cl_int
clSetCommandQueueProperty_test(cl_command_queue command_queue,
cl_command_queue_properties properties,
cl_bool enable,
cl_command_queue_properties * old_properties)
{
cl_int status;
printf("clSetCommandQueueProperty_test: properties==%ld enable==%d\n", (long) properties, enable);
status = clSetCommandQueueProperty(command_queue, properties, enable, old_properties);
return status;
}
int main(int argc, char **argv)
#endif
{
int i, niter, step;
double mflops, t, tmax;
logical verified;
char class;
double tsum[t_last+2], t1[t_last+2],
tming[t_last+2], tmaxg[t_last+2];
char *t_recs[t_last+2] = {
"total", "rhs", "xsolve", "ysolve", "zsolve",
"bpack", "exch", "xcomm", "ycomm", "zcomm",
" totcomp", " totcomm" };
//---------------------------------------------------------------------
// Root node reads input file (if it exists) else takes
// defaults from parameters
//---------------------------------------------------------------------
printf("\n\n NAS Parallel Benchmarks (NPB3.3-OCL-MD) - SP Benchmark\n\n");
FILE *fp;
fp = fopen("timer.flag", "r");
timeron = false;
if (fp != NULL) {
timeron = true;
fclose(fp);
}
if ((fp = fopen("inputsp.data", "r")) != NULL) {
int result;
printf(" Reading from input file inputsp.data\n");
result = fscanf(fp, "%d", &niter);
while (fgetc(fp) != '\n');
result = fscanf(fp, "%*f");
while (fgetc(fp) != '\n');
result = fscanf(fp, "%d%d%d", &grid_points[0], &grid_points[1],
&grid_points[2]);
fclose(fp);
} else {
printf(" No input file inputsp.data. Using compiled defaults\n");
niter = NITER_DEFAULT;
grid_points[0] = PROBLEM_SIZE;
grid_points[1] = PROBLEM_SIZE;
grid_points[2] = PROBLEM_SIZE;
}
setup_opencl(argc, argv);
printf(" Size: %4dx%4dx%4d\n",
grid_points[0], grid_points[1], grid_points[2]);
printf(" Iterations: %4d", niter);
if (num_devices != MAXCELLS*MAXCELLS)
printf(" WARNING: compiled for %5d devices \n", MAXCELLS*MAXCELLS);
printf(" Number of active devices: %5d\n\n", num_devices);
make_set();
for (i = 0; i < t_last; i++) {
timer_clear(i);
}
set_constants();
initialize();
lhsinit();
exact_rhs();
compute_buffer_size(5);
set_kernel_args();
//---------------------------------------------------------------------
// do one time step to touch all code, and reinitialize
//---------------------------------------------------------------------
#ifdef MINIMD_SNUCL_OPTIMIZATIONS
// set cmd queue property
for(i = 0; i < num_devices; i++) {
clSetCommandQueueProperty(cmd_queue[i],
CL_QUEUE_AUTO_DEVICE_SELECTION |
//CL_QUEUE_ITERATIVE |
CL_QUEUE_COMPUTE_INTENSIVE,
true,
NULL);
}
#endif
adi();
#ifdef MINIMD_SNUCL_OPTIMIZATIONS
for(i = 0; i < num_devices; i++) {
clSetCommandQueueProperty(cmd_queue[i],
0,
true,
NULL);
}
#endif
initialize();
//---------------------------------------------------------------------
// Synchronize before placing time stamp
//---------------------------------------------------------------------
for (i = 0; i < t_last; i++) {
timer_clear(i);
}
timer_clear(0);
timer_start(0);
for (step = 1; step <= niter; step++) {
if ((step % 20) == 0 || step == 1) {
printf(" Time step %4d\n", step);
}
adi();
}
timer_stop(0);
t = timer_read(0);
verify(niter, &class, &verified);
tmax = t;
if( tmax != 0.0 ) {
mflops = (881.174*(double)( PROBLEM_SIZE*PROBLEM_SIZE*PROBLEM_SIZE )
-4683.91*(double)( PROBLEM_SIZE*PROBLEM_SIZE )
+11484.5*(double)( PROBLEM_SIZE )
-19272.4) * (double)( niter ) / (tmax*1000000.0);
} else {
mflops = 0.0;
}
c_print_results("SP", class, grid_points[0],
grid_points[1], grid_points[2], niter,
tmax, mflops, " floating point",
verified, NPBVERSION,COMPILETIME, CS1, CS2, CS3, CS4, CS5,
CS6, CS7, clu_GetDeviceTypeName(device_type), device_name, num_devices);
if (timeron) {
/*
for (i = 0; i < t_last; i++) {
t1[i] = timer_read(i);
}
t1[t_xsolve] = t1[t_xsolve] - t1[t_xcomm];
t1[t_ysolve] = t1[t_ysolve] - t1[t_ycomm];
t1[t_zsolve] = t1[t_zsolve] - t1[t_zcomm];
t1[t_last+2] = t1[t_xcomm]+t1[t_ycomm]+t1[t_zcomm]+t1[t_exch];
t1[t_last+1] = t1[t_total] - t1[t_last+2];
MPI_Reduce(&t1, tsum, t_last+2, dp_type, MPI_SUM, 0, comm_setup);
MPI_Reduce(&t1, tming, t_last+2, dp_type, MPI_MIN, 0, comm_setup);
MPI_Reduce(&t1, tmaxg, t_last+2, dp_type, MPI_MAX, 0, comm_setup);
if (node == 0) {
printf(" nprocs =%6d minimum maximum average\n",
total_nodes);
for (i = 0; i < t_last+2; i++) {
tsum[i] = tsum[i] / total_nodes;
printf(" timer %2d(%8s) : %10.4f %10.4f %10.4f\n",
i+1, t_recs[i], tming[i], tmaxg[i], tsum[i]);
}
}
*/
}
release_opencl();
return 0;
}