fenics team mailing list archive

["Robert C. Kirby" <kirby@xxxxxxxxxxxx>]

> The main point I was attacking with my post was the suggestion that we
> should sacrifice performance by not pre-compiling code, since the
1.) We don't know that not-precompiling code sacrifices significant  
performance.  This is not a settled question.  It could be that the  
optimal run-time code is only epsilon more expensive than the optimal  
compiled code.  It largely depends on how the systems are  
engineered.  An interesting comparison (don't want to ruffle any  
feathers though) would be to make a run of solving some problem in  
DOLFIN and in Sundance and compare the time to assemble the matrix.

> Linux version Blue Gene runs (and perhaps other supercomputer
> operating systems) apparently can't do dynamic loading. This would
Your comment about "only running on supercomputers" is well-taken but  
slightly overstated -- there are lots of places besides the laptop/ 
desktop/cluster/supercomputer spectrum where we might be interested  
in running.  Kevin's design decision were to make something portable  
across the spectrum of machines that he runs on -- it can run on a  
linux box/cluster, but he doesn't have to do too much to do his  
production runs.

> impact your laptop user, since his code would run slower. A better
> solution as far as I can see would be to fix Linux on Blue Gene so
> that it can do dynamic loading, or alternatively avoid dynamic loading
> by recompiling and reuploading the code to the nodes between runs.

The first is neither possible nor practical.
1.) Blue Gene runs a very stripped-down kernel (parts of it are  
highly guarded as they contain proprietary IBM secrets, I believe) to  
allow maximum memory usage per node.  The claim is that when you  
start adding stuff to the kernel (like dynamic loading) you will lose  
available memory for the highest-end runs.  Hence, even if you wanted  
to spend all your time rewriting the BlueGene kernel, they wouldn't  
let you.

2.) cross-compiling on the other hand, including generating code for  
forms instead of having code to interpret them, could be an eventual  
possibility.


Rob Kirby

"Mathematical software should be mathematical."