I wouldn’t normally post here, but just looking at some of the new techniques coming out from various manufacturers I believe a few DO make sense to finally add to the OpenGL pipeline.
3Dlabs has already proposed the shader additions, and although I have no ear in the OpenGL sessions, I would hope that the shading capability would be integrated into the OpenGL pipeline as it does make more sense there. Even if the end product is not quiet what 3Dlabs proposed, it makes sense to add it. Another one, is Displacement Map capability, currently only one company has it used, but I sincerely doubt if that will take long for others. Nvidia and others have admitted that Displacement maps are well suited for hardware acceleration; and I think, like shaders, this would be better set into the OpenGL pipeline so that it can be tesselated, textured, and shaded in a well ordered and well known manor. I think Nvidia’s point-sprites are also a good example of new technology that makes sense as a primitive, though it is also simple enough as an extension I fully expect it to make the ARB list and never make it as a primitive, but I think like points, lines, tris, quads, etc the point-sprite makes a good primitive using the current loaded texture it just falls in line with the other primitives.
Last, and then I will be quiet, Occlusion culling has been pondered over and over again, I do not know if Nvidia’s current method is the best, but I think it an obvious need to send a list of primitives down the pipe and then check to see if any were actually drawn. I think the concept is sound. As I understand it Nvidia built a queue and review system so that you request the feedback, but don’t get it until it makes sense to send it back (so as not to halt the pipeline as most user read’s do). I would like to see at least the process of queueing a read for the end-of-cycle and getting the results later a part of the OpenGL pipeline, not just for an occlusion test, but for other user reads as well. The so-called “do NOT list”: never read content from the OpenGL pipeline, the cost is prohibitive, etc… Admitadly, reading an entire screen full of information would NOT make sense in a queue and review system, it would tie up the AGP bus (or any IO bus on other platforms) with information back. But I do think some make sense in a method that is designed into the OpenGL pipeline so as not to disrupt it.
just my .02, from the for what it’s worth department. 