[Ncep.list.nems.announce] Correct MPI tag collision issue and increase array sizes to support integration of more domains
Thiago Quirino - NOAA Federal
thiago.quirino at noaa.gov
Tue Oct 14 15:07:27 UTC 2014
Good morning, all. I would like to commit some changes after Jun Wang's
(1) A small fix to a rare event that can take place during nest motion. MPI
tags generated by subroutine PARENT_SENDS_CHILD_DATA_LIMITS in
"module_PARENT_CHILD_CPL_COMP.F90" may collide with the MPI tag expected by
subroutine CHILDREN_RECV_PARENT_DATA. We increased the constant
"PARENT_SHIFT_TAG=1113" used in the generation of tags used by the related
subroutines PARENTS_SEND_CHILD_DATA and CHILDREN_RECV_PARENT_DATA to
"PARENT_SHIFT_TAG=1e8". This simple solution causes the MPI tags to be
shifted out of the collision range. It also does not induce tag collisions
with other MPI calls in the code.
(2) We are starting to run NMMB with large number of domains. Tom has
indicated a few arrays whose dimension should be increased to successfully
run the model with increased number of domains. The modifications are to
"module_CLOCKTIMES.F90" and "module_NMM_GRID_COMP.F90".
(3) In "src/atmos/phys/module_SF_GFDL.F90", three total calls to subroutine
"wrf_debug" seem to be generating a lot unnecessary log output. We propose
to comment those out.
I will post the code and the list of modified files after Jun Wang's commit.
Thank you so much. This is my first time submitting a ticket so please let
me know of any issues or if I am doing something wrong.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Ncep.list.nems.announce