As I get used to CUDA Fortran, I have a simple question about syntax. Does pinned memory have to be allocatable? That is are:
real, pinned :: ta(m,np)
and
real, allocatable, pinned :: ta(:,:)
allocate(ta(m,np))
equivalent? Or do you need the explicit allocate to use pinned memory? (Note: I understand the latter is probably better, no matter what, since you can use the PINNED keyword with it.)