Skip to content

Separate the Glide and Glissade dycores; remove obsolete code - #94

Closed
whlipscomb wants to merge 81 commits into
mainfrom
lipscomb/separate_glide_glissade
Closed

whlipscomb wants to merge 81 commits into
mainfrom
lipscomb/separate_glide_glissade

Conversation

@whlipscomb

@whlipscomb whlipscomb commented Nov 2, 2026 •

Copy link
Copy Markdown
Contributor

Branch: lipscomb/separate_glide_glissade (includes the earlier lipscomb/remove_externals commits)
Base: cism_main_3.00.002
81 commits, October 14 - November 7, 2026 (updated November 7: 11 commits added at the end; see Added November 7)

Summary

This PR does three things:

  1. Removes obsolete code and build options: serial (non-MPI) builds, the BISICLES and Albany-FELIX external dycore interfaces, the Trilinos solver interface, build scripts for retired machines, and the long-unsupported Glam dycore option.
  2. Separates the Glide and Glissade dycores. Before, both dycores shared one model type (glide_global_type), one config reader (glide_setup), one set of I/O definitions (glide_vars.def) and one diagnostics module, and each dycore used modules from the other. Now each dycore is self-contained, with its own types, setup, I/O and diagnostics modules; neither uses the other's code, and both use the shared code in libglimmer. Each copy was then trimmed to its own dycore. GLAD (used by CESM) now supports only Glissade, and Glint supports only Glide.
  3. Cleans up and renames: removes dead code and unused options, fields and modules, and renames the source directories libglide, libglissade, libglad and libglint to glide, glissade, glad and glint.

It also includes some lateral melt changes, several Glide bug fixes found while testing Glide after the split, and several Glissade bug fixes and cleanups found while preparing test baselines for the config reorganization.

The main motivation is the planned reorganization of the Glissade config file into short, physics-based sections. That is much easier with a Glissade-only glissade_setup (now about 4,000 lines) than with the shared glide_setup (formerly about 4,400 lines, covering both dycores). The Glide config format is frozen and will not be part of the reorganization.

Splitting the shared modules did not add much code, because each copy was trimmed to its own dycore. Line counts for the modules that were split:

File Start (cism_main_3.00.002) Glide, now Glissade, now
*_types.F90 4,376 1,292 3,570
*_setup.F90 4,430 1,255 3,972
*_diagnostics.F90 1,491 760 3,588
glissade_diagnostics.F90 (already existed) 2,245 — (merged into the row above)
*_vars.def 1,930 584 1,832
*_nc_custom.F90 225 198 236
Total 14,697 4,089 13,198

The total goes from 14,697 lines in 6 files to 17,287 lines in 10 files (+18%). For the whole PR, about 31,600 lines are deleted and 9,900 added: a net decrease of about 21,600 lines, mostly from removing the external dycore and Trilinos interfaces, the serial-build code, the old build scripts and the unsupported test cases.

Most commits do not change answers. The exceptions are listed under Answer changes below.

Main changes

Removal of obsolete code (from lipscomb/remove_externals)

  • Remove support for serial (non-MPI) builds; CISM is always built with MPI (single-task runs still work) (f3f2495)
  • Remove build scripts for retired machines; keep betzy-gnu, derecho-gnu, derecho-intel, linux-gnu and mac-gnu (e22c7aa)
  • Remove the BISICLES and Albany-FELIX external dycore interfaces (66b763b)
  • Remove the Trilinos solver interface (which_ho_sparse = 4 now gives a fatal error) (ea924cb)
  • Update the in-tree BATS regression scripts for Derecho (not yet tested by running BATS) (992433e)
  • Turn off GPTL in the Derecho build scripts, where a broken setting meant it was never used (3104810)

Lateral melt

  • Continued work on the lateral melt scheme, including removal of which_ho_calving_front = 2 (partial marine-grounded cells), new input fields runoff_applied and thermal_forcing_applied, and renamed variables (355cc53, 88dd672, db92563, e95206c)
  • Fix: always allocate latmelt_rate and latmelt_rate_tavg, which are used even without lateral melt (0ee9cde)

Glide fixes

  • Glide velocities were wrong by orders of magnitude, because of a spurious factor of scyr in cflow (probably introduced when the Glide scaling factors were removed). Also, Glide runs crashed at the first diagnostic output (unallocated ice_age). Both are fixed (8559727). Glide and Glint results change accordingly.
  • Fix the grid coordinates written by Glide runs with no input file, which made EISMINT-2 restarts fail (e45f336)
  • Dome and Halfar test configs: one config file per velocity solver (6c98aaa)

Separating Glide and Glissade

  • Decouple the shared netCDF I/O code in libglimmer from glide_global_type: I/O routines take the file lists and other values as arguments, and dimension sizes are set in the generated I/O code (d1f5a8f, f74a257, bf014ce, 329b554)
  • Remove each dycore's use of the other's modules (5f7d21f, 4749c5e, b953d2c, f4f4ff3, 432ca5c, f428478)
  • Add Glissade's own modules: glissade_types, glissade_setup, glissade_io (from glissade_vars.def), glissade_nc_custom and Glissade diagnostics (aee3889, 0dc3073, 7e7537d, 3f2973e, d8ed7f5)
  • GLAD supports only Glissade; Glint supports only Glide (ef2e775, 0fef8eb)
  • Make the model registry independent of the model type, and split the standalone driver into Glide and Glissade paths (d7550de, ba9295c, e3a6310)
  • Give Glissade its own model instance type, glissade_global_type, with subtypes glissade_*_type (18bf2ae, 7b4fffc)

Trimming each dycore to its own code

Renames

  • Rename the source directories libglide -> glide, libglissade -> glissade, libglad -> glad, libglint -> glint (b2b2643, rename-only). These have not been separate libraries for many years: CMake builds all source directories into one library.
  • isostasy*.F90 -> glide_isostasy*.F90, glide_ground.F90 -> glide_calve.F90, parallel_mpi.F90 -> cism_parallel.F90, profile.F90 -> cism_profile.F90 (f48575c, c55ac84, f462028)
  • libglimmer and libglimmer-solve are not renamed here. We are considering cism_shared and glimmer-solve or cism_solve, possibly with a glimmer _* -> cism_* renaming of files, modules and identifiers; this is deferred to a future PR.

Added November 7: bug fixes and cleanups from preparing test baselines

These commits came out of setting up standalone test baselines (including multi-task runs) for the planned config reorganization, and an audit of the April 2025 commits that removed CISM's internal scale parameters.

  • Fix the units of the external beta (which_ho_babc = 5). glissade_velo_higher passed beta*scyr to glissade_calcbeta instead of beta/scyr, so the basal traction was too large by a factor of scyr² (~1e15) and the bed was effectively frozen wherever the external beta was nonzero. Introduced in 9d16020 (April 2025). The slab test cases of Robinson et al. (2022) again match the analytic solution (172ba53).
  • Fix the scaling of several basal melt parameters (bmlt_float_const, bmlt_float_omega, bmlt_float_depth_meltmax, bmlt_float_depth_frzmax, bmlt_float_depth_meltmin), which were divided by scyr twice. Melt rates were too small by orders of magnitude, e.g. in MISMIP+ (1c59cbf). Also fix the depth test for maximum melting in the depth-dependent option, which compared the ice draft with a melt rate (meltmax) instead of a depth (zmeltmax); this did not change answers (fbb55e2).
  • Fix the timescale test for immediate calving with a calving mask (CALVING_GRID_MASK or apply_calving_mask). The test compared calving%timescale, which is in seconds, with 1.0 instead of 1 yr, so a timescale between 0 and 1 yr gave relaxed rather than immediate calving (8b3bac8).
  • Fix the units of acab in the EISMINT exact verification test (eismint_type = 4, Glide only), which was set in m/yr instead of m/s (b9bb654).
  • Make PCG the default sparse solver in Glissade (which_ho_sparse = 3, was 0 = serial SLAP), so that config files without which_ho_sparse can run on more than one task; and allow the local SIA solver, which uses no sparse solver, to run in parallel (785cb1a).
  • Remove tests/unsupported and utils/isostasy, which no longer compiled (984976a).
  • Update comments with obsolete references to Glide code in Glissade modules (239a961).
  • Tests: add dome config files for SSA and local SIA, use DIVA by default in ismip-hom, update the ISMIP-HOM URLs (the old sites are gone), and update the MISMIP+ scripts, README and default options (10ef85d, ba7388b, fe965a0).

Glint example

  • The Glint example test case (tests/glint-example, Greenland 20 km PDD and SMB cases) runs again; its README is updated for Derecho (ba81a07). The 5 km case is unstable with the available input file; this is documented.

Answer changes

  • CESM (GLAD with Glissade), at startup only (02eae8d): GLAD now finds ice-free ocean cells (where it sets acab = 0) from the current geometry, instead of from a mask that could be out of date at the first time step. Answers change slightly for cold starts whose input file has ice that is removed durin
    g initialization (e.g., icebergs or ice caps). In the aux_glc suite, this affects all tests with an Antarctic ice sheet, and the solid ice runoff in 2 cells in three Greenland-only tests. Restart comparisons pass.
  • Glide (standalone and Glint): answers change completely with the cflow fix (8559727), and at roundoff level with n_glen replacing the integer gn (c409e81).
  • Lateral melt: configurations with lateral melt turned on may change (355cc53 through e95206c). Configurations without lateral melt, including all aux_glc tests, are unchanged.
  • External beta (which_ho_babc = 5 with nonzero beta): answers change completely (172ba53). Not used in CESM.
  • Basal melt options using bmlt_float_const, bmlt_float_omega or the depth-dependent parameters (e.g., MISMIP+): answers change completely (1c59cbf). CESM's thermal-forcing option (bmlt_float = 6) is unchanged.
  • Config files without which_ho_sparse: single-task runs now use the PCG solver instead of SLAP, which changes answers at the level of the solver tolerance (785cb1a). CESM sets which_ho_sparse = 3 and is unchanged.
  • EISMINT exact test (eismint_type = 4, Glide): the SMB changes (b9bb654).
  • Calving mask with a calving timescale between 0 and 1 yr: ice is now removed immediately, as intended (8b3bac8). The default timescale is 0, and CESM does not use a calving mask.

All other commits are bit-for-bit, apart from the changes in output files and log files described below.

Changes for users

Building

  • CISM must be built with MPI. Build scripts are provided for betzy-gnu, derecho-gnu, derecho-intel, linux-gnu and mac-gnu.
  • After updating to this version, rerun cmake in existing build directories (source files are found with a glob, and several were renamed or moved).

Config files (Glissade)

  • Removed options, which are ignored if present: basal_water, vertical_integration, periodic_ew, hydro_time, basal_tract_max, basal_tract_slope , basal_tract_tanh, profile, expand_calving_mask, damage_threshold, forcewrite_restart. (forcewrite_restart was replaced by the hybrid restart option, restart = 2, in 2023.)
  • Fatal errors: dycore other than 2; gthf = 2; Glide evolution options (0-2); which_ho_sparse = 4; an [external_dycore_options] section.
  • The default evolution is now 3 (incremental remapping). Before, a config file with no evolution setting stopped with an error.
  • The default which_ho_sparse is now 3 (parallel PCG). Before, it was 0 (serial SLAP), which stopped with an error on more than one task.

Config files (Glide)

  • The Glide config format is frozen. Glissade-only sections ([ho_options], [glaciers], [basal_hydro], [lateral_melt], [grid_ocn], [grid_atm]) and options are no longer read; unused sections give a warning.
  • Huybrechts calving is now marine_margin = 5 (was 12).

Output

  • Glissade output fields removed (Glissade never computed them): wgrd, diffu, taux, tauy, dthckdtm, dusrfdtm, uflx, vflx, thkmask, glacier_area_factor, calving_lateral, tau_c. Requesting them gives a warning. The empty lithoz dimension and coordinate are no longer written. removal_rate_tavg is now available.
  • Glide output files no longer have Glissade fields or the empty Glissade coordinate variables.
  • The log no longer lists options that the dycore does not use.

Testing

  • Standalone (Derecho, intel): dome (DIVA, SLAP, Glide), dome restart tests (DIVA and Glide), Glide dome variants, Halfar, EISMINT-2 A and B, and (since ba81a07) the Glint example, with all output fields and log files compared to a baseline for every commit. Also EISMINT-1 and the Glide isostasy tests, and Antarctica (8 km, ISMIP6-style ocean forcing) and Alps (200 m, 3633 glaciers) restart runs, old vs new executable, for the commits that affect them.
  • CESM (with the matching CISM-wrapper branch): a 4-test aux_glc subset (intel and gnu) for commits that touch GLAD, the I/O or the build; and the full aux_glc suite at f3ebc16 (except the known NCK_Ly3 failure, CISM-wrapper issue #110), plus a mid-year restart test with time-averaged output (34 tests in all), compared to the cesm_timevars_step3 baseline (from the code of PR CESM-compliant time variables and time-average output; output options for an external driver #92): all builds, runs and restart comparisons pass, and all baseline differences are explained by the GLAD change above. The subset passes at b2b2643, compared to a baseline generated at f3ebc16.
  • Added November 7: new standalone baselines with multi-task runs in a PBS job: dome with DIVA, L1L2, local SIA, SSA (4 tasks) and BP (8 tasks), a 4-task DIVA restart test, Halfar with DIVA, ISMIP-HOM A and C at 20 and 80 km, the confined and circular shelves, and the two slab cases. At the head (8b3bac8): RESULT_STANDALONE. CESM 4-test subset at fbb55e2 compared to the f3ebc16 baseline: RESULT_CESM.

Notes for reviewers

  • Each commit is self-contained and was tested on its own; the commit messages describe the details and testing. The commits could be split into separate PRs (e.g., the remove_externals commits, the separation, and the renames) if preferred.
  • A matching CISM-wrapper PR (lipscomb/separate_glide_glissade) is needed: the wrapper uses CISM's I/O routines and GLAD, whose interfaces change here.
  • Planned follow-ups: the Glissade config reorganization (including renumbering Glissade as dycore = 1), new aux_glc tests for ocean forcing and glaciers, and possibly the libglimmer renaming.

🤖 Generated with Claude Code

whlipscomb and others added 30 commits October 19, 2026 17:02
This commit continues work on the lateral melt scheme, with several changes:

* I removed the option which_ho_calving_front = 2, which allowed for partial
  marine-grounded cells. I couldn't get this scheme to give sensible effective
  thicknesses at melt fronts. Marine-grounded cells are now considered to be
  full cells with thck_effective = thck; only floating cells can be partial.

* This change simplified the lateral melt logic, since it no longer needs to be
  patterned after the subgrid calving schemes. I moved the high-level lateral melt
  code from the calving module to the lateral melt module in a driver subroutine
  called glissade_lateral_melt_solve.

* I introduced a new 2d field, 'runoff', in the climate derived type.
  This is the meltwater runoff from the ice surface to the ocean in kg/m^2/s.
  It is loaded if present in the input file.

* I changed the option subglacial_discharge_from_ablation to runoff_from ablation.
  When set to T, CISM computes runoff based on the negative part of acab_applied.
  When set to F, CISM reads runoff from the input file (or else aborts).

* I changed subroutine glissade_subglacial_discharge so that runoff is an input
  and subglacial_discharge is an output. The units of subglacial discharge
  are now kg/m^2/s throughout the code.

These changes still need to be tested in various lateral-melt scenarios.
This commit renames some lateral melt variables to be consistent with the NORCE names.

There are two key input fields:
* runoff_applied (kg/m^2/s), the surface runoff applied as subglacial discharge
  at the grounding line
* thermal_forcing applied (deg C), the average thermal forcing over a given depth range

Both fields are now in the lateral_melt derived type. Either can be read from an input file.
If not in an input file, than runoff_applied can be computed from the surface ablation,
and thermal_forcing_applied can be computed from the 3d thermal_forcing.

I renamed thermal_forcing_2d to thermal_forcing_applied and removed the 'runoff' field.
This commit adds runoff_applied and thermal_forcing_applied (or thermal_forcing)
to the restart file as needed for the option which_lateral_melt = LATERAL_MELT_ISMIP6.
I verified exact restart.

For the case marine_margin = 1 (calve all floating ice), I added a call to subroutine
calve_ice_basic after the main lateral melt calculation.
This call removes any ice that is newly afloat after lateral melting.

I added the field melt_front_mask to the lateral_melt derived type and to glide_vars.def.

I changed some subroutine names to satisfy the convention that the 'glissade' prefix
is reserved for public subroutines.
This commit changes the names of lateral_melt%melt_thck and lateral_melt%melt_rate
to lateral_melt%latmelt_thck and lateral_melt%latmelt_rate.
This follows the convention (followed for most though not all input/output fields)
of having the same field name in glide_vars.def as within the derived type.

I also turned off some verbose diagnostics.
CISM is now always built with MPI (CISM_MPI and GLIMMER_MPI always defined).
Single-processor runs still work, using an MPI build with 1 task.

- Delete libglimmer/parallel_slap.F90
- CMakeLists.txt: remove options CISM_MPI_MODE, CISM_SERIAL_MODE,
  CISM_USE_MPI_WITH_SLAP and the unused CISM_BUILD_EXTRA_EXECUTABLES,
  along with the serial-mode source selection logic
- Delete the 6 *-serial-cmake build scripts, and remove the
  CISM_MPI_MODE/CISM_SERIAL_MODE flags from the remaining build scripts
- Update README files that referred to serial builds

Does not change answers.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Keep builds for betzy-gnu, derecho-gnu, derecho-intel, linux-gnu, and
mac-gnu. Remove builds for retired machines (cartesius, cheyenne, cori,
edison, grizzly, lisa, titan) and the linux-gnu-bisicles and
linux-gnu-felix builds. Scripts for new machines can be added on request.

Also fix the usage comment at the top of several remaining scripts.

Does not change answers.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The in-tree copy of BATS (tests/regression) supported only titan, hopper,
and cheyenne, whose build scripts have been removed. Add derecho support,
based on the derecho changes in https://github.com/Katetc/BATS
(commits 5385814 and 8af4bbb):

- util/dicts.py: add derecho_dict; remove titan, hopper, and cheyenne
- util/derecho_job.template: new PBS job template, used in place of
  util/job.template when platform is derecho. Loads the same modules as
  builds/derecho-intel/derecho-intel-modules (rather than intel-mpi, as
  in Katetc/BATS), to match the modules used for the build.
- util/runnit.py: choose a platform-specific job template if present;
  request full derecho nodes for the large and timing jobs; fix
  Python 2 constructs (iteritems, file.next) that fail under Python 3
- setup_derecho: load the derecho-intel modules plus conda;
  remove setup_titan, setup_hopper, and setup_cheyenne
- README.md: use derecho-intel as the worked example; fix some typos

Also fix builds/derecho-{intel,gnu}/derecho-*-cmake.sh to source the
modules file from the CISM source tree instead of the current directory,
so the scripts work for out-of-source builds (as used by BATS).

NOTE: These changes have not yet been tested by running the BATS scripts.
I checked only that the scripts parse under Python 3, that BATS finds the
derecho-intel build script, that a derecho job script is generated
correctly, and that derecho-intel-cmake.sh configures CISM in an
out-of-source build directory.

Does not change answers.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
These interfaces have not been used or maintained for many years.

- Delete libdycore/, libglide/felix_dycore_interface.F90, and
  cism_driver/cism_external_dycore_interface.F90
- glide_types: remove DYCORE_ALBANYFELIX (3), DYCORE_BISICLES (4),
  external_dycore_type, external_dycore_model_index, and dycore_input_file
- glide_setup: remove handle_dycore_options and the dycore 3/4 names and
  warnings. A config file with an [external_dycore_options] section now
  gives a fatal error, as does dycore = 3 or 4 (out of range).
- cism_front_end, glissade: remove the BISICLES and FELIX branches
- CMakeLists.txt, cism_driver/CMakeLists.txt: remove CISM_ENABLE_BISICLES
  and the BISICLES link logic; also remove the dead CISM_BUILD_SIMPLE_GLIDE
  option
- Build scripts: remove CISM_ENABLE_BISICLES, CISM_ENABLE_FELIX, and
  BISICLES_INTERFACE_DIR settings
- Remove other stale BISICLES comments

Does not change answers. Verified BFB for a 10-year dome test (all output
fields identical before and after this change).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Commit e95206c moved the allocation of lateral_melt%latmelt_rate and
latmelt_rate_tavg inside 'if (which_lateral_melt /= LATERAL_MELT_NONE)'.
But latmelt_rate is used unconditionally in glissade_mass_balance_diagnostics
and glide_write_diag, so runs without lateral melt used an unassociated
pointer. In CESM tests this caused failures with debug flags (intel and gnu)
and a segfault with gnu optimization; intel optimized runs happened to
survive.

Allocate these arrays unconditionally, as was done before e95206c.

Does not change answers (dome test BFB, with identical log output).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Delete libglimmer-trilinos/, glissade_velo_higher_trilinos.F90,
  utils/trilinos_config_scripts_examples/, and the trilinosOptions.xml
  files in tests/
- glissade_velo_higher: remove the #ifdef TRILINOS blocks and
  Trilinos-only variables
- Remove HO_SPARSE_TRILINOS (which_ho_sparse = 4); this option now gives
  a fatal error. Previously, in builds without Trilinos, it silently
  used a SLAP solver.
- print_options: check that which_ho_sparse is in range before using it
  as an array index
- CMakeLists.txt: remove CISM_USE_TRILINOS and C++ support; rename
  CISM_TRILINOS_LIBS (which also held the implicit Fortran link
  libraries) to CISM_IMPLICIT_LINK_LIBS
- Build scripts: remove Trilinos and C++ compiler settings
- Test scripts: no longer copy trilinosOptions.xml
- Update comments

Does not change answers (dome test BFB, with identical log output).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
derecho-intel-cmake.sh and derecho-gnu-cmake.sh set
CISM_USE_GPTL_INSTRUMENTATION=ON by default, but passed
'-D CISM_GPTL_DIR= "utils/libgptl"' (with a space after '='). This set
CISM_GPTL_DIR to an empty string, so GPTL was never actually used, and
cmake warned about the stray "utils/libgptl" argument.

Set CISM_USE_GPTL_INSTRUMENTATION=OFF, remove the CISM_GPTL_DIR line, and
add a comment on how to enable GPTL. Builds are unchanged (same compile
definitions), minus the cmake warning.

Note: The in-tree BATS scripts check the cmake cache for
CISM_USE_GPTL_INSTRUMENTATION=ON before performance runs (always on for
HPC platforms), so BATS on derecho now stops with a message that CISM
was not built with GPTL. Previously this check passed, but no timing
data were produced.

Does not change answers.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit fixes a major bug in glide_velo.F90.
In subroutine init_velo, the parameter cflow was initialized with an expression
that included a spurious factor of scyr. I think this factor was introduced
when the Glide scaling factors (thk0, vel0, etc.) were removed.
As a result, velocity computations were off by several orders of magnitude.
With the factor removed, the velocity computation is correct.

Also in Glide, I removed the redundant field model%velocity%bed_softness
(known as 'soft' in glide_vars.def). The basal traction coefficient is now
contained only in model%velocity%btrc (known as 'btrc' in glide_vars.def).
I removed btrac_const and btrac_max from the velowk derived type;
they are now part of the paramets derived type.

I added logic in glide_setup to prevent Glide from running with which_ho_ice_age = 1
(since Glide does not support ice age calculations).
I added logic in glide_diagnostics to turn off ice age diagnostics when which_ho_ice_age = 0.

I tested these Glide changes in dome tests (both no-slip and with slip_coeff = 1),
a Halfar test, and EISMINT-2 Experiment A. In these tests I was able to reproduce
earlier results within roundoff.
Other EISMINT-2 tests fail on restart; this still needs to be fixed.

Glissade answers are unchanged.
For the dome test, this commit replaces dome.config with four similar files:
dome.config.diva, dome.config.l1l2, dome.config.bp, and dome.config.glide.
This makes it easier to test multiple velocity solvers with the same test problem.
The default is dome.config.diva. (The old default, dome.config, used the BP solver.)

For the Halfar test, this commit replaces halfar.config and halfar-HO.config
with halfar.config.glide and halfar.config.diva. The new default is halfar.config.glide.
Recall that for the Halfar test, Glide is more accurate than any of the Glissade solvers.

I also modified runHalfar.py so that config files do not need semicolons before comments.
EISMINT-2 tests that start from the output of an earlier test (e.g., e2.b reads
e2.a.restart.nc) were aborting at startup with a fatal error:

  deltax1 of file e2.a.restart.nc does not match with config deltax: 9.9692100E+36 25000.0

The cause was in subroutine glide_nc_filldvars. When there is no input file
(as in e2.a), the grid coordinates x0, y0, x1 and y1 are computed from the grid
spacing. The old code, which dated from early Glimmer, wrote one value per
nf90_put_var call without a 'start' index, so each call overwrote the first
element. Only the first element was set (to the last value computed), and the
rest of each array was left as fill values. The x1 and y1 loops also stopped
one point short.

This commit computes each coordinate array in full and writes it with a single
parallel_put_var call, as is done when there is an input file. The coordinate
conventions are unchanged: x1 = (i-1)*dew and x0 = (i-1/2)*dew, and likewise
for y.

The fix changes only the coordinate values in output files of runs with no
input file. Model answers are unchanged. Glissade is not affected, since it
requires the coordinates to be read from an input file.

Tested with the full EISMINT-2 suite, which now runs through.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This is the first step in making libglimmer and libglimmer-solve independent
of glide_types, as part of the work to make Glide and Glissade self-contained.

The restart options (NO_RESTART, STANDARD_RESTART, HYBRID_RESTART) are now
defined in libglimmer/glimmer_ncdf.F90, and the sparse solver options
(HO_SPARSE_PCG_INCH, HO_SPARSE_BICG, HO_SPARSE_GMRES, HO_SPARSE_PCG_STANDARD,
HO_SPARSE_PCG_CHRONGEAR) in libglimmer-solve/glimmer_sparse_type.F90.
glide_types uses these modules and makes the constants public, so code that
gets them from glide_types is unchanged.

With this change, libglimmer-solve no longer uses glide_types.

No answer changes. Dome (DIVA with native PCG, DIVA with SLAP GMRES, and
Glide) is BFB with the previous commit, including log files. Standalone
Greenland (4 km) and Antarctica (8 km) restart tests are also BFB.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This is the second step in making libglimmer independent of glide_types.

The netCDF descriptors (glimmer_nc_stat) hold the sizes of the vertical and
other non-horizontal dimensions (nlevel, nstaglevel, nstagwbndlevel, nzocn,
nzatm, nglacier, nbasin, naxis). These sizes are used only by the generated
read and write loops in the *_io.F90 modules. Previously they were copied from
the model in glimmer_ncio.F90, in subroutines glimmer_nc_openappend,
glimmer_nc_createfile and glimmer_nc_openfile.

This commit removes those copies and adds a private subroutine,
NAME_set_nc_dims, to ncdf_template.F90.in. It is called in NAME_io_create
(for output files) and before each call to NAME_io_read, in NAME_io_readall
and NAME_read_forcing_once (for input and forcing files). Thus the sizes are
set when a file is defined or read, rather than when it is opened.

No public interfaces change, so no changes are needed in CISM-wrapper.

No answer changes. Dome (DIVA with native PCG, DIVA with SLAP GMRES, and
Glide) is BFB with the previous commit, including log files. Also BFB:
EISMINT-2 experiments A and B (B restarts from A); standalone Greenland (4 km)
and Antarctica (8 km) restart tests; and a 200-m Alps glacier restart test,
which includes read_once forcing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This is the third step in making libglimmer independent of glide_types.

The derived type holding a model instance's lists of netCDF input, output and
forcing files (glide_funits in glide_types.F90) is now glimmer_funits in
libglimmer/glimmer_ncdf.F90, alongside the glimmer_nc_input and
glimmer_nc_output types it points to. The component name is unchanged, so
model%funits%out_first, etc., work as before, in CISM and in CISM-wrapper.

This prepares for the next step, in which the subroutines in glimmer_ncio.F90
and glimmer_ncparams.F90 will take model%funits (and other values as needed)
as arguments, instead of the full glide_global_type.

No answer changes. Dome (DIVA with native PCG, DIVA with SLAP GMRES, and
Glide) is BFB with the previous commit, including log files.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This is the fourth step in making libglimmer independent of glide_types.
With this commit, no libglimmer or libglimmer-solve module uses glide_types.
(The template ncdf_template.F90.in still does; the generated *_io modules are
compiled into the dycore and coupling libraries, and the template will be
parameterized on the model type when Glissade gets its own I/O module.)

The subroutines in glimmer_ncio.F90 and glimmer_ncparams.F90 previously took
the full model instance (glide_global_type) as an argument. They now take the
model's file lists (model%funits, of type glimmer_funits) and only the other
values they need:

- openall_out:   funits, is_restart, projection, tstart, tstep_count
- closeall_out:  funits
- openall_in:    funits, projection, and the grid values used to check input
                 files (global_ewn, global_nsn, dew, dns, upn)
- closeall_in:   funits
- glimmer_nc_openappend, glimmer_nc_tavg_streams, glimmer_nc_write_tavg_state,
  glimmer_nc_restore_tavg_state: funits
- glimmer_nc_createfile: funits, projection
- glimmer_nc_openfile: projection and grid values (as for openall_in)
- glimmer_nc_checkwrite: funits, time, tinc, tstep_count. The time argument
  is now required.
- glimmer_nc_write_due: time, tinc, tstep_count
- glimmer_nc_write_timeslice: funits, time, external_time, tstep_count
- glimmer_nc_checkread: is_restart, and tstart, time, tinc and tstep_count.
  tstart, time and tstep_count are intent(inout), since they are reset when
  reading a restart file. The unused optional time argument is removed.
- glimmer_nc_readparams: funits, is_restart
- check_for_tempstag: a logical (true if temp, flwa and dissip are on the
  staggered vertical grid) instead of the dycore option

The callers are updated: ncdf_template.F90.in, glide.F90, glide_stop.F90,
glissade.F90, glad_initialise.F90 and glint_initialise.F90.

CISM-wrapper must be updated at the same time (glc_io.F90 calls
glimmer_nc_createfile, glimmer_nc_write_timeslice and glimmer_nc_checkwrite).

No answer changes. BFB with the previous commit, including log files:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide)
- Dome DIVA and Glide restart tests (5 years, then a standard restart to
  year 10), with history, time-average and restart files. The restarted runs
  are also BFB with 10-year runs without a restart.
- EISMINT-2 experiments A and B (B restarts from A)
- Standalone Greenland (4 km), Antarctica (8 km) and Alps glacier (200 m)
  restart tests; the Alps test includes read_once forcing

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step toward making Glide and Glissade self-contained dycores. After this
commit, no module in libglide uses a libglissade module.

- glide_bwater: Remove an unused 'use glissade_grid_operators' statement.
- glide_diagnostics: Replace calls to glissade_usrf_to_thck and
  glissade_rms_error with private copies, diag_usrf_to_thck and
  diag_rms_error. These are used only by Glissade diagnostics (rms errors
  relative to observed thickness and surface speed, and glacier inversion
  diagnostics). The copies are temporary; they will be removed when the
  Glissade-only diagnostics move out of glide_diagnostics.

No answer changes. BFB with the previous commit, including log files:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide)
- Dome DIVA and Glide restart tests (5 years, then a standard restart to
  year 10), with history, time-average and restart files
- EISMINT-2 experiments A and B (B restarts from A)
- Standalone Greenland (4 km), Antarctica (8 km) and Alps glacier (200 m)
  restart tests, which write rms-error and glacier inversion diagnostics

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step toward making Glissade independent of libglide procedures.

- glissade_velo: Remove the two calls to glide_set_mask that computed
  model%geometry%stagmask. Glissade does not use stagmask; it was only
  pointer-assigned in glissade_velo_higher (and was formerly an input to
  calcbeta).
- glissade_velo_higher: Remove the unused stagmask pointer.
- glissade_velo, glissade_therm: Remove unused 'use glide_mask' statements.
- glissade_velo, glissade_therm, glissade_basal_traction: Remove
  '#include "glide_mask.inc"'; these modules use none of its macros.

The remaining Glissade calls to glide_set_mask (for thkmask, in glissade.F90)
are removed in a following commit.

No answer changes. BFB with the previous commit, including log files:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide)
- Dome DIVA and Glide restart tests (5 years, then a standard restart to
  year 10), with history, time-average and restart files
- EISMINT-2 experiments A and B (B restarts from A)
- Standalone Greenland (4 km), Antarctica (8 km) and Alps glacier (200 m)
  restart tests

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step toward making Glissade independent of libglide procedures.

Glissade computed model%geometry%thkmask by calling glide_set_mask (three
times: at initialization, after initial calving, and each time step after
transport and calving), but never used it. Glissade uses its own masks
(glissade_get_masks). The only use was in GLAD, which zeroes acab over
ice-free ocean cells (GLIDE_IS_OCEAN).

- glissade: Remove the calls to glide_set_mask, the 'use glide_mask'
  statements and '#include "glide_mask.inc"'. At the same three points,
  call a new subroutine, glissade_set_glad_ocean_mask, which sets
  model%geometry%glad_ocean_mask = 1 for ice-free ocean cells, using the
  same test as GLIDE_IS_OCEAN.
- glide_types: Add glad_ocean_mask (temporary).
- glide_setup: Remove thkmask from the Glissade restart variables.
- glad_timestep: For Glissade, zero acab where glad_ocean_mask = 1; for
  Glide, keep GLIDE_IS_OCEAN(thkmask).

With this commit, no libglissade module uses glide_mask.

glad_ocean_mask is temporary. The mask set during initialization can be out
of date at GLAD's first step, if init later removes icebergs or ice caps; in
that case GLAD keeps acab in cells that are already ice-free ocean. A
following commit will have GLAD test the current geometry instead, changing
answers slightly at startup.

No answer changes. Glissade restart files no longer contain thkmask, and
thkmask is no longer meaningful as a Glissade output field.
Otherwise BFB with the previous commit, including log files:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide)
- Dome DIVA and Glide restart tests (5 years, then a standard restart to
  year 10), with history, time-average and restart files
- EISMINT-2 experiments A and B (B restarts from A)
- Standalone Greenland (4 km), Antarctica (8 km) and Alps glacier (200 m)
  restart tests
- CESM: 4 aux_glc tests (intel and gnu) compared against baselines

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ANSWER-CHANGING for CESM (GLAD) runs with Glissade, at startup only.
Standalone runs are not affected.

GLAD sets acab = 0 over ice-free ocean cells before each ice-sheet time step.
With Glissade, it found these cells with glad_ocean_mask (formerly thkmask),
which Glissade set at three points: during initialization, after initial
calving, and at the end of each time step (after transport and calving).
After a time step the mask matched the current geometry. At GLAD's first
step, however, the mask could be out of date: if initialization later
removed ice (e.g., icebergs or ice caps with remove_icebergs or
remove_ice_caps), GLAD did not zero acab in the newly ice-free ocean cells,
and positive SMB grew ice there.

Now GLAD tests the current geometry directly, zeroing acab where
thck <= 0 and topg - eus < 0. Remove glad_ocean_mask and subroutine
glissade_set_glad_ocean_mask. GLAD with Glide still uses
GLIDE_IS_OCEAN(thkmask).

CESM answers change slightly for cold starts whose input file has ice that
is removed during initialization. In the aux_glc subset, compared to
baseline cesm_timevars_step3:
- SMS_D_Ly1.f09_g17_gris4 (intel and gnu): solid ice runoff (rofi) differs
  in 2 cells (ice-free, bed below sea level, with isolated ice in the input
  file); the ice sheet state is identical.
- ERS_Ly5.f09_g17_ais8: the ice sheet state differs slightly after 5 years
  (e.g., thk normalized RMS ~1e-6).
- ERI_Ly9.f09_g17_ais8gris4 (noevolve_ais): no change.
Restart comparisons pass.

Standalone tests are BFB with the previous commit, including log files:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide)
- Dome DIVA and Glide restart tests (5 years, then a standard restart to
  year 10), with history, time-average and restart files
- EISMINT-2 experiments A and B (B restarts from A)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step toward making Glissade independent of libglide procedures.

Glissade computed the diagnostic vertical velocity, wvel, by calling
wvelintg in glide_velo. This required calling init_velo during
initialization to set up the work arrays in model%velowk; the rest of
init_velo is needed only by the Glide SIA solver.

- glissade_velo: Add subroutine glissade_wvelintg, adapted from wvelintg.
  The velowk work arrays (dupsw, depthw, suvel, svvel) are now local
  arrays, computed from numerics%sigma with the same expressions as in
  init_velo.
- glissade: Call glissade_wvelintg instead of wvelintg, and remove the call
  to init_velo. Glissade no longer uses glide_velo or model%velowk.

No answer changes. BFB with the previous commit, including log files:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide)
- Dome DIVA and SLAP runs with wvel added to the output (wvel is not in the
  standard dome output): wvel is bitwise identical
- Dome DIVA and Glide restart tests (5 years, then a standard restart to
  year 10), with history, time-average and restart files
- EISMINT-2 experiments A and B (B restarts from A)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step toward making Glissade independent of libglide procedures.

Glissade initialized its diagnostics (the diagnostic output interval ndiag
and the local indices of the diagnostic point) by calling glide_init_diag
in glide_diagnostics.

- glissade_diagnostics: Add subroutine glissade_init_diag, a copy of
  glide_init_diag.
- glissade: Call glissade_init_diag instead of glide_init_diag.

Glide still calls glide_init_diag. Glissade no longer uses
glide_diagnostics.

No answer changes. BFB with the previous commit, including log files:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide)
- Dome DIVA and Glide restart tests (5 years, then a standard restart to
  year 10), with history, time-average and restart files
- EISMINT-2 experiments A and B (B restarts from A)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step toward making Glissade independent of libglide procedures.

The option gthf = 2 (GTHF_COMPUTE) computes the geothermal heat flux by
solving for the lithosphere temperature (glide_lithot). With Glissade, this
worked only on one processor (not_parallel). The lithosphere temperature
calculation is now supported only for Glide.

- glissade: Remove the calls to init_lithot and calc_lithot, and the use of
  glide_lithot and not_parallel.
- glide_setup: Stop with a fatal error if gthf = 2 with the Glissade dycore.

The lithot arrays and the litho_temp restart variable are allocated and
added only for gthf = 2, so they are not affected.

No answer changes for supported options. Glissade runs with gthf = 2 now
stop at setup. BFB with the previous commit, including log files:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide)
- Dome DIVA and Glide restart tests (5 years, then a standard restart to
  year 10), with history, time-average and restart files
- EISMINT-2 experiments A and B (B restarts from A)
- Dome Glide with gthf = 2
Dome DIVA with gthf = 2 stops with the new error message.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step toward making Glissade a self-contained dycore.

- Add libglissade/glissade_types.F90. For now it simply re-exports
  glide_types; the Glissade types will be defined here when Glissade gets
  its own model instance type.
- In libglissade, replace all 'use glide_types' statements with
  'use glissade_types' (34 statements in 23 files), keeping 'only' lists.

The CMake build finds libglissade sources with file(GLOB), so no CMake
change is needed, but existing build directories must rerun cmake once
to pick up the new file.

No answer changes. BFB with the previous commit, including log files:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide)
- Dome DIVA and Glide restart tests (5 years, then a standard restart to
  year 10), with history, time-average and restart files
- EISMINT-2 experiments A and B (B restarts from A)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step toward making Glide and Glissade self-contained dycores. CESM runs CISM
only with Glissade, so GLAD (the coupling layer used by CESM) no longer
supports Glide.

- glad_initialise: Stop with a fatal error if dycore is not Glissade.
  Remove the Glide branch (glide_initialise, glide_init_state_diagnostic).
- glad_timestep: Remove the Glide branches: the glide_tstep_p1/p2/p3 calls
  and the GLIDE_IS_OCEAN(thkmask) test for ice-free ocean cells. Remove
  '#include "glide_mask.inc"'.

GLAD still gets glide_config, the glide_io accessors and glide_finalise
from libglide; these will move to Glissade modules in later commits.

No answer changes. BFB with the previous commit:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide), dome DIVA
  and Glide restart tests, and EISMINT-2 experiments A and B, including log
  files (standalone runs do not use GLAD)
- CESM: 4 aux_glc tests (intel and gnu) compared against baseline
  dycore_split_3b2

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step toward making Glide and Glissade self-contained dycores. Glint (the
older GCM coupling layer, with built-in downscaling) is kept, but as in the
original Glimmer it now supports only Glide. GLAD is the coupling layer for
Glissade.

- glint_initialise: In both initialization subroutines, stop with a fatal
  error if dycore is not Glide, and remove the Glissade branch
  (glissade_initialise, glissade_diagnostic_variable_solve).
- glint_timestep: In both time-step subroutines, remove the Glissade branch
  (glissade_tstep).
- Remove the 'use glissade' statements.

With this commit, no Glint module uses a libglissade module.

No answer changes. The Glint test (tests/glint-example, whose configs
already use dycore = 0) was not run; testing Glint is deferred until later.
Standalone runs do not use Glint. BFB with the previous commit, including
log files:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide)
- Dome DIVA and Glide restart tests (5 years, then a standard restart to
  year 10), with history, time-average and restart files
- EISMINT-2 experiments A and B (B restarts from A)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Step toward making Glissade a self-contained dycore, and toward the
reorganization of the Glissade config file (which will be done in
glissade_setup only; the Glide config format is frozen).

- Add libglissade/glissade_setup.F90, a verbatim copy of glide_setup.F90
  except that:
  - the public subroutines are renamed glissade_readconfig,
    glissade_printconfig, glissade_scale_params, glissade_load_sigma,
    glissade_read_sigma, glissade_calc_sigma, glissade_get_zocn and
    glissade_get_zatm (check_fill_values keeps its name);
  - it uses glissade_types instead of glide_types, and does not include
    glide_mask.inc (whose macros it does not use);
  - it adds glissade_config, a copy of glide_config (in glide.F90).
  Glide-only options will be removed from glissade_setup later.
- glissade: Use glissade_setup instead of glide_setup.
- cism_front_end: Read the dycore option from the [options] section first,
  then call glide_config for Glide and glissade_config otherwise.
- glad_initialise: Call glissade_config instead of glide_config.

Glide and Glint still use glide_setup and glide_config. Both config
readers still add restart variables with glide_add_to_restart_variable_list
(glide_io); Glissade will get its own I/O module in a following commit.

No answer changes. BFB with the previous commit, including log files:
- Dome (DIVA with native PCG, DIVA with SLAP GMRES, and Glide)
- Dome DIVA and Glide restart tests (5 years, then a standard restart to
  year 10), with history, time-average and restart files
- EISMINT-2 experiments A and B (B restarts from A)
- Standalone Greenland (4 km), Antarctica (8 km) and Alps glacier (200 m)
  restart tests
- CESM: 4 aux_glc tests (intel and gnu) compared against baseline
  dycore_split_3b2

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The pattern 'cism_driver' (meant for the cism_driver executable and the
links to it in test directories) also matched the top-level cism_driver/
source directory. Tracked files there stayed tracked, but 'git add' with
an explicit path under cism_driver/ printed a warning and returned an
error, and new source files there would have been ignored.

Add '!/cism_driver/' to re-include the top-level source directory. The
executable (e.g., builds/*/cism_driver/cism_driver) and the cism_driver
links in test directories are still ignored.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
whlipscomb and others added 4 commits November 2, 2026 09:10
The Glint example (Greenland 20 km, PDD and SMB cases) runs correctly
again. It could not run on the code before the Glide/Glissade split:
Glide crashed in its diagnostics (unallocated ice_age), and Glide
velocities were wrong by orders of magnitude (spurious factor of scyr
in cflow); both were fixed in 8559727. Since c409e81, the 20 km
results are unchanged (identical in all fields).

README.md:
- Note that the input files can be copied from the CESM inputdata
  directory on Derecho, and add the Derecho build path
- The 5 km case now uses gland5.input.nc from CESM inputdata (instead of
  Greenland_5km_v1.1.nc)
- Note that the 5 km case is currently unstable with gland5.input.nc
  (not investigated); the 20 km cases are used to test Glint
- Glint is no longer described as required for coupling to a climate
  model (CESM uses GLAD)

greenland_5km.config.pdd: read gland5.input.nc.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Rename isostasy.F90, isostasy_elastic.F90 and isostasy_kelvin.F90 in
libglide to glide_isostasy.F90, glide_isostasy_elastic.F90 and
glide_isostasy_kelvin.F90, and rename the modules accordingly, for
consistency with the other Glide modules (and with glissade_isostasy*
in Glissade).

Existing build directories need to rerun cmake, since CMake collects
the source files with a glob.

No answer changes.

Tested with the standalone tests (identical to baseline 4f87262) and
the four Glide isostasy configs (tests/EISMINT/isos, 5000 years), old vs
new: identical outputs and logs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
In libglimmer:
- Remove four modules that are not used anywhere in CISM or
  CISM-wrapper: glimmer_map_proj4, glimmer_searchcircle, glimmer_ts and
  ncdf_utils
- Rename parallel_mpi.F90 to cism_parallel.F90, to match the name of the
  module it contains (cism_parallel)
- Rename profile.F90 to cism_profile.F90, and the module profile to
  cism_profile, and update the 'use' statements

Existing build directories need to rerun cmake, since CMake collects
the source files with a glob.

No answer changes.

Tested with the standalone tests: identical to baseline 4f87262.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Rename the source directories
  libglide    -> glide
  libglissade -> glissade
  libglad     -> glad
  libglint    -> glint

The 'lib' prefix suggested separate libraries, but these directories
have not been built as separate libraries for many years: CMake builds
all the CISM source directories into a single library
(glimmercismfortran), which is linked into cism_driver (and used by
CESM). The new names match the module prefixes in each directory.

This is a rename-only commit: the files are moved with git mv, and the
only content changes are path updates in CMakeLists.txt, the scripts in
utils/build, a few READMEs and comments, and the (unbuilt) Makefile.am in
tests/unsupported/exact-isothermal. Existing build directories need to
rerun cmake.

The libglimmer and libglimmer-solve directories are not renamed here.
We are considering renaming libglimmer to cism_shared and
libglimmer-solve to glimmer-solve or cism_solve, possibly together with
renaming the glimmer_* files, modules and identifiers with a cism_
prefix. This decision is deferred to a future PR.

No answer changes.

Tested with:
- A fresh out-of-source build and the in-place build
- Standalone tests: identical to baseline 4f87262
- CESM: 4 aux_glc tests (intel and gnu) compared against baseline
  dycore_split_phase7: all pass, including NLCOMP and the baseline
  comparisons (this also covers f462028)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
whlipscomb and others added 11 commits November 2, 2026 16:34
Creating a more complete set of dome config files to test each
of the most important Glissade approximations.
Set which_ho_approx = 4 (DIVA) in ismip-hom.config, so that the
ISMIP-HOM experiments run with DIVA by default instead of BP.

The ISMIP-HOM web pages at homepages.ulb.ac.be and homepages.vub.ac.be
no longer respond. In the README and in the messages and comments of
runISMIP_HOM.py and plotISMIP_HOM.py, refer instead to the paper
(Pattyn et al., 2008, https://doi.org/10.5194/tc-2-95-2008) and to its
supplement on the Copernicus site, which contains the results of the
other models. The README now explains that the supplement zip file
contains tc-2007-0019-sp2.zip, which unzips to the ismip_all directory.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With which_ho_babc = 5 (HO_BABC_BETA_EXTERNAL), beta is read from the
input file in Pa yr/m and stored in model%velocity%beta in SI units
(Pa s/m; factor = 1.0/scyr in glissade_vars.def). glissade_calcbeta
expects Pa yr/m, but glissade_velo_higher passed beta*scyr instead of
beta/scyr, making the basal traction too large by a factor of scyr^2
(~1e15). As a result, the bed was effectively no-slip wherever the
external beta was nonzero.

The bug was introduced in 9d16020 ("Removed scale parameters tau0,
vis0, and evs0", April 2025), which changed scale_beta from tau0/scyr
to 1/scyr; the argument beta*tau0/scyr should have become beta/scyr.

With the fix, the slab test cases of Robinson et al. (2022) (thin
sliding and thick shearing) again agree with the analytic solution
and with results from March 2024. In the circular shelf test, the
velocity at the four pinned center vertices (beta = 1e8 Pa yr/m)
changes from ~1e-18 to ~1e-3 m/yr. Answers are unchanged for runs
with which_ho_babc /= 5, and for runs with zero external beta.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tests/unsupported held older shallow-ice test cases (benchmarks, erosion,
exact-isothermal, lin-non_lin) that have not been supported for many
years; its README says "Use at your own risk!". The exact-isothermal
test used old autotools files, referred to Trilinos, and still used the
scale parameters len0, thk0 and vis0, which were removed from
glimmer_paramets in April 2025, so it no longer compiles.

utils/isostasy held relaxed_topo.F90, a utility from the original
Glimmer release to add relaxed bedrock topography to input files. It is
not built by CMake, calls glimmer_nc_read (which no longer exists) and
uses len0, so it no longer compiles either.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
In exact_surfmass (eismint_type = 4, Glide only), acab was set to
M*SperA (m/yr) inside the ice and to -0.1 (meant as 0.1 m/yr ablation)
outside, but acab has internal units of m/s. The mass balance M from
the testsFG module is already in m/s, so acab is now set to M, and the
ablation outside the ice to -0.1d0/scyr.

This bug predates the April 2025 removal of the scale parameters;
before that, acab was off by the old scale factor scale_acab. The
exact test is not part of the supported test suite, and this fix
has not been tested in a run. Other answers are unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…n parallel

Change the default of which_ho_sparse from 0 (SLAP biconjugate gradient,
serial only) to HO_SPARSE_PCG_CHRONGEAR (3, native parallel PCG). With
the old default, a Glissade run on more than one task stopped with
"SLAP solver not supported for more than one processor" unless the
config file set which_ho_sparse. CESM already sets which_ho_sparse = 3,
and all the standard test configs set it explicitly, so their answers
are unchanged. Single-task runs whose config files do not set
which_ho_sparse will now use the PCG solver instead of SLAP.

Also skip that check when which_ho_approx = HO_APPROX_LOCAL_SIA, since
the local SIA solver does not use a sparse linear solver. Update the
comment for which_ho_sparse in the dome local SIA config.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit modifies obsolete references to Glide code in several Glissade modules.

No answer changes
Several basal melt parameters were erroneously divided by scyr twice:
first in subroutine glissade_scale_params, then again locally in
glissade_bmlt_float.F90. This resulted in melt rates that were too small
by several orders of magnitude in MISMIP+ and other applications.

I removed the scaling in glissade_scale_params. Now the scaling is done
only once, with comments noting a division by scyr to convert to mks units.

This commit is answer-changing for MISMIP+, whose results are now as expected.
This commit updates the scripts, default options, and README file
for the MISMIP+ experiments.

I updated several default options:
* upn = 5 instead of 3
* which_ho_nlinear = 1 (Picard acceleration) instead of 0
* restart = 2 (hybrid restart) instead of 1 for all experiments except the Spinup
* which_ho_flotation_function = 2 (replacing the old option 3)

When I first ran the experiments, the melt rates were too small by several
orders of magnitude because there was one too many divisions by scyr.
This was fixed in a preceding commit.

After the bug fix, the results for all experiments are as expected,
similar to those described in the MISMIP+ paper by Cornford et al. (2020).
…oat option

In the depth-dependent basal melt scheme in glissade_bmlt_float.F90, the
branches for maximum melting (warm-ocean and standard profiles) compared
the ice draft z_draft (m, negative) with the melt rate
bmlt_float_depth_meltmax instead of the depth bmlt_float_depth_zmeltmax.
Since the preceding branches cover z_draft > zmeltmax and meltmax is a
non-negative rate, the test was always true and acted like a plain else,
so answers should be unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With a binary calving mask (CALVING_GRID_MASK or apply_calving_mask,
without a subgrid calving front), ice in masked cells is removed
immediately if calving%timescale is at most 1 yr, and otherwise thinned
with a relaxation timescale. The test compared calving%timescale with
1.0, but calving%timescale has units of s (it is converted from yr in
glissade_scale_params), so the test was effectively timescale = 0.
Now the test compares with scyr, as intended.

This changes answers for runs with a calving mask and a calving
timescale between 0 and 1 yr. The default timescale is 0, and CESM runs
do not use a calving mask, so they are unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@whlipscomb

Copy link
Copy Markdown
Contributor Author

Superseded by #97 (merged) and the stacked series #99–#104, which contain the same changes split into smaller PRs, plus a few later commits.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants