[Mvapich-discuss] Build Failure with NVHPC 23.3
Louise Spellacy
louise.spellacy at linaro.org
Wed May 17 10:00:09 EDT 2023
Hi Sylvain,
I forced it with a patch to the configure file - but I wouldn't recommend
it as a long term fix:
--- a/mvapich2-2.3.7/configure
> +++ b/mvapich2-2.3.7/configure
> @@ -34260,7 +34260,7 @@ ac_compile='$F77 -c $FFLAGS conftest.$ac_ext >&5'
> ac_link='$F77 -o conftest$ac_exeext $FFLAGS $LDFLAGS conftest.$ac_ext
> $LIBS >&5'
> ac_compiler_gnu=$ac_cv_f77_compiler_gnu
>
> -
> +f08_works=no
> if test "$f08_works" = "yes" ; then
> f08_works=yes
>
else
This was the only way I could get mpif90 to work.
Thanks,
Louise
On Mon, 15 May 2023 at 15:29, Korzennik, Sylvain <skorzennik at cfa.harvard.edu>
wrote:
> Thx Louise. I hadn't yet check the build. How did you disable have f08 in
> the config script?
>
> Cheers,
> Sylvain
> ---
>
> On Mon, May 15, 2023, 10:12 AM Louise Spellacy <louise.spellacy at linaro.org>
> wrote:
>
>> Hi All,
>>
>> The above patch solves the -rdynamic issue but doesn't resolve the
>> Fortran 2008 isuse.
>>
>> With the above patch, mvapich2 is still being configured
>> with HAVE_F08_BINDING set to 1. The configure and make steps now work but
>> mpif90 now has a undefined reference:
>>
>>> mpif90 -g -o test test.f90
>>> /home/user/install/mvapich2-install/lib/libmpifort.so: undefined
>>> reference to `CFI_is_contiguous'
>>
>>
>> Forcibly changing the configure script to set this to zero results in a
>> usable mpif90.
>>
>> Thanks,
>>
>> Louise
>>
>>
>> On Fri, 12 May 2023 at 16:37, Korzennik, Sylvain <
>> skorzennik at cfa.harvard.edu> wrote:
>>
>>> Same here:
>>> 1- I could not run patch on the file you supplied, but edited the files
>>> directly (am I missing something?)
>>> 2- I get the same "'aclocal-1.15' is missing on your system." err mgs
>>>
>>> So I reverted to the original .m4 & .mk, removed the -rdynamic flags
>>> from the top Makefile after running configure and was able to build mvapich2-2.3.7
>>> w/ nvidia 23.3 compilers.
>>> I have yet to check that all runs fine, tho.
>>>
>>> Also, I had to downgrade the optimization to -O1 (instead of -O4 or -O2)
>>> since I was getting
>>>
>>>> NVC++-F-0007-Source file too large to compile at this optimization
>>>> level (src/mpi/coll/allreduce_tuning.c)
>>>> NVC++/x86-64 Linux 22.9-0: compilation aborted
>>>
>>> that I assume is an issue w/ the NVIDIA compilers (aargh) that seems to
>>> has creeped up since ver 22.9 (ver 22.3 was happy w/ -O4).
>>>
>>> Cheers,
>>> Sylvain
>>> --
>>>
>>>
>>>
>>> On Fri, May 12, 2023 at 9:41 AM Louise Spellacy <
>>> louise.spellacy at linaro.org> wrote:
>>>
>>>> Thanks Nat,
>>>>
>>>> I took the patch and applied it to the mvapich2-2.3.7.tar.gz but it
>>>> now requires that aclocal and automake be installed on the system:
>>>>
>>>> CDPATH="${ZSH_VERSION+.}:" && cd mvapich2-2.3.7 && /bin/bash
>>>>> confdb/missing aclocal-1.15 -I confdb
>>>>> confdb/missing: line 81: aclocal-1.15: command not found
>>>>> WARNING: 'aclocal-1.15' is missing on your system.
>>>>
>>>>
>>>> and
>>>>
>>>> cd mvapich2-2.3.7 && /bin/bash confdb/missing automake-1.15 --foreign
>>>>> Makefile
>>>>> confdb/missing: line 81: automake-1.15: command not found
>>>>> WARNING: 'automake-1.15' is missing on your system.
>>>>
>>>>
>>>> Because the subconfigure.m4 and Makefile.mk files are changed, a new
>>>> configure and Makefile.in file need to be generated. Would it be possible
>>>> to include a patch with the changes to configure and Makefile.in instead of
>>>> the m4 and Makefile.mk file? I know long term they will need to be merged
>>>> as changes to the subconfigure.m4 and Makefile.mk but for testing purposes
>>>> they won't work with the tar ball.
>>>>
>>>> Thanks,
>>>>
>>>> Louise
>>>>
>>>> On Thu, 11 May 2023 at 17:12, Korzennik, Sylvain <
>>>> skorzennik at cfa.harvard.edu> wrote:
>>>>
>>>>> Hi Nat,
>>>>>
>>>>> Thanks for getting back to us. Shall I apply it to mvapich2-2.3.7
>>>>> or mvapich2-2.3.7-1?
>>>>>
>>>>> Cheers,
>>>>> Sylvain
>>>>> --
>>>>>
>>>>>
>>>>>
>>>>> On Thu, May 11, 2023 at 11:02 AM Shineman, Nat <shineman.5 at osu.edu>
>>>>> wrote:
>>>>>
>>>>>> Louise and Sylvain,
>>>>>>
>>>>>> Please try the attached patch to resolve both the real128 and the
>>>>>> rdynamic issues. Let me know if you have any issues.
>>>>>>
>>>>>> Thanks,
>>>>>> Nat
>>>>>> ------------------------------
>>>>>> *From:* Mvapich-discuss <mvapich-discuss-bounces at lists.osu.edu> on
>>>>>> behalf of Subramoni, Hari via Mvapich-discuss <
>>>>>> mvapich-discuss at lists.osu.edu>
>>>>>> *Sent:* Friday, May 5, 2023 11:23
>>>>>> *To:* Korzennik, Sylvain <skorzennik at cfa.harvard.edu>; Announcement
>>>>>> about MVAPICH2 (MPI over InfiniBand, RoCE, Omni-Path, iWARP and EFA)
>>>>>> Libraries developed at NBCL/OSU <mvapich-discuss at lists.osu.edu>;
>>>>>> Louise Spellacy <louise.spellacy at linaro.org>
>>>>>> *Subject:* Re: [Mvapich-discuss] Build Failure with NVHPC 23.3
>>>>>>
>>>>>>
>>>>>> Hi, Sylvain/Louise.
>>>>>>
>>>>>>
>>>>>>
>>>>>> Thanks for the report. We will take a look at this and get back to
>>>>>> you soon.
>>>>>>
>>>>>>
>>>>>>
>>>>>> Best,
>>>>>> Hari.
>>>>>>
>>>>>>
>>>>>>
>>>>>> *From:* Mvapich-discuss <mvapich-discuss-bounces+subramon=
>>>>>> cse.ohio-state.edu at lists.osu.edu> *On Behalf Of *Korzennik, Sylvain
>>>>>> via Mvapich-discuss
>>>>>> *Sent:* Friday, May 5, 2023 11:20 AM
>>>>>> *To:* Louise Spellacy <louise.spellacy at linaro.org>; Announcement
>>>>>> about MVAPICH2 (MPI over InfiniBand, RoCE, Omni-Path, iWARP and EFA)
>>>>>> Libraries developed at NBCL/OSU <mvapich-discuss at lists.osu.edu>
>>>>>> *Subject:* Re: [Mvapich-discuss] Build Failure with NVHPC 23.3
>>>>>>
>>>>>>
>>>>>>
>>>>>> I ran into the exact same problem - unclear to me why config w/
>>>>>> NVIDIA 23. x turns on "supports fortran 2008" and then fails miserably,
>>>>>> while NVIDIA 22. x compilers did not. Also, the config adds -rdynamic that
>>>>>> is not supported by NVIDIA's
>>>>>>
>>>>>> I ran into the exact same problem - unclear to me why config w/
>>>>>> NVIDIA 23.x turns on "supports fortran 2008" and then fails miserably,
>>>>>> while NVIDIA 22.x compilers did not.
>>>>>>
>>>>>> Also, the config adds -rdynamic that is not supported by NVIDIA's
>>>>>> compilers.
>>>>>>
>>>>>> Cheers,
>>>>>> Sylvain
>>>>>> --
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Fri, May 5, 2023 at 11:14 AM Louise Spellacy via Mvapich-discuss <
>>>>>> mvapich-discuss at lists.osu.edu> wrote:
>>>>>>
>>>>>> Hi, I am building MVAPICH2 2. 3. 7 with NVHPC 23. 3 which is
>>>>>> encountering a build error. The configure script runs successfully but when
>>>>>> building with make I get the following error: make -j 8 MOD
>>>>>> src/binding/fortran/use_mpi_f08/mpi_f08_types.
>>>>>> stampNVFORTRAN-S-0081-Illegal
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>>
>>>>>>
>>>>>> I am building MVAPICH2 2.3.7 with NVHPC 23.3 which is encountering a
>>>>>> build error.
>>>>>>
>>>>>>
>>>>>>
>>>>>> The configure script runs successfully but when building with make I
>>>>>> get the following error:
>>>>>>
>>>>>>
>>>>>>
>>>>>> make -j 8
>>>>>> MOD src/binding/fortran/use_mpi_f08/mpi_f08_types.stamp
>>>>>> NVFORTRAN-S-0081-Illegal selector - KIND value must be non-negative
>>>>>> (../mvapich2-2.3.7/src/binding/fortran/use_mpi_f08/mpi_f08_types.f90: 355)
>>>>>> 0 inform, 0 warnings, 1 severes, 0 fatal for mpi_sizeof_xreal128
>>>>>> NVFORTRAN-S-0081-Illegal selector - KIND value must be non-negative
>>>>>> (../mvapich2-2.3.7/src/binding/fortran/use_mpi_f08/mpi_f08_types.f90: 385)
>>>>>> 0 inform, 0 warnings, 1 severes, 0 fatal for
>>>>>> mpi_sizeof_xcomplex128
>>>>>> make: *** [Makefile:45401:
>>>>>> src/binding/fortran/use_mpi_f08/mpi_f08_types.stamp] Error 1
>>>>>>
>>>>>>
>>>>>>
>>>>>> These lines refer to code using real128 datatypes.
>>>>>>
>>>>>>
>>>>>>
>>>>>> When I compared the status of the configure, with a previously
>>>>>> successful build with NVHPC 22.5 - the HAVE_F08_BINDING is set to 1 (with
>>>>>> 23.3) not 0 (with 22.5).
>>>>>>
>>>>>>
>>>>>>
>>>>>> I attempted to work around this by setting --enable-fortran=f77 but
>>>>>> this removes the mpif90 compiler so this wasn't a viable work-around.
>>>>>>
>>>>>>
>>>>>>
>>>>>> Is there a way of enabling f90 support without explicitly enabling
>>>>>> f08 support? The only option I could see was --enable-fortran=fc which
>>>>>> enables both.
>>>>>>
>>>>>>
>>>>>>
>>>>>> Or adding a check for real128 support to the configure script for
>>>>>> Fortran 2008 support checking?
>>>>>>
>>>>>>
>>>>>> All the best,
>>>>>>
>>>>>> Louise
>>>>>>
>>>>>> --------
>>>>>>
>>>>>> *Louise Spellacy*
>>>>>> Principal Engineer
>>>>>>
>>>>>>
>>>>>>
>>>>>> https://urldefense.com/v3/__http://www.linaroforge.com__;!!KGKeukY!xsGJkIHHQxfcYGufgfVaeaVyEMh09_nBbZQ3ndrqCb05-e4UwFCbDIa-MBpoA2sxPxfY72fE6mUH3nFuPu5nWzG1VKtkZlIqM3FBLA$
>>>>>> <https://urldefense.com/v3/__http:/www.linaroforge.com/__;!!KGKeukY!3iNLn14b76jc9LvUWa_ex-6vU1H2l0eOVemLx4Ifpj9f27VMFsusddvwZOz1dFkdJDARoTJq5Q-WqB_Eq39J_cZ2AhNmp4Tl0oJ8Iw$>
>>>>>>
>>>>>>
>>>>>>
>>>>>> [image: Image removed by sender.]
>>>>>>
>>>>>> _______________________________________________
>>>>>> Mvapich-discuss mailing list
>>>>>> Mvapich-discuss at lists.osu.edu
>>>>>> https://lists.osu.edu/mailman/listinfo/mvapich-discuss
>>>>>>
>>>>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osu.edu/pipermail/mvapich-discuss/attachments/20230517/6560152b/attachment-0006.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ~WRD2516.jpg
Type: image/jpeg
Size: 823 bytes
Desc: not available
URL: <http://lists.osu.edu/pipermail/mvapich-discuss/attachments/20230517/6560152b/attachment-0006.jpg>
More information about the Mvapich-discuss
mailing list