Hi Vince,

Yeah, I noticed that when I was stracing a working perf but nothing else worked.

Also, I spent some time actually getting together a libpfm that works with 
Cray's distro perf_event.h in /usr/include/linux/perf_event.h. That required a 
number of changes and eventually worked. However, I decided to go back to the 
original and start again, and attached is the patch that I eventually got 
working.

This is a Cray XC30 system, with paranoid == 2. 

Attachment: libpfm.diff
Description: Binary data


> On Jan 5, 2016, at 2:00 PM, Vince Weaver <vincent.wea...@maine.edu> wrote:
> 
> On Mon, 4 Jan 2016, Philip Mucci wrote:
> 
>> Right. The default however is currently to exclude_guest = 1, which is 
>> the problem. PAPI did not seem to rectify that setting…
> 
> just a further comment, I complained to the kernel people about this mess 
> again but there doesn't seem to be much that can be done about it.
> 
> The perf tool when it gets an EINVAL actually tries rolling back flags one 
> by one to handle this issue (see perf_missing_features in 
> perf/util/evesel.c).  I'm not sure if it's worth moving something 
> similar into PAPI or libpfm4.
> 
> Vince

------------------------------------------------------------------------------
_______________________________________________
perfmon2-devel mailing list
perfmon2-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/perfmon2-devel

Reply via email to