Difference between revisions of "Minutes 12th Feb 2010"
| (One intermediate revision by the same user not shown) | |||
| Line 1: | Line 1: | ||
We looked at the html and diagrams generated by doxygen. It looks very useful already. The 'program's do not appear to get picked up by doxygen. The call diagrams work well though there may be problems with the caller diagrams (i.e. you can see which subroutines a given subroutine calls, but you can't always see which subroutines a subroutine is called by). This html is now on the web. | We looked at the html and diagrams generated by doxygen. It looks very useful already. The 'program's do not appear to get picked up by doxygen. The call diagrams work well though there may be problems with the caller diagrams (i.e. you can see which subroutines a given subroutine calls, but you can't always see which subroutines a subroutine is called by). This html is now on the web. | ||
| − | Profiling was discussed. Sarah showed some more profiling results. Seems that vast majority of | + | Profiling was discussed. Sarah showed some more profiling results. Seems that vast majority of time in HO simulation is spent in the HO code, so parallelising just the relevant sections (that is, the linear system solver, and the viscosity calculations) should give big efficiency improvements. Steph outlined approach to parallelising velocity solver. We'd need to parallelise construction of the sparse matrix. |
| + | '''At the CCSM meeting in Boulder, we found that the US DOE ISICLES project has already begun this, using Trilinos''' | ||
Actions: | Actions: | ||
Latest revision as of 18:26, 22 February 2010
We looked at the html and diagrams generated by doxygen. It looks very useful already. The 'program's do not appear to get picked up by doxygen. The call diagrams work well though there may be problems with the caller diagrams (i.e. you can see which subroutines a given subroutine calls, but you can't always see which subroutines a subroutine is called by). This html is now on the web.
Profiling was discussed. Sarah showed some more profiling results. Seems that vast majority of time in HO simulation is spent in the HO code, so parallelising just the relevant sections (that is, the linear system solver, and the viscosity calculations) should give big efficiency improvements. Steph outlined approach to parallelising velocity solver. We'd need to parallelise construction of the sparse matrix.
At the CCSM meeting in Boulder, we found that the US DOE ISICLES project has already begun this, using Trilinos
Actions:
- Sarah to put a link to the doxygen html on this wiki.
- Gethin (and possibly Rupert) to contribute to glimmer-cism scaling discussion clarifying Jed's point about CF compliant NetCDF files being designed to handling scaling at the file level, and, perhaps more importantly, to advise against saving the same data in different ways.
- Gethin to try debugging with gdb in emacs on Fedora or similar OS to see whether that allows derived types to be interrogated.
Did I miss anything?