ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Sat, Jan 20, 2007 at 10:13:30PM -0600, Andy Terrel wrote:
> Hello,
> 
> Here is a patch for optimize.py.  So the code generated for me using the 
> --optimize flag was three times as long and ran twice as slow.  I 
> thought it was in part because of all the 0.0 values in the generated 
> code.  This patch updates ffc so it doesn't generate all the tmp = 0.0 
> lines.
> 
> Alas it really didn't fix my runtime problem.  I would be interested to 
> know if people have gotten any speed up with this flag or not.
> 
> Thanks,
> Andy

What this flag does is that it calls FErari to compute an optimized
code and yes, it does give a speedup (for a range of forms). Rob and I
have just finished benchmarking a series of test cases and the maximum
speedup we get is a factor 3.29 for the local tensor and 1.47 for the
global assembly.

/Anders