Open bobsomers opened 13 years ago
After toying with the occupancy calculator and reworking some memory allocation stuff, I think the kernel is way too big and branchy at the moment. Splitting the problem into much smaller parts than just intersection and shading, while leaving all intermediate data on the GPU may allow us to increase occupancy and pipeline a lot of operations with streams.
Break intersection tests and shading into separate kernels. Prep work for future volume integration kernel in between these two steps.