On Sun, May 30, 2010 at 3:37 PM, Carlo de Falco <[email protected]> wrote:
>
> On 26 May 2010, at 09:10, Jaroslav Hajek wrote:
>
>>
>> OK. Please note that your C++ version only works in real doubles,
>> whereas a properly written m-file function will probably work well in
>> both single and double (that is typically automatic, one only needs to
>> be careful about eps() and tolerances) and both real and complex (that
>> is sometimes harder, requires using ' and .' properly, sometimes it is
>> not possible).
>
> Jaroslav,
> I just commited an pgmres.m and mgorth.[m,oct] to linear-algebra.
> pgmres.m seems to be almost as fast as pgmres.oct, as long as mgorth.oct is
> used
> but is slower if mgorth.m is used.

Good, that works as expected, then.

> On the other hand the oct file version of
> mgorth
> suffers from the problems you noted above.
> What do you suggest?
> c.
>

I improved mgorth using a template to compute in single/double
real/complex. 3.3.51+ has even more powerful tools to reduce the
amount of boilerplate code for such tasks, but I tried to avoid those,
so I hope the file is compilable by 3.2.x though I haven't tried it. I
also let mgorth normalize the resulting vector and store it as the
last component of h, and modified pgmres.m accordingly.
mgorth.m is removed.

regards

-- 
RNDr. Jaroslav Hajek, PhD
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz

------------------------------------------------------------------------------

_______________________________________________
Octave-dev mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/octave-dev

Reply via email to