Hello,
Im using the FVP model RevC-AEMv8a, and i'm trying to verify the performance of my system with it. It is possible to verify the cache misses values?And another question, i found it in the documentation, and can't understand this statement, can anyone explain it?
Thanks :)
FVPs are functional models for software development, but they are not performance models. You can measure instruction counts, but cycle counts will be inaccurate because caches and the downstream memory system are not modelled.
HTH, Pete
This is right, even if i enable this model parameter?
Hi GGFreitas,
Some level of functional modelling of the caches is possible in the fast models; see Functional caches in Fast Models. This will allow to report cache misses to the software for example.
When combined with Timing Annotation and when the cache miss latencies are configured, it is possible to see an impact of the cache misses on software performances (even if the resulting timings will be very different compared to a real hardware).
Hi,
I'm using this parameters, in order to check the PMU values (e.g. cache misses) but his value does not incremet... (my l2 cache size is 512k, and im using a 1M working set)
The configuration I was using by the time was:
cache_state_modelled=truecluster0.dcache-miss_latency=1270000