Thank you Bernard.
This has been fixed.
A couple weeks ago I synchronized the BLAS netlib page with the
reference BLAS include in LAPACK, adding also a Makfile.
Since 3.1.1 LAPACK includes the "whole" reference blas implementation.
dsdot was part of these "new" routines. It looks like I forgot to add
that one in the Makefile.
Again thank you for your contribution
On Apr 4, 2007, at 2:56 PM, Bernard Leak wrote:
Dear LAPACK (what foresighted parents you must have had!),
just been swearing over an ATLAS test-run which threw up the
unresolved symbol 'dsdot_' when using the netlib implementation
of BLAS as the reference implementation. Much chasing round the
houses later, it turns out that although dsdot.f is in the current
(as of yesterday, 2007-04-03) version of the BLAS source distribution
from netlib, it is not actually used - the only FORTRAN source-file
in the tarball of which this is true.
Presumably this is an oversight? The file time-stamps are recent,
which may explain why I haven't found a similar complaint for
you or ATLAS. All seems to work both hunkily and dorily if I simply
apply the attached patch, which excitingly inserts dsdot.o amongst
the 'd' sources.
I've bzipped the patch to minimise mangling in transmission: it uses
UNIX-style (LF) line-breaks. For the foreseeable future it should
be available at http://www.brenda-arkle.demon.co.uk/patches/blas/ .
Lapack mailing list