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;
}
int main(int argc, char *argv[])
{
int i, niter, step, n3;
double mflops, t, tmax, trecs[t_last+1];
logical verified;
char Class;
char *t_names[t_last+1];
//---------------------------------------------------------------------
// Read input file (if it exists), else take
// defaults from parameters
//---------------------------------------------------------------------
FILE *fp;
if ((fp = fopen("timer.flag", "r")) != NULL) {
timeron = true;
t_names[t_total] = "total";
t_names[t_rhsx] = "rhsx";
t_names[t_rhsy] = "rhsy";
t_names[t_rhsz] = "rhsz";
t_names[t_rhs] = "rhs";
t_names[t_xsolve] = "xsolve";
t_names[t_ysolve] = "ysolve";
t_names[t_zsolve] = "zsolve";
t_names[t_rdis1] = "redist1";
t_names[t_rdis2] = "redist2";
t_names[t_tzetar] = "tzetar";
t_names[t_ninvr] = "ninvr";
t_names[t_pinvr] = "pinvr";
t_names[t_txinvr] = "txinvr";
t_names[t_add] = "add";
fclose(fp);
} else {
timeron = false;
}
printf("\n\n NAS Parallel Benchmarks (NPB3.3-SER-C) - SP Benchmark\n\n");
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, "%lf", &dt);
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;
dt = DT_DEFAULT;
grid_points[0] = PROBLEM_SIZE;
grid_points[1] = PROBLEM_SIZE;
grid_points[2] = PROBLEM_SIZE;
}
printf(" Size: %4dx%4dx%4d\n",
grid_points[0], grid_points[1], grid_points[2]);
printf(" Iterations: %4d dt: %10.6f\n", niter, dt);
printf("\n");
if ((grid_points[0] > IMAX) ||
(grid_points[1] > JMAX) ||
(grid_points[2] > KMAX) ) {
printf(" %d, %d, %d\n", grid_points[0], grid_points[1], grid_points[2]);
printf(" Problem size too big for compiled array sizes\n");
return 0;
}
nx2 = grid_points[0] - 2;
ny2 = grid_points[1] - 2;
nz2 = grid_points[2] - 2;
set_constants();
for (i = 1; i <= t_last; i++) {
timer_clear(i);
}
exact_rhs();
initialize();
//---------------------------------------------------------------------
// do one time step to touch all code, and reinitialize
//---------------------------------------------------------------------
adi();
initialize();
for (i = 1; i <= t_last; i++) {
timer_clear(i);
}
timer_start(1);
for (step = 1; step <= niter; step++) {
if ((step % 20) == 0 || step == 1) {
printf(" Time step %4d\n", step);
}
adi();
}
timer_stop(1);
tmax = timer_read(1);
verify(niter, &Class, &verified);
if (tmax != 0.0) {
n3 = grid_points[0]*grid_points[1]*grid_points[2];
t = (grid_points[0]+grid_points[1]+grid_points[2])/3.0;
mflops = (881.174 * (double)n3
- 4683.91 * (t * t)
+ 11484.5 * t
- 19272.4) * (double)niter / (tmax*1000000.0);
} else {
mflops = 0.0;
}
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, "(none)");
//---------------------------------------------------------------------
// More timers
//---------------------------------------------------------------------
if (timeron) {
for (i = 1; i <= t_last; i++) {
trecs[i] = timer_read(i);
}
if (tmax == 0.0) tmax = 1.0;
printf(" SECTION Time (secs)\n");
for (i = 1; i <= t_last; i++) {
printf(" %-8s:%9.3f (%6.2f%%)\n",
t_names[i], trecs[i], trecs[i]*100./tmax);
if (i == t_rhs) {
t = trecs[t_rhsx] + trecs[t_rhsy] + trecs[t_rhsz];
printf(" --> %8s:%9.3f (%6.2f%%)\n", "sub-rhs", t, t*100./tmax);
t = trecs[t_rhs] - t;
printf(" --> %8s:%9.3f (%6.2f%%)\n", "rest-rhs", t, t*100./tmax);
} else if (i == t_zsolve) {
t = trecs[t_zsolve] - trecs[t_rdis1] - trecs[t_rdis2];
printf(" --> %8s:%9.3f (%6.2f%%)\n", "sub-zsol", t, t*100./tmax);
} else if (i == t_rdis2) {
t = trecs[t_rdis1] + trecs[t_rdis2];
printf(" --> %8s:%9.3f (%6.2f%%)\n", "redist", t, t*100./tmax);
}
}
}
return 0;
}
int main(int argc, char *argv[])
{
int i, niter, step;
double navg, mflops, n3;
double tmax, t, trecs[t_last+1];
logical verified;
char Class;
char *t_names[t_last+1];
//---------------------------------------------------------------------
// Root node reads input file (if it exists) else takes
// defaults from parameters
//---------------------------------------------------------------------
FILE *fp;
if ((fp = fopen("timer.flag", "r")) != NULL) {
timeron = true;
t_names[t_total] = "total";
t_names[t_rhsx] = "rhsx";
t_names[t_rhsy] = "rhsy";
t_names[t_rhsz] = "rhsz";
t_names[t_rhs] = "rhs";
t_names[t_xsolve] = "xsolve";
t_names[t_ysolve] = "ysolve";
t_names[t_zsolve] = "zsolve";
t_names[t_rdis1] = "redist1";
t_names[t_rdis2] = "redist2";
t_names[t_add] = "add";
fclose(fp);
} else {
timeron = false;
}
printf("\n\n NAS Parallel Benchmarks (NPB3.3-OMP-C) - BT Benchmark\n\n");
if ((fp = fopen("inputbt.data", "r")) != NULL) {
int result;
printf(" Reading from input file inputbt.data\n");
result = fscanf(fp, "%d", &niter);
while (fgetc(fp) != '\n');
result = fscanf(fp, "%lf", &dt);
while (fgetc(fp) != '\n');
result = fscanf(fp, "%d%d%d\n",
&grid_points[0], &grid_points[1], &grid_points[2]);
fclose(fp);
} else {
printf(" No input file inputbt.data. Using compiled defaults\n");
niter = NITER_DEFAULT;
dt = DT_DEFAULT;
grid_points[0] = PROBLEM_SIZE;
grid_points[1] = PROBLEM_SIZE;
grid_points[2] = PROBLEM_SIZE;
}
printf(" Size: %4dx%4dx%4d\n",
grid_points[0], grid_points[1], grid_points[2]);
printf(" Iterations: %4d dt: %11.7f\n", niter, dt);
printf(" Number of available threads: %5d\n", omp_get_max_threads());
printf("\n");
if ( (grid_points[0] > IMAX) ||
(grid_points[1] > JMAX) ||
(grid_points[2] > KMAX) ) {
printf(" %d, %d, %d\n", grid_points[0], grid_points[1], grid_points[2]);
printf(" Problem size too big for compiled array sizes\n");
return 0;
}
set_constants();
for (i = 1; i <= t_last; i++) {
timer_clear(i);
}
initialize();
exact_rhs();
//---------------------------------------------------------------------
// do one time step to touch all code, and reinitialize
//---------------------------------------------------------------------
adi();
initialize();
for (i = 1; i <= t_last; i++) {
timer_clear(i);
}
timer_start(1);
// Do not do inlining, to avoid huge loops, scops are kept separate and are
// distributed among files.
//#pragma scop
for (step = 1; step <= niter; step++) {
adi();
}
//#pragma endscop
timer_stop(1);
tmax = timer_read(1);
verify(niter, &Class, &verified);
n3 = 1.0*grid_points[0]*grid_points[1]*grid_points[2];
navg = (grid_points[0]+grid_points[1]+grid_points[2])/3.0;
if(tmax != 0.0) {
mflops = 1.0e-6 * (double)niter *
(3478.8 * n3 - 17655.7 * (navg*navg) + 28023.7 * navg)
/ tmax;
} else {
mflops = 0.0;
}
print_results("BT", Class, grid_points[0],
grid_points[1], grid_points[2], niter,
tmax, mflops, " floating point",
verified, NPBVERSION,COMPILETIME, CS1, CS2, CS3, CS4, CS5,
CS6, "(none)");
//---------------------------------------------------------------------
// More timers
//---------------------------------------------------------------------
if (timeron) {
for (i = 1; i <= t_last; i++) {
trecs[i] = timer_read(i);
}
if (tmax == 0.0) tmax = 1.0;
printf(" SECTION Time (secs)\n");
for (i = 1; i <= t_last; i++) {
printf(" %-8s:%9.3f (%6.2f%%)\n",
t_names[i], trecs[i], trecs[i]*100./tmax);
if (i == t_rhs) {
t = trecs[t_rhsx] + trecs[t_rhsy] + trecs[t_rhsz];
printf(" --> %8s:%9.3f (%6.2f%%)\n", "sub-rhs", t, t*100./tmax);
t = trecs[t_rhs] - t;
printf(" --> %8s:%9.3f (%6.2f%%)\n", "rest-rhs", t, t*100./tmax);
} else if (i==t_zsolve) {
t = trecs[t_zsolve] - trecs[t_rdis1] - trecs[t_rdis2];
printf(" --> %8s:%9.3f (%6.2f%%)\n", "sub-zsol", t, t*100./tmax);
} else if (i==t_rdis2) {
t = trecs[t_rdis1] + trecs[t_rdis2];
printf(" --> %8s:%9.3f (%6.2f%%)\n", "redist", t, t*100./tmax);
}
}
}
return 0;
}
