I'm processing WRF ARW netcdf output files with unipost.exe, compiled with option "Linux x86_64, Intel compiler (dmpar)".
I run unipost with the command:
mpiexec -n N unipost.exe
When N > 1 the resulting 3d field Total Cloud Cover [%] looks wrong.
In particular a 2d plot onto a fixed z lelel looks like segmented into horizontal bands.
The problem disappear if I run with N = 1.
On my system the problem affect all the version UPPV2.0, UPPV2.1, UPPV2.2 coupled with WRFV3.4.x, WRFV3.5.x, WRFV3.6.x.
My system has two Intel Xeon CPU, Centos 6.5 x86_64.
On my system Hyper-Threading is disabled.
WPS - WRF (ARW) - UPP was compiled with Intel Compilers:
Intel(R) Fortran Compiler XE 14.0 Update 3 for Linux
Intel(R) C++ Compiler XE 14.0 Update 3 for Linux
And with the following libraries:
netcdf-4.1.1 (I have tried also 3.6.3)
zlib-1.2.8 (I have also tried 1.2.3)
libpng-1.6.10 (I have also tried 1.5.51)
The following GrADS plots are from GRIB files produced by UPP from an ARW run (4Km grid size, Lambert projection) and I think these may explain the problem better than my description.
mpiexec -n 2 unipost.exe
mpiexec -n 12 unipost.exe
The cloud cover grid looks like segmented in N (= number of used cores) horizontal tiles.
Have you any suggestion that could help me resolving this?