Hi, I saw one of paper (Cycle Accounting Analysis on Intel Core2 Processors). In this paper it's given RS_UOPS_DISPATCHED (CMASK=1,INV=1) gives the
number of stall cycles for a processor in a given interval. Suppose
stall cycles given for a program's execution , by the above assumption
is C(stall) and the useful cycles are C(useful) at frequency F. Total cycles
C= C(useful) + C (Stall)
If I modify the frequency to f, the total number of cycles taken above is
C" = C(useful) * (f/F) + C(Stall)
Is this above formulation correct w.r.t to the stall cycles given by the event RS_UOPS_DISPATCHED?
The errant assumption here is that C(Stall) is independent of F. In fact, if you increase F, you're more likely to see an increase in C(Stall)the nature of whichI won't try to predict, beyond guessing it's pretty nonlinear.