Yes, I'll do that. Thank you.

Subhankar Pal


On Mar 13, 2017 4:47 AM, "Andreas Hansson" <[email protected]> wrote:

Hi Subhankar,

I think the best option would be to resolve any outstanding issues to
properly support the AddrRange interleaving in the cache. It has definitely
worked in the past, at least with the caveats I mentioned earlier. Could
you do a bit more digging?

Andreas

From: Subhankar Pal <[email protected]>
Date: Sunday, 12 March 2017 at 04:24
To: gem5 users mailing list <[email protected]>, Andreas Hansson <
[email protected]>
Subject: Re: [gem5-users] Caches with interleaved address ranges connected
to HBM

Hi Andreas,

Thank you for answering my question. I looked at MemConfig and I see that
it assigns interleaved address ranges to memory controllers. The major
difference I found with respect to assigning interleaved address ranges to
a cache was that the cache seems to accept a Vector of address ranges,
whereas the mem controller takes a single address range, which causes
different functions to get called in the C++ code (src/base/addr_range.hh).
I get a fatal error if I try to assign an address range to a cache with
some interleave bits. It seems to have the requirement of* # address ranges
= 2^(# of interleave bits)*.

Say, I want to split and interleave my address range between four caches (#
of interleave bits = 2). I am thinking of trying to manually append partial
address ranges to each of the four caches (populate the Vector of address
ranges) to create interleaved pattern, instead of using the interleave
mechanism present in the C++ codebase. Can you tell me if that would work?

Thank you!

Subhankar Pal

On March 11, 2017 at 1:43:16 PM, Andreas Hansson ([email protected])
wrote:

Hi Subhankar,

Each cache has an address range, and it is by configuring these ranges that
you can create a number of interleaved caches. I would suggest having a
look at configs/common/MemConfig for an example of how this is done. Each
range supports “simple” interleaving by simply using a stride size, and it
also supports XOR-based hashing.

Something you need to keep in mind is that the caches currently are not
taking the interleaving into account when determining the set/way. This is
something that has been a “todo” for quite some time, and unfortunately we
have not yet created a simple and elegant way of doing this.

In the case of simple interleaving you should simply ignore those bits in
the cache to avoid a “loss of entropy”. For the hashing it gets more
complicated, as no specific bits can be dropped.

I hope this e-mail can perhaps also start a discussion on how to sensibly
accommodate more advanced interleaving, and do so without the
aforementioned caveats. If anyone has suggestions for a good design, please
speak up. It would be good to make sure that it “just works”.

Andreas

From: gem5-users <[email protected]> on behalf of Subhankar Pal <
[email protected]>
Reply-To: gem5 users mailing list <[email protected]>
Date: Saturday, 11 March 2017 at 01:56
To: "[email protected]" <[email protected]>
Subject: [gem5-users] Caches with interleaved address ranges connected to
HBM

Hi all,

I am trying to build a system where processors are connected (via a
crossbar) to multiple caches having interleaved address ranges. These
caches are each connected to one channel of an HBM. How do I interleave the
address ranges over the caches and how do I connect them to an HBM memory?

Your help is highly appreciated!

Subhankar Pal
IMPORTANT NOTICE: The contents of this email and any attachments are
confidential and may also be privileged. If you are not the intended
recipient, please notify the sender immediately and do not disclose the
contents to any other person, use it for any purpose, or store or copy the
information in any medium. Thank you. ______________________________
_________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

IMPORTANT NOTICE: The contents of this email and any attachments are
confidential and may also be privileged. If you are not the intended
recipient, please notify the sender immediately and do not disclose the
contents to any other person, use it for any purpose, or store or copy the
information in any medium. Thank you.
_______________________________________________
gem5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users

Reply via email to