[petsc-users] Full blas-lapack on Windows
Barry Smith
bsmith at mcs.anl.gov
Tue Feb 10 10:34:30 CST 2015
The reason it (superlu_dist) doesn't build is that it uses C99 compiler features while Microsoft C compiler only supports C89. The line of code
int nlsupers = nsupers/Pc;
is not valid c89 since it declares a new variable after other code.
You can try to fix all the C99 uses in that file (by moving the declarations of all the variables to the top of the routine) and run the configure again. I don't know how much Sheri used C99 so it may mean changing many things.
Barry
> On Feb 10, 2015, at 8:44 AM, Fabien RAPHEL <fabien.raphel at etu.univ-nantes.fr> wrote:
>
> Thanks,
> I had an error when I used the --download-f2cblaslapack command but now it
> works.
> The configuration and compilation work well with the superlu library, but
> not with superlu_dist.
> I don't think it's a version compatibility problem.
>
> I have some errors with the pdgstrf.c file during the configuration.
> Have I to change the version of the library? (I tried with the
> SuperLU_DIST_2.5 version, but I still have the same error).
>
>
> Thanks,
>
> Fabien
>
>
>
>
>> I think we had this conversation before.
>>
>> --download-f2cblaslapack will give you a full blas/lapack.
>>
>> And you can't use MUMPS without a fortran compiler [as far as I know]
>>
>> You should be able to use superlu_dist
>>
>> Satish
>>
>> On Mon, 9 Feb 2015, Fabien RAPHEL wrote:
>>
>>> Hello,
>>>
>>> I can configure, compile and use PETSc on Windows with Visual Studio
>>> 2008
>>> (in serial and parallel).
>>> But I would like to use a LU factorization in parallel (for example,
>>> using
>>> Superlu_dist or MUMPS library).
>>> I don't have FORTRAN compiler on my machine, so I can't compile the full
>>> version of BLAS/LAPACK (with the slamch() routine for example).
>>>
>>> I found a precompiled version of the full libraries (I can run an sample
>>> in VS2008).
>>> But I have a PETSc configure error: "--with-blas-lapack-lib..... cannot
>>> be
>>> used" and the configure.log returns a lot of undefined references
>>> whereas
>>> I set the libraries and the include file during configuration.
>>> Do I forgot something in the command line? Or is the error comes from
>>> the
>>> library?
>>>
>>> Or, is there a C/C++ library who can do that?
>>>
>>> Thanks in advance,
>>>
>>> Fabien
>>>
>>
>>
> <configure.log>
More information about the petsc-users
mailing list