* perf_event validation tests
@ 2011-05-27 21:53 Vince Weaver
2011-05-27 22:45 ` Arun Sharma
0 siblings, 1 reply; 3+ messages in thread
From: Vince Weaver @ 2011-05-27 21:53 UTC (permalink / raw)
To: linux-perf-users
Hello
a while back it was mentioned it would be nice to have some perf_event
validation tests.
I've posted some here:
http://web.eecs.utk.edu/~vweaver1/projects/perf-events/validation.html
It's a mix of tests that test the perf_events interface. They don't test
the perf tool itself, but the kernel ABI. Some people might still find
them useful. And if you have any tests to suggest or contribute, feel
free to drop me an e-mail.
Vince
vweaver1@eecs.utk.edu
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: perf_event validation tests
2011-05-27 21:53 perf_event validation tests Vince Weaver
@ 2011-05-27 22:45 ` Arun Sharma
2011-05-28 2:03 ` Vince Weaver
0 siblings, 1 reply; 3+ messages in thread
From: Arun Sharma @ 2011-05-27 22:45 UTC (permalink / raw)
To: Vince Weaver; +Cc: linux-perf-users
On Fri, May 27, 2011 at 05:53:41PM -0400, Vince Weaver wrote:
> Hello
>
> a while back it was mentioned it would be nice to have some perf_event
> validation tests.
>
> I've posted some here:
> http://web.eecs.utk.edu/~vweaver1/projects/perf-events/validation.html
>
cache-references.c:
pe.config=PERF_COUNT_HW_CACHE_REFERENCES;
...
for(i=0; i<ARRAYSIZE; i++) {
array[i]=(double)i;
}
...
if (!quiet) printf("\tL1 D accesseses: %lld %lf\n",count,aSumm);
PERF_COUNT_HW_CACHE_REFERENCES is counting LLC references on many
platforms. Did you mean to use PERF_COUNT_HW_CACHE_L1D here?
-Arun
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: perf_event validation tests
2011-05-27 22:45 ` Arun Sharma
@ 2011-05-28 2:03 ` Vince Weaver
0 siblings, 0 replies; 3+ messages in thread
From: Vince Weaver @ 2011-05-28 2:03 UTC (permalink / raw)
To: Arun Sharma; +Cc: linux-perf-users
On Fri, 27 May 2011, Arun Sharma wrote:
> On Fri, May 27, 2011 at 05:53:41PM -0400, Vince Weaver wrote:
> cache-references.c:
>
> pe.config=PERF_COUNT_HW_CACHE_REFERENCES;
> if (!quiet) printf("\tL1 D accesseses: %lld %lf\n",count,aSumm);
>
> PERF_COUNT_HW_CACHE_REFERENCES is counting LLC references on many
> platforms. Did you mean to use PERF_COUNT_HW_CACHE_L1D here?
probably. That was a last minute addition. I have a much more complete
set of tests writted for PAPI and I'm trying to port over the ones that
make sense to be perf-event direct. Properly programming perf_event
generalized cache events isn't documented very well though.
I've been distracted by the fact that my overflow tests start failing as
of kernel 2.6.36 :( I hadn't noticed because I was primarily developing
the tests on 2.6.32 and 2.6.35 machines.
Vince
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2011-05-28 2:03 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2011-05-27 21:53 perf_event validation tests Vince Weaver
2011-05-27 22:45 ` Arun Sharma
2011-05-28 2:03 ` Vince Weaver
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).