[QE-users] Immediate segfault in pw.x when compiled on SGI machine

[Nikolas Antolin]

Hello,

I am attempting to build/install QE using the CMake infrastructure on a SGI
8600 cluster. I have successfully built and worked on the testsuite on my
local CentOS 7 machine using CMake with minimal difficulty; when I attempt
to reproduce this build on the cluster, pw.x immediates segfaults when
attempting to run tests.

Some build specifics:
Local build is using GCC 10.2, OpenMPI 4.0.5, and IntelMKL (2021 version
provided by OneAPI).
SGI build has been attempted with GCC 10.1 and 9.2, IntelMPI 2020.2 and HPE
MPT 2.20, and IntelMKL (both 2020 and 2019 versions).
Both builds are attempting to build with QE_ENABLE_LIBXC.

IntelMPI build segfaults immediately with no output. Core stacktrace is as
follows (basically worthless as far as I can tell):
#0  0x00002af86df05817 in atan () from /lib64/libm.so.6
#1  0x00002af86888c6c8 in _dl_relocate_object () from
/lib64/ld-linux-x86-64.so.2
#2  0x00002af86888499a in dl_main () from /lib64/ld-linux-x86-64.so.2
#3  0x00002af86889800e in _dl_sysdep_start () from
/lib64/ld-linux-x86-64.so.2
#4  0x00002af868881bd1 in _dl_start () from /lib64/ld-linux-x86-64.so.2
#5  0x00002af868881148 in _start () from /lib64/ld-linux-x86-64.so.2
#6  0x0000000000000002 in ?? ()
#7  0x00007ffc6faf92fe in ?? ()
#8  0x00007ffc6faf9327 in ?? ()
#9  0x0000000000000000 in ?? ()

SGI MPT build also segfaults *nearly* immediately but points to a possible
issue with an MPI allreduce operation in FFTXlib.

Are there known issues with QE builds against MPI implementations other
than OpenMPI? Known issues with specific versions of Intel MKL?

Any suggestions of possible next steps in debugging this issue would be
greatly appreciated.

Regards,
Nik Antolin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20211013/e92b498a/attachment.html>