[petsc-users] flops performed on each processor

[Barry Smith]

  You will need to upgrade to 3.5  We don’t provide any convenient way to dump the information per process for 3.4


On Aug 22, 2014, at 1:58 PM, Xiangdong <epscodes at gmail.com> wrote:

> simply replacing -log_summary with -log_view ::ascii_info_detail does not produce anything. (Sams as -log_view ascii:filename:ascii_info_detail.) Where does it dump the log? Is this the feature of new petsc? I am using petsc-3.4.3.
> 
> Thanks.
> 
> Xiangdong 
> 
> 
> On Fri, Aug 22, 2014 at 2:53 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
> On Aug 22, 2014, at 1:28 PM, Matthew Knepley <knepley at gmail.com> wrote:
> 
> > On Fri, Aug 22, 2014 at 1:24 PM, Xiangdong <epscodes at gmail.com> wrote:
> > Hello everyone,
> >
> > In the PETSc log_summary, it outputs the Max flops, the ratio of Max/Min flops. If it is far from 1.0, I would like to know the distributions of flops among all the procs. How can I obtain the flops performed on each proc?
> >
> > -log_view ::ascii_info_detail
> 
>   We don’t have it properly documented yet but this outputs python dictionaries that contain the various performance counters for each process and stages and events. You can then write simple Python code to process the data any way you like, for example plotting distributions etc. I recommend starting by collecting data for just 2 or 3 processes and using that data to write your python scripts. Once you have the scripts you want you can make parallel runs with more processes and use the already debugged scripts.
> >
> >   Thanks,
> >
> >      Matt
> >
> > Thank you.
> >
> > Best,
> > Xiangdong
> >
> >
> >
> > --
> > 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
> 
>