sample2.cu

#include <iostream>
#include <cstdlib>
#include <ctime>

#include <thrust/host_vector.h>
#include <thrust/device_vector.h>
#include <thrust/copy.h>
#include <thrust/functional.h>
#include <thrust/transform.h>

const int SIZE=2048*512;

int main(int argc, char**argv){
  clock_t start,stop;
  printf("GPU:\n");
  int i;
  const int ishow=16;
  int n;
  const int nite=1000;
  
  cudaSetDevice(0);
  /* variables in device */
  thrust::device_vector<float> d_InA(SIZE);
  thrust::device_vector<float> d_InB(SIZE);
  thrust::device_vector<float> d_Out(SIZE);
  /* variables in host */
  thrust::  host_vector<float> h_InA(SIZE);
  thrust::  host_vector<float> h_InB(SIZE);
  thrust::  host_vector<float> h_Out(SIZE);
  
  srand(0);
  /* initialize */
  for(i=0;i<SIZE;i++) h_InA[i] = (float)(rand()%10)/10.0f;
  for(i=0;i<SIZE;i++) h_InB[i] = (float)(rand()%10)/10.0f; 

  /* confirm */
  printf("InA: "); for(i=0;i<ishow;i++) printf(" %.2f",h_InA[i]); printf("\n");
  printf("InB: "); for(i=0;i<ishow;i++) printf(" %.2f",h_InB[i]); printf("\n");
  
  /* transfer from host to device */
  thrust::copy(h_InA.begin(),h_InA.end(),d_InA.begin());/* d_InA = h_InA */ 
  thrust::copy(h_InB.begin(),h_InB.end(),d_InB.begin());/* d_InB = h_InB */

  cudaDeviceSynchronize();

  /* call kernel functions, specify grid and block as <<< grid, block >>> */
 
  start=clock();
  for(n=0;n<nite;n++){
  thrust::transform(d_InA.begin(), d_InA.end(), d_InB.begin(), d_Out.begin(), thrust::plus<float>());
  }
  cudaDeviceSynchronize();
  stop=clock();
  
  /* transfer from device to host */
  thrust::copy(d_Out.begin(),d_Out.end(),h_Out.begin());
 
  /* confirm */
  printf("Out: "); for(i=0;i<ishow;i++) printf(" %.2f",h_Out[i]); printf("\n");
  
  printf("Time: %.2f [ms]\n",(double)(stop-start));
  return 0;
}