I have implemented segment tree on CUDA. The code compiles and runs without any problem. Now I am extending that to implement some algorithm using segment tree and unfortunately now I need cuda-debugger as a kernel has some logical/programmatic error. So, consequently, I have to compile my code with -g -G. The extended code did not compile with the debugger and it halted during compilation. When I press ctrl+C, it gives the message - nvcc error : ‘ptxas’ died due to signal 2. I tried with the working code, that also gives the same error. I have used cuda-GDB quite often, but never faced such problem. I am using cuda 3.2, gcc 4.4.5 and working on a cc 2.1 gpu. I would be thankful for any suggestion.
Try compile with make -j 2 or make -j 1. ptxas died commonly because os run out of swap space. Reduce num of parallel build is the easiest work around.
1 Like