[petsc-dev] Mark Adam's performance kernel logging

[Barry Smith]

On Nov 30, 2012, at 4:10 PM, Matthew Knepley <knepley at gmail.com> wrote:

> On Fri, Nov 30, 2012 at 3:45 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
>> 
>>   I think we can implement Mark Adam's idea of generating performance information for individual kernels (multiply, dot, daxpy …) by using the XXView() code since that handles the recursive traversal  to the sub objects. So MatView() with the performance viewer format would do a MatMult(),   KSPView() would do a axpy,  KSPView_GMRES() would do an mdot and maxpy etc. … PCMG would call the viewer recursively on the subksps etc etc etc.
> 
> Yes, we can use Viewer for top level traversal, and then the low-level
> Log API for
> extracting the performance info online and processing it.

   Just run all the things related to a particular object (level) in its own stage then log summary will just display all this stuff after the data for the run.




> 
>   Matt
> 
>>   Barry
>> 
> 
> 
> 
> --
> What most experimenters take for granted before they begin their
> experiments is infinitely more interesting than any results to which
> their experiments lead.
> -- Norbert Wiener