[mpich-discuss] Fortran 08 tests with 'use mpi'

Fri Jan 14 08:10:50 CST 2022

Gregor,

I see your issue now. Thanks for reporting. We'll fix it.

--
Hui Zhou

________________________________
From: Gregor Corbin <g.corbin at fz-juelich.de>
Sent: Friday, January 14, 2022, 6:27 AM
To: Zhou, Hui; discuss at mpich.org; Raffenetti, Ken
Subject: Re: [mpich-discuss] Fortran 08 tests with 'use mpi'

Dear Hui,

I suspect that the issue I reported is a different one. To confirm this, I compiled MPICH (from 4.0rc3 tarball) once with GCC and once with Intel and then compiled the attached test programs each with both MPICHs.

The Intel-MPICH does not have the gcc-bug from issue 4170, but the indices returned by MPI_Testsome are still wrong.

Gregor

On 13.01.22 15:39, Zhou, Hui wrote:
Hi Gregor,

This issue was reported before. You can track the issue here: https://github.com/pmodels/mpich/issues/4170

--
Hui Zhou

From: Gregor Corbin via discuss <discuss at mpich.org><mailto:discuss at mpich.org>
Date: Wednesday, January 12, 2022 at 5:19 AM
To: Raffenetti, Ken <raffenet at anl.gov><mailto:raffenet at anl.gov>, discuss at mpich.org<mailto:discuss at mpich.org> <discuss at mpich.org><mailto:discuss at mpich.org>
Cc: Gregor Corbin <g.corbin at fz-juelich.de><mailto:g.corbin at fz-juelich.de>
Subject: Re: [mpich-discuss] Fortran 08 tests with 'use mpi'

Ken,

while updating the F08 tests with this bugfix, I may have found another bug inside the 'use mpi_f08' wrappers.

The fix removes (superficially) the error in irsendf08, but I still get the same error in pssendf08. This has to do with the array of indices returned by MPI_Testsome: With the 'use mpi_f08' wrappers, the indices in this array are 0-indexed although they should start from 1 according to the Standard (they do start from 1 with the 'use mpi' and 'mpif.h' functions). I have attached a minimal working example to confirm this.

Indeed, the function mpi_testsome_ (in src/binding/fortran/mpif_h/fortran_binding.c) converts the indices:

    for (int i = 0; i < (*outcount); i++) {
        array_of_indices[i] += 1;
    }

The subroutine MPI_Testsome_f08 (in src/binding/fortran/use_mpi_f08/wrappers_f/f08ts.f90) does no such conversion.

The bug is also in MPI_Testany (and potentially other functions I did not test yet).

However, without detailed knowledge about internals of MPICH I might not be the best person to fix the issue.

Greetings,

Gregor

On 11.01.22 22:52, Raffenetti, Ken wrote:

Gregor,

I believe these are test bugs. A fix for the f77 versions was added previously in https://github.com/pmodels/mpich/pull/3962/commits/69b02eaabcdb0008f3879372629cfe92f98e8d56, though it does not cover the prsendf test. Perhaps you could add a commit to your branch to fix these tests in a similar way.

Ken

On 1/11/22, 2:30 AM, "Gregor Corbin" <g.corbin at fz-juelich.de><mailto:g.corbin at fz-juelich.de> wrote:

        Updating the F08 tests to 'use mpi_f08' was mostly

          straightforward. However, i ran into problems with three of the

          tests: irsendf08.f90, pssendf08.f90 and prsendf08.f90.

        With my modified irsendf08 (attached) for example, the test

          produces these error messages:

           Nonnull request in irsend and irecv

             [           0 ] Unexpected source:           0  in irsend and

            irecv

             [           0 ] Unexpected tag:           0  in irsend and

            irecv

             [           0 ] Unexpected count:           0  in irsend and

            irecv

            Invalid data   0.0 at    1 of  666 in irsend and irecv

            Invalid data   0.0 at    2 of  666 in irsend and irecv

            ...

          Invalid data   0.0 at  666 of  666 in irsend and irecv

            Found          669  errors

        That is, the last message from process 1 to 0 is not received

        correctly. When I replace the call to MPI_Waitany with Waitall or

        Wait there are not errors.

        Is there maybe a problem with Waitany or did I just do something

          wrong?

        Gregor

        On 07.01.22 17:59, Raffenetti, Ken

          wrote:

          We would consider a patch, yes. First, you would need to sign and return the contributors agreement. https://www.mpich.org/documentation/contributor-docs/

    As for whether or not automatic conversion is practical, I think if we structured the tests such that the types needing conversion were easily identifiable, then it could be done. It would require some churn of the existing tests. It may also complicate the addition of new, as the author would need to adhere to guidelines that may not be obvious without prior knowledge.

    I gave myself a task to support automatic conversion of the tests the `use mpi_f08`, but it is not very high priority at the moment.

    Ken

    On 1/7/22, 3:10 AM, "Gregor Corbin via discuss" <discuss at mpich.org><mailto:discuss at mpich.org> <mailto:discuss at mpich.org><mailto:discuss at mpich.org> wrote:

        Hi MPICH,

        I noticed that some of the tests in test/mpi/f08 still work with 'use

        mpi' instead of  'use mpi_f08', mostly in the pt2pt folder. These seem

        to be copies of the tests in f90 which in turn have been generated by

        f77tof90.

        Shouldn't these tests be converted to 'use mpi_f08'? If so, would you be

        interested in a PR that fixes this? I would need to do this anyway.

        Kind regards,

        Gregor Corbin

        PS.: The interfaces for 'use mpi' and 'use mpi_f08' are probably too

        different for an automatic conversion to be practical (how to change all

        the integers to type(mpi_xxx) for example?).

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpich.org/pipermail/discuss/attachments/20220114/cae53d06/attachment-0001.html>