void verif_cumuls(int ** img_in, uint64 ** d_img_1, uint64 ** d_img_x, uint64 ** d_img_x2){
//allocation memoire CPU
  uint64  img_1[H][L];
  uint64  img_x[H][L];
  uint64  img_x2[H][L];

  /*pour test comparaison*/
  uint64  img_1b[H][L];
  uint64  img_xb[H][L];
  uint64  img_x2b[H][L];
  cudaMemcpy( img_1b, *d_img_1, taille*sizeof(uint64), cudaMemcpyDeviceToHost );
  cudaMemcpy( img_xb, *d_img_x, taille*sizeof(uint64), cudaMemcpyDeviceToHost);
  cudaMemcpy( img_x2b, *d_img_x2, taille*sizeof(uint64), cudaMemcpyDeviceToHost);
 
  for (int i=0; i<H; i++)
	{
	  img_1[i][0] = 1 ;
	  img_x[i][0] = img_in[i][0] ;
	  img_x2[i][0]= img_in[i][0]*img_in[i][0] ;
	  for (int j=1; j<L; j++)
		{
		  img_1[i][j]  = img_1[i][j-1]  + 1 ;
		  img_x[i][j]  = img_x[i][j-1]  + img_in[i][j] ;
		  img_x2[i][j] = img_x2[i][j-1] + img_in[i][j]*img_in[i][j] ;
		}
	}
  
  int cpt = 0;
  int cpt_err = 0;
  for (int i=0; i<H; i++){
	for (int j=0; j<L; j++){
	  if( (img_1[i][j] !=  img_1b[i][j]) ||  (img_x[i][j] !=  img_xb[i][j]) ||  (img_x2[i][j] !=  img_x2b[i][j]) )
		{
		  if (!(j%BS)) printf("\n");
#ifdef DISPLAY_ERR_IMG_CUMUL
		  printf("(%d,%d)CPU:%lu GPU:%lu\n",i, j, img_x2[i][j], img_x2b[i][j]);
		  printf("(%d,%d):CPU=%lu  GPU=%lu\n",i,j,img_x[i][j], img_xb[i][j]);
#endif
		  cpt_err++;
		}
	  cpt++;
	}
  }
  printf("%d erreurs / %d points\n", cpt_err, cpt );
}