| 1 | /********** |
| 2 | This library is free software; you can redistribute it and/or modify it under |
| 3 | the terms of the GNU Lesser General Public License as published by the |
| 4 | Free Software Foundation; either version 3 of the License, or (at your |
| 5 | option) any later version. (See <http://www.gnu.org/copyleft/lesser.html>.) |
| 6 | |
| 7 | This library is distributed in the hope that it will be useful, but WITHOUT |
| 8 | ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS |
| 9 | FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for |
| 10 | more details. |
| 11 | |
| 12 | You should have received a copy of the GNU Lesser General Public License |
| 13 | along with this library; if not, write to the Free Software Foundation, Inc., |
| 14 | 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 15 | **********/ |
| 16 | // "liveMedia" |
| 17 | // Copyright (c) 1996-2020 Live Networks, Inc. All rights reserved. |
| 18 | // A filter that produces a sequence of I-frame indices from a MPEG-2 Transport Stream |
| 19 | // Implementation |
| 20 | |
| 21 | #include "MPEG2IndexFromTransportStream.hh" |
| 22 | |
| 23 | ////////// IndexRecord definition ////////// |
| 24 | |
| 25 | enum RecordType { |
| 26 | RECORD_UNPARSED = 0, |
| 27 | RECORD_VSH = 1, // a MPEG Video Sequence Header |
| 28 | RECORD_GOP = 2, |
| 29 | RECORD_PIC_NON_IFRAME = 3, // includes slices |
| 30 | RECORD_PIC_IFRAME = 4, // includes slices |
| 31 | RECORD_NAL_H264_SPS = 5, // H.264 |
| 32 | RECORD_NAL_H264_PPS = 6, // H.264 |
| 33 | RECORD_NAL_H264_SEI = 7, // H.264 |
| 34 | RECORD_NAL_H264_NON_IFRAME = 8, // H.264 |
| 35 | RECORD_NAL_H264_IFRAME = 9, // H.264 |
| 36 | RECORD_NAL_H264_OTHER = 10, // H.264 |
| 37 | RECORD_NAL_H265_VPS = 11, // H.265 |
| 38 | RECORD_NAL_H265_SPS = 12, // H.265 |
| 39 | RECORD_NAL_H265_PPS = 13, // H.265 |
| 40 | RECORD_NAL_H265_NON_IFRAME = 14, // H.265 |
| 41 | RECORD_NAL_H265_IFRAME = 15, // H.265 |
| 42 | RECORD_NAL_H265_OTHER = 16, // H.265 |
| 43 | RECORD_JUNK |
| 44 | }; |
| 45 | |
| 46 | class IndexRecord { |
| 47 | public: |
| 48 | IndexRecord(u_int8_t startOffset, u_int8_t size, |
| 49 | unsigned long transportPacketNumber, float pcr); |
| 50 | virtual ~IndexRecord(); |
| 51 | |
| 52 | RecordType& recordType() { return fRecordType; } |
| 53 | void setFirstFlag() { fRecordType = (RecordType)(((u_int8_t)fRecordType) | 0x80); } |
| 54 | u_int8_t startOffset() const { return fStartOffset; } |
| 55 | u_int8_t& size() { return fSize; } |
| 56 | float pcr() const { return fPCR; } |
| 57 | unsigned long transportPacketNumber() const { return fTransportPacketNumber; } |
| 58 | |
| 59 | IndexRecord* next() const { return fNext; } |
| 60 | void addAfter(IndexRecord* prev); |
| 61 | void unlink(); |
| 62 | |
| 63 | private: |
| 64 | // Index records are maintained in a doubly-linked list: |
| 65 | IndexRecord* fNext; |
| 66 | IndexRecord* fPrev; |
| 67 | |
| 68 | RecordType fRecordType; |
| 69 | u_int8_t fStartOffset; // within the Transport Stream packet |
| 70 | u_int8_t fSize; // in bytes, following "fStartOffset". |
| 71 | // Note: fStartOffset + fSize <= TRANSPORT_PACKET_SIZE |
| 72 | float fPCR; |
| 73 | unsigned long fTransportPacketNumber; |
| 74 | }; |
| 75 | |
| 76 | #ifdef DEBUG |
| 77 | static char const* recordTypeStr[] = { |
| 78 | "UNPARSED" , |
| 79 | "VSH" , |
| 80 | "GOP" , |
| 81 | "PIC(non-I-frame)" , |
| 82 | "PIC(I-frame)" , |
| 83 | "SPS (H.264)" , |
| 84 | "PPS (H.264)" , |
| 85 | "SEI (H.264)" , |
| 86 | "H.264 non-I-frame" , |
| 87 | "H.264 I-frame" , |
| 88 | "other NAL unit (H.264)" , |
| 89 | "VPS (H.265)" , |
| 90 | "SPS (H.265)" , |
| 91 | "PPS (H.265)" , |
| 92 | "H.265 non-I-frame" , |
| 93 | "H.265 I-frame" , |
| 94 | "other NAL unit (H.265)" , |
| 95 | "JUNK" |
| 96 | }; |
| 97 | |
| 98 | UsageEnvironment& operator<<(UsageEnvironment& env, IndexRecord& r) { |
| 99 | return env << "[" << ((r.recordType()&0x80) != 0 ? "1" : "" ) |
| 100 | << recordTypeStr[r.recordType()&0x7F] << ":" |
| 101 | << (unsigned)r.transportPacketNumber() << ":" << r.startOffset() |
| 102 | << "(" << r.size() << ")@" << r.pcr() << "]" ; |
| 103 | } |
| 104 | #endif |
| 105 | |
| 106 | |
| 107 | ////////// MPEG2IFrameIndexFromTransportStream implementation ////////// |
| 108 | |
| 109 | MPEG2IFrameIndexFromTransportStream* |
| 110 | MPEG2IFrameIndexFromTransportStream::createNew(UsageEnvironment& env, |
| 111 | FramedSource* inputSource) { |
| 112 | return new MPEG2IFrameIndexFromTransportStream(env, inputSource); |
| 113 | } |
| 114 | |
| 115 | // The largest expected frame size (in bytes): |
| 116 | #define MAX_FRAME_SIZE 400000 |
| 117 | |
| 118 | // Make our parse buffer twice as large as this, to ensure that at least one |
| 119 | // complete frame will fit inside it: |
| 120 | #define PARSE_BUFFER_SIZE (2*MAX_FRAME_SIZE) |
| 121 | |
| 122 | // The PID used for the PAT (as defined in the MPEG Transport Stream standard): |
| 123 | #define PAT_PID 0 |
| 124 | |
| 125 | MPEG2IFrameIndexFromTransportStream |
| 126 | ::MPEG2IFrameIndexFromTransportStream(UsageEnvironment& env, |
| 127 | FramedSource* inputSource) |
| 128 | : FramedFilter(env, inputSource), |
| 129 | fIsH264(False), fIsH265(False), |
| 130 | fInputTransportPacketCounter((unsigned)-1), fClosureNumber(0), fLastContinuityCounter(~0), |
| 131 | fFirstPCR(0.0), fLastPCR(0.0), fHaveSeenFirstPCR(False), |
| 132 | fPMT_PID(0x10), fVideo_PID(0xE0), // default values |
| 133 | fParseBufferSize(PARSE_BUFFER_SIZE), |
| 134 | fParseBufferFrameStart(0), fParseBufferParseEnd(4), fParseBufferDataEnd(0), |
| 135 | fHeadIndexRecord(NULL), fTailIndexRecord(NULL) { |
| 136 | fParseBuffer = new unsigned char[fParseBufferSize]; |
| 137 | } |
| 138 | |
| 139 | MPEG2IFrameIndexFromTransportStream::~MPEG2IFrameIndexFromTransportStream() { |
| 140 | delete fHeadIndexRecord; |
| 141 | delete[] fParseBuffer; |
| 142 | } |
| 143 | |
| 144 | void MPEG2IFrameIndexFromTransportStream::doGetNextFrame() { |
| 145 | // Begin by trying to deliver an index record (for an already-parsed frame) |
| 146 | // to the client: |
| 147 | if (deliverIndexRecord()) return; |
| 148 | |
| 149 | // No more index records are left to deliver, so try to parse a new frame: |
| 150 | if (parseFrame()) { // success - try again |
| 151 | doGetNextFrame(); |
| 152 | return; |
| 153 | } |
| 154 | |
| 155 | // We need to read some more Transport Stream packets. Check whether we have room: |
| 156 | if (fParseBufferSize - fParseBufferDataEnd < TRANSPORT_PACKET_SIZE) { |
| 157 | // There's no room left. Compact the buffer, and check again: |
| 158 | compactParseBuffer(); |
| 159 | if (fParseBufferSize - fParseBufferDataEnd < TRANSPORT_PACKET_SIZE) { |
| 160 | envir() << "ERROR: parse buffer full; increase MAX_FRAME_SIZE\n" ; |
| 161 | // Treat this as if the input source ended: |
| 162 | handleInputClosure1(); |
| 163 | return; |
| 164 | } |
| 165 | } |
| 166 | |
| 167 | // Arrange to read a new Transport Stream packet: |
| 168 | fInputSource->getNextFrame(fInputBuffer, sizeof fInputBuffer, |
| 169 | afterGettingFrame, this, |
| 170 | handleInputClosure, this); |
| 171 | } |
| 172 | |
| 173 | void MPEG2IFrameIndexFromTransportStream |
| 174 | ::afterGettingFrame(void* clientData, unsigned frameSize, |
| 175 | unsigned numTruncatedBytes, |
| 176 | struct timeval presentationTime, |
| 177 | unsigned durationInMicroseconds) { |
| 178 | MPEG2IFrameIndexFromTransportStream* source |
| 179 | = (MPEG2IFrameIndexFromTransportStream*)clientData; |
| 180 | source->afterGettingFrame1(frameSize, numTruncatedBytes, |
| 181 | presentationTime, durationInMicroseconds); |
| 182 | } |
| 183 | |
| 184 | #define TRANSPORT_SYNC_BYTE 0x47 |
| 185 | |
| 186 | void MPEG2IFrameIndexFromTransportStream |
| 187 | ::afterGettingFrame1(unsigned frameSize, |
| 188 | unsigned numTruncatedBytes, |
| 189 | struct timeval presentationTime, |
| 190 | unsigned durationInMicroseconds) { |
| 191 | if (frameSize < TRANSPORT_PACKET_SIZE || fInputBuffer[0] != TRANSPORT_SYNC_BYTE) { |
| 192 | if (fInputBuffer[0] != TRANSPORT_SYNC_BYTE) { |
| 193 | envir() << "Bad TS sync byte: 0x" << fInputBuffer[0] << "\n" ; |
| 194 | } |
| 195 | // Handle this as if the source ended: |
| 196 | handleInputClosure1(); |
| 197 | return; |
| 198 | } |
| 199 | |
| 200 | ++fInputTransportPacketCounter; |
| 201 | |
| 202 | // Figure out how much of this Transport Packet contains PES data: |
| 203 | u_int8_t adaptation_field_control = (fInputBuffer[3]&0x30)>>4; |
| 204 | u_int8_t |
| 205 | = adaptation_field_control <= 1 ? 4 : 5 + fInputBuffer[4]; |
| 206 | if ((adaptation_field_control == 2 && totalHeaderSize != TRANSPORT_PACKET_SIZE) || |
| 207 | (adaptation_field_control == 3 && totalHeaderSize >= TRANSPORT_PACKET_SIZE)) { |
| 208 | envir() << "Bad \"adaptation_field_length\": " << fInputBuffer[4] << "\n" ; |
| 209 | doGetNextFrame(); |
| 210 | return; |
| 211 | } |
| 212 | |
| 213 | // Check for a PCR: |
| 214 | if (totalHeaderSize > 5 && (fInputBuffer[5]&0x10) != 0) { |
| 215 | // There's a PCR: |
| 216 | u_int32_t pcrBaseHigh |
| 217 | = (fInputBuffer[6]<<24)|(fInputBuffer[7]<<16) |
| 218 | |(fInputBuffer[8]<<8)|fInputBuffer[9]; |
| 219 | float pcr = pcrBaseHigh/45000.0f; |
| 220 | if ((fInputBuffer[10]&0x80) != 0) pcr += 1/90000.0f; // add in low-bit (if set) |
| 221 | unsigned short pcrExt = ((fInputBuffer[10]&0x01)<<8) | fInputBuffer[11]; |
| 222 | pcr += pcrExt/27000000.0f; |
| 223 | |
| 224 | if (!fHaveSeenFirstPCR) { |
| 225 | fFirstPCR = pcr; |
| 226 | fHaveSeenFirstPCR = True; |
| 227 | } else if (pcr < fLastPCR) { |
| 228 | // The PCR timestamp has gone backwards. Display a warning about this |
| 229 | // (because it indicates buggy Transport Stream data), and compensate for it. |
| 230 | envir() << "\nWarning: At about " << fLastPCR-fFirstPCR |
| 231 | << " seconds into the file, the PCR timestamp decreased - from " |
| 232 | << fLastPCR << " to " << pcr << "\n" ; |
| 233 | fFirstPCR -= (fLastPCR - pcr); |
| 234 | } |
| 235 | fLastPCR = pcr; |
| 236 | } |
| 237 | |
| 238 | // Get the PID from the packet, and check for special tables: the PAT and PMT: |
| 239 | u_int16_t PID = ((fInputBuffer[1]&0x1F)<<8) | fInputBuffer[2]; |
| 240 | if (PID == PAT_PID) { |
| 241 | analyzePAT(&fInputBuffer[totalHeaderSize], TRANSPORT_PACKET_SIZE-totalHeaderSize); |
| 242 | } else if (PID == fPMT_PID) { |
| 243 | analyzePMT(&fInputBuffer[totalHeaderSize], TRANSPORT_PACKET_SIZE-totalHeaderSize); |
| 244 | } |
| 245 | |
| 246 | // Ignore transport packets for non-video programs, |
| 247 | // or packets with no data, or packets that duplicate the previous packet: |
| 248 | u_int8_t continuity_counter = fInputBuffer[3]&0x0F; |
| 249 | if ((PID != fVideo_PID) || |
| 250 | !(adaptation_field_control == 1 || adaptation_field_control == 3) || |
| 251 | continuity_counter == fLastContinuityCounter) { |
| 252 | doGetNextFrame(); |
| 253 | return; |
| 254 | } |
| 255 | fLastContinuityCounter = continuity_counter; |
| 256 | |
| 257 | // Also, if this is the start of a PES packet, then skip over the PES header: |
| 258 | Boolean payload_unit_start_indicator = (fInputBuffer[1]&0x40) != 0; |
| 259 | if (payload_unit_start_indicator && totalHeaderSize < TRANSPORT_PACKET_SIZE - 8 |
| 260 | && fInputBuffer[totalHeaderSize] == 0x00 && fInputBuffer[totalHeaderSize+1] == 0x00 |
| 261 | && fInputBuffer[totalHeaderSize+2] == 0x01) { |
| 262 | u_int8_t = fInputBuffer[totalHeaderSize+8]; |
| 263 | totalHeaderSize += 9 + PES_header_data_length; |
| 264 | if (totalHeaderSize >= TRANSPORT_PACKET_SIZE) { |
| 265 | envir() << "Unexpectedly large PES header size: " << PES_header_data_length << "\n" ; |
| 266 | // Handle this as if the source ended: |
| 267 | handleInputClosure1(); |
| 268 | return; |
| 269 | } |
| 270 | } |
| 271 | |
| 272 | // The remaining data is Video Elementary Stream data. Add it to our parse buffer: |
| 273 | unsigned vesSize = TRANSPORT_PACKET_SIZE - totalHeaderSize; |
| 274 | memmove(&fParseBuffer[fParseBufferDataEnd], &fInputBuffer[totalHeaderSize], vesSize); |
| 275 | fParseBufferDataEnd += vesSize; |
| 276 | |
| 277 | // And add a new index record noting where it came from: |
| 278 | addToTail(new IndexRecord(totalHeaderSize, vesSize, fInputTransportPacketCounter, |
| 279 | fLastPCR - fFirstPCR)); |
| 280 | |
| 281 | // Try again: |
| 282 | doGetNextFrame(); |
| 283 | } |
| 284 | |
| 285 | void MPEG2IFrameIndexFromTransportStream::handleInputClosure(void* clientData) { |
| 286 | MPEG2IFrameIndexFromTransportStream* source |
| 287 | = (MPEG2IFrameIndexFromTransportStream*)clientData; |
| 288 | source->handleInputClosure1(); |
| 289 | } |
| 290 | |
| 291 | #define VIDEO_SEQUENCE_START_CODE 0xB3 // MPEG-1 or 2 |
| 292 | #define VISUAL_OBJECT_SEQUENCE_START_CODE 0xB0 // MPEG-4 |
| 293 | #define GROUP_START_CODE 0xB8 // MPEG-1 or 2 |
| 294 | #define GROUP_VOP_START_CODE 0xB3 // MPEG-4 |
| 295 | #define PICTURE_START_CODE 0x00 // MPEG-1 or 2 |
| 296 | #define VOP_START_CODE 0xB6 // MPEG-4 |
| 297 | |
| 298 | void MPEG2IFrameIndexFromTransportStream::handleInputClosure1() { |
| 299 | if (++fClosureNumber == 1 && fParseBufferDataEnd > fParseBufferFrameStart |
| 300 | && fParseBufferDataEnd <= fParseBufferSize - 4) { |
| 301 | // This is the first time we saw EOF, and there's still data remaining to be |
| 302 | // parsed. Hack: Append a Picture Header code to the end of the unparsed |
| 303 | // data, and try again. This should use up all of the unparsed data. |
| 304 | fParseBuffer[fParseBufferDataEnd++] = 0; |
| 305 | fParseBuffer[fParseBufferDataEnd++] = 0; |
| 306 | fParseBuffer[fParseBufferDataEnd++] = 1; |
| 307 | fParseBuffer[fParseBufferDataEnd++] = PICTURE_START_CODE; |
| 308 | |
| 309 | // Try again: |
| 310 | doGetNextFrame(); |
| 311 | } else { |
| 312 | // Handle closure in the regular way: |
| 313 | handleClosure(); |
| 314 | } |
| 315 | } |
| 316 | |
| 317 | void MPEG2IFrameIndexFromTransportStream |
| 318 | ::analyzePAT(unsigned char* pkt, unsigned size) { |
| 319 | // Get the PMT_PID: |
| 320 | while (size >= 17) { // The table is large enough |
| 321 | u_int16_t program_number = (pkt[9]<<8) | pkt[10]; |
| 322 | if (program_number != 0) { |
| 323 | fPMT_PID = ((pkt[11]&0x1F)<<8) | pkt[12]; |
| 324 | return; |
| 325 | } |
| 326 | |
| 327 | pkt += 4; size -= 4; |
| 328 | } |
| 329 | } |
| 330 | |
| 331 | void MPEG2IFrameIndexFromTransportStream |
| 332 | ::analyzePMT(unsigned char* pkt, unsigned size) { |
| 333 | // Scan the "elementary_PID"s in the map, until we see the first video stream. |
| 334 | |
| 335 | // First, get the "section_length", to get the table's size: |
| 336 | u_int16_t section_length = ((pkt[2]&0x0F)<<8) | pkt[3]; |
| 337 | if ((unsigned)(4+section_length) < size) size = (4+section_length); |
| 338 | |
| 339 | // Then, skip any descriptors following the "program_info_length": |
| 340 | if (size < 22) return; // not enough data |
| 341 | unsigned program_info_length = ((pkt[11]&0x0F)<<8) | pkt[12]; |
| 342 | pkt += 13; size -= 13; |
| 343 | if (size < program_info_length) return; // not enough data |
| 344 | pkt += program_info_length; size -= program_info_length; |
| 345 | |
| 346 | // Look at each ("stream_type","elementary_PID") pair, looking for a video stream: |
| 347 | while (size >= 9) { |
| 348 | u_int8_t stream_type = pkt[0]; |
| 349 | u_int16_t elementary_PID = ((pkt[1]&0x1F)<<8) | pkt[2]; |
| 350 | if (stream_type == 1 || stream_type == 2 || |
| 351 | stream_type == 0x1B/*H.264 video*/ || stream_type == 0x24/*H.265 video*/) { |
| 352 | if (stream_type == 0x1B) fIsH264 = True; |
| 353 | else if (stream_type == 0x24) fIsH265 = True; |
| 354 | fVideo_PID = elementary_PID; |
| 355 | return; |
| 356 | } |
| 357 | |
| 358 | u_int16_t ES_info_length = ((pkt[3]&0x0F)<<8) | pkt[4]; |
| 359 | pkt += 5; size -= 5; |
| 360 | if (size < ES_info_length) return; // not enough data |
| 361 | pkt += ES_info_length; size -= ES_info_length; |
| 362 | } |
| 363 | } |
| 364 | |
| 365 | Boolean MPEG2IFrameIndexFromTransportStream::deliverIndexRecord() { |
| 366 | IndexRecord* head = fHeadIndexRecord; |
| 367 | if (head == NULL) return False; |
| 368 | |
| 369 | // Check whether the head record has been parsed yet: |
| 370 | if (head->recordType() == RECORD_UNPARSED) return False; |
| 371 | |
| 372 | // Remove the head record (the one whose data we'll be delivering): |
| 373 | IndexRecord* next = head->next(); |
| 374 | head->unlink(); |
| 375 | if (next == head) { |
| 376 | fHeadIndexRecord = fTailIndexRecord = NULL; |
| 377 | } else { |
| 378 | fHeadIndexRecord = next; |
| 379 | } |
| 380 | |
| 381 | if (head->recordType() == RECORD_JUNK) { |
| 382 | // Don't actually deliver the data to the client: |
| 383 | delete head; |
| 384 | // Try to deliver the next record instead: |
| 385 | return deliverIndexRecord(); |
| 386 | } |
| 387 | |
| 388 | // Deliver data from the head record: |
| 389 | #ifdef DEBUG |
| 390 | envir() << "delivering: " << *head << "\n" ; |
| 391 | #endif |
| 392 | if (fMaxSize < 11) { |
| 393 | fFrameSize = 0; |
| 394 | } else { |
| 395 | fTo[0] = (u_int8_t)(head->recordType()); |
| 396 | fTo[1] = head->startOffset(); |
| 397 | fTo[2] = head->size(); |
| 398 | // Deliver the PCR, as 24 bits (integer part; little endian) + 8 bits (fractional part) |
| 399 | float pcr = head->pcr(); |
| 400 | unsigned pcr_int = (unsigned)pcr; |
| 401 | u_int8_t pcr_frac = (u_int8_t)(256*(pcr-pcr_int)); |
| 402 | fTo[3] = (unsigned char)(pcr_int); |
| 403 | fTo[4] = (unsigned char)(pcr_int>>8); |
| 404 | fTo[5] = (unsigned char)(pcr_int>>16); |
| 405 | fTo[6] = (unsigned char)(pcr_frac); |
| 406 | // Deliver the transport packet number (in little-endian order): |
| 407 | unsigned long tpn = head->transportPacketNumber(); |
| 408 | fTo[7] = (unsigned char)(tpn); |
| 409 | fTo[8] = (unsigned char)(tpn>>8); |
| 410 | fTo[9] = (unsigned char)(tpn>>16); |
| 411 | fTo[10] = (unsigned char)(tpn>>24); |
| 412 | fFrameSize = 11; |
| 413 | } |
| 414 | |
| 415 | // Free the (former) head record (as we're now done with it): |
| 416 | delete head; |
| 417 | |
| 418 | // Complete delivery to the client: |
| 419 | afterGetting(this); |
| 420 | return True; |
| 421 | } |
| 422 | |
| 423 | Boolean MPEG2IFrameIndexFromTransportStream::parseFrame() { |
| 424 | // At this point, we have a queue of >=0 (unparsed) index records, representing |
| 425 | // the data in the parse buffer from "fParseBufferFrameStart" |
| 426 | // to "fParseBufferDataEnd". We now parse through this data, looking for |
| 427 | // a complete 'frame', where a 'frame', in this case, means: |
| 428 | // for MPEG video: a Video Sequence Header, GOP Header, Picture Header, or Slice |
| 429 | // for H.264 or H.265 video: a NAL unit |
| 430 | |
| 431 | // Inspect the frame's initial 4-byte code, to make sure it starts with a system code: |
| 432 | if (fParseBufferDataEnd-fParseBufferFrameStart < 4) return False; // not enough data |
| 433 | unsigned numInitialBadBytes = 0; |
| 434 | unsigned char const* p = &fParseBuffer[fParseBufferFrameStart]; |
| 435 | if (!(p[0] == 0 && p[1] == 0 && p[2] == 1)) { |
| 436 | // There's no system code at the beginning. Parse until we find one: |
| 437 | if (fParseBufferParseEnd == fParseBufferFrameStart + 4) { |
| 438 | // Start parsing from the beginning of the frame data: |
| 439 | fParseBufferParseEnd = fParseBufferFrameStart; |
| 440 | } |
| 441 | unsigned char nextCode; |
| 442 | if (!parseToNextCode(nextCode)) return False; |
| 443 | |
| 444 | numInitialBadBytes = fParseBufferParseEnd - fParseBufferFrameStart; |
| 445 | fParseBufferFrameStart = fParseBufferParseEnd; |
| 446 | fParseBufferParseEnd += 4; // skip over the code that we just saw |
| 447 | p = &fParseBuffer[fParseBufferFrameStart]; |
| 448 | } |
| 449 | |
| 450 | unsigned char curCode = p[3]; |
| 451 | if (fIsH264) curCode &= 0x1F; // nal_unit_type |
| 452 | else if (fIsH265) curCode = (curCode&0x7E)>>1; |
| 453 | |
| 454 | RecordType curRecordType; |
| 455 | unsigned char nextCode; |
| 456 | if (fIsH264) { |
| 457 | switch (curCode) { |
| 458 | case 1: // Coded slice of a non-IDR picture |
| 459 | curRecordType = RECORD_NAL_H264_NON_IFRAME; |
| 460 | if (!parseToNextCode(nextCode)) return False; |
| 461 | break; |
| 462 | case 5: // Coded slice of an IDR picture |
| 463 | curRecordType = RECORD_NAL_H264_IFRAME; |
| 464 | if (!parseToNextCode(nextCode)) return False; |
| 465 | break; |
| 466 | case 6: // Supplemental enhancement information (SEI) |
| 467 | curRecordType = RECORD_NAL_H264_SEI; |
| 468 | if (!parseToNextCode(nextCode)) return False; |
| 469 | break; |
| 470 | case 7: // Sequence parameter set (SPS) |
| 471 | curRecordType = RECORD_NAL_H264_SPS; |
| 472 | if (!parseToNextCode(nextCode)) return False; |
| 473 | break; |
| 474 | case 8: // Picture parameter set (PPS) |
| 475 | curRecordType = RECORD_NAL_H264_PPS; |
| 476 | if (!parseToNextCode(nextCode)) return False; |
| 477 | break; |
| 478 | default: |
| 479 | curRecordType = RECORD_NAL_H264_OTHER; |
| 480 | if (!parseToNextCode(nextCode)) return False; |
| 481 | break; |
| 482 | } |
| 483 | } else if (fIsH265) { |
| 484 | switch (curCode) { |
| 485 | case 19: // Coded slice segment of an IDR picture |
| 486 | case 20: // Coded slice segment of an IDR picture |
| 487 | curRecordType = RECORD_NAL_H265_IFRAME; |
| 488 | if (!parseToNextCode(nextCode)) return False; |
| 489 | break; |
| 490 | case 32: // Video parameter set (VPS) |
| 491 | curRecordType = RECORD_NAL_H265_VPS; |
| 492 | if (!parseToNextCode(nextCode)) return False; |
| 493 | break; |
| 494 | case 33: // Sequence parameter set (SPS) |
| 495 | curRecordType = RECORD_NAL_H265_SPS; |
| 496 | if (!parseToNextCode(nextCode)) return False; |
| 497 | break; |
| 498 | case 34: // Picture parameter set (PPS) |
| 499 | curRecordType = RECORD_NAL_H265_PPS; |
| 500 | if (!parseToNextCode(nextCode)) return False; |
| 501 | break; |
| 502 | default: |
| 503 | curRecordType = (curCode <= 31) ? RECORD_NAL_H265_NON_IFRAME : RECORD_NAL_H265_OTHER; |
| 504 | if (!parseToNextCode(nextCode)) return False; |
| 505 | break; |
| 506 | } |
| 507 | } else { // MPEG-1, 2, or 4 |
| 508 | switch (curCode) { |
| 509 | case VIDEO_SEQUENCE_START_CODE: |
| 510 | case VISUAL_OBJECT_SEQUENCE_START_CODE: |
| 511 | curRecordType = RECORD_VSH; |
| 512 | while (1) { |
| 513 | if (!parseToNextCode(nextCode)) return False; |
| 514 | if (nextCode == GROUP_START_CODE || |
| 515 | nextCode == PICTURE_START_CODE || nextCode == VOP_START_CODE) break; |
| 516 | fParseBufferParseEnd += 4; // skip over the code that we just saw |
| 517 | } |
| 518 | break; |
| 519 | case GROUP_START_CODE: |
| 520 | curRecordType = RECORD_GOP; |
| 521 | while (1) { |
| 522 | if (!parseToNextCode(nextCode)) return False; |
| 523 | if (nextCode == PICTURE_START_CODE || nextCode == VOP_START_CODE) break; |
| 524 | fParseBufferParseEnd += 4; // skip over the code that we just saw |
| 525 | } |
| 526 | break; |
| 527 | default: // picture |
| 528 | curRecordType = RECORD_PIC_NON_IFRAME; // may get changed to IFRAME later |
| 529 | while (1) { |
| 530 | if (!parseToNextCode(nextCode)) return False; |
| 531 | if (nextCode == VIDEO_SEQUENCE_START_CODE || |
| 532 | nextCode == VISUAL_OBJECT_SEQUENCE_START_CODE || |
| 533 | nextCode == GROUP_START_CODE || nextCode == GROUP_VOP_START_CODE || |
| 534 | nextCode == PICTURE_START_CODE || nextCode == VOP_START_CODE) break; |
| 535 | fParseBufferParseEnd += 4; // skip over the code that we just saw |
| 536 | } |
| 537 | break; |
| 538 | } |
| 539 | } |
| 540 | |
| 541 | if (curRecordType == RECORD_PIC_NON_IFRAME) { |
| 542 | if (curCode == VOP_START_CODE) { // MPEG-4 |
| 543 | if ((fParseBuffer[fParseBufferFrameStart+4]&0xC0) == 0) { |
| 544 | // This is actually an I-frame. Note it as such: |
| 545 | curRecordType = RECORD_PIC_IFRAME; |
| 546 | } |
| 547 | } else { // MPEG-1 or 2 |
| 548 | if ((fParseBuffer[fParseBufferFrameStart+5]&0x38) == 0x08) { |
| 549 | // This is actually an I-frame. Note it as such: |
| 550 | curRecordType = RECORD_PIC_IFRAME; |
| 551 | } |
| 552 | } |
| 553 | } |
| 554 | |
| 555 | // There is now a parsed 'frame', from "fParseBufferFrameStart" |
| 556 | // to "fParseBufferParseEnd". Tag the corresponding index records to note this: |
| 557 | unsigned frameSize = fParseBufferParseEnd - fParseBufferFrameStart + numInitialBadBytes; |
| 558 | #ifdef DEBUG |
| 559 | envir() << "parsed " << recordTypeStr[curRecordType] << "; length " |
| 560 | << frameSize << "\n" ; |
| 561 | #endif |
| 562 | for (IndexRecord* r = fHeadIndexRecord; ; r = r->next()) { |
| 563 | if (numInitialBadBytes >= r->size()) { |
| 564 | r->recordType() = RECORD_JUNK; |
| 565 | numInitialBadBytes -= r->size(); |
| 566 | } else { |
| 567 | r->recordType() = curRecordType; |
| 568 | } |
| 569 | if (r == fHeadIndexRecord) r->setFirstFlag(); |
| 570 | // indicates that this is the first record for this frame |
| 571 | |
| 572 | if (r->size() > frameSize) { |
| 573 | // This record contains extra data that's not part of the frame. |
| 574 | // Shorten this record, and move the extra data to a new record |
| 575 | // that comes afterwards: |
| 576 | u_int8_t newOffset = r->startOffset() + frameSize; |
| 577 | u_int8_t newSize = r->size() - frameSize; |
| 578 | r->size() = frameSize; |
| 579 | #ifdef DEBUG |
| 580 | envir() << "tagged record (modified): " << *r << "\n" ; |
| 581 | #endif |
| 582 | |
| 583 | IndexRecord* newRecord |
| 584 | = new IndexRecord(newOffset, newSize, r->transportPacketNumber(), r->pcr()); |
| 585 | newRecord->addAfter(r); |
| 586 | if (fTailIndexRecord == r) fTailIndexRecord = newRecord; |
| 587 | #ifdef DEBUG |
| 588 | envir() << "added extra record: " << *newRecord << "\n" ; |
| 589 | #endif |
| 590 | } else { |
| 591 | #ifdef DEBUG |
| 592 | envir() << "tagged record: " << *r << "\n" ; |
| 593 | #endif |
| 594 | } |
| 595 | frameSize -= r->size(); |
| 596 | if (frameSize == 0) break; |
| 597 | if (r == fTailIndexRecord) { // this shouldn't happen |
| 598 | envir() << "!!!!!Internal consistency error!!!!!\n" ; |
| 599 | return False; |
| 600 | } |
| 601 | } |
| 602 | |
| 603 | // Finally, update our parse state (to skip over the now-parsed data): |
| 604 | fParseBufferFrameStart = fParseBufferParseEnd; |
| 605 | fParseBufferParseEnd += 4; // to skip over the next code (that we found) |
| 606 | |
| 607 | return True; |
| 608 | } |
| 609 | |
| 610 | Boolean MPEG2IFrameIndexFromTransportStream |
| 611 | ::parseToNextCode(unsigned char& nextCode) { |
| 612 | unsigned char const* p = &fParseBuffer[fParseBufferParseEnd]; |
| 613 | unsigned char const* end = &fParseBuffer[fParseBufferDataEnd]; |
| 614 | while (p <= end-4) { |
| 615 | if (p[2] > 1) p += 3; // common case (optimized) |
| 616 | else if (p[2] == 0) ++p; |
| 617 | else if (p[0] == 0 && p[1] == 0) { // && p[2] == 1 |
| 618 | // We found a code here: |
| 619 | nextCode = p[3]; |
| 620 | fParseBufferParseEnd = p - &fParseBuffer[0]; // where we've gotten to |
| 621 | return True; |
| 622 | } else p += 3; |
| 623 | } |
| 624 | |
| 625 | fParseBufferParseEnd = p - &fParseBuffer[0]; // where we've gotten to |
| 626 | return False; // no luck this time |
| 627 | } |
| 628 | |
| 629 | void MPEG2IFrameIndexFromTransportStream::compactParseBuffer() { |
| 630 | #ifdef DEBUG |
| 631 | envir() << "Compacting parse buffer: [" << fParseBufferFrameStart |
| 632 | << "," << fParseBufferParseEnd << "," << fParseBufferDataEnd << "]" ; |
| 633 | #endif |
| 634 | memmove(&fParseBuffer[0], &fParseBuffer[fParseBufferFrameStart], |
| 635 | fParseBufferDataEnd - fParseBufferFrameStart); |
| 636 | fParseBufferDataEnd -= fParseBufferFrameStart; |
| 637 | fParseBufferParseEnd -= fParseBufferFrameStart; |
| 638 | fParseBufferFrameStart = 0; |
| 639 | #ifdef DEBUG |
| 640 | envir() << "-> [" << fParseBufferFrameStart |
| 641 | << "," << fParseBufferParseEnd << "," << fParseBufferDataEnd << "]\n" ; |
| 642 | #endif |
| 643 | } |
| 644 | |
| 645 | void MPEG2IFrameIndexFromTransportStream::addToTail(IndexRecord* newIndexRecord) { |
| 646 | #ifdef DEBUG |
| 647 | envir() << "adding new: " << *newIndexRecord << "\n" ; |
| 648 | #endif |
| 649 | if (fTailIndexRecord == NULL) { |
| 650 | fHeadIndexRecord = fTailIndexRecord = newIndexRecord; |
| 651 | } else { |
| 652 | newIndexRecord->addAfter(fTailIndexRecord); |
| 653 | fTailIndexRecord = newIndexRecord; |
| 654 | } |
| 655 | } |
| 656 | |
| 657 | ////////// IndexRecord implementation ////////// |
| 658 | |
| 659 | IndexRecord::IndexRecord(u_int8_t startOffset, u_int8_t size, |
| 660 | unsigned long transportPacketNumber, float pcr) |
| 661 | : fNext(this), fPrev(this), fRecordType(RECORD_UNPARSED), |
| 662 | fStartOffset(startOffset), fSize(size), |
| 663 | fPCR(pcr), fTransportPacketNumber(transportPacketNumber) { |
| 664 | } |
| 665 | |
| 666 | IndexRecord::~IndexRecord() { |
| 667 | IndexRecord* nextRecord = next(); |
| 668 | unlink(); |
| 669 | if (nextRecord != this) delete nextRecord; |
| 670 | } |
| 671 | |
| 672 | void IndexRecord::addAfter(IndexRecord* prev) { |
| 673 | fNext = prev->fNext; |
| 674 | fPrev = prev; |
| 675 | prev->fNext->fPrev = this; |
| 676 | prev->fNext = this; |
| 677 | } |
| 678 | |
| 679 | void IndexRecord::unlink() { |
| 680 | fNext->fPrev = fPrev; |
| 681 | fPrev->fNext = fNext; |
| 682 | fNext = fPrev = this; |
| 683 | } |
| 684 | |