[HPGMG Forum] Do we want the benchmark to go into intrinsics?

Mark Adams mfadams at lbl.gov
Wed Apr 30 18:59:58 UTC 2014


On Wed, Apr 30, 2014 at 8:25 AM, Brian Van Straalen <bvstraalen at lbl.gov>wrote:

> ugh, you guys really get all over the map  :-)
>
> I think we don’t drift too far from the benchmark we are looking to track
> behind, which is really HPL.  Our hope is that HPCG is eventually buried
> once the primary actors have gotten their publications.
>
>
I have seemed to be the one that likes to say "just like [the great] HPL"
but now you are hitting me from the other side :)

LU and FMG are so different in terms of the computational processing -- by
design almost -- that I am wary of trying the "track" them that closely.
 We all have a lot of experience in deploying software to the masses.  I
think we can trust our judgment to some degree and be willing to learn and
adjust as we go along.

These issues seem to be the most pressing and I don't want to get locked
into a design without Jed being in on the conversation.  So if he can not
make it tom. I'd like to defer in depth design discussion until a later
meeting.

Two issues that have come up that Erich needs to give us guidance on are:

1) Would a "optimized" vs "non-opimized" distinction in the code base be
viable.  This came up yesterday.  Perhaps we have already made progress on
that or it morphed into the design conversation.

2) Can we insist that all code written for the benchmark be made public
after publication?  They could just make a library out of all the kernels,
package it, charge a $1 for the object files, say it is a "product" and so
they can't make it public.  Then we would have to decide if it is a fake
product (like what just described) or a real thing like MKL.  That is not a
good position for us to be in.  I also hate to forgo the publics ability to
vet codes and do it all ourselves.  We should discuss a policy on this.
Maybe Erich can tell us what HPL does and certainly his thoughts.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://hpgmg.org/lists/archives/hpgmg-forum/attachments/20140430/694e7661/attachment.html>


More information about the HPGMG-Forum mailing list