[Ptools-perfapi] Papi on Westmere using perf gives me only 3 events

Philip Mucci mucci at eecs.utk.edu
Thu Aug 4 13:52:38 EDT 2011


Ok, until it's fixed we can use one of the papi locks around the pfm_call...

Phil

On Aug 4, 2011, at 12:43 PM, Vince Weaver wrote:

> On Thu, 4 Aug 2011, Jason Riedy wrote:
>> 
>> The segfault occurs in the strcmp at papi_pfm4_events.c:585
>> within libpfm4 during PAPI_event_name_to_code.  Single-threaded
>> works, so I suspect a simple race and not something
>> CPU-specific.
> 
> yes.  The problem is in the new libpfm4 code.  The older code built
> up the event tables at init time, and then had read-only access to them 
> after.  The libpfm4 code dynamically allocates events, so there needs
> to be locking around the allocations.
> 
>> I'll check on access to ours if you'd like.  Having 10 cores
>> behind a single L3 sure makes memory bottlenecks apparent...
> 
> I'll never turn down accesses on various machines, especially on a 
> Westmere as we've had lots of trouble finding one to test PAPI on.
> 
> Vince
> vweaver1 at eecs.utk.edu
> 
> _______________________________________________
> Ptools-perfapi mailing list
> Ptools-perfapi at eecs.utk.edu
> http://lists.eecs.utk.edu/mailman/listinfo/ptools-perfapi



More information about the Ptools-perfapi mailing list