cuda-gdb-blog

This is the source code to reproduce the experiments shown in the blog

Requirements: CUDA (Full CUDA Toolkit) cuda-gdb cmake gcc g++

Steps

1. Compile the source with\n nvcc -g -G test__memory_err.cu -o test__memory_err or sh build.sh

2. Execute program ./test__memory_err

3. Execute executible with cuda-gdb cuda-gdb test__memory_err

4. Inside cuda-gdb execute the full program run

5. Isolate line by using autostep autostep 5 for 7 run (Press y when asked whether it should restart execution)

6. Focus on segment In my case with the following error statement [Current focus set to CUDA kernel 0, grid 1, block (4,0,0), thread (0,0,0), device 0, sm 8, warp 0, lane 0] Autostep precisely caught exception at test__memory_err.cu:10 (0x555555b2ac10)

   cuda device 0 sm 8 warp 0 lane 0

7. Check focus cuda device sm warp lane block thread

8. Get lane information info cuda lanes