Re: 2.9.5 and 2.1.73

[David Bateman]

John W. Eaton wrote:

>I'd like to make a new set of snapshots soon (this week, if
>possible).  Are there any outstanding bugs that you think must be
>fixed before I do that?
>
>Thanks,
>
>jwe
>
>
>  
>
Yes. I have quite a few changes, in particular to the sparse rectangular
solver code I recently introduced. In fact in is partially disabled in
the current CVS. I have a couple of issues (ie seg-faults) with the code
at the moment but will send it hopefully before the end of the week.

There are also three features I have patches for I'd like to add for 2.9.5

* Disable the calculation of the condition number for the diagonal and
triangular sparse solvers. Given the manner in which these solvers work,
its generally not possible to pass these singular matrices. You have to
force the matrix type to be able to. The calculation of the condition
number can often take orders of magnitude longer than the calculation of
the solution for these two cases, and so disabling the makes the speed
comparable (or slightly faster) than matlab.
* Special case sparse permutations like A = B(p,q) where p and q are
strict permutations (ie the elements of p and q are unique). This gives
several orders of magnitude is speedup of the above indexing operation
* Sparse Dulmange-Mendelsohn solver. I'd like to have this in 2.9.5 so
that my Octave 2006 paper can refer to 2.9.5 as the version the
benchmarks are created against. However I have a question on how to use
C++ template functions in octave. My problem is I need to write
"dmsolve<SparseMatrix,SparseMatrix,SparseMatrix>(...)" rather than
"dmsolve(....)" to call the specific version of a template function. I
can find no instances of the use of such calls to specific template
functions in octave and so I'm not sure of the best way to include the
code. I attach my current version of the code as a separate function
dmsol, so if you can comment on the best way to include this I'd
appreciate it..

Cheers
David


-- 
David Bateman                                address@hidden
Motorola Labs - Paris                        +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 6 72 01 06 33 (Mob) 
91193 Gif-Sur-Yvette FRANCE                  +33 1 69 35 77 01 (Fax) 

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary