Missing referenced architecture template linux-gnu-openmpi-nogpu

[Sam-Bieberich]

The top-level Makefile comments state that a template architecture file named linux-gnu-openmpi-nogpu is provided as an example:

# A templete "linux-gnu-openmpi-nogpu" is given as an example architecture file.

However, this file does not appear to exist in the architecture/ directory. The available architecture files seem more system-specific, and I can't tell if any serve as a generic GNU + OpenMPI + CPU-only template.

Could you clarify whether this file was removed/renamed, or if another architecture file is intended to serve this role? I am attempting to make a containerized installation method to allow for benchmarking across various architectures, and have had trouble getting the installation setup to work (using an NVIDIA DGX Spark for reference).

Thank you!

[vishnu2709]

That specific architecture file indeed doesn't exist, but there are others in the architecture directory that would be functionally similar, like ubuntu-gnu-nogpu or ubuntu-gnu-openmpi. I would recommend using one of these.

Additionally, if you are interested in a containerized installation, we have a docker file present in the main directory.

If you're still having trouble, you may have to make modifications to an existing architecture file to work for your system, or opening an issue with the specific error messages you are receiving.

[Sam-Bieberich]

I have tried to use the Dockerfile, but I get an error saying that DefaultParameters.h is missing. I thought installing from source would be better, since I know that it has been tested recently, but I have had poor luck trying that as well.

Related to the container, is there a new path that needs to be defined for the .h file that's missing? It is referenced here in the CMake.

Thanks again.

[wangy2014]

The Docker file is somewhat outdated. It recommended to build the executables from the source code. What error messages do you get when trying to install from the source code?

…

On Tue, Jan 20, 2026 at 13:56 Sam B ***@***.***> wrote: *Sam-Bieberich* left a comment (mstsuite/MuST#74) <#74 (comment)> I have tried to use the Dockerfile, but I get an error saying that DefaultParameters.h is missing. I thought installing from source would be better, since I know that it has been tested recently, but I have had poor luck trying that as well. Related to the container, is there a new path that needs to be defined for the .h file that's missing? It is referenced here <https://github.com/mstsuite/MuST/blob/01c191144d6444f756c0dac952158f9cda7ff993/MST/src/CMakeLists.txt#L31> in the CMake. Thanks again. — Reply to this email directly, view it on GitHub <#74 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB4HQPHXYD7PSVJMJIWZUFL4HZ267AVCNFSM6AAAAACSJPPFN2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTONZUGQ2TQMRSGY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Sam-Bieberich]

I have had a few dependency and installation errors related to NVHPC, but I installed that from source too. Now, the latest is that the C compiler fails

checking whether the C compiler works... no and it leads me to MuST/external/libxc/LibXC and says the C compiler can't create executables. When I review the config.log file, the error says that `lscalapack-mpich' cannot be found, though I have installed it and all of its dependencies. This is using the ubuntu-gnu-nogpu with no modification to the highest level Makefile.

[Sam-Bieberich]

Configure command:

./configure --prefix=/tmp/MuST_build/external/libxc/LibXC \
            CC=mpicc CFLAGS=-O3 \
            FC=mpif90 FFLAGS=-O3 -I.

Environment:

CC=mpicc
FC=mpif90
CFLAGS=-O3
FFLAGS=-O3
LIBS='-L/usr/lib/x86_64-linux-gnu -lblas -llapack -lscalapack-mpich'

Compiler version:

$ mpicc --version
nvc 25.11-0 linuxarm64 target on aarch64 Linux -tp gb10
Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.

Error output during configure:

checking whether the C compiler works... no
configure: error: C compiler cannot create executables
/usr/bin/ld: cannot find -lscalapack-mpich: No such file or directory
collect2: error: ld returned 1 exit status

Failed test program (conftest.c):

[wangy2014]

Since you are using nvhpc, I would suggest you take a look at vista-nvhpc-cpu or bridges2-nvhpc-cpu and modify one of these two architecture files. Nevertheless, you will need to have a MPI library installed with your nvhpc compiler.

…

On Tue, Jan 20, 2026 at 14:52 Sam B ***@***.***> wrote: *Sam-Bieberich* left a comment (mstsuite/MuST#74) <#74 (comment)> I have had a few dependency and installation errors related to NVHPC, but I installed that from source too. Now, the latest is that the C compiler fails checking whether the C compiler works... no and it leads me to MuST/external/libxc/LibXC and says the C compiler can't create executables. When I review the config.log file, the error says that `lscalapack-mpich' cannot be found, though I have installed it and all of its dependencies. This is using the ubuntu-gnu-nogpu with no modification to the highest level Makefile. — Reply to this email directly, view it on GitHub <#74 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB4HQPD6LYONJPYPHGGQRV34H2BP7AVCNFSM6AAAAACSJPPFN2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTONZUGY3DINJXGE> . You are receiving this because you commented.Message ID: ***@***.***>

[wangy2014]

This is a linking error. It shows that you do have a scalapack built around mpich library available in your system. Actually, if you use nvhpc, you do not need to link with a separate scalapack library since nvhpc package comes with its own Lapack/Scalapack libraries.

…

On Tue, Jan 20, 2026 at 15:04 Sam B ***@***.***> wrote: *Sam-Bieberich* left a comment (mstsuite/MuST#74) <#74 (comment)> Configure command: ./configure --prefix=/tmp/MuST_build/external/libxc/LibXC \ CC=mpicc CFLAGS=-O3 \ FC=mpif90 FFLAGS=-O3 -I. Environment: CC=mpicc FC=mpif90 CFLAGS=-O3 FFLAGS=-O3 LIBS='-L/usr/lib/x86_64-linux-gnu -lblas -llapack -lscalapack-mpich' Compiler version: $ mpicc --version nvc 25.11-0 linuxarm64 target on aarch64 Linux -tp gb10 Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved. Error output during configure: checking whether the C compiler works... no configure: error: C compiler cannot create executables /usr/bin/ld: cannot find -lscalapack-mpich: No such file or directory collect2: error: ld returned 1 exit status Failed test program (conftest.c): — Reply to this email directly, view it on GitHub <#74 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB4HQPCOQ56VY4N3IUZHDRL4H2C3VAVCNFSM6AAAAACSJPPFN2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTONZUG4YDOMJSG4> . You are receiving this because you commented.Message ID: ***@***.***>

[Sam-Bieberich]

When I've worked with the vista arch file locally, I get Fortran compiler errors:

configure:18202: ./conftest
./conftest: symbol lookup error: /opt/nvidia/hpc_sdk/Linux_aarch64/25.11/comm_libs/13.0/hpcx/hpcx-2.25.1/ompi/lib/libmpi_mpifh.so.40: undefined symbol: mpi_conversion_fn_null_
configure:18202: $? = 127
configure: program exited with status 127
configure: failed program was:
|       program main
|
|   integer    :: i
|   integer(8) :: i8
|
|   i8 = huge(i)
|
|   select case(i8)
|   case(127_8);                 i = 1
|   case(32767_8);               i = 2
|   case(2147483647_8);          i = 4
|   case(9223372036854775807_8); i = 8
|   case default; write(*,'(a,i20)') "unrecognized size ", i8; stop 999
|   end select
|
|   open(1, file='conftest.out')
|   write(1,'(i1)') i
|   close(1)
|
|       end
configure:18205: error: in `/tmp/MuST_build/external/libxc/LibXC':
configure:18207: error: f90 program to find the size of a Fortran integer failed

[wangy2014]

This is strange! You may copy that piece of test code, from “program main” until “end”, to a separate file and compile it yourself to determine if there is a problem with your compiler.

…

On Tue, Jan 20, 2026 at 15:20 Sam B ***@***.***> wrote: *Sam-Bieberich* left a comment (mstsuite/MuST#74) <#74 (comment)> When I've worked with the vista arch file locally, I get Fortran compiler errors: configure:18202: ./conftest ./conftest: symbol lookup error: /opt/nvidia/hpc_sdk/Linux_aarch64/25.11/comm_libs/13.0/hpcx/hpcx-2.25.1/ompi/lib/libmpi_mpifh.so.40: undefined symbol: mpi_conversion_fn_null_ configure:18202: $? = 127 configure: program exited with status 127 configure: failed program was: | program main | | integer :: i | integer(8) :: i8 | | i8 = huge(i) | | select case(i8) | case(127_8); i = 1 | case(32767_8); i = 2 | case(2147483647_8); i = 4 | case(9223372036854775807_8); i = 8 | case default; write(*,'(a,i20)') "unrecognized size ", i8; stop 999 | end select | | open(1, file='conftest.out') | write(1,'(i1)') i | close(1) | | end configure:18205: error: in `/tmp/MuST_build/external/libxc/LibXC': configure:18207: error: f90 program to find the size of a Fortran integer failed — Reply to this email directly, view it on GitHub <#74 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB4HQPF6ZCXIDUGHOCDSEDT4H2EZNAVCNFSM6AAAAACSJPPFN2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTONZUG43TCNJRGA> . You are receiving this because you commented.Message ID: ***@***.***>

[Sam-Bieberich]

Somehow it works in the terminal directly, but not in that configuration. Maybe a library path is missing somewhere?

[wangy2014]

You may try using architecture file ubuntu-gnu-openmpi which assumes that 1) you use gcc/gfortran compiler and openmpi built for gcc/gfortran, and 2) have libxc and scalapack installed under /opt/packages/. I will create an architecture file for using nvhpc on ubuntu system and make it available for you to download from github. Stay tuned.

…

On Tue, Jan 20, 2026 at 4:06 PM Sam B ***@***.***> wrote: *Sam-Bieberich* left a comment (mstsuite/MuST#74) <#74 (comment)> Somehow it works in the terminal directly, but not in that configuration. Maybe a library path is missing somewhere? — Reply to this email directly, view it on GitHub <#74 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB4HQPHSANG2V4WLJP43ERT4H2KD7AVCNFSM6AAAAACSJPPFN2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTONZUHE2DMNJZGU> . You are receiving this because you commented.Message ID: ***@***.***>

[Sam-Bieberich]

I got an error yesterday afternoon that implied something wrong with my nvhpc source installation, so I am redoing that again today. Once I get something working, I will let you know. Thanks

[wangy2014]

I just uploaded ubuntu-nvhpc-cpu and ubuntu-nvhpc-gpu to the github. They have been tested on a ubuntu system with nvhpc-25.5 with openmpi in hpcx-2.22.1. The LibXC library was pre-installed though.

…

On Wed, Jan 21, 2026 at 11:31 AM Sam B ***@***.***> wrote: *Sam-Bieberich* left a comment (mstsuite/MuST#74) <#74 (comment)> I got an error yesterday afternoon that implied something wrong with my nvhpc source installation, so I am redoing that again today. Once I get something working, I will let you know. Thanks — Reply to this email directly, view it on GitHub <#74 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB4HQPEHESETGW7ET7BZLIT4H6SUNAVCNFSM6AAAAACSJPPFN2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTONZZGQZTKMZQGU> . You are receiving this because you commented.Message ID: ***@***.***>

[Sam-Bieberich]

I believe I have it working with vista-nvhpc-cpu. I had to relocate everything to /tmp and modify some of the paths to match the architecture file to get it to work. I think the NVHPC installation I had was also messed up the first time; there were several non-fatal warnings when I originally installed it from source that I ignored, but the second time, there were no warnings.
Thank you for your help! I am planning to benchmark on more CPUs in the future, and will use the new architecture files in that process.