Andy Porter
Andy Porter
In making lfric/eg14 work in #1669 I had to extend the `field%log_minmax` method to ensure that it brought data back from the GPU before doing the calculation. This needs doing...
Currently the psyclone-kern algorithm-generation functionality initialises every field argument that it finds to unity. However, LFRic kernels often have arguments that contain geometric information and this should be correctly set-up....
Currently, InlinedKern *has* a Schedule where, really, it ought to *be* a (subclass of) Schedule as that would then be consistent with a Routine.
While delivering training on the NEMO API, it has become clear that the use of 'invoke' language is just confusing since it is a hangover from the PSyKAl APIs. It...
Functionally, this is a small PR that just adds a 'force' option to ParallelLoopTrans. However, I've also take the opportunity to improve the testing of this class.
This requires #1782 to go onto master first.
Changes the default OMP schedule to be "none" which means that no `schedule` clause is added to the OpenMP directive. Also refactors things slightly so that the `OMPScheduleClause` is the...
Asif has reported that he can create invalid code by combining the `HoistLocalArrayTrans` with the `ACCRoutineTrans` because the former adds an `allocate` to the routine. Since the routine is being...
A small PR that just tightens-up the fparser2 subroutine handler to make sure that we catch a subroutine that has arguments but no declarations. Since PSyclone does not support implicitly-declared...
Chris has found that if PSyclone encounters code such as: module my_mod contains subroutine my_sub(iarg) write(*,*) iarg end subroutine my_sub end module my_mod then PSyclone loses the `iarg` argument to...