NVidia CUDA Trace or Debug Function

Question

Is there a trace or debug function that I can place in a CUDA kernel? I'm writing a program on Windows 7, VS2010 and I just discovered that to use NSIGHT Monitor I have to have 2 GPUs. I'm on a laptop unfortunately so this isn't really an option. I figured that I would fall back to tried and true debug/trace functions. Something akin to OutputDebugString. I don't see if one exists though.

Any help is appreciated. Thanks,

mj

http://stackoverflow.com/questions/2173771/printf-inside-cuda-global-function — Roger Dahl, May 10 '12 at 13:13

score 1 · Accepted Answer · answered May 11 '12 at 02:23

1

Parallel Nsight 2.2 supports local single GPU debugging.

CUDA compute capability >= 2.0 support printf from device code.

answered May 11 '12 at 02:23

Greg Smith

11,007
2
36
37

On another note, in Visual Studio, how is it that you jump between GPU threads while debugging? I created one block of size 8 to run through my data and I can see the threadIdx.x where the id = 0, but I can't see anything else. – mj_ May 13 '12 at 02:53

NVidia CUDA Trace or Debug Function

1 Answers1