[Scilab-Dev] Building scilab-master-1286272063 on Fedora 13

[Dean S. Messing]

On Mon, 18 Oct 2010 09:53:18 +0200, Clément David wrote:

> Just to record that this problem may be linked to #2741 [1] `The "atlas"
> lib is not automatically detected; Scilab uses the internal BLAS
> library.'.
> 
> 
> [1] http://bugzilla.scilab.org/show_bug.cgi?id=2741

Thanks Clément.

That bug report in the comments says

  "This bug does not block Scilab compilation and usage.  Atlas is used in
  place of user installed or built in BLAS ONLY for better numerical speed
  performances"

But, in fact it _did_ block my compilation of the Master Branch for
5.3.0.  I had included these switches on my "configure" line:

PATH=${PATH}:/usr/share/pvm3/lib ./configure\
                 --with-blas-library=/usr/lib64/atlas\
                 --with-lapack-library=/usr/lib64/atlas\
                 --with-umfpack-library=/usr/lib64/atlas 

yet the compilation died  at the compilation (the arnoldi module).


By the way, if I leave off any of the above "--with" switch, then the
configure itself will die somewhere.  For example, if I leave off the
last switch it dies like this:

checking for sgemm_... no
checking for ATL_xerbla in -latlas... yes
checking for sgemm_ in -lf77blas... yes
checking for cblas_dgemm in -lcblas... yes
Atlas found
checking if LAPACK is available... 
checking for cheev_... no
checking for cheev_ in -llapack... yes
Library -llapack found
checking suitesparse/umfpack.h usability... yes
checking suitesparse/umfpack.h presence... yes
checking for suitesparse/umfpack.h... yes
checking for amd_info in -lamd... no
configure: error: libamd: Library missing (Cannot find symbol amd_info). Check if libamd (sparse matrix minimum degree ordering) is installed and if the version is correct

>From the config.log, the problem is really

/usr/bin/ld: cannot find -lcblas
collect2: ld returned 1 exit status

So I added the bogus --with-umfpack to force -L/usr/lib64/atlas to be
included on the conftest.c compile line.  I know it's a hack but I don't
know a better way.

Anyway, there appear to be bugs other than #2741 involved here.
But I'm a novice at this.

Dean