I’m new with cuda, and I have a question about cuda surface and npp.
I have some mathematical functions to write, to be fast I need to read data from texture and write result using a surface, I only find on cuda documentation that is possible to bind a surface to an Array “cudaBindSurfaceToArray” (not a memory pointer), so in my application I use Array for texture reading and surface writing.
(For information I don’t find a mean to ubind an Array from a surface, maybe there is a function like “cudaUnbindTexture”)
To not write all the mathematical functions myself, I plan to use some functions from npp, but I d’ont find the way to send a cudaArray to npp, I can’t maybe use pointer on cudaArray, but I don’t know how to get the pitch of my yhe cuda array.
I know it’s possible to create a useless kernel to copy data from a cuda array to a memory pointer, but with this kind of kernel I will lose some time…
So my question is, how it’s possible to send a cuda array to npp, or how to use a surface with a memory pointer (not a cudaArray).
Best regards and thanks.
For information, I use a GTX460.