int main(int argc, char *argv[])
{
int tmax = TMAX;
int nx = NX;
int ny = NY;
POLYBENCH_1D_ARRAY_DECL(_fict_,DATA_TYPE,TMAX,TMAX);
POLYBENCH_2D_ARRAY_DECL(ex,DATA_TYPE,NX,NY,nx,ny);
POLYBENCH_2D_ARRAY_DECL(ey,DATA_TYPE,NX,NY,nx,ny);
POLYBENCH_2D_ARRAY_DECL(hz,DATA_TYPE,NX,NY,nx,ny);
POLYBENCH_2D_ARRAY_DECL(hz_outputFromGpu,DATA_TYPE,NX,NY,nx,ny);
init_arrays(tmax, nx, ny, POLYBENCH_ARRAY(_fict_), POLYBENCH_ARRAY(ex), POLYBENCH_ARRAY(ey), POLYBENCH_ARRAY(hz));
read_cl_file();
cl_initialization();
cl_mem_init(POLYBENCH_ARRAY(_fict_), POLYBENCH_ARRAY(ex), POLYBENCH_ARRAY(ey), POLYBENCH_ARRAY(hz));
cl_load_prog();
cl_launch_kernel(tmax, nx, ny);
errcode = clEnqueueReadBuffer(clCommandQue, hz_mem_obj, CL_TRUE, 0, NX * NY * sizeof(DATA_TYPE), POLYBENCH_ARRAY(hz_outputFromGpu), 0, NULL, NULL);
if(errcode != CL_SUCCESS) printf("Error in reading GPU mem\n");
#ifdef RUN_ON_CPU
/* Start timer. */
polybench_start_instruments;
runFdtd(tmax, nx, ny, POLYBENCH_ARRAY(_fict_), POLYBENCH_ARRAY(ex), POLYBENCH_ARRAY(ey), POLYBENCH_ARRAY(hz));
/* Stop and print timer. */
printf("CPU Time in seconds:\n");
polybench_stop_instruments;
polybench_print_instruments;
compareResults(nx, ny, POLYBENCH_ARRAY(hz), POLYBENCH_ARRAY(hz_outputFromGpu));
#else //prevent dead code elimination
polybench_prevent_dce(print_array(nx, ny, POLYBENCH_ARRAY(hz_outputFromGpu)));
#endif //RUN_ON_CPU
POLYBENCH_FREE_ARRAY(_fict_);
POLYBENCH_FREE_ARRAY(ex);
POLYBENCH_FREE_ARRAY(ey);
POLYBENCH_FREE_ARRAY(hz);
POLYBENCH_FREE_ARRAY(hz_outputFromGpu);
cl_clean_up();
return 0;
}
int main(void)
{
double t_start, t_end;
DATA_TYPE* _fict_;
DATA_TYPE* ex;
DATA_TYPE* ey;
DATA_TYPE* hz;
DATA_TYPE* hz_outputFromGpu;
_fict_ = (DATA_TYPE*)malloc(TMAX*sizeof(DATA_TYPE));
ex = (DATA_TYPE*)malloc(NX*(NY+1)*sizeof(DATA_TYPE));
ey = (DATA_TYPE*)malloc((NX+1)*NY*sizeof(DATA_TYPE));
hz = (DATA_TYPE*)malloc(NX*NY*sizeof(DATA_TYPE));
hz_outputFromGpu = (DATA_TYPE*)malloc(NX*NY*sizeof(DATA_TYPE));
int i;
init_arrays(_fict_, ex, ey, hz);
read_cl_file();
cl_initialization();
cl_mem_init(_fict_, ex, ey, hz);
cl_load_prog();
cl_launch_kernel();
errcode = clEnqueueReadBuffer(clCommandQue, hz_mem_obj, CL_TRUE, 0, NX * NY * sizeof(DATA_TYPE), hz_outputFromGpu, 0, NULL, NULL);
if(errcode != CL_SUCCESS) printf("Error in reading GPU mem\n");
t_start = rtclock();
runFdtd(_fict_, ex, ey, hz);
t_end = rtclock();
fprintf(stdout, "CPU Runtime: %0.6lfs\n", t_end - t_start);
compareResults(hz, hz_outputFromGpu);
cl_clean_up();
free(_fict_);
free(ex);
free(ey);
free(hz);
free(hz_outputFromGpu);
return 0;
}
