dan.stanger@ieee.org wrote:
> In a previous message to the list, I saw a bench mark that indicated
> that clisp was only about 30% slower then gcl. Wouldnt our time
> be better served by building the stuff in clisp that gcl has,
> which I thought was tk support, and then putting our time into cmucl
> and making it work on more platforms? clisp is a gpl'd product also,
> works on many more platforms than gcl. Plus, it implements CLOS,
> and many more features then gcl.
> Dan Stanger
> _______________________________________________
> Maxima mailing list
> Maxima@www.math.utexas.edu
> http://www.math.utexas.edu/mailman/listinfo/maxima
I do not think that damping GCL support is really good idea.
First of
all we do not have so many Lisps readily available as base Lisps
to
Maxima and I'm not sure what will happen to any of them in
future.
Having just three viable (some of them barely viable at present)
alternatives is really minimal number to fill safe.
As for performance. I have Maxima running on gcl 2.3.8, gcl
2.4.0, clisp
2.27,
cmucl 2.4.17, cmucl 18c on one Linux machine with 384Mb RAM and
Pentium III 733MHz CPU.
In all my tests clisp is 2-3 times (not 30% !!!) slower than gcl
and
cmucl.
The performance of gcl and cmucl is approximately the same but
fastest is gcl 2.4.0 with gmp support. Note that with gcl I pre
allocate
large chunk of ram at start since without this gcl slows down
substantially.
Vadim
--
[ Vadim V. Zhytnikov <vvzhy@mail.ru> <vvzhy@td.lpi.ac.ru> ]