[Ktechlab-devel] Converting to Eigen - patch for revision, ver.1

[P Zoltan]

On Fri, 21 Aug 2009 21:55:58 +0200, Alan Grimes <agrimes at speakeasy.net>  
wrote:

> P Zoltan wrote:
>>  A flowcode (let's do things with style :D) made with ktechlab,
>> attached, shows the current operations when one step in the simulation
>> is done. It seem to me that it's too complicated and inefficient. It can
>> be seen that in Simulator::step() there are quite a few loops, and in
>> each loop ElementSet::doNonLinear also makes  internal iterations.
>> That's not needed, as far I know.
>
> I had forgotten about flowcode. =P
>
> Good start, looks like a great test for the flowcode code too. =)
>

  It crashes quite often ;)

  You said before that there are some book about modified nodal analysis  
(?). What is the theoretical complexiti of that algorithm? Looking at the  
current implementation, it seem to me that ktechlab implementation is  
needlessly too complex.

>>  My goal now is to make the test circiut (also attached) to run
>> decently, without 100% processor usage. The big question is how to do  
>> it?
>
> The circuit runs just fine for me right now with less than 50% of a
> single CPU (2-way SMP machine). I'm using the old engine. Naturally,
> things slow down if I put in a voltage probe... (which might be a
> problem. probably a good target for multithreading...)
>
>

  With the old engine it's not that interesting, maybe for comparison. With  
the new engine ktechlab uses one core 100% if the processor is on 800Mhz,  
and 60% on 1800Mhz. It would also be interesting, to make a release or  
just normal build, without debug information, and see how fast is that.

  What other people think about the patch? Is it fast enough to be included?