/** Log performance counters into "pstats.dat" */
static void log_counters( FILE *out, const char *phase, long long *pTic, long long values[] ) {
long long toc = PAPI_get_real_usec();
#ifdef CACHE_PROFILE
if( PAPI_read_counters( values, 4 ) != PAPI_OK ) {
handle_error( 1 );
}
fprintf( out, "%s PAPI_L2_TCM %lld\n", phase, values[0] );
fprintf( out, "%s PAPI_L2_TCA %lld\n", phase, values[2] );
fprintf( out, "%s L2MissRate %.4lf%\n", phase, ( double )values[0] / ( double )values[2] );
fprintf( out, "%s PAPI_L3_TCM %lld\n", phase, values[1] );
fprintf( out, "%s PAPI_L3_TCA %lld\n", phase, values[3] );
fprintf( out, "%s L3MissRate %.4lf%\n", phase, ( double )values[1] / ( double )values[3] );
#else
if( PAPI_read_counters( values, 1 ) != PAPI_OK ) {
handle_error( 1 );
}
fprintf( out, "%s PAPI_FP_OPS %lld\n", phase, values[0] );
#endif
fprintf( out, "%s RealTime %.4lfs\n", phase, ( toc - *pTic ) * 1e-6 );
*pTic = PAPI_get_real_usec();
}
int main(int argc, char *argv[]) {
double a[MAXVSIZE], b[MAXVSIZE], c[MAXVSIZE];
int i,n;
long long before, after;
if (PAPI_VER_CURRENT !=
PAPI_library_init(PAPI_VER_CURRENT))
ehandler("PAPI_library_init error.");
const size_t EVENT_MAX = PAPI_num_counters();
printf("# Max counters = %zd\n", EVENT_MAX);
if (PAPI_OK != PAPI_query_event(PAPI_TOT_INS))
ehandler("Cannot count PAPI_TOT_INS.");
if (PAPI_OK != PAPI_query_event(PAPI_FP_OPS))
ehandler("Cannot count PAPI_FP_OPS.");
size_t EVENT_COUNT = 2;
int events[] = { PAPI_TOT_INS, PAPI_FP_OPS };
long long values[EVENT_COUNT];
printf("Enter vector size: ");
scanf("%d",&n);
for (i=0;i<n;i++) {
a[i] = i;
b[i] = n-i;
}
PAPI_start_counters(events, EVENT_COUNT);
if(PAPI_OK != PAPI_read_counters(values, EVENT_COUNT))
ehandler("Problem reading counters.");
loop(c,a,b,n);
if(PAPI_OK != PAPI_read_counters(values, EVENT_COUNT))
ehandler("Problem reading counters.");
printf("Number of instructions = %lld\n",values[0]);
printf("Number of fp operations = %lld\n",values[1]);
return 0;
}
double
papi_generic_call()
{
long_long tmp[1];
/* Read and reset the counters.
* The commented out conditional affects the reading of the performance
* counters, but might be good for debugging.
* NOTE: PAPI_accum_counters does not work properly.
* */
#if 0
if (PAPI_read_counters(tmp, 1) != PAPI_OK)
papi_eprintf("Problem reading counters %s:%d.\n", __FILE__, __LINE__);
#else
PAPI_read_counters(tmp, 1);
#endif
return ((double) tmp[0]);
}
int main(int argc, char **argv) {
int retval;
retval = PAPI_library_init(PAPI_VER_CURRENT);
if (retval != PAPI_VER_CURRENT) {
fprintf(stderr,"Error! PAPI_library_init %d\n", retval);
}
retval = PAPI_query_event(PAPI_TOT_INS);
if (retval != PAPI_OK) {
fprintf(stderr,"PAPI_TOT_INS not supported\n");
exit(1);
}
int i;
int events[1],result;
long long counts[1];
long long total=0,average,max=0,min=0x7ffffffffffffffULL;
events[0]=PAPI_TOT_INS;
PAPI_start_counters(events,1);
for(i=0;i<NUM_RUNS;i++) {
result=instructions_million();
PAPI_read_counters(counts,1);
results[i]=counts[0];
}
PAPI_stop_counters(counts,1);
PAPI_shutdown();
for(i=0;i<NUM_RUNS;i++) {
total+=results[i];
if (results[i]>max) max=results[i];
if (results[i]<min) min=results[i];
}
average=total/NUM_RUNS;
printf("Average=%lld max=%lld min=%lld\n",average,max,min);
(void) result;
return 0;
}
int main (int argc, char *argv[]) {
int i, count;
int *array = (int*) malloc (SIZE * sizeof(int));
uint64_t start, end;
int events[3] = { PAPI_L1_DCM, PAPI_L2_DCM, PAPI_L3_DCM };
long long misses[3];
int papilevels = 3;
if (PAPI_library_init(PAPI_VER_CURRENT) != PAPI_VER_CURRENT) {
exit(1);
}
//Initialization
count = 0;
srand(time(NULL));
init_time();
for (i = 0; i < SIZE; i++)
array[i] = rand();
// Uncomment this line to
qsort(array, SIZE, sizeof (int), compare_ints);
//Measurement
while (PAPI_start_counters(events, papilevels) != PAPI_OK) {
papilevels--;
}
start = get_time();
/*
* É possível, em um vetor ordenado, fazer a contagem
* em tempo O(lg(n)) em vez de O(n) utilizando busca
* binária. Embora isto diminua sensivelmente o tempo
* este não é o intuito deste exercício.
*
*/
for (i = 0; i < SIZE; i++)
if (array[i] < RAND_MAX / 2)
count++;
end = get_time();
uint64_t exec_time = diff_time(start, end);
if (PAPI_read_counters(misses, papilevels) != PAPI_OK) {
fprintf(stderr, "Erro em PAPI_read_counters\n");
exit(1);
}
printf("Time: %" PRIu64 " Count %d\n", exec_time, count);
for (i = 0; i < papilevels; i++) {
printf("Cache misses (L%d): %lld\n", i+1, misses[i]);
}
free(array);
return 0;
}
int test_measure(char* phase)
{
if ( PAPI_read_counters( values, NUM_EVENTS ) != PAPI_OK )
return -1;
else {
float rate = (float)values[0] / (float)(values[1]);
fprintf(res_file, "%s_PAPI_L1_TCM=%lld\n", phase, values[0]);
fprintf(res_file, "%s_PAPI_L1_TCA=%lld\n", phase, values[1]);
fprintf(res_file, "%s_L1MissRate=%f%%\n", phase, (rate * 100.0));
return 0;
}
}
inline
void
papi_reset(size_t n)
{
long_long *papi_tmp;
papi_tmp = malloc(sizeof(*papi_tmp) * n);
if (PAPI_read_counters(papi_tmp, n) != PAPI_OK)
papi_eprintf("Problem reading counters %s:%d.\n", __FILE__, __LINE__);
free(papi_tmp);
}
int main ()
{
float t0, t1;
int iter, i, j;
int events[2] = {PAPI_L1_DCM, PAPI_FP_OPS }, ret;
long_long values[2];
if (PAPI_num_counters() < 2) {
fprintf(stderr, "No hardware counters here, or PAPI not supported.\n");
exit(1);
}
for (i = 0; i < MX; i++) {
if ((ad[i] = malloc(sizeof(double)*MX)) == NULL) {
fprintf(stderr,"malloc failed\n");
exit(1);
}
}
for (j = 0; j < MX; j++) {
for (i = 0; i < MX; i++) {
ad[i][j] = 1.0/3.0; /* Initialize the data */
}
}
t0 = gettime();
if ((ret = PAPI_start_counters(events, 2)) != PAPI_OK) {
fprintf(stderr, "PAPI failed to start counters: %s\n", PAPI_strerror(ret));
exit(1);
}
for (iter = 0; iter < NITER; iter++) {
for (j = 0; j < MX; j++) {
for (i = 0; i < MX; i++) {
ad[i][j] += ad[i][j] * 3.0;
}
}
}
if ((ret = PAPI_read_counters(values, 2)) != PAPI_OK) {
fprintf(stderr, "PAPI failed to read counters: %s\n", PAPI_strerror(ret));
exit(1);
}
t1 = gettime();
printf("Total software flops = %f\n",(float)TOT_FLOPS);
printf("Total hardware flops = %lld\n",(float)values[1]);
printf("MFlop/s = %f\n", (float)(TOT_FLOPS/MEGA)/(t1-t0));
printf("L1 data cache misses is %lld\n", values[0]);
}
int main()
{
//this will fail if some counters can't be accessed
if (PAPI_start_counters(papi_events, n_papi_events) != PAPI_OK)
{
printf("failed to start papi\n");
return 1;
}
doWork(123);
if (PAPI_read_counters(papi_values[0], n_papi_events) != PAPI_OK)
{
printf("failed to read countess\n");
return 1;
}
printf("counters' values: misses = %d, accesses = %d\n", papi_values[0][0], papi_values[0][1]);
return 0;
}
static void start_sssp(FibHeap<size_t, size_t> *pq,
vertex_t *graph)
{
#ifdef PAPI
if (PAPI_OK != PAPI_start_counters(g_events, G_EVENT_COUNT)) {
std::cout << ("Problem starting counters 1.\n");
}
#endif
while (!pq->empty()) {
size_t distance;
size_t node;
pq->pop(distance, node);
vertex_t *v = &graph[node];
size_t v_dist = v->distance;
for (size_t i = 0; i < v->num_edges; i++) {
const edge_t *e = &v->edges[i];
const size_t new_dist = v_dist + e->weight;
vertex_t *w = &graph[e->target];
size_t w_dist = w->distance;
if (new_dist < w_dist) {
w->distance = new_dist;
if (w->n == NULL) {
w->n = pq->push(new_dist, e->target);
} else {
pq->decrease_key(w->n, new_dist);
}
}
}
}
#ifdef PAPI
if (PAPI_OK != PAPI_read_counters(g_values[0], G_EVENT_COUNT)) {
std::cout << ("Problem reading counters 2.\n");
}
#endif
}
void my_papi_stop(int *events, int NUM_EVENTS)
{
int j;
/* Read the counters */
if (PAPI_read_counters(values, NUM_EVENTS) != PAPI_OK) {
fprintf(stderr, "PAPI_read_counters - FAILED\n");
exit(1);
}
for (j=0; j<NUM_EVENTS; j++)
{
printf("GG: %d : %lld\n", events[j], values[j]);
}
/* Stop counting events */
if (PAPI_stop_counters(values, NUM_EVENTS) != PAPI_OK) {
fprintf(stderr, "PAPI_stoped_counters - FAILED\n");
exit(1);
}
if (values != NULL) {
free(values);
}
}
void
papi_set_events(char *metric)
{
const size_t n = 1;
int max;
long_long *papi_tmp;
int papi_events[1];
int code;
max = PAPI_num_counters();
if (n > max)
papi_eprintf("Too many counters requested.\n");
papi_tmp = malloc(sizeof(*papi_tmp) * n);
PAPI_reset(max);
PAPI_stop_counters(papi_tmp, n);
if (PAPI_event_name_to_code(metric, &code) != PAPI_OK)
papi_eprintf("Unknown PAPI event %s.\n", metric);
if (code == 0)
papi_eprintf("Unknown PAPI event %s.\n", metric);
papi_events[0] = code;
PAPI_start_counters(papi_events, n);
if (PAPI_read_counters(papi_tmp, n) != PAPI_OK)
papi_eprintf("Problem reading counters %s:%d.\n", __FILE__, __LINE__);
free(papi_tmp);
}
void* test(void *data) {
int unext, last = -1;
val_t val = 0;
pval_t pval = 0;
thread_data_t *d = (thread_data_t *)data;
/* Create transaction */
TM_THREAD_ENTER(d->id);
set_cpu(the_cores[d->id]);
/* Wait on barrier */
ssalloc_init();
PF_CORRECTION;
seeds = seed_rand();
#ifdef PIN
int id = d->id;
int cpu = 40*(id/40) + 4*(id%10) + (id%40)/10;
// printf("Pinning %d to %d\n",id,cpu);
pin(pthread_self(), cpu);
// pin(pthread_self(), id);
#endif
#ifdef PAPI
if (PAPI_OK != PAPI_start_counters(g_events, G_EVENT_COUNT))
{
printf("Problem starting counters 1.");
}
#endif
barrier_cross(d->barrier);
/* Is the first op an update? */
unext = (rand_range_re(&d->seed, 100) - 1 < d->update);
#ifdef DISTRIBUTION_EXPERIMENT
while (1)
#else
while (*running)
#endif
{
if (d->es) { // event simulator experiment
if (d->lin) {
if (!empty(d->linden_set)) {
d->nb_remove++;
pval_t pval = deletemin(d->linden_set, d);
d->nb_removed++;
// printf("%d %d\n", pval, deps[pval][0]);
int i = 0;
val_t dep;
while ((dep = deps[pval][i]) != -1 && i < MAX_DEPS) {
d->nb_add++;
if (insert(d->linden_set, dep, dep)) {
d->nb_added++;
}
i++;
}
}
} else {
if (d->set->head->next[0]->next[0] != NULL) {// set not empty
d->nb_remove++;
if (d->sl) { // spray list
if (spray_delete_min(d->set, &val, d)) {
d->nb_removed++;
} else {
continue;
}
} else if (d->pq) { // lotan_shavit pq
if (lotan_shavit_delete_min(d->set, &val, d)) {
d->nb_removed++;
// continue; // TODO: maybe try remove this to simulate task handling (dependency checks still occur)
} else {
continue;
}
}
// struct timespec ten_usec;
// ten_usec.tv_sec = 0;
// ten_usec.tv_nsec = 10000;
// nanosleep(&ten_usec, NULL);
// dependency handling
int i = 0;
val_t dep;
while ((dep = deps[val][i]) != -1 && i < MAX_DEPS) {
if (!sl_contains(d->set, dep, TRANSACTIONAL)) { // dependent has been removed, need to add it again
if (sl_add(d->set, dep, TRANSACTIONAL)) { // check if insert actually succeeded (otherwise someone else did it first)
d->nb_added++;
}
d->nb_add++;
}
i++;
}
}
}
} else { // not event simulator
if (unext) { // update
if (last < 0) { // add
val = rand_range_re(&d->seed, d->range);
if (d->lin) {
pval = val;
insert(d->linden_set, pval, pval);
d->nb_added++;
last = pval;
} else { // not linden
if (sl_add(d->set, val, TRANSACTIONAL)) {
d->nb_added++;
last = val;
}
}
d->nb_add++;
} else { // remove
if (d->pq) {
if (lotan_shavit_delete_min(d->set, &val, d)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
}
last = -1;
}
else if (d->sl) {
if (spray_delete_min(d->set, &val, d)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
last = -1;
}
}
else if (d->lin) {
if ((pval = deletemin(d->linden_set, d))) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = pval;
}
last = -1;
}
}
else if (d->alternate) { // alternate mode (default)
if (sl_remove(d->set, last, TRANSACTIONAL)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
}
last = -1;
} else {
/* Random computation only in non-alternated cases */
val = rand_range_re(&d->seed, d->range);
/* Remove one random value */
if (sl_remove_succ(d->set, val, TRANSACTIONAL)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
/* Repeat until successful, to avoid size variations */
last = -1;
}
}
d->nb_remove++;
}
} else { // read
if (d->alternate) {
if (d->update == 0) {
if (last < 0) {
val = d->first;
last = val;
} else { // last >= 0
val = rand_range_re(&d->seed, d->range);
last = -1;
}
} else { // update != 0
if (last < 0) {
val = rand_range_re(&d->seed, d->range);
//last = val;
} else {
val = last;
}
}
} else val = rand_range_re(&d->seed, d->range);
PF_START(2);
if (sl_contains(d->set, val, TRANSACTIONAL))
d->nb_found++;
PF_STOP(2);
d->nb_contains++;
}
/* Is the next op an update? */
if (d->effective) { // a failed remove/add is a read-only tx
unext = ((100 * (d->nb_added + d->nb_removed))
< (d->update * (d->nb_add + d->nb_remove + d->nb_contains)));
} else { // remove/add (even failed) is considered as an update
unext = (rand_range_re(&d->seed, 100) - 1 < d->update);
}
}
#ifdef DISTRIBUTION_EXPERIMENT
if (d->first_remove != -1) {
break; //only one run
}
#endif
}
#ifdef PAPI
if (PAPI_OK != PAPI_read_counters(g_values[d->id], G_EVENT_COUNT))
{
printf("Problem reading counters 2.");
}
#endif
/* Free transaction */
TM_THREAD_EXIT();
PF_PRINT;
return NULL;
}
main(int argc, char *argv[])
{
float **a,**b,**c;
int n,n1,n2;
int i,j;
//double t0,t1;
struct timeval t0,t1;
long mtime, seconds, useconds;
// Using PAPI - from countloop.c
if (PAPI_VER_CURRENT !=
PAPI_library_init(PAPI_VER_CURRENT))
ehandler("PAPI_library_init error.");
const size_t EVENT_MAX = PAPI_num_counters();
// Suppressing output
// printf("# Max counters = %zd\n", EVENT_MAX);
if (PAPI_OK != PAPI_query_event(PAPI_TOT_INS))
ehandler("Cannot count PAPI_TOT_INS.");
if (PAPI_OK != PAPI_query_event(PAPI_FP_OPS))
ehandler("Cannot count PAPI_FP_OPS.");
if (PAPI_OK != PAPI_query_event(PAPI_L1_DCM))
ehandler("Cannot count PAPI_L1_DCM.");
size_t EVENT_COUNT = 3;
int events[] = { PAPI_TOT_INS, PAPI_FP_OPS, PAPI_L1_DCM };
long long values[EVENT_COUNT];
// Take size from args, not prompt
// printf("Enter n: "); scanf("%d",&n); printf("n = %d\n",n);
n = atoi(argv[1]);
//printf("Enter n1: "); scanf("%d",&n1); printf("n1 = %d\n",n1);
//printf("Enter n2: "); scanf("%d",&n2); printf("n2 = %d\n",n2);
// To conform to the other matrix functions
n1 = floor(sqrt(n));
n2 = n1;
n = n1*n2;
//printf("n = %d X %d = %d\n",n1,n2,n);
a = matrix(1,n,1,n);
for (i=1;i<=n;i++)
for (j=1;j<=n;j++)
a[i][j] = i+j;
b = matrix(1,n,1,n);
for (i=1;i<=n;i++)
for (j=1;j<=n;j++)
b[i][j] = i-j;
//#ifdef PRINT
//print_matrix(a,1,n,1,n);
//printf("\n"); */
//print_matrix(b,1,n,1,n);
//printf("\n"); */
//#endif
//t0 = get_seconds();
//c = matrix_prod(n,n,n,n,a,b);
//t1 = get_seconds();
//printf("Time for matrix_prod = %f sec\n",t1-t0);
//t0 = get_seconds();
gettimeofday(&t0, NULL);
// Start PAPI
PAPI_start_counters(events, EVENT_COUNT);
if (PAPI_OK != PAPI_read_counters(values, EVENT_COUNT))
ehandler("Problem reading counters.");
c = block_prod(n1,n1,n1,n2,n2,n2,a,b);
if (PAPI_OK != PAPI_read_counters(values, EVENT_COUNT))
ehandler("Problem reading counters.");
//t1 = get_seconds();
//printf("Time for block_prod = %f sec\n",t1-t0);
gettimeofday(&t1, NULL);
seconds = t1.tv_sec - t0.tv_sec;
useconds = t1.tv_usec - t0.tv_usec;
mtime = ((seconds) * 1000 + useconds/1000.0) + 0.5;
//printf("Time for matrix_prod = %f sec\n",t1-t0);
printf("%d\t%lld\t%lld\t%lld\t%ld\n", n, values[0], values[1],
values[2], mtime);
}
void* Thread(void *userData) {
ThreadInfo *info = (ThreadInfo*) userData;
Context *c = info->c;
int index = info->index;
int threadCount = c->threadCount;
int64_t repetitionCount = c->repetitionCount;
uint64_t me = 0x1 << index;
uint64_t full = 0x0000000000000000;
uint64_t copy; //thread local copy of the entry/exit barrier
for (int i = 0; i < threadCount; ++i) {
full |= 0x1 << i;
}
// set thread affinity
cpu_set_t cpuset;
CPU_ZERO(&cpuset);
CPU_SET(index, &cpuset);
assert(pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset) == 0);
//DEBUG
//pthread_getaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset);
//printf("%i uses cpus: ", index);
//for (int i = 0; i < threadCount; ++i) {
// if (CPU_ISSET(i, &cpuset)) {
// printf("%i, ", i);
// }
//}
//printf("\n");
int threadToBeRecorded = -1;
int papiEvents[3] = {0x8000003b, 0x80000000, 0x80000002};
long long papiStart[3] = {0, 0, 0};
long long papiEnd[3] = {0, 0, 0};
if (index == threadToBeRecorded) {
int ret = PAPI_start_counters(papiEvents, 3);
if (ret != 0) {
printf("thread %i: PAPI_start_counters %i\n", index, ret);
assert(0);
}
ret = PAPI_read_counters(papiStart, 3);
if (ret != 0) {
printf("thread %i: PAPI_read_counters %i\n", index, ret);
assert(0);
}
}
//unlink("a");
//FILE *log = fopen("a", "a");
for(int64_t repetition = 0; repetition < repetitionCount; repetition++){
if (c->left == 0) { /* *** if () { UNIFIED ENTRY *********************/
/* run to wall and wait busily */
do {
copy = c->entry;
//fprintf(log, "%i r %lli\n", prime, (long long) copy);
//fflush(log);
if ((copy & me) == 0) {
copy |= me;
c->entry = copy;
//fprintf(log, "%i w %lli\n", prime, (long long) copy);
//fflush(log);
}
}while (copy != full && c->left == 0);
c->left = 1;
c->exit = 0x0000000000000000;
} else if (c->left == 1) { /* *** } else if () { UNIFIED ENTRY *******/
for (int i = 0; i < threadCount - 1; ++i) {
if (c->successfulBarrierVisitsCount[i] != c->successfulBarrierVisitsCount[i+1]) {
printf("thread %i and %i are not equal at %lli %lli\n", i, i+1,
(long long)c->successfulBarrierVisitsCount[i],
(long long)c->successfulBarrierVisitsCount[i+1]);
++c->outOfSyncCount;
assert(0);
}
}
/* wait busily until everyone has left the barrier */
do {
copy = c->exit;
if ((copy & me) == 0) {
copy |= me;
c->exit = copy;
}
}while (copy != full && c->left == 1);
c->left = 0;
c->entry = 0x0000000000000000;
++(c->successfulBarrierVisitsCount[index]);
} /* *** } UNIFIED ENTRY *********************************************/
}
if (index == threadToBeRecorded) {
int ret = PAPI_stop_counters(papiEnd, 3);
if (ret != 0) {
printf("%i: PAPI_stop_counters %i\n", index, ret);
assert(0);
}
printf("thread %i: papi counter 0: %lli - %lli = %lli\n", index, papiEnd[0], papiStart[0], papiEnd[0] - papiStart[0]);
printf("thread %i: papi counter 1: %lli - %lli = %lli\n", index, papiEnd[1], papiStart[1], papiEnd[1] - papiStart[1]);
printf("thread %i: papi counter 2: %lli - %lli = %lli\n", index, papiEnd[2], papiStart[2], papiEnd[2] - papiStart[2]);
printf("\n");
}
return NULL;
}
int main() {
int level, numlevels;
/* Variaveis das medicoes reais (hwloc) */
hwloc_uint64_t real_cache[3];
unsigned real_line[3];
hwloc_topology_t topology;
hwloc_obj_t obj;
/* Variaveis das medicoes estimadas (PAPI) */
int events[3] = { PAPI_L1_DCM, PAPI_L2_DCM, PAPI_L3_DCM };
long long misses[3];
int papilevels = 3;
/* Inicializa hwloc */
if (hwloc_topology_init(&topology)) {
fprintf(stderr, "Erro em hwloc_topology_init\n");
exit(1);
}
hwloc_topology_load(topology);
/* Pega tamanhos reais de cache e line */
level = 0;
for (obj = hwloc_get_obj_by_type(topology, HWLOC_OBJ_PU, 0); obj; obj = obj->parent) {
if (obj->type == HWLOC_OBJ_CACHE) {
real_cache[level] = obj->attr->cache.size;
real_line[level] = obj->attr->cache.linesize;
level++;
}
}
numlevels = level;
/* Comeca a contar eventos */
while (PAPI_start_counters(events, papilevels) != PAPI_OK) {
papilevels--;
}
/* TODO: alguma coisa para estimar */
/* Le os contadores */
if (PAPI_read_counters(misses, papilevels) != PAPI_OK) {
fprintf(stderr, "Erro em PAPI_read_counters\n");
exit(1);
}
for (level = 0; level < papilevels; level++) {
printf("L%d cache miss: %lld\n", level+1, misses[level]);
}
printf("\n");
/* Imprime tabela */
printf("Nível Linha Real Cache Real Linha Est. Cache Est.\n");
printf("----- ---------- ---------- ---------- ----------\n");
for (level = 0; level < 3; level++) {
printf("%5d ", level+1);
if (level < numlevels) {
printf("%10u ", real_line[level]);
printf("%9lluK ", (unsigned long long) real_cache[level] / 1024);
if (level < papilevels) {
/* TODO: imprimir valores estimados no lugar dos reais */
printf("%10u ", real_line[level]);
printf("%9lluK", (unsigned long long) real_cache[level] / 1024);
} else {
printf(" N/D N/D");
}
} else {
printf(" N/D N/D N/D N/D");
}
printf("\n");
}
return 0;
}
int main(int argc, char *argv[])
{
double *collideField = NULL;
double *streamField = NULL;
char problem[100];
char pgmInput[1000];
int *flagField = NULL;
clock_t begin, end;
double time_spent;
struct timeval time_start, time_end;
int xlength[3], timesteps, timestepsPerPlotting;
double tau, bddParams[7];
long long counters[3];
int PAPI_events[] = {
PAPI_TOT_CYC,
PAPI_L2_DCM,
PAPI_L2_DCA };
PAPI_library_init(PAPI_VER_CURRENT);
#ifdef DEBUG
double * exactCollideField; // used for debugging
#endif
if(readParameters(xlength, &tau, bddParams, ×teps, ×tepsPerPlotting, problem, pgmInput, argc, argv) == 0)
{
begin = clock();
gettimeofday(&time_start, NULL);
collideField = (double*) malloc((size_t) sizeof(double) * PARAMQ * (xlength[0] + 2)*(xlength[1] + 2)*(xlength[2] + 2));
streamField = (double*) malloc((size_t) sizeof(double) * PARAMQ * (xlength[0] + 2)*(xlength[1] + 2)*(xlength[2] + 2));
flagField = (int *) malloc((size_t) sizeof (int) * (xlength[0] + 2)*(xlength[1] + 2)*(xlength[2] + 2));
initialiseFields(collideField, streamField, flagField, xlength, problem, pgmInput);
/** debugging code */
// /* output the flagField */
// char szFileName2[80];
// FILE *fp2 = NULL;
// sprintf( szFileName2, "Testdata/%s/flagField.dat", problem);
// fp2 = fopen(szFileName2,"w");
// for (int i = 0; i < (xlength[0] + 2) * (xlength[1] + 2) * (xlength[2] + 2); i++)
// fprintf(fp2, "%d\n", flagField[i]);
/** debugging code end */
printf("Progress: ");
PAPI_start_counters( PAPI_events, 3 );
for(int t = 0; t < timesteps; t++)
{
double *swap = NULL;
#ifdef _AVX_
doStreamingAndCollisionAVX(collideField, streamField, flagField, xlength, tau);
#else
doStreamingAndCollision(collideField, streamField, flagField, xlength, tau);
#endif // _AVX_
swap = collideField;
collideField = streamField;
streamField = swap;
treatBoundary(collideField, flagField, bddParams, xlength);
if (t % timestepsPerPlotting == 0)
{
// writeVtkOutput(collideField, flagField, "./Paraview/output", (unsigned int) t / timestepsPerPlotting, xlength);
/** debugging code */
// /* create reference files */
// FILE *fp = NULL;
// char szFileName[80];
// sprintf( szFileName, "Testdata/%s/%i.dat", problem, t / timestepsPerPlotting );
// fp = fopen(szFileName,"w");
// for (int i = 0; i < PARAMQ * (xlength[0] + 2) * (xlength[1] + 2) * (xlength[2] + 2); i++)
// fprintf(fp, "%0.7f\n", collideField[i]);
/* check correctness */
#ifdef DEBUG
exactCollideField = (double *) malloc ( ( size_t ) sizeof(double) * PARAMQ * (xlength[0] + 2) * (xlength[1] + 2) * (xlength[2] + 2));
FILE *fp = NULL;
unsigned int line = 0;
int noOfReadEntries;
int error = 0;
char szFileName[80];
sprintf( szFileName, "Testdata/%s/%i.dat", problem, t / timestepsPerPlotting );
fp = fopen(szFileName,"r");
if (fp != NULL)
{
for (line = 0; line < PARAMQ * (xlength[0] + 2) * (xlength[1] + 2) * (xlength[2] + 2); line++)
{
noOfReadEntries = fscanf(fp,"%lf",&exactCollideField[line]);
if (noOfReadEntries != 1)
continue;
}
}
fclose(fp);
for (int i = 0; i < PARAMQ; i++)
for (int z = 1; z <= xlength[2]; z++)
for (int y = 1; y <= xlength[1]; y++)
for(int x = 1; x <= xlength[0]; x++)
if (flagField[z * (xlength[0] + 2) * (xlength[1] + 2) + y * (xlength[0] + 2) + x] == FLUID)
if (fabs(collideField[z * (xlength[0] + 2) * (xlength[1] + 2) + y * (xlength[0] + 2 ) + x + (xlength[0] + 2) * (xlength[1] + 2) * (xlength[2] + 2) * i] - exactCollideField[PARAMQ * (z * (xlength[0] + 2) * (xlength[1] + 2) + y * (xlength[0] + 2) + x) + i]) > 1e-4)
error = 1;
if (error)
printf("ERROR: Different collideField in timestep %d\n", t);
free(exactCollideField);
#endif // DEBUG
/** end of debugging code */
}
PAPI_read_counters( counters, 3 );
int pct = ((float) t / timesteps) * 100;
printf("\b\b\b%02d%%", pct);
fflush(stdout);
}
printf("\b\b\b\b100%%\n");
end = clock();
gettimeofday(&time_end, NULL);
time_spent = (double) (end - begin) / CLOCKS_PER_SEC;
printf("Running time: %.2fs\n", time_spent);
printf("Running time (Wall clock): %.2fs\n", ( (double) (( time_end.tv_sec - time_start.tv_sec) * 1000000u + time_end.tv_usec - time_start.tv_usec) )/ 1e6);
printf("MLUPS: %.3f\n", ((double) (xlength[0] + 2) * (xlength[1] + 2) * (xlength[2] + 2) * timesteps) / (1000000.0 * ((time_end.tv_sec - time_start.tv_sec) * 1000000u + time_end.tv_usec - time_start.tv_usec) / 1e6));
printf("%lld L2 cache misses (%.3lf%% misses) in %lld cycles\n",
counters[1],
(double)counters[1] / (double)counters[2] * 100,
counters[0] );
free(collideField);
free(streamField);
free(flagField);
}
return 0;
}
int main(int argc, char *argv[]) {
if (argc < 4) {
printf("Usage: %s data_type(text or bin) input_file output_file\n", argv[0]);
return EXIT_FAILURE;
}
char *file_type = argv[1];
char *file_in = argv[2];
char *file_out = argv[3];
char *str1 = "SU.vtk";
char *str2 = "VAR.vtk";
char *str3 = "CGUP.vtk";
char *file_perf = "pstats.dat";
int status = 0;
/** internal cells start and end index*/
int nintci, nintcf;
/** external cells start and end index. The external cells are only ghost cells.
They are accessed only through internal cells*/
int nextci, nextcf;
/** link cell-to-cell array. Stores topology information*/
int **lcc;
/** red-black colouring of the cells*/
int *nboard;
/** boundary coefficients for each volume cell */
double *bs, *be, *bn, *bw, *bl, *bh, *bp, *su;
/**parameter used for volmesh and reading binary input file */
int* nodeCnt;
int*** points;
int*** elems;
/**Measured Performance and Papi parameters*/
long long *values_i = (long long *) calloc(sizeof(long long), 4);
long long *values_c = (long long *) calloc(sizeof(long long), 4);
long long *values_o = (long long *) calloc(sizeof(long long), 4);
double *mflops = (double *) calloc(sizeof(double), 3);
double *L1mira = (double *) calloc(sizeof(double), 3);
double *Lmirate = (double *) calloc(sizeof(double), 3);
double *util = (double *) calloc(sizeof(double), 3);
long long *et = (long long *) calloc(sizeof(long long), 3);
long long start_cycles, start_usec,end_cycles_1, end_usec_1, end_cycles_2, end_cycles_3, end_usec_2, end_usec_3;
/**In cluster mpp_inter L1 and L2 events can not computed at the same time,
so set into two groups*/
int Events[NUM_EVENTS]={PAPI_L2_TCM,PAPI_L2_TCA,PAPI_FP_INS,PAPI_TOT_CYC};
// int Events[NUM_EVENTS]={PAPI_L1_TCM,PAPI_L1_TCA,PAPI_FP_INS,PAPI_TOT_CYC};
/**start HW counters and execution time recorder*/
if ( PAPI_start_counters( Events, NUM_EVENTS ) != PAPI_OK )
printf("Fail to start PAPI counter\n");
start_cycles = PAPI_get_real_cyc(); // Gets the starting time in clock cycles
start_usec = PAPI_get_real_usec(); // Gets the starting time in microseconds
/* initialization */
// read-in the input file
int f_status;
if (strcmp(file_type,"text") == 0) {
f_status = read_formatted(file_in, &nintci, &nintcf, &nextci, &nextcf, &lcc,
&bs, &be, &bn, &bw, &bl, &bh, &bp, &su, &nboard);
} else if (strcmp(file_type,"bin") == 0) {
f_status = read_formatted_bin(file_in, &nintci, &nintcf, &nextci,
&nextcf, &lcc, &bs, &be, &bn, &bw,
&bl, &bh, &bp, &su,&nboard);
} else {
printf ("Input file format is nor correct\n");
return EXIT_FAILURE;
}
if (f_status != 0){
printf("failed to initialize data!\n");
return EXIT_FAILURE;
}
// allocate arrays used in gccg
int nomax = 3;
/** the reference residual*/
double resref = 0.0;
/** the ratio between the reference and the current residual*/
double ratio;
/** array storing residuals */
double* resvec = (double *) calloc(sizeof(double), (nintcf + 1));
/** the variation vector -> keeps the result in the end */
double* var = (double *) calloc(sizeof(double), (nextcf + 1));
/** the computation vectors */
double* direc1 = (double *) calloc(sizeof(double), (nextcf + 1));
double* direc2 = (double *) calloc(sizeof(double), (nextcf + 1));
/** additional vectors */
double* cgup = (double *) calloc(sizeof(double), (nextcf + 1));
double* oc = (double *) calloc(sizeof(double), (nintcf + 1));
double* cnorm = (double *) calloc(sizeof(double), (nintcf + 1));
double* adxor1 = (double *) calloc(sizeof(double), (nintcf + 1));
double* adxor2 = (double *) calloc(sizeof(double), (nintcf + 1));
double* dxor1 = (double *) calloc(sizeof(double), (nintcf + 1));
double* dxor2 = (double *) calloc(sizeof(double), (nintcf + 1));
/**store volume information*/
int nc=0;
// initialize the reference residual
for ( nc = nintci; nc <= nintcf; nc++) {
resvec[nc] = su[nc];
resref = resref + resvec[nc] * resvec[nc];
}
resref = sqrt(resref);
if (resref < 1.0e-15){
printf("i/o - error: residue sum less than 1.e-15 - %lf\n", resref);
return EXIT_FAILURE;
}
// initialize the arrays
for (nc = 0; nc <= 10; nc++){
oc[nc] = 0.0;
cnorm[nc] = 1.0;
}
for (nc = nintci; nc <= nintcf; nc++){
cgup[nc] = 0.0;
var[nc] = 0.0;
}
for (nc = nextci; nc <= nextcf; nc++){
var[nc] = 0.0;
cgup[nc] = 0.0;
direc1[nc] = 0.0;
bs[nc] = 0.0;
be[nc] = 0.0;
bn[nc] = 0.0;
bw[nc] = 0.0;
bl[nc] = 0.0;
bh[nc] = 0.0;
}
for (nc = nintci; nc <= nintcf; nc++){
cgup[nc] = 1.0 / bp[nc];
}
int if1 = 0;
int if2 = 0;
int iter = 1;
int nor = 1;
int nor1 = nor - 1;
/* finished initalization */
/*read PAPI HW counters and caculate performance of input phase*/
if ( PAPI_read_counters( values_i, NUM_EVENTS ) != PAPI_OK ){
printf("fail to stop papi counter");
}
Lmirate[0] = (double) values_i[0] / values_i[1];
end_usec_1 = PAPI_get_real_usec();
mflops[0] = (double) values_i[2] / (end_usec_1-start_usec);
util[0] = mflops[0] / PEAKPER;
/* start computation loop */
while (iter < 10000){
/* start phase 1 */
// update the old values of direc
for (nc = nintci; nc <= nintcf; nc++){
direc1[nc] = direc1[nc] + resvec[nc] * cgup[nc];
}
// compute new guess (approximation) for direc
for (nc = nintci; nc <= nintcf; nc++){
direc2[nc] = bp[nc] * direc1[nc] - bs[nc] * direc1[lcc[0][nc]]
- bw[nc] * direc1[lcc[3][nc]] - bl[nc] * direc1[lcc[4][nc]]
- bn[nc] * direc1[lcc[2][nc]] - be[nc] * direc1[lcc[1][nc]]
- bh[nc] * direc1[lcc[5][nc]];
} /* end phase 1 */
/* start phase 2 */
// execute normalization steps
double oc1, oc2, occ;
if (nor1 == 1){
oc1 = 0;
occ = 0;
for (nc = nintci; nc <= nintcf; nc++){
occ = occ + adxor1[nc] * direc2[nc];
}
oc1 = occ / cnorm[1];
for (nc = nintci; nc <= nintcf; nc++){
direc2[nc] = direc2[nc] - oc1 * adxor1[nc];
direc1[nc] = direc1[nc] - oc1 * dxor1[nc];
}
if1++;
}else if (nor1 == 2){
oc1 = 0;
occ = 0;
for (nc = nintci; nc <= nintcf; nc++){
occ = occ + adxor1[nc] * direc2[nc];
}
oc1 = occ / cnorm[1];
oc2 = 0;
occ = 0;
for (nc = nintci; nc <= nintcf; nc++){
occ = occ + adxor2[nc] * direc2[nc];
}
oc2 = occ / cnorm[2];
for (nc = nintci; nc <= nintcf; nc++){
direc2[nc] = direc2[nc] - oc1 * adxor1[nc] - oc2 * adxor2[nc];
direc1[nc] = direc1[nc] - oc1 * dxor1[nc] - oc2 * dxor2[nc];
}
if2++;
}
cnorm[nor] = 0;
double omega = 0;
// compute the new residual
for (nc = nintci; nc <= nintcf; nc++){
cnorm[nor] = cnorm[nor] + direc2[nc] * direc2[nc];
omega = omega + resvec[nc] * direc2[nc];
}
omega = omega / cnorm[nor];
double resnew = 0.0;
for (nc = nintci; nc <= nintcf; nc++){
var[nc] = var[nc] + omega * direc1[nc];
resvec[nc] = resvec[nc] - omega * direc2[nc];
resnew = resnew + resvec[nc] * resvec[nc];
}
resnew = sqrt(resnew);
ratio = resnew / resref;
// exit on no improvements of residual
if (ratio <= 1.0e-10){
break;
}
iter++;
// prepare additional arrays for the next iteration step
if (nor == nomax){
nor = 1;
}else{
if (nor == 1){
for (nc = nintci; nc <= nintcf; nc++){
dxor1[nc] = direc1[nc];
adxor1[nc] = direc2[nc];
}
} else if (nor == 2){
for (nc = nintci; nc <= nintcf; nc++){
dxor2[nc] = direc1[nc];
adxor2[nc] = direc2[nc];
}
}
nor++;
}
nor1 = nor - 1;
}/* end phase 2 */
/* finished computation loop */
/*read PAPI HW counters and caculate performance of computation phase*/
end_cycles_2 = PAPI_get_real_cyc(); // Gets the ending time in clock cycles
end_usec_2 = PAPI_get_real_usec(); // Gets the ending time in microseconds
if ( PAPI_read_counters( values_c, NUM_EVENTS ) != PAPI_OK ){
printf("fail to read papi counter");
}
Lmirate[1] = (double) values_c[0]/values_c[1];
mflops[1] = (double) values_c[2] / ( end_usec_2-end_usec_1 );
util[1] = mflops[1] / PEAKPER;
/* write output file */
if ( write_result(file_in, file_out, nintci, nintcf, var, iter, ratio) != 0 )
printf("error when trying to write to file %s\n", file_out);
//transfer volume to mesh
if (vol2mesh(nintci, nintcf, lcc, &nodeCnt, &points, &elems) != 0 ){
printf("error when trying to converge topology to volume");
}
//write output to vtk file
if (write_result_vtk(str1, nintci, nintcf, nodeCnt, points, elems, su) != 0){
printf("error when write SU to vtk file");
}
if (write_result_vtk(str2, nintci, nintcf, nodeCnt, points, elems, var) != 0){
printf("error when write VAR to vtk file");
}
if (write_result_vtk(str3, nintci, nintcf, nodeCnt, points, elems, cgup) != 0){
printf("error when write CGUP to vtk file");
}
/*read PAPI HW counters and caculate performance of output phase*/
if ( PAPI_stop_counters( values_o, NUM_EVENTS ) != PAPI_OK ){
printf("fail to stop papi counter");
}
Lmirate[2] = (double) values_o[0]/values_o[1];
end_cycles_3 = PAPI_get_real_cyc(); // Gets the ending time in clock cycles
end_usec_3 = PAPI_get_real_usec(); // Gets the ending time in microseconds
mflops[2] = (double) (values_o[2])/(end_usec_3-end_usec_2);
util[2] = mflops[2] / PEAKPER;
/** Write all measured performance to pstats.dat*/
et[0] = end_usec_1-start_usec;
et[1] = end_usec_2-end_usec_1;
et[2] = end_usec_3-end_usec_2;
if (write_result_dat(file_perf, values_i,values_c, values_o,Lmirate, et, mflops, util) != 0 ){
printf("error when write measured performance to data file");
}
/* Free all the dynamically allocated memory */
free(direc2); free(direc1); free(dxor2); free(dxor1); free(adxor2); free(adxor1);
free(cnorm); free(oc); free(var); free(cgup); free(resvec); free(su); free(bp);
free(bh); free(bl); free(bw); free(bn); free(be); free(bs);
printf("Simulation completed successfully!\n");
return EXIT_SUCCESS;
}
void* sssp(void *data) {
thread_data_t *d = (thread_data_t *)data;
/* Create transaction */
set_cpu(the_cores[d->id]);
/* Wait on barrier */
ssalloc_init();
PF_CORRECTION;
seeds = seed_rand();
#ifdef PIN
int id = d->id;
// int cpu = 40*(id/40) + 4*(id%10) + (id%40)/10;
int cpu = 4*(id%20) + id/20;
// printf("Pinning %d to %d\n",id,cpu);
pin(pthread_self(), cpu);
// pin(pthread_self(), id);
#endif
#ifdef PAPI
if (PAPI_OK != PAPI_start_counters(g_events, G_EVENT_COUNT))
{
printf("Problem starting counters 1.");
}
#endif
barrier_cross(d->barrier);
// Begin SSSP
int fail = 0;
// int radius = 0;
while (1) {
val_t node;
slkey_t dist_node;
// print_skiplist(d->set);
while (1) {
if (d->sl) {
if (spray_delete_min_key(d->set, &dist_node, &node, d)) break; // keep trying until get a node
} else if (d->pq) {
if (lotan_shavit_delete_min_key(d->set, &dist_node, &node, d)) break;
} else if (d->lin) {
node = (val_t) deletemin_key(d->linden_set, &dist_node, d); break;
} else {
printf("error: no queue selected\n");
exit(1); // TODO: grace
}
if (dist_node == -1) { // flag that list is empty
break;
}
dist_node = 0;
}
if (dist_node == -1) { // list is empty; TODO make sure threads don't quit early
fail++;
if (fail > 20*d->nb_threads) { // TODO: really need a better break condition...
break;
}
continue;
}
fail = 0;
if (dist_node != nodes[node].dist) continue; // dead node
nodes[node].times_processed++;
int i;
for (i = 0;i < nodes[node].deg;i++) {
int v = nodes[node].adj[i];
int w = nodes[node].weights[i];
slkey_t dist_v = nodes[v].dist;
// printf("v=%d dist_v=%d\n", v, dist_v);
if (dist_v == -1 || dist_node + w < dist_v) { // found better path to v
// printf("attempting cas...\n");
// printf("nodes[v].dist=%d dist_v=%d dist_node=%d\n", nodes[v].dist, dist_v, dist_node);
int res = ATOMIC_CAS_MB(&nodes[v].dist, dist_v, dist_node+w);
// printf("%d nodes[%d].dist=%d\n", res, v, nodes[v].dist);
if (res) {
if (d->pq || d->sl) {
sl_add_val(d->set, dist_node+w, v, TRANSACTIONAL); // add to queue only if CAS is successful
} else if (d->lin) {
insert(d->linden_set, dist_node+w, v);
}
d->nb_add++;
// if (dist_node+1 > radius) {
// radius = dist_node+1;
// printf("radius %d\n", radius);
// }
}
}
}
}
// End SSSP
#ifdef PAPI
if (PAPI_OK != PAPI_read_counters(g_values[d->id], G_EVENT_COUNT))
{
printf("Problem reading counters 2.");
}
#endif
PF_PRINT;
return NULL;
}
main(int argc, char *argv[])
{
float **a,**b,**c;
int n;
int NB;
int i,j;
int x;
//double t0,t1;
struct timeval t0,t1;
long mtime, seconds, useconds;
// Using PAPI - from countloop.c
if (PAPI_VER_CURRENT !=
PAPI_library_init(PAPI_VER_CURRENT))
ehandler("PAPI_library_init error.");
const size_t EVENT_MAX = PAPI_num_counters();
// Suppressing output
// printf("# Max counters = %zd\n", EVENT_MAX);
if (PAPI_OK != PAPI_query_event(PAPI_TOT_INS))
ehandler("Cannot count PAPI_TOT_INS.");
if (PAPI_OK != PAPI_query_event(PAPI_FP_OPS))
ehandler("Cannot count PAPI_FP_OPS.");
if (PAPI_OK != PAPI_query_event(PAPI_L1_DCM))
ehandler("Cannot count PAPI_L1_DCM.");
size_t EVENT_COUNT = 3;
int events[] = { PAPI_TOT_INS, PAPI_FP_OPS, PAPI_L1_DCM };
long long values[EVENT_COUNT];
// Take size from args, not prompt
// printf("Enter n: "); scanf("%d",&n); printf("n = %d\n",n);
n = atoi(argv[1]);
NB = atoi(argv[2]);
a = matrix(1,n,1,n);
for (i=1; i<=n; i++)
for (j=1; j<=n; j++)
a[i][j] = i+j;
b = matrix(1,n,1,n);
for (i=1; i<=n; i++)
for (j=1; j<=n; j++)
b[i][j] = i-j;
//t0 = get_seconds();
gettimeofday(&t0, NULL);
// Start PAPI
PAPI_start_counters(events, EVENT_COUNT);
if (PAPI_OK != PAPI_read_counters(values, EVENT_COUNT))
ehandler("Problem reading counters.");
//for (x=0;x<1000;x++){
c = matrix_prod(n,n,n,n,a,b,NB);
//}
if (PAPI_OK != PAPI_read_counters(values, EVENT_COUNT))
ehandler("Problem reading counters.");
//t1 = get_seconds();
gettimeofday(&t1, NULL);
seconds = t1.tv_sec - t0.tv_sec;
useconds = t1.tv_usec - t0.tv_usec;
mtime = ((seconds) * 1000 + useconds/1000.0) + 0.5;
//printf("Time for matrix_prod = %f sec\n",t1-t0);
printf("%d\t%lld\t%lld\t%lld\t%ld\n", n, values[0], values[1],
values[2], mtime);
}
