Update MSMPI to version 10.1.3

[mmuetzel]

Apparently, Microsoft released version 10.1.3 of MSMPI in June 2023:
https://learn.microsoft.com/en-us/message-passing-interface/microsoft-mpi-release-notes

Would it be possible that this new version gets installed by this action?

[dalcinl]

Yes, it can be done. It is basically a matter of changing the download URL to point to the MS download center.
It is unfortunate that Microsoft does not keep their GitHub repository releases in sync with their published packaged downloads.

[dalcinl]

@mmuetzel Could you please run a quick test on your side temporarily using mpi4py/setup-mpi@msmpi ?

[mmuetzel]

Thank you for looking into this so swiftly. 🚀

I've updated to that branch in a fork of the ElmerFEM repository:
https://github.com/mmuetzel/elmerfem/actions/runs/19389003085/job/55484053366
(Ignore the "MSYS2 (CLANG64, bundled dependencies)" runner. That is an experiment that fails to build for an unrelated reason.)

Version 10.13 of MSMPI was downloaded and installed correctly by the action:

Installing msmpi
  Downloading Microsoft MPI 10.1.3
  Downloading https://download.microsoft.com/download/7/2/7/72731ebb-b63c-4170-ade7-836966263a8f/msmpisdk.msi
  Downloading https://download.microsoft.com/download/7/2/7/72731ebb-b63c-4170-ade7-836966263a8f/msmpisetup.exe
  Installing Microsoft MPI 10.1.3
  Adding environment variables to $GITHUB_ENV
  MSMPI_BIN=C:\Program Files\Microsoft MPI\Bin\
  MSMPI_INC=C:\Program Files (x86)\Microsoft SDKs\MPI\Include\
  MSMPI_LIB32=C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86\
  MSMPI_LIB64=C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x64\
  Adding $MSMPI_BIN to $GITHUB_PATH

The CMake build rules successfully detected (and identified) MSMPI:

-- Found MPI_C: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0")
-- Found MPI_CXX: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0")
-- Found MPI_Fortran: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0")
-- Found MPI: TRUE (found version "2.0")
[...]
-- ------------------------------------------------
--   MPI Fortran:             TRUE
--   MPI Fortran compiler:    D:/a/_temp/msys64/mingw64/bin/mpif90.exe
--   MPI Fortran flags:       
--   MPI Fortran include dir: D:/a/_temp/msys64/mingw64/include
--   MPI Fortran libraries:   D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a
--   MPI Fortran link flags:  
-- ------------------------------------------------
--   MPI C:             TRUE
--   MPI C compiler:    D:/a/_temp/msys64/mingw64/bin/mpicc.exe
--   MPI C flags:       
--   MPI C include dir: D:/a/_temp/msys64/mingw64/include
--   MPI C libraries:   D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a
--   MPI C flags:       
-- ------------------------------------------------

ElmerFEM linked successfully against the updated version.
Some tests failed randomly. (But that was already the case with the previous version. I had the shallow hope that updating MSMPI would improve that. But apparently there's no change in that respect.)

All-in-all, I'd say the update looks good. 👍

[dalcinl]

The CMake build rules successfully detected (and identified) MSMPI:

-- Found MPI_C: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0")
-- Found MPI_CXX: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0")
-- Found MPI_Fortran: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0")
-- Found MPI: TRUE (found version "2.0")

This is a bit confusing, though I'm not a Windows expert. What's the connection between the stuff in D:/a/_temp/msys64/mingw64/ and the MS MPI installation in C:\Program Files\ ?
Is the first some sort of shallow wrapper of the second to provide a POSIX-like install tree?

Some tests failed randomly.

Are those failures MPI-related? If yes, any chance you can pinpoint which MPI call is failing?

I had the shallow hope that updating MSMPI would improve that.

Do you really need to stick to Microsoft MPI? IMHO, everyone on Windows should be using Intel MPI.

All-in-all, I'd say the update looks good. 👍

Then I guess that I'll move this update forward and publish a new minor release.

[mmuetzel]

Sorry for the late reply.

-- Found MPI_C: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0")
-- Found MPI_CXX: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0")
-- Found MPI_Fortran: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0")
-- Found MPI: TRUE (found version "2.0")

This is a bit confusing, though I'm not a Windows expert. What's the connection between the stuff in D:/a/_temp/msys64/mingw64/ and the MS MPI installation in C:\Program Files\ ? Is the first some sort of shallow wrapper of the second to provide a POSIX-like install tree?

Linking to shared libraries on Windows works a bit differently compared to other platforms.
There are three "types" of libraries on Windows (with their default file extensions for MinGW):

• static libraries (*.a): Those are very similar to (static) archives on Linux.
• dynamic link libraries (*.dll): These libraries export symbols which can be loaded when starting a program or on runtime. They are probably the closest to what shared libraries (*.so) are on Linux.
• import libraries (*.dll.a): Windows needs to resolve all symbols on link-time. (That is somewhat similar to static linking on Linux.) The link libraries contain the "glue" that is needed to link to the symbols that are exported by DLLs. That "glue" might need to look different depending on the toolchain (e.g., MSVC vs. binutils vs. LLVM). Afaict, there is no equivalent to that concept on Linux.

MSYS2 provides the link library for MSMPI that contains the "glue" that is needed to link to the "actual" .dll that is installed in C:\Windows\System. The MS MPI installation in C:\Program Files contains some executables that are needed when using MPI like mpiexec.exe.

Are those failures MPI-related? If yes, any chance you can pinpoint which MPI call is failing?

Unfortunately, I don't know why these tests fail randomly. The frequency of the failures increases if multiple tests are run in parallel. That might indicate that it relates to something multithreading. Afaict, there is only a vague assumption that these failures could be related to MSMPI.
Maybe, @juharu (one of the developers of ElmerFEM) found out something more concrete in the meantime?

Do you really need to stick to Microsoft MPI? IMHO, everyone on Windows should be using Intel MPI.

There is probably no real reason to stick to MSMPI.
ElmerFEM is primarily developed on Linux. That's why I thought it would be easiest to use MinGW (instead of MSVC) when porting it to Windows.
Additionally, it is written primarily in Fortran. That was another reason why I thought that using a toolchain that includes a Fortran compiler (i.e., GCC instead of MSVC) would be reasonable.
I seem to recall that I chose MSMPI when I first set up the CI on Windows for ElmerFEM because MSYS2 provides the "glue" for that implementation of MPI. I didn't look into how difficult it would be to use Intel MPI with gcc/gfortran/ld.bfd/etc. on Windows.
If most everyone on Windows uses Intel MPI, it might be worth having a look into switching to that, too. (I don't know when I'll be able to come around to that though.)

Then I guess that I'll move this update forward and publish a new minor release.

Sounds good to me. 👍

[juharu]

Hi Markus I've no idea why MS MPI is used, probably something that has been easily available at some point in the past. Nothing against swiching / updating this. Regards, Juha

…

________________________________ Lähettäjä: Markus Mützel ***@***.***> Lähetetty: tiistai 18. marraskuuta 2025 14.15 Vastaanottaja: mpi4py/setup-mpi ***@***.***> Kopio: juharu ***@***.***>; Mention ***@***.***> Aihe: Re: [mpi4py/setup-mpi] Update MSMPI to version 10.1.3 (Issue #21) [https://avatars.githubusercontent.com/u/65065102?s=20&v=4]mmuetzel left a comment (mpi4py/setup-mpi#21)<#21 (comment)> Sorry for the late reply.

-- Found MPI_C: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0") -- Found MPI_CXX: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0") -- Found MPI_Fortran: D:/a/_temp/msys64/mingw64/lib/libmsmpi.dll.a (found version "2.0") -- Found MPI: TRUE (found version "2.0") This is a bit confusing, though I'm not a Windows expert. What's the connection between the stuff in D:/a/_temp/msys64/mingw64/ and the MS MPI installation in C:\Program Files\ ? Is the first some sort of shallow wrapper of the second to provide a POSIX-like install tree? Linking to shared libraries on Windows works a bit differently compared to other platforms. There are three "types" of libraries on Windows (with their default file extensions for MinGW): * static libraries (*.a): Those are very similar to (static) archives on Linux. * dynamic link libraries (*.dll): These libraries export symbols which can be loaded when starting a program or on runtime. They are probably the closest to what shared libraries (*.so) are on Linux. * link libraries (*.dll.a): Windows needs to resolve all symbols on link-time. (That is somewhat similar to static linking on Linux.) The link libraries contain the "glue" that is needed to link to the symbols that are exported by DLLs. That "glue" might need to look different depending on the toolchain (e.g., MSVC vs. binutils vs. LLVM). Afaict, there is no equivalent to that concept on Linux. MSYS2 provides the link library for MSMPI that contains the "glue" that is needed to link to the "actual" .dll that is installed in C:\Windows\System. The MS MPI installation in C:\Program Files contains some executables that are needed when using MPI like mpiexec.exe. Are those failures MPI-related? If yes, any chance you can pinpoint which MPI call is failing? Unfortunately, I don't know why these tests fail randomly. The frequency of the failures increases if multiple tests are run in parallel. That might indicate that it relates to something multithreading. Afaict, there is only a vague assumption that these failures could be related to MSMPI. Maybe, @juharu<https://github.com/juharu> (one of the developers of ElmerFEM) found out something more concrete in the meantime? Do you really need to stick to Microsoft MPI? IMHO, everyone on Windows should be using Intel MPI. There is probably no real reason to stick to MSMPI. ElmerFEM is primarily developed on Linux. That's why I thought it would be easiest to use MinGW (instead of MSVC) when porting it to Windows. Additionally, it is written primarily in Fortran. That was another reason why I thought that using a toolchain that includes a Fortran compiler (i.e., GCC instead of MSVC) would be reasonable. I seem to recall that I chose MSMPI when I first set up the CI on Windows for ElmerFEM because MSYS2 provides the "glue" for that implementation of MPI. I didn't look into how difficult it would be to use Intel MPI with gcc/gfortran/ld.bfd/etc. on Windows. If most everyone on Windows uses Intel MPI, it might be worth having a look into switching to that, too. (I don't know when I'll be able to come around to that though.) Then I guess that I'll move this update forward and publish a new minor release. Sounds good to me. 👍 — Reply to this email directly, view it on GitHub<#21 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ACTOMSXYDY3UCZOZDY2UWZT35MEV3AVCNFSM6AAAAACMDAFVDKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTKNBXGM2DQNBRGM>. You are receiving this because you were mentioned.Message ID: ***@***.***> The information in this email may be confidential and is intended solely for the use of the individual or entity to whom it is intended. If you are not the intended recipient of this message, please delete the message and notify the sender immediately. For information on how we process personal data and our contact information, please see CSC's website: Privacy<https://csc.fi/en/privacy> Tämän sähköpostin tiedot voivat olla luottamuksellisia ja ne on tarkoitettu yksinomaan sen henkilön tai yhteisön käyttöön, jolle ne on osoitettu. Jos et ole viestissä tarkoitettu vastaanottaja, tuhoa viesti ja ilmoita asiasta välittömästi viestin lähettäjälle. Tietoja henkilötietojen ja yhteystietojen käsittelystä löydät CSC:n verkkosivuilta: Tietosuoja<https://csc.fi/tietosuoja>

[mmuetzel]

@juharu: The main reason to @-mention you was in case you could give @dalcinl some hints as to what exactly is causing the random test errors.

[juharu]

I'm sorry, no further ideas atm. Obivously limiting concurrently running processes makes a difference, but still random failures now and then. Recently we've had some failures also for macos-15 runner. This seemed to heal when setting OMP_NUM_THREADS=1 before running ctest (not REALLY sure, but almost). I see my 6cpu laptop running ubuntu 25.04 launchning 12 threads if the env. variable not set. I think this is somewhat recent feature (of either elmersolver or some library called).

…

________________________________ Lähettäjä: Markus Mützel ***@***.***> Lähetetty: keskiviikko 19. marraskuuta 2025 12.07 Vastaanottaja: mpi4py/setup-mpi ***@***.***> Kopio: juharu ***@***.***>; Mention ***@***.***> Aihe: Re: [mpi4py/setup-mpi] Update MSMPI to version 10.1.3 (Issue #21) [https://avatars.githubusercontent.com/u/65065102?s=20&v=4]mmuetzel left a comment (mpi4py/setup-mpi#21)<#21 (comment)> @juharu<https://github.com/juharu>: The main reason to @-mention you was in case you could give @dalcinl<https://github.com/dalcinl> some hints as to what exactly is causing the random test errors. — Reply to this email directly, view it on GitHub<#21 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ACTOMSTGZ64HDQ2ENE7XGXL35Q6P5AVCNFSM6AAAAACMDAFVDKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTKNJRHA3DSMRXGM>. You are receiving this because you were mentioned.Message ID: ***@***.***> The information in this email may be confidential and is intended solely for the use of the individual or entity to whom it is intended. If you are not the intended recipient of this message, please delete the message and notify the sender immediately. For information on how we process personal data and our contact information, please see CSC's website: Privacy<https://csc.fi/en/privacy> Tämän sähköpostin tiedot voivat olla luottamuksellisia ja ne on tarkoitettu yksinomaan sen henkilön tai yhteisön käyttöön, jolle ne on osoitettu. Jos et ole viestissä tarkoitettu vastaanottaja, tuhoa viesti ja ilmoita asiasta välittömästi viestin lähettäjälle. Tietoja henkilötietojen ja yhteystietojen käsittelystä löydät CSC:n verkkosivuilta: Tietosuoja<https://csc.fi/tietosuoja>

[mmuetzel]

@dalcinl: It looks like the randomly failing tests disappeared after an error in the "glue" from MSYS2 was fixed.

It would probably still make sense to update to the latest version of MS-MPI. Could you please merge that update to the master branch of this repository?

[dalcinl]

Done. Thanks!

[mmuetzel]

Thanks!