[petsc-users] Warning while compiling Fortran with PETSc

Bojan Niceno bojan.niceno.scientist at gmail.com
Mon Feb 14 01:19:46 CST 2022 

Dear both,

It was the compiler options for integer lengths after all, thanks,  Now I
corrected it all in my code, all integers have explicitly defined lengths,
and I am using the MPI_F08 module instead of obsolete mpi.f.  It was a
daunting task (> 800 files, > 64000 lines of code), but I am happy with the
outcome.  Now I can continue with PETSc :-)

    Cheers

    Bojan




On Fri, Feb 11, 2022 at 7:26 AM Bojan Niceno <
bojan.niceno.scientist at gmail.com> wrote:

> What does seem to work is:
>
> use mpi_f08
>
> and using:
>
>   call Mpi_Init_08(error)
>
>   call Mpi_Comm_Size_08(MPI_COMM_WORLD, n_proc, error)·
>
>   call Mpi_Comm_Rank_08(MPI_COMM_WORLD, this_proc, error)
>
>
> That is, function calls with extension _08
>
>
>
>
>
> On Fri, Feb 11, 2022 at 6:51 AM Bojan Niceno <
> bojan.niceno.scientist at gmail.com> wrote:
>
>> Thanks for the example.
>>
>> I do use -i8 -r8 (I know it is a bad practice and am planning to get rid
>> of it in mid terms), and I did suspect that, but when I tried to compile
>> without those options, the warning remained.
>>
>> When I switched from "include mpif.h" to:
>>
>> use mpi,
>>
>> or even:
>>
>> #include <petsc/finclude/petscsys.h>
>>
>>   use petscmpi
>>
>>   use petscsys
>>
>>
>> I get the following type of messages:
>>
>> Error: There is no specific subroutine for the generic ‘mpi_barrier’ at
>> (1)
>> Comm_Mod/Parallel/Start.f90:11:22:
>>
>> Error: There is no specific subroutine for the generic ‘mpi_init’ at (1)
>> Comm_Mod/Parallel/Start.f90:14:52:
>>
>> Error: There is no specific subroutine for the generic ‘mpi_comm_size’ at
>> (1)
>> Comm_Mod/Parallel/Start.f90:17:54:
>>
>> I was googling for a solution, but StackOverflow seems to be down for
>> maintenance at the moment :-(
>>
>> I did manage to find that "include mpif.h" is obsolete, which I didn't
>> know before :-)
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> On Fri, Feb 11, 2022 at 5:29 AM Satish Balay <balay at mcs.anl.gov> wrote:
>>
>>> 1. you can call MPI_Init() before calling PetscInitialize() For
>>> example - check src/sys/tutorials/ex4f90.F90
>>>
>>> 2. Are you using -i8 -r8 type flags when compiling your code? That
>>> might case issues when using mpif.h. Perhaps you can switch from
>>> "include 'mpif.h'" to "use mpi" - in your module file - and see if
>>> that helps.
>>>
>>> Satish
>>>
>>> On Fri, 11 Feb 2022, Bojan Niceno wrote:
>>>
>>> > Dear both,
>>> >
>>> > Allow me to update you on the issue.  I tried to re-compile PETSc with
>>> > different configuration options as Satish suggested, and went further
>>> on by
>>> > specifying exact location of OpenMPI libraries and include files to the
>>> > ones installed by PETSc (for those configurations for which I used
>>> > "--download-openmpi=1") and the original problem, the warning Named
>>> COMMON
>>> > block ‘mpi_fortran_bottom’ at (1) shall be of the same size as
>>> elsewhere (4
>>> > vs 8 bytes), prevailed.
>>> >
>>> > In desperation, I completely removed OpenMPI from my workstation to
>>> make
>>> > sure that only those which are downloaded with PETSc are used, yet the
>>> > warning was still there.  (That resolved the Invalid MIT-MAGIC-COOKIE-1
>>> > warning at least)
>>> >
>>> > Now I am wondering if the problem originates from the fact that I
>>> already
>>> > have all the necessary MPI routines developed in Fortran?  All calls,
>>> > including the basic MPI_Init, MPI_Comm_Size and MPI_Comm_Rank, are done
>>> > from Fortran.  I actually have a module called Comm_Mod which does all
>>> > MPI-related calls, and this module contains line include 'mpif.h'.
>>> That
>>> > include statement does take the file from PETSc installation as no
>>> other
>>> > MPI installation is left on my system, but still it somehow seems to
>>> be the
>>> > origin of the warning on common blocks I observe.  Now I am wondering
>>> if
>>> > the include 'mpif.h' from Fortran somehow collides with the option
>>> include
>>> > ${PETSC_DIR}/lib/petsc/conf/variables I put in my makefile in order to
>>> > compile with PETSc.
>>> >
>>> > I am really not sure if it is possible to have main program and all MPI
>>> > initialization done from Fortran (as I have now) and then plug PETSc
>>> on top
>>> > of it?  Should that be possible?
>>> >
>>> >     Kind regards,
>>> >
>>> >     Bojan
>>> >
>>> > P.S. The sequential version works fine, I can compile without warning
>>> and
>>> > can call PETSc solvers from Fortran without a glitch.
>>> >
>>> > On Thu, Feb 10, 2022 at 5:08 PM Bojan Niceno <
>>> > bojan.niceno.scientist at gmail.com> wrote:
>>> >
>>> > > Dear Satish,
>>> > >
>>> > > Thanks for the advice.  I will try in a few hours because it is
>>> almost
>>> > > dinner time with me (I am in Europe) and I am supposed to go out
>>> with a
>>> > > friend this evening.
>>> > >
>>> > > Will let you know.  Thanks for help, I highly appreciate it.
>>> > >
>>> > >
>>> > >     Kind regards,
>>> > >
>>> > >     Bojan
>>> > >
>>> > >
>>> > > On Thu, Feb 10, 2022 at 5:06 PM Satish Balay <balay at mcs.anl.gov>
>>> wrote:
>>> > >
>>> > >> Hm - this is strange.
>>> > >>
>>> > >> Do you have 'xauth' installed?
>>> > >>
>>> > >> I would make sure xauth is installed, delete ~/.Xauthority - and
>>> reboot
>>> > >> [or restart the X server]
>>> > >>
>>> > >> Yeah - it might not work - but perhaps worth a try..
>>> > >>
>>> > >> Or perhaps its not X11 related..
>>> > >>
>>> > >> I would also try 'strace' on an application that is producing this
>>> > >> message - to see if I can narrow down further..
>>> > >>
>>> > >> Do you get this message with both (runs)?:
>>> > >>
>>> > >> cd src/ksp/ksp/tutorials
>>> > >> make ex2
>>> > >> mpiexec -n 1 ./ex2
>>> > >> ./ex2
>>> > >>
>>> > >> Satish
>>> > >>
>>> > >> On Thu, 10 Feb 2022, Bojan Niceno wrote:
>>> > >>
>>> > >> > Dear both,
>>> > >> >
>>> > >> > I work on an ASUS ROG laptop and don't use any NFS.  Everything
>>> is on
>>> > >> one
>>> > >> > computer, one disk.  That is why I couldn't resolve the Invalid
>>> Magic
>>> > >> > Cookie, because all the advice I've found about it concerns the
>>> remote
>>> > >> > access/display.  It is not an issue for me.  My laptop has an
>>> Nvidia
>>> > >> > GeForce RTX graphical card, maybe Ubuntu drivers are simply not
>>> able to
>>> > >> > cope with it.  I am out of ideas, really.
>>> > >> >
>>> > >> >
>>> > >> >     Cheers,
>>> > >> >
>>> > >> >     Bojan
>>> > >> >
>>> > >> > On Thu, Feb 10, 2022 at 4:53 PM Satish Balay <balay at mcs.anl.gov>
>>> wrote:
>>> > >> >
>>> > >> > > Do the compute nodes and frontend share the same NFS?
>>> > >> > >
>>> > >> > > I would try the following [to see if they work):
>>> > >> > >
>>> > >> > > - delete ~/.Xauthority [first check with 'xauth list')
>>> > >> > > - setup ssh to not use X - i.e add the following to
>>> ~/.ssh/config
>>> > >> > >
>>> > >> > > ForwardX11 no
>>> > >> > > ForwardX11Trusted no
>>> > >> > >
>>> > >> > > [this can be tailored to apply only to your specific compute
>>> nodes -
>>> > >> if
>>> > >> > > needed]
>>> > >> > >
>>> > >> > > Satish
>>> > >> > >
>>> > >> > > On Thu, 10 Feb 2022, Matthew Knepley wrote:
>>> > >> > >
>>> > >> > > > On Thu, Feb 10, 2022 at 10:40 AM Bojan Niceno <
>>> > >> > > > bojan.niceno.scientist at gmail.com> wrote:
>>> > >> > > >
>>> > >> > > > > Thanks a lot, now I feel much better.
>>> > >> > > > >
>>> > >> > > > > By the way, I can't get around the invalid magic cookie.
>>> It is
>>> > >> > > occurring
>>> > >> > > > > ever since I installed the OS (Ubuntu 20.04) so I
>>> eventually gave
>>> > >> up
>>> > >> > > and
>>> > >> > > > > decided to live with it :-D
>>> > >> > > > >
>>> > >> > > >
>>> > >> > > >
>>> > >> > >
>>> > >>
>>> https://unix.stackexchange.com/questions/199891/invalid-mit-magic-cookie-1-key-when-trying-to-run-program-remotely
>>> > >> > > >
>>> > >> > > >   Thanks,
>>> > >> > > >
>>> > >> > > >     Matt
>>> > >> > > >
>>> > >> > > >
>>> > >> > > > >     Cheers,
>>> > >> > > > >
>>> > >> > > > >     Bojan
>>> > >> > > > >
>>> > >> > > > > On Thu, Feb 10, 2022 at 4:37 PM Matthew Knepley <
>>> > >> knepley at gmail.com>
>>> > >> > > wrote:
>>> > >> > > > >
>>> > >> > > > >> On Thu, Feb 10, 2022 at 10:34 AM Bojan Niceno <
>>> > >> > > > >> bojan.niceno.scientist at gmail.com> wrote:
>>> > >> > > > >>
>>> > >> > > > >>> Dear Satish,
>>> > >> > > > >>>
>>> > >> > > > >>> Thanks for the answer.  Your suggestion makes a lot of
>>> sense,
>>> > >> but
>>> > >> > > this
>>> > >> > > > >>> is what I get as a result of that:
>>> > >> > > > >>>
>>> > >> > > > >>> Running check examples to verify correct installation
>>> > >> > > > >>> Using PETSC_DIR=/home/niceno/Development/petsc-debug and
>>> > >> > > > >>> PETSC_ARCH=arch-linux-c-debug
>>> > >> > > > >>> Possible error running C/C++ src/snes/tutorials/ex19 with
>>> 1 MPI
>>> > >> > > process
>>> > >> > > > >>> See http://www.mcs.anl.gov/petsc/documentation/faq.html
>>> > >> > > > >>> Invalid MIT-MAGIC-COOKIE-1 keylid velocity = 0.0016,
>>> prandtl #
>>> > >> = 1.,
>>> > >> > > > >>> grashof # = 1.
>>> > >> > > > >>> Number of SNES iterations = 2
>>> > >> > > > >>> Possible error running C/C++ src/snes/tutorials/ex19 with
>>> 2 MPI
>>> > >> > > processes
>>> > >> > > > >>> See http://www.mcs.anl.gov/petsc/documentation/faq.html
>>> > >> > > > >>> Invalid MIT-MAGIC-COOKIE-1 keylid velocity = 0.0016,
>>> prandtl #
>>> > >> = 1.,
>>> > >> > > > >>> grashof # = 1.
>>> > >> > > > >>> Number of SNES iterations = 2
>>> > >> > > > >>> Possible error running Fortran example
>>> src/snes/tutorials/ex5f
>>> > >> with 1
>>> > >> > > > >>> MPI process
>>> > >> > > > >>> See http://www.mcs.anl.gov/petsc/documentation/faq.html
>>> > >> > > > >>> Invalid MIT-MAGIC-COOKIE-1 keyNumber of SNES iterations
>>> =     4
>>> > >> > > > >>> Completed test examples
>>> > >> > > > >>>
>>> > >> > > > >>> I am getting the "Possible error running Fortran example"
>>> > >> warning
>>> > >> > > with
>>> > >> > > > >>> this.  This somehow looks more severe to me.  But I could
>>> be
>>> > >> wrong.
>>> > >> > > > >>>
>>> > >> > > > >>
>>> > >> > > > >> You are getting this message because your MPI
>>> implementation is
>>> > >> > > printing
>>> > >> > > > >>
>>> > >> > > > >>   Invalid MIT-MAGIC-COOKIE-1 key
>>> > >> > > > >>
>>> > >> > > > >> It is still running fine, but this is an MPI configuration
>>> issue.
>>> > >> > > > >>
>>> > >> > > > >>   Thanks,
>>> > >> > > > >>
>>> > >> > > > >>      Matt
>>> > >> > > > >>
>>> > >> > > > >> Any suggestions what to do?
>>> > >> > > > >>>
>>> > >> > > > >>>
>>> > >> > > > >>>     Kind regards,
>>> > >> > > > >>>
>>> > >> > > > >>>     Bojan
>>> > >> > > > >>>
>>> > >> > > > >>>
>>> > >> > > > >>>
>>> > >> > > > >>> On Wed, Feb 9, 2022 at 5:49 PM Satish Balay <
>>> balay at mcs.anl.gov>
>>> > >> > > wrote:
>>> > >> > > > >>>
>>> > >> > > > >>>> To clarify:
>>> > >> > > > >>>>
>>> > >> > > > >>>> you are using --download-openmpi=yes with petsc. However
>>> you
>>> > >> say:
>>> > >> > > > >>>>
>>> > >> > > > >>>> > > The mpif90 command which
>>> > >> > > > >>>> > > I use to compile the code, wraps gfortran with
>>> OpenMPI
>>> > >> > > > >>>>
>>> > >> > > > >>>> This suggests a different install of OpenMPI is used to
>>> build
>>> > >> your
>>> > >> > > code.
>>> > >> > > > >>>>
>>> > >> > > > >>>> One way to resolve this is - delete current build of
>>> PETSc -
>>> > >> and
>>> > >> > > > >>>> rebuild it with this same MPI [that you are using with
>>> your
>>> > >> > > application]
>>> > >> > > > >>>>
>>> > >> > > > >>>> ./configure --with-cc=mpicc --with-cxx=mpicxx
>>> --with-fc=mpif90
>>> > >> > > > >>>> --download-fblaslapack --download-metis
>>> --download-parmetis
>>> > >> > > --download-cmake
>>> > >> > > > >>>>
>>> > >> > > > >>>> Also PETSc provides makefile format that minimizes such
>>> > >> conflicts..
>>> > >> > > > >>>>
>>> > >> > > > >>>>
>>> > >> > > > >>>>
>>> > >> > >
>>> > >>
>>> https://petsc.org/release/docs/manual/getting_started/#writing-c-c-or-fortran-applications
>>> > >> > > > >>>>
>>> > >> > > > >>>> Satish
>>> > >> > > > >>>>
>>> > >> > > > >>>> On Wed, 9 Feb 2022, Balay, Satish via petsc-users wrote:
>>> > >> > > > >>>>
>>> > >> > > > >>>> > Are you using the same MPI to build both PETSc and your
>>> > >> > > appliation?
>>> > >> > > > >>>> >
>>> > >> > > > >>>> > Satish
>>> > >> > > > >>>> >
>>> > >> > > > >>>> > On Wed, 2022-02-09 at 05:21 +0100, Bojan Niceno wrote:
>>> > >> > > > >>>> > > To whom it may concern,
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > > I am working on a Fortran (2003) computational fluid
>>> > >> dynamics
>>> > >> > > > >>>> solver,
>>> > >> > > > >>>> > > which is actually quite mature, was parallelized
>>> with MPI
>>> > >> from
>>> > >> > > the
>>> > >> > > > >>>> > > very beginning and it comes with its own suite of
>>> Krylov
>>> > >> > > solvers.
>>> > >> > > > >>>> > > Although the code is self-sustained, I am inclined to
>>> > >> believe
>>> > >> > > that
>>> > >> > > > >>>> it
>>> > >> > > > >>>> > > would be better to use PETSc instead of my own
>>> home-grown
>>> > >> > > solvers.
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > > In the attempt to do so, I have installed PETSc
>>> 3.16.4 with
>>> > >> > > > >>>> following
>>> > >> > > > >>>> > > options:
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > > ./configure --with-debugging=yes
>>> --download-openmpi=yes
>>> > >> > > --download-
>>> > >> > > > >>>> > > fblaslapack=yes --download-metis=yes
>>> > >> --download-parmetis=yes --
>>> > >> > > > >>>> > > download-cmake=yes
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > > on a workstation running Ubuntu 20.04 LTS.  The
>>> mpif90
>>> > >> command
>>> > >> > > which
>>> > >> > > > >>>> > > I use to compile the code, wraps gfortran with
>>> OpenMPI,
>>> > >> hence
>>> > >> > > the
>>> > >> > > > >>>> > > option "--download-openmpi=yes" when configuring
>>> PETSc.
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > > Anyhow, installation of PETSc went fine, I managed
>>> to link
>>> > >> and
>>> > >> > > run
>>> > >> > > > >>>> it
>>> > >> > > > >>>> > > with my code, but I am getting the following messages
>>> > >> during
>>> > >> > > > >>>> > > compilation:
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > > Petsc_Mod.f90:18:6:
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > >    18 |   use PetscMat, only: tMat,
>>> MAT_FINAL_ASSEMBLY
>>> > >> > > > >>>> > >       |      1
>>> > >> > > > >>>> > > Warning: Named COMMON block ‘mpi_fortran_bottom’ at
>>> (1)
>>> > >> shall
>>> > >> > > be of
>>> > >> > > > >>>> > > the same size as elsewhere (4 vs 8 bytes)
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > > Petsc_Mod.f90 is a module I wrote for interfacing
>>> PETSc.
>>> > >> All
>>> > >> > > works,
>>> > >> > > > >>>> > > but these messages give me a reason to worry.
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > > Can you tell what causes this warnings?  I would
>>> guess they
>>> > >> > > might
>>> > >> > > > >>>> > > appear if one mixes OpenMPI with MPICH, but I don't
>>> think
>>> > >> I even
>>> > >> > > > >>>> have
>>> > >> > > > >>>> > > MPICH on my system.
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > > Please let me know what you think about it?
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > >     Cheers,
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > >     Bojan
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> > >
>>> > >> > > > >>>> >
>>> > >> > > > >>>> >
>>> > >> > > > >>>>
>>> > >> > > > >>>
>>> > >> > > > >>
>>> > >> > > > >> --
>>> > >> > > > >> What most experimenters take for granted before they begin
>>> their
>>> > >> > > > >> experiments is infinitely more interesting than any
>>> results to
>>> > >> which
>>> > >> > > their
>>> > >> > > > >> experiments lead.
>>> > >> > > > >> -- Norbert Wiener
>>> > >> > > > >>
>>> > >> > > > >> https://www.cse.buffalo.edu/~knepley/
>>> > >> > > > >> <http://www.cse.buffalo.edu/~knepley/>
>>> > >> > > > >>
>>> > >> > > > >
>>> > >> > > >
>>> > >> > > >
>>> > >> > >
>>> > >> >
>>> > >>
>>> > >
>>> >
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20220214/5bfdb907/attachment-0001.html>

More information about the petsc-users mailing list