(L) [2015/10/05] [szellmann] [Re: Separate GPU compute kernels vs. a single GPU kernel] Wayback!Just came across this very interesting paper:
[LINK https://mediatech.aalto.fi/~samuli/publications/laine2013hpg_paper.pdf]
The authors basically just confirm the advice you people here gave me. Seems like I have to do further testing with more complex materials and/or multiple materials, and eventually provide a mechanism to evaluate those in separate kernels. Remains the question how a good abstraction for that may look, but that's probably a different topic.