MITgcm / verification_other

Additional verification experiments for MITgcm
MIT License
10 stars 19 forks source link

Post pr611 ad output update #54

Closed jm-c closed 2 years ago

jm-c commented 2 years ago

After merging PR 611 ( https://github.com/MITgcm/MITgcm/pull/611 ), update all 4 global_oce_llc90 ADM ref output.

This is a side effect of using single prec tapes (isbyte=4), since switching to local tapes in gad_advection.F with less recomputations results in differences (vs recomputing some terms in full prec). For the same reason (isbyte=4), AD-monitor in exp. global_oce_cs32 are also affected but not as much (not updated here).

Note: AD-grad (or Cost or FWD grad) were not affected, except for ecco_v4 AD-test with only 6 matching digits remains (--> fail to pass). So updating the 3 others ADM outp was not strictly necessary. However, since AD-monitor were affected in all 4 tests (with only 5 matching digits remaining for some of them) this would have made detection of future changes more tricky to detect without this update.