A bug fixing release 4.3.5.1 has been uploaded in the File Section of the
gforge.ictp.it site.

It contains the following modifications:

- Fix problem with wind variable names in GrADS helper programs (Yamin)

After the name change to CORDEX variable naming convention, the VECTORPAIR
GrADS command was misaligned with the names of variables in the actual
output file, making also crash some versions of GrADS (2.0.9a 32bit). The
names are now consistent. The problem was introduced in 4.3.5.

- Fix problem with MPICH2 and CLM on a printout of hostnames (Yamin)

This was the root problem of the impossibility to run RegCM+CLM on machine
with MPICH2. A buffer used both for send and receive in a gather operation
to do just a printout works with openmpi but screws things with MPICH2.
Note this bug was also present in the original CLM 3.5 code. The problem
was there from the introduction of CLM in RegCM 4.0.

- Fix initialization of STS variables on first timestep

The storage used to write in the STS output the variables tsmin, tasmin
and psmin was initialized in the wrong point of the code (init instead of
output), preventing the correct computation of those variables on the
first timestep, both on initial run and on restart. The problem was
introduced in 4.3.5.

- Fix topo variable on SUB output

The SUB output file construction of the topo variable was wrong, and the
output file did contain a wrong field. The problem was introduced in
4.3.5.

- Update Documentation (both User and Reference Manual)

The Documentation has been updated: the Reference Manual now has a
description of the UW pbl scheme (Thanks Travis!), and the User Guide has
been updated to the 4.3.5 release. The User Guide in 4.3.5 was still
having bits related to the previous releases. The updated PDF documents
have been uploaded to the Gforge Doc section.

- Fixed PGI compilation (Shalaby)

An if condition present in the mod_mppparam was first checking for the
presence of an optional arguments, and then checking it. The PGI compiler
optimized this out to have a segfault (probably checking the two
conditions together instead of in the given order). Splitting this in two
ifs did the trick. The bug is relative to the 4.3.5 release.

Now the CLM+MPICH is working, and we have it tested also with PGI 11
compiler on a SuSE system (Thanks Ahmed!). I have also did a test with
latest PGI compiler to success.

Other note, with Intel 11 compiler, on systems with low memory, the
mod_ncout module may make the compiler crash with segfault (reported by
James Ciarḷ). Solution is to remove in this case interprocedural
optimization, i.e. remove the -ip option on the command line. As this is
not happening on all systems (on our argo cluster masternode the model
compiles with same intel compiler version) and with newer intel compilers
(12 and 13) it gives a boost to performances, we have kept it anyway by
default.

Users affected by any of the above bugs are invited to upgrade.

Ciao!

Graziano.

_______________________________________________
RegCNET mailing list
RegCNET@lists.ictp.it
https://lists.ictp.it/mailman/listinfo.cgi/regcnet