static double avg_gradient(pa_smoother *s, pa_usec_t x) {
unsigned i, j, c = 0;
int64_t ax = 0, ay = 0, k, t;
double r;
drop_old(s, x);
/* First, calculate average of all measurements */
i = s->history_idx;
for (j = s->n_history; j > 0; j--) {
ax += s->history_x[i];
ay += s->history_y[i];
c++;
i++;
while (i >= HISTORY_MAX)
i -= HISTORY_MAX;
}
/* Too few measurements, assume gradient of 1 */
if (c < 2)
return 1;
ax /= c;
ay /= c;
/* Now, do linear regression */
k = t = 0;
i = s->history_idx;
for (j = s->n_history; j > 0; j--) {
int64_t dx, dy;
dx = (int64_t) s->history_x[i] - ax;
dy = (int64_t) s->history_y[i] - ay;
k += dx*dy;
t += dx*dx;
i++;
while (i >= HISTORY_MAX)
i -= HISTORY_MAX;
}
r = (double) k / t;
return s->monotonic && r < 0 ? 0 : r;
}
static void add_to_history(pa_smoother *s, uint64_t x, uint64_t y) {
unsigned j, i;
pa_assert(s);
/* First try to update an existing history entry */
i = s->history_idx;
for (j = s->n_history; j > 0; j--) {
if (s->history_x[i] == x) {
s->history_y[i] = y;
return;
}
REDUCE_INC(i);
}
/* Drop old entries */
drop_old(s, x);
/* Calculate position for new entry */
j = s->history_idx + s->n_history;
REDUCE(j);
/* Fill in entry */
s->history_x[j] = x;
s->history_y[j] = y;
/* Adjust counter */
s->n_history ++;
/* And make sure we don't store more entries than fit in */
if (s->n_history > HISTORY_MAX) {
s->history_idx += s->n_history - HISTORY_MAX;
REDUCE(s->history_idx);
s->n_history = HISTORY_MAX;
}
}
static void add_to_history(pa_smoother *s, pa_usec_t x, pa_usec_t y) {
unsigned j;
pa_assert(s);
drop_old(s, x);
/* Calculate position for new entry */
j = s->history_idx + s->n_history;
while (j >= HISTORY_MAX)
j -= HISTORY_MAX;
/* Fill in entry */
s->history_x[j] = x;
s->history_y[j] = y;
/* Adjust counter */
s->n_history ++;
/* And make sure we don't store more entries than fit in */
if (s->n_history >= HISTORY_MAX) {
s->history_idx += s->n_history - HISTORY_MAX;
s->n_history = HISTORY_MAX;
}
}
