Hi
I've been having trouble with declared subarray types, so am trying to switch my code over to using the direct subarray notation that's now allowed by the mpi_f08 module.
Using lines like:
call MPI_Irecv(Array(0, 1:ksizey_l), ksizey_l, MPI_Real, ip_xdn, 0 + tag_offset, comm, reqs(2)) call MPI_Isend(Array(1, 1:ksizey_l), ksizey_l, MPI_Real, ip_xdn, 1 + tag_offset, comm, reqs(3))
give warnings along the lines of:
test_mpif08.F90(51): warning #8100: The actual argument is an array section or assumed-shape array, corresponding dummy argument that has either the VOLATILE or ASYNCHRONOUS attribute shall be an assumed-shape array. [ARRAY] call MPI_Irecv(Array(0, 1:ksizey_l), ksizey_l, MPI_Real, ip_xdn, 0 + tag_offset, comm, reqs(2))
(I've gisted a full minimal working example here.)
While I can't find much documentation on mpi_f08, reading examples like this suggest that what I'm doing should be fine - can anyone tell me why am I getting this warning, and how I can avoid it (without just suppressing it with a flag)?
Thanks
Ed
Link Copied
Having pushed through ignoring the warnings, I now find that this seems to work fine with my rank 4 and 5 arrays (large numbers of warnings notwithstanding), but corrupts data in rank 6 arrays. I've put up another gist here that shows the program I'm using in this case. I'd appreciate any pointers on how to avoid this!
Edit to add: I've noticed I was using ifort 2017 (and Intel MPI 2017). Switching to ifort 2018 (and Intel MPI 2018) gives the following error in addition to the above warning:
test_halo_6_aio.F90(60): error #7505: If an actual argument is an array section with vector subscript and corresponding dummy argument does not have VALUE attribute, it must not have ASYNCHRONOUS / VOLATILE attribute. [BUF]
For more complete information about compiler optimizations, see our Optimization Notice.