How to use CUarray in kernels ?


The CUDA Driver API has special functions for “CUarray” handles. (like 1d,2d,3d “cuda arrays”, it comes with descriptors and such and channels (presumable for r,g,b and such).

I was just wondering a couple of things about this:

  1. How to use CUarray in kernels ? How to access the elements ?
    (Maybe these handles are not supposed to be accessed, and are supposed to be only passed to special inside-kernel-api’s-functions-instructions (probably not though ;)) maybe also texture related or so… )

  2. What benefits does using this API provide compared to cuMalloc and such ?