CCPP Framework Meeting Minutes 2020 03 19

Agenda:

• RRTMGP merge to NCAR dtc/develop - required some development work on ccpp-framework (ccpp_prebuild.py) to make the CCPP kind and type definitions available to the host model (so that it can allocate the memory)
• NCAR/CGD is only interested in the RRTMGP version (this is what they are using now) - will try to do some testing with the version that goes into NCAR dtc/develop
• Collaboration with NRL/NEPTUNE:
  • Jim Doyle and Alex Reinecke have some money to develop CCPP (framework in particular)
  • discussion was around availability of staff, how the work they are interested in is related to cap_gen
  • NRL interests: SCM (spectral element method for the vertical coordinate; currently SCM is taking the vertical velocity to do vertical advection); CCPP framework to provide information which arrays to allocate/perform advection/... (this is something Steve is currently working on); work related to the SIP proposal (GPU, NUOPC caps from CCPP)
  • invite NRL to one of the next ccpp-framework meetings
  • difficulty is cross-agency funding (how to get the money from NRL to NOAA - easier to get to NCAR)
  • but do we have the people to do the work???
  • alternative arrangement is that NRL hires someone and we are taking an advisory role
  • NCAR/CGD will soon be looking for a software engineer to work on CCPP - funding from NRL would be useful

GPU preparedness (Steve was talking with CISL performance group)

• code is written in a way that can be ported to GPUs quite easily
• need to conduct some experiments
• there is a version of MG that has been ported to GPUs
• GPU discussion with MPAS team would be useful to get their insight and experience

Invite NRL to the meeting in two weeks (April 2nd), then have the GPU conversion with NRL and MPAS a week or two later?