Call CUDA kernel without parallel computing toolbox?

Can anyone confirm whether or not it is possible to call a cuda kernel (ptx file) in Matlab without having the parallel computing toolbox (ie without using parallel.gpu.CUDAKernel) ?

I've seen "rumors" that it's possible to write a mex file that calls a kernel, and that this would not require the parallel computing toolbox... is this true?

I'm not looking for a detailed description of how to do it, just confirmation that it's possible.

Answer 1

[Matt J]

Oliver Woodford insists here

http://www.mathworks.com/matlabcentral/answers/110697-is-there-a-way-to-make-device-memory-persist-between-cuda-mex-calls

that he has done it. It seems like a very painful thing to do, however, if you want data to persist on the graphics card between MEX calls, which I imagine to be very common. But, the thread talks about how to do that, too.