void report_mflops(const bench_problem *p, double *t, int st)
{
struct stats s;
mkstat(t, st, &s);
ovtpvt("(%g %g %g %g)\n",
mflops(p, s.max), mflops(p, s.avg),
mflops(p, s.min), mflops(p, s.median));
}
void report_verbose(const bench_problem *p, double *t, int st)
{
struct stats s;
char bmin[64], bmax[64], bavg[64], bmedian[64], btmin[64];
char bsetup[64];
int copyp = tensor_sz(p->sz) == 1;
mkstat(t, st, &s);
sprintf_time(s.min, bmin, 64);
sprintf_time(s.max, bmax, 64);
sprintf_time(s.avg, bavg, 64);
sprintf_time(s.median, bmedian, 64);
sprintf_time(time_min, btmin, 64);
sprintf_time(p->setup_time, bsetup, 64);
ovtpvt("Problem: %s, setup: %s, time: %s, %s: %.5g\n",
p->pstring, bsetup, bmin,
copyp ? "fp-move/us" : "``mflops''",
mflops(p, s.min));
if (verbose) {
ovtpvt("Took %d measurements for at least %s each.\n", st, btmin);
ovtpvt("Time: min %s, max %s, avg %s, median %s\n",
bmin, bmax, bavg, bmedian);
}
}
void test_speed_aux(int n, fftw_direction dir, int flags, int specific)
{
fftw_complex *in, *out;
fftw_plan plan;
double t;
fftw_time begin, end;
in = (fftw_complex *) fftw_malloc(n * howmany_fields
* sizeof(fftw_complex));
out = (fftw_complex *) fftw_malloc(n * howmany_fields
* sizeof(fftw_complex));
if (specific) {
begin = fftw_get_time();
plan = fftw_create_plan_specific(n, dir,
speed_flag | flags
| wisdom_flag | no_vector_flag,
in, howmany_fields,
out, howmany_fields);
end = fftw_get_time();
} else {
begin = fftw_get_time();
plan = fftw_create_plan(n, dir, speed_flag | flags
| wisdom_flag | no_vector_flag);
end = fftw_get_time();
}
CHECK(plan != NULL, "can't create plan");
t = fftw_time_to_sec(fftw_time_diff(end, begin));
WHEN_VERBOSE(2, printf("time for planner: %f s\n", t));
WHEN_VERBOSE(2, fftw_print_plan(plan));
if (paranoid && !(flags & FFTW_IN_PLACE)) {
begin = fftw_get_time();
test_ergun(n, dir, plan);
end = fftw_get_time();
t = fftw_time_to_sec(fftw_time_diff(end, begin));
WHEN_VERBOSE(2, printf("time for validation: %f s\n", t));
}
FFTW_TIME_FFT(fftw(plan, howmany_fields,
in, howmany_fields, 1, out, howmany_fields, 1),
in, n * howmany_fields, t);
fftw_destroy_plan(plan);
WHEN_VERBOSE(1, printf("time for one fft: %s", smart_sprint_time(t)));
WHEN_VERBOSE(1, printf(" (%s/point)\n", smart_sprint_time(t / n)));
WHEN_VERBOSE(1, printf("\"mflops\" = 5 (n log2 n) / (t in microseconds)"
" = %f\n", howmany_fields * mflops(t, n)));
fftw_free(in);
fftw_free(out);
WHEN_VERBOSE(1, printf("\n"));
}
void test_speed_nd_aux(struct size sz,
fftw_direction dir, int flags, int specific)
{
fftw_complex *in;
fftwnd_plan plan;
double t;
fftw_time begin, end;
int i, N;
/* only bench in-place multi-dim transforms */
flags |= FFTW_IN_PLACE;
N = 1;
for (i = 0; i < sz.rank; ++i)
N *= (sz.narray[i]);
in = (fftw_complex *) fftw_malloc(N * howmany_fields *
sizeof(fftw_complex));
if (specific) {
begin = fftw_get_time();
plan = fftwnd_create_plan_specific(sz.rank, sz.narray, dir,
speed_flag | flags
| wisdom_flag | no_vector_flag,
in, howmany_fields, 0, 1);
} else {
begin = fftw_get_time();
plan = fftwnd_create_plan(sz.rank, sz.narray,
dir, speed_flag | flags
| wisdom_flag | no_vector_flag);
}
end = fftw_get_time();
CHECK(plan != NULL, "can't create plan");
t = fftw_time_to_sec(fftw_time_diff(end, begin));
WHEN_VERBOSE(2, printf("time for planner: %f s\n", t));
WHEN_VERBOSE(2, printf("\n"));
WHEN_VERBOSE(2, (fftwnd_print_plan(plan)));
WHEN_VERBOSE(2, printf("\n"));
FFTW_TIME_FFT(fftwnd(plan, howmany_fields,
in, howmany_fields, 1, 0, 0, 0),
in, N * howmany_fields, t);
fftwnd_destroy_plan(plan);
WHEN_VERBOSE(1, printf("time for one fft: %s", smart_sprint_time(t)));
WHEN_VERBOSE(1, printf(" (%s/point)\n", smart_sprint_time(t / N)));
WHEN_VERBOSE(1, printf("\"mflops\" = 5 (N log2 N) / (t in microseconds)"
" = %f\n", howmany_fields * mflops(t, N)));
fftw_free(in);
WHEN_VERBOSE(1, printf("\n"));
}
int
main(int argc,char *argv[])
{
int i,j,k,nn;
int mx,my,mz,it;
float gosa;
double cpu,cpu0,cpu1,flop,target;
target= 60.0;
omega= 0.8;
mx= MX0-1;
my= MY0-1;
mz= MZ0-1;
ndx= NDX0;
ndy= NDY0;
ndz= NDZ0;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &npe);
MPI_Comm_rank(MPI_COMM_WORLD, &id);
int namelen;
char processor_name[MPI_MAX_PROCESSOR_NAME];
MPI_Get_processor_name(processor_name,&namelen);
fprintf(stderr, "[%d] %s\n", id, processor_name);
initcomm(ndx,ndy,ndz);
it= initmax(mx,my,mz);
/*
* Initializing matrixes
*/
initmt(mx,it);
float *sendp2_buf = (float*)malloc(MIMAX*MKMAX*sizeof(float)*4);
sendp2_lo_sendbuf = &sendp2_buf[MIMAX*MKMAX*0];
sendp2_lo_recvbuf = &sendp2_buf[MIMAX*MKMAX*1];
sendp2_hi_sendbuf = &sendp2_buf[MIMAX*MKMAX*2];
sendp2_hi_recvbuf = &sendp2_buf[MIMAX*MKMAX*3];
#pragma acc enter data create(sendp2_buf[0:MIMAX*MKMAX*4])
if(id==0){
printf("Sequential version array size\n");
printf(" mimax = %d mjmax = %d mkmax = %d\n",MX0,MY0,MZ0);
printf("Parallel version array size\n");
printf(" mimax = %d mjmax = %d mkmax = %d\n",MIMAX,MJMAX,MKMAX);
printf("imax = %d jmax = %d kmax =%d\n",imax,jmax,kmax);
printf("I-decomp = %d J-decomp = %d K-decomp =%d\n",ndx,ndy,ndz);
}
nn= 3;
if(id==0){
printf(" Start rehearsal measurement process.\n");
printf(" Measure the performance in %d times.\n\n",nn);
}
#pragma acc data copyin(p, bnd, wrk1, wrk2, a, b, c) present(sendp2_buf[0:MIMAX*MKMAX*4])
{
MPI_Barrier(MPI_COMM_WORLD);
cpu0= gettime();
gosa= jacobi(nn);
cpu1= gettime();
cpu = cpu1 - cpu0;
MPI_Allreduce(MPI_IN_PLACE,
&cpu,
1,
MPI_DOUBLE,
MPI_MAX,
MPI_COMM_WORLD);
flop= fflop(mz,my,mx);
if(id == 0){
printf(" MFLOPS: %f time(s): %f %e\n\n",
mflops(nn,cpu,flop),cpu,gosa);
}
nn= (int)(target/(cpu/3.0));
nn= LOOP_TIMES;
halo_time = 0.0;
if(id == 0){
printf(" Now, start the actual measurement process.\n");
printf(" The loop will be excuted in %d times\n",nn);
printf(" This will take about one minute.\n");
printf(" Wait for a while\n\n");
}
/*
* Start measuring
*/
MPI_Barrier(MPI_COMM_WORLD);
cpu0= gettime();
gosa= jacobi(nn);
cpu1= gettime();
cpu = cpu1 - cpu0;
MPI_Allreduce(MPI_IN_PLACE,
&cpu,
1,
MPI_DOUBLE,
MPI_MAX,
MPI_COMM_WORLD);
MPI_Allreduce(&halo_time,
&max_halo_time,
1,
MPI_DOUBLE,
MPI_MAX,
MPI_COMM_WORLD);
MPI_Allreduce(&halo_time,
&ave_halo_time,
1,
MPI_DOUBLE,
MPI_SUM,
MPI_COMM_WORLD);
ave_halo_time /= npe;
}//end of acc data
if(id == 0){
printf("cpu : %f sec. halo(AVE.) %f sec. halo(MAX) %f sec.\n", cpu, ave_halo_time, max_halo_time);
printf("Loop executed for %d times\n",nn);
printf("Gosa : %e \n",gosa);
printf("MFLOPS measured : %f\n",mflops(nn,cpu,flop));
printf("Score based on Pentium III 600MHz : %f\n",
mflops(nn,cpu,flop)/82.84);
}
free(sendp2_buf);
MPI_Finalize();
return (0);
}
void report_benchmark(const bench_problem *p, double *t, int st)
{
struct stats s;
mkstat(t, st, &s);
ovtpvt("%.5g %.8g %g\n", mflops(p, s.min), s.min, p->setup_time);
}
int
main(int argc,char *argv[])
{
int i,j,k,nn;
int mx,my,mz,it;
float gosa;
double cpu,cpu0,cpu1,flop,target;
target= 60.0;
omega= 0.8;
mx= MX0-1;
my= MY0-1;
mz= MZ0-1;
ndx= NDX0;
ndy= NDY0;
ndz= NDZ0;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &npe);
MPI_Comm_rank(MPI_COMM_WORLD, &id);
hime_err_init(id);
if (argc != 3) {
if (id == 0) {
printf("./bmt <Restart #> <Checkpoint interval (steps)>\n");
printf("\n");
printf(" Restart #:\n");
printf(" Checkpiont id at which bmt starts\n");
printf(" Checkpoint interval (steps):\n");
printf(" # of Steps to skip checkpointing\n");
printf("");
}
MPI_Finalize();
exit(0);
}
restart_id = atoi(argv[1]);
interval = atoi(argv[2]);
hime_dbgi(0, "Checkpoint directory: %s", CHECKPOINT_DIR);
hime_dbgi(0, "Checkpoint interval: %d", interval);
if (restart_id > 0) {
hime_dbgi(0, "Restart ID: %d", restart_id);
restart(restart_id);
}
initcomm(ndx,ndy,ndz);
it= initmax(mx,my,mz);
/*
* Initializing matrixes
*/
initmt(mx,it);
if(id==0){
printf("Sequential version array size\n");
printf(" mimax = %d mjmax = %d mkmax = %d\n",MX0,MY0,MZ0);
printf("Parallel version array size\n");
printf(" mimax = %d mjmax = %d mkmax = %d\n",MIMAX,MJMAX,MKMAX);
printf("imax = %d jmax = %d kmax =%d\n",imax,jmax,kmax);
printf("I-decomp = %d J-decomp = %d K-decomp =%d\n",ndx,ndy,ndz);
}
nn= 3;
if(id==0){
printf(" Start rehearsal measurement process.\n");
printf(" Measure the performance in %d times.\n\n",nn);
}
MPI_Barrier(MPI_COMM_WORLD);
cpu0= MPI_Wtime();
gosa= jacobi(nn);
cpu1= MPI_Wtime() - cpu0;
MPI_Allreduce(&cpu1,
&cpu,
1,
MPI_DOUBLE,
MPI_MAX,
MPI_COMM_WORLD);
flop= fflop(mz,my,mx);
if(id == 0){
printf(" MFLOPS: %f time(s): %f %e\n\n",
mflops(nn,cpu,flop),cpu,gosa);
}
nn= (int)(target/(cpu/3.0));
if(id == 0){
printf(" Now, start the actual measurement process.\n");
printf(" The loop will be excuted in %d times\n",nn);
printf(" This will take about one minute.\n");
printf(" Wait for a while\n\n");
}
/*
* Start measuring
*/
MPI_Barrier(MPI_COMM_WORLD);
cpu0 = MPI_Wtime();
gosa = jacobi(nn);
cpu1 = MPI_Wtime() - cpu0;
MPI_Allreduce(&cpu1,
&cpu,
1,
MPI_DOUBLE,
MPI_MAX,
MPI_COMM_WORLD);
if(id == 0){
printf("cpu : %f sec.\n", cpu);
printf("Loop executed for %d times\n",nn);
printf("Gosa : %e \n",gosa);
printf("MFLOPS measured : %f\n",mflops(nn,cpu,flop));
printf("Score based on Pentium III 600MHz : %f\n",
mflops(nn,cpu,flop)/82.84);
}
MPI_Finalize();
return (0);
}
void test_speed_aux(int n, fftw_direction dir, int flags, int specific)
{
int local_n, local_start, local_n_after_transform,
local_start_after_transform, total_local_size, nalloc;
fftw_complex *in, *work;
fftw_plan plan = 0;
fftw_mpi_plan mpi_plan;
double t, t0 = 0.0;
if (specific || !(flags & FFTW_IN_PLACE))
return;
if (io_okay && !only_parallel)
plan = fftw_create_plan(n, dir, speed_flag | flags
| wisdom_flag | no_vector_flag);
mpi_plan = fftw_mpi_create_plan(MPI_COMM_WORLD, n, dir,
speed_flag | flags
| wisdom_flag | no_vector_flag);
CHECK(mpi_plan, "failed to create plan!");
fftw_mpi_local_sizes(mpi_plan, &local_n, &local_start,
&local_n_after_transform,
&local_start_after_transform,
&total_local_size);
if (io_okay && !only_parallel)
nalloc = n;
else
nalloc = total_local_size;
in = (fftw_complex *) fftw_malloc(nalloc * howmany_fields
* sizeof(fftw_complex));
work = (fftw_complex *) fftw_malloc(nalloc * howmany_fields
* sizeof(fftw_complex));
if (io_okay) {
WHEN_VERBOSE(2, fftw_mpi_print_plan(mpi_plan));
}
if (io_okay && !only_parallel) {
FFTW_TIME_FFT(fftw(plan, howmany_fields,
in, howmany_fields, 1, work, 1, 0),
in, n * howmany_fields, t0);
fftw_destroy_plan(plan);
WHEN_VERBOSE(1, printf("time for one fft (uniprocessor): %s\n", smart_sprint_time(t0)));
}
MPI_TIME_FFT(fftw_mpi(mpi_plan, howmany_fields, in, NULL),
in, total_local_size * howmany_fields, t);
if (io_okay) {
WHEN_VERBOSE(1, printf("time for one fft (%d cpus): %s", ncpus, smart_sprint_time(t)));
WHEN_VERBOSE(1, printf(" (%s/point)\n", smart_sprint_time(t / n)));
WHEN_VERBOSE(1, printf("\"mflops\" = 5 (n log2 n) / (t in microseconds)"
" = %f\n", howmany_fields * mflops(t, n)));
if (!only_parallel)
WHEN_VERBOSE(1, printf("parallel speedup: %f\n", t0 / t));
}
MPI_TIME_FFT(fftw_mpi(mpi_plan, howmany_fields, in, work),
in, total_local_size * howmany_fields, t);
if (io_okay) {
WHEN_VERBOSE(1, printf("w/WORK: time for one fft (%d cpus): %s", ncpus, smart_sprint_time(t)));
WHEN_VERBOSE(1, printf(" (%s/point)\n", smart_sprint_time(t / n)));
WHEN_VERBOSE(1, printf("w/WORK: \"mflops\" = 5 (n log2 n) / (t in microseconds)"
" = %f\n", howmany_fields * mflops(t, n)));
if (!only_parallel)
WHEN_VERBOSE(1, printf("w/WORK: parallel speedup: %f\n", t0 / t));
}
fftw_free(in);
fftw_free(work);
fftw_mpi_destroy_plan(mpi_plan);
WHEN_VERBOSE(1, my_printf("\n"));
}
void test_speed_nd_aux(struct size sz,
fftw_direction dir, int flags, int specific)
{
int local_nx, local_x_start, local_ny_after_transpose,
local_y_start_after_transpose, total_local_size;
fftw_complex *in, *work;
fftwnd_plan plan = 0;
fftwnd_mpi_plan mpi_plan;
double t, t0 = 0.0;
int i, N;
if (sz.rank < 2)
return;
/* only bench in-place multi-dim transforms */
flags |= FFTW_IN_PLACE;
N = 1;
for (i = 0; i < sz.rank; ++i)
N *= (sz.narray[i]);
if (specific) {
return;
} else {
if (io_okay && !only_parallel)
plan = fftwnd_create_plan(sz.rank, sz.narray,
dir, speed_flag | flags
| wisdom_flag | no_vector_flag);
mpi_plan = fftwnd_mpi_create_plan(MPI_COMM_WORLD, sz.rank, sz.narray,
dir, speed_flag | flags
| wisdom_flag | no_vector_flag);
}
CHECK(mpi_plan != NULL, "can't create plan");
fftwnd_mpi_local_sizes(mpi_plan, &local_nx, &local_x_start,
&local_ny_after_transpose,
&local_y_start_after_transpose,
&total_local_size);
if (io_okay && !only_parallel)
in = (fftw_complex *) fftw_malloc(N * howmany_fields *
sizeof(fftw_complex));
else
in = (fftw_complex *) fftw_malloc(total_local_size * howmany_fields *
sizeof(fftw_complex));
work = (fftw_complex *) fftw_malloc(total_local_size * howmany_fields *
sizeof(fftw_complex));
if (io_okay && !only_parallel) {
FFTW_TIME_FFT(fftwnd(plan, howmany_fields,
in, howmany_fields, 1, 0, 0, 0),
in, N * howmany_fields, t0);
fftwnd_destroy_plan(plan);
WHEN_VERBOSE(1, printf("time for one fft (uniprocessor): %s\n",
smart_sprint_time(t0)));
}
MPI_TIME_FFT(fftwnd_mpi(mpi_plan, howmany_fields,
in, NULL, FFTW_NORMAL_ORDER),
in, total_local_size * howmany_fields, t);
if (io_okay) {
WHEN_VERBOSE(1, printf("NORMAL: time for one fft (%d cpus): %s",
ncpus, smart_sprint_time(t)));
WHEN_VERBOSE(1, printf(" (%s/point)\n", smart_sprint_time(t / N)));
WHEN_VERBOSE(1, printf("NORMAL: \"mflops\" = 5 (N log2 N) / "
"(t in microseconds)"
" = %f\n", howmany_fields * mflops(t, N)));
if (!only_parallel)
WHEN_VERBOSE(1, printf("NORMAL: parallel speedup: %f\n", t0 / t));
}
MPI_TIME_FFT(fftwnd_mpi(mpi_plan, howmany_fields,
in, NULL, FFTW_TRANSPOSED_ORDER),
in, total_local_size * howmany_fields, t);
if (io_okay) {
WHEN_VERBOSE(1, printf("TRANSP.: time for one fft (%d cpus): %s",
ncpus, smart_sprint_time(t)));
WHEN_VERBOSE(1, printf(" (%s/point)\n", smart_sprint_time(t / N)));
WHEN_VERBOSE(1, printf("TRANSP.: \"mflops\" = 5 (N log2 N) / "
"(t in microseconds)"
" = %f\n", howmany_fields * mflops(t, N)));
if (!only_parallel)
WHEN_VERBOSE(1, printf("TRANSP.: parallel speedup: %f\n", t0 / t));
}
MPI_TIME_FFT(fftwnd_mpi(mpi_plan, howmany_fields,
in, work, FFTW_NORMAL_ORDER),
in, total_local_size * howmany_fields, t);
if (io_okay) {
WHEN_VERBOSE(1, printf("NORMAL,w/WORK: time for one fft (%d cpus): %s",
ncpus, smart_sprint_time(t)));
WHEN_VERBOSE(1, printf(" (%s/point)\n", smart_sprint_time(t / N)));
WHEN_VERBOSE(1, printf("NORMAL,w/WORK: \"mflops\" = 5 (N log2 N) / "
"(t in microseconds)"
" = %f\n", howmany_fields * mflops(t, N)));
if (!only_parallel)
WHEN_VERBOSE(1, printf("NORMAL,w/WORK: parallel speedup: %f\n", t0 / t));
}
MPI_TIME_FFT(fftwnd_mpi(mpi_plan, howmany_fields,
in, work, FFTW_TRANSPOSED_ORDER),
in, total_local_size * howmany_fields, t);
if (io_okay) {
WHEN_VERBOSE(1, printf("TRANSP.,w/WORK: time for one fft (%d cpus): %s",
ncpus, smart_sprint_time(t)));
WHEN_VERBOSE(1, printf(" (%s/point)\n", smart_sprint_time(t / N)));
WHEN_VERBOSE(1, printf("TRANSP.,w/WORK: \"mflops\" = 5 (N log2 N) / "
"(t in microseconds)"
" = %f\n", howmany_fields * mflops(t, N)));
if (!only_parallel)
WHEN_VERBOSE(1, printf("TRANSP.,w/WORK: parallel speedup: %f\n", t0 / t));
}
fftwnd_mpi_destroy_plan(mpi_plan);
fftw_free(in);
fftw_free(work);
WHEN_VERBOSE(1, my_printf("\n"));
}
int
main(int argc,char *argv[])
{
int i,j,k,nn;
int mx,my,mz,it;
float gosa;
double cpu,cpu0,cpu1,flop,target;
target= 60.0;
omega= 0.8;
mx= MX0-1;
my= MY0-1;
mz= MZ0-1;
ndx= NDX0;
ndy= NDY0;
ndz= NDZ0;
MPI_Init(&argc, &argv);
#ifdef SCR_ENABLE
SCR_Init();
#endif
MPI_Comm_size(MPI_COMM_WORLD, &npe);
MPI_Comm_rank(MPI_COMM_WORLD, &id);
initcomm(ndx,ndy,ndz);
it= initmax(mx,my,mz);
/*
* Initializing matrixes
*/
initmt(mx,it);
if(id==0){
printf("Sequential version array size\n");
printf(" mimax = %d mjmax = %d mkmax = %d\n",MX0,MY0,MZ0);
printf("Parallel version array size\n");
printf(" mimax = %d mjmax = %d mkmax = %d\n",MIMAX,MJMAX,MKMAX);
printf("imax = %d jmax = %d kmax =%d\n",imax,jmax,kmax);
printf("I-decomp = %d J-decomp = %d K-decomp =%d\n",ndx,ndy,ndz);
}
nn= 3;
if(id==0){
printf(" Start rehearsal measurement process.\n");
printf(" Measure the performance in %d times.\n\n",nn);
}
MPI_Barrier(MPI_COMM_WORLD);
cpu0= MPI_Wtime();
gosa= jacobi(nn);
cpu1= MPI_Wtime() - cpu0;
MPI_Allreduce(&cpu1,
&cpu,
1,
MPI_DOUBLE,
MPI_MAX,
MPI_COMM_WORLD);
flop= fflop(mz,my,mx);
if(id == 0){
printf(" MFLOPS: %f time(s): %f %e\n\n",
mflops(nn,cpu,flop),cpu,gosa);
}
nn= (int)(target/(cpu/3.0));
if(id == 0){
printf(" Now, start the actual measurement process.\n");
printf(" The loop will be excuted in %d times\n",nn);
printf(" This will take about one minute.\n");
printf(" Wait for a while\n\n");
}
/*
* Start measuring
*/
MPI_Barrier(MPI_COMM_WORLD);
cpu0 = MPI_Wtime();
// nn = 10000000;
gosa = jacobi(nn);
cpu1 = MPI_Wtime() - cpu0;
MPI_Allreduce(&cpu1,
&cpu,
1,
MPI_DOUBLE,
MPI_MAX,
MPI_COMM_WORLD);
if(id == 0){
fprintf(stderr, "cpu : %f sec.\n", cpu);
fprintf(stderr, "Loop executed for %d times\n",nn);
fprintf(stderr, "Gosa : %e \n",gosa);
fprintf(stderr, "GFLOPS measured : %f\n",mflops(nn,cpu,flop)/1000.0);
fprintf(stderr, "Score based on Pentium III 600MHz : %f\n",
mflops(nn,cpu,flop)/82.84);
}
#ifdef SCR_ENABLE
SCR_Finalize();
#endif
MPI_Finalize();
return (0);
}
int
main()
{
int i, j, k, nn;
float gosa;
double cpu, cpu0, cpu1, flop, target;
int myrank = xmp_node_num() - 1;
target = 60.0;
omega = 0.8;
imax = MIMAX;
jmax = MJMAX;
kmax = MKMAX;
/*
* Initializing matrixes
*/
initmt();
if (myrank == 0) {
printf("mimax = %d mjmax = %d mkmax = %d\n",MIMAX, MJMAX, MKMAX);
printf("imax = %d jmax = %d kmax =%d\n",imax,jmax,kmax);
}
nn= 3;
if (myrank == 0) {
printf(" Start rehearsal measurement process.\n");
printf(" Measure the performance in %d times.\n\n",nn);
}
cpu0= xmp_wtime();
gosa= jacobi(nn);
cpu1= xmp_wtime();
cpu= cpu1 - cpu0;
flop= fflop(imax,jmax,kmax);
if (myrank == 0) {
printf(" MFLOPS: %f time(s): %f %e\n\n", mflops(nn,cpu,flop),cpu,gosa);
}
nn= (int)(target/(cpu/3.0));
#pragma xmp reduction (max:nn)
if (myrank == 0) {
printf(" Now, start the actual measurement process.\n");
printf(" The loop will be excuted in %d times\n",nn);
printf(" This will take about one minute.\n");
printf(" Wait for a while\n\n");
}
/*
* Start measuring
*/
cpu0 = xmp_wtime();
gosa = jacobi(nn);
cpu1 = xmp_wtime();
cpu= cpu1 - cpu0;
if (myrank == 0) {
printf(" Loop executed for %d times\n",nn);
printf(" Gosa : %e \n",gosa);
printf(" MFLOPS measured : %f\tcpu : %f\n",mflops(nn,cpu,flop),cpu);
printf(" Score based on Pentium III 600MHz : %f\n",
mflops(nn,cpu,flop)/82.84);
}
return (0);
}