Re: About ITSOL - ILUTP - MILU and packaging

[saad]

Hello,

Sorry I did not get the chance yet to look carefully at these messages yet.
I do not have a copy of the  stand-alone versions Ruipeng sent you..
Regarding releasing a separate library which will be the same as the one used in Octave,
[for maintenance purposes] -- it may be doable but as you know in academia
students come and go -- so maintenance by people in my group is unlikely to be
reliable.. It will help if the codes are not too different from the ones we have - so I can
direct students to make fixes if they are simple ones.

For the question regarding ``a minimal usage example of ILUTP"  -- there is normally
one within the ITSOL package which can be converted to the stand-alone version.
This may be a little more tricky than the other preconditioners because of the additional
 permutation..
I can take a look at this but the problem is that I do not  know what Ruipeng sent you...
It may be best to wait until he returns [he is currently in China]..

I do want to help within our means

Yousef Saad


On 07/09/13 10:30, Kai Torben Ohlhus wrote:

Hi Ruipeng,

about one month has passed and the ITSOL interface for GNU/Octave has grown. Here are three excerpts from past conversations, which I want to refer to.

On 05/31/2013 04:32 PM, Yousef Saad wrote:

> 

>   Hi,

> 

>   Both Ruipeng  and I will  be able to  help with advice  and guidance.

>   For one thing  we have stand-along implementations of  ILUT (+ ILUTP)

>   and ILUK  [not available in matlab-  except for ILU0]  which are more

>   efficient and  easier to implement  than those you will  extract from

>   ITSOL.  I will be fairly available after July 15th.

> 

>   With best regards,

> 

>   Yousef Saad

> 

On 05/31/2013 05:04 PM, Ruipeng Li wrote:

> Hi, Kai

> 

> sure we can talk about it anytime you like and also I can provide you 

>   the standalone C code of ILUTP and ILUK.

> 

On 29 June 2013 06:12, c. <address@hidden> wrote:

Marco, Kai,

We don't want to be maintaining a separate version of the
ITSOL distribution, so, before going too far in patching ITSOL
I think it is worth discussing your changes with the upstream
maintainers so that they are aware of what is going on and can
consider making the changes in their package directly.

Ruipeng,

It seems that the ITSOL bulding and linking process needs to
be changed a little bit to improve portability, in particular
(but not only) to windows systems. Would you mind considering
to include those changes in your distribution of ITSOL?

c.

As c. points out, the Octave project will benefit a lot from being able to use the ITSOL library. Currently I can run ILUK, ILUT and ILUC from Octave without problems. But in order to get a shared dynamically linked ITSOL library version, one has to change it's original source code and the Octave project's intention is not to become a maintainer an ITSOL-fork. So if you planned to release a new version of ITSOL, maybe including the new standalone implementations of ILUTP and ILUK, I would be appreciated to support the integration of the GNU build system.

Another question I have is regarding a minimal usage example for the ILUTP implementation of the current ITSOL version, which I could not use satisfactorily so far. It differs from the other ILU-versions as it doesn't offer the typical L-D-U output. Especially I need guidance on how to initialize the input and output data structures correctly.

Finally I would like to ask you for advice about the implementation of the MILU-option MATLAB offers for all ILU-types http://www.mathworks.com/help/matlab/ref/ilu.html. As far as I understood MILU, you sum up all dropped elements from the factorization process and subtract them from the diagonal. Do I have access to these elements from the "outside" of the function or would it be necessary to create own implementations of the ILU-types to archive this?

Thank you very much for your help. If you are interested in the progress of the Octave interface, I have prepared a little demo version of the current state.

https://docs.google.com/file/d/0B4V9W1l15-xMVTFRamFkWkhUbEU/edit?usp=sharing

With best regards,

Kai

-- 
-----------------------------------------------------------------------
Yousef Saad
University of Minnesota
Dept of computer Science & Eng.           address@hidden
4-192 Keller Hall     	      	      	  ph:  (612) 624 7804
200 Union st. SE.                         http://www.cs.umn.edu/~saad
Minneapolis, MN 55155 	      	      	  fax: (612) 625 0572
-----------------------------------------------------------------------