[MPlayer-dev-eng] [PATCH] more accurate sleeping for audio-only

[Reimar Döffinger]

Hello,
attached patch should reduce the number of useless wakeups in audio-only
mode (in my tests from about 90 to about 30 per second overall).
Not that much, but the code is not particularly complex either and has
caused no issues in my tests with different sample rates and -speed
values.

Greetings,
Reimar Döffinger
-------------- next part --------------
diff --git a/mplayer.c b/mplayer.c
index 66f1665..18be663 100644
--- a/mplayer.c
+++ b/mplayer.c
@@ -2015,6 +2015,7 @@ static int fill_audio_out_buffers(void)
     current_module="play_audio";
 
     while (1) {
+	int sleep_time;
 	// all the current uses of ao_data.pts seem to be in aos that handle
 	// sync completely wrong; there should be no need to use ao_data.pts
 	// in get_space()
@@ -2026,7 +2027,9 @@ static int fill_audio_out_buffers(void)
 	// handle audio-only case:
 	// this is where mplayer sleeps during audio-only playback
 	// to avoid 100% CPU use
-	usec_sleep(10000); // Wait a tick before retry
+	sleep_time = (ao_data.outburst - bytes_to_write) * 1000 / ao_data.samplerate;
+	if (sleep_time < 10) sleep_time = 10; // limit to 100 wakeups per second
+	usec_sleep(sleep_time * 1000);
     }
 
     while (bytes_to_write) {