1. 11 Mar, 2022 2 commits
  2. 25 Feb, 2022 2 commits
  3. 05 Jan, 2022 2 commits
  4. 07 Oct, 2021 7 commits
  5. 02 Sep, 2021 2 commits
    • Christopher Subich's avatar
      Add system file for the new belize3 · d63d8368
      Christopher Subich authored
      This adds belize3.sh, a system file for the newly-built belize3, using
      the g++ compiler.
      As a usage note, blitz requires python to build successfully.  As of
      this writing, python is only available on belize3 with `module load
    • Christopher Subich's avatar
      Update dependencies to Blitz 1.0.2 and update to new MPI standard · 937c04ae
      Christopher Subich authored
      This commit updates make_deps.sh to build newer versions of Blitz (1.0)
      and fftw (3.3.9).  The 2010 version of Blitz no longer builds with the
      newest versions of GCC, and there is no harm in updating FFTW.
      The newest versions of Blitz (github head) require cmake to
      build, and some systems still in use have old, incompatible versions of
      cmake.  Blitz 1.0.2 requires python to complete its build process (some
      files are auto-generated).
      The new version of Blitz dropped the blitz::Vector type, being
      equivalent to a 1-dimensional array, so it was replaced with
      blitz::Array<type,1> instead.
      Additionally, the Parformer (parallel transpose) code was adjusted to no
      longer use MPI_UB to define the upper bound of the array section
      datatype.  MPI_UB was deprecated in the MPI-2 standard and removed in
      the MPI-3 standard, so SPINS could fail to build with some new MPI
      libraries (OpenMPI 4.0.3 is confirmed to break).
      Because this commit updates dependencies, the patch version is
  6. 12 Jul, 2021 2 commits
  7. 21 May, 2021 2 commits
  8. 30 Apr, 2021 2 commits
  9. 16 Apr, 2021 2 commits
  10. 11 Jan, 2021 13 commits
  11. 02 Jun, 2020 1 commit
  12. 27 May, 2020 2 commits
    • Christopher Subich's avatar
      wave_reader: compute hill properly with z-decreasing · 4fba6227
      Christopher Subich authored
      Previously, wave_reader computed the hill height as a fixed
        zgrid(:,1,1) - MinZ // matlab notation
      which implicitly assumes that z increases along the third index.  If
      that assumption is violated, the BPE calculation breaks (in ways that
      can cause assertion failures based on roundoff error, at that).
      Fixing the calculation as min(zgrid(i,1,:)) at initialization time
      alleviates this problem.
    • Christopher Subich's avatar
      wave_reader: allow FULL input types · 471ccd94
      Christopher Subich authored
      Previously, wave_reader locked 'FULL' (3D) input behind enabling
      restart.  This restriction was unnecessary, and doing so in fact made
      the specification of u/v/w/rho filenames useless.
      Deleting this check allows proper read-in of FULL datatypes, without
      faking a restart.  The only functional difference is that if
      perturbations are also enabled, they will be applied on top of the 3D
      data read.  This seems to be desirable, since it makes the application
      of the config-file settings more straightforward.
  13. 04 May, 2020 1 commit
    • Christopher Subich's avatar
      Updates to solver parameters · 79b17389
      Christopher Subich authored
      Based on experience with a poorly-converging, steep hill case, the
      number of inner iterations in GMRES is increased to 40, and the number
      of outer iterations drops to 2.
      Please pay attention to GMRES warnings and report cases where things
      don't work well, there's probably an optimal parameter range (or else we
      could make the parameters a runtime config option).