Juan 16/11/2021: mode_openacc_set_device.f90,tools.f90, add _FAKEOPENACC for compile in OPENACCDEFONLY with gfortran/ifort etc ...

Juan 16/11/2021: ZSOLVER/ new source from MNH only to change intent(IN) -> INOUT for PDZX/Y , for get_halo2_d

Juan 16/11/2021: add orig source from MNH , to change intent(IN) -> INOUT for PDZX/Y , for get_halo2_d

Juan 21/10/2021:ZSOLVER/mode_exchange2_ll.f90 , For GPU=OPENACC , add orig for no managed memory init_halo2 allocation/adaptation

Juan 15/10/2021:tensor,prepare for nomanaged compilation -> replace %{dimx/y/z} in loop to avoid copyin of derived type structure

Juan 15/10/2021:tensor,prepare for nomanaged compilation ->  add enter/exit data , for allocate variables

Juan 12/10/2021:communication.f90, for GPU use async N/S/E/W kernels for haloswap_mnh , but not sure better performance

Juan 21/09/2021:advection_uvw.f90, use pointer contiguous + MNH_ALLOCATE , & move kernels outside IF to avoid copy/Out of memory allocating nvhpc BUG !!

Juan 21/09/2021:p_abs.f90, use pointer contiguous + MNH_ALLOCATE , and add loop seq to avoid collapse(4) no reproductible computation

Juan 21/09/2021:ZSOLVER/turb_hor_thermo_flux.f90 , remove some empty acc kernels/end kernels ,generating empty serial kernels

Juan 21/09/2021:ZSOLVER/tridiag_thermo.f90 , use acc ... gang, vector , or vectorisation in only 1 dim on GPU

Juan 21/09/2021:ZSOLVER/update_lm.f90 , use GET_HALO_DDC + kernels async + wait for boundaries opt on GPU