Re: Linking problems with mutex and zsyrk_

[Daniel J Sebald]

John W. Eaton wrote:
> On 15-Dec-2008, Daniel J Sebald wrote:
>
> | The list of functions missing during linking is growing.  Now there are eight 
> LAPACK routines (zggev, dsygv, etc.) associated with a file EIG.cc that 
> liboctave.so requires.
>
> Yes, we should fix this, but does your system not have an accelerated
> BLAS+LAPACK library like ATLAS available?

It had the BLAS but not LAPACK library.  After Jarkko's change to check for 
these libraries, the result is still a little arcane:

make[3]: Leaving directory `/usr/local/src/octave-hg/octave/libcruft/fftpack'
make -C lapack all
make[3]: Entering directory `/usr/local/src/octave-hg/octave/libcruft/lapack'
Makefile:30: *** commands commence before first target.  Stop.

but at least there is a hint there that lapack might be missing.  OK, so after 
installing the LAPACK library I see

  BLAS libraries:       -llapack -lblas

where there were none before.  Nice.

All compiled well except for the pthread_mutex_attr routines.  Even though the 
pthread library is present on the machine, Octave compile isn't finding it.  I 
add

 ./configure LIBS=-lpthread

and the source code compiles all the way through.  (Some strange things in 
M-scripts, but that's not so critical.)


> Perhaps we should add a warning at the end of the configure step
> saying something like
>
>   WARNING: No accelerated BLAS+LAPACK library detected.  Using Fortran
>   reference version instead.  For better performance, install ATLAS
>   and re-run configure.

Yes, what you are suggesting is that Octave remain compilable without the LAPACK.  That's 
fine/good.  But if that isn't the case, then a nice error message like "You need 
LAPACK library version A.Z or greater." either during config or build.  (Config is 
preferred because no lengthy compilation is required to find out.)

Dan