[mpas-developers] vector reconstruction in MPAS trunk

[Michael Duda]

Hi, All.

The switch to using a common module_vector_reconstruction.F in
a new operators directory seems to be working for me. I'll wait
until tomorrow in case anyone still needs a bit more time to 
raise any objections; in case I don't hear any, I'll go ahead
and commit the change tomorrow.

Cheers,
Michael


On Thu, Mar 25, 2010 at 11:31:58AM -0600, Michael Duda wrote:
> Hi, All.
> 
> Xylar and Todd have completed a vector reconstruction module that
> we'd like to be able to share between MPAS cores. I'm proposing to
> add this module as part of an operators directory that can be
> expanded in future to contain other non-infrastructure code that
> is usable by any core for performing general computations on MPAS
> grids. Essentially, then, the change to the trunk would create the
> directory mpas/src/operators, add module_vector_reconstruction.F
> to this directory and remove it from core_* directories, and make
> the necessary changes to Makefiles to compile code in
> src/operators as a library. Since a number of files will be
> modified, and to make it easier for others to test, I've placed a
> tar file in
> http://www.mmm.ucar.edu/people/duda/files/mpas_vector_commit.tar.gz
> that contains a checked-out version of trunk/mpas (so anyone can,
> for example, use 'svn status' or 'svn diff <file>' to see the
> changes) with the proposed changes included.
> 
> The only expected changes to results come from changes in the
> vector reconstruction, which is currently commented out in the
> ocean core and used only in the shallow water core.
> 
> I did make changes in the core_ocean directory, so ocean model
> developers might want to give the changes a closer review.
> Specifically:
> 
> * src/core_ocean/Registry - remove fields matrix_reconstruct, 
> normal, and rbf_value, which are not needed by the new version of 
> module_vector_reconstruction.F
> 
> * src/core_ocean/Makefile - remove rules and dependencies for
> module_vector_reconstruction, which is now found in the
> src/operators directory; add ../src/operators to the module search
> path
> 
> * src/core_ocean/module_vector_reconstruction.F - remove this file
> in order to use the shared version in the operators directory
> 
> I have checked that all three cores compile with these changes,
> and I did a quick run with each, although I'm not entirely sure
> what to look for in the ocean model output to check for
> correctness. A couple of comments about the ocean core code to
> consider: 
> 
> * In module_time_integration.F, MPI_Abort() is called directly,
> and it would probably be better to use dmpar_abort() from
> module_dmpar.F to keep as much code independent of the particular
> message passing library as possible
> 
> * also in module_time_integration.F, there is a call to isNaN(), 
> which isn't an intrinsic defined in the Fortran standard, causing
> some compilers to balk; using IEEE_IS_NAN() from Fortran2003 might
> be an option, although I don't know how many compilers currently
> support the IEEE_ARITHMETIC module
> 
> * in the default namelist.input.ocean, the strings "grid.nc",
> "restart.nc", and "output.nc" should be in quotes ("character
> input must be delimited" according to the Fortran standard); 
> executables produced by some compilers fail when reading
> un-delimited strings in namelists.
> 
> If anyone has any comments or suggestions regarding the proposed
> changes, please don't hesitate to let me know.
> 
> Cheers,
> Michael