YAC coupling: mesh coordinates from Elmer internals

[ClemensSchannwell]

In the yac-coupling branch, where we couple ICON and the EBFM to Elmer/Ice, the elmer grid information that YAC needs to interpolate the fields is currently computed using an external c-function (https://github.com/ElmerCSC/elmerfem/blob/yac-coupling/fem/src/elmer_grid.c). This function reads in the coordinates from the partitioned mesh files. The proj-string is currently also hard-coded (https://github.com/ElmerCSC/elmerfem/blob/yac-coupling/fem/src/elmer_grid.c#L59-L61). While this is working for the moment, it makes running Elmer/Ice on different domains (e.g. Svalbard, Antarctica) cumbersome for now.
From my point of view it would be ideal, if the user can provide the proj-string, while the mesh coordinates, cell-centers etc. are used from Elmer internals.

A shopping list of what YAC needs is the following:

int num_nodes // number of nodes stored on the local process
int num_elem // number of elements stored on the local process
int node_ids[num_nodes] // global ids of all local nodes
int elem_ids[num_elem] // global ids of all elem
double node_x[num_nodes] // x coords of local nodes
double node_y[num_nodes] // y coords  of local nodes
int num_node_per_elem[num_elem] or int num_node_per_elem // number of nodes for each elem
int elem_to_node[SUM(num_node_per_elem[:])] or int elem_to_node[num_node_per_elem * num_elem] // local index of nodes for each elem; node of each elem have to be order either clock or counter-clock wise; values are in the range [0;num_nodes-1]; nodes of i'th elem are stored at elem_to_node[SUM(num_node_per_elem[0:i-1]):SUM(num_node_per_elem[0:i])]

some kind of information on the projection being used

If available (I can also compute it from node_x/node_y):
double elem_x[num_elem] // x coords of center of elem
double elem_y[num_elem] // y coords of center of elem

[raback]

I sent a suggested code directly as it was still just a template.

[BenjaminRodenberg]

I will start working on this issue on the following branch of my fork: https://github.com/BenjaminRodenberg/elmerfem/tree/yac-coupling-direct-mesh BenjaminRodenberg#1

[fgillet]

Hi don't know if it can be useful for you; but for the coupling with XIOS I also need to send the grid information to xios;
This is done in the SUBROUTINE DomainDefinitions() in XIOSOutPutSolver in the Elmer/Ice solvers;
conversion from (x,y) to (lon,lat) is done using internal routines in the ProjUtils module (under ElmerIce/Utils); I have hard coded polar Steoreographic projection (north and sud) and allow the possibility to use the fortrangis library (https://fortrangis.sourceforge.net/) for more flexibility;
Maybe you could use parts of those codes.
BTW thanks on spotting and solving the issue to compile Elmer with XIOS on Ubuntu.

[BenjaminRodenberg]

@fgillet that's definitely a helpful piece of information! Currently, we are reading the Elmer mesh independently from file with another code (EBFM, see here) and we are planning to transfer this information directly over our coupling interface (YAC) from Elmer to EBFM.

Thanks for the feedback! Very happy to hear it's worth the trouble and the fix helps. I was already afraid I'm the only person trying to get this running on Ubuntu 😀

[BenjaminRodenberg]

I started working on this feature on my fork: BenjaminRodenberg#1 I will give a brief summary here to keep everybody who is interested in the loop.

BenjaminRodenberg#1 is a prototype that will end up on the branch yac-coupling occasionally. I still need to do some polishing and look into the failing CI jobs.

There are two cases that we want to ultimately cover:

1. SSA in 2D (corresponds roughly to GatherBulkData in the draft provided by @raback); Get coupling grid information from Elmer internals instead of file #772
2. Stokes in 3D (corresponds roughly to GatherBoundaryData in the draft provided by @raback); TODO

From my perspective 1. should generally work with the current state of my prototype @ BenjaminRodenberg@909366a. I will probably start implementing 2. in a follow-up PR as soon as the functionality for 1. is verified and merged into yac-coupling.