[petsc-users] PAPI availability

[Matthew Knepley]

On Wed, Feb 18, 2015 at 10:01 AM, Justin Chang <jychang48 at gmail.com> wrote:

> Matt, thank you for the response.
>
> For flop monitoring, are they hardware counts or manually counted? That
> is, would the flops documented by PETSc be affected by overhead factors
> like memory bandwidth, latency, etc, thus potentially giving you "poor
> efficiency" with respect to the theoretical performance?
>

Its always best to look at the code. Here is the event creation


https://bitbucket.org/petsc/petsc/src/17ccb9bd50733b8f00ffcc6fc4ee6e9be3a3168d/src/sys/logging/plog.c?at=master#cl-211

This measures 'flops', not 'flops/s'.


> Also, are these flops the flops noted in -log_summary?
>

Here is the log_summary code:


https://bitbucket.org/petsc/petsc/src/17ccb9bd50733b8f00ffcc6fc4ee6e9be3a3168d/src/sys/logging/utils/eventlog.c?at=master#cl-661

so we are using the PAPI count instead of our manual count in log_summary.

I am not sure what you mean about overheads. Memory bandwidth, cache usage,
machine latency, etc. are not overheads, they are how
the machine functions. They are reasons you might not obtain the floating
point peak.

  Thanks,

     Matt


> Thanks,
> Justin
>
> On Wed, Feb 18, 2015 at 5:42 AM, Matthew Knepley <knepley at gmail.com>
> wrote:
>
>> On Tue, Feb 17, 2015 at 8:12 PM, Justin Chang <jychang48 at gmail.com>
>> wrote:
>>
>>> Hi all,
>>>
>>> I want to document some profiling metrics like cache hits/misses using
>>> PAPI. Does PETSc automatically come with PAPI, or is it something I have to
>>> declare when configuring PETSc?
>>>
>>
>> You can turn on PAPI by configuring using --with-papi, just like other
>> packages (might need the --with-papi-dir too). Right now, we only monitor
>> flops.
>> Feel free to add other events.
>>
>>   Thanks,
>>
>>    Matt
>>
>>
>>> Thanks,
>>> Justin
>>>
>>
>>
>>
>> --
>> 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
>>
>
>


-- 
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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150218/66cd840e/attachment-0001.html>