Actually, a factor of 2 or even 3 is not so bad, considering the
convenience of not having to import, export data files in other formats.
Of course there will always be some program which handles a particular
routine better than others.
I, too, want to thank you for porting lapack. Whatever the speed, it
will be a real plus to have it available.
-sen
On Wed, 3 Jan 2007, Raymond Toy wrote:
>>>>>> "sen1" == sen1 <sen1 at math.msu.edu> writes:
>
> sen1> 3. Perhaps the ensuing discussion simply made alternate proposals
> sen1> which would not have a degrading effect on the speed of the final
> sen1> routines. If so, that is fine.
>
> sen1> If not, then
> sen1> I would rather see things done right (even if it takes longer to
> sen1> implement).
>
> sen1> Otherwise, I don't see the point for real utility. Serious users
> sen1> will simply go to octave, python, matlab, etc. or some other tool
> sen1> after possibly doing some testing in maxima.
>
> Even without testing, I am very confident in telling you that you will
> be disappointed if you are expecting these LAPACK routines to be as
> fast as octave or matlab. You'll be even more disappointed if you're
> doing large matrices and octave/matlab are using an optimized ATLAS
> library.
>
> In addition, I didn't even compile the code with (speed 3) (safety 0)
> so there will some loss in speed anyway because there will be type
> checks in many places. And the converted routines aren't using
> specialized arrays, so you get hit there as well. I would guess maybe
> a factor of 2(?) loss in speed compared to Fortran, and even more if
> you're used to ATLAS or some other optimized BLAS package.
>
> If all you want to do is crunch numbers, maxima may not be what you
> want.
>
> Ray
>
--
---------------------------------------------------------------------------
| Sheldon E. Newhouse | e-mail: sen1 at math.msu.edu |
| Mathematics Department | |
| Michigan State University | telephone: 517-355-9684 |
| E. Lansing, MI 48824-1027 USA | FAX: 517-432-1562 |
---------------------------------------------------------------------------