trouble with LAPACK + SBCL



On ??????? 14 ?????? 2009 00:18:58 Robert Dodier wrote:
> On 4/12/09, Valery Pipin <pip at iszf.irk.ru> wrote:
> > I've check out  dlaed6.lisp from 5.18.0 release tag.
> > Confirm that there are problem for compiling lapack's dgesvd with
> > sbcl-1.0.23(25) while 1.0.20 is ok. The issue exists for 32 bit linux
> > platform only. My mashine has 4Gb RAM.
>
> Thanks for the information. Can you try it w/ the SBCL command line
> option: --dynamic-space-size 1024 (or some other large number).
> You'll have to modify the maxima script (src/maxima) to put the
> option on the sbcl command line.
Did you mean /usr/bin/maxima?
I put that option there, i.e.,
elif [ "$MAXIMA_LISP" = "sbcl" ]; then
    exec "sbcl" --dynamic-space-size 2560 ....

> When I try it with that option,
> compiling dgesvd.lisp succeeds (and the rest of lapack too).
This works! I'm remembering that I already stepped to this rake somewhere 
half-year ago. While the given option indeed helps to compile lapack, the  
computations with large bigfloats matrices can be a problem still (the same 
issue -  not enough memory, heap exhausted, welome to ldb). Migrating to 64 
bit solved my problems that time.

>
> I notice there doesn't appear to be a way to pass command line options
> to the Lisp program. Extra arguments to the maxima script are passed
> on as user options, not Lisp options. I wonder if we should invent a way
> to pass options to Lisp.
This woud be nice!

best regards
Valery