Conflicts with SuperLU-6.0.0
sagitter opened this issue · 18 comments
Hi all.
I'm rebuilding freefem-4.12 against SuperLU-6.0.0 but i get these errors:
In file included from SuperLu.cpp:32:
/usr/include/SuperLU/slu_zdefs.h: At global scope:
/usr/include/SuperLU/slu_zdefs.h:239:17: error: conflicting declaration of C function 'double* doubleMalloc(int)'
239 | extern double *doubleMalloc(int);
| ^~~~~~~~~~~~
In file included from SuperLu.cpp:29:
/usr/include/SuperLU/slu_ddefs.h:236:17: note: previous declaration 'double* doubleMalloc(size_t)'
236 | extern double *doubleMalloc(size_t);
| ^~~~~~~~~~~~
/usr/include/SuperLU/slu_zdefs.h:240:17: error: conflicting declaration of C function 'double* doubleCalloc(int)'
240 | extern double *doubleCalloc(int);
| ^~~~~~~~~~~~
/usr/include/SuperLU/slu_ddefs.h:237:17: note: previous declaration 'double* doubleCalloc(size_t)'
237 | extern double *doubleCalloc(size_t);
| ^~~~~~~~~~~~
tetgen.cpp: In function 'Fem2D::Mesh3* Convexhull_3Dpoints(char*, const int&, const double*, const double*, const double*, const int&, const int&)':
tetgen.cpp:872:13: warning: variable 'jtet' set but not used [-Wunused-but-set-variable]
872 | int itet, jtet;
| ^~~~
tetgen.cpp: In function 'Fem2D::Mesh3* RemplissageSurf3D_tetgen(char*, const Fem2D::Mesh3&, const int&)':
tetgen.cpp:939:13: warning: variable 'jtet' set but not used [-Wunused-but-set-variable]
939 | int itet, jtet;
| ^~~~
tetgen.cpp: In function 'Fem2D::Mesh3* RemplissageSurf3D_tetgen_new(char*, const Fem2D::MeshS&, const int&, const int&, const double*, const int&, const double*, const int&, const double*)':
tetgen.cpp:1037:13: warning: variable 'jtet' set but not used [-Wunused-but-set-variable]
1037 | int itet, jtet;
| ^~~~
tetgen.cpp: In function 'Fem2D::Mesh3* RemplissageSurf3D_tetgen_new(char*, const Fem2D::Mesh3&, const int&, const int&, const double*, const int&, const double*, const int&, const double*, const int&, const double*, const int&, const double*)':
tetgen.cpp:1163:13: warning: variable 'jtet' set but not used [-Wunused-but-set-variable]
1163 | int itet, jtet;
| ^~~~
tetgen.cpp: In function 'Fem2D::Mesh3* ReconstructionRefine_tetgen(char*, const Fem2D::Mesh3&, const int&, const double*, const int&, const double*, const int&, const double*, const double*)':
tetgen.cpp:1309:9: warning: unused variable 'lenswitch' [-Wunused-variable]
1309 | int lenswitch;
| ^~~~~~~~~
tetgen.cpp:1346:13: warning: variable 'jtet' set but not used [-Wunused-but-set-variable]
1346 | int itet, jtet;
| ^~~~
tetgen.cpp: In function 'Fem2D::Mesh3* ReconstructionRefine_tetgen(char*, const Fem2D::Mesh3&, const int&, const double*, const int&, const double*, const int&, const double*, const double*, const int&, const double*)':
tetgen.cpp:1475:9: warning: unused variable 'lenswitch' [-Wunused-variable]
1475 | int lenswitch;
| ^~~~~~~~~
tetgen.cpp:1512:13: warning: variable 'jtet' set but not used [-Wunused-but-set-variable]
1512 | int itet, jtet;
| ^~~~
tetgen.cpp: In member function 'virtual AnyType Remplissage_Op::operator()(Stack) const':
tetgen.cpp:1792:7: warning: unused variable 'sizeofmetric' [-Wunused-variable]
1792 | int sizeofmetric = metric.N( ) / Th.nv;
| ^~~~~~~~~~~~
tetgen.cpp:1899:7: warning: unused variable 'nbinside' [-Wunused-variable]
1899 | int nbinside = InsidePoint.N( ) / 3;
The full build log: here
You should reconfigure with --disable-superlu.
I skip the problem in this way, cant it be resolved yet?
You'll have to ask SuperLU developers why we can't include both headers while we previously could.
Looks like an oversight to me. Can you give xiaoyeli/superlu#100 a try and confirm, whether this fixes your issue?
I just tried to compile FreeFem, and ran into the same issue. For me, the configure option --disable-superlu fixed the issue
@aszaboa Can you please double-check that you are using SuperLU Git master and not the 6.0 release? Further, can you share the compiler output?
@gruenich I tried to configure PETSc with the option --download-superlu --download-superlu-commit=4e88a0b, but it did not work. I am sharing the config.log file.
configure.log
Do not waste your time, just disable SuperLU. For some weeks now, it looks like every merged PR on either SuperLU or SuperLU_DIST repository is breaking something, so frustrating.
Sorry @aszaboa , for systems with broken or missing Doxygen, I introduced a regression. Your run was plagued by my mistaken. Can you please retry with commit d812a001a6d8e16906fe7376c1688d605391c368 which includes a fix?
@prj- Sorry for the inconvenience my contributions might cause. I see some room for improving SuperLU, this includes a better system to detect regressions. I hope I will find to address this.
@gruenich I tried it and it does not work for me. I think you can sort this issue out on your own. To compile PETSc/FreeFem, I use the following tutorial: https://joliv.et/FreeFem-tutorial/main.html#[197,%22XYZ%22,28.35,272.13,null] (the compilation is in the appendix)
I need your assistance. I cloned PETSc, configured it (./configure --dowload=superlu --download-superlu-commit=d812a001a6d8e16906fe7376c1688d605391c368 --with-mpi=0), make all, make check. It just worked. I need your configure.log file.
These are the exact commands that I use to compile FreeFem+PETSc. These should reproduce the issue.
----------------------------
export FF_DIR=${PWD}/FreeFem-sources
export PETSC_DIR=${PWD}/petsc
export PETSC_ARCH=arch-FreeFem
export PETSC_VAR=${PETSC_DIR}/${PETSC_ARCH}
git clone https://github.com/FreeFem/FreeFem-sources.git
git clone https://gitlab.com/petsc/petsc
cd ${PETSC_DIR} && ./configure --download-mumps --download-parmetis --download-metis --download-hypre --download-superlu --download-slepc --download-hpddm --download-ptscotch --download-suitesparse --download-scalapack --download-tetgen --with-fortran-bindings=no --with-scalar-type=real --with-debugging=no
make -j5
export PETSC_ARCH=arch-FreeFem-complex
./configure --with-mumps-dir=arch-FreeFem --with-parmetis-dir=arch-FreeFem --with-metis-dir=arch-FreeFem --with-superlu-dir=arch-FreeFem --download-slepc --download-hpddm --with-ptscotch-dir=arch-FreeFem --with-suitesparse-dir=arch-FreeFem --with-scalapack-dir=arch-FreeFem --with-tetgen-dir=arch-FreeFem --with-fortran-bindings=no --with-scalar-type=complex --with-debugging=no
make -j5
cd ${FF_DIR}
git checkout develop
autoreconf -i
./configure --without-hdf5 --disable-superlu --enable-download_metis --enable-download_parmetis --enable-download_mmg --enable-download_mmg3d --enable-download_parmmg --enable-download_mshmet --enable-download_nlopt --enable-optim --with-petsc=${PETSC_VAR}/lib --with-petsc_complex=${PETSC_VAR}-complex/lib
make -j5
----------------------------
I have a very shallow understanding but I think the difference may be related to --with-mpi=0. My guess is that the overwhelming majority of PETSc users uses it with mpi.
You are not downloading the right version of SuperLU. You need to have -download-superlu-commit=d812a001a6d8e16906fe7376c1688d605391c368 in you PETSc configure call. Not sure how to achieve this for FreeFem.
I configured PETSc with my locally installed OpenMPI. It passes make all (I had to remove two configure checks), so I cannt reproduce the problem.
I think I was not precise enough. The error occurs in the FreeFem configuration - there, the --disable-superlu ˛ flag circumvents the issue. So actually the commands above may reproduce the issue if you skip the --disable-superlu ˛ flag. Anyway, as I do not use superlu, I will just continue excluding it from the compilation.
Probably I did not make clear what I expected from you. I understand that it did not work with SuperLU and disabling SuperLU made it working for you. I think I fixed the issue. I was hoping that you can try your setup with the latest SuperLU master and confirm that this fixes your SuperLU issue, too.
xiaoyeli/superlu#100 fixes this issue for me
I will close this issue, it's not related to FreeFEM, hopefully PETSc 3.20.0 (scheduled this September) will ship a functioning SuperLU version...