Re: [DuMuX] Turn off MPI support for the dune-uggrid module
Hi Timo, thanks for your advice! I think the server is running Ubuntu 14.04, so then the system is approximately 4 years old I guess? By the way I am using DUNE modules dune-common, dune-istl, dune-grid, dune-uggrid, dune-geometry, and dune-localfunctions all version 2.5.1 together with DuMuX version 2.12.0, and the master branch of dumux-devel; all cloned by git from https://github.com/dune-project/ and https://git.iws.uni-stuttgart.de/dumux-repositories/ except for dune-uggrid, which was cloned from https://github.com/dune-mirrors/dune-uggrid.git I create and issue at the dune-uggrid repository: https://gitlab.dune-project.org/staging/dune-uggrid/issues/39 Best regards, Håkon ___ Dumux mailing list Dumux@listserv.uni-stuttgart.de https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
Re: [DuMuX] Turn off MPI support for the dune-uggrid module
Hi Håkon, I think you are going in the right direction of trying to compile on the cluster, which should give you less headache then linker issues. Out of interest, is the server running macOS? I never heard of such difficulties installing a C++ 14 conforming compiler with the corresponding standard library implementation. Often compiler packages include the right standard library implementation. The UGGrid error might be related to a missing BLAS library. However, UG should handle this I think. Would you post your problem on the dune mailing list? Or open an issue on the dune-uggrid repository with the error message when compiling "dune-uggrid/np/algebra/blasm.cc". You can also try dune-alugrid if you just want an unstructured grid implementation. Best wishes Timo On 07.03.2018 11:48, Håkon Hægland wrote: > Hi Timo, > > here follows a short update on my previous post. > > By investigating the the cmake module CheckCXXFeatures.cmake, it was > found > that libc++abi was missing on the server. After installing the package > libc++abi-dev > on the server, dune-control actually made some progress in the > build-process. > > It was able to build the dune-common module, then it started on > dune-uggrid: > > [...] > [ 83%] Building CXX object np/algebra/CMakeFiles/algebra2.dir/blasm.cc.o > /scratch/fcihh/dune/dune-uggrid/np/algebra/blasm.cc:408:19: warning: > cast from 'char *' to 'DOUBLE *' (aka 'double *') increases required > alignment from 1 to 8 [-Wcast-align] > M_ptr = (DOUBLE *) ((char *) mat + M_start_off_D); > ^ > [...] > /scratch/fcihh/dune/dune-uggrid/np/algebra/blasm.cc:1237:36: error: > unexpected type name 'MATRIX': expected expression > M_start_off[type] = offsetof(MATRIX, value) > > > And there follows several errors regarding unexpected type name for > MATRIX or VECTOR.. > > Then dune control aborts with the message: > > 61 warnings and 14 errors generated. > make[2]: *** [np/algebra/CMakeFiles/algebra2.dir/blasm.cc.o] Error 1 > make[1]: *** [np/algebra/CMakeFiles/algebra2.dir/all] Error 2 > make: *** [all] Error 2 > --- Failed to build dune-uggrid --- > Terminating dunecontrol due to previous errors! > > > Best regards, > Håkon Hægland > > ___ > Dumux mailing list > Dumux@listserv.uni-stuttgart.de > https://listserv.uni-stuttgart.de/mailman/listinfo/dumux ___ Dumux mailing list Dumux@listserv.uni-stuttgart.de https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
Re: [DuMuX] Turn off MPI support for the dune-uggrid module
Hi Timo, here follows a short update on my previous post. By investigating the the cmake module CheckCXXFeatures.cmake, it was found that libc++abi was missing on the server. After installing the package libc++abi-dev on the server, dune-control actually made some progress in the build-process. It was able to build the dune-common module, then it started on dune-uggrid: [...] [ 83%] Building CXX object np/algebra/CMakeFiles/algebra2.dir/blasm.cc.o /scratch/fcihh/dune/dune-uggrid/np/algebra/blasm.cc:408:19: warning: cast from 'char *' to 'DOUBLE *' (aka 'double *') increases required alignment from 1 to 8 [-Wcast-align] M_ptr = (DOUBLE *) ((char *) mat + M_start_off_D); ^ [...] /scratch/fcihh/dune/dune-uggrid/np/algebra/blasm.cc:1237:36: error: unexpected type name 'MATRIX': expected expression M_start_off[type] = offsetof(MATRIX, value) And there follows several errors regarding unexpected type name for MATRIX or VECTOR.. Then dune control aborts with the message: 61 warnings and 14 errors generated. make[2]: *** [np/algebra/CMakeFiles/algebra2.dir/blasm.cc.o] Error 1 make[1]: *** [np/algebra/CMakeFiles/algebra2.dir/all] Error 2 make: *** [all] Error 2 --- Failed to build dune-uggrid --- Terminating dunecontrol due to previous errors! Best regards, Håkon Hægland ___ Dumux mailing list Dumux@listserv.uni-stuttgart.de https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
Re: [DuMuX] Turn off MPI support for the dune-uggrid module
Hello Timo! Thanks for the support. The problem with compiling DuMuX on the computational server seems to be that it is running an old version of Ubuntu (and an old version of gcc). When I try the following command on the computation server: $ ./dune-common/bin/dunecontrol --opts=dumux/debug.opts all [...] CMake Error at cmake/modules/CheckCXXFeatures.cmake:117 (message): dune-common requires compiler support for C++14, but your compiler only supports C++11. It aborts due to too old version of the compiler. I had the server admin install clang-3.9 and libc++-dev on the server and then tried again using clang instead of gcc. I used these flags with dune-control: CMAKE_FLAGS="\ -DCMAKE_CXX_COMPILER=clang++-3.9 \ -DCMAKE_C_COMPILER=clang-3.9 \ -DCMAKE_CXX_FLAGS=\"$GXX_WARNING_OPTS $GXX_OPTS -stdlib=libc++\" \ " However, still the test for C++14 fails (even if the clang compiler itself supports C++14) in dune-control. So now I tried instead to compile the binary on my laptop and then copy the binary and some shared libraries onto the server. This worked well for structured grid (YaspGrid) but for UGGrid I got the following error message when I tried to run the executable: It looks like opal_init failed for some reason; your parallel process is likely to abort. There are many reasons that a parallel process can fail during opal_init; some of which are due to configuration or environment problems. This failure appears to be an internal failure; here's some additional information (which may only be relevant to an Open MPI developer): opal_shmem_base_select failed --> Returned value -1 instead of OPAL_SUCCESS [...] *** An error occurred in MPI_Init *** on a NULL communicator *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort, ***and potentially your MPI job) [cipr-cmg-sim:67060] Local abort before MPI_INIT completed completed successfully, but am not able to aggregate error messages, and not able to guarantee that all other processes were killed! So this is the reason why I am trying to compile uggrid without MPI support. Best regards, Håkon Hægland ___ Dumux mailing list Dumux@listserv.uni-stuttgart.de https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
Re: [DuMuX] Turn off MPI support for the dune-uggrid module
Hi Håkon, if MPI is found on your system then MPI support will be enabled per default, no matter if you use YaspGrid or UGGrid. The standard option is also that all libraries that are found will be linked to the executable which is why you can see the libs when you run "ldd executable" in there. The problem is also most likely not only MPI but other linked libraries too. As you link the libraries of your laptop when compiling, you will usually not be able use an executable compiled on your laptop and run it on another machine. However, you seem to be able to copy your executable to your cluster. So can't you copy the Dune/Dumux source files there and compile it on the cluster? AFAIK, this is the standard way to do it. If you are using a workflow where you install the dune modules, you can also install it locally by setting the CMAKE_INSTALL_PREFIX variable. If you are using the Dune Debian packages, you either have to convince your administrator to install them on the cluster, or switch to an installation from source (works just like you configured/compiled Dumux in that case). I'll move the discussion to the mailing list. Please answer there if possible so that others can profit from the discussion. Best wishes Timo -- ___ Timo Koch phone: +49 711 685 64676 IWS, Universität Stuttgart fax: +49 711 685 60430 Pfaffenwaldring 61email: timo.k...@iws.uni-stuttgart.de D-70569 Stuttgarturl: www.hydrosys.uni-stuttgart.de ___ ___ Dumux mailing list Dumux@listserv.uni-stuttgart.de https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
