Advanced constructors fix
This standardizes the advanced constructors where you pass in preallocated device memory. I don't want users to have to deal with the dev_mem_t
struct since that's internal, so I created two functions:
wrap_mem_cl()
wrap_mem_cuda()
and so depending on the backend you are using, you would call, e.g., mat(wrap_mem_cl(my_cl_mem), n_rows, n_cols)
or similar.