versions.
> If a given event type is not measured, the formula gets truncated
> accordingly.
>
> > I mention this because I remember at some time the LO crew were
> > complaining that the cycle count from Callgrind was completely misleading
> > due to not taking into
not taking into account global bus locks.
Hm, ok. The "20 Ge" estimates that locking often can be resolved at the
L2/L3 level. However, I never did any measurements myself. I think you
mentioned something ...
Josef
>
> J
>
> (*) eg. the first version of the memcheck leak c
