Does that mean you've fixed the environment module, and now everything is working as expected / your program compiles properly with Open MPI v5.0.x? ________________________________ From: Miroslav Iliaš <miro.il...@gmail.com> Sent: Wednesday, May 28, 2025 4:22 PM To: Open MPI users <users@lists.open-mpi.org> Cc: Jeff Squyres (jsquyres) <jsquy...@cisco.com> Subject: Re: [OMPI users] openmpi5.0.7 with Intel2021 can not compile simple MPI program, error #6633
Dear Jeff, indeed, you are right, the module is not properly built. mpifort --showme ifort -I/cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/include -I/cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/lib -L/cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/lib -Wl,-rpath -Wl,/cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/lib -Wl,--enable-new-dtags -lmpi_usempif08 -lmpi_usempi_ignore_tkr -lmpi_mpifh -lmpi mpicc --showme icc -std=gnu11 -I/cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/include -pthread -L/cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/lib -Wl,-rpath -Wl,/cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/lib -Wl,--enable-new-dtags -lmpi mpicxx --showme icpc -I/cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/include -pthread -L/cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/lib -Wl,-rpath -Wl,/cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/lib -Wl,--enable-new-dtags -lmpi mpirun --version mpirun (Open MPI) 5.0.7 module -v display openmpi/v5.0.7_icc2021 ------------------------------------------------------------------- /cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/modulefiles/openmpi/v5.0.7_icc2021: module-whatis Modulefile for openmpi v5.0.7 with icc 2021 module load BASE/1.0 intel/oneapi setenv OMPI_ROOT /cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021 prepend-path PATH /cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/bin prepend-path C_INCLUDE_PATH /cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/include prepend-path CPLUS_INCLUDE_PATH /cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/include prepend-path LIBRARY_PATH /cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/lib prepend-path LD_LIBRARY_PATH /cvmfs/hybrilit.jinr.ru/sw/slc7_x86-64/openmpi/v5.0.7_icc2021/lib On Wednesday, 28 May 2025 at 21:50:20 UTC+3 Jeff Squyres (jsquyres) wrote: How did you build Open MPI? Did you specify "-i8" for the FCFLAGS, perchance? What is the output of mpifort --showme? Open MPI definitely defines the mpi module MPI_INIT interface to take an integer as its parameter: https://github.com/open-mpi/ompi/blob/v5.0.x/ompi/mpi/fortran/use-mpi-ignore-tkr/mpi-ignore-tkr-interfaces.h.in#L2284-L2286 (or, if the ifort 2021 compiler falls into the "we don't support ignore TKR" types of Fortran compilers, then https://github.com/open-mpi/ompi/blob/v5.0.x/ompi/mpi/fortran/use-mpi-tkr/mpi-f90-interfaces.h#L1109-L1111) Either way, somehow your program is not matching what Open MPI was built with. One possible way for this is if Open MPI was configured and built with something like -i8, which would make all default integers be 8 bytes instead of 4 bytes. Or perhaps mpifort is automatically adding something like -i8. IIRC / I'm pretty sure that neither of these things should have happened unless configure was told to build Open MPI this way, though. ________________________________ From: us...@lists.open-mpi.org <us...@lists.open-mpi.org> on behalf of Miroslav Iliaš <miro....@gmail.com> Sent: Wednesday, May 28, 2025 2:39 PM To: Open MPI users <us...@lists.open-mpi.org> Cc: Jeff Squyres (jsquyres) <jsqu...@cisco.com> Subject: Re: [OMPI users] openmpi5.0.7 with Intel2021 can not compile simple MPI program, error #6633 Hello, declaration of ierr does not help. OpenMPI 5.0.7 has some strange deklaration of "dummy argument", giving this error: pi_upd.F90(40): error #6633: The type of the actual argument differs from the type of the dummy argument. [IERR] pi_upd.F90(44): error #6633: The type of the actual argument differs from the type of the dummy argument. [MYID] pi_upd.F90(45): error #6633: The type of the actual argument differs from the type of the dummy argument. [NUMPROCS] Miro On Wednesday, 28 May 2025 at 20:13:00 UTC+3 Jeff Squyres (jsquyres) wrote: I am not a Fortran expert, but you did not declare ierr to be a specific type (e.g., integer). If you declare it properly, does the problem go away? ________________________________ From: us...@lists.open-mpi.org <us...@lists.open-mpi.org> on behalf of Miroslav Iliaš <miro....@gmail.com> Sent: Wednesday, May 28, 2025 11:22 AM To: Open MPI users <us...@lists.open-mpi.org> Subject: [OMPI users] openmpi5.0.7 with Intel2021 can not compile simple MPI program, error #6633 Dear OpenMPI community, we have new openmpi5.0.7 library compiled with Intel2021 compilers. Hower, compiling simple program fails with errors #6633 . Here it is: https://gitlab.com/miro.ilias/mi_programming/-/blob/master/mpi_tests/pi_parallel/readme_govorun.rst and the whole directory : https://gitlab.com/miro.ilias/mi_programming/-/tree/master/mpi_tests/pi_parallel . Best, Miro To unsubscribe from this group and stop receiving emails from it, send an email to users+un...@lists.open-mpi.org. To unsubscribe from this group and stop receiving emails from it, send an email to users+un...@lists.open-mpi.org. To unsubscribe from this group and stop receiving emails from it, send an email to users+unsubscr...@lists.open-mpi.org.
