This repository has been archived by the owner on Oct 23, 2020. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 0
Agenda Notes Monday August 28th 2017 11:00 AM
Michael Duda edited this page Aug 28, 2017
·
9 revisions
Date: 28 August 2017
Time: 11:00 MDT / 19:00 CET
Call-in number: 1-866-740-1260
Access Code: 4978161
- Update on setting
_FillValueattribute in output files- See Issue MPAS NetCDF output does not include the _FillValue attribute
- Apparently, the netCDF library is clever enough to look for
_FillValueas the name of an attribute and to use it for certain additional purposes; see the nc_put_att documentation - Unless it's not documented, there is no separate call in the PIO API for specifying fill values; see the PIO API documentation
- Solution: this may just be a matter of specifying the
missing_valueattribute for fields in theRegistry.xmlfile and uncommenting the following lines of code:
- Framework modifications:
Participants: Mark, Matt, Dom, Xylar, Michael
- Several framework PRs, all related to streams, need to be handled in a specific order: 1392, 1284, 1395, and then 1396
- 1389 resolves issues on Mira and appears to do no harm for Intel and GNU compilers; there are just a few questions about the details of the changes that would be good to answer before final approval
Conclusion: we should have missing_value and _FillValue added to all variables by default
- Dom: CMOR software (used by CMIP5 and CORDEX) suggests both
missing_valueand_FillValueattributes for coordinate variables - How are missing values handled in MPAS-Ocean?
- Cells below the bathymetry are just assigned a value
- Perhaps this behavior has changed recently? now values are something like -1e34
- Is the behavior specific to analysis members?
- MPAS_field_accessor could help cores to fully integrate the use of
missing_valuethroughout code -
Make a PR to add missing_value and _FillValue to all variables by default
- Also investigate how fill values are getting set in "empty" variables
Mark: A developer working on KNL was wondering whether we have considered making "negative halos"?
- a negative halo would identify elements on the interior of a block that need to be communicated
- the idea would be to first compute over the negative halo, then begin an async halo exchange, and continue computing over the interior elements
- The framework/cell_ordering branch included work in progress to try to enable general element ordering within blocks, and this could be used as a basis for implementing "negative halos"