Recently I migrated to OpenCV 3.0 from 2.4.11 and I noticed that when using gpu (cuda) accelerated algorithms I do not get nvcuda.dll threads (when using VisualStudio debugger). Instead I get nvopencl.dll threads. I also noticed that algorithms are running slower. As I'm using a Nvidia gpu, I would like to know how I can get to run code optimized for cuda architecture.
I compiled openCV setting the flag WITH_CUDA and the architecture for my graphics card (gtx970), that is 5.2.