----- Original Message ----- > Looks good to me though I wonder if it would be an option to do this at > vertex fetch time instead after vs? I guess though that way it would > have to be done separate for the llvm and non-llvm code as the fetch is > integrated into the llvm code. I think that the idea for d3d10 was to do > it in the input assembler, so the unneeded vertices don't have to go > through the vs. I doubt though it makes much of a difference.
Yea, I thought about that too but then it becomes /really/ complicated. Because if you split into primitives before the vertex shader then you'll endup either running vertex shader up to almost 3x times you would normally for stripped primitives, or you need to figure what vertices you ran the vertex shader on and try to skip them (very hard in SoA). And adding primitive decomposition to LLVM's is going to be a pain (although that we might need anyway because we might have to start skipping the adjacency vertices in the VS). So for the general case this is a lot cleaner. z _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
