I was wondering about the possibility to create the shadow map(s) for dual-paraboloid shadow mapping in one single pass. I’m probably way off and that’s why I’m posting it here. Maybe there wouldn’t even be any performance gains, but it could be worth trying.
Drawing from the method presented by Brabec et al. in their paper, couldn’t one instead use a single shadow map of proportions 2:1 (e.g. 2048x1024) and instead of culling fragments in the wrong hemisphere by looking at the sign of the z-coordinate, use this to choose whether to write in the left or right half of the shadow map. Geometry would only have to be transformed once, and one could probably still make use of 4-sample PCF with free depth comparison in hardware, plus double speed Z-only rendering.
But as I said, maybe I’m completely off, so don’t “cut my head off” for posting a stupid idea.