account for.
I've been thinking, would it be acceptable to lose the hardware overlay scaler when FSAA is turned on, and/or to only have FSAA in full-screen OpenGL mode? The way the hardware overlay scaler works is that it takes the framebuffer, and a second buffer with video data. It copies the framebuffer to output, except for pixels with a certain colour key where it switches to a colour-converted (YUV->RGB), scaled (with linear interpolation?) version of the second buffer. So, that has to happen right before the DAC. Is that all correct? In that case, you could add a small amount of extra logic to the overlay scaler (basically, you have to be able to tell it to ignore the framebuffer input altogether and just give a scaled version of the second input, and you need to be able to turn off the YUV->RGB conversion, which you need to do anyway) and you could trade off quality against speed arbitrarily. For example, you could render the scene at width*sqrt(2), height*sqrt(2) to get one half the fillrate but still some AA. Lourens _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
