- mpi-comments - mailman3.cels.anl.gov

Re: [Mpi-comments] On the use of environment variables
by Bruno Kinoshita 18 Mar '26

18 Mar '26

This is exactly what I needed. Thank you so much for the answer, pointers and links, Christoph! Best regards Bruno On Wed, 18 Mar 2026 at 10:42, Christoph Niethammer <niethammer(a)hlrs.de> wrote: > Dear Bruno, > > As you correctly noted, the MPI Standard does not define and make any > environment variables mandatory for implementations. > This differs from, e.g., OpenMP, which specifies several environment > variables to control aspects such as the number of threads or thread > binding. [1] > > That said, MPI implementations, and the communication libraries they use > under the hood (e.g., UCX), commonly use environment variables to control > various runtime behaviors and tuning parameters. > > Here some pointers if you want to explore this further: > * Open MPI: > - Variables set by Open MPI: > https://urldefense.us/v3/__https://docs.open-mpi.org/en/main/tuning-apps/en… > - Variables used to control the wrapper compilers: > https://urldefense.us/v3/__https://docs.open-mpi.org/en/v5.0.x/building-app… > - Variables to control MCA parameters: > https://urldefense.us/v3/__https://docs.open-mpi.org/en/v5.0.x/mca.html*env… > * MPICH: > - READMES and the mpiexec man pages of mpich > - Variables in the HPE MPICH documentation: > https://urldefense.us/v3/__https://cpe.ext.hpe.com/docs/24.07/mpt/mpich/int… > * UCX: > - You can list all variables via the command `ucx_info -f` > > Best regards > Christoph Niethammer > > [1] https://urldefense.us/v3/__https://www.openmp.org/spec-html/5.2/openmpch21.… > > > > ----- Original Message ----- > From: "Bruno Kinoshita via mpi-comments" <mpi-comments(a)lists.mpi-forum.org > > > To: "mpi-comments" <mpi-comments(a)lists.mpi-forum.org> > Cc: "Bruno Kinoshita" <brunodepaulak(a)gmail.com> > Sent: Tuesday, 17 March, 2026 21:46:59 > Subject: [Mpi-comments] On the use of environment variables > > Hello, I am working on a masters thesis about workflows with CWL and MPI. > The workflows I am working with can be executed with tools (CWL runners) > that try to isolate the execution of each step. This isolation can be done, > for instance, with > ZjQcmQRYFpfptBannerStart > This Message Is From an External Sender > This message came from outside your organization. > ZjQcmQRYFpfptBannerEnd > Hello, > > I am working on a masters thesis about workflows with CWL and MPI. The > workflows I am working with can be executed with tools (CWL runners) that > try to isolate the execution of each step. > > This isolation can be done, for instance, with containers. When using the > containers, some CWL runners normally remove all environment variables > (e.g., with --cleanenv with Singularity). > > Before passing all variables from the host or creating a wrapper script to > control which variables are sent, I am searching for confirmation that the > environment variables used in MPI implementations are not defined in the > standard. > > I read parts of the MPI 5.0 specification, and I could only find > environment variables being mentioned as one of the possible mechanisms to > solve some sort of problem. So my understanding right now is that the MPI > specification does not say implementations MUST use environment variables, > only they MAY. > > Could anyone please confirm this? This is my first message in the list; I > hope I am posting in the right place. If not, I apologize. > > Thank you, > > Bruno > > > > _______________________________________________ > mpi-comments mailing list > mpi-comments(a)lists.mpi-forum.org > https://urldefense.us/v3/__https://lists.mpi-forum.org/mailman/listinfo/mpi… > >

1 0

Re: [Mpi-comments] On the use of environment variables
by Christoph Niethammer 18 Mar '26

18 Mar '26

1 0

On the use of environment variables
by Bruno Kinoshita 17 Mar '26

17 Mar '26

Hello, I am working on a masters thesis about workflows with CWL and MPI. The workflows I am working with can be executed with tools (CWL runners) that try to isolate the execution of each step. This isolation can be done, for instance, with containers. When using the containers, some CWL runners normally remove all environment variables (e.g., with --cleanenv with Singularity). Before passing all variables from the host or creating a wrapper script to control which variables are sent, I am searching for confirmation that the environment variables used in MPI implementations are not defined in the standard. I read parts of the MPI 5.0 specification, and I could only find environment variables being mentioned as one of the possible mechanisms to solve some sort of problem. So my understanding right now is that the MPI specification does not say implementations MUST use environment variables, only they MAY. Could anyone please confirm this? This is my first message in the list; I hope I am posting in the right place. If not, I apologize. Thank you, Bruno

1 0

Re: [Mpi-comments] Comment on example 6.36 of MPI 4.1 document
by Jim Dinan 17 Nov '24

17 Nov '24

I created an issue for this here: https://urldefense.us/v3/__https://github.com/mpi-forum/mpi-issues/issues/8… On Fri, Jan 19, 2024 at 9:57 AM Arham Amouei via mpi-comments < mpi-comments(a)lists.mpi-forum.org> wrote: > Hi > > This comment is about example 6.36 of MPI 4.1 document (on page 303) which > concerns the usage of collective operations with overlapping communicators. > I have already posted my comment as a question on stackoverflow.com, so I > won't repeat it here. Here is the link to the post: > https://urldefense.us/v3/__https://stackoverflow.com/questions/77844104/ove… > > The link to the standard: > https://urldefense.us/v3/__https://www.mpi-forum.org/docs/mpi-4.1/mpi41-rep… > > Kind regards, > Arham > _______________________________________________ > mpi-comments mailing list > mpi-comments(a)lists.mpi-forum.org > https://urldefense.us/v3/__https://lists.mpi-forum.org/mailman/listinfo/mpi… >

1 0

Comment on example 6.36 of MPI 4.1 document
by Arham Amouei 19 Jan '24

19 Jan '24

Hi This comment is about example 6.36 of MPI 4.1 document (on page 303) which concerns the usage of collective operations with overlapping communicators. I have already posted my comment as a question on stackoverflow.com, so I won't repeat it here. Here is the link to the post: https://stackoverflow.com/questions/77844104/overlapping-communicators-and-… The link to the standard: https://www.mpi-forum.org/docs/mpi-4.1/mpi41-report.pdf Kind regards, Arham

1 0

Re: [Mpi-comments] Use of MPI_DATATYPE_NULL
by George Bosilca 08 Dec '22

08 Dec '22

Thanks Gilles for pointing this discussion out. It appears that the use of the MPI_DATATYPE_NULL in the example 5.23 is correct, the standard clearly mentions the receiver datatype is only significant at the root. George. On Thu, Dec 8, 2022 at 7:03 PM Jim Edwards via mpi-comments < mpi-comments(a)lists.mpi-forum.org> wrote: > Right, so how do I formally request that the standard be updated to > clarify this issue? > > On Thu, Dec 8, 2022 at 4:56 PM gilles--- via mpi-comments < > mpi-comments(a)lists.mpi-forum.org> wrote: > >> I have a different recollection. >> >> >> >> The link is broken, here is one that works for me >> >> https://lists.mpi-forum.org/pipermail/mpi-forum/2016-January/003152.html >> >> TL;DR from Bill Gropp: >> >> "The standard says that the null handles are invalid for input unless explicitly permitted. There’s no exemption for a datatype argument where the associated count is zero." >> >> >> >> IIRC there was no objection to updating the standard in order to allow >> this, but unless this has been completed, I am afraid MPI_DATATYPE_NULL is >> not a valid datatype here >> >> >> >> Gilles >> >> ----- Original Message ----- >> >> You are correct. The MPI_DATATYPE_NULL is an acceptable datatype for >> cases where the target processor sends or receives no data. The standard >> uses MPI_DATATYPE_NULL in 2 instances (in example 5.23) to highlight no >> data movement, but fails to make a clear reference to the fact that this is >> an acceptable use. >> >> George. >> >> On Thu, Dec 8, 2022 at 5:31 PM Jim Edwards via mpi-comments < >> mpi-comments(a)lists.mpi-forum.org> wrote: >> >>> There seems to be some ambiguity in the standard with respect to usage >>> of the MPI_DATATYPE_NULL variable. >>> It would seem to me that if an MPI routine such as MPI_Send or >>> MPI_ALLTOALLW has a data count argument of 0 >>> then MPI_DATATYPE_NULL should be an acceptable value for the datatype. >>> Here is a discussion from several years >>> ago with openmpi developers. It references a discussion in the >>> standards forum but I can no longer find it. >>> >>> http://lists.mpi-forum.org/mpi-forum/2016/01/3152.php >>> >>> Would it be possibly to clarify this case in the next iteration of the >>> standard? >>> >>> -- >>> Jim Edwards >>> >>> CESM Software Engineer >>> National Center for Atmospheric Research >>> Boulder, CO >>> _______________________________________________ >>> mpi-comments mailing list >>> mpi-comments(a)lists.mpi-forum.org >>> https://lists.mpi-forum.org/mailman/listinfo/mpi-comments >> >> >> _______________________________________________ >> mpi-comments mailing list >> mpi-comments(a)lists.mpi-forum.org >> https://lists.mpi-forum.org/mailman/listinfo/mpi-comments >> > > > -- > Jim Edwards > > CESM Software Engineer > National Center for Atmospheric Research > Boulder, CO > _______________________________________________ > mpi-comments mailing list > mpi-comments(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpi-comments >

1 0

Re: [Mpi-comments] Use of MPI_DATATYPE_NULL
by Jim Edwards 08 Dec '22

08 Dec '22

Right, so how do I formally request that the standard be updated to clarify this issue? On Thu, Dec 8, 2022 at 4:56 PM gilles--- via mpi-comments < mpi-comments(a)lists.mpi-forum.org> wrote: > I have a different recollection. > > > > The link is broken, here is one that works for me > > https://lists.mpi-forum.org/pipermail/mpi-forum/2016-January/003152.html > > TL;DR from Bill Gropp: > > "The standard says that the null handles are invalid for input unless explicitly permitted. There’s no exemption for a datatype argument where the associated count is zero." > > > > IIRC there was no objection to updating the standard in order to allow > this, but unless this has been completed, I am afraid MPI_DATATYPE_NULL is > not a valid datatype here > > > > Gilles > > ----- Original Message ----- > > You are correct. The MPI_DATATYPE_NULL is an acceptable datatype for cases > where the target processor sends or receives no data. The standard > uses MPI_DATATYPE_NULL in 2 instances (in example 5.23) to highlight no > data movement, but fails to make a clear reference to the fact that this is > an acceptable use. > > George. > > On Thu, Dec 8, 2022 at 5:31 PM Jim Edwards via mpi-comments < > mpi-comments(a)lists.mpi-forum.org> wrote: > >> There seems to be some ambiguity in the standard with respect to usage of >> the MPI_DATATYPE_NULL variable. >> It would seem to me that if an MPI routine such as MPI_Send or >> MPI_ALLTOALLW has a data count argument of 0 >> then MPI_DATATYPE_NULL should be an acceptable value for the datatype. >> Here is a discussion from several years >> ago with openmpi developers. It references a discussion in the standards >> forum but I can no longer find it. >> >> http://lists.mpi-forum.org/mpi-forum/2016/01/3152.php >> >> Would it be possibly to clarify this case in the next iteration of the >> standard? >> >> -- >> Jim Edwards >> >> CESM Software Engineer >> National Center for Atmospheric Research >> Boulder, CO >> _______________________________________________ >> mpi-comments mailing list >> mpi-comments(a)lists.mpi-forum.org >> https://lists.mpi-forum.org/mailman/listinfo/mpi-comments > > > _______________________________________________ > mpi-comments mailing list > mpi-comments(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpi-comments > -- Jim Edwards CESM Software Engineer National Center for Atmospheric Research Boulder, CO

1 0

Re: [Mpi-comments] Use of MPI_DATATYPE_NULL
by gilles＠rist.or.jp 08 Dec '22

08 Dec '22

I have a different recollection. The link is broken, here is one that works for me https://lists.mpi-forum.org/pipermail/mpi-forum/2016-January/003152.html TL;DR from Bill Gropp: "The standard says that the null handles are invalid for input unless explicitly permitted. There’s no exemption for a datatype argument where the associated count is zero." IIRC there was no objection to updating the standard in order to allow this, but unless this has been completed, I am afraid MPI_DATATYPE_NULL is not a valid datatype here Gilles ----- Original Message ----- You are correct. The MPI_DATATYPE_NULL is an acceptable datatype for cases where the target processor sends or receives no data. The standard uses MPI_DATATYPE_NULL in 2 instances (in example 5.23) to highlight no data movement, but fails to make a clear reference to the fact that this is an acceptable use. George. On Thu, Dec 8, 2022 at 5:31 PM Jim Edwards via mpi-comments <mpi- comments(a)lists.mpi-forum.org> wrote: There seems to be some ambiguity in the standard with respect to usage of the MPI_DATATYPE_NULL variable. It would seem to me that if an MPI routine such as MPI_Send or MPI_ ALLTOALLW has a data count argument of 0 then MPI_DATATYPE_NULL should be an acceptable value for the datatype. Here is a discussion from several years ago with openmpi developers. It references a discussion in the standards forum but I can no longer find it. http://lists.mpi-forum.org/mpi-forum/2016/01/3152.php Would it be possibly to clarify this case in the next iteration of the standard? -- Jim Edwards CESM Software Engineer National Center for Atmospheric Research Boulder, CO _______________________________________________ mpi-comments mailing list mpi-comments(a)lists.mpi-forum.org https://lists.mpi-forum.org/mailman/listinfo/mpi-comments

1 0

Re: [Mpi-comments] Use of MPI_DATATYPE_NULL
by George Bosilca 08 Dec '22

08 Dec '22

You are correct. The MPI_DATATYPE_NULL is an acceptable datatype for cases where the target processor sends or receives no data. The standard uses MPI_DATATYPE_NULL in 2 instances (in example 5.23) to highlight no data movement, but fails to make a clear reference to the fact that this is an acceptable use. George. On Thu, Dec 8, 2022 at 5:31 PM Jim Edwards via mpi-comments < mpi-comments(a)lists.mpi-forum.org> wrote: > There seems to be some ambiguity in the standard with respect to usage of > the MPI_DATATYPE_NULL variable. > It would seem to me that if an MPI routine such as MPI_Send or > MPI_ALLTOALLW has a data count argument of 0 > then MPI_DATATYPE_NULL should be an acceptable value for the datatype. > Here is a discussion from several years > ago with openmpi developers. It references a discussion in the standards > forum but I can no longer find it. > > http://lists.mpi-forum.org/mpi-forum/2016/01/3152.php > > Would it be possibly to clarify this case in the next iteration of the > standard? > > > -- > Jim Edwards > > CESM Software Engineer > National Center for Atmospheric Research > Boulder, CO > _______________________________________________ > mpi-comments mailing list > mpi-comments(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpi-comments >

1 0

Use of MPI_DATATYPE_NULL
by Jim Edwards 08 Dec '22

08 Dec '22

There seems to be some ambiguity in the standard with respect to usage of the MPI_DATATYPE_NULL variable. It would seem to me that if an MPI routine such as MPI_Send or MPI_ALLTOALLW has a data count argument of 0 then MPI_DATATYPE_NULL should be an acceptable value for the datatype. Here is a discussion from several years ago with openmpi developers. It references a discussion in the standards forum but I can no longer find it. http://lists.mpi-forum.org/mpi-forum/2016/01/3152.php Would it be possibly to clarify this case in the next iteration of the standard? -- Jim Edwards CESM Software Engineer National Center for Atmospheric Research Boulder, CO

1 0