/*
* Runs benchmark.
*/
int main(int argc, char **argv)
{
int i; /* Loop index. */
double *mask; /* Mask. */
uint64_t end; /* End time. */
uint64_t start; /* Start time. */
unsigned char *img; /* Image. */
readargs(argc, argv);
timer_init();
srandnum(seed);
/* Benchmark initialization. */
if (verbose)
printf("initializing...\n");
start = timer_get();
img = smalloc(p->imgsize*p->imgsize*sizeof(char));
for (i = 0; i < p->imgsize*p->imgsize; i++)
img[i] = randnum() & 0xff;
mask = smalloc(p->masksize*p->masksize*sizeof(double));
generate_mask(mask);
end = timer_get();
if (verbose)
printf(" time spent: %f\n", timer_diff(start, end)*MICROSEC);
/* Apply filter. */
if (verbose)
printf("applying filter...\n");
start = timer_get();
gauss_filter(img, p->imgsize, mask, p->masksize);
end = timer_get();
total = timer_diff(start, end);
/* Print timing statistics. */
printf("timing statistics:\n");
printf(" master: %f\n", master*MICROSEC);
for (i = 0; i < nclusters; i++)
printf(" slave %d: %f\n", i, slave[i]*MICROSEC);
printf(" communication: %f\n", communication*MICROSEC);
printf(" total time: %f\n", total*MICROSEC);
/* House keeping. */
free(mask);
free(img);
return (0);
}
int main(int argc, char **argv)
{
timer_init();
((void) argc);
total = 0;
rank = atoi(argv[0]);
open_noc_connectors();
getwork();
start = timer_get();
friendly_numbers();
end = timer_get();
syncNumbers();
total = timer_diff(start, end);
data_send(outfd, &total, sizeof(uint64_t));
/* Close channels. */
mppa_close(infd);
mppa_close(outfd);
mppa_exit(0);
return (0);
}
/*
* Compute the target congestion window for the next RTT according to
* cubic equation when an ack is received.
*
* W(t) = C(t-K)^3 + W(last_max)
*/
static uint32_t
tcp_cubic_update(struct tcpcb *tp, u_int32_t rtt)
{
float K, var;
u_int32_t elapsed_time, win;
win = min(tp->snd_cwnd, tp->snd_wnd);
if (tp->t_ccstate->cub_last_max == 0)
tp->t_ccstate->cub_last_max = tp->snd_ssthresh;
if (tp->t_ccstate->cub_epoch_start == 0) {
/*
* This is the beginning of a new epoch, initialize some of
* the variables that we need to use for computing the
* congestion window later.
*/
tp->t_ccstate->cub_epoch_start = tcp_now;
if (tp->t_ccstate->cub_epoch_start == 0)
tp->t_ccstate->cub_epoch_start = 1;
if (win < tp->t_ccstate->cub_last_max) {
VERIFY(current_task() == kernel_task);
/*
* Compute cubic epoch period, this is the time
* period that the window will take to increase to
* last_max again after backoff due to loss.
*/
K = (tp->t_ccstate->cub_last_max - win)
/ tp->t_maxseg / tcp_cubic_coeff;
K = cbrtf(K);
tp->t_ccstate->cub_epoch_period = K * TCP_RETRANSHZ;
/* Origin point */
tp->t_ccstate->cub_origin_point =
tp->t_ccstate->cub_last_max;
} else {
tp->t_ccstate->cub_epoch_period = 0;
tp->t_ccstate->cub_origin_point = win;
}
tp->t_ccstate->cub_target_win = 0;
}
VERIFY(tp->t_ccstate->cub_origin_point > 0);
/*
* Compute the target window for the next RTT using smoothed RTT
* as an estimate for next RTT.
*/
elapsed_time = timer_diff(tcp_now, 0,
tp->t_ccstate->cub_epoch_start, 0);
if (tcp_cubic_use_minrtt)
elapsed_time += max(tcp_cubic_use_minrtt, rtt);
else
elapsed_time += rtt;
var = (elapsed_time - tp->t_ccstate->cub_epoch_period) / TCP_RETRANSHZ;
var = var * var * var * (tcp_cubic_coeff * tp->t_maxseg);
tp->t_ccstate->cub_target_win = tp->t_ccstate->cub_origin_point + var;
return (tp->t_ccstate->cub_target_win);
}
/*
* Runs benchmark.
*/
int main(int argc, char **argv)
{
int i; /* Loop index. */
int *a; /* Array to be sorted. */
uint64_t end; /* End time. */
uint64_t start; /* Start time. */
readargs(argc, argv);
timer_init();
srandnum(seed);
/* Benchmark initialization. */
if (verbose)
printf("initializing...\n");
start = timer_get();
a = smalloc(p->n*sizeof(int));
for (i = 0; i < p->n; i++)
a[i] = randnum() & 0xfffff;
end = timer_get();
if (verbose)
printf(" time spent: %f\n", timer_diff(start, end)*MICROSEC);
/* Cluster data. */
if (verbose)
printf("sorting...\n");
start = timer_get();
bucketsort(a, p->n);
end = timer_get();
total = timer_diff(start, end);
/* Print timing statistics. */
printf("timing statistics:\n");
printf(" master: %f\n", master*MICROSEC);
for (i = 0; i < nclusters; i++)
printf(" slave %d: %f\n", i, slave[i]*MICROSEC);
printf(" communication: %f\n", communication*MICROSEC);
printf(" total time: %f\n", total*MICROSEC);
/* House keeping. */
free(a);
return (0);
}
int timer_get_shortest(egg_timeval_t *howlong)
{
egg_timer_t *timer = timer_repeat_head;
/* No timers? Boo. */
if (!timer) return(1);
timer_diff(&now, &timer->trigger_time, howlong);
return(0);
}
/*
* Asserts if another iteration is needed.
*/
static int again(void)
{
int i;
start = timer_get();
/* Checks if another iteration is needed. */
for (i = 0; i < nprocs*NUM_THREADS; i++)
{
if (has_changed[i] && too_far[i])
{
end = timer_get();
total += timer_diff(start, end);
return (1);
}
}
end = timer_get();
total += timer_diff(start, end);
return (0);
}
/*
* Runs benchmark.
*/
int main(int argc, char **argv)
{
int i; /* Loop index. */
int *map; /* Map of clusters. */
uint64_t end; /* End time. */
uint64_t start; /* Start time. */
vector_t *data; /* Data points. */
readargs(argc, argv);
printf("CAPBench - KM kernel\n");
printf(" # of threads: %d \n", nthreads);
timer_init();
srandnum(seed);
omp_set_num_threads(nthreads);
/* Benchmark initialization. */
start = timer_get();
data = smalloc(p->npoints*sizeof(vector_t));
for (i = 0; i < p->npoints; i++)
{
data[i] = vector_create(p->dimension);
vector_random(data[i]);
}
end = timer_get();
/* Cluster data. */
printf("Entering in KM ROI - Clustering data...\n");
start = timer_get();
m5_reset_stats(0,0);
map = kmeans(data, p->npoints, p->ncentroids, p->mindistance);
end = timer_get();
m5_dump_stats(0,0);
printf("KM total time: %f\n", timer_diff(start, end)*MICROSEC);
/* House keeping. */
free(map);
for (i = 0; i < p->npoints; i++)
vector_destroy(data[i]);
free(data);
return (0);
}
static void test_basic(void)
{
char buf[64];
struct timespec ts = { .tv_sec = 0, .tv_nsec = 500000000 };
timer *t = timer_new();
info("Starting");
timer_start(t);
nanosleep(&ts, NULL);
timer_stop(t);
info("Diff: %s", timer_diff(t, buf, 64));
}
int main(int argc, char **argv)
{
run_test("basic", &test_basic);
return 0;
}
/*
* Populates clusters.
*/
static void populate(void)
{
int i, j; /* Loop indexes. */
float tmp; /* Auxiliary variable. */
float distance; /* Smallest distance. */
start = timer_get();
memset(&too_far[rank*NUM_THREADS], 0, NUM_THREADS*sizeof(int));
/* Iterate over data points. */
#pragma omp parallel for schedule(static) default(shared) private(i, j, tmp, distance)
for (i = 0; i < lnpoints; i++)
{
distance = vector_distance(CENTROID(map[i]), POINT(i));
/* Look for closest cluster. */
for (j = 0; j < ncentroids; j++)
{
/* Point is in this cluster. */
if (j == map[i])
continue;
tmp = vector_distance(CENTROID(j), POINT(i));
/* Found. */
if (tmp < distance)
{
map[i] = j;
distance = tmp;
}
}
/* Cluster is too far away. */
if (distance > mindistance)
too_far[rank*NUM_THREADS + omp_get_thread_num()] = 1;
}
end = timer_get();
total += timer_diff(start, end);
}
/*
* Computes clusters' centroids.
*/
static void compute_centroids(void)
{
int i, j; /* Loop indexes. */
int population; /* Centroid population. */
start = timer_get();
memcpy(lcentroids, CENTROID(rank*(ncentroids/nprocs)), lncentroids[rank]*dimension*sizeof(float));
memset(&has_changed[rank*NUM_THREADS], 0, NUM_THREADS*sizeof(int));
memset(centroids, 0, (ncentroids + DELTA*nprocs)*dimension*sizeof(float));
memset(ppopulation, 0, (ncentroids + nprocs*DELTA)*sizeof(int));
/* Compute partial centroids. */
#pragma omp parallel for schedule(static) default(shared) private(i, j)
for (i = 0; i < lnpoints; i++)
{
j = map[i]%NUM_THREADS;
omp_set_lock(&lock[j]);
vector_add(CENTROID(map[i]), POINT(i));
ppopulation[map[i]]++;
omp_unset_lock(&lock[j]);
}
end = timer_get();
total += timer_diff(start, end);
sync_pcentroids();
sync_ppopulation();
start = timer_get();
/* Compute centroids. */
#pragma omp parallel for schedule(static) default(shared) private(i, j, population)
for (j = 0; j < lncentroids[rank]; j++)
{
population = 0;
for (i = 0; i < nprocs; i++)
{
if (*POPULATION(i, j) == 0)
continue;
population += *POPULATION(i, j);
if (i == rank)
continue;
vector_add(PCENTROID(rank, j), PCENTROID(i, j));
}
if (population > 1)
vector_mult(PCENTROID(rank, j), 1.0/population);
/* Cluster mean has changed. */
if (!vector_equal(PCENTROID(rank, j), LCENTROID(j)))
{
has_changed[rank*NUM_THREADS + omp_get_thread_num()] = 1;
vector_assign(LCENTROID(j), PCENTROID(rank, j));
}
}
end = timer_get();
total += timer_diff(start, end);
sync_centroids();
sync_status();
}
/*
* When a new ack with SACK is received, check if it indicates packet
* reordering. If there is packet reordering, the socket is marked and
* the late time offset by which the packet was reordered with
* respect to its closest neighboring packets is computed.
*/
static void
tcp_sack_detect_reordering(struct tcpcb *tp, struct sackhole *s,
tcp_seq sacked_seq, tcp_seq snd_fack)
{
int32_t rext = 0, reordered = 0;
/*
* If the SACK hole is past snd_fack, this is from new SACK
* information, so we can ignore it.
*/
if (SEQ_GT(s->end, snd_fack))
return;
/*
* If there has been a retransmit timeout, then the timestamp on
* the SACK segment will be newer. This might lead to a
* false-positive. Avoid re-ordering detection in this case.
*/
if (tp->t_rxtshift > 0)
return;
/*
* Detect reordering from SACK information by checking
* if recently sacked data was never retransmitted from this hole.
*/
if (SEQ_LT(s->rxmit, sacked_seq)) {
reordered = 1;
tcpstat.tcps_avoid_rxmt++;
}
if (reordered) {
if (tcp_detect_reordering == 1 &&
!(tp->t_flagsext & TF_PKTS_REORDERED)) {
tp->t_flagsext |= TF_PKTS_REORDERED;
tcpstat.tcps_detect_reordering++;
}
tcpstat.tcps_reordered_pkts++;
tp->t_reordered_pkts++;
/*
* If reordering is seen on a connection wth ECN enabled,
* increment the heuristic
*/
if (TCP_ECN_ENABLED(tp)) {
INP_INC_IFNET_STAT(tp->t_inpcb, ecn_fallback_reorder);
tcpstat.tcps_ecn_fallback_reorder++;
tcp_heuristic_ecn_aggressive(tp);
}
VERIFY(SEQ_GEQ(snd_fack, s->rxmit));
if (s->rxmit_start > 0) {
rext = timer_diff(tcp_now, 0, s->rxmit_start, 0);
if (rext < 0)
return;
/*
* We take the maximum reorder window to schedule
* DELAYFR timer as that will take care of jitter
* on the network path.
*
* Computing average and standard deviation seems
* to cause unnecessary retransmissions when there
* is high jitter.
*
* We set a maximum of SRTT/2 and a minimum of
* 10 ms on the reorder window.
*/
tp->t_reorderwin = max(tp->t_reorderwin, rext);
tp->t_reorderwin = min(tp->t_reorderwin,
(tp->t_srtt >> (TCP_RTT_SHIFT - 1)));
tp->t_reorderwin = max(tp->t_reorderwin, 10);
}
}
/*
* RT kernel.
*/
int main(int argc, char **argv)
{
/* Number of spheres. */
#define NR_SPHERES 6
int i; /* Loop index. */
image_t img; /* Image. */
sphere_t spheres[NR_SPHERES]; /* Spheres. */
uint64_t end; /* End time. */
uint64_t start; /* Start time. */
#ifdef _XEON_PHI_
double power;
#endif
readargs(argc, argv);
timer_init();
omp_set_num_threads(nthreads);
/* Benchmark initialization. */
if (verbose)
printf("initializing...\n");
start = timer_get();
/* Ground sphere. */
spheres[0] = sphere_create(
VECTOR(0, -10004, -20), /* Center */
10000, /* Radius */
VECTOR(0.2, 0.2, 0.2), /* Surface Color */
0, /* Reflection */
0, /* Transparency */
VECTOR(0, 0, 0)); /* Emission Color */
/* Red sphere. */
spheres[1] = sphere_create(
VECTOR(0, 0, -20), /* Center */
4, /* Radius */
VECTOR(1.00, 0.32, 0.36), /* Surface Color */
1, /* Reflection */
0.5, /* Transparency */
VECTOR(0, 0, 0)); /* Emission Color */
/* Yellow sphere. */
spheres[2] = sphere_create(
VECTOR(5, -1, -15),
2,
VECTOR(0.90, 0.76, 0.46),
1,
0.0,
VECTOR(0, 0, 0));
/* Blue sphere. */
spheres[3] = sphere_create(
VECTOR(5, 0, -25),
3,
VECTOR(0.65, 0.77, 0.97),
1,
0.0,
VECTOR(0, 0, 0));
/* Gray sphere. */
spheres[4] = sphere_create(
VECTOR(-5.5, 0, -15),
3,
VECTOR(0.90, 0.90, 0.90),
1,
0.0,
VECTOR(0, 0, 0));
/* Light source. */
spheres[5] = sphere_create(
VECTOR(0, 30, -30),
3,
VECTOR(0, 0, 0),
0,
0,
VECTOR(3, 3, 3));
end = timer_get();
if (verbose)
printf(" time spent: %f\n", timer_diff(start, end)*MICROSEC);
#ifdef _XEON_PHI_
power_init();
#endif
/* Ray tracing. */
if (verbose)
printf("rendering scene...\n");
start = timer_get();
img = render(spheres, NR_SPHERES, p->height, p->width, p->depth);
end = timer_get();
#ifdef _XEON_PHI_
power = power_end();
#endif
if (verbose)
image_export("out.ppm", img, IMAGE_PPM);
printf("timing statistics:\n");
printf(" total time: %f\n", timer_diff(start, end)*MICROSEC);
#ifdef _XEON_PHI_
printf(" average power: %f\n", power*0.000001);
#endif
/* Hous keeping. */
for (i = 0; i < NR_SPHERES; i++)
sphere_destroy(spheres[i]);
image_destroy(img);
return (EXIT_SUCCESS);
}
int main(int argc, char **argv)
{
Timer *render_timer;
Sdl *sdl;
FILE *out;
Colour *buffer;
Pixel *pixels;
int num_pixels;
SDL_Event event = {0};
if (argc < 2)
return 1;
sdl = sdl_load(argv[1]);
if (sdl == NULL)
return 1;
if (!init_SDL())
return 1;
num_pixels = config->width * config->height;
buffer = calloc(num_pixels, sizeof(Colour));
pixels = calloc(num_pixels, sizeof(Pixel));
for (int j = 0; j < config->height; j++)
for (int i = 0; i < config->width; i++)
{
pixels[j*config->width + i].x = i;
pixels[j*config->width + i].y = j;
}
srand(time(NULL));
shuffle_pixels(pixels, config->width, config->height);
srand(0x20071208);
/* START */
render_timer = timer_start("Rendering");
for (int i = 0; i < num_pixels; i++)
{
Camera *cam = scene->camera;
Colour c;
Ray r;
int x = pixels[i].x, y = pixels[i].y;
/* The last parameter is the near plane, which is irrelevant for
* the moment. */
r = camera_ray(cam, x, y, 1);
c = ray_colour(r, 0);
buffer[config->width*y + x] = c;
put_pixel(display_surface, x, y, c);
if (i % config->width == 0)
{
SDL_Flip(display_surface);
while (SDL_PollEvent(&event))
if (event.type == SDL_QUIT)
return 0;
}
}
/* STOP */
timer_stop(render_timer);
timer_diff_print(render_timer);
printf("%.2f kilopixels per second\n",
num_pixels/1000./(timer_diff(render_timer)));
out = fopen("ray.ppm", "w");
ppm_write(buffer, config->width, config->height, out);
free(buffer);
fclose(out);
SDL_Flip(display_surface);
while(1)
{
while (SDL_WaitEvent(&event))
if (event.type == SDL_QUIT)
return 0;
else if (event.type == SDL_VIDEOEXPOSE)
SDL_Flip(display_surface);
}
return 0;
}
/*
* Bucket-sort algorithm.
*/
extern void bucketsort(int *array, int n)
{
int max; /* Maximum number. */
int i, j; /* Loop indexes. */
int range; /* Bucket range. */
struct minibucket *minib; /* Working mini-bucket. */
struct message *msg; /* Working message. */
struct bucket **todo; /* Todo buckets. */
struct bucket **done; /* Done buckets. */
uint64_t start, end; /* Timers. */
/* Setup slaves. */
open_noc_connectors();
spawn_slaves();
sync_slaves();
todo = smalloc(NUM_BUCKETS*sizeof(struct bucket *));
done = smalloc(NUM_BUCKETS*sizeof(struct bucket *));
for (i = 0; i < NUM_BUCKETS; i++)
{
done[i] = bucket_create();
todo[i] = bucket_create();
}
/* Find max number in the array. */
start = timer_get();
max = INT_MIN;
for (i = 0; i < n; i++)
{
/* Found. */
if (array[i] > max)
max = array[i];
}
/* Distribute numbers. */
range = max/NUM_BUCKETS;
for (i = 0; i < n; i++)
{
j = array[i]/range;
if (j >= NUM_BUCKETS)
j = NUM_BUCKETS - 1;
bucket_insert(&todo[j], array[i]);
}
end = timer_get();
master += timer_diff(start, end);
/* Sort buckets. */
j = 0;
for (i = 0; i < NUM_BUCKETS; i++)
{
while (bucket_size(todo[i]) > 0)
{
minib = bucket_pop(todo[i]);
/* Send message. */
msg = message_create(SORTWORK, i, minib->size);
message_send(outfd[j], msg);
message_destroy(msg);
/* Send data. */
communication +=
data_send(outfd[j], minib->elements, minib->size*sizeof(int));
minibucket_destroy(minib);
j++;
/*
* Slave processes are busy.
* So let's wait for results.
*/
if (j == nclusters)
{
/* Receive results. */
for (/* NOOP */ ; j > 0; j--)
{
/* Receive message. */
msg = message_receive(infd[nclusters - j]);
/* Receive mini-bucket. */
minib = minibucket_create();
minib->size = msg->u.sortresult.size;
communication += data_receive(infd[nclusters -j], minib->elements,
minib->size*sizeof(int));
bucket_push(done[msg->u.sortresult.id], minib);
message_destroy(msg);
}
}
}
}
/* Receive results. */
for (/* NOOP */ ; j > 0; j--)
{
/* Receive message. */
msg = message_receive(infd[j - 1]);
/* Receive bucket. */
minib = minibucket_create();
minib->size = msg->u.sortresult.size;
communication +=
data_receive(infd[j - 1], minib->elements, minib->size*sizeof(int));
bucket_push(done[msg->u.sortresult.id], minib);
message_destroy(msg);
}
start = timer_get();
rebuild_array(done, array);
end = timer_get();
master += timer_diff(start, end);
/* House keeping. */
for (i = 0; i < NUM_BUCKETS; i++)
{
bucket_destroy(todo[i]);
bucket_destroy(done[i]);
}
free(done);
free(todo);
join_slaves();
close_noc_connectors();
}
/*
* Runs benchmark.
*/
int main(int argc, char **argv)
{
int i; /* Loop index. */
int *mask; /* Mask. */
uint64_t end; /* End time. */
uint64_t start; /* Start time. */
char *img; /* Image. */
int numcorners=0; /* Total corners detected */
#ifdef _XEON_PHI_
double power;
#endif
readargs(argc, argv);
timer_init();
srandnum(seed);
omp_set_num_threads(nthreads);
/* Benchmark initialization. */
if (verbose)
printf("initializing...\n");
start = timer_get();
img = smalloc(p->imgsize*p->imgsize*sizeof(char));
for (i = 0; i < p->imgsize*p->imgsize; i++){
char val = randnum() & 0xff;
img[i] = (val>0) ? val : val*(-1);
}
mask = smalloc(p->maskrows*p->maskcolumns*sizeof(int));
generate_mask(mask);
end = timer_get();
if (verbose)
printf(" time spent: %f\n", timer_diff(start, end)*MICROSEC);
#ifdef _XEON_PHI_
power_init();
#endif
/* Detect corners. */
if (verbose)
printf("detecting corners...\n");
start = timer_get();
numcorners = fast(img, p->imgsize, mask);
end = timer_get();
#ifdef _XEON_PHI_
power = power_end();
#endif
printf("timing statistics:\n");
printf(" total time: %f\n", timer_diff(start, end)*MICROSEC);
#ifdef _XEON_PHI_
printf(" average power: %f\n", power*0.000001);
#endif
printf(" corners detected: %d\n", numcorners);
/* House keeping. */
free(mask);
free(img);
return (0);
}
//-----------------------------------------------------------------
// usb_control_send: Perform a transfer via IN
//-----------------------------------------------------------------
int usb_control_send(unsigned char *buf, int size)
{
t_time tS;
int send;
int remain;
int count = 0;
int err = 0;
log_printf(USBLOG_SETUP_IN, "USB: usb_control_send %d\n", size);
// Loop until partial packet sent
do
{
remain = size - count;
send = MIN(remain, EP0_MAX_PACKET_SIZE);
log_printf(USBLOG_SETUP_IN_DBG, " Remain %d, Send %d\n", remain, send);
usbhw_load_tx_buffer(ENDPOINT_CONTROL, buf, (unsigned char) send);
buf += send;
count += send;
log_printf(USBLOG_SETUP_IN_DBG, " Sent %d, Remain %d\n", send, (size - count));
tS = timer_now();
while ( !usbhw_has_tx_space( ENDPOINT_CONTROL ) )
{
if (timer_diff(timer_now(), tS) > USB_CTRL_TX_TIMEOUT)
{
log_printf(USBLOG_ERR, "USB: Timeout sending IN data\n");
err = 1;
break;
}
}
}
while (send >= EP0_MAX_PACKET_SIZE);
if (!err)
{
log_printf(USBLOG_SETUP_IN, "USB: Sent total %d\n", count);
// Wait for ACK from host
tS = timer_now();
do
{
if (timer_diff(timer_now(), tS) > USB_CTRL_TX_TIMEOUT)
{
log_printf(USBLOG_ERR, "USB: ACK not received\n");
err = 1;
break;
}
}
while (!usbhw_is_rx_ready(ENDPOINT_CONTROL));
usbhw_clear_rx_ready(ENDPOINT_CONTROL);
if (!err)
{
log_printf(USBLOG_SETUP_IN, "USB: ACK received\n");
}
}
return !err;
}
