[MITgcm-devel] small changes in output from version 3.5.2 and new version 3.5.13
Jean-Michel Campin
jmc at mit.edu
Thu Mar 2 00:43:14 EST 2017
Hi,
After the switch to the new TAF version yesterday, from 3.5.2 to 3.5.13,
the results of 2 tests (both using the same executable) have changed
in a significant way, although these 2 tests are not our most robust ones.
version 3.5.13:
< Y Y Y Y 16> 7<16 FAIL lab_sea (e=0, w=8)
< Y Y Y Y 16> 7< 5 FAIL lab_sea.noseaicedyn
---
compared to version 3.5.2 from yesterday:
> Y Y Y Y 16>16<16 pass lab_sea (e=0, w=8)
> Y Y Y Y 16>13< 5 pass lab_sea.noseaicedyn
Note that these tests use what we call the "Divided adjoint"
I am confident that the source code is the same (zero diff),
and noticed some differences in the adjoint code (but I admit that
I am not good at figuring out how significant it is).
I am providing, attached, the input source (gzip: ad_input_code.f.gz)
as well as the "taf command" used to generate the AD code.
I would appreciate if you could confirm that these changes are
benign or if there is something to worry about.
Thanks,
Jean-Michel
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ad_input_code.f.gz
Type: application/octet-stream
Size: 2320746 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-devel/attachments/20170302/9f086c83/attachment-0001.obj>
-------------- next part --------------
tafx -f77 -reverse -pure -i4 -r4 -intrinsic system,flush -l taf_ad.log -toplevel the_main_loop -input xx_theta_dummy,xx_salt_dummy,xx_tr1_dummy,xx_hflux_dummy,xx_sflux_dummy,xx_tauu_dummy,xx_tauv_dummy,xx_atemp_dummy,xx_aqh_dummy,xx_precip_dummy,xx_swflux_dummy,xx_swdown_dummy,xx_lwflux_dummy,xx_lwdown_dummy,xx_evap_dummy,xx_snowprecip_dummy,xx_apressure_dummy,xx_runoff_dummy,xx_uwind_dummy,xx_vwind_dummy,xx_obcsn_dummy,xx_obcss_dummy,xx_obcsw_dummy,xx_obcse_dummy,xx_sst_dummy,xx_sss_dummy,xx_depth_dummy,xx_diffkr_dummy,xx_kapgm_dummy,xx_kapredi_dummy,xx_bottomdrag_dummy,xx_efluxy_dummy,xx_efluxp_dummy,xx_edtaux_dummy,xx_edtauy_dummy,xx_uvel_dummy,xx_vvel_dummy,xx_etan_dummy,xx_gen2d_dummy,xx_gen3d_dummy,xx_genarr2d_dummy,xx_genarr3d_dummy,xx_gentim2d_dummy,xx_siarea_dummy,xx_siheff_dummy,xx_sihsnow_dummy,xx_shifwflx_dummy,xx_relaxsst_dummy,xx_relaxsss_dummy,xx_atemp_mean_dummy,xx_aqh_mean_dummy,xx_precip_mean_dummy,xx_swdown_mean_dummy,xx_uwind_mean_dummy,xx_vwind_mean_dummy -output fc ad_input_code.f
More information about the MITgcm-devel
mailing list