Finding a seg fault in a small code

Question

I am unable to get the source of segmentation fault in this short cuda code. I am using it to test the sorting speed of the Thrust library versus the STL library for sorting integers. I am passing the size of the array of doubles to be sorted as a command line argument.

Here is the code

inline void check_cuda_error(char *message)

    {
      cudaThreadSynchronize();
      cudaError_t error = cudaGetLastError();
      if(error != cudaSuccess)
      {
        printf("CUDA error after %s: %s\n", message, cudaGetErrorString(error));
      }
    }



            int main(int argc, char *argv[])
        {
          int  N = atoi(argv[1]);
          double* h = new double[N];
          for (int i = 0; i < N; ++i)
            {
              h[i] = (double)rand()/RAND_MAX; //std::cout << h[i] << " " ;
            }

          clock_t start , stop;

          std::cout << std::endl;

          // Start timing
          start = clock();
          std::sort(h, h+N);
          stop  = clock();  
          std::cout << "Host sorting took " << (stop - start) /(double)CLOCKS_PER_SEC << std::endl ; 


          // Start the GPU work. Initialize to random numbers again.
          for (int i = 0; i < N; ++i)
            {
              h[i] = (double)rand()/RAND_MAX; //std::cout << h[i] << " " ;
            }
          double* d = 0; 
          const size_t num_bytes = N * sizeof( double ); 
          cudaMalloc((void**)&d, num_bytes);
          check_cuda_error("Memory Allocation"); 

          cudaMemcpy(d ,h , N * sizeof(double), cudaMemcpyHostToDevice); // Transfer data
          thrust::sort( d, d+ N ) ;
            return 0;
        }

I get the following errors

[BeamerLatex/Farber]$ nvcc -arch=sm_20 sortcompare.cu  ; ./a.out 16777216

Host sorting took 3.77
[1]    4661 segmentation fault  ./a.out 16777216
[BeamerLatex/Farber]$ 

Answer 1

Seems that you cannot run thrust::sort on a raw pointer, you need to cast it to device_ptr first, ie.:

thrust::device_ptr< double > dv = thrust::device_pointer_cast(d);
thrust::sort( dv, dv+ NN ) ;

this works fine for me.