// includes, system
#include <stdio.h>
#include <assert.h>
#include <cutil.h>

void checkCUDAError(const char *msg);

///////////////////////////////////////////////////////////////////////////////
// Program main
///////////////////////////////////////////////////////////////////////////////
int main( int argc, char** argv) 
{
  int n, dimA;
  float *h_a;

  // pointer to gpu memory
  float *d_a, *d_b;

  //timing
  unsigned int t1= 0;
  cutCreateTimer(&t1);

  // alloc and init
  dimA = 10000000;
  h_a = (float *) malloc(dimA*sizeof(float));
  for (n=0; n<dimA; n++)
    {
      h_a[n] = (float) n;
    }

  //step 1 memory allocation on gpu for d_a and d_b
  size_t memSize = dimA*sizeof(float);
  cudaMalloc((void **) &d_a, memSize );
  cudaMalloc((void **) &d_b, memSize );

  // step 2: copy host h-a to gpu d_a
  cudaMemcpy(d_a, h_a, memSize, cudaMemcpyHostToDevice) ;

  // step 3: copy gpu d_a to gpu d_b
  cudaMemcpy(d_b, d_a, memSize,  cudaMemcpyDeviceToDevice);

  // reset cpu memory
  for (n=0; n<dimA; n++)
    {
      h_a[n] = 0.f;
    }

  // step 4: copy gpu d_b to host h_a
  cudaMemcpy(h_a, d_b, memSize, cudaMemcpyDeviceToHost );

  // 
  checkCUDAError("cudaMemcpy calls");

  // assert correct data
  for (n=0; n<dimA; n++)
    {
      assert(h_a[n] == (float) n);
    }

  // step 5 free memory
  cudaFree(d_a);
  cudaFree(d_b);


  // check CUDA error
  checkCUDAError("cudaFree");

  cutDeleteTimer(t1);

  free(h_a);

  printf("Ok!\n");

  return 0;
}


void checkCUDAError(const char *msg)
{
  cudaError_t err = cudaGetLastError();
  if( cudaSuccess != err) 
    {
      fprintf(stderr, "Cuda error: %s: %s.\n", msg, cudaGetErrorString( err) );
      exit(-1);
    }                         
}