Hello All!
I'm wondering if anyone managed to successfully compile (and run!)
MatLisp on FC5 (that's Fedora Core 5 for non-Linux folks :-) )?
I'm using stock CMU Common Lisp 19c Fedora Extras release 4.fc5.
MatLisp configured/compiled just fine (all I had to change was adding
(load "packages") in the beginning of start.lisp -- otherwise it
complained about unknown "F2CL-LIB" package), but when I try any
non-trivial matrix operation (such as m+, which, I think is the
simplest
one making an actual Fortran call) it gets into an infinite loop
(breakable with ^C though). When I run it under gdb it seems to be
getting SIGSEGVs when (unsuccessfully) trying to load a symbol in
dl_-something.
It worked fine for me on FC4. One of the obvious differences is that
FC4
used gcc 4.0, while FC5 uses 4.1. I've tried downgrading C compiler to
gcc32, upgrading Fortran compiler to gfortran (failed in even more
spectacular way!), nothing helps.
CMUCL docs mention that "There are problems running CMUCL on certain
kernels that are built with address-space randomization techniques,
such
as certain Fedora kernels. If you run into problems, please use a
vanilla kernel." -- I have 2.6.16-1.2111_FC5 and I *assume* that Fedora
folks would make their CMUCL rpm work with their stock kernel -- is it
too much to assume? :-) I did do echo 0 > /proc/sys/kernel/exec-shield
just
to be on the safe side, did not help...
Thanks in advance!
Paul Bunyk
--
Paul Bunyk <pbunyk_no_spam_dwavesys.com>
D-Wave: The Quantum Computing Company
You should probably try asking this on the matlisp mailing list on
gmane:
http://dir.gmane.org/gmane.lisp.matlisp.user
However, based on what you wrote, have you verified that you have the
proper LAPACK & BLAS libraries installed?
······@gmail.com wrote:
> Hello All!
>
> I'm wondering if anyone managed to successfully compile (and run!)
> MatLisp on FC5 (that's Fedora Core 5 for non-Linux folks :-) )?
...
> it gets into an infinite loop
> (breakable with ^C though). When I run it under gdb it seems to be
> getting SIGSEGVs when (unsuccessfully) trying to load a symbol in
> dl_-something.
·············@specastro.com wrote:
> You should probably try asking this on the matlisp mailing list on
> gmane: http://dir.gmane.org/gmane.lisp.matlisp.user
I did ask there -- and someone suggested asking on cmucl list, which
seems to be mostly spam these days... ;-( I thought I'd go to the
source
of all LISP knowledge. ;-)
> However, based on what you wrote, have you verified that you have the
> proper LAPACK & BLAS libraries installed?
MatLisp comes with its own copy of LAPACK and BLAS (and it claims it
can
work with system-optimized ATLAS libraries, but it did not really
manage to
successfully link against them). I tried both.
On the other hand, I started looking into SBCL in addition to CMUCL and
was pleasantly surprized that 1) unlike CMUCL it is available for
x86-64
(which by itself gave me a warm and fuzzy feeling about its
portability)
2) it is much more verbose with what it is doing "under the hood" and
3) MatLisp seems to compile and work just fine with it on both my
desktop
64-bit box and server 32-bit (talk about priority inversion! ;-) ).
So, for now I am happy and will recommend everyone to use SBCL with
MatLisp -- now time to actually make ATLAS libraries work.
Paul B.