Replying to myself with an additional request for help. (in case anyone has any
insights into the issues below)
Unfortunately, despite some reading I was unsuccessful at building OpenMPI
4.1.1 with support for current Intel Fortran 21.3 on current MacOS Big Sur 11.4
.
1) The issue I mentioned earlier could be overcome by using
../configure --with-hwloc=internal --with-libevent=internal
--prefix=/usr/local/openmpi-4.1.1 FC=ifort
Somewhat tricky, but findable (as I said earlier).
2) Nevertheless, the resulting mpif90 executable does not work (command line
sequence broken up for clarity):
blum:/Users/blum/codes> which ifort
/opt/intel/oneapi/compiler/2021.3.0/mac/bin/intel64/ifort
blum:/Users/blum/codes> ifort test.f90
blum:/Users/blum/codes> mpif90 --version
ifort (IFORT) 2021.3.0 20210609
Copyright (C) 1985-2021 Intel Corporation. All rights reserved.
blum:/Users/blum/codes> mpif90 test.f90
ld: warning: dylib (/usr/local/openmpi-4.1.1/lib/libmpi_mpifh.dylib) was built
for newer macOS version (11.0) than being linked (10.16)
ld: warning: dylib (/usr/local/openmpi-4.1.1/lib/libmpi.dylib) was built for
newer macOS version (11.0) than being linked (10.16)
ld: warning: dylib (/usr/local/openmpi-4.1.1/lib/libopen-rte.40.dylib) was
built for newer macOS version (11.0) than being linked (10.16)
ld: warning: dylib (/usr/local/openmpi-4.1.1/lib/libopen-orted-mpir.dylib) was
built for newer macOS version (11.0) than being linked (10.16)
ld: warning: dylib (/usr/local/openmpi-4.1.1/lib/libopen-pal.40.dylib) was
built for newer macOS version (11.0) than being linked (10.16)
Undefined symbols for architecture x86_64:
"_ompi_buffer_detach_f08", referenced from:
import-atom in libmpi_usempif08.dylib
ld: symbol(s) not found for architecture x86_64
blum:/Users/blum/codes> cat test.f90
program test
write (6,*) "Hello"
end program test
***
This issue is reminiscent of issue
https://github.com/open-mpi/ompi/issues/7615
which I tried to address by modifying configure, as described here:
https://ntq1982.github.io/files/20200621.html
However, that doesn’t seem to be all - see the above build failure. Unless I
misapplied the patch to configure (could be, although I’m not sure how),
something else is wrong.
As an aside, I also do not quite understand the ld warnings above. (This
Macbook never had 10.16 so it’s unlikely to be some leftover library.)
Does anyone know of a way forward here? On the bright side, the software
combination I’m using is fairly current so it might be worth enabling it.
Additionally, Intel Fortran 21.3 now no longer requires a paid license.
Thanks
Best wishes
Volker
> On Jul 18, 2021, at 4:29 PM, Volker Blum via users <users@lists.open-mpi.org>
> wrote:
>
> Hi,
>
> This is a quick note regarding FAQ 41/42 on building OpenMPI:
>
> https://urldefense.com/v3/__https://www.open-mpi.org/faq/?category=building*libevent-or-hwloc-errors-when-linking-fortran__;Iw!!OToaGQ!-B4Kna7yfhCDcqLh2WYzWffcD7deQ22z3KSuB55LhRdl4HL_t-rkm7g8h4hUiygt-Jc$
>
>
> I was glad to find the information. However, since this is a somewhat
> predictable issue … would it make sense to address this in configure instead?
>
> (i.e., just use the internal versions of hwloc and libevent as default if FC
> and CC are known to come from different compiler suites? e.g., gcc and ifort?)
>
> Thank you for keeping OpenMPI running!
>
> Best wishes
> Volker
>
>
Volker Blum
Associate Professor
Ab Initio Materials Simulations
Thomas Lord Department of Mechanical Engineering and Materials Science
Duke University
https://aims.pratt.duke.edu
volker.b...@duke.edu
Twitter: Aimsduke
Office: 4308 Chesterfield Building
