Re: improved matrix_type check

[Jaroslav Hajek]

On Thu, May 22, 2008 at 2:33 PM, David Bateman
<address@hidden> wrote:
> Jaroslav Hajek wrote:
>> I didn't mean my question as asking for them. I guessed that part of
>> the reason was the fact that UMFPack et al. doesn't support single
>> precision, so the linear algebra stuff would have to be done via
>> conversion anyway. OTOH, single precision support available for BLAS,
>> LAPACK, FFTW and libraries in libcruft.
>>
> Yes that is one of the major pains of trying to port single precision
> types to sparse matrices/
>
>>> It seems to me that float sparse matrices are
>>> probably more useful than float matrices.
>>>
>>
>> Why?
>>
> If you are using sparse matrices, one of your concerns is that you don't
> have enough memory to store the dense version of the matrix... So having
> single precision would allow a matrix size roughly sqrt(2) large at the
> same density without having to use specialized out-of-core techniques.
>
But you can say the same for dense matrices. I don't see why
stretching your memory to its limits should be more common with sparse
matrices.

Another concern may be speed - especially if it's memory traffic that
slows the computations down.

> Regards
> 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
>
>



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