void
neighbor_get(int *target, int *src, int elements, int me, int npes, int loops)
{
int i, neighbor_pe;
double start_time, elapsed_time;
long total_bytes = loops * elements * sizeof(*src);
if (me==0 && Verbose) {
fprintf(stdout, "%s: %d loops of get(%ld bytes) from neighbor, %d PEs: ",
__FUNCTION__, loops, (elements*sizeof(*src)), npes);
fflush(stdout);
}
shmem_barrier_all();
neighbor_pe = (me + 1) % npes;
start_time = shmemx_wtime();
for(i = 0; i < loops; i++)
shmem_int_get(target, src, elements, neighbor_pe);
elapsed_time = shmemx_wtime() - start_time;
if (me==0 && Verbose) {
printf("%7.3f secs\n", elapsed_time);
printf(" %7.5f usecs / get(), %ld Kbytes @ %7.4f MB/sec\n\n",
(elapsed_time/((double)loops*npes))*1000000.0,
(total_bytes/1024),
((double)total_bytes/(1024.0*1024.0)) / elapsed_time );
}
shmem_barrier_all();
}
void
many2one_get(int *target, int *src, int elements, int me, int npes, int loops)
{
int i, pe;
double start_time, elapsed_time;
long total_bytes = loops * elements * sizeof(*src) * (npes - 1);
if (me == 0) {
fprintf(stdout,"%s: %d loops of get(%ld bytes) from %d PEs: ",
__FUNCTION__, loops, (elements*sizeof(*src)), npes-1);
fflush(stdout);
}
shmem_barrier_all();
if (me == 0) {
start_time = shmemx_wtime();
for(i = 0; i < loops; i++) {
for(pe = 1; pe < npes; pe++)
shmem_int_get(target, src, elements, pe);
}
elapsed_time = shmemx_wtime() - start_time;
if (Verbose) {
printf("%7.3f secs\n", elapsed_time);
printf(" %7.5f usecs / get(), %ld Kbytes @ %7.4f MB/sec\n\n",
(elapsed_time/((double)loops*(npes-1)))*1000000.0,
(total_bytes/1024),
((double)total_bytes/(1024.0*1024.0)) / elapsed_time );
}
}
shmem_barrier_all();
}
void
fcollect(int *target, int *src, int elements, int me, int npes, int loops)
{
int i;
double start_time, elapsed_time;
long total_bytes = loops * elements * sizeof(*src);
long *ps, *pSync, *pSync1;
pSync = (long*) shmem_malloc( 2 * sizeof(long) * _SHMEM_COLLECT_SYNC_SIZE );
pSync1 = &pSync[_SHMEM_COLLECT_SYNC_SIZE];
for (i = 0; i < _SHMEM_COLLECT_SYNC_SIZE; i++) {
pSync[i] = pSync1[i] = _SHMEM_SYNC_VALUE;
}
target = (int *) shmem_malloc( elements * sizeof(*target) * npes );
if (me==0 && Verbose) {
fprintf(stdout,"%s: %d loops of fcollect32(%ld bytes) over %d PEs: ",
__FUNCTION__,loops,(elements*sizeof(*src)),npes);
fflush(stdout);
}
shmem_barrier_all();
start_time = shmemx_wtime();
for(i = 0; i < loops; i++) {
ps = &pSync[(i&1)];
shmem_fcollect32( target, src, elements, 0, 0, npes, ps );
}
elapsed_time = shmemx_wtime() - start_time;
if (me==0 && Verbose) {
printf("%7.3f secs\n", elapsed_time);
printf(" %7.5f usecs / fcollect32(), %ld Kbytes @ %7.4f MB/sec\n\n",
(elapsed_time/((double)loops*npes))*1000000.0,
(total_bytes/1024),
((double)total_bytes/(1024.0*1024.0)) / elapsed_time );
}
shmem_barrier_all();
shmem_free(target);
shmem_free( pSync );
shmem_barrier_all();
}
double FORTRANIFY (shmemx_wtime) (void)
{
return shmemx_wtime ();
}
static inline double
timer(void)
{
return shmemx_wtime();
}
int
main(int argc, char **argv)
{
int loops=DFLT_LOOPS;
char *pgm;
int *Target;
int *Source;
int i, me, npes;
int target_PE;
long bytes;
double start_time, *total_time;
shmem_init();
me = shmem_my_pe();
npes = shmem_n_pes();
if ((pgm=strrchr(argv[0],'/')))
pgm++;
else
pgm = argv[0];
while ((i = getopt (argc, argv, "hve:l:st")) != EOF) {
switch (i)
{
case 'v':
Verbose++;
break;
case 'e':
if ((elements = atoi_scaled(optarg)) <= 0) {
fprintf(stderr,"ERR: Bad elements count %d\n",elements);
shmem_finalize();
return 1;
}
break;
case 'l':
if ((loops = atoi_scaled(optarg)) <= 0) {
fprintf(stderr,"ERR: Bad loop count %d\n",loops);
shmem_finalize();
return 1;
}
break;
case 's':
Sync++;
break;
case 't':
Track++;
break;
case 'h':
if (me == 0)
usage(pgm);
return 0;
default:
if (me == 0) {
fprintf(stderr,"%s: unknown switch '-%c'?\n",pgm,i);
usage(pgm);
}
shmem_finalize();
return 1;
}
}
for(i=0; i < SHMEM_REDUCE_SYNC_SIZE; i++)
pSync[i] = SHMEM_SYNC_VALUE;
target_PE = (me+1) % npes;
total_time = (double *) shmem_malloc( npes * sizeof(double) );
if (!total_time) {
fprintf(stderr,"ERR: bad total_time shmem_malloc(%ld)\n",
(elements * sizeof(double)));
shmem_global_exit(1);
}
for(i=0; i < npes; i++)
total_time[i] = -1.0;
Source = (int *) shmem_malloc( elements * sizeof(*Source) );
if (!Source) {
fprintf(stderr,"ERR: bad Source shmem_malloc(%ld)\n",
(elements * sizeof(*Target)));
shmem_free(total_time);
shmem_global_exit(1);
}
Target = (int *) shmem_malloc( elements * sizeof(*Target) );
if (!Target) {
fprintf(stderr,"ERR: bad Target shmem_malloc(%ld)\n",
(elements * sizeof(*Target)));
shmem_free(Source);
shmem_free(total_time);
shmem_global_exit(1);
}
for (i = 0; i < elements; i++) {
Target[i] = -90;
Source[i] = i + 1;
}
bytes = loops * sizeof(int) * elements;
if (Verbose && me==0) {
fprintf(stderr,
"%s: INFO - %d loops, put %d (int) elements to PE+1 Max put ??\n",
pgm, loops, elements);
}
shmem_barrier_all();
for(i=0; i < loops; i++) {
start_time = shmemx_wtime();
shmem_int_put(Target, Source, elements, target_PE);
time_taken += (shmemx_wtime() - start_time);
if (me==0) {
if ( Track && i > 0 && ((i % 200) == 0))
fprintf(stderr,".%d",i);
}
if (Sync)
shmem_barrier_all();
}
// collect time per node.
shmem_double_put( &total_time[me], &time_taken, 1, 0 );
shmem_double_sum_to_all(&sum_time, &time_taken, 1, 0, 0, npes, pWrk, pSync);
shmem_barrier_all();
for (i = 0; i < elements; i++) {
if (Target[i] != i + 1) {
printf("%d: Error Target[%d] = %d, expected %d\n",
me, i, Target[i], i + 1);
shmem_global_exit(1);
}
}
if ( Track && me == 0 ) fprintf(stderr,"\n");
if(Verbose && me == 0) {
double rate, comp_time;
if (Verbose > 1)
fprintf(stdout,"Individule PE times: (seconds)\n");
for(i=0,comp_time=0.0; i < npes; i++) {
comp_time += total_time[i];
if (Verbose > 1)
fprintf(stdout," PE[%d] %8.6f\n",i,total_time[i]);
}
sum_time /= (double)npes;
comp_time /= (double)npes;
if (sum_time != comp_time)
printf("%s: computed_time %7.5f != sum_to_all_time %7.5f)\n",
pgm, comp_time, sum_time );
rate = ((double)bytes/(1024.0*1024.0)) / comp_time;
printf("%s: shmem_int_put() %7.4f MB/sec (bytes %ld secs %7.4f)\n",
pgm, rate, bytes, sum_time);
}
shmem_free(total_time);
shmem_free(Target);
shmem_free(Source);
shmem_finalize();
return 0;
}
int
main(int argc, char **argv)
{
int loops=DFLT_LOOPS;
char *pgm;
int *Target;
int *Source;
int i, me, npes;
int target_pe;
long bytes;
double time_taken=0.0, start_time;
shmem_init();
me = shmem_my_pe();
npes = shmem_n_pes();
if ((pgm=strrchr(argv[0],'/')))
pgm++;
else
pgm = argv[0];
while ((i = getopt (argc, argv, "hve:l:st")) != EOF) {
switch (i)
{
case 'v':
Verbose++;
break;
case 'e':
if ((elements = atoi_scaled(optarg)) <= 0) {
fprintf(stderr,"ERR: Bad elements count %d\n",elements);
shmem_finalize();
return 1;
}
break;
case 'l':
if ((loops = atoi_scaled(optarg)) <= 0) {
fprintf(stderr,"ERR: Bad loop count %d\n",loops);
shmem_finalize();
return 1;
}
break;
case 's':
Sync++;
break;
case 't':
Track++;
break;
case 'h':
if (me == 0)
usage(pgm);
return 0;
default:
if (me == 0) {
fprintf(stderr,"%s: unknown switch '-%c'?\n",pgm,i);
usage(pgm);
}
shmem_finalize();
return 1;
}
}
target_pe = (me+1) % npes;
total_time = (double *) shmem_malloc( npes * sizeof(double) );
if (!total_time) {
fprintf(stderr,"ERR: bad total_time shmem_malloc(%ld)\n",
(elements * sizeof(double)));
shmem_global_exit(1);
}
Source = (int *) shmem_malloc( elements * sizeof(*Source) );
if (!Source) {
fprintf(stderr,"ERR: bad Source shmem_malloc(%ld)\n",
(elements * sizeof(*Target)));
shmem_free(total_time);
shmem_global_exit(1);
}
Target = (int *) shmem_malloc( elements * sizeof(*Target) );
if (!Target) {
fprintf(stderr,"ERR: bad Target shmem_malloc(%ld)\n",
(elements * sizeof(*Target)));
shmem_free(Source);
shmem_free(total_time);
shmem_global_exit(1);
}
for (i = 0; i < elements; i++) {
Target[i] = -90;
Source[i] = i + 1;
}
bytes = loops * sizeof(int) * elements;
if (Verbose && me==0)
fprintf(stderr, "%s: INFO - %d loops, get %d (int) elements from PE+1\n",
pgm, loops, elements);
shmem_barrier_all();
for(i=0; i < loops; i++) {
start_time = shmemx_wtime();
shmem_int_get( Target, Source, elements, target_pe );
time_taken += shmemx_wtime() - start_time;
if (me==0) {
if ( Track && i > 0 && ((i % 200) == 0))
fprintf(stderr,".%d",i);
}
if (Sync)
shmem_barrier_all();
}
// collect time per node elapsed time.
shmem_double_put( &total_time[me], &time_taken, 1, 0 );
shmem_barrier_all();
for (i = 0; i < elements; i++) {
if (Target[i] != i + 1) {
printf("%d: Error Target[%d] = %d, expected %d\n",
me, i, Target[i], i + 1);
shmem_global_exit(1);
}
}
if ( Track && me == 0 )
fprintf(stderr,"\n");
if (Verbose && me == 0) {
double rate,secs;
// average time
for(i=0,secs=0.0; i < npes; i++)
secs += total_time[i];
secs /= (double)npes;
rate = ((double)bytes/(1024.0*1024.0)) / secs;
printf("%s: ave %5.3f MB/sec (bytes %ld secs %5.3f)\n",
pgm, rate, bytes, secs);
}
shmem_free(total_time);
shmem_free(Target);
shmem_free(Source);
shmem_finalize();
return 0;
}
int
main(int argc, char **argv)
{
int i,ps,ps_cnt=2;
int *target;
int *source;
int me, npes, elements=N_ELEMENTS, loops=DFLT_LOOPS;
char *pgm;
double start_time, time_taken;
shmem_init();
me = shmem_my_pe();
npes = shmem_n_pes();
if ((pgm=strrchr(argv[0],'/')))
pgm++;
else
pgm = argv[0];
while ((i = getopt (argc, argv, "hve:l:p:s")) != EOF) {
switch (i)
{
case 'v':
Verbose++;
break;
case 'e':
if ((elements = atoi_scaled(optarg)) <= 0) {
fprintf(stderr,"ERR: Bad elements count %d\n",elements);
shmem_finalize();
return 1;
}
break;
case 'l':
if ((loops = atoi_scaled(optarg)) <= 0) {
fprintf(stderr,"ERR: Bad loop count %d\n",loops);
shmem_finalize();
return 1;
}
break;
case 'p':
if ((ps_cnt = atoi_scaled(optarg)) <= 0) {
fprintf(stderr,"ERR: Bad pSync[] elements %d\n",loops);
shmem_finalize();
return 1;
}
break;
case 's':
Serialize++;
break;
case 'h':
if (me == 0)
usage(pgm);
return 0;
default:
if (me == 0) {
fprintf(stderr,"%s: unknown switch '-%c'?\n",pgm,i);
usage(pgm);
}
shmem_finalize();
return 1;
}
}
ps_cnt *= _SHMEM_BCAST_SYNC_SIZE;
pSync = shmem_malloc( ps_cnt * sizeof(long) );
for (i = 0; i < ps_cnt; i++)
pSync[i] = _SHMEM_SYNC_VALUE;
source = (int *) shmem_malloc( elements * sizeof(*source) );
target = (int *) shmem_malloc( elements * sizeof(*target) );
for (i = 0; i < elements; i += 1) {
source[i] = i + 1;
target[i] = -90;
}
if (me==0 && Verbose)
fprintf(stderr,"ps_cnt %d loops %d nElems %d\n",
ps_cnt,loops,elements);
shmem_barrier_all();
for(time_taken = 0.0, ps = i = 0; i < loops; i++) {
start_time = shmemx_wtime();
shmem_broadcast32(target, source, elements, 0, 0, 0, npes, &pSync[ps]);
if (Serialize) shmem_barrier_all();
time_taken += (shmemx_wtime() - start_time);
if (ps_cnt > 1 ) {
ps += _SHMEM_BCAST_SYNC_SIZE;
if ( ps >= ps_cnt ) ps = 0;
}
}
if(me == 0 && Verbose) {
printf("%d loops of Broadcast32(%ld bytes) over %d PEs: %7.3f secs\n",
loops, (elements*sizeof(*source)), npes, time_taken);
elements = (elements * loops * sizeof(*source)) / (1024*1024);
printf(" %7.5f secs per broadcast() @ %7.4f MB/sec\n",
(time_taken/(double)loops), ((double)elements / time_taken) );
}
if (Verbose > 1) fprintf(stderr,"[%d] pre B1\n",me);
shmem_barrier_all();
if (Verbose > 1) fprintf(stderr,"[%d] post B1\n",me);
shmem_free(pSync);
shmem_free(target);
shmem_free(source);
shmem_finalize();
return 0;
}
int main( int argc, char *argv[])
{
int rc=0, my_pe, npes, neighbor;
int loops=LOOPS;
int j;
size_t data_sz=sizeof(long) * 3;
double start_time;
long *data, lval=0;
if (argc > 1)
loops = atoi(argv[1]);
shmem_init();
my_pe = shmem_my_pe();
npes = shmem_n_pes();
data = shmem_malloc(data_sz);
if (!data) {
fprintf(stderr,"[%d] shmem_malloc(%ld) failure? %d\n",
my_pe,data_sz,errno);
shmem_global_exit(1);
}
memset((void*)data,0,data_sz);
shmem_barrier_all();
neighbor = (my_pe + 1) % npes;
start_time = shmemx_wtime();
for(j=0,elapsed=0.0; j < loops; j++) {
start_time = shmemx_wtime();
lval = shmem_long_finc( (void*)&data[1], neighbor );
elapsed += shmemx_wtime() - start_time;
if (lval != (long) j) {
fprintf(stderr,"[%d] Test: FAIL previous val %ld != %d Exit.\n",
my_pe, lval, j);
shmem_global_exit(1);
}
}
shmem_barrier_all();
rc = 0;
if (data[1] != (long)loops) {
fprintf(stderr,"[%d] finc neighbot: FAIL data[1](%p) %ld != %d Exit.\n",
my_pe, (void*)&data[1], data[1], loops);
rc--;
}
/* check if adjancent memory locations distrubed */
assert(data[0] == 0);
assert(data[2] == 0);
if (my_pe == 0 ) {
if (rc == 0 && Verbose)
fprintf(stderr,"[%d] finc neighbor: PASSED.\n",my_pe);
fprintf(stderr,"[%d] %d loops of shmem_long_finc() in %6.4f secs\n"
" %2.6f usecs per shmem_long_finc()\n",
my_pe,loops,elapsed,((elapsed*100000.0)/(double)loops));
}
shmem_free(data);
shmem_finalize();
return rc;
}
