Modification in interpolation to consume less memory

[jgipsl]

Laurent Li, working with the coupled LMDZ-OR on a zoomed grid with resolution 184x179x19 detected a high memory pic while interpolating the files in ORCHIDEE, specially carteveg file. The memory consumption is a probleme specially with zoomed grid where some grid boxes outside the zoom becomes very big.

He proposed therefor a modified version where he loops over all point and calls aggregate in this loop. This avoids to allocate the tables sub_index and sub_area.

At Ada, original sources code is found here : corresponds to orchidee_1_9_2 on CVS
/workgpfs/rech/ces/rces111/tibet/models_global_regional_20140625.tar.gz

Modified sources : (only interpol_help.f90 and slowproc.f90)
/linkhome/rech/ces/rces111/test

Test case:
/workgpfs/rech/ces/rces111/tibet/test/redem_sechiba

This work is based on tagged version orchidee_1_9_2 on CVS.

To do
Test if this zoomed version can run on ada without any modifications.
A first analyze using the trunk must be done to examine the current memory consumption before decision to take into account these modifications or not. The difference in time consumtion must also be verified. Calling aggregate on each point will take longer time and this has to be compaired by the gain on memory.

[jgipsl]

This will not be investigated because we will move towards XIOS for all interpolations in ORCHIDEE.