[MAGEEC] BEEBS characterization wrt Worst Case vs Average Case Performance / Energy

Simon Hollis cssjh at bristol.ac.uk
Fri Jul 25 12:08:01 BST 2014


K: this dovetails into the very extensive discussion we had in Lymington
about what extent it is possible to generate multiple input data values for
benchmarks.

This issue needs much more discussion (i.e. it is hard!), and directly
impacts the ability to do WC / AC analysis.

H


On 23 July 2014 19:13, Kerstin Eder <Kerstin.Eder at bristol.ac.uk> wrote:

> Dear All,
>
> As BEEBS is being characterized this summer, this is a good opportunity to
> see whether the benchmarks (or a subset of them) show significant
> differences wrt Worst Case and Average Case Performance / Energy
> Consumption. If not, can we find some that do?
>
> Here is why I think this may be interesting: Provided the difference is
> large enough, an interesting research question is whether MAGEEC will find
> different optimizations for the WC vs the AC, even though the source code
> is the same. ;)
>
> Would it not be cool if we could demonstrate that MAGEEC automatically
> finds different optimizations for WC vs AC Time / Energy for programs where
> there is a significant difference between these two cases? To allow MAGEEC
> to do this, we need a few benchmarks characterized wrt WC/AC behaviour and
> there being a significant difference between these.
>
> If MAGEEC can indeed to that, I think this would add significant value to
> a MAGEEC evaluation paper.
>
> Greg, please can you look into this and let me know your observations, and
> how difficult/easy you think this would be to do?
>
> Best wishes,
> Kerstin
>
>
>
> _______________________________________________
> mageec mailing list
> mageec at mageec.org
> http://mageec.org/cgi-bin/mailman/listinfo/mageec
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mageec.org/pipermail/mageec/attachments/20140725/c0f15389/attachment.html>


More information about the mageec mailing list