Changes

#12 (Oct 6, 2025, 3:55:06 AM)

Must add -fopenmp back for now, since this is used for the openmp test and it also broke the BFB tests for an obnoxious reason - that test compiles each commit without cleaning the build, so the baseline commit had -fopenmp and compiled with openmp stuff enabled, but then the subsequent commits had it turned off, but no changes in the files, so it didn't recompile with it off, and only turned it off in the linking stage, causing a build error. RESOLVED:320551df13ee84bdb9280db0be788a36e1f03494 RESOLVED:6c0201d44694fadf7f4bb0d4d58c3a9871096a22 RESOLVED:26644de2ea83ee6335b273f54dd57afc8f6f4b0c — Gunther Huebler / githubweb

Adding -Mstack_arrays to the nvfortran flags causes us to use more stack space, which means we need to allow processes to use more stack space (when using lots of columns, since that increases memory footprint). We do by using using 'ulimit BIGNUMBER'. — Gunther Huebler / githubweb

Adding variables to openacc data statements. Without this, the code crashes when using the GPU with l_godunov_upwind_xpyp_ta=.true. — Gunther Huebler / githubweb
Modifying compiler configurations (mainly optimization level) in order to prepare for switching scipts and tests to cmake, specifically separating out this BIT_CHANGING commit from the rest — Gunther Huebler / githubweb
thlm_ed and rtm_ed caused GPU errors since those variables were added to a acc loop, but never added to the acc data statement. This change makes it so they're only calculated on the CPU, since they are only used for stats, fixing the issue without adding the GPU memory footprint. — Gunther Huebler / githubweb

Making restart test not overwrite time_initial. The time_initial should be defined by the _model.in file and not overwritten, but this script was setting time_initial to time_restart, which resulted in the initial iteration value (iinit) to be 1 in the restart regardless of the restart time. — Gunther Huebler / githubweb
Making the l_random_k_lh_start set the random generator seed before generating a random. This should fix the generalized grid test, which failed for the only flag set that turned on l_random_k_lh_start. Before, the test passed because we used the same random seed for all timesteps, preventing differences in the random call. But now we use a different seed each timestep, causing differences. To ensure reproducibility, we use the seed each timestep to generate rand_pool, which is an effective solution, but we need to do that for the rand_integer_in_range call in the l_random_k_lh_start code for the same reason. Since this does change the random that comes out, this is BIT_CHANGING, but only for runs using both SILHS and l_random_k_lh_start=.true. — Gunther Huebler / githubweb

Bug fix. This is BIT_CHANGING, but only on GPUs and only when l_soil_veg=.true. (which is only gabls3 currently). — Gunther Huebler / githubweb

Change tolerance in way that affects only single precision builds (#1256) — noreply / githubweb

Making cmake only optionally use ninja. Now we only use it if it's available. — Gunther Huebler / githubweb