Hi Brian,
Nice to hear from you. I'm in Nepal at the moment on volunteer work,
but I had a chance to reply. My work at BSC was basically preliminary
investigative work...after looking at the code for oprofile and
reviewing the Cell specs, it was decided that two weeks would not be
sufficient for a Cell perfmon port, so I punted and decided to do a
little work on the PPC32 port in prep for this. I did a bit, but it
wasn't tested and not complete. Thankfully, some IBM'ers on this list
have been picking things up and are hard at work on a Perfmon port.
Once perfmon is complete, then libpfm must be done. Once that is
done, PAPI/perfmon will basically work out of the box. I suspect by
the end of the summer we should have something running.
PAPI/perfmon is in a bit of flux. I've got a hardened version here
for SiCortex that has been extensively tested and I will be working
with Dan to integrate it into the mainline PAPI releases. So that's
the news... I would subscribe to the perfmon2 list to keep up to date.
Regards,
Phil
On Apr 19, 2007, at 3:29 PM, Brian Wylie wrote:
Phil,
I spent March at BSC under the HPC Europa program, to work with
MareNostrum
and Paraver, and I learnt that you had also visited some months
earlier to
port PAPI to the Cell processor, though it sounded like that was
incomplete.
We have some Cell processors in Juelich, and we'd be interested in
having
PAPI provide access to their counters: is there a version of PAPI
for Cell
available that we could test and/or do you know when one can be
expected?
(By the way, I was able to use PAPI on MareNostrum without problem,
both with
Paraver and our Scalasca measurement system, which was helpful in
convincing
folks to patch/install PAPI on our own much smaller JS21-based blade
cluster :-)
Regards,
Brian.
--
John von Neumann Institute for Computing (NIC)
Forschungszentrum Jülich GmbH, D-52425 Jülich, Germany
_______________________________________________
perfmon mailing list
[email protected]
http://www.hpl.hp.com/hosted/linux/mail-archives/perfmon/
