I don't know much about the HSAIL model, and it was deprecated a while ago
in favor of the more up-to-date GCN3 model (which is in the process of
being merged into the develop branch, and is also fully available on its
own staging branch).  If HSAIL is the same as GCN3 though, there weren't
any cache stats in the protocol (which is why they were 0).  Support for
TCP and TCC hits/misses were added here:
https://gem5-review.googlesource.com/c/public/gem5/+/30174.  I have no idea
if this would apply cleanly to HSAIL though.  If not, you'd need to add
similar support.

Matt

On Sat, Jul 25, 2020 at 1:28 PM Marco Bianchi via gem5-dev <
[email protected]> wrote:

> Which stats specify the various L1, L2 and L3 hit and miss rates for the
> AMD GPU?
>
> Stats such as:
>   system.tcp_cntrl0.L1cache.demand_accesses,
> system.dir_cntrl0.L3CacheMemory.demand_accesses,
> system.tccdir_cntrl0.directory.demand_accesses
> are always 0.
>
> So, I am wondering whether I am looking at the right cache stats, or is
> there some issue in the simulation itself,
> or the stats for GPU related caches are never actually updated (I couldn't
> find a location where these are updated for the TCP, TCC, Directories.
> But they are updated for the cpu core pairs.)
>
> P.S.: I am using the GCN HSAIL model.
>
> Thanks in advance,
> Marco Bianchi
>
>
>
> <https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail&utm_term=icon>
>  Virus-free.
> www.avast.com
> <https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail&utm_term=link>
> <#m_4084755903454649598_DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>
> _______________________________________________
> gem5-dev mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
> %(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s
_______________________________________________
gem5-dev mailing list -- [email protected]
To unsubscribe send an email to [email protected]
%(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s

Reply via email to