I want to double check on some of the latency settings made in GEMs
before I run some tests. I am trying to determine how the latencies
are calculated.
The L1 cache access latency should just be
SEQUENCER_TO_CONTROLLER_LATENCY.
If the data misses in the L1, but is in the local L2, the latency
should also add in L1_REQUEST_LATENCY
if the data misses in the L2, but is found in another chips L2 (or
L1's) caches, what is the latency added, I think this is
L2_REQUEST_LATENCY, but I could be wrong here.
The latency for an L2 cache responding with data to another L2 cache
is ?
And I think the response the latency for the local chip's L2 to send
data to the L1 cache is just L2_RESPONSE_LATENCY
and the DIRECTORY latency is just DIRECTORY_LATENCY, however I'm not
too concerned with this latency when compared to the other ones.
If anyone can clarify what the above latencies are, I'd greatly
appreciate it. I'm currently playing with them to try and figure out
exactly which latencies are which, however I'd rather know for sure
what each of these corresponds to.
thanks,
Phil
|