[HPGMG Forum] Do we want the benchmark to go into intrinsics?
mfadams at lbl.gov
Tue Apr 29 17:32:29 UTC 2014
On Tue, Apr 29, 2014 at 1:14 PM, Sam Williams <swwilliams at lbl.gov> wrote:
> optimized = simd intrinsics, array padding to avoid cache aliasing
> cognizant of line size, associativity, cache size, replacing omp workshare
> with static parallel regions and manually fusing operations/communication
> together, etc...
> I think you can have a couple reference implementations...
> - C with MPI and/or OpenMP (so you can run on distributed memory machines
> or SGI UV ccNUMA)
> - CUDA with MPI (little hesitant on this given there is either a new
> version or new release candidate of CUDA every 6 months)
> The kiviats were constructed with code as is.
What is "as is". Many of these apps were LLNL apps that have been running
on BG for years. They are certainly optimized for it. All apps are
optimized to some degree or they would not qualify as an app.
> HPCG defenders promise better performance with optimization.
> Showing HPGMG optimized performance and performance characteristics looks
> better than vanilla code/HPCG doesn't address the argument. It only says
> optimization is beneficial.
I hate to try to define "optimized". I don't think we have the resources
to adjudicate this and it will just piss people off and cause debate (like
this) that is not useful as far as I can see.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the HPGMG-Forum