void loop(void* arg) {
int (*test)(int, int) = (int (*)(int, int))arg;
int i, j;
int res;
//communicator_t* loc_com = &comm;
communicator_t* loc_com = &comm_world;
//if (test == barrier_master || test == barrier_slave ||
// test == broadcast_master || test == broadcast_slave) {
// loc_com = &comm_world;
//}
//communicator_t stack_com;
//stack_com.barrier_set = loc_com->barrier_set;
//for(int i = 0; i < loc_com->count; i++) {
// stack_com.addr[i] = loc_com->addr[i];
//}
//stack_com.count = loc_com->count;
//stack_com.msg_size = loc_com->msg_size;
DEBUGGER("Initial test run\n");
/* Allow routing/cache setup ahead of time */
test(1,1024);
DEBUGGER("Initial test run done\n");
for( i=0; i < num_sizes; i++){
if (flush & FLUSH_BETWEEN_SIZES)
{
inval_dcache();
inval_mcache();
}
for( j=1; j <= repeat_count; j++){
if (flush & FLUSH_BETWEEN_REPEATS) {
inval_dcache();
inval_mcache();
}
mp_barrier(loc_com);
//mp_barrier(&stack_com);
res = test(iterations, sizes[i]);
my_two_printf("%u\t%i\n",sizes[i],res);
mp_barrier(loc_com);
//mp_barrier(&stack_com);
}
}
inval_dcache();
inval_mcache();
if(get_cpuid() != 0){
int ret = 0;
corethread_exit(&ret);
}
return;
}
int __patmos_lock_acquire(_LOCK_T *lock) {
const unsigned cnt = get_cpucnt();
if (cnt > 1) {
const unsigned char id = get_cpuid();
_UNCACHED _LOCK_T *ll = (_UNCACHED _LOCK_T *)lock;
ll->entering[id] = 1;
unsigned n = 1 + max(ll);
ll->number[id] = n;
ll->entering[id] = 0;
for (unsigned j = 0; j < cnt; j++) {
while (ll->entering[j]) {
/* busy wait */
}
unsigned m = ll->number[j];
while ((m != 0) &&
((m < n) || ((m == n) && (j < id)))) {
/* busy wait, only update m */
m = ll->number[j];
}
}
// invalidate data cache to establish cache coherence
inval_dcache();
}
return 0;
}
int main() {
for (int i = 0; i < 9; ++i) {
done[i] = 0;
}
print_cpuinfo();
inval_dcache();
core_count = print_processor_info();
//done = malloc(sizeof(unsigned int _SPM *)*core_count);
com_spm_size = com_spm_test();
main_mem_size = main_mem_test();
int param = 0;
printf("Creating corethreads...");
for(int i = 0; i < get_cpucnt(); i++) {
if (i != NOC_MASTER) {
corethread_t ct = (corethread_t) i;
if(corethread_create(&ct,&slave_tester,(void*)param) != 0){
printf("Corethread %d not created\n",i);
}
}
}
puts("OK");
printf("Performing main mem load test...");
fflush(stdout);
mem_load_test();
printf("OK\n");
noc_test_master();
int* ret;
for (int i = 0; i < get_cpucnt(); ++i) {
if (i != NOC_MASTER) {
corethread_join((corethread_t)i,(void**)&ret);
}
}
printf("Joined with other cores\n");
return 0;
}
int latency_slave(int cnt, int bytes) {
int i;
for (i=0; i<cnt; i++) {
if (flush & FLUSH_BETWEEN_ITERATIONS) {
inval_dcache();
inval_mcache();
}
mp_recv_size(&m2s,bytes);
}
mp_send(&s2m);
return 0;
}
int calibrate_cache_flush(int cnt) {
int tmp = 1;
int i;
TIMER_START;
for (i=0;i<cnt;i++) {
if (flush & FLUSH_BETWEEN_ITERATIONS){
inval_dcache();
inval_mcache();
}
}
TIMER_STOP;
tmp = TIMER_ELAPSED;
return tmp;
}
int barrier_slave(int cnt, int bytes)
{
int i;
for (i=0; i<cnt; i++) {
if (flush & FLUSH_BETWEEN_ITERATIONS) {
inval_dcache();
//inval_mcache();
}
mp_barrier(&comm_world);
}
return 0;
}
int broadcast_slave(int cnt, int bytes)
{
int i;
for (i=0; i<cnt; i++) {
if (flush & FLUSH_BETWEEN_ITERATIONS) {
inval_dcache();
inval_mcache();
}
mp_broadcast(&comm_world, 1);
}
return 0;
}
int roundtrip_slave(int cnt, int bytes) {
int i;
for (i=0; i<cnt; i++) {
if (flush & FLUSH_BETWEEN_ITERATIONS)
{
inval_dcache();
inval_mcache();
}
mp_recv_size(&m2s, bytes);
mp_send_size(&s2m, bytes);
}
return 0;
}
int barrier_master(int cnt, int bytes)
{
int i;
int total = 0;
TIMER_START;
for (i=0; i<cnt; i++) {
if (flush & FLUSH_BETWEEN_ITERATIONS) {
inval_dcache();
//inval_mcache();
}
mp_barrier(&comm_world);
}
TIMER_STOP;
total = TIMER_ELAPSED;
total -= calibrate_cache_flush(cnt);
return(total/cnt); /* usec */
}
int broadcast_master(int cnt, int bytes)
{
int i;
unsigned long long total = 0;
TIMER_START;
for (i=0; i<cnt; i++) {
if (flush & FLUSH_BETWEEN_ITERATIONS) {
inval_dcache();
inval_mcache();
}
mp_broadcast(&comm_world, 1);
}
TIMER_STOP;
total = TIMER_ELAPSED;
total -= calibrate_cache_flush(cnt);
return(((unsigned long long)cnt*bytes*1000000)/(total*1024.0)); /* KB/sec */
}
int roundtrip_master(int cnt, int bytes) {
int i;
unsigned long long total = 0;
TIMER_START;
for (i=0; i<cnt; i++) {
if (flush & FLUSH_BETWEEN_ITERATIONS)
{
inval_dcache();
inval_mcache();
}
mp_send_size(&m2s, bytes);
mp_recv_size(&s2m, bytes);
}
TIMER_STOP;
total = TIMER_ELAPSED;
total -= calibrate_cache_flush(cnt);
return(((unsigned long long)cnt*1000000) / total); /* Transactions/sec */
}
int latency_master(int cnt, int bytes) {
int i;
unsigned long long total = 0;
TIMER_START;
for (i=0; i<cnt; i++) {
if (flush & FLUSH_BETWEEN_ITERATIONS) {
inval_dcache();
inval_mcache();
}
mp_send_size(&m2s,bytes);
}
TIMER_STOP;
mp_recv(&s2m);
mp_ack(&s2m);
total = TIMER_ELAPSED;
total -= calibrate_cache_flush(cnt);
return(total/cnt);
}
int bandwidth_master(int cnt, int bytes)
{
int i;
unsigned long long total = 0;
TIMER_START;
for (i=0; i<cnt; i++)
{
if (flush & FLUSH_BETWEEN_ITERATIONS)
{
inval_dcache();
inval_mcache();
}
mp_send_size(&m2s, bytes);
}
mp_recv(&s2m);
mp_ack(&s2m);
TIMER_STOP;
total = TIMER_ELAPSED;
total -= calibrate_cache_flush(cnt);
return(((unsigned long long)cnt*bytes*1000000)/(total*1024)); /* KB/sec */
}
int main() {
corethread_t worker_1 = SLAVE_CORE; // For now the core ID
corethread_create(&worker_1,&func_worker_1,(void*)&worker_1);
puts("Corethread created");
unsigned short int local_phase = 0;
min_time = ULONG_MAX;
max_time = 0;
accum_time = 0;
cnt_time = 0;
unsigned long long int start = 0;
unsigned long long int stop = 0;
spd_t * sport1 = mp_create_sport(CHAN_ID_ONE,SINK,SAMPLE_SIZE*sizeof(short));
spd_t * sport2 = mp_create_sport(CHAN_ID_TWO,SOURCE,SAMPLE_SIZE*sizeof(short));
if (sport1 == NULL || sport2 == NULL) {
//exit(1);
}
volatile short _SPM * sample = mp_alloc(SAMPLE_SIZE*sizeof(short));
mp_init_ports();
done = 1;
int balance = 0;
for (int i = 0; i < SAMPLE_SIZE; ++i) {
sample[i] = i;
}
for (int i = 0; i < ITERATIONS/2; ++i) {
mp_write(sport2,sample);
for (int i = 0; i < SAMPLE_SIZE; ++i) {
sample[i] = i;
}
}
for (int i = 0; i < ITERATIONS/2; ++i) {
mp_write(sport2,sample);
for (int i = 0; i < SAMPLE_SIZE; ++i) {
sample[SAMPLE_SIZE-1-i] = i;
}
}
for (int i = 0; i < ITERATIONS; ++i) {
start = get_cpu_usecs();
int ret = mp_read(sport1,sample);
stop = get_cpu_usecs();
if (ret == 0)
{
puts("No value written yet.");
} else {
unsigned long long int exe_time = stop - start;
min_time = (exe_time < min_time) ? exe_time : min_time;
max_time = (exe_time > max_time) ? exe_time : max_time;
accum_time += exe_time;
cnt_time++;
if (sample[0] == 0) {
balance++;
for (int i = 0; i < SAMPLE_SIZE; ++i) {
if(sample[i] != i) {
printf("Error: sample[%i] = %i\n",i,sample[i]);
break;
}
}
} else if (sample[0] == SAMPLE_SIZE-1) {
balance--;
for (int i = 0; i < SAMPLE_SIZE; ++i) {
if(sample[SAMPLE_SIZE-1-i] != i) {
printf("Error: sample[%i] = %i\n",i,sample[i]);
break;
}
}
} else {
printf("Wrong sample values sample[0] = %i\n",sample[0]);
}
}
}
printf("Local phase: %d\n",local_phase);
inval_dcache();
int* res;
corethread_join(worker_1,&res);
printf("Balance: %i\n",balance);
printf("Min time: %llu\tMax time: %llu\tAccumulated time: %llu\nCount time: %llu\tAverage time: %llu\n", min_time,max_time,accum_time,cnt_time,accum_time/cnt_time);
puts("Corethread joined");
return *res;
}
