I've encountered some wired problem that might be mixing Intel MPI 2017, MKL 2017 and glibc 2.26. A fortran program using `mpiifort -mkl=sequential -g -traceback -lstdc++`. It can run when executed directly, but fails with SIGSEGV immediately with mass of Unknown source and line on backtrace, even with `-np 1`.
I guess attaching gdb can help, so I ran `mpirun -np 1 -gdb a.out`, but the mpigdb just become more wired: (bold for my input)
~:server> mpirun -np 1 -gdb sleep 5 mpigdb: np = 1 mpigdb: attaching to 15762 sleep w003  (mpigdb) <b>r</b>  Undefined command: "". Try "help".  (mpigdb) <b>help</b>  Undefined command: "". Try "help".  (mpigdb) <b>aaa</b>  Undefined command: "". Try "help".  (mpigdb)
It seems that mpigdb just ignores all my input. Tried with several Intel Parallel Studio versions, ranging from 2016 to 2020 update 1, all of which suffers the same problem.
Any idea on it? OS is openSUSE 15.2, command executed from ssh session, bash as shell.
May I know the reason you are using angular brackets(tags) while giving gdb commands.
You were saying they are bold tags, have you inserted them manually?
Clearly, gdb is not recognising the tags and giving the error Undefined command.
Can you try without using them?
Sorry, I thought you are providing the tags as input to gdb.
I have tried with various MPI versions and gdb is working fine for me, but I haven't tested it on OpenSUSE.
I am transferring this query to internal team.
I'm sorry that I should mention that I've tested on an CentOS 7 machine, which behaves as expected. However I have totally no idea how to debug it.