Following #926 -- the substitute of GPU_PARALLEL() for !$acc kernels does not work. A fix is needed. I do notice that the current fypp macro creates this kind of code

!$acc parallel default(present)
<code>
!$acc end parallel

I wonder if default(present) is the issue here. I think !$acc parallel alone might be OK, but have not tested it. Either way, it seems like having kernels be an option is somehow useful.