Re: [RFCv2 0/8] perf tool: Add new event group management

[Peter Zijlstra]

On Mon, 2012-04-16 at 14:16 +0200, Jiri Olsa wrote:
> also, any thoughts about the displaying question? ;)
> 
> > I'm looking on how to present this data in perf and it seems we need
> > to reset all siblings once we read/store them (in kernel) to the
> > leader sample.
> > 
> > My current thinking is to store siblings' sum values for each
> > hists entry of the sample (perf report count unit) .. and display
> > them in similar way we display callchains: for each hists entry
> > display the sum value for each sibling.
> > 
> > Could you provide more of your world examples? Your expectations about
> > presenting this..
> > 
> > Maybe we want to make the reset optional, and do some do some other
> > math with siblings' values..?


My thought was to do the exact same thing with it that we do with the
regular multi-event thing (which could be improved too). Simply create
individual event views with variable period based on the sample delta
for that particular event.

So suppose you have a group of 3 with only the leader sampling and you
get tuples like:

 IP=x {1000,  500,  750}
 IP=y {2000,  800, 1500}
 IP=z {3000, 1100, 2000}

This could be decomposed into 3 event views like:

 IP=x,period=1000	IP=x,period=500		IP=x,period=750
 IP=y,period=1000	IP=y,period=300		IP=y,period=750
 IP=z,period=1000	IP=z,period=300		IP=z,period=500

Just like what would've happened if you'd used multiple events like:

 attr = {
	.sample_type = PERF_SAMPLE_IP|PERF_SAMPLE_PERIOD,
	.freq = 1,
	.sample_period = 1000,
 }


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/