Skip to content

8 - Run CAM-SIMA

It's time to try to run CAM-SIMA!

Configure CAM-SIMA

  • Either navigate to the case you created when you were validating your metadata or create a new case
  • Your user_nl_cam needs to contain the following (to bring in the cam_snapshot for input and to activate the validation tool):
ncdata='/FullPathName/your_before_snapshot_file.nc'
ncdata_check='/FullPathName/your_output_filename.nc'
debug_output = 0  !Turns off debug messages which interfere with the validation tool output

  • If the number of vertical levels in your snapshot file is something other than 30, then you will also need to add the following to your user_nl_cam:

    pver = N

    where “N” is the number of vertical levels in your snapshot file (likely 32 levels for CAM6)

Build and run

  • Run ./case.build and ./case.submit
    • Consult the table below for common errors. Also see debugging tips for help if you're getting errors
Error text Description Possible Fix
ERROR: cam_get_file: FAILED to get UNSET_PATH No initial data file specified (or defaulted to) Specify ncdata = '<file>' in your user_nl_cam file
ERROR: read_field_3d: No variable foun din (/XXX, /) Variable not found when we're reading in the initial data or snapshot file
  • Double-check your standard name
  • Determine if CAM-SIMA needs to handle/initialize the variable. If it does, follow the procedure for the error below
ERROR: Cannot read XXX from file, XXX has no horizontal dimension Variable is a scalar; we don't have an interface to read it in from a file (and it's not likely to be present on the file)
  • If "XXX" is static during the run, add access="protected" to the variable's XML entry in the registry
  • if the variable is not static, either:
    • Initialize the variable in the init phase of your scheme (intent=out), or
    • initialize "XXX" in CAM-SIMA (NOT in the physics code) and add a call to mark_as_initialized(<standard_name>) so CAM-SIMA doesn't try to read in the variable
ERROR in '<path/to/cesm.exe>': corrupted size vs. prev_size: 0xXXXXXXXXXXXXXX Possibly an out of bounds error
  • Find the line that the stack trace is pointing to (likely in the generated code) and trace the variable in question, looking for out-of-bounds errors
    • Specifically, check that the variable has the correct dimensions in the metadata (e.g. vertical_layer_dimension vs vertical_interface_dimension)
  • Also, check the variables immediately before and after the line for out-of-bounds errors (we have seen this error reported on the line immediately before the problem line).
  • Try running in DEBUG mode if not already to include bounds checking
  • Once the model completes without error, the results from the validation tool will appear in the atm_log file for each timestep (under the header ********** Physics Check Data Results **********)
    • The message No differences found! should be logged for each timestep.
    • If there are differences found, that indicates there are issues with your ported code
    • See the section on Tips for uncovering unexpected answer changes for debugging strategies

Additional step for NAG compiler

If we have a working NAG compiler with CAM-SIMA (not working as of 8/26/2024), perform this extra step:

  • Make at least one run using the NAG compiler on izumi in debug mode and specifying “-nan” as an additional compilation option. This option initializes all variables to NaN so that if something is uninitialized and is used, it will be trapped. To enable this flag:
    • After executing the ./create_newcase and ./case.setup commands and BEFORE executing the ./case.build, edit the file in your case directory cmake_macros/nag.cmake (- If you want to change the file so that it runs every time you execute a ./create_newcase, you can find the file at ccs_config/machines/cmake_macros/nag.cmake).
    • Change the line Append line so that it contains ‘-nan’ and it should look like this: if (DEBUG) string(APPEND FFLAGS " -nan -C=all -g -time -f2003 -ieee=stop") endif()

Once you have successfully run CAM-SIMA and all answer changes have either accepted or fixed, proceed to 9 - Bring back into CAM

Tips for uncovering unexpected answer changes

  • Make sure that your snapshot is taken with an updated atmospheric_physics, especially if answers changed during development
  • Check all of your standard_names. CAM-SIMA will run successfully, but if a standard_name is incorrect, you may be running with an incorrect variation of the variable that you intended
  • If comparing printed values on the physics grid between CAM and CAM-SIMA, then it might be good to turn-off chunking in CAM, which can be done by setting -pcols in CAM_CONFIG_OPTS to be a large number (e.g. >500 for the ne3 grid): ./xmlchange CAM_CONFIG_OPTS="-pcols 1352" --append (use the --append option to not lose existing options.)
    • You can also utilize the tools/find_max_nonzero_index.F90 tool to (more) easily compare values between CAM and CAM-SIMA at a specific rank and index that is known to be non-zero
  • Note that the way latitude in radians is calculated for the null dycore in CAM-SIMA is different (at a round-off level) to the way it is calculated in the SE dycore in CAM. This means if you are validating a physics scheme that has latitude (in radians) as an input, then in CAM you’ll need to change the scheme interface to read in latitude in degrees, and then convert to radians by multiplying by pi/180 (which you’ll only want to do when creating the snapshot file, not when validating CAM itself).
  • Sometimes CAM-SIMA will set a state variable directly, while CAM will only set a tendency variable and then use the “physics_update” routine to update the state using the tendency. This can result in round-off level differences. If this happens then simply try setting the state variable directly in CAM when creating the snapshot files for CAM-SIMA testing.
  • Namelist settings may be incompatible with what is being run in CAM-SIMA with it being a single physics package versus the complete CAM code. For instance, a normal CAM run may have a default namelist setting and then a different value for "clubb_sgs=1 microphys=mg3" which is what a CAM7 run would be. To temporarily work around this, set the values which are evaluating incorrectly in your user_nl_cam file for CAM-SIMA to the values which match a CAM run. The easy way to find these is to compare your atm_in values with your CAM and CAM-SIMA runs and add the appropriate settings.

  • Compare numbers between a run in CAM and another in CAM-SIMA. A debug run with the --pecount 1 argument provided to create_newcase gives you an executable which works well in the Totalview debugger.

    timestep discrepancy

    The cam_snapshot file will be written starting with the second timestep, so your CAM run will need to advance past the first timestep before you can start comparing.

  • If you run into a "tip" that should be added here, please confer with the other CAM SEs to get it added to the documentation!!