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 main() {
// clear communication areas
// cannot use memset() for _SPM pointers!
for(int i = 0; i < sizeof(struct msg_t); i++) {
((volatile _SPM char *)spm_in)[i] = 0;
((volatile _SPM char *)spm_out)[i] = 0;
}
int slave_param = 1;
for(int i = 0; i < get_cpucnt(); i++) {
if (i != NOC_MASTER) {
corethread_t ct = (corethread_t) i;
if(corethread_create(&ct,&slave,(void*)slave_param) != 0){
}
}
}
master();
int* ret;
for (int i = 0; i < get_cpucnt(); ++i) {
if (i != NOC_MASTER) {
corethread_join((corethread_t)i,(void**)&ret);
}
}
}
int main() {
printf("Hello CMP\n");
int core_id = 1; // The core number
static int parameter = 1000;
corethread_create(core_id, &work, (void *) ¶meter);
blink(2000);
// the folowing is not executed in this example
int* res;
corethread_join( core_id, (void *) &res );
return 0;
}
int main() {
int cpucnt = get_cpucnt();
printf("cpus:%d\n",cpucnt);
int dum;
for(int i = 1; i < cpucnt; i++) {
corethread_create(i,&writer_init,(void *)i);
}
flag = 1;
int sum = reader();
int retries = 0;
for(int i = 1; i < cpucnt; i++) {
int _res;
corethread_join(i, (void **)&_res);
retries += _res;
}
printf("Sum: %d Retries: %d\n",sum, retries);
return sum;
}
int main() {
corethread_t slave1 = 1;
corethread_t slave2 = 2;
corethread_t slave3 = 3;
corethread_t slave4 = 4;
corethread_t slave5 = 5;
corethread_t slave6 = 6;
corethread_t slave7 = 7;
corethread_t slave8 = 8;
if (!mp_chan_init(&m2s,
get_cpuid(),
slave4,
BUFFER_SIZE,
2)) {
abort();
}
if (!mp_chan_init(&s2m,
slave4,
get_cpuid(),
BUFFER_SIZE,
2)) {
abort();
}
if (!mp_communicator_init(&comm,
2,
cores,
0)) {
abort();
}
if (!mp_communicator_init(&comm_world,
sizeof(cores_world)/sizeof(cores_world[0]),
cores_world,
BUFFER_SIZE)) {
abort();
}
int* ret;
/* set up the run */
//corethread_create(&slave,&loop,(void*)roundtrip_slave);
/* run appropriate test */
// // TEST_LATENCY
// puts("Latency (usecs)");
// corethread_create(&slave4,&loop,(void*)latency_slave);
// loop(latency_master);
// corethread_join(slave4,(void**)&ret);
//
// // TEST_ROUNDTRIP
// puts("Roundtrip (Transactions/sec)");
// corethread_create(&slave4,&loop,(void*)roundtrip_slave);
// loop(roundtrip_master);
// corethread_join(slave4,(void**)&ret);
//
// // TEST_BANDWIDTH
// puts("Bandwidth (KB/sec)");
// corethread_create(&slave4,&loop,(void*)bandwidth_slave);
// loop(bandwidth_master);
// corethread_join(slave4,(void**)&ret);
// // TEST_BIBANDWIDTH
// puts("Bibandwidth");
// corethread_create(&slave4,&loop,(void*)bibandwidth_slave);
// loop(bibandwidth_master);
// corethread_join(slave4,(void**)&ret);
//
// // TEST_REDUCE
// puts("Reduce");
// corethread_create(&slave4,&loop,(void*)reduce_slave);
// loop(reduce_master);
// corethread_join(slave4,(void**)&ret);
//
// // TEST_ALLREDUCE
// puts("Allreduce");
// corethread_create(&slave4,&loop,(void*)allreduce_slave);
// loop(allreduce_master);
// corethread_join(slave4,(void**)&ret);
//
// // TEST_ALLTOALL
// puts("Alltoall");
// corethread_create(&slave4,&loop,(void*)alltoall_slave);
// loop(alltoall_master);
// corethread_join(slave4,(void**)&ret);
/////////////////////////////////////////////////////////////////////////////
// TEST_BARRIER
/////////////////////////////////////////////////////////////////////////////
puts("Barrier (usecs)");
for(int i = 0; i < sizeof(cores_world)/sizeof(cores_world[0]); i++) {
if (i != NOC_MASTER) {
if(corethread_create((corethread_t*)&cores_world[i],&loop,(void*)barrier_slave) != 0){
printf("Corethread %d not created\n",i);
}
}
}
loop(barrier_master);
//puts("Master finished");
for (int i = 0; i < sizeof(cores_world)/sizeof(cores_world[0]); ++i) {
if (i != NOC_MASTER) {
corethread_join((corethread_t)cores_world[i],(void**)&ret);
//printf("Slave %d joined\n",i);
}
}
// /////////////////////////////////////////////////////////////////////////////
// // TEST_BROADCAST
// /////////////////////////////////////////////////////////////////////////////
// puts("Broadcast (KB/sec)");
// for(int i = 0; i < sizeof(cores_world)/sizeof(cores_world[0]); i++) {
// if (i != NOC_MASTER) {
// if(corethread_create((corethread_t*)&cores_world[i],&loop,(void*)broadcast_slave) != 0){
// printf("Corethread %d not created\n",i);
// }
// }
// }
// loop(broadcast_master);
// //puts("Master finished");
// for (int i = 0; i < sizeof(cores_world)/sizeof(cores_world[0]); ++i) {
// if (i != NOC_MASTER) {
// corethread_join((corethread_t)cores_world[i],(void**)&ret);
// //printf("Slave %d joined\n",i);
// }
// }
exit(0);
}
int main() {
puts("Master");
corethread_t worker_1 = 1; // For now the core ID
int worker_1_param = 1;
corethread_create(&worker_1,&func_worker_1,(void*)&worker_1_param);
// Create the queuing ports
spd_t * chan = mp_create_sport(MP_CHAN_1_ID, SINK, MP_CHAN_1_MSG_SIZE);
volatile unsigned long long _SPM * time_sample = mp_alloc(MP_CHAN_1_MSG_SIZE);
if (chan == NULL || time_sample == NULL) {
DEBUGF(chan);
abort();
}
// Initialize the communication channels
int retval = mp_init_ports();
// TODO: check on retval
puts("Initialized ports");
while(slave != 1) {
;
}
puts("Slave is ready");
unsigned long long min_time_diff = -1;
unsigned long long max_time_diff = 0;
unsigned long long accum_time_diff = 0;
unsigned long long cnt_time_diff = 0;
unsigned long long percent = 0;
int done = 0;
unsigned long long start = get_cpu_usecs();
while(!done) {
int success = mp_read(chan,time_sample);
unsigned long long time_diff = get_cpu_usecs() - (*time_sample);
if (success == 0) {
printf("No sample received\n");
} else if ((*time_sample) == 0) {
printf("Received empty sample, newest: %u, sample size: %u\n",chan->newest,chan->sample_size);
} else {
if (time_diff > 2000 ) {
// Time difference is larger than a micro second
printf("Time sample: %llu\tdiff: %llu\n",*time_sample,time_diff);
}
cnt_time_diff++;
if (time_diff < min_time_diff) {
min_time_diff = time_diff;
}
if (time_diff > max_time_diff) {
max_time_diff = time_diff;
}
accum_time_diff += time_diff;
}
if (start + percent < get_cpu_usecs()) {
percent += RUNTIME/10;
printf("+");
fflush(stdout);
}
if ( start + RUNTIME < get_cpu_usecs()) {
done = 1;
}
}
printf("\n");
printf("Status:\n\tMin time diff: %llu\n\tMax time diff: %llu\n\tAvg time diff: %llu\n", min_time_diff,max_time_diff,accum_time_diff/cnt_time_diff);
int* res;
corethread_join(worker_1,&res);
return *res;
}
int main() {
puts("Master");
corethread_t worker_1 = 2; // For now the core ID
int worker_1_param = 1;
char send_data[] = "Hello World!, Sending messages is cool!";
char recv_data[40];
// Initialization of message passing buffers
// mp_chan_init() return false if local and remote
// addresses are not aligned to words
if (!mp_chan_init(&chan1,
get_cpuid(),
worker_1,
MP_CHAN_1_BUF_SIZE,
MP_CHAN_1_NUM_BUF)) {
abort();
}
if (!mp_chan_init(&chan2,
worker_1,
get_cpuid(),
MP_CHAN_2_BUF_SIZE,
MP_CHAN_2_NUM_BUF)) {
abort();
}
puts("Initialized buffers");
if (!mp_communicator_init(&comm,
2,
cores,
0)) {
abort();
}
puts("Initialized barrier");
corethread_create(&worker_1,&func_worker_1,(void*)&worker_1_param);
int i = 0;
// While there is still data to be sent
while(i < sizeof(send_data)) {
int chunk = 0;
if ( sizeof(send_data)-i >= chan1.buf_size) {
// If the remaining data is more than the size of the buffer
chunk = chan1.buf_size;
} else {
// The remaining data all fits in a buffer
chunk = sizeof(send_data)-i;
}
// Copy the chunk of data to the write buffer
for (int j = 0; j < chunk; ++j) {
*((volatile char _SPM *)chan1.write_buf+j) = send_data[i+j];
}
// Send the chunk of data
mp_send(&chan1);
i += chunk;
}
puts("Messages sent");
mp_barrier(&comm);
puts("Barrier reached");
mp_recv(&chan2);
puts("Message recv");
// Copy the received data to the recv_data array
for(int i = 0; i < sizeof(recv_data)-1; i++) {
recv_data[i] = (*((volatile char _SPM *)chan2.read_buf+i)-worker_1_param);
}
// Acknowledge the received data
mp_ack(&chan2);
recv_data[39] = '\0';
puts(recv_data);
int* res;
corethread_join(worker_1,&res);
return *res;
}
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;
}
void bench_noc() {
// Pointer to the deadline device
volatile _IODEV int *dead_ptr = (volatile _IODEV int *) PATMOS_IO_DEADLINE;
// Measure execution time with the clock cycle timer
volatile _IODEV int *timer_ptr = (volatile _IODEV int *) (PATMOS_IO_TIMER+4);
printf("Hello NoC\n");
printf("We use %d bytes buffers\n", BUF_SIZE);
int core_id = 1; // The core number
corethread_create(core_id, &work, NULL);
int start, val;
int data = 42;
// create a channel
qpd_t *channel = mp_create_qport(1, SOURCE, BUF_SIZE, NUM_BUF);
// init
mp_init_ports();
start = *timer_ptr;
// write data into the send buffer
*(volatile int _SPM *) channel->write_buf = data;
start = *timer_ptr;
// send the buffer
mp_send(channel, 0);
printf("Data sent\n");
printf("Returned data is: %d\n", field);
printf("Took %d cycles\n", end_time - start - 1);
int min = 999999;
int max = 0;
printf("NoC in a loop:\n");
for (int i=0; i<CNT; ++i) {
start = *timer_ptr;
*(volatile int _SPM *) channel->write_buf = i;
start = *timer_ptr;
mp_send(channel, 0);
*dead_ptr = 10000; // some delay to see the result
val = *dead_ptr;
val = end_time - start - 1;
// printf("%d ", val);
if (min>val) min = val;
if (max<val) max = val;
}
printf("\n");
printf("Min: %d max: %d\n", min, max);
min = 999999;
max = 0;
do_delay_times();
printf("NoC in a loop with random delay:\n");
for (int i=0; i<CNT; ++i) {
start = *timer_ptr;
*(volatile int _SPM *) channel->write_buf = i;
*dead_ptr = data_spm[i];
val = *dead_ptr; // delay by a random value
start = *timer_ptr;
mp_send(channel, 0);
*dead_ptr = 3000; // some delay to see the result
val = *dead_ptr;
val = end_time - start;
// printf("%d ", val);
if (min>val) min = val;
if (max<val) max = val;
}
printf("\n");
printf("Min: %d max: %d\n", min, max);
// not really as the worker runs forever
int* res;
corethread_join( core_id, (void *) &res );
}
// The main acts as producer
int main() {
int val = 0;
end_flag = 0;
for (int i=0; i<CONSUMERS; ++i) {
result[i] = 0;
started_consumer[i] = 0;
finished_consumer[i] = 0;
}
started_producer = 0;
started_fork = 0;
finished_producer = 0;
finished_fork = 0;
printf("Producer/fork/n-consumers benchmark for the S4NOC paper:\n");
printf(" Delay: %d\n", DELAY);
printf(" Number of cores: %d\n", get_cpucnt());
printf(" Total packets sent: %d\n", LEN);
printf(" Buffer size: %d\n", BUF_LEN);
printf("Runnning test:\n");
for (int k=0; k<CONSUMERS; ++k) {
corethread_create(CONSUMER_CORE[k], &consumer, (void*) &CONSUMER_ID[k] );
*dead_ptr = 8000;
val = *dead_ptr;
while(started_consumer[k] == 0) {;}
printf(" Consumer-%d is ready.\n", k+1);
}
corethread_create(FORK_CORE, &fork, NULL);
while(started_fork == 0) {;}
printf(" Fork is ready.\n");
corethread_create(PRODUCER_CORE, &producer, NULL);
while(started_producer == 0) {;}
printf(" Producer has started.\n [...]\n");
while(finished_producer == 0) {;}
printf(" Producer has finished.\n");
while(finished_fork == 0) {;}
printf(" Fork has finished.\n");
for (int i=0; i<CONSUMERS; ++i) {
while(finished_consumer[i] == 0) {;}
printf(" Consumer-%d has finished.\n", i+1);
}
/*
for (int i=0; i<CONSUMERS; ++i) {
printf(" %d %d \n", started_consumer[i], finished_consumer[i] );
}
*/
*dead_ptr = 8000000;
val = *dead_ptr;
for (int i=0; i<CONSUMERS; ++i) {
printf("Results Consumer-%d: \n", i+1);
printf(" %d valid pakets out of of %d received.\n", result[i], LEN);
printf(" Reception time of %d cycles -> %g cycles per received packet.\n", time[i], 1. * time[i]/LEN);
}
// Join threads
int *retval;
end_flag = 1;
corethread_join(PRODUCER_CORE, (void **)&retval);
corethread_join(FORK_CORE, (void **)&retval);
for (int i=0; i<CONSUMERS; ++i) {
corethread_join(CONSUMER_CORE[i], (void **)&retval);
}
printf("End of program.\n");
return val;
}