Cannot write cpl aux hist files due to bug in format statement #2504
Comments
|
@bishtgautam does the land group use auxilary cpl hist files for land spinups? |
|
Yes, ELM is using cpl history files for offline BGC spinup. |
|
@bishtgautam you'll need billsacks@4cc4f2b in your release branch. |
billsacks
added a commit
that referenced
this issue
May 4, 2018
Fix generation and use of forcing data for T compsets Both the generation and use of forcing data for T compsets (glc-only, typically forced by cplhist output) have been somewhat broken for a long time. This PR fixes both the generation and use of this cplhist dlnd forcing data. I have confirmed that this is working properly by running companion I and T compset cases, where I generate data from the I compset and then use it in the T compset (making sure that the processor count is the same for the T compset as for the I compset): - Generating data from a 3-year run of I1850Clm50Sp at resolution f10_f10_musgs with histaux_l2x1yrg = .true. and histaux_double_precision = .true., then running 3 years of T1850G with `./xmlchange CISM_EVOLVE=FALSE`, I get bit-for-bit results. - When I instead used I1850Clm50SpG (which has ice evolution turned on), and ran the T1850G compset with `glc_renormalize_smb = 'on'`, I got bit-for-bit results for the first 2 years, but then roundoff-level differences in qice forcing in many grid cells in year 3 (and thus small differences in many CISM fields in year 3). Since the differences were only roundoff-level, I did not pursue this further. Test suite: - scripts_regression_tests on hobart (ran from cdd1686) - aux_cime_baselines from cesm2_0_alpha10e (compared this cime branch with the version of master on which it was based: 974839b) - Tests pass except for two that failed in the baseline: ERP_D_Ld10.f10_f10_musgs.IHistClm50BgcCrop.cheyenne_intel.clm-ciso_decStart and SMS_N3_PM3_Ld2.f19_g17.BHISTWs.cheyenne_gnu.allactive-defaultio - NLCOMP failures as expected - BASELINE comparisons bit-for-bit except for SMS_Ly2.f09_g16_gl20.T1850G.cheyenne_intel, as expected - SMS_Ly3.f09_g17_gl4.T1850G.hobart_intel from ESCOMP/CISM-wrapper@bc15238 - SMS_D_Ld3.f09_f09_mg17.A1850DLND.hobart_intel - Note: This out-of-the-box A1850DLND case passes. However, if I set GLC_NEC to 0 (as suggested by @mvertens), this fails (despite passing on master). I think this is because runs with GLC_NEC=0 now have no fields listed in the streams file. This seems okay to me: running A1850DLND with a manual setting of GLC_NEC=0 seems like a non-sensical thing to do in practice. (This was more important in the past when GLC_NEC=0 was the default; now GLC_NEC=10 is the default.) Test baseline: N/A Test namelist changes: Namelists will differ for cases with dlnd Test status: climate changing for cases with dlnd (T compsets), should otherwise be bit-for-bit Fixes #2446 Fixes #2504 User interface changes?: - Renamed drv namelist variable histaux_l2x1yr to histaux_l2x1yrg to denote that it's just for glc fields (following the naming convention used for histaux_a2x3hrp) - Added drv namelist variable histaux_double_precision: if true, use double-precision rather than single-precision for coupler auxiliary history files Update gh-pages html (Y/N)?: Y Code review: Mariana Vertenstein
rljacob
pushed a commit
to E3SM-Project/E3SM
that referenced
this issue
Apr 12, 2021
rljacob
pushed a commit
to E3SM-Project/E3SM
that referenced
this issue
Apr 12, 2021
Fix generation and use of forcing data for T compsets Both the generation and use of forcing data for T compsets (glc-only, typically forced by cplhist output) have been somewhat broken for a long time. This PR fixes both the generation and use of this cplhist dlnd forcing data. I have confirmed that this is working properly by running companion I and T compset cases, where I generate data from the I compset and then use it in the T compset (making sure that the processor count is the same for the T compset as for the I compset): - Generating data from a 3-year run of I1850Clm50Sp at resolution f10_f10_musgs with histaux_l2x1yrg = .true. and histaux_double_precision = .true., then running 3 years of T1850G with `./xmlchange CISM_EVOLVE=FALSE`, I get bit-for-bit results. - When I instead used I1850Clm50SpG (which has ice evolution turned on), and ran the T1850G compset with `glc_renormalize_smb = 'on'`, I got bit-for-bit results for the first 2 years, but then roundoff-level differences in qice forcing in many grid cells in year 3 (and thus small differences in many CISM fields in year 3). Since the differences were only roundoff-level, I did not pursue this further. Test suite: - scripts_regression_tests on hobart (ran from cdd1686) - aux_cime_baselines from cesm2_0_alpha10e (compared this cime branch with the version of master on which it was based: 974839b) - Tests pass except for two that failed in the baseline: ERP_D_Ld10.f10_f10_musgs.IHistClm50BgcCrop.cheyenne_intel.clm-ciso_decStart and SMS_N3_PM3_Ld2.f19_g17.BHISTWs.cheyenne_gnu.allactive-defaultio - NLCOMP failures as expected - BASELINE comparisons bit-for-bit except for SMS_Ly2.f09_g16_gl20.T1850G.cheyenne_intel, as expected - SMS_Ly3.f09_g17_gl4.T1850G.hobart_intel from ESCOMP/CISM-wrapper@bc15238 - SMS_D_Ld3.f09_f09_mg17.A1850DLND.hobart_intel - Note: This out-of-the-box A1850DLND case passes. However, if I set GLC_NEC to 0 (as suggested by @mvertens), this fails (despite passing on master). I think this is because runs with GLC_NEC=0 now have no fields listed in the streams file. This seems okay to me: running A1850DLND with a manual setting of GLC_NEC=0 seems like a non-sensical thing to do in practice. (This was more important in the past when GLC_NEC=0 was the default; now GLC_NEC=10 is the default.) Test baseline: N/A Test namelist changes: Namelists will differ for cases with dlnd Test status: climate changing for cases with dlnd (T compsets), should otherwise be bit-for-bit Fixes ESMCI/cime#2446 Fixes ESMCI/cime#2504 User interface changes?: - Renamed drv namelist variable histaux_l2x1yr to histaux_l2x1yrg to denote that it's just for glc fields (following the naming convention used for histaux_a2x3hrp) - Added drv namelist variable histaux_double_precision: if true, use double-precision rather than single-precision for coupler auxiliary history files Update gh-pages html (Y/N)?: Y Code review: Mariana Vertenstein
rljacob
pushed a commit
to E3SM-Project/E3SM
that referenced
this issue
Apr 21, 2021
rljacob
pushed a commit
to E3SM-Project/E3SM
that referenced
this issue
Apr 21, 2021
Fix generation and use of forcing data for T compsets
Both the generation and use of forcing data for T compsets (glc-only,
typically forced by cplhist output) have been somewhat broken for a long
time. This PR fixes both the generation and use of this cplhist dlnd
forcing data.
I have confirmed that this is working properly by running companion I
and T compset cases, where I generate data from the I compset and then
use it in the T compset (making sure that the processor count is the
same for the T compset as for the I compset):
- Generating data from a 3-year run of I1850Clm50Sp at resolution
f10_f10_musgs with histaux_l2x1yrg = .true. and
histaux_double_precision = .true., then running 3 years of T1850G with
`./xmlchange CISM_EVOLVE=FALSE`, I get bit-for-bit results.
- When I instead used I1850Clm50SpG (which has ice evolution turned on),
and ran the T1850G compset with `glc_renormalize_smb = 'on'`, I got
bit-for-bit results for the first 2 years, but then roundoff-level
differences in qice forcing in many grid cells in year 3 (and thus
small differences in many CISM fields in year 3). Since the
differences were only roundoff-level, I did not pursue this further.
Test suite:
- scripts_regression_tests on hobart (ran from 024637618)
- aux_cime_baselines from cesm2_0_alpha10e (compared this cime branch
with the version of master on which it was based: 4de3835)
- Tests pass except for two that failed in the baseline:
ERP_D_Ld10.f10_f10_musgs.IHistClm50BgcCrop.cheyenne_intel.clm-ciso_decStart
and SMS_N3_PM3_Ld2.f19_g17.BHISTWs.cheyenne_gnu.allactive-defaultio
- NLCOMP failures as expected
- BASELINE comparisons bit-for-bit except for
SMS_Ly2.f09_g16_gl20.T1850G.cheyenne_intel, as expected
- SMS_Ly3.f09_g17_gl4.T1850G.hobart_intel from ESCOMP/CISM-wrapper@bc15238
- SMS_D_Ld3.f09_f09_mg17.A1850DLND.hobart_intel
- Note: This out-of-the-box A1850DLND case passes. However, if I set
GLC_NEC to 0 (as suggested by @mvertens), this fails (despite
passing on master). I think this is because runs with GLC_NEC=0 now
have no fields listed in the streams file. This seems okay to me:
running A1850DLND with a manual setting of GLC_NEC=0 seems like a
non-sensical thing to do in practice. (This was more important in
the past when GLC_NEC=0 was the default; now GLC_NEC=10 is the
default.)
Test baseline: N/A
Test namelist changes: Namelists will differ for cases with dlnd
Test status: climate changing for cases with dlnd (T compsets), should
otherwise be bit-for-bit
Fixes ESMCI/cime#2446
Fixes ESMCI/cime#2504
User interface changes?:
- Renamed drv namelist variable histaux_l2x1yr to histaux_l2x1yrg to
denote that it's just for glc fields (following the naming convention
used for histaux_a2x3hrp)
- Added drv namelist variable histaux_double_precision: if true, use
double-precision rather than single-precision for coupler auxiliary
history files
Update gh-pages html (Y/N)?: Y
Code review: Mariana Vertenstein
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
When trying to write a coupler auxiliary history file, I got this death:
The problem is that, in the following,
a6should be6a:cime/src/drivers/mct/main/seq_hist_mod.F90
Lines 1150 to 1151 in 2e0ba3b
I'll fix this in the course of fixing #2446 .
cc @jedwards4b
The text was updated successfully, but these errors were encountered: