One way of debugging segmentation fault is using GDB. It is such a deep learning curve that no one ever claims he or she fully understands the complete usages. Here is just my experience of debugging a C++ code with GDB.


The most important compilation flag is -g: this enables the program to store information that can be used when backtracing. Nowadays often you don’t need to turn off many compilation flags. For safety, use -O0; if you want to give it a try, use -O2.


If you set ulimit -c unlimited, the program is allowed to use all the available resources in the current platform. Therefore Core dump files can be generated at runtime. Core files are created when a program encounters a run-time error. It is an image of the memory used by the program, and debuggers such as gdb can access it to find out the state of the program at the time of the error.

Then we can enter GDB REPL by

gdb executable core

and type bt or backtrace to see the lines of codes that crash. Note that line numbers display is a magic, and I rarely get the luck to see it. However, there is a higher chance that you can see the variable or function name instead of question marks or hexacodes.

Parallel Debugging

At least Intel MPI libraries allow

mpirun <mpi options> -gdb program.exe


It is relatively easy to deal with memory errors in simple codes. However, for more complicated codes, you can easily get lost in the data structures and pointers. Read the code if you can to learn the basic ideas. Then use a development tool such as VTune to do a systematic check if you have time.

If you run gdb directly at runtime, you may encounter many false-positives in memory issues, and the real problem will be buried underneath. This is not ideal for debugging.