Changes between Initial Version and Version 1 of Ticket #2694
- Timestamp:
- 2021-06-18T12:09:48+02:00 (22 months ago)
Legend:
- Unmodified
- Added
- Removed
- Modified
-
Ticket #2694 – Description
initial v1 3 3 4 4 In the actual version of the code, in iom_put (and because of the way we defined our interface with xios): 5 6 we give to xios arrays defined over the full domain (jpi,jpi) 7 we say to xios to write data only over the inner domain (Ni_0, Nj_0) 5 - we give to xios arrays defined over the full domain (jpi,jpi) 6 - we say to xios to write data only over the inner domain (Ni_0, Nj_0) 8 7 The problem is that in many cases (in most of cases?), output variables are diagnostics that could be computed only over the inner domain. Having to use full domain arrays forces us to do more computation, to call lbc_lbk (that is incompatible with a proper use of the tiles) or to use arrays with uninitialized halos which prevents to use the debug options. 9 8 … … 22 21 2) the second solution is good for performances but bad for users and maintenance: 23 22 The idea is to define 2 set of grids: gridT/U/V/F over the full domain and gridT/U/V/F over the inner domain. The modifications are quite easy to do in the code. The problem is that we would have to modify the xml files: 24 25 We would need to duplicated the contain of grid_def_nemo.xml to have for example grid_T_2D_full and grid_T_2D_inner instead than a simple grid_T_2D 26 next in all the field_def_nemo-*.xml files, we sould make sure that the grid linked to the variable is the proper one… there is more than 1200 variables... 23 - We would need to duplicated the contain of grid_def_nemo.xml to have for example grid_T_2D_full and grid_T_2D_inner instead than a simple grid_T_2D 24 - next in all the field_def_nemo-*.xml files, we sould make sure that the grid linked to the variable is the proper one… there is more than 1200 variables... 27 25 28 26 ==== Recommendation
