//msg_Dbg(aout, "desync: %+"PRId64" us (variation: %+"PRId64" us)",
// delta, change);
+ const unsigned inrate = aout->format.i_rate;
+ unsigned outrate = sys->rate;
+ bool sync = false;
+
if (delta < -AOUT_MAX_PTS_DELAY)
msg_Warn(aout, "too late by %"PRId64" us", -delta);
else if (delta > +AOUT_MAX_PTS_ADVANCE)
msg_Warn(aout, "too early by %"PRId64" us", delta);
+ else if (outrate == inrate)
+ return; /* In sync, do not add unnecessary disturbance! */
+ else
+ sync = true;
/* Compute playback sample rate */
- const unsigned inrate = aout->format.i_rate;
- int limit = inrate / 100; /* max varation per iteration */
-
-#define ADJUST_FACTOR 4
/* This is empirical. Feel free to define something smarter. */
- int adj = sys->rate * (delta + change) / (CLOCK_FREQ * ADJUST_FACTOR);
-
- /* This avoids too fast rate variation. They sound ugly as hell and they
- * make the algorithm unstable (e.g. oscillation around inrate). */
+ int adj = sync ? (outrate - inrate)
+ : outrate * (delta + change) / (CLOCK_FREQ << 4);
+ /* This avoids too quick rate variation. It sounds really bad and
+ * causes unstability (e.g. oscillation around the correct rate). */
+ int limit = inrate >> 10;
+ /* However, to improve stability and try to converge, closing to the
+ * nominal rate is favored over drifting from it. */
+ if ((adj > 0) == (sys->rate > inrate))
+ limit *= 2;
if (adj > +limit)
adj = +limit;
if (adj < -limit)
adj = -limit;
-
- unsigned outrate = sys->rate - adj;
- /* Favor native rate to avoid resampling (FIXME: really a good idea?) */
- if (abs(outrate - inrate) < limit)
- outrate = inrate;
+ outrate -= adj;
/* This keeps the effective rate within specified range
* (+/-AOUT_MAX_RESAMPLING% - see <vlc_aout.h>) of the nominal rate. */
- limit *= AOUT_MAX_RESAMPLING;
+ limit = inrate * AOUT_MAX_RESAMPLING / 100;
if (outrate > inrate + limit)
outrate = inrate + limit;
if (outrate < inrate - limit)