PGI answers for OM4 do not match across restart #257
Comments
|
When I set #override DEBUG = True the model came down with: FATAL from PE 20: NaN detected: after ePBL Kd_ePBL stdout: |
|
The restart issue also exists on c2 with pgi14 |
|
The problem exists in the much smaller (32 pes) test case MOM6_SIS2_bergs_cgrid Also, I think the issue exists only in prod mode (-O3) and goes away in repro mode (-O2). |
- Kd_ePBL has intent(out) from the PBL routine but was being
initialized by the calling routine and conditionally on whether
a diagnostic is active:
- Kd_ePBL is a mandatory output, no-longer conditional on the
diagnostic;
- initializing values by the caller is not guaranteed to work when the
callee has intent(out);
- ePBL was only setting values for non-land points.
- The conditional initialization has been removed entirely.
- ePBL now sets values on all computational points.
- Addresses issue discussed in #257 where DEBUG=True was detecting
NaN's in Kd_ePBL.
|
DEBUG=True issue has been fixed in b4f8cdb -A. Dr Alistair Adcroft ([email protected]) On Wed, Mar 16, 2016 at 4:38 PM, Niki Zadeh [email protected]
|
* Move mct_cap/ to STALE_mct_cap/. mct cap is no longer supported and will soon be removed for good. * remove mct from CI testing * Remove mct test from github workflows
MOM6 tag 2016.03.02 (the issue is probably much older)
When compiled with pgi15 on c3 the experiment OM4_SIS2_low_mixing3 does not reproduce when a restart is involved
(i.e., 2 days restart files do not match 1+1 days restart files) (irrespective of mask_table being used or not).
There is no such issue for the intel15 runs on c3.
I have not seen a non-repeatability issue (answer changes between dual runs) with pgi15 for this experiment (3 runs so far) . So this is most probably a restart issue (75%).
The text was updated successfully, but these errors were encountered: