Hello,
I am working on hardware prefetching and cache coherence. Using a the
FullCPU model (with detailedconfig.py) the hardware prefetcher works fine,
and issues several prefetches. However, I would like to use the simple CPU
model so mye simulations would run faster, however, copying the cache
configuration seems to not work, as no prefetches are issued.
I have used Simpleconfig as a template, and use only a single level of
cache, like this:
class L1Cache(BaseCache):
size = '64kB'
assoc = 2
block_size = 64
latency = Parent.clock.period
mshrs = 4
tgts_per_mshr = 8
protocol = CoherenceProtocol(protocol='msi')
prefetch_miss = True
prefetch_degree = 8
prefetch_latency = 1
prefetch_policy = 'tagged'
In the working FullCpu model, I have used this:
class BaseL1Cache(BaseCache):
in_bus = NULL
size = '64kB'
assoc = 2
block_size = 64
tgts_per_mshr = 16
protocol = CoherenceProtocol(protocol=env['PROTOCOL'])
class DL1(BaseL1Cache):
latency = 3 * Parent.clock.period
mshrs = 32
prefetch_miss = True
prefetch_degree = 8
prefetch_latency = 1
prefetch_policy = 'tagged'
Which is similar to the example config provided with m5 v.1.1
I have tried to trace the problem by looking at which functions are called.
It seems like the constructor for the prefetcher is called, but the
function generating the prefetch is never called, after looking at the
functions calling the prefetcher, it seems that the call graph is different
when using SimpleCpu. Can anyone shed some light on this? Or am I doing
something completely wrong?
--
Marius Grannæs
_______________________________________________
m5-users mailing list
[EMAIL PROTECTED]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
