// Default read ahead buffer size
mlt_properties_set_int( properties, "buffer", 25 );
+ mlt_properties_set_int( properties, "drop_max", 5 );
// Default audio frequency and channels
mlt_properties_set_int( properties, "frequency", 48000 );
static void mlt_consumer_property_changed( mlt_properties owner, mlt_consumer self, char *name )
{
- if ( !strcmp( name, "profile" ) )
+ if ( !strcmp( name, "mlt_profile" ) )
{
// Get the properies
mlt_properties properties = MLT_CONSUMER_PROPERTIES( self );
// For worker threads implementation, buffer must be at least # threads
if ( abs( self->real_time ) > 1 && mlt_properties_get_int( properties, "buffer" ) <= abs( self->real_time ) )
- mlt_properties_set_int( properties, "buffer", abs( self->real_time ) + 1 );
+ mlt_properties_set_int( properties, "_buffer", abs( self->real_time ) + 1 );
// Start the service
if ( self->start != NULL )
mlt_properties_set_double( frame_properties, "consumer_aspect_ratio", mlt_properties_get_double( properties, "aspect_ratio" ) );
mlt_properties_set_int( frame_properties, "consumer_deinterlace", mlt_properties_get_int( properties, "progressive" ) | mlt_properties_get_int( properties, "deinterlace" ) );
mlt_properties_set( frame_properties, "deinterlace_method", mlt_properties_get( properties, "deinterlace_method" ) );
+ mlt_properties_set_int( frame_properties, "consumer_tff", mlt_properties_get_int( properties, "top_field_first" ) );
}
// Return the frame
struct timeval ante;
// Average time for get_frame and get_image
- int count = 1;
+ int count = 0;
int skipped = 0;
- int64_t time_wait = 0;
- int64_t time_frame = 0;
int64_t time_process = 0;
int skip_next = 0;
mlt_position pos = 0;
mlt_position start_pos = 0;
mlt_position last_pos = 0;
+ int frame_duration = mlt_properties_get_int( properties, "frame_duration" );
+ int drop_max = mlt_properties_get_int( properties, "drop_max" );
if ( preview_off && preview_format != 0 )
self->format = preview_format;
// Continue to read ahead
while ( self->ahead )
{
- // Fetch width/height again
- width = mlt_properties_get_int( properties, "width" );
- height = mlt_properties_get_int( properties, "height" );
-
// Put the current frame into the queue
pthread_mutex_lock( &self->queue_mutex );
while( self->ahead && mlt_deque_count( self->queue ) >= buffer )
pthread_cond_broadcast( &self->queue_cond );
pthread_mutex_unlock( &self->queue_mutex );
- time_wait += time_difference( &ante );
-
// Get the next frame
frame = mlt_consumer_get_frame( self );
- time_frame += time_difference( &ante );
// If there's no frame, we're probably stopped...
if ( frame == NULL )
continue;
pos = mlt_frame_get_position( frame );
- // Increment the count
+ // Increment the counter used for averaging processing cost
count ++;
- // All non normal playback frames should be shown
+ // All non-normal playback frames should be shown
if ( mlt_properties_get_int( MLT_FRAME_PROPERTIES( frame ), "_speed" ) != 1 )
{
#ifdef DEINTERLACE_ON_NOT_NORMAL_SPEED
mlt_properties_set_int( MLT_FRAME_PROPERTIES( frame ), "consumer_deinterlace", 1 );
#endif
- skipped = 0;
- time_frame = 0;
- time_process = 0;
- time_wait = 0;
- count = 1;
- skip_next = 0;
+ // Indicate seeking or trick-play
start_pos = pos;
}
- // Get the image
+ // If skip flag not set or frame-dropping disabled
if ( !skip_next || self->real_time == -1 )
{
- // Get the image, mark as rendered and time it
if ( !video_off )
{
+ // Reset width/height - could have been changed by previous mlt_frame_get_image
+ width = mlt_properties_get_int( properties, "width" );
+ height = mlt_properties_get_int( properties, "height" );
+
+ // Get the image
mlt_events_fire( MLT_CONSUMER_PROPERTIES( self ), "consumer-frame-render", frame, NULL );
mlt_frame_get_image( frame, &image, &self->format, &width, &height, 0 );
}
+
+ // Indicate the rendered image is available.
mlt_properties_set_int( MLT_FRAME_PROPERTIES( frame ), "rendered", 1 );
+
+ // Reset consecutively-skipped counter
skipped = 0;
}
- else
+ else // Skip image processing
{
- // Increment the number of sequentially skipped frames
- skipped ++;
- skip_next = 0;
+ // Increment the number of consecutively-skipped frames
+ skipped++;
- // If we've reached an unacceptable level, reset everything
- if ( skipped > fps * 2 )
+ // If too many (1 sec) consecutively-skipped frames
+ if ( skipped > drop_max )
{
- skipped = 0;
- time_frame = 0;
+ // Reset cost tracker
time_process = 0;
- time_wait = 0;
count = 1;
+ mlt_log_verbose( self, "too many frames dropped - forcing next frame\n" );
}
}
mlt_frame_get_audio( frame, &audio, &afmt, &frequency, &channels, &samples );
}
- // Increment the time take for self frame
- time_process += time_difference( &ante );
+ // Get the time to process this frame
+ int64_t time_current = time_difference( &ante );
+
+ // If the current time is not suddenly some large amount
+ if ( time_current < time_process / count * 20 || !time_process || count < 5 )
+ {
+ // Accumulate the cost for processing this frame
+ time_process += time_current;
+ }
+ else
+ {
+ mlt_log_debug( self, "current %"PRId64" threshold %"PRId64" count %d\n",
+ time_current, (int64_t) (time_process / count * 20), count );
+ // Ignore the cost of this frame's time
+ count--;
+ }
- // Determine if the next frame should be skipped
+ // Determine if we started, resumed, or seeked
if ( pos != last_pos + 1 )
start_pos = pos;
last_pos = pos;
- if ( pos - start_pos <= buffer/5 )
+
+ // Do not skip the first 20% of buffer at start, resume, or seek
+ if ( pos - start_pos <= buffer / 5 + 1 )
{
- // Do not skip the first 20% of buffer at start, resuming, or seeking
- skipped = 0;
- time_frame = 0;
+ // Reset cost tracker
time_process = 0;
- time_wait = 0;
count = 1;
}
- if ( mlt_deque_count( self->queue ) <= buffer/5 )
+
+ // Reset skip flag
+ skip_next = 0;
+
+ // Only consider skipping if the buffer level is low (or really small)
+ if ( mlt_deque_count( self->queue ) <= buffer / 5 + 1 )
{
- int frame_duration = mlt_properties_get_int( properties, "frame_duration" );
- if ( ( ( time_wait + time_frame + time_process ) / count ) > frame_duration )
+ // Skip next frame if average cost exceeds frame duration.
+ if ( time_process / count > frame_duration )
skip_next = 1;
+ if ( skip_next )
+ mlt_log_debug( self, "avg usec %"PRId64" (%"PRId64"/%d) duration %d\n",
+ time_process/count, time_process, count, frame_duration);
}
}
// Frame to return
mlt_frame frame = NULL;
- int size = abs( self->real_time );
- int buffer = mlt_properties_get_int( properties, "buffer" );
+ double fps = mlt_properties_get_double( properties, "fps" );
+ int threads = abs( self->real_time );
+ int buffer = mlt_properties_get_int( properties, "_buffer" );
+ buffer = buffer > 0 ? buffer : mlt_properties_get_int( properties, "buffer" );
// This is a heuristic to determine a suitable minimum buffer size for the number of threads.
- int headroom = 2 + size * size;
+ int headroom = 2 + threads * threads;
buffer = buffer < headroom ? headroom : buffer;
// Start worker threads if not already started.
pthread_cond_wait( &self->done_cond, &self->done_mutex );
pthread_mutex_unlock( &self->done_mutex );
}
- self->process_head = size;
+ self->process_head = threads;
}
// mlt_log_verbose( MLT_CONSUMER_SERVICE(self), "size %d done count %d work count %d process_head %d\n",
-// size, first_unprocessed_frame( self ), mlt_deque_count( self->queue ), self->process_head );
+// threads, first_unprocessed_frame( self ), mlt_deque_count( self->queue ), self->process_head );
// Feed the work queue
while ( self->ahead && mlt_deque_count( self->queue ) < buffer )
}
// Wait if not realtime.
- while( self->ahead && self->real_time < 0 &&
- ! mlt_properties_get_int( MLT_FRAME_PROPERTIES( MLT_FRAME( mlt_deque_peek_front( self->queue ) ) ), "rendered" ) )
+ mlt_frame head_frame = MLT_FRAME( mlt_deque_peek_front( self->queue ) );
+ while ( self->ahead && self->real_time < 0 &&
+ !( head_frame && mlt_properties_get_int( MLT_FRAME_PROPERTIES( head_frame ), "rendered" ) ) )
{
pthread_mutex_lock( &self->done_mutex );
pthread_cond_wait( &self->done_cond, &self->done_mutex );
if ( frame && mlt_properties_get_int( MLT_FRAME_PROPERTIES( frame ), "rendered" ) )
{
self->consecutive_dropped = 0;
- if ( self->process_head > size && self->consecutive_rendered >= self->process_head )
+ if ( self->process_head > threads && self->consecutive_rendered >= self->process_head )
self->process_head--;
else
self->consecutive_rendered++;
else
{
self->consecutive_rendered = 0;
- if ( self->process_head < buffer - size && self->consecutive_dropped > size )
+ if ( self->process_head < buffer - threads && self->consecutive_dropped > threads )
self->process_head++;
else
self->consecutive_dropped++;
}
// mlt_log_verbose( MLT_CONSUMER_SERVICE(self), "dropped %d rendered %d process_head %d\n",
// self->consecutive_dropped, self->consecutive_rendered, self->process_head );
+
+ // Check for too many consecutively dropped frames
+ if ( self->consecutive_dropped > mlt_properties_get_int( properties, "drop_max" ) )
+ {
+ int orig_buffer = mlt_properties_get_int( properties, "buffer" );
+ int prefill = mlt_properties_get_int( properties, "prefill" );
+ mlt_log_verbose( self, "too many frames dropped - " );
+
+ // If using a default low-latency buffer level (SDL) and below the limit
+ if ( ( orig_buffer == 1 || prefill == 1 ) && buffer < (threads + 1) * 10 )
+ {
+ // Auto-scale the buffer to compensate
+ mlt_log_verbose( self, "increasing buffer to %d\n", buffer + threads );
+ mlt_properties_set_int( properties, "_buffer", buffer + threads );
+ self->consecutive_dropped = fps / 2;
+ }
+ else
+ {
+ // Tell the consumer to render it
+ mlt_log_verbose( self, "forcing next frame\n" );
+ mlt_properties_set_int( MLT_FRAME_PROPERTIES( frame ), "rendered", 1 );
+ self->consecutive_dropped = 0;
+ }
+ }
}
return frame;
projects / mlt / blobdiff