major refactor stream of stream usage #376
Conversation
…mode specific streams
…d_stream_usage' into feature/refactor_stream_usage
…t field pointers to nans
…ams' into feature/refactor_stream_usage
…efactor_stream_usage
…al_streams' into feature/refactor_stream_usage
|
Thanks @mvertens ! dice cplhist test is ok in UFS with refactor UFS RTs pass, except for datm_cdeps_lnd_era5_intel and its restart. These fail in run with Especially since this was PR from @uturuncoglu in ufs-community/ufs-weather-model#1845 , tagging here. Currently in this test, I think tdew needs to be calculated in datm/datm_datamode_era5_mod.F90 |
For UFS, I think docn cplhist and sstdata can be the same, if can add units_CToK |
|
@NickSzapiro-NOAA @mvertens I wonder if there is a way to get rid of lsize completely such as querying mesh dimensions etc. It would be better solution and stream will not depend on any particular variable. |
|
@NickSzapiro-NOAA- thanks for your testing! @NickSzapiro-NOAA @uturuncoglu - I think I am missing something in datm_datamode_era5_mod.F90. In the the original code for this: if (first_time)
! determine t2max (see below for use)
if (associated(Sa_t2m)) then
rtmp(1) = maxval(Sa_t2m(:))
call ESMF_VMAllReduce(vm, rtmp, rtmp(2:), 1, ESMF_REDUCE_MAX, rc=rc)
t2max = rtmp(2)
if (mainproc) write(logunit,*) subname,' t2max = ',t2max
end if
! determine tdewmax (see below for use)
rtmp(1) = maxval(strm_tdew(:))
call ESMF_VMAllReduce(vm, rtmp, rtmp(2:), 1, ESMF_REDUCE_MAX, rc=rc)
td2max = rtmp(2)
if (mainproc) write(logunit,*) subname,' td2max = ',td2max
! reset first_time
first_time = .false.
endif
In the new code:
What does the ESMF configuration file look like for this datamode? |
|
@NickSzapiro-NOAA It would be also nice to see actual error. |
|
@NickSzapiro-NOAA We have just have a short chat with @mvertens about the issue. It seems that UFS WM is using CDEPS develop branch (https://github.com/NOAA-EMC/CDEPS/blob/develop/datm/datm_datamode_era5_mod.F90) and the ERA5 mode in that branch is different from the one found in ESCOMP. So, that was the issue. @mvertens will check the new code found in develop branch and try to fix the issue in ERA5 mode. Then once this PR merged, UFS WM needs to sync with ESCOMP to bring these changes. |
Great, thank you for clarifying that. |
@mvertens - Great, thanks! If the purpose of this is software testing, would it be reasonable for CESM to use the NorESM cplhist files for testing, so that we're running consistent tests and don't need to do double-maintenance on this test? |
|
@mvertens @billsacks is this PR about ready for prealpha testing? |
drof/drof_datamode_copyall.F90
Outdated
| if (associated(strm_Forr_rofi)) then | ||
| do ni = 1, size(Forr_rofi) | ||
| if (abs(strm_Forr_rofi(ni)) < 1.e28_r8) then | ||
| Forr_rofi(:) = strm_Forr_rofi(:) |
There was a problem hiding this comment.
@mvertens - you commented that this is fixed (changing the whole-array assignment of Forr_rofi to just Forr_rofi(ni)), but it looks like you haven't yet pushed a commit fixing this? (You fixed other, similar issues.)
|
@billsacks - I have pushed the drof change back. |
It looks like you changed one of the |
Oh, never mind - I see your second commit. |
|
@fischer-ncar - thanks for checking in. I talked to @mvertens and we both feel this is ready for your testing. See #376 (comment) for some notes on testing - specifically a few tests to add beyond the standard prealphas on derecho. I'm in the midst of a deep dive into the different data model modes to make sure we're running testing covering everything we want to test, so I may come up with a few additional tests in the next day or so, but the prealphas plus the tests noted above should be close to what we want. Thank you! |
|
@fischer-ncar and @mvertens - I have done a pretty detailed analysis of the different data modes and their coverage in prealpha testing. In addition to the CESM prealpha testing, I want to make sure the following are covered: @fischer-ncar - can you please add the following two tests to your testing, along with baseline comparisons (these tests are present in the aux_mom or aux_cice test lists but will need baselines for your testing) (I have also updated my testing comment with this request so it's all in one place):
@mvertens - I don't think we have a way to test the following in CESM, so I wanted to double-check that you have tested these (sorry, you might have answered this before):
@mvertens - also, I think you said you have tested DWAV... just wanted to double-check that, since we don't have any DWAV testing in our prealpha test list (there's a test of ADWAV in aux_cmeps). (EDIT: I see that you ran aux_cdeps_noresm, which covers this.) |
|
Just flagging this, as we'll want to test with CTSM when we update our submodules, @ekluzek. |
|
|
@fischer-ncar - can you please add one more test, with baseline comparisons: |
|
Here's the prealpha test results I have so far. I'm still running the extra tests that @billsacks wanted me to run. The above tests all failed with no comp_calss for rof. Then there was a couple of small answer changes. |
|
Here's the three other tests that @billsacks wanted me to do. |
|
I had a look at test SMS_Ld40.TL319_t232.C_JRA.derecho_intel
which has a compset
longname: 2000_DATM%JRA-1p5-2023_SLND_DICE%SSMI_MOM6_DROF%JRA-1p5-2023_SGLC_SWAV_SESP
the DROF part of that is
DROF%JRA-1p5-2023
and I don't see a description for JRA-1p5-2023 in the config_component.xml
file. The error message is a bit misleading I think.
I confirmed that adding a description for JRA-1p5-2023 solves the problem.
…On Thu, Jan 22, 2026 at 4:57 PM fischer-ncar ***@***.***> wrote:
*fischer-ncar* left a comment (ESCOMP/CDEPS#376)
<#376 (comment)>
Here's the three other tests that @billsacks
<https://github.com/billsacks> wanted me to do.
PASS SMS_D_Ld2.T62_t232.G.derecho_gnu RUN time=351
FAIL SMS_D_Ld5.f10_f10_mg37.ISSP245Clm60BgcCropCrujra.derecho_intel.clm-datm_rcp45_anom_forc BASELINE cesm3_0_alpha08b: DIFF
lndExp_Sa_co2prog (lndExp_nx,lndExp_ny,time) t_index = 1 1
456 456 ( 1, 1, 1) ( 1, 1, 1) ( 1, 1, 1) ( 1, 1, 1)
456 3.999488830566406E+02 3.999488830566406E+02 4.0E+02 3.999488830566406E+02 1.0E+00 3.999488830566406E+02
456 0.000000000000000E+00 0.000000000000000E+00 0.000000000000000E+00 0.000000000000000E+00
456 ( 1, 1, 1) ( 1, 1, 1)
avg abs field values: 3.999488830566406E+02 rms diff: 4.0E+02 avg rel diff(npos): 1.0E+00
0.000000000000000E+00 avg decimal digits(ndif): 0.0 worst: 0.0
RMS lndExp_Sa_co2prog 3.9995E+02 NORMALIZED 2.0000E+00
FAIL ERS_Ld5.TL319_t232.G_IAF.derecho_intel.cice-default CREATE_NEWCASE
ERROR: grid alias TL319_t232 not valid for compset 2000_DATM%IAF_SLND_CICE_MOM6_DROF%IAF_SGLC_SWAV_SESP
—
Reply to this email directly, view it on GitHub
<#376 (comment)>, or
unsubscribe
<https://github.com/notifications/unsubscribe-auth/ABOXUGAXZRTJUACCEWBZUUL4IFIVZAVCNFSM6AAAAACRMTMMLKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTOOBXGE2TIOBTGE>
.
You are receiving this because your review was requested.Message ID:
***@***.***>
--
Jim Edwards
STAND UP FOR SCIENCE
CESM Software Engineer
National Center for Atmospheric Research
Boulder, CO
|
|
@fischer-ncar - thanks a lot for this testing report! And @jedwards4b - thanks for your input. I have just pushed a commit that restores the JRA-1p5-2023 mode for DROF that I think should fix the issues with G_JRA and C_JRA compsets. (@mvertens - I assume your removal of that mode for DROF was accidental? As far as I can tell the mode should still work but it looks like it was accidentally removed from DROF's config_component... but please let us know if there was a reason why you removed it.) @fischer-ncar - can you please retry the failed tests with this latest version? The changes in lndExp_Sa_o3 and lndExp_Sa_co2prog are expected based on changes from #373 (but are expected not to feedback to cause any other differences in typical configurations). For the failing G_IAF case, can you instead try @fischer-ncar - I have poked around in Have you run the extra tests noted in #376 (comment) - the ones under (1), (2), and the additional DTEST and E1850TEST test, in addition to the more recent additions that I copied at the bottom of that comment? Thanks again for all of your testing of this! |
|
@billsacks - the removal of JRA-1p5-2023 was a mistake that I made in trying to sync the changes between the PR to NorESMHub and ESCOMP. @jedwards4b - thanks for suggesting the fix. @fischer-ncar - thanks for all of your testing. |
|
@billsacks I missed those tests, I'll run them now. |
|
All of the tests with the missing JRA-1p5-2023 have passed. So all of the prealpha tests are now good. For the extra tests @billsacks wanted me to do, the following passed. Of the tests that passed, two of the tests failed to generate complete baselines, which is not an issue with this PR. This test passed, with expected answer difference caused by #373. Then this test failed with a missing grid alias. I believe I've run all of the extra tests. |
7 participants
Description of changes
Major refactor of stream usage
Specific notes
All data mode files across the data components(except for
datm_datamode_gefs_mod.F90anddice_datamode_cplhist_mod.F90) now have explicit setting of streams rather than using the implicit copy used bydfields. Copies that used to be done implicitly using the naming convention assumed by dfields arenow done explictly. This makes it clear what export fields are direct copies and what export fields are
derived fields that have no corresponding stream field.
In each datamode module - there are now two sections in the module variable list:
export (and sometimes import) state pointersandstream pointer.The following changes have been made throughout the code:
character(*) => character(len=*),trim(subname) => subname(the latter is done since parameters, e.g.subname, don't need trim functions)The following new routines have been introduced or deleted:
drof_datamode_cplhist_mod.F90) to support this addition.docn_datamode_copyall_mod.F90anddocn_datamode_iaf_mod.F90into a new filedocn_datamode_sstdata_mod.F90. The two filesdocn_datamode_copyall_mod.F90anddocn_datamode_iaf_mod.F90were effectively the same and two different datamodes are not needed since in both cases prescribed SST data was read in. (In addition, docn_datamode_iaf_mod.F90` set pointers to importState data which was never used and not needed).NOTE: This PR also provides a new mapfile algorithm as a new option for mapping the streams input to the component model resolution. The new scheme does not have a default but is enabled via adding the following in the appropriate user_nl_XXX_streams where the user needs to fill in the entries in <> below:
<stream_name>:mapalgo = "mapfile:
As an example:
rof.ryf8485_jra:mapalgo="mapfile:/cluster/shared/noresm/inputdata/cpl/cpl6/map_JRA025_to_tnx1v4_e1000r300_170928.nc". This new capability enables a significant speedup in stand-alone ocean simulations. See NorESMhub/BLOM#686 (comment).NOTE*:
Contributors other than yourself, if any: None
CDEPS Issues Fixed: #377
Are there dependencies on other component PRs (if so list):
Are changes expected to change answers (bfb, different to roundoff, more substantial):
Any User Interface Changes (namelist or namelist defaults changes):
Testing performed (will describe the noresm testing here)
Hashes used for testing: noresm3_0_beta09 plus this CDEPS branch