Unless Halo is not wrapping his "decal" to the surface of the level... Which would produce really stupid looking results, (a big fuzzy circle pressed up against the wall that sticks out when it is a near a corner) then the method he suggests is likely too slow to use.
If you're talking about rendering Quake-like dynamic lights, by using a spherical lightmap, comprised of one 2d map and one 1d map mapped on perpendicular axes to produce a "spherical" region of light, then that is stimply not going to be fast enough fro realtime except on the simplest of levels.
I can say this with fair certainty, because in essence, doing this is very much like rendering shadows in my shadow system. For maximum speed, what you would end up doing is having to clip all the polygons in your level, or at least, all those in sections that are currently visible, using sutherland-hodgeman clipping. You would clip them to a cube which defines the region of the spherical light, with the cube's borders being the maximum extent of the light. The result of this clipping is a list of clipped polygons that you then need to build into a new mesh with proper lightmap coordinates.
But clipping polygons is not free, nor is building new meshes. And because this is so similar to how my shadow system works... except for the fact that you will be "shadowing" roughly six times as many polygons per light, as each light effectively casts in all directions whereas a shadow casts in only one, this is how I can predict the speed of such a system as being slower than my shadow system by a significant degree.
Plus, because lights will cover such large areas in all likelyhood, they will create more overdraw than my shadow system as well.
If you're talking about rendering Quake-like dynamic lights, by using a spherical lightmap, comprised of one 2d map and one 1d map mapped on perpendicular axes to produce a "spherical" region of light, then that is stimply not going to be fast enough fro realtime except on the simplest of levels.
I can say this with fair certainty, because in essence, doing this is very much like rendering shadows in my shadow system. For maximum speed, what you would end up doing is having to clip all the polygons in your level, or at least, all those in sections that are currently visible, using sutherland-hodgeman clipping. You would clip them to a cube which defines the region of the spherical light, with the cube's borders being the maximum extent of the light. The result of this clipping is a list of clipped polygons that you then need to build into a new mesh with proper lightmap coordinates.
But clipping polygons is not free, nor is building new meshes. And because this is so similar to how my shadow system works... except for the fact that you will be "shadowing" roughly six times as many polygons per light, as each light effectively casts in all directions whereas a shadow casts in only one, this is how I can predict the speed of such a system as being slower than my shadow system by a significant degree.
Plus, because lights will cover such large areas in all likelyhood, they will create more overdraw than my shadow system as well.