[Pkg-scicomp-devel] Bug#502168: Bug#502168: New upstream package

David Bateman David.Bateman at motorola.com
Tue Oct 14 12:33:33 UTC 2008


Thomas Weber wrote:
> Am Dienstag, den 14.10.2008, 08:59 +0100 schrieb David Bateman:
>   
>> Package: suitesparse
>> Version: 3.1.0-3
>>
>> There is a new upstream release of suitesparse available that includes a 
>> new sparse QR factorization and solver that is significantly faster than 
>> anything else available. I'd like to be able to include this in Octave 
>> to improve the rank deficient, under- and over-determined solvers. Could 
>> I request that this new version of suitesparse be packaged and in 
>> particular that the new SPQR package be built. The upstream version 
>> number of this new release is 3.2.0.
>>     
>
> Attached a changed 13-UFconfig_UFconfig.mk.dpatch and a new
> 14-memcpy_needs_stringH.dpatch. 
>
> The latter should probably go upstream.
>   
Reading the installation note of

http://www.cise.ufl.edu/research/sparse/SuiteSparse/

I believe upstream is already aware of this issue.

> The former has two changes: upstream changed the line for BLAS = 
> and the new SPQR stuff needs -DNOPARTITION as well (I didn't check the
> timing stuff, is it useful?).
>   
-DNOPARTITION basically just removes METIS from the build, which is a 
good idea as METIS although being a very good graph partitioner is not 
GPL compatible. I'd say don't include the timing stuff, as it seems to 
just be there for testing the code, and will slowup the factorization.

Cheers
David



> 	Thomas
>   


-- 
David Bateman                                David.Bateman at motorola.com
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






More information about the Pkg-scicomp-devel mailing list