Then I run the command make -j32. I’m doing all these on a system with Intel Xeon Gold 5218 CPU, memory space 31.3 GiB, and OS Ubuntu 20.04.2 LTS with Open MPI installed.
The making process went smoothly, and I got ~/build/src/98_main/abinit executable after that. But when I tried to run:
abinit -b
the terminal returned:
Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
Backtrace for this error:
#0 0x14acc82d677d in ???
#1 0x14acc82d5993 in ???
#2 0x14acc84b341f in ???
#3 0x14acc82d5e9b in ???
#4 0x14acc8454a8d in ???
#5 0x14acc84425fb in ???
#6 0x55a54ac8c851 in __m_xmpi_MOD_xmpi_init
at ../../../../../shared/common/src/12_hide_mpi/m_xmpi.F90:763
#7 0x55a5497afaa6 in abinit
at ../../../src/98_main/abinit.F90:202
#8 0x55a5497ba9da in main
at ../../../src/98_main/abinit.F90:88
Segmentation fault (core dumped)
Trying to run it in parallel by using:
mpirun -np 4 abinit -b
also returned a similar error, all four processes were killed by signal 11 (segmentation fault).
I’ve also installed mkl (libmkl-full-dev) as the default alternative for blas and lapack, could this be the problem?
And speaking of the library, I remember another issue occurred during the installation of hdf5-dev via apt. The terminal reported a dependency error about libjpeg-dev that apt cannot solve, so I tried to install it first, which led to a series of dependency errors, at the end of that series is a similar issue like this. I tried to install gfortran-7 as that post suggested, then finished the installation of hdf5-dev without further issues. Could this be related?
In the output of the ldd command, I notice that there are 2 versions of the libgfortran.so library : libgfortran.so.4 and libgfortran.so.5 : it’s the problem… but how it happened… during an OS upgrade?
I started from scratch :
-minimum installation of Ubuntu 20.04.4 LTS
-installation of these packages :
Thanks very much! Seems indeed an env problem. This system has too many messy packages to fix on my own, but I’m the sole user of it now, so I just reinstalled everything and started from scratch, and problem solved I think? At least test run was finished cleanly.
Thanks again, I know this (restarting everything in a clean environment) is not a doable solution for everyone who might also be reading this thread, but the problem is very much like the different versions of the same library.