[petsc-dev] Fwd: [petsc-maint] help with PETSc compilation

Barry Smith bsmith at mcs.anl.gov
Fri Aug 9 13:24:59 CDT 2013


   petsc-dev folks,

   If PETSc is configured to generate a shared library then ./configure should check each externally needed library and make sure they can be handled with a shared library. This shouldn't be something discovered at PETSc library compilation time.

   Barry


Begin forwarded message:

> From: Barry Smith <bsmith at mcs.anl.gov>
> Subject: Re: [petsc-maint] help with PETSc compilation
> Date: August 9, 2013 1:23:14 PM CDT
> To: Brendan C Lyons <bclyons at princeton.edu>
> Cc: petsc-maint at mcs.anl.gov
> 
> 
>   Brendan,
> 
>   The scalapack you are linking against is not compiled to go into a shared library. 
> 
> /usr/bin/ld: /usr/pppl/pathscale/4.0.X-pkgs/openmpi-1.6-pkgs/scalapack-2.0.2/lib/libscalapack.a(blacs_map_.o): relocation R_X86_64_32S against `ompi_mpi_comm_null' can not be used when making a shared object; recompile with -fPIC
> 
>  You can either 
> 
> 1) not build PETSc shared libraries by adding the configure option --with-shared-libraries=0   With this option linking executables will be slower. 
> 
> or
> 
> 2) use --download-scalapack instead of --with-scalapack-dir
> 
>   Barry
> 
> 
> On Aug 9, 2013, at 12:58 PM, Brendan C Lyons <bclyons at princeton.edu> wrote:
> 
>> Dear Sir or Madam,
>> 
>> I'm trying to compile a debug version of PETSc to track down an error with SuperLU_DIST using multiple processors, but I'm running into an error.  The error message said to send you the attached files.  If you have any advice, I would appreciate it.
>> 
>> Thank you,
>> 
>> ~Brendan Lyons 
>> Graduate Student
>> Princeton Plasma Physics Laboratory
>> <configure.log><make.log>
> 




More information about the petsc-dev mailing list