Re: performance measurement and caches

Andre Santos <alms@pesqueira.di.ufpe.br> writes:


> The cache behaviour plays a big role in the timings, and
> it is affected by far too many reasons (machine load, locality of
> other running processes etc.). The only solutions I know of are the
> two obvious ones:


[first solution deleted]


> - Use a simulator to count clock cycles (?). ...


> ... This way, in one go, you have a repeatable result.


Of course, it depends on how you plan to use the results. In
particular, if you are searching for an optimal solution to a problem
that is independent of the cache, then using a simulator that does not
model the cache is reasonable. However, if the cache significantly
affects or dominates the problem you are trying to solve, then a
simulator that does not model the cache will produce incorrect
repeatable results.


Regards,
Christopher Glaeser cdg@nullstone.com
--

Return to the comp.compilers page.
Search the comp.compilers archives again.