Commit by Torsten Albrecht on dev :: r737a4a5 / (4 files in 2 dirs): (link)

Modified partgrid mass flux.

1) partgrid mass flux should symmetric at ice margin (in and out), also
for grounded margins
2) for floating ice boundaries next to ice free bedrock the flux should
vanish (that's a specific case butrelevant in regional ice shelf
simulations at PIK)
3) Minor changes in cfl_eigencalving routine - http://git.io/2w9ilQ

Commit by Torsten Albrecht on dev :: r18105ee / (4 files in 2 dirs): (link)

Modified partgrid mass flux.

1) partgrid mass flux should symmetric at ice margin (in and out), also
for grounded margins
2) for floating ice boundaries next to ice free bedrock the flux should
vanish (that's a specific case butrelevant in regional ice shelf
simulations at PIK)
3) Minor changes in cfl_eigencalving routine - http://git.io/PJmmdg

Commit by Torsten Albrecht on dev :: r25196b2 / (814 files in 129 dirs): (link)

Loaded up accidentally bin, build and lib files by using GitHub for Mac, now git rm them manually. - http://git.io/MMCAeg

Commit by Torsten Albrecht on dev :: r98862df / src/base/iMcalving.cc : (link)

Modified EigenCalving. - http://git.io/EK-Pug

Commit by Torsten Albrecht on dev :: r5e7d551 / src/base/iMcalving.cc : (link)

Modified EigenCalving. - http://git.io/6sBMzw

Commit by Torsten Albrecht on dev :: r3e7c045 / (815 files in 129 dirs): (link)

Modified eigenCalving.

1) There was a discrepancy in calculation of H_average (with respect to
the old mask and the newly updated H field)
2) In the calculation of principal strain rates as offset-neighbor
average avoid taking into account margin cells, which are affected by
the inward scheme in SSA.cc - http://git.io/sjAPYg

Commit by Torsten Albrecht on dev :: r6016cee / src/base/iMcalving.cc : (link)

Modified eigenCalving.

1) There was a discrepancy in calculation of H_average (with respect to
the old mask and the newly updated H field)
2) In the calculation of principal strain rates as offset-neighbor
average avoid taking into account margin cells, which are affected by
the inward scheme in SSA.cc - http://git.io/t9sbCg

Commit by Torsten Albrecht on dev :: rf3e110e / (3 files in 2 dirs): (link)

Reverted eigen_calving option(s) and moved cfl_eigencalving and nuBedrock options to pism_options.cc. - http://git.io/00BvCQ

Commit by Ed Bueler on dev :: rc3508da / doc/ice_bib.bib : (link)

Ref re inversion - http://git.io/BVwZEA

Commit by Constantine Khroulev on dev :: rc544561 / src/base/stressbalance/SSA.cc : (link)

Addressing issue #26 (by fixing typos in SSA.cc in stable0.4). - http://git.io/hpw6AA

Commit by Constantine Khroulev on dev :: r577fbdb / src/base/util/io/PISMNC3File.cc : (link)

Fixes a minor bug in PISMNC3File. - http://git.io/p76loQ

Commit by Torsten Albrecht on dev :: rca98e22 / (7 files in 3 dirs): (link)

1) Added adaptive timestep based on CFL-criterion for eigenCalving rate. 2) Fixed a bug in iMcalving concerning calved partially filled grid cells (vHref<0). 3) Changed eigencalving option, so that only -eigen_calving_K is needed in order to switch on do_eigen_calving. - http://git.io/h9l5gQ

Commit by David Maxwell on dev :: rc4272fa / (14 files in 4 dirs): (link)

Housekeeping of python inversion code.

Lots of code from examples/inverse has been properly promoted to code in site_packages/PISM

The unholy examples/inverse/pismssaforward.py has been refactored and is dead.

Python "executables" vel2tauc.py and a few others are now installed in the PISM binaries directory.

Still more PISMIO->PIO fixes (including fixing some of my broken fixes…) - http://git.io/u7Ta6g

Commit by David Maxwell on dev :: rb56980f / (3 files in 2 dirs): (link)

PISMIO -> PIO updates in a few python places - http://git.io/SR-UGA

Commit by Torsten Albrecht on dev :: ra06685e / (2 files): (link)

modification in CFBC when applied to a setup of an ice shelf confined in a bay of ice-free land avoiding the uphill creep of ice... - http://git.io/h-uFJw

Commit by Torsten Albrecht on dev :: r6dd5ed7 / (4 files in 3 dirs): (link)

changes concerning the ice schelf - bedrock interface - http://git.io/0FXM4w

Commit by David Maxwell on dev :: rfc7efaa / (5 files in 2 dirs): (link)

Language has been cleaned up in InvSSAForwardProblem. We are technically inverting for some parameter zeta, and tauc=tauc(zeta) (e.g. zeta*zeta or exp(zeta) or just plain old zeta). So InvSSAForwardProblem::set_tauc becomes InvSSAForwardProblem::set_zeta, etc.

InvSSAForwardProblem now allows zeta/tauc to be fixed at certain nodes (e.g. where ice is floating or absent). This can be turned on for now in vel2tauc with the option -use_zeta_fixed_mask; this unhelpfully named option is going to go away soon when this feature is turned on by default. This feature will be used for inversions in regional models, where tauc in the no_model_mask zone will be fixed.

plotting tweaks in vel2tauc

make_synth_ssa has new options: -misfit_type = grounded or fast to build a misfit functional based on all grounded ice or just the fast stuff (as determined by -fast_ice_speed) - http://git.io/39c-wg

Commit by David Maxwell on dev :: r5e182da / (10 files in 3 dirs): (link)

Addressed weirdness with mpiexec and ssa python tests:
Call mpiexec without an absolute path and it inherits a PATH from the environment.
Call it with an absolute path (/usr/bin/mpiexec) and you set the PATH to /usr/bin.
Our tests call mpiexec with an absolute path to ensure we use the correct version, but this means there is a path side effect.
Calling a python script with a shebang with mpiexec can then result in a different python executable being used than the user's standard environment would find. If a different python version is picked up (e.g. 2.6 rather than 2.7), you can fail to find python modules such as PISM and petsc4py if they were compiled for a mismatched version of python. Consequently, tests fail.
Now any mpiexec calls in the test suite that run a python script actually run python explicitly, with an absolute path to the user's python executable. (i.e /usr/bin/mpiexec /opt/local/bin/python my_test.py rather than /usr/bin/mpiexec mytest.py)

Fixed bug in parallel SSA test reporting messages.

All SSA tests are now run in parallel. Only prose was being run in parallel. Maybe I unintentionally broke this last summer. - http://git.io/m8YN6w

Commit by Constantine Khroulev on dev :: rc6a27ac / src/base/util/io/PIO.hh : (link)

Adds "#define MPI_INCLUDED" to trick NetCDF. - http://git.io/iQyc9A

Commit by Constantine Khroulev on dev :: r4b2f151 / (115 files in 26 dirs): (link)

Merge branch 'issue-2' into dev. Closes #2.

Finally!

Everything seems to work the way it should, but please try it out on low
resolution and/or short runs before submitting runs using thousands of
processor hours.

Most changes are deep inside and aren't visible. Here's a summary.

• Adds the PISMNCFile purely virtual class. This class defines the interface of the low-level NetCDF wrapper. There are two implementations: - PISMNC3File -- most of the old NCTool code, but re-wrapped and improved. - PISMNC4File -- the new code using NetCDF's parallel I/O. Unlike PISMNC3File, this class can read and write NetCDF-4 files only.
• The PIO class combines methods of NCTool related to metadata, plus adds methods previously contained in PISMIO. This class does not use NetCDF directly. Its constructor chooses one of the low-level NetCDF wrappers.
• There is one new command-line option: -o_format {netcdf3, netcdf4_parallel} and one new configuration parameter: output_format.
• Most PISM code should avoid creating PISMNC3File and PISMNC4File instances and use PIO instead.
• We always read data using the "netcdf3" wrapper to allow initializing from NetCDF-3 files even if "-o_format netcdf4_parallel" was selected. This can (and probably should) change. I can see adding the -i_format option if using parallel I/O for re-starting will save us some processor hours. I'm curious to learn if NetCDF parallel I/O is faster on pacman, which should have the hardware necessary to support MPI-IO. On computers I have access to right now it is significantly (about 2x) slower. I need to get on pacman and do some tests... (+6 more commits...) - http://git.io/U0Xl0w