int main(int argc, char *argv[]) {
double results[NUM_TRIALS];
int i, papi_setnum;
// initialize papi
int desired_events[] = {PAPI_TOT_CYC, PAPI_FP_INS, PAPI_L2_DCA, PAPI_L2_DCM, PAPI_L3_DCA, PAPI_L3_DCM, PAPI_TLB_DM, PAPI_LD_INS, PAPI_SR_INS};
int num_desired = 9;
PAPI_event_set_wrapper_t* event_sets;
int num_sets;
papi_init(desired_events, num_desired, &event_sets, &num_sets);
// input parameters
int log2_stanzaLength = atoi(argv[1]);
int log2_numIterations = atoi(argv[2]);
// compute actual values from base 2 logs
stanzaLength = 1;
for (i=0; i<log2_stanzaLength; i++) {
stanzaLength *= 2;
}
numIterations = 1;
for (i=0; i<log2_numIterations; i++) {
numIterations *= 2;
}
int arrayLength = stanzaLength;
printf("\nstanzaLength = %d\n", stanzaLength);
printf("arrayLength = %d\n", arrayLength);
printf("numIterations = %d\n", numIterations);
printf("num_sets = %d\n\n", num_sets);
// allocate working arrays
A = (double *) malloc(arrayLength * sizeof(double));
B = (double *) malloc(arrayLength * sizeof(double));
if (A==NULL) {
printf("Error on array A malloc.\n");
exit(EXIT_FAILURE);
}
if (B==NULL) {
printf("Error on array B malloc.\n");
exit(EXIT_FAILURE);
}
// initialize arrays
init_flush_cache_array();
initArrays();
for (papi_setnum=0; papi_setnum < num_sets; papi_setnum++) {
PAPI_MAKE_MEASUREMENTS(event_sets[papi_setnum].set, cacheBenchmark(), NUM_TRIALS, results);
print_measurements(&(event_sets[papi_setnum]), results, NUM_TRIALS);
}
papi_cleanup(event_sets, num_sets);
return 0;
}
int main(int argc, char** argv)
{
/* declearation */
DT *a, b;
long arr_bytes, arr_size, p;
int i, samples;
double t_start, t_end, deltaT;
/* initialization */
b = 0.0;
arr_bytes = 1024 * 1024 * 1024; // 1GB
arr_size = arr_bytes/sizeof(DT);
samples = 3;
/* memory allocation */
#ifndef MIC
a = (DT *)malloc(arr_bytes);
#else
a = (DT *)_mm_malloc(arr_bytes, 64);
#endif
if(a==NULL)
{
DB(RT_LVL, "array 'a' allocation failed");
}
fill(a, arr_size, 5.0);
#pragma omp parallel
{
init_flush_cache_array();
}
/* measurement */
for(i=0; i<samples; i++)
{
#pragma omp parallel
{
flush_cache();
}
t_start = timer();
#pragma omp parallel for private(p) shared(a, arr_size)
for(p=0; p<arr_size; p++)
{
//b += a[p];
a[p] = b;
}
b = b + 1.0;
t_end = timer();
if(i==(samples-1))
{
deltaT = t_end - t_start;
SAVE_DATA("%lf\t", arr_bytes/deltaT)
printf("bw: %lf\t", arr_bytes/deltaT);
}
}
b = a[0] + a[arr_size-1];
save_results(&b, 1); // save results to avoid the aggressive optimizations
SAVE_DATA("\n")
/* post-process */
#ifndef MIC
if(a!=NULL) free(a);
#else
if(a!=NULL) _mm_free(a);
#endif
return 0;
}
int main(int argc, char *argv[]) {
pthread_t *threads;
pthread_attr_t attr;
uint32_t **ranks;
void *status;
#if defined(PAPI_ENABLED) && !defined(DEBUG)
int num_sets;
PAPI_event_set_wrapper_t* event_sets;
#endif
int rc;
uint32_t t;
printf("Optimized Stream benchmark (using SSE intrinsics)\n");
init_flush_cache_array();
malloc_arrays(argv);
print_array_parameters();
select_code_variant(argv);
print_code_variant_parameters();
threads = (pthread_t *) malloc(numThreads * sizeof(pthread_t));
ranks = (uint32_t **) malloc(numThreads * sizeof(uint32_t *));
#if !defined(DEBUG)
#if defined(PAPI_ENABLED)
papi_init(desired_events, num_desired, &event_sets, &num_sets);
// initialize threaded PAPI
if (PAPI_thread_init((unsigned long (*)(void)) (pthread_self)) != PAPI_OK) {
printf("Error with PAPI_thread_init().\n");
exit(EXIT_FAILURE);
}
results = (double *) malloc(num_sets * numThreads * NUM_TRIALS * sizeof(double));
if (results==NULL) {
printf("Error on array results malloc.\n");
exit(EXIT_FAILURE);
}
#else
results = (double *) malloc(numThreads * NUM_TRIALS * sizeof(double));
if (results==NULL) {
printf("Error on array results malloc.\n");
exit(EXIT_FAILURE);
}
#if defined(CYCLE_TIME)
// calculate clock rate
GET_CLOCK_RATE(results, NUM_TRIALS);
median_counts_per_sec = find_median(results, NUM_TRIALS);
//printf("Median ticks per second = %e\n", median_counts_per_sec);
#else
timer_init();
median_counts_per_sec = 1.0;
#endif
#endif
#endif
pthread_attr_init(&attr);
pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
barrier_init(&my_barrier, numThreads);
#if defined(AFFINITY_ENABLED)
Affinity_Init();
#endif
// run stream tests
for (t=0; t < numThreads; t++) {
ranks[t] = (uint32_t *) malloc(sizeof(uint32_t));
*ranks[t] = t;
}
for (t=1; t < numThreads; t++) {
#if defined(DEBUG)
printf("Creating thread %u\n", t);
#endif
rc = pthread_create(&threads[t], &attr, pthreads_each, (void *) ranks[t]);
if (rc) {
printf("ERROR; return code from pthread_create() is %d\n", rc);
exit(EXIT_FAILURE);
}
}
pthreads_each((void *) ranks[0]);
// join the other threads
for (t=1; t < numThreads; t++) {
pthread_join(threads[t], &status);
}
#if defined(PAPI_ENABLED) && !defined(DEBUG)
papi_cleanup(event_sets, num_sets);
#endif
pthread_attr_destroy(&attr);
pthread_exit(NULL);
barrier_destroy(&my_barrier);
free_arrays();
return EXIT_SUCCESS;
}
int main(int argc, char *argv[]) {
double results[NUM_TRIALS];
#if !defined(DEBUG)
#if defined(PAPI_ENABLED)
int papi_setnum, num_desired, num_sets;
#else
double median_counts_per_sec;
#endif
#endif
int i;
printf("7-point stencil, no add, naive C code with non-periodic boundary conditions\n");
#if !defined(DEBUG)
#if defined(PAPI_ENABLED)
// initialize papi
int desired_events[] = {PAPI_TOT_CYC, PAPI_FP_INS, PAPI_L2_DCA, PAPI_L2_DCM, PAPI_L3_DCM, PAPI_TLB_DM, PAPI_LD_INS, PAPI_SR_INS};
num_desired = 9;
PAPI_event_set_wrapper_t* event_sets;
papi_init(desired_events, num_desired, &event_sets, &num_sets);
#else
// calculate clock rate
GET_CLOCK_RATE(results, NUM_TRIALS);
median_counts_per_sec = find_median(results, NUM_TRIALS);
#endif
#endif
// initialize arrays
init_flush_cache_array();
malloc_grids(argv);
printf("\n");
#if defined(DEBUG)
init_grids();
printf("SINGLY NESTED LOOP:\n");
printf("\nGRID A BEFORE:");
print_grid(A);
printf("\nGRID B BEFORE:");
print_grid(B);
naive_singly_nested_loop();
printf("\nGRID A AFTER:");
print_grid(A);
printf("\nGRID B AFTER:");
print_grid(B);
init_grids();
printf("TRIPLY NESTED LOOPS:\n");
printf("\nGRID A BEFORE:");
print_grid(A);
printf("\nGRID B BEFORE:");
print_grid(B);
naive_triply_nested_loops();
printf("\nGRID A AFTER:");
print_grid(A);
printf("\nGRID B AFTER:");
print_grid(B);
#else
#if defined(PAPI_ENABLED)
printf("SINGLY NESTED LOOP:\n");
for (papi_setnum=0; papi_setnum < num_sets; papi_setnum++) {
PAPI_MAKE_MEASUREMENTS(event_sets[papi_setnum].set, naive_singly_nested_loop(), NUM_TRIALS, results);
print_papi_measurements(&(event_sets[papi_setnum]), results, NUM_TRIALS);
}
printf("\n");
printf("TRIPLY NESTED LOOPS:\n");
for (papi_setnum=0; papi_setnum < num_sets; papi_setnum++) {
PAPI_MAKE_MEASUREMENTS(event_sets[papi_setnum].set, naive_triply_nested_loops(), NUM_TRIALS, results);
print_papi_measurements(&(event_sets[papi_setnum]), results, NUM_TRIALS);
}
printf("\n");
papi_cleanup(event_sets, num_sets);
#else
printf("SINGLY NESTED LOOP:\n");
TIMER_MAKE_MEASUREMENTS(naive_singly_nested_loop(), results, NUM_TRIALS);
print_timer_measurements(results, NUM_TRIALS, median_counts_per_sec);
printf("\n");
printf("TRIPLY NESTED LOOPS:\n");
TIMER_MAKE_MEASUREMENTS(naive_triply_nested_loops(), results, NUM_TRIALS);
print_timer_measurements(results, NUM_TRIALS, median_counts_per_sec);
printf("\n");
printf("\n");
#endif
#endif
printf("\nFinal interior values: A[%lu, %lu, %lu] = %4.2e, B[%lu, %lu, %lu] = %4.2e\n", nx/2, ny/2, nz/2, A[Index3D(nx/2, ny/2, nz/2)], nx/2, ny/2, nz/2, B[Index3D(nx/2, ny/2, nz/2)]);
fc_checksum();
free(A);
free(B);
return EXIT_SUCCESS;
}
