Add cmake option to not build OpenMP by default.

[jamiebresch]

On Cheyenne, when submitting jobs (for example, #PBS -l select=6:ncpus=32:mpiprocs=32), OMP_NUM_THREADS will default to the value of ncpus (=32 in this example). Either add ompthreads=xx setting to PBS, or setenv OMP_NUM_THREADS xx in the script to specify desired threads.
When mpas/mpas_jedi is compiled with OpenMP and the job is run with more than 1 OMP_NUM_THREADS, it hangs somewhere after ----- done checking limited-area settings ----- in log.atmosphere.0000.out.
This PR changes the default behavior of cmake from Enable OpenMP to disabled.

[mmiesch]

It looks reasonable, though I have not tested it

[svahl991]

Tested on Cheyenne with GNU.

[liujake]

FYI. I tried to build the model inside mpas-bundle without OpenMP using jedi-gnu modules. This does NOT solve the problem: the executable still can not do 30-km mesh forecast. For my case, it hangs at

 Reading streams configuration from file streams.atmosphere
 Found mesh stream with filename template static.nc
 Using default io_type for mesh stream
  ** Attempting to bootstrap MPAS framework using stream: static
 Bootstrapping framework with mesh fields from input file 'static.nc'

When I build the same source code with the traditional way using intel+mpt, it can run the forecast with the same initial condition file.

Related issues are documented at https://github.com/JCSDA-internal/mpas-bundle/issues/90.

[jjguerrette]

Is this ready to merge?

[jamiebresch]

Is this ready to merge?

I think so.

[jjguerrette]

FYI. I tried to build the model inside mpas-bundle without OpenMP using jedi-gnu modules. This does NOT solve the problem: the executable still can not do 30-km mesh forecast. For my case, it hangs at

 Reading streams configuration from file streams.atmosphere
 Found mesh stream with filename template static.nc
 Using default io_type for mesh stream
  ** Attempting to bootstrap MPAS framework using stream: static
 Bootstrapping framework with mesh fields from input file 'static.nc'

When I build the same source code with the traditional way using intel+mpt, it can run the forecast with the same initial condition file.

Related issues are documented at JCSDA-internal/mpas-bundle#90.

@liujake, I started seeing this same hanging issue in some mpas_hofx_nomodel runs. It is unrelated to the MPI_Finalize issue that is documented in JCSDA-internal/mpas-bundle#90. I will try a new build soon to see if the issue goes away.