| 1 | /* |
| 2 | * seek utility functions for use within format handlers |
| 3 | * |
| 4 | * Copyright (c) 2009 Ivan Schreter |
| 5 | * |
| 6 | * This file is part of FFmpeg. |
| 7 | * |
| 8 | * FFmpeg is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU Lesser General Public |
| 10 | * License as published by the Free Software Foundation; either |
| 11 | * version 2.1 of the License, or (at your option) any later version. |
| 12 | * |
| 13 | * FFmpeg is distributed in the hope that it will be useful, |
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 16 | * Lesser General Public License for more details. |
| 17 | * |
| 18 | * You should have received a copy of the GNU Lesser General Public |
| 19 | * License along with FFmpeg; if not, write to the Free Software |
| 20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 21 | */ |
| 22 | |
| 23 | #include <stdint.h> |
| 24 | |
| 25 | #include "seek.h" |
| 26 | #include "libavutil/mathematics.h" |
| 27 | #include "libavutil/mem.h" |
| 28 | #include "internal.h" |
| 29 | |
| 30 | // NOTE: implementation should be moved here in another patch, to keep patches |
| 31 | // separated. |
| 32 | |
| 33 | /** |
| 34 | * helper structure describing keyframe search state of one stream |
| 35 | */ |
| 36 | typedef struct { |
| 37 | int64_t pos_lo; ///< position of the frame with low timestamp in file or INT64_MAX if not found (yet) |
| 38 | int64_t ts_lo; ///< frame presentation timestamp or same as pos_lo for byte seeking |
| 39 | |
| 40 | int64_t pos_hi; ///< position of the frame with high timestamp in file or INT64_MAX if not found (yet) |
| 41 | int64_t ts_hi; ///< frame presentation timestamp or same as pos_hi for byte seeking |
| 42 | |
| 43 | int64_t last_pos; ///< last known position of a frame, for multi-frame packets |
| 44 | |
| 45 | int64_t term_ts; ///< termination timestamp (which TS we already read) |
| 46 | AVRational term_ts_tb; ///< timebase for term_ts |
| 47 | int64_t first_ts; ///< first packet timestamp in this iteration (to fill term_ts later) |
| 48 | AVRational first_ts_tb; ///< timebase for first_ts |
| 49 | |
| 50 | int terminated; ///< termination flag for the current iteration |
| 51 | } AVSyncPoint; |
| 52 | |
| 53 | /** |
| 54 | * Compute a distance between timestamps. |
| 55 | * |
| 56 | * Distances are only comparable, if same time bases are used for computing |
| 57 | * distances. |
| 58 | * |
| 59 | * @param ts_hi high timestamp |
| 60 | * @param tb_hi high timestamp time base |
| 61 | * @param ts_lo low timestamp |
| 62 | * @param tb_lo low timestamp time base |
| 63 | * @return representation of distance between high and low timestamps |
| 64 | */ |
| 65 | static int64_t ts_distance(int64_t ts_hi, |
| 66 | AVRational tb_hi, |
| 67 | int64_t ts_lo, |
| 68 | AVRational tb_lo) |
| 69 | { |
| 70 | int64_t hi, lo; |
| 71 | |
| 72 | hi = ts_hi * tb_hi.num * tb_lo.den; |
| 73 | lo = ts_lo * tb_lo.num * tb_hi.den; |
| 74 | |
| 75 | return hi - lo; |
| 76 | } |
| 77 | |
| 78 | /** |
| 79 | * Partial search for keyframes in multiple streams. |
| 80 | * |
| 81 | * This routine searches in each stream for the next lower and the next higher |
| 82 | * timestamp compared to the given target timestamp. The search starts at the current |
| 83 | * file position and ends at the file position, where all streams have already been |
| 84 | * examined (or when all higher key frames are found in the first iteration). |
| 85 | * |
| 86 | * This routine is called iteratively with an exponential backoff to find the lower |
| 87 | * timestamp. |
| 88 | * |
| 89 | * @param s format context |
| 90 | * @param timestamp target timestamp (or position, if AVSEEK_FLAG_BYTE) |
| 91 | * @param timebase time base for timestamps |
| 92 | * @param flags seeking flags |
| 93 | * @param sync array with information per stream |
| 94 | * @param keyframes_to_find count of keyframes to find in total |
| 95 | * @param found_lo ptr to the count of already found low timestamp keyframes |
| 96 | * @param found_hi ptr to the count of already found high timestamp keyframes |
| 97 | * @param first_iter flag for first iteration |
| 98 | */ |
| 99 | static void search_hi_lo_keyframes(AVFormatContext *s, |
| 100 | int64_t timestamp, |
| 101 | AVRational timebase, |
| 102 | int flags, |
| 103 | AVSyncPoint *sync, |
| 104 | int keyframes_to_find, |
| 105 | int *found_lo, |
| 106 | int *found_hi, |
| 107 | int first_iter) |
| 108 | { |
| 109 | AVPacket pkt; |
| 110 | AVSyncPoint *sp; |
| 111 | AVStream *st; |
| 112 | int idx; |
| 113 | int flg; |
| 114 | int terminated_count = 0; |
| 115 | int64_t pos; |
| 116 | int64_t pts, dts; // PTS/DTS from stream |
| 117 | int64_t ts; // PTS in stream-local time base or position for byte seeking |
| 118 | AVRational ts_tb; // Time base of the stream or 1:1 for byte seeking |
| 119 | |
| 120 | for (;;) { |
| 121 | if (av_read_frame(s, &pkt) < 0) { |
| 122 | // EOF or error, make sure high flags are set |
| 123 | for (idx = 0; idx < s->nb_streams; ++idx) { |
| 124 | if (s->streams[idx]->discard < AVDISCARD_ALL) { |
| 125 | sp = &sync[idx]; |
| 126 | if (sp->pos_hi == INT64_MAX) { |
| 127 | // no high frame exists for this stream |
| 128 | (*found_hi)++; |
| 129 | sp->ts_hi = INT64_MAX; |
| 130 | sp->pos_hi = INT64_MAX - 1; |
| 131 | } |
| 132 | } |
| 133 | } |
| 134 | break; |
| 135 | } |
| 136 | |
| 137 | idx = pkt.stream_index; |
| 138 | st = s->streams[idx]; |
| 139 | if (st->discard >= AVDISCARD_ALL) |
| 140 | // this stream is not active, skip packet |
| 141 | continue; |
| 142 | |
| 143 | sp = &sync[idx]; |
| 144 | |
| 145 | flg = pkt.flags; |
| 146 | pos = pkt.pos; |
| 147 | pts = pkt.pts; |
| 148 | dts = pkt.dts; |
| 149 | if (pts == AV_NOPTS_VALUE) |
| 150 | // some formats don't provide PTS, only DTS |
| 151 | pts = dts; |
| 152 | |
| 153 | av_free_packet(&pkt); |
| 154 | |
| 155 | // Multi-frame packets only return position for the very first frame. |
| 156 | // Other frames are read with position == -1. Therefore, we note down |
| 157 | // last known position of a frame and use it if a frame without |
| 158 | // position arrives. In this way, it's possible to seek to proper |
| 159 | // position. Additionally, for parsers not providing position at all, |
| 160 | // an approximation will be used (starting position of this iteration). |
| 161 | if (pos < 0) |
| 162 | pos = sp->last_pos; |
| 163 | else |
| 164 | sp->last_pos = pos; |
| 165 | |
| 166 | // Evaluate key frames with known TS (or any frames, if AVSEEK_FLAG_ANY set). |
| 167 | if (pts != AV_NOPTS_VALUE && |
| 168 | ((flg & AV_PKT_FLAG_KEY) || (flags & AVSEEK_FLAG_ANY))) { |
| 169 | if (flags & AVSEEK_FLAG_BYTE) { |
| 170 | // for byte seeking, use position as timestamp |
| 171 | ts = pos; |
| 172 | ts_tb.num = 1; |
| 173 | ts_tb.den = 1; |
| 174 | } else { |
| 175 | // otherwise, get stream time_base |
| 176 | ts = pts; |
| 177 | ts_tb = st->time_base; |
| 178 | } |
| 179 | |
| 180 | if (sp->first_ts == AV_NOPTS_VALUE) { |
| 181 | // Note down termination timestamp for the next iteration - when |
| 182 | // we encounter a packet with the same timestamp, we will ignore |
| 183 | // any further packets for this stream in next iteration (as they |
| 184 | // are already evaluated). |
| 185 | sp->first_ts = ts; |
| 186 | sp->first_ts_tb = ts_tb; |
| 187 | } |
| 188 | |
| 189 | if (sp->term_ts != AV_NOPTS_VALUE && |
| 190 | av_compare_ts(ts, ts_tb, sp->term_ts, sp->term_ts_tb) > 0) { |
| 191 | // past the end position from last iteration, ignore packet |
| 192 | if (!sp->terminated) { |
| 193 | sp->terminated = 1; |
| 194 | ++terminated_count; |
| 195 | if (sp->pos_hi == INT64_MAX) { |
| 196 | // no high frame exists for this stream |
| 197 | (*found_hi)++; |
| 198 | sp->ts_hi = INT64_MAX; |
| 199 | sp->pos_hi = INT64_MAX - 1; |
| 200 | } |
| 201 | if (terminated_count == keyframes_to_find) |
| 202 | break; // all terminated, iteration done |
| 203 | } |
| 204 | continue; |
| 205 | } |
| 206 | |
| 207 | if (av_compare_ts(ts, ts_tb, timestamp, timebase) <= 0) { |
| 208 | // keyframe found before target timestamp |
| 209 | if (sp->pos_lo == INT64_MAX) { |
| 210 | // found first keyframe lower than target timestamp |
| 211 | (*found_lo)++; |
| 212 | sp->ts_lo = ts; |
| 213 | sp->pos_lo = pos; |
| 214 | } else if (sp->ts_lo < ts) { |
| 215 | // found a better match (closer to target timestamp) |
| 216 | sp->ts_lo = ts; |
| 217 | sp->pos_lo = pos; |
| 218 | } |
| 219 | } |
| 220 | if (av_compare_ts(ts, ts_tb, timestamp, timebase) >= 0) { |
| 221 | // keyframe found after target timestamp |
| 222 | if (sp->pos_hi == INT64_MAX) { |
| 223 | // found first keyframe higher than target timestamp |
| 224 | (*found_hi)++; |
| 225 | sp->ts_hi = ts; |
| 226 | sp->pos_hi = pos; |
| 227 | if (*found_hi >= keyframes_to_find && first_iter) { |
| 228 | // We found high frame for all. They may get updated |
| 229 | // to TS closer to target TS in later iterations (which |
| 230 | // will stop at start position of previous iteration). |
| 231 | break; |
| 232 | } |
| 233 | } else if (sp->ts_hi > ts) { |
| 234 | // found a better match (actually, shouldn't happen) |
| 235 | sp->ts_hi = ts; |
| 236 | sp->pos_hi = pos; |
| 237 | } |
| 238 | } |
| 239 | } |
| 240 | } |
| 241 | |
| 242 | // Clean up the parser. |
| 243 | ff_read_frame_flush(s); |
| 244 | } |
| 245 | |
| 246 | int64_t ff_gen_syncpoint_search(AVFormatContext *s, |
| 247 | int stream_index, |
| 248 | int64_t pos, |
| 249 | int64_t ts_min, |
| 250 | int64_t ts, |
| 251 | int64_t ts_max, |
| 252 | int flags) |
| 253 | { |
| 254 | AVSyncPoint *sync, *sp; |
| 255 | AVStream *st; |
| 256 | int i; |
| 257 | int keyframes_to_find = 0; |
| 258 | int64_t curpos; |
| 259 | int64_t step; |
| 260 | int found_lo = 0, found_hi = 0; |
| 261 | int64_t min_distance, distance; |
| 262 | int64_t min_pos = 0; |
| 263 | int first_iter = 1; |
| 264 | AVRational time_base; |
| 265 | |
| 266 | if (flags & AVSEEK_FLAG_BYTE) { |
| 267 | // for byte seeking, we have exact 1:1 "timestamps" - positions |
| 268 | time_base.num = 1; |
| 269 | time_base.den = 1; |
| 270 | } else { |
| 271 | if (stream_index >= 0) { |
| 272 | // we have a reference stream, which time base we use |
| 273 | st = s->streams[stream_index]; |
| 274 | time_base = st->time_base; |
| 275 | } else { |
| 276 | // no reference stream, use AV_TIME_BASE as reference time base |
| 277 | time_base.num = 1; |
| 278 | time_base.den = AV_TIME_BASE; |
| 279 | } |
| 280 | } |
| 281 | |
| 282 | // Initialize syncpoint structures for each stream. |
| 283 | sync = av_malloc_array(s->nb_streams, sizeof(AVSyncPoint)); |
| 284 | if (!sync) |
| 285 | // cannot allocate helper structure |
| 286 | return -1; |
| 287 | |
| 288 | for (i = 0; i < s->nb_streams; ++i) { |
| 289 | st = s->streams[i]; |
| 290 | sp = &sync[i]; |
| 291 | |
| 292 | sp->pos_lo = INT64_MAX; |
| 293 | sp->ts_lo = INT64_MAX; |
| 294 | sp->pos_hi = INT64_MAX; |
| 295 | sp->ts_hi = INT64_MAX; |
| 296 | sp->terminated = 0; |
| 297 | sp->first_ts = AV_NOPTS_VALUE; |
| 298 | sp->term_ts = ts_max; |
| 299 | sp->term_ts_tb = time_base; |
| 300 | sp->last_pos = pos; |
| 301 | |
| 302 | st->cur_dts = AV_NOPTS_VALUE; |
| 303 | |
| 304 | if (st->discard < AVDISCARD_ALL) |
| 305 | ++keyframes_to_find; |
| 306 | } |
| 307 | |
| 308 | if (!keyframes_to_find) { |
| 309 | // no stream active, error |
| 310 | av_free(sync); |
| 311 | return -1; |
| 312 | } |
| 313 | |
| 314 | // Find keyframes in all active streams with timestamp/position just before |
| 315 | // and just after requested timestamp/position. |
| 316 | step = s->pb->buffer_size; |
| 317 | curpos = FFMAX(pos - step / 2, 0); |
| 318 | for (;;) { |
| 319 | avio_seek(s->pb, curpos, SEEK_SET); |
| 320 | search_hi_lo_keyframes(s, |
| 321 | ts, time_base, |
| 322 | flags, |
| 323 | sync, |
| 324 | keyframes_to_find, |
| 325 | &found_lo, &found_hi, |
| 326 | first_iter); |
| 327 | if (found_lo == keyframes_to_find && found_hi == keyframes_to_find) |
| 328 | break; // have all keyframes we wanted |
| 329 | if (!curpos) |
| 330 | break; // cannot go back anymore |
| 331 | |
| 332 | curpos = pos - step; |
| 333 | if (curpos < 0) |
| 334 | curpos = 0; |
| 335 | step *= 2; |
| 336 | |
| 337 | // switch termination positions |
| 338 | for (i = 0; i < s->nb_streams; ++i) { |
| 339 | st = s->streams[i]; |
| 340 | st->cur_dts = AV_NOPTS_VALUE; |
| 341 | |
| 342 | sp = &sync[i]; |
| 343 | if (sp->first_ts != AV_NOPTS_VALUE) { |
| 344 | sp->term_ts = sp->first_ts; |
| 345 | sp->term_ts_tb = sp->first_ts_tb; |
| 346 | sp->first_ts = AV_NOPTS_VALUE; |
| 347 | } |
| 348 | sp->terminated = 0; |
| 349 | sp->last_pos = curpos; |
| 350 | } |
| 351 | first_iter = 0; |
| 352 | } |
| 353 | |
| 354 | // Find actual position to start decoding so that decoder synchronizes |
| 355 | // closest to ts and between ts_min and ts_max. |
| 356 | pos = INT64_MAX; |
| 357 | |
| 358 | for (i = 0; i < s->nb_streams; ++i) { |
| 359 | st = s->streams[i]; |
| 360 | if (st->discard < AVDISCARD_ALL) { |
| 361 | sp = &sync[i]; |
| 362 | min_distance = INT64_MAX; |
| 363 | // Find timestamp closest to requested timestamp within min/max limits. |
| 364 | if (sp->pos_lo != INT64_MAX |
| 365 | && av_compare_ts(ts_min, time_base, sp->ts_lo, st->time_base) <= 0 |
| 366 | && av_compare_ts(sp->ts_lo, st->time_base, ts_max, time_base) <= 0) { |
| 367 | // low timestamp is in range |
| 368 | min_distance = ts_distance(ts, time_base, sp->ts_lo, st->time_base); |
| 369 | min_pos = sp->pos_lo; |
| 370 | } |
| 371 | if (sp->pos_hi != INT64_MAX |
| 372 | && av_compare_ts(ts_min, time_base, sp->ts_hi, st->time_base) <= 0 |
| 373 | && av_compare_ts(sp->ts_hi, st->time_base, ts_max, time_base) <= 0) { |
| 374 | // high timestamp is in range, check distance |
| 375 | distance = ts_distance(sp->ts_hi, st->time_base, ts, time_base); |
| 376 | if (distance < min_distance) { |
| 377 | min_distance = distance; |
| 378 | min_pos = sp->pos_hi; |
| 379 | } |
| 380 | } |
| 381 | if (min_distance == INT64_MAX) { |
| 382 | // no timestamp is in range, cannot seek |
| 383 | av_free(sync); |
| 384 | return -1; |
| 385 | } |
| 386 | if (min_pos < pos) |
| 387 | pos = min_pos; |
| 388 | } |
| 389 | } |
| 390 | |
| 391 | avio_seek(s->pb, pos, SEEK_SET); |
| 392 | av_free(sync); |
| 393 | return pos; |
| 394 | } |
| 395 | |
| 396 | AVParserState *ff_store_parser_state(AVFormatContext *s) |
| 397 | { |
| 398 | int i; |
| 399 | AVStream *st; |
| 400 | AVParserStreamState *ss; |
| 401 | AVParserState *state = av_malloc(sizeof(AVParserState)); |
| 402 | if (!state) |
| 403 | return NULL; |
| 404 | |
| 405 | state->stream_states = av_malloc_array(s->nb_streams, sizeof(AVParserStreamState)); |
| 406 | if (!state->stream_states) { |
| 407 | av_free(state); |
| 408 | return NULL; |
| 409 | } |
| 410 | |
| 411 | state->fpos = avio_tell(s->pb); |
| 412 | |
| 413 | // copy context structures |
| 414 | state->packet_buffer = s->packet_buffer; |
| 415 | state->parse_queue = s->parse_queue; |
| 416 | state->raw_packet_buffer = s->raw_packet_buffer; |
| 417 | state->raw_packet_buffer_remaining_size = s->raw_packet_buffer_remaining_size; |
| 418 | |
| 419 | s->packet_buffer = NULL; |
| 420 | s->parse_queue = NULL; |
| 421 | s->raw_packet_buffer = NULL; |
| 422 | s->raw_packet_buffer_remaining_size = RAW_PACKET_BUFFER_SIZE; |
| 423 | |
| 424 | // copy stream structures |
| 425 | state->nb_streams = s->nb_streams; |
| 426 | for (i = 0; i < s->nb_streams; i++) { |
| 427 | st = s->streams[i]; |
| 428 | ss = &state->stream_states[i]; |
| 429 | |
| 430 | ss->parser = st->parser; |
| 431 | ss->last_IP_pts = st->last_IP_pts; |
| 432 | ss->cur_dts = st->cur_dts; |
| 433 | ss->probe_packets = st->probe_packets; |
| 434 | |
| 435 | st->parser = NULL; |
| 436 | st->last_IP_pts = AV_NOPTS_VALUE; |
| 437 | st->cur_dts = AV_NOPTS_VALUE; |
| 438 | st->probe_packets = MAX_PROBE_PACKETS; |
| 439 | } |
| 440 | |
| 441 | return state; |
| 442 | } |
| 443 | |
| 444 | void ff_restore_parser_state(AVFormatContext *s, AVParserState *state) |
| 445 | { |
| 446 | int i; |
| 447 | AVStream *st; |
| 448 | AVParserStreamState *ss; |
| 449 | ff_read_frame_flush(s); |
| 450 | |
| 451 | if (!state) |
| 452 | return; |
| 453 | |
| 454 | avio_seek(s->pb, state->fpos, SEEK_SET); |
| 455 | |
| 456 | // copy context structures |
| 457 | s->packet_buffer = state->packet_buffer; |
| 458 | s->parse_queue = state->parse_queue; |
| 459 | s->raw_packet_buffer = state->raw_packet_buffer; |
| 460 | s->raw_packet_buffer_remaining_size = state->raw_packet_buffer_remaining_size; |
| 461 | |
| 462 | // copy stream structures |
| 463 | for (i = 0; i < state->nb_streams; i++) { |
| 464 | st = s->streams[i]; |
| 465 | ss = &state->stream_states[i]; |
| 466 | |
| 467 | st->parser = ss->parser; |
| 468 | st->last_IP_pts = ss->last_IP_pts; |
| 469 | st->cur_dts = ss->cur_dts; |
| 470 | st->probe_packets = ss->probe_packets; |
| 471 | } |
| 472 | |
| 473 | av_free(state->stream_states); |
| 474 | av_free(state); |
| 475 | } |
| 476 | |
| 477 | static void free_packet_list(AVPacketList *pktl) |
| 478 | { |
| 479 | AVPacketList *cur; |
| 480 | while (pktl) { |
| 481 | cur = pktl; |
| 482 | pktl = cur->next; |
| 483 | av_free_packet(&cur->pkt); |
| 484 | av_free(cur); |
| 485 | } |
| 486 | } |
| 487 | |
| 488 | void ff_free_parser_state(AVFormatContext *s, AVParserState *state) |
| 489 | { |
| 490 | int i; |
| 491 | AVParserStreamState *ss; |
| 492 | |
| 493 | if (!state) |
| 494 | return; |
| 495 | |
| 496 | for (i = 0; i < state->nb_streams; i++) { |
| 497 | ss = &state->stream_states[i]; |
| 498 | if (ss->parser) |
| 499 | av_parser_close(ss->parser); |
| 500 | } |
| 501 | |
| 502 | free_packet_list(state->packet_buffer); |
| 503 | free_packet_list(state->parse_queue); |
| 504 | free_packet_list(state->raw_packet_buffer); |
| 505 | |
| 506 | av_free(state->stream_states); |
| 507 | av_free(state); |
| 508 | } |