Hi!
I enhanced QuickTimeFileSink to detect h264 IDR Frames.
If -y is set in openRTSP it now waits
*) until Audio is in sync first
*) then until h264 is in sync and an IDR Frame is received
This ensures that the recorded stream starts with an IDR Frame.
I've limited this detection to feeds with 2 sources (mostly audio+h264 I
guess).
And I've introduced class SyncFrame to store frame numbers on sync frames
for the stss atom. In case of h264 I store the IDR frame numbers and write
those instead of the fixed += 12 interval.
Last but not least I changed the hinf.payt atom payload format from 8 bit to
32 bit since IMO the specs says so. See Table 3-12 at
http://developer.apple.com/mac/library/documentation/QuickTime/QTFF/QTFFChap3/qtff3.html#//apple_ref/doc/uid/TP40000939-CH205-85342
Makes mp4dump happy so far;-)
Greetings, Wolfgang
PS: the patch works with --fuzz=0 for 2010.03.15
--
Wolfgang Breyha <[email protected]> | http://www.blafasel.at/
Vienna University Computer Center | Austria
--- live.orig/liveMedia/QuickTimeFileSink.cpp 2010-03-14 14:02:15.000000000
+0100
+++ live/liveMedia/QuickTimeFileSink.cpp 2010-03-14 13:58:10.000000000
+0100
@@ -32,6 +32,8 @@
#define fourChar(x,y,z,w) ( ((x)<<24)|((y)<<16)|((z)<<8)|(w) )
+#define H264_IDR_FRAME 0x65 //bit 8 == 0, bits 7-6 (ref) == 3, bits 5-0
(type) == 5
+
////////// SubsessionIOState, ChunkDescriptor ///////////
// A structure used to represent the I/O state of each input 'subsession':
@@ -84,6 +86,16 @@
unsigned fBytesInUse;
};
+class SyncFrame {
+public:
+ SyncFrame(unsigned frameNum);
+ virtual ~SyncFrame();
+
+public:
+ class SyncFrame *nextSyncFrame;
+ unsigned sfFrameNum;
+};
+
// A 64-bit counter, used below:
class Count64 {
@@ -155,6 +167,7 @@
// if there's a pause at the beginning
ChunkDescriptor *fHeadChunk, *fTailChunk;
+ SyncFrame *fHeadSyncFrame, *fTailSyncFrame;
unsigned fNumChunks;
// Counters to be used in the hint track's 'udta'/'hinf' atom;
@@ -511,7 +524,8 @@
MediaSubsession& subsession)
: fHintTrackForUs(NULL), fTrackHintedByUs(NULL),
fOurSink(sink), fOurSubsession(subsession),
- fLastPacketRTPSeqNum(0), fHaveBeenSynced(False), fQTTotNumSamples(0),
+ fLastPacketRTPSeqNum(0), fHaveBeenSynced(False), fQTTotNumSamples(0),
+ fHeadSyncFrame(NULL), fTailSyncFrame(NULL),
fHeadChunk(NULL), fTailChunk(NULL), fNumChunks(0) {
fTrackID = ++fCurrentTrackNumber;
@@ -529,7 +543,7 @@
SubsessionIOState::~SubsessionIOState() {
delete fBuffer; delete fPrevBuffer;
- delete fHeadChunk;
+ delete fHeadChunk, delete fHeadSyncFrame;
}
Boolean SubsessionIOState::setQTstate() {
@@ -787,6 +801,20 @@
sampleNumberOfFrameStart = fQTTotNumSamples + 1;
}
+ if (avcHack && (*frameSource == H264_IDR_FRAME))
+ {
+ SyncFrame* newSyncFrame = new SyncFrame(fQTTotNumSamples + 1);
+ if (fTailSyncFrame == NULL)
+ {
+ fHeadSyncFrame = newSyncFrame;
+ }
+ else
+ {
+ fTailSyncFrame->nextSyncFrame = newSyncFrame;
+ }
+ fTailSyncFrame = newSyncFrame;
+ }
+
// Remember the current frame for next time:
fPrevFrameState.frameSize = frameSize;
fPrevFrameState.presentationTime = presentationTime;
@@ -1057,6 +1085,25 @@
if (!fHaveBeenSynced) {
// We weren't synchronized before
if (fOurSubsession.rtpSource()->hasBeenSynchronizedUsingRTCP()) {
+ // H264 ?
+ if (fQTMediaDataAtomCreator == &QuickTimeFileSink::addAtom_avc1)
+ {
+ // special case: audio + H264 video: wait until audio is in sync
+ if ((s.fNumSubsessions == 2) && (s.fNumSyncedSubsessions <
(s.fNumSubsessions - 1)))
+ {
+ return false;
+ }
+ // if audio is in sync, wait for the next IDR frame to start
+ unsigned char* const frameSource = fBuffer->dataStart();
+ if (*frameSource != H264_IDR_FRAME)
+ {
+ return false;
+ }
+ else
+ {
+ fprintf(stderr, "IDR Frame found. Go!\n");
+ }
+ }
// But now we are
fHaveBeenSynced = True;
fSyncTime = presentationTime;
@@ -1082,6 +1129,15 @@
if (hintTrack != NULL) hintTrack->fTrackHintedByUs = hintedTrack;
}
+SyncFrame
+::SyncFrame(unsigned frameNum)
+ : nextSyncFrame(NULL), sfFrameNum(frameNum) {
+}
+
+SyncFrame::~SyncFrame() {
+ delete nextSyncFrame;
+}
+
void Count64::operator+=(unsigned arg) {
unsigned newLo = lo + arg;
if (newLo < lo) { // lo has overflowed
@@ -1953,29 +2009,42 @@
int64_t numEntriesPosition = TellFile64(fOutFid);
size += addWord(0); // dummy for "Number of entries"
- // Then, run through the chunk descriptors, counting up the total nuber of
samples:
unsigned numEntries = 0, numSamplesSoFar = 0;
- unsigned const samplesPerFrame = fCurrentIOState->fQTSamplesPerFrame;
- ChunkDescriptor* chunk = fCurrentIOState->fHeadChunk;
- while (chunk != NULL) {
- unsigned const numSamples = chunk->fNumFrames*samplesPerFrame;
- numSamplesSoFar += numSamples;
- chunk = chunk->fNextChunk;
- }
-
- // Then, write out the sample numbers that we deem correspond to 'sync
samples':
- unsigned i;
- for (i = 0; i < numSamplesSoFar; i += 12) {
- // For an explanation of the constant "12", see
http://lists.live555.com/pipermail/live-devel/2009-July/010969.html
- // (Perhaps we should really try to keep track of which 'samples'
('frames' for video) really are 'key frames'?)
- size += addWord(i+1);
- ++numEntries;
+ if (fCurrentIOState->fHeadSyncFrame != NULL)
+ {
+ SyncFrame* currentSyncFrame = fCurrentIOState->fHeadSyncFrame;
+ while(currentSyncFrame != NULL)
+ {
+ ++numEntries;
+ size += addWord(currentSyncFrame->sfFrameNum);
+ currentSyncFrame = currentSyncFrame->nextSyncFrame;
+ }
}
-
- // Then, write out the last entry (if we haven't already done so):
- if (i != (numSamplesSoFar - 1)) {
- size += addWord(numSamplesSoFar);
- ++numEntries;
+ else
+ {
+ // Then, run through the chunk descriptors, counting up the total nuber of
samples:
+ unsigned const samplesPerFrame = fCurrentIOState->fQTSamplesPerFrame;
+ ChunkDescriptor* chunk = fCurrentIOState->fHeadChunk;
+ while (chunk != NULL) {
+ unsigned const numSamples = chunk->fNumFrames*samplesPerFrame;
+ numSamplesSoFar += numSamples;
+ chunk = chunk->fNextChunk;
+ }
+
+ // Then, write out the sample numbers that we deem correspond to 'sync
samples':
+ unsigned i;
+ for (i = 0; i < numSamplesSoFar; i += 12) {
+ // For an explanation of the constant "12", see
http://lists.live555.com/pipermail/live-devel/2009-July/010969.html
+ // (Perhaps we should really try to keep track of which 'samples'
('frames' for video) really are 'key frames'?)
+ size += addWord(i+1);
+ ++numEntries;
+ }
+
+ // Then, write out the last entry (if we haven't already done so):
+ if (i != (numSamplesSoFar - 1)) {
+ size += addWord(numSamplesSoFar);
+ ++numEntries;
+ }
}
// Now go back and fill in the "Number of entries" field:
@@ -2236,7 +2305,7 @@
addAtom(payt);
MediaSubsession& ourSubsession = fCurrentIOState->fOurSubsession;
RTPSource* rtpSource = ourSubsession.rtpSource();
- size += addByte(rtpSource->rtpPayloadFormat());
+ size += addWord(rtpSource->rtpPayloadFormat());
// Also, add a 'rtpmap' string: <mime-subtype>/<rtp-frequency>
unsigned rtpmapStringLength = strlen(ourSubsession.codecName()) + 20;
_______________________________________________
live-devel mailing list
[email protected]
http://lists.live555.com/mailman/listinfo/live-devel
