我如何调试 C++ MPI 代码？

Question

My Boost MPI code compiles with no problem:

But when running, I have issues. I am not sure which line in my code causes this problem. What could I do?

[dhcp-18-189-36-128:80590] *** Process received signal ***
[dhcp-18-189-36-128:80590] Signal: Floating point exception: 8 (8)
[dhcp-18-189-36-128:80590] Signal code: Integer divide-by-zero (7)
[dhcp-18-189-36-128:80590] Failing at address: 0x10e4cab9f
[dhcp-18-189-36-128:80590] [ 0] 0   libsystem_platform.dylib            0x00007fff8d0185aa _sigtramp + 26
[dhcp-18-189-36-128:80590] *** End of error message ***

Answer 1

My recommendation would be to put unit tests around your code to verify it functions correctly in a context smaller than your entire application. If you are new to unit testing and test-driven development, you can look at my tutorial on test-driven development from C++ Now! 2014 .

To elaborate further: when faced with a bug, I try to write a failing unit test that demonstrates the bug. It's hard to make a specific recommendation of what test to write without seeing any code, but I suggest you start by running your code in the debugger and setting the debugger to treat any floating point exception signal as a breakpoint. When the breakpoint is hit due to the signal, examine the call stack at that point. Then you will know what code is causing the signal. Then you can write a unit test that calls that code with the same arguments. Your unit test should now fail by reproducing the problem and causing the same divide-by-zero signal. Then you can change your implementation and re-run the test to verify that the signal is no longer caused.