* Starting to integrate jax with dash more and making the run_scm -jax flag a little more useful by being able to choose to run on the CPU or GPU via -jax=cpu or -jax=gpu rather than requireing environment variables
* Making the jax option even fancier, dash runner chooser is better/smarter now too
* Make CLUBB-JAX launcher portable across macOS and Linux
* Making the jax device detection work without needing the venvs setup first.
* Keep JAX radiation compiled with static timestep scheduling
Preserve the host radiation flag as a static JIT argument and mirror the Fortran conditional directly. Remove the Metal-only eager execution path and redundant branch helpers.
Cover both schedule states, statistics cadence, soil update ordering, and alternating time-dependent radiation calls. Validated with CPU and Metal tests and three-step standalone runs.
* Microphys loop push (#1355)
* First set of loop pushes - feature addition to stats too to allow for calling stats_update multiple times in a timestep and averaging the results, which is quite nice for silhs and allows us to get rid of the special microphysics stats type
* More loop pushes
* Tiny changes to restore BFBness
* Pushing loop into advance_microphys
* Little more loop push
* More basic loop pushing
* More loop pushing, also found what seems to be a bug with additional columns, theres code that uses only the first one. Im leaving this in for now to maintain BFBness, but labeled the problem areas with TODOs
* Potentially final batch of loop pushes
* Adding acc fixes for microphys stuff - the 2D versions called are GPUized, so they need acc data statements surrounding them
* Removing unneccesary single col versions of lots of things
* Restoring some comments and formatting
* A little more restoration, maybe unnecceary
* Little OpenACC fixes for microphys code RESOLVED:900681c91bdbdf8d6cd069bab01663ebac1a3279
* renamed jax jenkins test to clubb_jax_tests
* Making column mirror test work by comparing two different output files rather than one. This test now runs once with 3 different columns in a certain order, say ABC, and again with reverse order CBA, then ensures that A=A B=B and C=C (bit-for-bit equivalence) in both files. This should fail now actually, as there is currently a bug, and this update is intented to improve this test so that it fails before we fix the bug
* Making column mirror test use CPU. This makes it much faster and doesn't change the test at all - not sure why I made it use GPU originally.
* Fixing microphys bug that has been around for quite a while. This happened when making clubb standalone use multiple columns, and only affects columns other than the first. This is BIT_CHANGING, but only for microphysics cases using multiple columns (#1358)
* Test different f90 vs jax flags (#1357)
* wip
* wip
* test is basically done -> find out why some jax runs fail, mabe also run fortran even if jax fails
* fortran is tested regardless of whether jax fails
* changed jenkinsfile + removed l_Lscale_plume_centered from flag file + flag_set readme + lil print F90 fix
* added comments and a new flagset for jax tests
* added new jax flag config file to Jenkinstest
* Small UI fixes to make entry boxes refresh better
* Making python/jax driver status prints happen right away
* Removing accidentally added file
* Improving readmes
* Simpliciation cleanup
---------
Co-authored-by: Daniel Ebel <debel@uwm.edu> Co-authored-by: Daniel Ebel <124399436+shellwitz@users.noreply.github.com>
