int main(int argc, char *argv[])
{
int ndim;
armci_msg_init(&argc, &argv);
ARMCI_Init_args(&argc, &argv);
nproc = armci_msg_nproc();
me = armci_msg_me();
ARMCI_Barrier();
if (me == 0) {
printf("\nTesting armci_notify\n");
fflush(stdout);
sleep(1);
}
ARMCI_Barrier();
for (ndim = 1; ndim <= MAXDIMS; ndim++) {
test_notify(ndim);
}
ARMCI_Barrier();
ARMCI_Finalize();
armci_msg_finalize();
return(0);
}
int main(int argc, char **argv)
{
armci_msg_init(&argc,&argv);
ARMCI_Init_args(&argc, &argv);
me = armci_msg_me();
nproc = armci_msg_nproc();
/* This test only works for two processes */
assert(nproc == 2);
if (0 == me) {
printf("msg size (bytes) avg time (us) avg b/w (MB/sec)\n");
}
if (0 == me) {
printf("#PNNL comex Put Test\n");
}
contig_test(MAX_MESSAGE_SIZE, PUT);
if (0 == me) {
printf("#PNNL comex Get Test\n");
}
contig_test(MAX_MESSAGE_SIZE, GET);
if (0 == me) {
printf("#PNNL comex Accumulate Test\n");
}
contig_test(MAX_MESSAGE_SIZE, ACC);
ARMCI_Finalize();
armci_msg_finalize();
return 0;
}
int main(int argc, char **argv)
{
MP_INIT(argc,argv);
MP_MYID(&me);
MP_PROCS(&nproc);
if(nproc < 2 || nproc> MAXPROC) {
if(me == 0)
fprintf(stderr,
"USAGE: 2 <= processes < %d - got %d\n", MAXPROC, nproc);
MP_BARRIER();
MP_FINALIZE();
exit(0);
}
/* initialize ARMCI */
ARMCI_Init_args(&argc, &argv);
if(!me)printf("\n Performance of Basic Blocking Communication Operations\n");
MP_BARRIER();
CHECK_RESULT=1; test_1D(); CHECK_RESULT=0; /* warmup run */
/* test 1 dimension array */
if(!me)printf("\n\t\t\tContiguous Data Transfer\n");
test_1D();
/* test 2 dimension array */
if(!me)printf("\n\t\t\tStrided Data Transfer\n");
test_2D();
MP_BARRIER();
if(me == 0){
if(warn_accuracy)
printf("\nWARNING: Your timer does not have sufficient accuracy for this test (%d)\n",warn_accuracy);
printf("\n\n------------ Now we test the same data transfer for correctness ----------\n");
fflush(stdout);
}
MP_BARRIER();
CHECK_RESULT=1;
if(!me)printf("\n\t\t\tContiguous Data Transfer\n");
test_1D();
if(me == 0) printf("OK\n");
MP_BARRIER();
if(!me)printf("\n\t\t\tStrided Data Transfer\n");
test_2D();
if(me == 0) printf("OK\n\n\nTests Completed.\n");
MP_BARRIER();
/* done */
ARMCI_Finalize();
MP_FINALIZE();
return(0);
}
int main(int argc, char **argv)
{
int i, j, rank, nranks, msgsize, dest;
int xdim, ydim;
long bufsize;
double **buffer;
double t_start, t_stop, t_latency;
int count[2], src_stride, trg_stride, stride_level;
int provided;
MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nranks);
ARMCI_Init_args(&argc, &argv);
ARMCI_Barrier();
int me = armci_msg_me();
int node = armci_domain_my_id(ARMCI_DOMAIN_SMP);
printf("MPI_Rank: %d, \
armci_msg_nproc: %d \
armci_msg_me: %d, \
armci_domain_id: %d, \
armci_domain_same_id: %d,\
armci_domain_my_id: %d, \
armci_domain_count: %d, \
armci_domain_nprocs: %d, \
armci_domain_glob_proc_id: %d \n",
rank, armci_msg_nproc(), me, armci_domain_id(ARMCI_DOMAIN_SMP, me),
armci_domain_same_id(ARMCI_DOMAIN_SMP, me), armci_domain_my_id(ARMCI_DOMAIN_SMP),
armci_domain_count(ARMCI_DOMAIN_SMP), armci_domain_nprocs(ARMCI_DOMAIN_SMP, node),
armci_domain_glob_proc_id(ARMCI_DOMAIN_SMP, node, 0));
fflush(stdout);
ARMCI_Free((void *) buffer[rank]);
ARMCI_Finalize();
MPI_Finalize();
return 0;
}
int main(int argc, char *argv[])
{
ARMCI_Init_args(&argc, &argv);
nproc = armci_msg_nproc();
me = armci_msg_me();
/* printf("nproc = %d, me = %d\n", nproc, me);*/
if (nproc > MAXPROC && me == 0) {
ARMCI_Error("Test works for up to %d processors\n", MAXPROC);
}
if (me == 0) {
printf("ARMCI test program (%d processes)\n", nproc);
fflush(stdout);
sleep(1);
}
if (me == 0) {
printf("\nAggregate put/get requests\n\n");
fflush(stdout);
}
test_aggregate(1); /* cold start */
test_aggregate(0); /* warm start */
ARMCI_AllFence();
ARMCI_Barrier();
if (me == 0) {
printf("\nSuccess!!\n");
fflush(stdout);
}
sleep(2);
ARMCI_Barrier();
ARMCI_Finalize();
armci_msg_finalize();
return(0);
}
int main(int argc, char* argv[])
{
int provided;
int i, rank, nranks, msgsize, target;
long bufsize;
int **counter;
int *complete;
int increment;
int counter_fetch;
int counters_received;
int t_start, t_stop, t_latency;
int expected;
MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nranks);
ARMCI_Init_args(&argc, &argv);
complete = (int *) malloc(sizeof(int) * COUNT);
counter = (int**) ARMCI_Malloc_local( nranks * sizeof(int*) );
ARMCI_Malloc((void *) counter[rank], sizeof(int));
if (rank == 0)
{
printf("ARMCI_RMW Test - in usec \n");
fflush(stdout);
}
target = 0;
for(i=0; i<COUNT; i++)
{
complete[i] = 0;
}
if(rank == target)
{
*(counter[rank]) = 0;
}
increment = 1;
counter_fetch = 0;
counters_received = 0;
MPI_Barrier(MPI_COMM_WORLD);
while(counter_fetch < COUNT)
{
ARMCI_Rmw(ARMCI_FETCH_AND_ADD,
(void *) &counter_fetch,
(void *) counter[target],
increment,
target);
/* s/1/rank/ means we will know who got the counter */
if (counter_fetch < COUNT) complete[counter_fetch] = rank;
counters_received++;
}
MPI_Allreduce(MPI_IN_PLACE,complete,COUNT,MPI_INT,MPI_SUM,MPI_COMM_WORLD);
for(i=0; i<COUNT; i++)
{
if (complete[i] == 0)
{
printf("[%d] The RMW update failed at index: %d \n", rank, i);
fflush(stdout);
exit(-1);
}
}
printf("[%d] The RMW update completed successfully \n", rank);
fflush(stdout);
MPI_Barrier(MPI_COMM_WORLD);
if (0==rank)
{
printf("Checking for fairness...\n", rank);
fflush(stdout);
for(i=0; i<COUNT; i++)
{
printf("counter value %d was received by process %d\n", i, complete[i]);
}
fflush(stdout);
}
MPI_Barrier(MPI_COMM_WORLD);
printf("process %d received %d counters\n", rank, counters_received);
fflush(stdout);
ARMCI_Free(counter[rank]);
ARMCI_Free_local(counter);
ARMCI_Finalize();
MPI_Finalize();
return 0;
}
int main(int argc, char **argv)
{
int i, j, rank, nranks, peer;
size_t xdim, ydim;
unsigned long bufsize;
double **buffer, *src_buf;
double t_start=0.0, t_stop;
int count[2], src_stride, trg_stride, stride_level;
double scaling;
int provided;
MPI_Init_thread(&argc, &argv, MPI_THREAD_SINGLE, &provided);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nranks);
if (nranks < 2) {
printf("%s: Must be run with at least 2 processes\n", argv[0]);
MPI_Abort(MPI_COMM_WORLD, 1);
}
ARMCI_Init_args(&argc, &argv);
buffer = (double **) malloc(sizeof(double *) * nranks);
bufsize = MAX_XDIM * MAX_YDIM * sizeof(double);
ARMCI_Malloc((void **) buffer, bufsize);
src_buf = ARMCI_Malloc_local(bufsize);
if (rank == 0)
{
printf("ARMCI_AccS Latency - local and remote completions - in usec \n");
printf("%30s %22s %22s\n",
"Dimensions(array of double)",
"Local Completion",
"Remote completion");
fflush(stdout);
}
ARMCI_Access_begin(buffer[rank]);
for (i = 0; i < bufsize / sizeof(double); i++)
{
*(buffer[rank] + i) = 1.0 + rank;
*(src_buf + i) = 1.0 + rank;
}
ARMCI_Access_end(buffer[rank]);
scaling = 2.0;
src_stride = MAX_YDIM * sizeof(double);
trg_stride = MAX_YDIM * sizeof(double);
stride_level = 1;
ARMCI_Barrier();
for (xdim = 1; xdim <= MAX_XDIM; xdim *= 2)
{
count[1] = xdim;
for (ydim = 1; ydim <= MAX_YDIM; ydim *= 2)
{
count[0] = ydim * sizeof(double);
if (rank == 0)
{
peer = 1;
for (i = 0; i < ITERATIONS + SKIP; i++)
{
if (i == SKIP) t_start = MPI_Wtime();
ARMCI_AccS(ARMCI_ACC_DBL,
(void *) &scaling,
/* (void *) buffer[rank] */ src_buf,
&src_stride,
(void *) buffer[peer],
&trg_stride,
count,
stride_level,
1);
}
t_stop = MPI_Wtime();
ARMCI_Fence(1);
char temp[10];
sprintf(temp, "%dX%d", (int) xdim, (int) ydim);
printf("%30s %20.2f ", temp, ((t_stop - t_start) * 1000000)
/ ITERATIONS);
fflush(stdout);
ARMCI_Barrier();
ARMCI_Barrier();
for (i = 0; i < ITERATIONS + SKIP; i++)
{
if (i == SKIP) t_start = MPI_Wtime();
ARMCI_AccS(ARMCI_ACC_DBL,
(void *) &scaling,
/* (void *) buffer[rank] */ src_buf,
&src_stride,
(void *) buffer[peer],
&trg_stride,
count,
stride_level,
1);
ARMCI_Fence(1);
}
t_stop = MPI_Wtime();
printf("%20.2f \n", ((t_stop - t_start) * 1000000) / ITERATIONS);
fflush(stdout);
ARMCI_Barrier();
ARMCI_Barrier();
}
else
{
peer = 0;
ARMCI_Barrier();
if (rank == 1)
{
ARMCI_Access_begin(buffer[rank]);
for (i = 0; i < xdim; i++)
{
for (j = 0; j < ydim; j++)
{
if (*(buffer[rank] + i * MAX_XDIM + j) != ((1.0 + rank)
+ scaling * (1.0 + peer) * (ITERATIONS + SKIP)))
{
printf("Data validation failed at X: %d Y: %d Expected : %f Actual : %f \n",
i,
j,
((1.0 + rank) + scaling * (1.0 + peer)),
*(buffer[rank] + i * MAX_YDIM + j));
fflush(stdout);
ARMCI_Error("Bailing out", 1);
}
}
}
for (i = 0; i < bufsize / sizeof(double); i++)
{
*(buffer[rank] + i) = 1.0 + rank;
}
ARMCI_Access_end(buffer[rank]);
}
ARMCI_Barrier();
ARMCI_Barrier();
if (rank == 1)
{
ARMCI_Access_begin(buffer[rank]);
for (i = 0; i < xdim; i++)
{
for (j = 0; j < ydim; j++)
{
if (*(buffer[rank] + i * MAX_XDIM + j) != ((1.0 + rank)
+ scaling * (1.0 + peer) * (ITERATIONS + SKIP)))
{
printf("Data validation failed at X: %d Y: %d Expected : %f Actual : %f \n",
i,
j,
((1.0 + rank) + scaling * (1.0 + peer)),
*(buffer[rank] + i * MAX_YDIM + j));
fflush(stdout);
ARMCI_Error("Bailing out", 1);
}
}
}
for (i = 0; i < bufsize / sizeof(double); i++)
{
*(buffer[rank] + i) = 1.0 + rank;
}
ARMCI_Access_end(buffer[rank]);
}
ARMCI_Barrier();
}
}
}
ARMCI_Barrier();
ARMCI_Free((void *) buffer[rank]);
ARMCI_Free_local(src_buf);
free(buffer);
ARMCI_Finalize();
MPI_Finalize();
return 0;
}
int main(int argc, char *argv[]) {
int i, j, rank, nranks;
int xdim, ydim;
long bufsize;
double **buffer;
double t_start=0.0, t_stop=0.0;
int count[2], src_stride, trg_stride, stride_level, peer;
double expected, actual;
int provided;
MPI_Init_thread(&argc, &argv, MPI_THREAD_SINGLE, &provided);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nranks);
if (nranks < 2) {
printf("%s: Must be run with at least 2 processes\n", argv[0]);
MPI_Abort(MPI_COMM_WORLD, 1);
}
ARMCI_Init_args(&argc, &argv);
bufsize = MAX_XDIM * MAX_YDIM * sizeof(double);
buffer = (double **) malloc(sizeof(double *) * nranks);
ARMCI_Malloc((void **) buffer, bufsize);
for(i=0; i< bufsize/sizeof(double); i++) {
*(buffer[rank] + i) = 1.0 + rank;
}
if(rank == 0) {
printf("ARMCI_PutS Latency - local and remote completions - in usec \n");
printf("%30s %22s %22s\n", "Dimensions(array of doubles)", "Latency-LocalCompeltion", "Latency-RemoteCompletion");
fflush(stdout);
}
src_stride = MAX_YDIM*sizeof(double);
trg_stride = MAX_YDIM*sizeof(double);
stride_level = 1;
ARMCI_Barrier();
for(xdim=1; xdim<=MAX_XDIM; xdim*=2) {
count[1] = xdim;
for(ydim=1; ydim<=MAX_YDIM; ydim*=2) {
count[0] = ydim*sizeof(double);
if(rank == 0)
{
peer = 1;
for(i=0; i<ITERATIONS+SKIP; i++) {
if(i == SKIP)
t_start = MPI_Wtime();
ARMCI_PutS((void *) buffer[rank], &src_stride, (void *) buffer[peer], &trg_stride, count, stride_level, peer);
}
t_stop = MPI_Wtime();
ARMCI_Fence(peer);
char temp[10];
sprintf(temp,"%dX%d", xdim, ydim);
printf("%30s %20.2f", temp, ((t_stop-t_start)*1000000)/ITERATIONS);
fflush(stdout);
ARMCI_Barrier();
ARMCI_Barrier();
for(i=0; i<ITERATIONS+SKIP; i++) {
if(i == SKIP)
t_start = MPI_Wtime();
ARMCI_PutS((void *) buffer[rank], &src_stride, (void *) buffer[peer], &trg_stride, count, stride_level, peer);
ARMCI_Fence(peer);
}
t_stop = MPI_Wtime();
printf("%20.2f \n", ((t_stop-t_start)*1000000)/ITERATIONS);
fflush(stdout);
ARMCI_Barrier();
ARMCI_Barrier();
}
else
{
peer = 0;
expected = (1.0 + (double) peer);
ARMCI_Barrier();
if (rank == 1)
{
for(i=0; i<xdim; i++)
{
for(j=0; j<ydim; j++)
{
actual = *(buffer[rank] + i*MAX_YDIM + j);
if(actual != expected)
{
printf("Data validation failed at X: %d Y: %d Expected : %f Actual : %f \n",
i, j, expected, actual);
fflush(stdout);
ARMCI_Error("Bailing out", 1);
}
}
}
}
for(i=0; i< bufsize/sizeof(double); i++) {
*(buffer[rank] + i) = 1.0 + rank;
}
ARMCI_Barrier();
ARMCI_Barrier();
if (rank == 1)
{
for(i=0; i<xdim; i++)
{
for(j=0; j<ydim; j++)
{
actual = *(buffer[rank] + i*MAX_YDIM + j);
if(actual != expected)
{
printf("Data validation failed at X: %d Y: %d Expected : %f Actual : %f \n",
i, j, expected, actual);
fflush(stdout);
ARMCI_Error("Bailing out", 1);
}
}
}
for(i=0; i< bufsize/sizeof(double); i++) {
*(buffer[rank] + i) = 1.0 + rank;
}
}
ARMCI_Barrier();
}
}
}
ARMCI_Barrier();
ARMCI_Free((void *) buffer[rank]);
free(buffer);
ARMCI_Finalize();
MPI_Finalize();
return 0;
}
int main(int argc, char *argv[])
{
size_t i, rank, nranks, msgsize, dest;
size_t iterations, max_msgsize;
int bufsize;
double **buffer;
double t_start, t_stop, t_total, d_total;
double expected, bandwidth;
int provided;
armci_hdl_t handle;
max_msgsize = MAX_MSGSIZE;
MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nranks);
ARMCI_Init_args(&argc, &argv);
bufsize = max_msgsize * ITERATIONS_LARGE;
buffer = (double **) malloc(sizeof(double *) * nranks);
ARMCI_Malloc((void **) buffer, bufsize);
for (i = 0; i < bufsize / sizeof(double); i++)
{
*(buffer[rank] + i) = 1.0 + rank;
}
ARMCI_INIT_HANDLE(&handle);
ARMCI_SET_AGGREGATE_HANDLE(&handle);
ARMCI_Barrier();
if (rank == 0)
{
printf("ARMCI_Put Bandwidth in MBPS \n");
printf("%20s %22s \n", "Message Size", "Bandwidth");
fflush(stdout);
dest = 1;
expected = 1 + dest;
for (msgsize = sizeof(double); msgsize <= max_msgsize; msgsize *= 2)
{
if (msgsize <= 16 * 1024) iterations = ITERATIONS_VERYSMALL;
else if (msgsize <= 64 * 1024) iterations = ITERATIONS_SMALL;
else if (msgsize <= 512 * 1024) iterations = ITERATIONS_MEDIUM;
else iterations = ITERATIONS_LARGE;
t_start = MPI_Wtime();
for (i = 0; i < iterations; i++)
{
ARMCI_NbPut((void *) ((size_t) buffer[dest] + (size_t)(i
* msgsize)), (void *) ((size_t) buffer[rank]
+ (size_t)(i * msgsize)), msgsize, dest, &handle);
}
ARMCI_Wait(&handle);
t_stop = MPI_Wtime();
d_total = (iterations * msgsize) / (1024 * 1024);
t_total = t_stop - t_start;
bandwidth = d_total / t_total;
printf("%20d %20.4lf \n", msgsize, bandwidth);
fflush(stdout);
ARMCI_Fence(dest);
}
}
ARMCI_Barrier();
ARMCI_UNSET_AGGREGATE_HANDLE(&handle);
ARMCI_Free((void *) buffer[rank]);
ARMCI_Finalize();
MPI_Finalize();
return 0;
}
int main(int argc, char **argv)
{
int i, j, rank, nranks, msgsize;
int *buffer;
int provided;
char op = '+';
MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nranks);
ARMCI_Init_args(&argc, &argv);
ARMCI_Barrier();
buffer = (int *) malloc(MAX_MSG_SIZE);
for(i=0; i<MAX_MSG_SIZE/sizeof(int); i++)
{
if(rank == 0)
buffer[i] = (2<<20 - 1);
else
buffer[i] = 0;
}
if(rank == 0)
{
printf("Testing functionality of ARMCI_Bcast \n");
fflush(stdout);
}
for(msgsize=sizeof(int); msgsize<=MAX_MSG_SIZE; msgsize*=2)
{
armci_msg_bcast(buffer, msgsize, 0);
for(i=0; i<msgsize/sizeof(int); i++)
{
if(buffer[i] != (2<<20 - 1))
{
printf("[%d] Validation failed for msg size: %d at index: %d expected: %d actual: %d \n",
rank, msgsize, i, (2<<20 - 1), buffer[i]);
fflush(stdout);
exit(-1);
}
}
for(i=0; i<MAX_MSG_SIZE/sizeof(int); i++)
{
if(rank == 0)
buffer[i] = (2<<20 - 1);
else
buffer[i] = 0;
}
ARMCI_Barrier();
if(rank == 0)
{
printf("Validation successful for msg size: %d\n", msgsize);
fflush(stdout);
}
}
free(buffer);
ARMCI_Finalize();
MPI_Finalize();
return 0;
}
int main(int argc, char *argv[])
{
int ch;
extern char *optarg;
int i, j, r;
thread_t threads[MAX_TPP];
/* init ARMCI */
ARMCI_Init_args(&argc, &argv);
size = armci_msg_nproc();
rank = armci_msg_me();
while ((ch = getopt(argc, argv, "t:s:i:d:h")) != -1) {
switch (ch) {
case 't': /* # of threads */
tpp = atoi(optarg);
if (tpp < 1 || tpp > MAX_TPP) {
PRINTF0("\"%s\" is improper value for -t, should be a "
"number between 1 and %d(MAX_TPP)\n",
optarg, MAX_TPP);
usage();
}
break;
case 'i': /* # of iterations */
iters = atoi(optarg);
if (iters < 1) {
PRINTF0("\"%s\" is improper value for -t, should be a "
"number equal or larger than 1\n", optarg);
usage();
}
break;
case 's': /* # of elements in the array */
asize = atoi(optarg);
if (iters < 1) {
PRINTF0("\"%s\" is improper value for -s, should be a "
"number equal or larger than 1\n", optarg);
usage();
}
break;
case 'd':
delay = atoi(optarg);
break; /* delay before start */
case 'h':
usage();
break; /* print usage info */
}
}
#ifdef NOTHREADS
tpp = 1;
PRINTF0("Warning: NOTHREADS debug symbol is set -- running w/o threads\n");
#endif
th_size = size * tpp;
PRINTF0("\nTest of multi-threaded capabilities:\n"
"%d threads per process (%d threads total),\n"
"%d array elements of size %d,\n"
"%d iteration(s)\n\n", tpp, th_size, asize, sizeof(atype_t), iters);
if (delay) {
printf("%d: %d\n", rank, getpid());
fflush(stdout);
sleep(delay);
ARMCI_Barrier();
}
TH_INIT(size, tpp);
for (i = 0; i < tpp; i++) {
th_rank[i] = rank * tpp + i;
}
#if defined(DEBUG) && defined(LOG2FILE)
for (i = 0; i < tpp; i++) {
fname[10] = '0' + th_rank[i] / 100;
fname[11] = '0' + th_rank[i] % 100 / 10;
fname[12] = '0' + th_rank[i] % 10;
dbg[i] = fopen(fname, "w");
}
#endif
for (i = 0; i < tpp; i++) {
prndbg(i, "proc %d, thread %d(%d):\n", rank, i, th_rank[i]);
}
/* set global seed (to ensure same random sequence across procs) */
time_seed = (unsigned)time(NULL);
armci_msg_brdcst(&time_seed, sizeof(time_seed), 0);
srand(time_seed);
rand();
prndbg(0, "seed = %u\n", time_seed);
/* random pairs */
pairs = calloc(th_size, sizeof(int));
for (i = 0; i < th_size; i++) {
pairs[i] = -1;
}
for (i = 0; i < th_size; i++) {
if (pairs[i] != -1) {
continue;
}
r = RND(0, th_size);
while (i == r || pairs[r] != -1) {
r = RND(0, th_size);
}
pairs[i] = r;
pairs[r] = i;
}
for (i = 0, cbufl = 0; i < th_size; i++)
cbufl += sprintf(cbuf + cbufl, " %d->%d|%d->%d",
i, pairs[i], pairs[i], pairs[pairs[i]]);
prndbg(0, "random pairs:%s\n", cbuf);
/* random targets */
rnd_tgts = calloc(th_size, sizeof(int));
for (i = 0, cbufl = 0; i < th_size; i++) {
rnd_tgts[i] = RND(0, th_size);
if (rnd_tgts[i] == i) {
i--;
continue;
}
cbufl += sprintf(cbuf + cbufl, " %d", rnd_tgts[i]);
}
prndbg(0, "random targets:%s\n", cbuf);
/* random one */
rnd_one = RND(0, th_size);
prndbg(0, "random one = %d\n", rnd_one);
assert(ptrs1 = calloc(th_size, sizeof(void *)));
assert(ptrs2 = calloc(th_size, sizeof(void *)));
#ifdef NOTHREADS
thread_main((void *)(long)0);
#else
for (i = 0; i < tpp; i++) {
THREAD_CREATE(threads + i, thread_main, (void *)(long)i);
}
for (i = 0; i < tpp; i++) {
THREAD_JOIN(threads[i], NULL);
}
#endif
ARMCI_Barrier();
PRINTF0("Tests Completed\n");
/* clean up */
#if defined(DEBUG) && defined(LOG2FILE)
for (i = 0; i < tpp; i++) {
fclose(dbg[i]);
}
#endif
ARMCI_Finalize();
TH_FINALIZE();
armci_msg_finalize();
return 0;
}
int main(int argc, char **argv)
{
int i, rank, nranks, msgsize, dest;
long bufsize;
double **buffer;
double t_start, t_stop, t_latency;
int provided;
ARMCI_Init_args(&argc, &argv);
rank = A1_Process_id(A1_GROUP_WORLD);
nranks = A1_Process_total(A1_GROUP_WORLD);
bufsize = MAX_MSG_SIZE * (ITERATIONS + SKIP);
buffer = (double **) malloc(sizeof(double *) * nranks);
ARMCI_Malloc((void **) buffer, bufsize);
for (i = 0; i < bufsize / sizeof(double); i++)
{
*(buffer[rank] + i) = 1.0 + rank;
}
A1_Barrier_group(A1_GROUP_WORLD);
if (rank == 0)
{
printf("ARMCI_Get Latency in usec \n");
printf("%20s %22s \n", "Message Size", "Latency");
fflush(stdout);
dest = 1;
for (msgsize = sizeof(double); msgsize <= MAX_MSG_SIZE; msgsize *= 2)
{
for (i = 0; i < ITERATIONS + SKIP; i++)
{
if (i == SKIP) t_start = A1_Time_seconds();
ARMCI_Get((void *) ((size_t) buffer[dest] + (size_t)(i
* msgsize)), (void *) ((size_t) buffer[rank]
+ (size_t)(i * msgsize)), msgsize, 1);
}
t_stop = A1_Time_seconds();
printf("%20d %20.2f \n", msgsize, ((t_stop - t_start) * 1000000)
/ ITERATIONS);
fflush(stdout);
for (i = 0; i < ((ITERATIONS + SKIP) * msgsize) / sizeof(double); i++)
{
if (*(buffer[rank] + i) != (1.0 + dest))
{
printf("Data validation failed At displacement : %d Expected : %f Actual : %f \n",
i,
(1.0 + dest),
*(buffer[rank] + i));
fflush(stdout);
return -1;
}
}
for (i = 0; i < bufsize / sizeof(double); i++)
{
*(buffer[rank] + i) = 1.0 + rank;
}
}
}
A1_Barrier_group(A1_GROUP_WORLD);
ARMCI_Free(buffer[rank]);
ARMCI_Finalize();
return 0;
}
int main(int argc, char *argv[])
{
int rank, nranks;
size_t i, msgsize, dest;
size_t iterations, max_msgsize;
int bufsize;
double **buffer;
double t_start, t_stop, t_total, d_total;
double expected, bandwidth;
int provided;
armci_hdl_t handle;
MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nranks);
max_msgsize = MAX_MSGSIZE;
ARMCI_Init_args(&argc, &argv);
bufsize = max_msgsize * ITERATIONS;
buffer = (double **) malloc(sizeof(double *) * nranks);
ARMCI_Malloc((void **) buffer, bufsize);
for (i = 0; i < bufsize / sizeof(double); i++)
{
*(buffer[rank] + i) = 1.0 + rank;
}
ARMCI_INIT_HANDLE(&handle);
ARMCI_SET_AGGREGATE_HANDLE(&handle);
ARMCI_Barrier();
if (rank == 0)
{
printf("ARMCI_Get Bandwidth in MBPS \n");
printf("%20s %22s \n", "Message Size", "Bandwidth");
fflush(stdout);
dest = 1;
expected = 1 + dest;
for (msgsize = sizeof(double); msgsize <= max_msgsize; msgsize *= 2)
{
iterations = bufsize/msgsize;
t_start = MPI_Wtime();
for (i = 0; i < iterations; i++)
{
ARMCI_NbGet((void *) ((size_t) buffer[dest] + (size_t)(i
* msgsize)), (void *) ((size_t) buffer[rank]
+ (size_t)(i * msgsize)), msgsize, dest, &handle);
}
ARMCI_Wait(&handle);
t_stop = MPI_Wtime();
d_total = (iterations * msgsize) / (1024 * 1024);
t_total = t_stop - t_start;
bandwidth = d_total / t_total;
printf("%20d %20.4lf \n", msgsize, bandwidth);
fflush(stdout);
#ifdef DATA_VALIDATION
{
for(j=0; j<((iterations*msgsize)/sizeof(double)); j++)
{
if(*(buffer[rank] + j) != expected)
{
printf("Data validation failed At displacement : %d Expected : %lf Actual : %lf \n",
j, expected, *(buffer[rank] + j));
fflush(stdout);
return -1;
}
}
for(j=0; j<bufsize/sizeof(double); j++)
{
*(buffer[rank] + j) = 1.0 + rank;
}
}
#endif
}
}
ARMCI_Barrier();
ARMCI_UNSET_AGGREGATE_HANDLE(&handle);
ARMCI_Free((void *) buffer[rank]);
ARMCI_Finalize();
MPI_Finalize();
return 0;
}
int main(int argc, char *argv[])
{
int i, j, rank, nranks, msgsize, dest;
int dim, iterations;
long bufsize;
double **buffer;
double t_start, t_stop, t_total, d_total, bw;
int count[2], src_stride, trg_stride, stride_level;
int provided;
armci_hdl_t handle;
MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nranks);
ARMCI_Init_args(&argc, &argv);
bufsize = MAX_DIM * MAX_DIM * sizeof(double);
buffer = (double **) malloc(sizeof(double *) * nranks);
ARMCI_Malloc((void **) buffer, bufsize);
for (i = 0; i < bufsize / sizeof(double); i++)
{
*(buffer[rank] + i) = 1.0 + rank;
}
ARMCI_INIT_HANDLE(&handle);
ARMCI_SET_AGGREGATE_HANDLE(&handle);
ARMCI_Barrier();
if (rank == 0)
{
printf("ARMCI_PutS Bandwidth in MBPS \n");
printf("%30s %22s \n", "Dimensions(array of doubles)", "Latency");
fflush(stdout);
dest = 1;
src_stride = MAX_DIM * sizeof(double);
trg_stride = MAX_DIM * sizeof(double);
stride_level = 1;
for (dim = 1; dim <= MAX_DIM; dim *= 2)
{
count[0] = dim*sizeof(double);
count[1] = dim;
iterations = 10*(MAX_DIM * MAX_DIM)/(dim * dim);
t_start = MPI_Wtime();
for (i = 0; i < iterations; i++)
{
ARMCI_NbPutS((void *) buffer[rank],
&src_stride,
(void *) buffer[dest],
&trg_stride,
count,
stride_level,
dest,
&handle);
}
ARMCI_Wait(&handle);
t_stop = MPI_Wtime();
ARMCI_Fence(1);
char temp[10];
sprintf(temp, "%dX%d", dim, dim);
t_total = t_stop - t_start;
d_total = (dim*dim*sizeof(double)*iterations)/(1024*1024);
bw = d_total/t_total;
printf("%30s %20.2f \n", temp, bw);
fflush(stdout);
}
}
ARMCI_Barrier();
ARMCI_UNSET_AGGREGATE_HANDLE(&handle);
ARMCI_Free((void *) buffer[rank]);
ARMCI_Finalize();
MPI_Finalize();
return 0;
}
int main(int argc, char *argv[])
{
int rc, i, j = 0, rid, ret;
armci_ckpt_ds_t ckptds;
ARMCI_Group grp;
ARMCI_Init_args(&argc, &argv);
nproc = armci_msg_nproc();
me = armci_msg_me();
if (me == 0) {
if (nproc > MAXPROCS) {
ARMCI_Error("nproc > MAXPROCS", nproc);
}
else {
printf("ARMCI test program (%d processes)\n", nproc);
fflush(stdout);
sleep(1);
}
}
armci_init_checkpoint2();
ARMCI_Group_get_world(&grp);
size = SIZE_;
rc = ARMCI_Malloc((void **)ptr_arr, size * 8);
printf("ARMCI test program (%d processes)\n", nproc);
fflush(stdout);
for (size = 1; size <= SIZE_; size *= 2) {
t1 = MPI_Wtime();
for (i = 0; i < 5; i++) {
for (rc = 0; rc < 15; rc++) {
do_work(size);
}
}
time_array[j++] = MPI_Wtime() - t1;
ARMCI_Barrier();
printf("%d:done for size %ld\n", me, size);
fflush(stdout);
}
(void)ARMCI_Ckpt_create_ds(&ckptds, 1);
ckptds.ptr_arr[0] = ptr_arr[me];
ckptds.sz[0] = SIZE_ * 8;
rid = ARMCI_Ckpt_init(NULL, &grp, 1, 0, &ckptds);
printf("%d: After ARMCI_Ckpt_init(): \n", me);
j = 0;
for (size = 128; size <= SIZE_; size *= 2) {
int rc;
int simulate_restart = 1;
t1 = MPI_Wtime();
ret = ARMCI_Ckpt(rid);
if (ret == ARMCI_CKPT) {
printf("%d: Performed CHECKPOINT @ size=%ld\n", me, size);
}
else if (ret == ARMCI_RESTART) {
simulate_restart = 0;
printf("%d: Performed RESTART @ size=%ld\n", me, size);
}
for (i = 0; i < 5; i++) {
for (rc = 0; rc < 15; rc++)
if (i == 3 && rc == 10) {
}
do_work(size);
}
time_array1[j++] = MPI_Wtime() - t1;
sleep(1);
if (simulate_restart && size == FAILURE_SIZE_) {
printf("%d: Simulating FAILURE @ size = %d\n", me, size);
ARMCI_Restart_simulate(rid, 1);
}
printf("%d: DONE for size=%ld regular=%f withckpt=%f\n\n",
me, size, time_array[j-1], time_array1[j-1]);
fflush(stdout);
}
ARMCI_Ckpt_finalize(rid);
printf("Before Finalize()\n");
ARMCI_Barrier();
ARMCI_Finalize();
armci_msg_finalize();
return(0);
}
int main(int argc, char *argv[])
{
int i, j;
int ch;
int edge;
int size;
/* ARMCI */
void **ptr;
double **ptr_loc;
armci_msg_init(&argc,&argv);
nproc = armci_msg_nproc();
me = armci_msg_me();
while ((ch = getopt(argc, argv, "n:b:p:h")) != -1) {
switch(ch) {
case 'n':
n = atoi(optarg);
break;
case 'b':
block_size = atoi(optarg);
break;
case 'p':
nproc = atoi(optarg);
break;
case 'h': {
printf("Usage: LU, or \n");
printf(" LU -nMATRIXSIZE -bBLOCKSIZE -pNPROC\n");
armci_msg_barrier();
armci_msg_finalize();
exit(0);
}
}
}
if(me == 0) {
printf("\n Blocked Dense LU Factorization\n");
printf(" %d by %d Matrix\n", n, n);
printf(" %d Processors\n", nproc);
printf(" %d by %d Element Blocks\n", block_size, block_size);
printf("\n");
}
/* num_rows = (int) sqrt((double) nproc); */
/* for (;;) { */
/* num_cols = nproc/num_rows; */
/* if (num_rows*num_cols == nproc) */
/* break; */
/* num_rows--; */
/* } */
nblocks = n/block_size;
if (block_size * nblocks != n) {
nblocks++;
}
nnodes = nproc / 4;
if((nnodes * 4) != nproc) {
num_cols = nproc - nnodes * 4;
nnodes++;
num_rows = 1;
}
else {
num_cols = 2;
num_rows = 2;
}
num = (nblocks * nblocks)/nnodes;
if((num * nnodes) != (nblocks * nblocks))
num++;
#ifdef DEBUG
if(me == 0)
for (i=0; i<nblocks; i++) {
for (j=0; j<nblocks; j++)
printf("%d ", block_owner(i, j));
printf("\n");
}
armci_msg_barrier();
armci_msg_finalize();
exit(0);
#endif
edge = n%block_size;
if (edge == 0) {
edge = block_size;
}
for (i=0; i<nblocks; i++) {
for (j=0; j<nblocks; j++) {
if(block_owner(i,j) == me) {
if ((i == nblocks-1) && (j == nblocks-1)) {
size = edge*edge;
}
else if ((i == nblocks-1) || (j == nblocks-1)) {
size = edge*block_size;
}
else {
size = block_size*block_size;
}
proc_bytes += size*sizeof(double);
}
}
}
/* initialize ARMCI */
ARMCI_Init_args(&argc, &argv);
ptr = (void **)malloc(nproc * sizeof(void *));
ARMCI_Malloc(ptr, proc_bytes);
a = (double **)malloc(nblocks*nblocks*sizeof(double *));
if (a == NULL) {
fprintf(stderr, "Could not malloc memory for a\n");
exit(-1);
}
ptr_loc = (double **)malloc(nproc*sizeof(double *));
for(i=0; i<nproc; i++) ptr_loc[i] = (double *)ptr[i];
for(i=0; i<nblocks; i ++) {
for(j=0; j<nblocks; j++) {
a[i+j*nblocks] = ptr_loc[block_owner(i, j)];
if ((i == nblocks-1) && (j == nblocks-1)) {
size = edge*edge;
} else if ((i == nblocks-1) || (j == nblocks-1)) {
size = edge*block_size;
} else {
size = block_size*block_size;
}
ptr_loc[block_owner(i, j)] += size;
}
}
/* initialize the array */
init_array();
/* barrier to ensure all initialization is done */
armci_msg_barrier();
/* to remove cold-start misses, all processors touch their own data */
touch_array(block_size, me);
armci_msg_barrier();
if(doprint) {
if(me == 0) {
printf("Matrix before LU decomposition\n");
print_array(me);
}
armci_msg_barrier();
}
/* Starting the timer */
if(me == 0) start_timer();
lu(n, block_size, me);
armci_msg_barrier();
/* Timer Stops here */
if(me == 0)
printf("\nRunning time = %f milliseconds.\n\n", elapsed_time());
if(doprint) {
if(me == 0) {
printf("after LU\n");
print_array(me);
}
armci_msg_barrier();
}
/* done */
ARMCI_Free(ptr[me]);
ARMCI_Finalize();
armci_msg_finalize();
return 0;
}
