I am in the final stages of eliminating reference to the pmgrid module in cam physics,
pmgrid provides the global size of the rectangular grid in plat and plon.
This is a nessesary step for inclusion of non-rectangular dynamics grids.
Outstanding problems exist in subroutine gavglook (physpkg.F90) and phys_gmean.F90 both of these routines involve doing calculations largely on the masterproc in SPMD mode which creates a bottleneck for parallel scaling. But there are issues about doing these in parallel because that has the potential to introduce order of operation differences which makes the validation tests more difficult to impliment.
This is a first attempt to open this for discussion via the forum rather than just to an email list... comments?
Jim
pmgrid provides the global size of the rectangular grid in plat and plon.
This is a nessesary step for inclusion of non-rectangular dynamics grids.
Outstanding problems exist in subroutine gavglook (physpkg.F90) and phys_gmean.F90 both of these routines involve doing calculations largely on the masterproc in SPMD mode which creates a bottleneck for parallel scaling. But there are issues about doing these in parallel because that has the potential to introduce order of operation differences which makes the validation tests more difficult to impliment.
This is a first attempt to open this for discussion via the forum rather than just to an email list... comments?
Jim