suppose that I have a openacc code that has to run on CPU (multicore) or on GPU, and that it uses curand API. My question is: which kind of API functions of curand should I use to run the same code on multicore CPU and GPU depending on the compilation target (-ta=mulitcore or -ta=nvidia)?
example: in the code there are references to curand_init (works on gpu target compilation) and when I compile it with -ta=multicore, it gives me this error on the compilation stage:
undefined reference to `__pgicudalib_curandInitXORWOW’.
Any curand usage will always require a GPU. There are no CPU-only generator functions in the curand library.