Hi Andreas,

My thoughts are inline below.

On Thu, Apr 6, 2017 at 4:04 AM Andreas Sandberg <[email protected]>
wrote:

> Hi Everyone,
>
> Gabe's heroic effort to refresh the reference stats (thanks Gabe!) have
> once again highlighted issues in the current test regime.
>
> My impression is that some of the affected tests are only marginally
> useful. Especially for architecture that are largely unmaintained. I
> don't want to rekindle the discussion about retiring ISAs. However, I
> would suggest that we reduce the amount of time we spend testing these
> ISAs. In particular, I would like to see these changes:
>
> 1. Remove all SPEC-based tests for poorly supported ISAs
>

Yes. In fact, I think that in the medium-term we should remove all
spec-based tests since we can't distribute the binaries.

>
> 2. Remove stat diffing for all remaining tests for poorly supported ISAs.
>

Sure. The stats for each component should be ISA-independent and covered by
other, better supported, ISAs. The only thing we would lose here is stats
diffing when these ISAs are updated (very rarely).


> 3. Remove stat diffing for the learning gem5 tests. We mainly want to
> make sure that these scripts can be run.
>

Definitely! I thought this was already done...

>
> 4. Consider removing O3 (and minor?) tests for poorly supported ISAs
>

We've said this many times... we should construct the matrix of what ISAs x
CPU models x coherence protocols we support. If it's marked as supported on
the matrix, we should test it. I don't know if these models are currently
"supported" for any of these ISAs.

>
> I would argue that "poorly supported ISAs" would be the following:
>    * SPARC
>    * POWER
>    * MIPS
>    * ALPHA?
>

Seems reasonable to me.


>
> Thoughts?
>
> //Andreas
>
> 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-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev

Reply via email to