On May 28, 11:01 pm, "Bill Page" <[EMAIL PROTECTED]> wrote:

Hi Bill,

<SNIP>

> > It seems to expect gcl to be installed, so it is likely that once you
> > install a system wide gcl it would get beyond that point. In the past
> > there was trouble building FriCAS/Axiom/OpenAxiom with the clisp
> > Sage provides since we do not build libsigsev [Maxima doesn't require
> > it] and we do wrap the binary clisp into a thin shell wrapper because
> > otherwise on relocation clisp does not find its default memory image.
> > Maxima is build with some switch telling it to use clisp as clisp.bin
> > for example.
>
> Apparently this is a result of changes made by Burcin Eroal.
>
> Fricas without X support can be built with the version of clisp
> included with Sage. X is only required to run Axiom graphics and
> Hyperdoc - neither of which are normally accessible from within Sage.

ok, I didn't know that, so thanks for clearing that up.

> http://groups.google.com/group/sage-devel/browse_thread/thread/90bd5c...
>
> > We will soon [a months or two] switch to using ecl [Maxima in
> > interpreted mode works already] and FriCAS's test suite on
> > the latest ecl is actually about 30% faster than clisp. And there
> > are already speed ups beyond that.
>
> Has this result for FriCAS been previously reported somewhere?

Yes, on the ecls mailing list on May 10th, 2008:

[Ecls-list] FriCAS timings

I have compared speed of running FriCAS testsuite.  All measurements
are on 1.8 GHz Athlon 64 (Athlon 3000) running Gentoo.  I compared:

- ecl CVS 2008-05-09 15:30, build with --enable-smallcons and
  setting safety to 0 during FriCAS build
- ecl CVS 2008-05-07 11:23, build with default settings and
  setting safety to 1 during FriCAS build
- Gentoo sbcl 1.0.9
- Gentoo clisp 2.43

Results (measured by 'time'):

            ecl 05-09     ecl 05-07       sbcl        clisp

real        36m16.106s    162m43.116s    10m1.985s   47m51.595s
user        35m49.750s    162m14.250s    9m38.560s   47m10.280s
sys          0m25.880s      0m28.280s    0m23.240s   0m41.010s


I did not measure gcl and Open MCL times, but past measurements
indicate that both are faster than ecl and slower than sbcl.

Nice things is that now ecl based FriCAs is faster than clisp
based one.

Remark: I had some noise measuring build time, so I do not report
it.  But it seems that ecl 05-09 gives about 20% saving in build
time compared to ecl 05-07.

-- Waldek Hebisch [EMAIL PROTECTED]
-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save
$100. Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________ Ecls-list mailing list
[EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/ecls-list
[end quote]


Notice today's email by Juan, so there are good things to come and he
has been responsive to eliminate bottle necks in FriCAS or Maxima code
that so far has escaped to be noticed by the other lisp code people
have used ecls with:

[Ecls-list] Status of ECL

Still pending to get CVS imported in Sourceforge, there is progress in
my local tree. I have optimized TYPEP and COERCE, at least in the
compiled versions, though the interpreted ones can also benefit if one
wishes so -- the mechanism is based on compiler macros --. In
addition, I discovered bottlenecks in GENSYM, GENTEMP, MAKE-ARRAY and
MAKE-VECTOR, which have been corrected.

Before, runing Paul's testsuite

real time : 572.967 secs
run time  : 396.084 secs
gc count  : 647 times
consed    : 5325200048 bytes

Right now

real time : 476.979 secs
run time  : 268.773 secs
gc count  : 590 times
consed    : 3833384924 bytes

None of the optimizations are usafe: it is just using information
about the known types to produce the simplest type checks and type
coercions possibles. And everything is rather standard Common Lisp, so
I expect it to be robust.

Juanjo
[end quote]

> What is the build time for the complete system?

Somthing about that come up today when Waldek inquired about the start
up time of ecl. It is slower than the other lisps, but that discussion
seems to be ongoing.

> Is anyone working on making preparing a Sage spkg for FriCAS based on ecl?

Not to my knowledge, but something in that direction was discussed
with Martin Rubey in sage-devel a couple weeks back.

Ecl in Sage is currently being held up by me working on the Solaris,
OSX 64 bit and Cygwin ports, but it should be fairly straight forward.
We only need to package boehmgc as dependency, so that is something
that will happen soon. Since M2 also depends on boehm gc it will
happen before their M2 meeting in Snowbird Utah at the end of June
since some Sage people will be there and then the current M2 will
exist as a working optional spkg again. It is on my to do list in the
medium priority, but the work is basically done.

>
> > ...
>
> Regards,
> Bill Page.

Cheers,

Michael
--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at http://groups.google.com/group/sage-devel
URLs: http://www.sagemath.org
-~----------~----~----~----~------~----~------~--~---

Reply via email to