int
main(int argc, char* argv[])
{
int c, j, loops, k, l;
int my_pe, nProcs, nWorkers;
int nWords=1;
int failures=0;
char *prog_name;
long *wp,work_sz;
for(j=0; j < SHMEM_BARRIER_SYNC_SIZE; j++) {
pSync0[j] = pSync1[j] = pSync2[j] = pSync3[j] =
pSync4[j] = SHMEM_SYNC_VALUE;
}
shmem_init();
my_pe = shmem_my_pe();
nProcs = shmem_n_pes();
nWorkers = nProcs - 1;
if (nProcs == 1) {
Rfprintf(stderr,
"ERR - Requires > 1 PEs\n");
shmem_finalize();
return 0;
}
for(j=0; j < nProcs; j++)
if ( shmem_pe_accessible(j) != 1 ) {
fprintf(stderr,
"ERR - pe %d not accessible from pe %d\n",
j, my_pe);
}
prog_name = strrchr(argv[0],'/');
if ( prog_name )
prog_name++;
else
prog_name = argv[0];
while((c=getopt(argc,argv,"hvM:s")) != -1) {
switch(c) {
case 's':
Slow++;
break;
case 'v':
Verbose++;
break;
case 'M':
output_mod = atoi(optarg);
if (output_mod <= 0) {
Rfprintf(stderr, "ERR - output modulo arg out of "
"bounds '%d'?\n", output_mod);
shmem_finalize();
return 1;
}
Rfprintf(stderr,"%s: output modulo %d\n",
prog_name,output_mod);
break;
case 'h':
Rfprintf(stderr,
"usage: %s {nWords-2-put(%d)K/M} {Loop-count(%d)K/M}\n",
prog_name, DFLT_NWORDS, DFLT_LOOPS);
shmem_finalize();
return 1;
default:
shmem_finalize();
return 1;
}
}
if (optind == argc)
nWords = DFLT_NWORDS;
else {
nWords = atoi_scaled(argv[optind++]);
if (nWords <= 0) {
Rfprintf(stderr, "ERR - Bad nWords arg '%d'?\n", nWords);
shmem_finalize();
return 1;
}
}
if (optind == argc)
loops = DFLT_LOOPS;
else {
loops = atoi_scaled(argv[optind++]);
if (loops <= 0 || loops > 1000000) {
Rfprintf(stderr,
"ERR - loops arg out of bounds '%d'?\n", loops);
shmem_finalize();
return 1;
}
}
work_sz = (nProcs*nWords) * sizeof(long);
work = shmem_malloc( work_sz );
if ( !work ) {
fprintf(stderr,"[%d] ERR - work = shmem_malloc(%ld) ?\n",my_pe,work_sz);
shmem_global_exit(1);
}
Target = shmem_malloc( 2 * nWords * sizeof(long) );
if ( !Target ) {
fprintf(stderr,"[%d] ERR - Target = shmem_malloc(%ld) ?\n",
my_pe, (nWords * sizeof(long)));
shmem_global_exit(1);
}
src = &Target[nWords];
#if _DEBUG
Rprintf("%s: %d loops of %d longs per put\n",prog_name,loops,nWords);
#endif
for(j=0; j < nWords; j++)
src[j] = VAL;
for(j=0; j < loops; j++) {
#if _DEBUG
if ( Verbose && (j==0 || (j % output_mod) == 0) )
fprintf(stderr,"[%d] +(%d)\n", my_pe,j);
#endif
shmem_barrier(0, 0, nProcs, pSync0);
if ( my_pe == 0 ) {
int p;
for(p=1; p < nProcs; p++)
shmem_long_put(Target, src, nWords, p);
}
else {
if (Slow) {
/* wait for each put to complete */
for(k=0; k < nWords; k++)
shmem_wait(&Target[k],my_pe);
} else {
/* wait for last word to be written */
shmem_wait(&Target[nWords-1],my_pe);
}
}
#if _DEBUG
if ( Verbose && (j==0 || (j % output_mod) == 0) )
fprintf(stderr,"[%d] -(%d)\n", shmem_my_pe(),j);
#endif
shmem_barrier(0, 0, nProcs, pSync1);
RDprintf("Workers[1 ... %d] verify Target data put by proc0\n",
nWorkers);
/* workers verify put data is expected */
if ( my_pe != 0 ) {
for(k=0; k < nWords; k++) {
if (Target[k] != VAL) {
fprintf(stderr, "[%d] Target[%d] %#lx "
"!= %#x?\n",
my_pe,k,Target[k],VAL);
failures++;
}
assert(Target[k] == VAL);
Target[k] = my_pe;
}
}
else /* clear results buffer, workers will put here */
memset(work, 0, work_sz);
shmem_barrier(0, 0, nProcs, pSync2);
RDprintf("Workers[1 ... %d] put Target data to PE0 work "
"vector\n",nWorkers);
if ( my_pe != 0 ) {
/* push nWords of val my_pe back to PE zero */
shmem_long_put(&work[my_pe * nWords], Target, nWords, 0);
}
else {
/* wait for procs 1 ... nProcs to complete put()s */
for(l=1; l < nProcs; l++) {
wp = &work[ l*nWords ]; // procs nWords chunk
#if 1
/* wait for last long to be written from each PE */
shmem_wait(&wp[nWords-1],0);
#else
for(k=0; k < nWords; k++)
shmem_wait(&wp[k],0);
#endif
}
}
shmem_barrier(0, 0, nProcs, pSync3);
if ( my_pe == 0 ) {
RDprintf("Loop(%d) PE0 verifing work data.\n",j);
for(l=1; l < nProcs; l++) {
wp = &work[ l*nWords ]; // procs nWords chunk
for(k=0; k < nWords; k++) {
if (wp[k] != l) {
fprintf(stderr,
"[0] PE(%d)_work[%d] %ld "
"!= %d?\n",
l,k,work[k],l);
failures++;
}
assert(wp[k] == l);
break;
}
if (failures)
break;
}
}
shmem_barrier(0, 0, nProcs, pSync4);
#if _DEBUG
if (loops > 1) {
Rfprintf(stderr,".");
RDprintf("Loop(%d) Pass.\n",j);
}
#endif
}
shmem_free( work );
shmem_free( Target );
#if _DEBUG
Rfprintf(stderr,"\n");fflush(stderr);
shmem_barrier_all();
RDprintf("%d(%d) Exit(%d)\n", my_pe, nProcs, failures);
#endif
shmem_finalize();
return failures;
}
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 c, j, cloop, loops = DFLT_LOOPS;
int mpe, num_pes;
int nWords=1;
int nIncr=1;
int failures=0;
char *pgm;
start_pes(0);
mpe = _my_pe();
num_pes = _num_pes();
if (num_pes == 1) {
Rfprintf(stderr,
"ERR - Requires > 1 PEs\n");
return 1;
}
pgm = strrchr(argv[0],'/');
if ( pgm )
pgm++;
else
pgm = argv[0];
while((c=getopt(argc,argv,"hqVvl:")) != -1) {
switch(c) {
case 'V':
case 'v':
Verbose++;
break;
case 'l':
loops = atoi(optarg);
break;
case 'h':
Rfprintf(stderr,
"usage: %s {-l loopcnt(%d)} {numLongs(%d)} {loopIncr(%d)}\n",
pgm,DFLT_LOOPS,DFLT_NWORDS,DFLT_INCR);
return 1;
default:
return 1;
}
}
if (optind == argc)
nWords = DFLT_NWORDS;
else {
nWords = atoi_scaled(argv[optind++]);
if (nWords <= 0) {
Rfprintf(stderr, "ERR - Bad nBytes arg?\n");
return 1;
}
}
if (optind == argc)
nIncr = DFLT_INCR;
else {
loops = atoi(argv[optind++]);
if (nIncr <= 0 ) {
Rfprintf(stderr, "ERR - incLongs arg out of bounds '%d'?\n", nIncr);
return 1;
}
}
if ( nWords % 8 ) { // integral multiple of longs
Rprintf("%s: nWords(%d) not a multiple of %ld?\n",
pgm,nWords,sizeof(long));
return 1;
}
for (c = 0; c < _SHMEM_COLLECT_SYNC_SIZE;c++)
pSync[c] = _SHMEM_SYNC_VALUE;
if (Verbose && mpe == 0)
fprintf(stderr,"loops(%d) nWords(%d) incr-per-loop(%d)\n",
loops,nWords,nIncr);
for(cloop=1; cloop <= loops; cloop++) {
c = (sizeof(long)*nWords) * (num_pes + 1); // src + dst allocation.
//nWords /= sizeof(long); // convert input of bytes --> longs.
src = (long*)shmalloc(c);
if ( !src ) {
Rprintf("[%d] %s: shmalloc(%d) failed?\n", mpe, pgm,c);
return 0;
}
dst = &src[nWords];
for(j=0; j < nWords; j++)
src[j] = (long) (j + mpe*nWords);
shmem_barrier_all();
shmem_fcollect64(dst,src,nWords,0,0,num_pes,pSync);
// Expect dst to be consecuative integers 0 ... (nLongs*num_pes)-1
for(j=0; j < (nWords*num_pes); j++) {
if ( dst[j] != (long) j ) {
fprintf(stderr,
"[%d] dst[%d] %ld != expected %d\n",mpe,j,dst[j],j);
return 1;
}
}
shmem_barrier_all();
if (Verbose && mpe == 0 && loops > 1) {
fprintf(stderr,".");
}
nWords += nIncr;
}
if (Verbose && mpe == 0) {
fprintf(stderr,"\n");fflush(stderr);
}
shfree( (void*)src );
shmem_barrier_all();
if (Verbose)
printf("%d(%d) Exit(%d)\n", mpe, num_pes, failures);
return failures;
}
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;
start_pes(0);
me = _my_pe();
npes = _num_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);
return 1;
}
break;
case 'l':
if ((loops = atoi_scaled(optarg)) <= 0) {
fprintf(stderr,"ERR: Bad loop count %d\n",loops);
return 1;
}
break;
case 'p':
if ((ps_cnt = atoi_scaled(optarg)) <= 0) {
fprintf(stderr,"ERR: Bad pSync[] elements %d\n",loops);
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);
}
return 1;
}
}
ps_cnt *= _SHMEM_BCAST_SYNC_SIZE;
pSync = shmalloc( ps_cnt * sizeof(long) );
for (i = 0; i < ps_cnt; i++)
pSync[i] = _SHMEM_SYNC_VALUE;
source = (int *) shmalloc( elements * sizeof(*source) );
target = (int *) shmalloc( 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 = shmem_wtime();
shmem_broadcast32(target, source, elements, 0, 0, 0, npes, &pSync[ps]);
if (Serialize) shmem_barrier_all();
time_taken += (shmem_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);
shfree(pSync);
shfree(target);
shfree(source);
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;
int *target;
int *source;
int me, npes, elements=N_ELEMENTS, loops=DFLT_LOOPS;
char *pgm;
shmem_init();
me = shmem_my_pe();
npes = shmem_n_pes();
if ((pgm=strrchr(argv[0],'/')))
pgm++;
else
pgm = argv[0];
/* lower-case switch enable only a specific test; otherwise run all tests */
while ((i = getopt (argc, argv, "hvqe:l:abcmn")) != EOF) {
switch (i)
{
case 'a':
All2++;
break;
case 'b':
Bcast++;
break;
case 'c':
Collect++;
break;
case 'm':
Many++;
break;
case 'n':
Neighbor++;
break;
case 'q':
Verbose=0;
break;
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 'h':
if (me == 0)
usage(pgm);
shmem_finalize();
return 0;
default:
if (me == 0) {
fprintf(stderr,"%s: unknown switch '-%c'?\n",pgm,i);
usage(pgm);
}
shmem_finalize();
return 1;
}
}
if (All2==0 && Bcast==0 && Collect==0 && Many==0 && Neighbor==0)
All2 = Bcast = Collect = Many = Neighbor = 1;
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;
}
shmem_barrier_all();
if (Neighbor) {
neighbor_put( target, source, elements, me, npes, loops );
neighbor_get( target, source, elements, me, npes, loops );
}
if (All2) {
all2all_put( target, source, elements, me, npes, loops );
all2all_get( target, source, elements, me, npes, loops );
}
if (Many) {
one2many_put( target, source, elements, me, npes, loops );
many2one_get( target, source, elements, me, npes, loops );
}
if (Bcast) bcast( target, source, elements, me, npes, loops );
if (Collect) {
collect( NULL, source, elements, me, npes, loops );
fcollect( NULL, source, elements, me, npes, loops );
}
shmem_barrier_all();
shmem_free(target);
shmem_free(source);
shmem_finalize();
return 0;
}
