I hope I am not miunderstanding this thread.
1. Richard proposed some suggestions for making the lapack (and other
numerical routines) fast.
2. It seems that some of his suggestions might be hard to implement,
(e.g., the ddmatrix or something similar)
given the current state of maxima source code.
3. Perhaps the ensuing discussion simply made alternate proposals
which would not have a degrading effect on the speed of the final
routines. If so, that is fine.
If not, then
I would rather see things done right (even if it takes longer to
implement).
Otherwise, I don't see the point for real utility. Serious users
will simply go to octave, python, matlab, etc. or some other tool
after possibly doing some testing in maxima.
-sen
On Wed, 3 Jan 2007, Raymond Toy wrote:
>>>>>> "Robert" == Robert Dodier <robert.dodier at gmail.com> writes:
>
> Robert> On 1/3/07, Raymond Toy <raymond.toy at ericsson.com> wrote:
>
> Robert> But even better would be to put them in src/lapack and compile
> Robert> them at the same time as everything else, but just omit them from
> Robert> the final image. Then load(lapack) would just load the lapack binaries
> Robert> instead of compiling them.
> >>
> >> I don't know how to do that. Maxima's build system for src is a bit
> >> messy. The defsystem file compiles and loads everything. Then it's
> >> done again, using the same defsystem to load everything. Hence,
> >> everything gets loaded. We'd have to rearrange the defsystem and
> >> build scripts.
>
> Robert> OK, let's not mess with the src build system. Let's put it in
> Robert> share/lapack. Can we compile it there so that users don't need to?
>
> Yeah, that we can probably do by running maxima one additional time to
> build it. I'm not sure what this will achieve though, unless you also
> ship the result fasls to the user. I don't think we do that today, do
> we?
>
> >> We could set *compile-verbose* and friends not to display these
> >> progress messages. However, on my ppc system it takes a long time
> >> (15+ minutes?) to compile, so having some messages go by gives the
> >> user a nice warm fuzzy that something is actually happening.
>
> Robert> 15 minutes?! I think users are going to be unhappy about that.
> Robert> I know I will be.
>
> I may have been exaggerating, but testing now on an unloaded 1.5 GHz
> sparc at work, cmucl takes 413 sec to compile lapack. IIRC, your
> machine is much slower than this, so not matter what, I think you'll
> be unhappy with the compile times. Clisp, however, is much faster, of
> course, and only takes 47 sec. I don't know how long gcl would take.
>
> Sheldon's suggestion about configure --with-lapack is also doable.
>
> But the bottom line is somebody has to take the hit. And I'd prefer
> it not be a developer since his time is scarce. :-)
>
> Ray
> _______________________________________________
> Maxima mailing list
> Maxima at math.utexas.edu
> http://www.math.utexas.edu/mailman/listinfo/maxima
>
--
---------------------------------------------------------------------------
| 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 |
---------------------------------------------------------------------------