On Mon, 29 Oct 2018, Michael Petlan wrote:
> Is there any way to detect what papi knows/uses apart from papi_avail?
well if you want the full details on an event you can do something like
./papi_avail -e PAPI_FP_OPS
I often end up poking around the papi_events.csv file manually.
> > Also if you look at the ARM ARM for a generic VFP_SPEC events it says
> > "The counter counts the last micro-operation of each data engine
> > floating-point instruction." So if somehow the fmadd instruction is
> > broken up into 3 vfp micro-ops internally then you might get the 3-count.
> >
> Shouldn't it be FP_OPS instead of FP_INS then? We're counting ops not insns
> in fact.
It could be. The X-gene support was contributed by Will Cohen, I don't
know if any of the PAPI devels actually have access to an X-gene machine
to test on.
> OK, so the conclusion is that the test is not reliable everywhere,
> since it does not adjust to HW, just makes basic assumptions based
> on whether it runs on ppc or not.
yes, and we'd be perfectly glad to extend the test to make it more
comprehensive and handle things better.
The problem with FP events is first determining what the result "should
be" and then trying to come up with some way to encode that into the test
without just having a lot of #ifdefs everywhere. This isn't always easy.
I do think ARM could use some work, I think the FP tests fail on Raspberry
Pi as well. It's just until recently not many people were using PAPI on
ARM so it hasn't had as much testing.
Vince
_______________________________________________
perfmon2-devel mailing list
perfmon2-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
