Debugging

Accelerated Computing CUDA CUDA Programming and Performance
1

how can i debuging a progrem runs on gpu?

2

There are multiple ways to debug in CUDA.

  1. run your application with cuda-memcheck. https://developer.nvidia.com/cuda-memcheck

  2. check the return status of CUDA API. http://stackoverflow.com/questions/14038589/what-is-the-canonical-way-to-check-for-errors-using-the-cuda-runtime-api

  3. you could also print out the variables in cuda kernel to see the values. e.g.

#include <stdio.h>
#include <assert.h>
#include <cuda.h>
#include <cuda_runtime.h>

__global__ void test(){
    printf("Hi Cuda World");
}

int main( int argc, char** argv )
{
    test<<<1,1>>>();
    cudaDeviceSynchronize();
    return 0;
}
3

There are debuggers available.

On linux you can use cuda-gdb or the debugger personality built into nsight eclipse edition (both are part of the CUDA toolkit)

[url]http://docs.nvidia.com/cuda/index.html#tools[/url]

On Windows, the nsight visual studio edition plug in for MS VS provides device code debugging support.

[url]https://developer.nvidia.com/nvidia-nsight-visual-studio-edition[/url]