Skip to content

Releases: gemini3d/gemini3d

enhance robustness of build system

23 Jul 15:55
1e9ad70

Choose a tag to compare

  • add self tests
  • give better user feedback when a library is missing or not working

Add numerous checks, simplify CLI

19 Jun 02:01
0ed6885

Choose a tag to compare

  • gemini.bin command line now has only one mandatory argument: the top-level output directory. The inputs/config.nml should already be in there when gemini.bin is run.
  • if the user doesn't specify an output file_format, Gemini.bin will use the same file format as the input/simsize.*.
  • Check for NaNs and impossible values in several more places.
  • correct potential source calculation error
  • eliminate stack overflow in tests by using allocate()
  • allow non-MPI unit tests to build even without MPI
  • auto-download and auto-setup PyGemini if needed
  • eliminate duplicative specification of "flags" in config.nml. Use presence of namelists to determine if an input is desired.
  • allow namelists to be read in a different order than in file
  • move all Python scripts to PyGemini except for bare minimum used for self-tests
  • add optional EIA namelist
  • make several .f90 files generated to cut down on linking for minimal parameters (faster parallel builds)
  • add -dryrun to CI
  • update reference data URL

robustness: add dryrun

07 May 08:57
cfd4e86

Choose a tag to compare

  • add gemini.bin -dryrun used by job.py to help avoid HPC queuing for simple errors on first time step
  • Make job.py / plasma.py build msis_setup more reliably
  • cleanup python packaging
  • remove needless file copying from Fortran

bugfix: nml flag defaults

20 Apr 09:35
2a8ea3e

Choose a tag to compare

set safe defaults for NML flags. Under certain conditions they could default non-zero otherwise and cause simulation to hang early just after first time step.

remove unneeded code from job.py

job.py: allow manual specify max cpu count, handy for HPC where you be submitting to different node.

bugfix and add check: MPI count

19 Apr 18:58
4a8b6ad

Choose a tag to compare

Fixed a bug where MPI image count was incorrect on some systems, and added test to mpi_basic to check for this.

add implicit none (type, external) for robustness

bugfix: hdf5/netcdf detect

17 Apr 22:25
48d2e8d

Choose a tag to compare

  • general code quality via use of Fortran 2018 implicit none (external)
  • fix bug where on some systems h5fortran / nc4fortran / hdf5 / netcdf4 would falsely not be found
  • remove many preprocessed files instead using configure_file generated Fortran source for clarity
  • make msis_setup more robust to use from external Python / Matlab programs
  • do compiler/library checks with clearer failure messages
  • add more feature printouts to final summary to help users understand their configuration

bugfix, quality

01 Apr 22:25
da70c0a

Choose a tag to compare

  • CMake: enhancements to make CDash and console output prettier
  • Docs: split Readme into several linked files and update to current info
  • grid: enhance grid_check. Still need to make checks stronger for manual MPI count specification #17
  • use gemini_cfg for more parameters
  • general code quality and lint

Bugfix: grid size check, add cli flag parser, output.nml

27 Mar 22:27

Choose a tag to compare

add netcdf, CDash, better CLI, fix memory leak

25 Mar 20:47
ec15118

Choose a tag to compare

  • add CDash
  • add NetCDF4 option, by default off, while HDF5 is default on
  • improve Gemini.bin CLI to be flag-based
  • fix memory leak in vendor/lapack95 that would silently crash larger sims with SIGKILL 9
  • fix GCC 9.3.0 by removing redundant use

multi-type file io: raw, hdf5. cleanup

17 Mar 08:59

Choose a tag to compare

  • allow raw, hdf5, netcdf file IO without recompile
  • make MSIS-Python interface use stdin/stdout pipes instead of files
  • enable Fortran HDF5 write compression automatic
  • NetCDF is not yet fully working, needs to be debugged, but it's mostly implemented.
  • move Matlab, archive to own repos
  • add h5diff-based test--in case Python numeric libraries are messed up, h5diff is standalone test.