Hi all,

Thanks for the interest in my question. :)

I want to know one additional detail. How does oprofile get these numbers? Does it somehow connect with PMU of the processor? The Sandy Bridge Architecture has MSRs that enable monitoring of uncore events. Does Oprofile measure them?

Thanks again

On Wed, Mar 6, 2013 at 9:45 PM, Andi Kleen <andi@firstfloor.org> wrote:
> IIUC, you're saying that for the unit mask files that you provided, we *can* assume there would never be more than one alias that does *not* have an extra parameter.  As far as I know, there aren't any unit mask files that have aliased UM values from anyone but you, so that gives me hope we can come up with an algorithmic solution to the problem versus having you go the hassle of hacking your unit mask files to add dummy extra fields.  And then for future submissions where aliasing is used, we would just need to ensure the "rule" was followed so our algorithmic solution doesn't break.  Sound reasonable?  If, on the other hand, you would rather add the dummy extra flag, I would be OK with that, too.  And then that would be the rule that future submissions would have to follow -- i.e., all aliased UM values *must* have an "extra" field, even if it's just a dummy field.

I can just script the changes of the unit masks. Need a small patch to
add a dummy flag though. I can look into it today.

I auto generate all the unit masks anyways, so yes it's easy enough to
implement that rule in the generation scripts.


Rakhi Hemani
Phd Scholar
Roll Number PhD1207