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

Sam Williams swwilliams at lbl.gov
Tue Apr 29 17:28:45 UTC 2014


There's also a deeper question on how you evaluate/rank.

I was thinking of requiring submitters to submit 6 numbers at a given concurrency.
- reference implementation for N, N/8, and N/64 DOF/process
- full fury implementation for N, N/8, and N/64 DOF/process
Ranking would be based on a linear combination of these (the first three could be weighted by 0) 



On Apr 29, 2014, at 10:24 AM, Mark Adams <mfadams at lbl.gov> wrote:

> 
> 
> 
> On Tue, Apr 29, 2014 at 1:06 PM, Sam Williams <swwilliams at lbl.gov> wrote:
> Bad compilers should be shamed.  Having it autodetect __bgq__ (or whetever) and run the best manually optimized implementation by defaults hides deficiencies.
> 
> 
> Humm, we really need to get on the same page with the HPGMG model and philosophy.  I will be in Berkeley Thursday all day and probably Wed PM.
> 
> We can start this with email perhaps. 
> 
> Vendors have to be able to use whatever the heck they want so what is the difference if Jed writes the BGQ kernels or John Gunnells?  John has to contribute it to the repo anyway so the only difference is who shows up with "blame" and when it gets released if John want to wait until the code is used for a Top500 list and is thus what we are calling "published" and must be made public.  At least this is the model that I had in mind.  You seem to have a different model.  We could keep some definition of optimized in branches perhaps ...



More information about the HPGMG-Forum mailing list