[tuhopuu-devel] Help for tuhopuu's uv mapping - continued (2)

[Bruno Levy]

>Thanks a lot for all the efforts you are doing! We will be using OpenNL
>for solving the lineair system, and I'll be commiting it to tuhopuu cvs
>in one of the following days. 
>

Cool ! I'm glad if OpenNL is useful and used !

>We decided to do the LSCM part in blender's code, with blender's data
>structures, but the main.cpp with LSCM implementation will certainly
>help us implementing it.
>
  Yes, I'm sure that the best thing to do (this will avoid copying data 
structures
and moving data around). That's exactly the spirit of OpenNL (i.e. 
facilitating the
integration of numerical methods in existing software).  The "main.cpp" is
just an example on how to do that (I'd not use main.cpp's IndexedMesh
data structure in real application, it's pretty unefficient !)
  Please tell me if you need more explanations on main.cpp.

>Symmetric storage indeed seems to be an interesting thing to
>investigate.
>

... unfortunately, SuperLU 3.0's symmetric mode is not what I thought,
but it is interresting (it does not use symmetric storage, it is just a 
different
pivoting strategy for nearly symmetric matrices). I've included support for
it in OpenNL (updated on my web page), the benefit for LSCM is not
clear (did not get significant increase of performance). 
   The current version of OpenNL takes approx. 20 seconds to LSCM-ize 100K
triangular facets (on my Laptop Pentium-M 1.7 GHz), in both symmetric and
regular mode.
 
-- Bruno.