Depends on how you decide to solve the coherence problem. There will
probably be some hacking on the caches as well (for example if you
decide that you need to enforce inclusion in the L2s).
Steve
Forum Mangal Parmar wrote:
got it. thanks :)
one last question. what files must be modified? only the coherance_protocol
files?
Thank you,
Forum
Quoting Steve Reinhardt <[EMAIL PROTECTED]>:
There shouldn't be any issues other than coherence. Structurally it's
no problem to build whatever kind of cache hierarchy you want (including
the one you describe), it's just that the existing coherence protocol
won't deal with it properly. You'll have to decide for yourself how you
want to manage coherence with your hierarchy and then implement that.
Steve
Forum Mangal Parmar wrote:
Hi,
I am not sure if I was clear in the last mail.
I am looking at a multi-cpu scenario.
Currently the M5 1.0 has private L1 caches and a shared L2 cache.
I want to make private L2 caches for every cpu and add another shared
L3
cache.
I just wanted to make sure that I was on the right track. I am planning
to
add the L2, L3 cache defnitions with appropriate bus interfaces in my
python scripts.
However, Im thinking there could be coherance issues that may need to
be
resolved if the L2 is now private (for example like mentioned in this
post:
http://www.mail-archive.com/[email protected]/msg00223.html).
It would be great if you could point out any other issues I need to
take
care of.
Thanks a ton for the help,
Forum
Quoting Forum Mangal Parmar <[EMAIL PROTECTED]>:
I mean split per-cpu.
and yes I meant coherence.
Quoting Steve Reinhardt <[EMAIL PROTECTED]>:
When you say "split", do you mean split I/D or split per-CPU?
Also, I'm not sure what you mean by "synchronization": are you
talking
about coherence?
Finally, please post m5 questions to the m5-users mailing list rather
than sending them directly to me.
Steve
Forum Mangal Parmar wrote:
Hi Steve,
I am trying to understand the way the split cache implementation.
I had one very specific question about m5 1.0. If I want to make the
L2
split instead of shared in a multicore situation my question is:
1. Is it sufficient to make a .py file with each cpu having a L1 and
L2
cache (ie declare the L2 cache in a similar way to L1) and then
replicate
multiple cpus. In this case I will just replicate the L1 to get the
L2
and
make one-to-one bus connections between every L1 and L2 and the next
level
will be a shared L3 just delared in the same way as L2 is currently.
or
2. Do I need to change the c++ code for synchronization of the split
caches?
(ie do something like what is done in m5 2.0)
Thank you,
Forum
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
