Philippe 18/01/2022: OpenACC: Cray compiler does not support directives with kernels and loop in the same one

Philippe 18/01/2022: OpenACC: store result of LNORTH_ll() and co in logicals to be used in kernels regions + do also this for HLBCX and co
The later is a workaround for a Cray compiler bug (CCE 13.0.0)

Philippe 13/01/2022: OpenACC: ACC_DEVICE_* is 64 bits for CCE and 32 bits for NVHPC + add detection of AMD GPUs

Philippe 07/12/2021: OpenACC: fix: remove acc loop independent if atomic not set (to compute correctly)
Ideally, the loop independent + atomic should be set but nvhpc 21.9 crashes