Converge/time

Tests before merging

• CI devel/trunk (passed with very small differences on both branches)
• GPU on JeanZay
• modipsl/LAM

CI devel/trunk

Very small differences on both branches.

• devel job: https://gitlab.in2p3.fr/ipsl/projets/dynamico/dynamico/-/jobs/799243
• devel plots: https://gitlab.in2p3.fr/ipsl/projets/dynamico/dynamico/-/jobs/799243/artifacts/file/spatial_differences.pdf
• trunk job: https://gitlab.in2p3.fr/ipsl/projets/dynamico/dynamico/-/jobs/799244
• trunk plots: https://gitlab.in2p3.fr/ipsl/projets/dynamico/dynamico/-/jobs/799244/artifacts/file/spatial_differences.pdf

GPU comparison

No differences, identical results.

• Reference GPU job: https://gitlab.in2p3.fr/ipsl/projets/dynamico/dynamico/-/jobs/799188
• Reference GPU output: https://gitlab.in2p3.fr/ipsl/projets/dynamico/dynamico/-/jobs/799188/artifacts/file/rundir/output_dcmip2016_regular.nc
• This merge request job: https://gitlab.in2p3.fr/ipsl/projets/dynamico/dynamico/-/jobs/799239
• This merge request output: https://gitlab.in2p3.fr/ipsl/projets/dynamico/dynamico/-/jobs/799239/artifacts/file/rundir/output_dcmip2016_regular.nc

modipsl/LAM

We compare start/restart files produced by the modipsl/LAM experiment with various versions of DYNAMICO:

• This merge request version: ed7431132f5a0888fa7e0a77f3d9134f071ebe8b (22)
• The last commit on DYNAMICO: 863853ff6674c73ca9b093be7741bc7865c4bcde (21)
• The official modipsl/LAM DYNAMICO version: 34d2c048093c7a608a36a4c0335848d1e08bfca2 (01)

21 vs 22 (last DYNAMICO vs this merge request)

• mr_compare_ICO_start_21_22.pdf - very small differences with some anomalies on the border (NaN); often similar in magnitude and pattern to differences in "01 vs 22" (in the next section)
• mr_compare_ICO_restart_21_22.pdf - mostly small differences, if the difference is big in absolute terms, relatively it is small; the biggest difference can be observed for wind u

01 vs 22 (last modipsl/LAM vs this merge request)

• mr_compare_ICO_start_01_22.pdf - similarly to "21 vs 22", very small differences and some anomalies on the border (NaN)
• mr_compare_ICO_restart_01_22.pdf - similarly to "21 vs 22", mostly small differences, if the value is big in absolute terms then it is small in relative terms

Reproducibility test 01 vs 01

As a reference for the possible magnitude and type of differences, we can analyze the start/restart file of the same experiment, but run twice from scratch. The graphs below shows that the differences we have seen so far are very similar to the differences obtained from the reproduction.

mr_compare_ICO_start_01_23_repro.pdf mr_compare_ICO_restart_01_23_repro.pdf