You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

767 lines
28KB

  1. /*
  2. * DV encoder
  3. * Copyright (c) 2003 Roman Shaposhnik
  4. *
  5. * This file is part of FFmpeg.
  6. *
  7. * FFmpeg is free software; you can redistribute it and/or
  8. * modify it under the terms of the GNU Lesser General Public
  9. * License as published by the Free Software Foundation; either
  10. * version 2.1 of the License, or (at your option) any later version.
  11. *
  12. * FFmpeg is distributed in the hope that it will be useful,
  13. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  15. * Lesser General Public License for more details.
  16. *
  17. * You should have received a copy of the GNU Lesser General Public
  18. * License along with FFmpeg; if not, write to the Free Software
  19. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20. */
  21. /**
  22. * @file
  23. * DV encoder
  24. */
  25. #include "config.h"
  26. #include "libavutil/attributes.h"
  27. #include "libavutil/pixdesc.h"
  28. #include "avcodec.h"
  29. #include "dv.h"
  30. #include "dv_profile_internal.h"
  31. #include "dv_tablegen.h"
  32. #include "fdctdsp.h"
  33. #include "internal.h"
  34. #include "me_cmp.h"
  35. #include "pixblockdsp.h"
  36. #include "put_bits.h"
  37. static av_cold int dvvideo_encode_init(AVCodecContext *avctx)
  38. {
  39. DVVideoContext *s = avctx->priv_data;
  40. FDCTDSPContext fdsp;
  41. MECmpContext mecc;
  42. PixblockDSPContext pdsp;
  43. int ret;
  44. s->sys = av_dv_codec_profile2(avctx->width, avctx->height, avctx->pix_fmt, avctx->time_base);
  45. if (!s->sys) {
  46. av_log(avctx, AV_LOG_ERROR, "Found no DV profile for %ix%i %s video. "
  47. "Valid DV profiles are:\n",
  48. avctx->width, avctx->height, av_get_pix_fmt_name(avctx->pix_fmt));
  49. ff_dv_print_profiles(avctx, AV_LOG_ERROR);
  50. return AVERROR(EINVAL);
  51. }
  52. if (avctx->height > 576) {
  53. av_log(avctx, AV_LOG_ERROR, "DVCPRO HD encoding is not supported.\n");
  54. return AVERROR_PATCHWELCOME;
  55. }
  56. ret = ff_dv_init_dynamic_tables(s, s->sys);
  57. if (ret < 0) {
  58. av_log(avctx, AV_LOG_ERROR, "Error initializing work tables.\n");
  59. return ret;
  60. }
  61. avctx->coded_frame = av_frame_alloc();
  62. if (!avctx->coded_frame)
  63. return AVERROR(ENOMEM);
  64. dv_vlc_map_tableinit();
  65. memset(&fdsp,0, sizeof(fdsp));
  66. memset(&mecc,0, sizeof(mecc));
  67. memset(&pdsp,0, sizeof(pdsp));
  68. ff_fdctdsp_init(&fdsp, avctx);
  69. ff_me_cmp_init(&mecc, avctx);
  70. ff_pixblockdsp_init(&pdsp, avctx);
  71. ff_set_cmp(&mecc, mecc.ildct_cmp, avctx->ildct_cmp);
  72. s->get_pixels = pdsp.get_pixels;
  73. s->ildct_cmp = mecc.ildct_cmp[5];
  74. s->fdct[0] = fdsp.fdct;
  75. s->fdct[1] = fdsp.fdct248;
  76. return ff_dvvideo_init(avctx);
  77. }
  78. /* bit budget for AC only in 5 MBs */
  79. static const int vs_total_ac_bits = (100 * 4 + 68 * 2) * 5;
  80. static const int mb_area_start[5] = { 1, 6, 21, 43, 64 };
  81. #if CONFIG_SMALL
  82. /* Convert run and level (where level != 0) pair into VLC, returning bit size */
  83. static av_always_inline int dv_rl2vlc(int run, int level, int sign,
  84. uint32_t *vlc)
  85. {
  86. int size;
  87. if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
  88. *vlc = dv_vlc_map[run][level].vlc | sign;
  89. size = dv_vlc_map[run][level].size;
  90. } else {
  91. if (level < DV_VLC_MAP_LEV_SIZE) {
  92. *vlc = dv_vlc_map[0][level].vlc | sign;
  93. size = dv_vlc_map[0][level].size;
  94. } else {
  95. *vlc = 0xfe00 | (level << 1) | sign;
  96. size = 16;
  97. }
  98. if (run) {
  99. *vlc |= ((run < 16) ? dv_vlc_map[run - 1][0].vlc :
  100. (0x1f80 | (run - 1))) << size;
  101. size += (run < 16) ? dv_vlc_map[run - 1][0].size : 13;
  102. }
  103. }
  104. return size;
  105. }
  106. static av_always_inline int dv_rl2vlc_size(int run, int level)
  107. {
  108. int size;
  109. if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {
  110. size = dv_vlc_map[run][level].size;
  111. } else {
  112. size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16;
  113. if (run)
  114. size += (run < 16) ? dv_vlc_map[run - 1][0].size : 13;
  115. }
  116. return size;
  117. }
  118. #else
  119. static av_always_inline int dv_rl2vlc(int run, int l, int sign, uint32_t *vlc)
  120. {
  121. *vlc = dv_vlc_map[run][l].vlc | sign;
  122. return dv_vlc_map[run][l].size;
  123. }
  124. static av_always_inline int dv_rl2vlc_size(int run, int l)
  125. {
  126. return dv_vlc_map[run][l].size;
  127. }
  128. #endif
  129. typedef struct EncBlockInfo {
  130. int area_q[4];
  131. int bit_size[4];
  132. int prev[5];
  133. int cur_ac;
  134. int cno;
  135. int dct_mode;
  136. int16_t mb[64];
  137. uint8_t next[64];
  138. uint8_t sign[64];
  139. uint8_t partial_bit_count;
  140. uint32_t partial_bit_buffer; /* we can't use uint16_t here */
  141. } EncBlockInfo;
  142. static av_always_inline PutBitContext *dv_encode_ac(EncBlockInfo *bi,
  143. PutBitContext *pb_pool,
  144. PutBitContext *pb_end)
  145. {
  146. int prev, bits_left;
  147. PutBitContext *pb = pb_pool;
  148. int size = bi->partial_bit_count;
  149. uint32_t vlc = bi->partial_bit_buffer;
  150. bi->partial_bit_count =
  151. bi->partial_bit_buffer = 0;
  152. for (;;) {
  153. /* Find suitable storage space */
  154. for (; size > (bits_left = put_bits_left(pb)); pb++) {
  155. if (bits_left) {
  156. size -= bits_left;
  157. put_bits(pb, bits_left, vlc >> size);
  158. vlc = vlc & ((1 << size) - 1);
  159. }
  160. if (pb + 1 >= pb_end) {
  161. bi->partial_bit_count = size;
  162. bi->partial_bit_buffer = vlc;
  163. return pb;
  164. }
  165. }
  166. /* Store VLC */
  167. put_bits(pb, size, vlc);
  168. if (bi->cur_ac >= 64)
  169. break;
  170. /* Construct the next VLC */
  171. prev = bi->cur_ac;
  172. bi->cur_ac = bi->next[prev];
  173. if (bi->cur_ac < 64) {
  174. size = dv_rl2vlc(bi->cur_ac - prev - 1, bi->mb[bi->cur_ac],
  175. bi->sign[bi->cur_ac], &vlc);
  176. } else {
  177. size = 4;
  178. vlc = 6; /* End Of Block stamp */
  179. }
  180. }
  181. return pb;
  182. }
  183. static av_always_inline int dv_guess_dct_mode(DVVideoContext *s, uint8_t *data,
  184. int linesize)
  185. {
  186. if (s->avctx->flags & CODEC_FLAG_INTERLACED_DCT) {
  187. int ps = s->ildct_cmp(NULL, data, NULL, linesize, 8) - 400;
  188. if (ps > 0) {
  189. int is = s->ildct_cmp(NULL, data, NULL, linesize << 1, 4) +
  190. s->ildct_cmp(NULL, data + linesize, NULL, linesize << 1, 4);
  191. return ps > is;
  192. }
  193. }
  194. return 0;
  195. }
  196. static const int dv_weight_bits = 18;
  197. static const int dv_weight_88[64] = {
  198. 131072, 257107, 257107, 242189, 252167, 242189, 235923, 237536,
  199. 237536, 235923, 229376, 231390, 223754, 231390, 229376, 222935,
  200. 224969, 217965, 217965, 224969, 222935, 200636, 218652, 211916,
  201. 212325, 211916, 218652, 200636, 188995, 196781, 205965, 206433,
  202. 206433, 205965, 196781, 188995, 185364, 185364, 200636, 200704,
  203. 200636, 185364, 185364, 174609, 180568, 195068, 195068, 180568,
  204. 174609, 170091, 175557, 189591, 175557, 170091, 165371, 170627,
  205. 170627, 165371, 160727, 153560, 160727, 144651, 144651, 136258,
  206. };
  207. static const int dv_weight_248[64] = {
  208. 131072, 262144, 257107, 257107, 242189, 242189, 242189, 242189,
  209. 237536, 237536, 229376, 229376, 200636, 200636, 224973, 224973,
  210. 223754, 223754, 235923, 235923, 229376, 229376, 217965, 217965,
  211. 211916, 211916, 196781, 196781, 185364, 185364, 206433, 206433,
  212. 211916, 211916, 222935, 222935, 200636, 200636, 205964, 205964,
  213. 200704, 200704, 180568, 180568, 175557, 175557, 195068, 195068,
  214. 185364, 185364, 188995, 188995, 174606, 174606, 175557, 175557,
  215. 170627, 170627, 153560, 153560, 165371, 165371, 144651, 144651,
  216. };
  217. static av_always_inline int dv_init_enc_block(EncBlockInfo *bi, uint8_t *data,
  218. int linesize, DVVideoContext *s,
  219. int bias)
  220. {
  221. const int *weight;
  222. const uint8_t *zigzag_scan;
  223. LOCAL_ALIGNED_16(int16_t, blk, [64]);
  224. int i, area;
  225. /* We offer two different methods for class number assignment: the
  226. * method suggested in SMPTE 314M Table 22, and an improved
  227. * method. The SMPTE method is very conservative; it assigns class
  228. * 3 (i.e. severe quantization) to any block where the largest AC
  229. * component is greater than 36. FFmpeg's DV encoder tracks AC bit
  230. * consumption precisely, so there is no need to bias most blocks
  231. * towards strongly lossy compression. Instead, we assign class 2
  232. * to most blocks, and use class 3 only when strictly necessary
  233. * (for blocks whose largest AC component exceeds 255). */
  234. #if 0 /* SMPTE spec method */
  235. static const int classes[] = { 12, 24, 36, 0xffff };
  236. #else /* improved FFmpeg method */
  237. static const int classes[] = { -1, -1, 255, 0xffff };
  238. #endif
  239. int max = classes[0];
  240. int prev = 0;
  241. av_assert2((((int) blk) & 15) == 0);
  242. bi->area_q[0] =
  243. bi->area_q[1] =
  244. bi->area_q[2] =
  245. bi->area_q[3] = 0;
  246. bi->partial_bit_count = 0;
  247. bi->partial_bit_buffer = 0;
  248. bi->cur_ac = 0;
  249. if (data) {
  250. bi->dct_mode = dv_guess_dct_mode(s, data, linesize);
  251. s->get_pixels(blk, data, linesize);
  252. s->fdct[bi->dct_mode](blk);
  253. } else {
  254. /* We rely on the fact that encoding all zeros leads to an immediate
  255. * EOB, which is precisely what the spec calls for in the "dummy"
  256. * blocks. */
  257. memset(blk, 0, 64 * sizeof(*blk));
  258. bi->dct_mode = 0;
  259. }
  260. bi->mb[0] = blk[0];
  261. zigzag_scan = bi->dct_mode ? ff_dv_zigzag248_direct : ff_zigzag_direct;
  262. weight = bi->dct_mode ? dv_weight_248 : dv_weight_88;
  263. for (area = 0; area < 4; area++) {
  264. bi->prev[area] = prev;
  265. bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :)
  266. for (i = mb_area_start[area]; i < mb_area_start[area + 1]; i++) {
  267. int level = blk[zigzag_scan[i]];
  268. if (level + 15 > 30U) {
  269. bi->sign[i] = (level >> 31) & 1;
  270. /* Weight it and shift down into range, adding for rounding.
  271. * The extra division by a factor of 2^4 reverses the 8x
  272. * expansion of the DCT AND the 2x doubling of the weights. */
  273. level = (FFABS(level) * weight[i] + (1 << (dv_weight_bits + 3))) >>
  274. (dv_weight_bits + 4);
  275. bi->mb[i] = level;
  276. if (level > max)
  277. max = level;
  278. bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, level);
  279. bi->next[prev] = i;
  280. prev = i;
  281. }
  282. }
  283. }
  284. bi->next[prev] = i;
  285. for (bi->cno = 0; max > classes[bi->cno]; bi->cno++)
  286. ;
  287. bi->cno += bias;
  288. if (bi->cno >= 3) {
  289. bi->cno = 3;
  290. prev = 0;
  291. i = bi->next[prev];
  292. for (area = 0; area < 4; area++) {
  293. bi->prev[area] = prev;
  294. bi->bit_size[area] = 1; // 4 areas 4 bits for EOB :)
  295. for (; i < mb_area_start[area + 1]; i = bi->next[i]) {
  296. bi->mb[i] >>= 1;
  297. if (bi->mb[i]) {
  298. bi->bit_size[area] += dv_rl2vlc_size(i - prev - 1, bi->mb[i]);
  299. bi->next[prev] = i;
  300. prev = i;
  301. }
  302. }
  303. }
  304. bi->next[prev] = i;
  305. }
  306. return bi->bit_size[0] + bi->bit_size[1] +
  307. bi->bit_size[2] + bi->bit_size[3];
  308. }
  309. static inline void dv_guess_qnos(EncBlockInfo *blks, int *qnos)
  310. {
  311. int size[5];
  312. int i, j, k, a, prev, a2;
  313. EncBlockInfo *b;
  314. size[0] =
  315. size[1] =
  316. size[2] =
  317. size[3] =
  318. size[4] = 1 << 24;
  319. do {
  320. b = blks;
  321. for (i = 0; i < 5; i++) {
  322. if (!qnos[i])
  323. continue;
  324. qnos[i]--;
  325. size[i] = 0;
  326. for (j = 0; j < 6; j++, b++) {
  327. for (a = 0; a < 4; a++) {
  328. if (b->area_q[a] != ff_dv_quant_shifts[qnos[i] + ff_dv_quant_offset[b->cno]][a]) {
  329. b->bit_size[a] = 1; // 4 areas 4 bits for EOB :)
  330. b->area_q[a]++;
  331. prev = b->prev[a];
  332. av_assert2(b->next[prev] >= mb_area_start[a + 1] || b->mb[prev]);
  333. for (k = b->next[prev]; k < mb_area_start[a + 1]; k = b->next[k]) {
  334. b->mb[k] >>= 1;
  335. if (b->mb[k]) {
  336. b->bit_size[a] += dv_rl2vlc_size(k - prev - 1, b->mb[k]);
  337. prev = k;
  338. } else {
  339. if (b->next[k] >= mb_area_start[a + 1] && b->next[k] < 64) {
  340. for (a2 = a + 1; b->next[k] >= mb_area_start[a2 + 1]; a2++)
  341. b->prev[a2] = prev;
  342. av_assert2(a2 < 4);
  343. av_assert2(b->mb[b->next[k]]);
  344. b->bit_size[a2] += dv_rl2vlc_size(b->next[k] - prev - 1, b->mb[b->next[k]]) -
  345. dv_rl2vlc_size(b->next[k] - k - 1, b->mb[b->next[k]]);
  346. av_assert2(b->prev[a2] == k && (a2 + 1 >= 4 || b->prev[a2 + 1] != k));
  347. b->prev[a2] = prev;
  348. }
  349. b->next[prev] = b->next[k];
  350. }
  351. }
  352. b->prev[a + 1] = prev;
  353. }
  354. size[i] += b->bit_size[a];
  355. }
  356. }
  357. if (vs_total_ac_bits >= size[0] + size[1] + size[2] + size[3] + size[4])
  358. return;
  359. }
  360. } while (qnos[0] | qnos[1] | qnos[2] | qnos[3] | qnos[4]);
  361. for (a = 2; a == 2 || vs_total_ac_bits < size[0]; a += a) {
  362. b = blks;
  363. size[0] = 5 * 6 * 4; // EOB
  364. for (j = 0; j < 6 * 5; j++, b++) {
  365. prev = b->prev[0];
  366. for (k = b->next[prev]; k < 64; k = b->next[k]) {
  367. if (b->mb[k] < a && b->mb[k] > -a) {
  368. b->next[prev] = b->next[k];
  369. } else {
  370. size[0] += dv_rl2vlc_size(k - prev - 1, b->mb[k]);
  371. prev = k;
  372. }
  373. }
  374. }
  375. }
  376. }
  377. static int dv_encode_video_segment(AVCodecContext *avctx, void *arg)
  378. {
  379. DVVideoContext *s = avctx->priv_data;
  380. DVwork_chunk *work_chunk = arg;
  381. int mb_index, i, j;
  382. int mb_x, mb_y, c_offset, linesize, y_stride;
  383. uint8_t *y_ptr;
  384. uint8_t *dif;
  385. LOCAL_ALIGNED_8(uint8_t, scratch, [128]);
  386. EncBlockInfo enc_blks[5 * DV_MAX_BPM];
  387. PutBitContext pbs[5 * DV_MAX_BPM];
  388. PutBitContext *pb;
  389. EncBlockInfo *enc_blk;
  390. int vs_bit_size = 0;
  391. int qnos[5] = { 15, 15, 15, 15, 15 }; /* No quantization */
  392. int *qnosp = &qnos[0];
  393. dif = &s->buf[work_chunk->buf_offset * 80];
  394. enc_blk = &enc_blks[0];
  395. for (mb_index = 0; mb_index < 5; mb_index++) {
  396. dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y);
  397. /* initializing luminance blocks */
  398. if ((s->sys->pix_fmt == AV_PIX_FMT_YUV420P) ||
  399. (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) ||
  400. (s->sys->height >= 720 && mb_y != 134)) {
  401. y_stride = s->frame->linesize[0] << 3;
  402. } else {
  403. y_stride = 16;
  404. }
  405. y_ptr = s->frame->data[0] +
  406. ((mb_y * s->frame->linesize[0] + mb_x) << 3);
  407. linesize = s->frame->linesize[0];
  408. if (s->sys->video_stype == 4) { /* SD 422 */
  409. vs_bit_size +=
  410. dv_init_enc_block(enc_blk + 0, y_ptr, linesize, s, 0) +
  411. dv_init_enc_block(enc_blk + 1, NULL, linesize, s, 0) +
  412. dv_init_enc_block(enc_blk + 2, y_ptr + 8, linesize, s, 0) +
  413. dv_init_enc_block(enc_blk + 3, NULL, linesize, s, 0);
  414. } else {
  415. vs_bit_size +=
  416. dv_init_enc_block(enc_blk + 0, y_ptr, linesize, s, 0) +
  417. dv_init_enc_block(enc_blk + 1, y_ptr + 8, linesize, s, 0) +
  418. dv_init_enc_block(enc_blk + 2, y_ptr + y_stride, linesize, s, 0) +
  419. dv_init_enc_block(enc_blk + 3, y_ptr + 8 + y_stride, linesize, s, 0);
  420. }
  421. enc_blk += 4;
  422. /* initializing chrominance blocks */
  423. c_offset = (((mb_y >> (s->sys->pix_fmt == AV_PIX_FMT_YUV420P)) * s->frame->linesize[1] +
  424. (mb_x >> ((s->sys->pix_fmt == AV_PIX_FMT_YUV411P) ? 2 : 1))) << 3);
  425. for (j = 2; j; j--) {
  426. uint8_t *c_ptr = s->frame->data[j] + c_offset;
  427. linesize = s->frame->linesize[j];
  428. y_stride = (mb_y == 134) ? 8 : (s->frame->linesize[j] << 3);
  429. if (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
  430. uint8_t *d;
  431. uint8_t *b = scratch;
  432. for (i = 0; i < 8; i++) {
  433. d = c_ptr + (linesize << 3);
  434. b[0] = c_ptr[0];
  435. b[1] = c_ptr[1];
  436. b[2] = c_ptr[2];
  437. b[3] = c_ptr[3];
  438. b[4] = d[0];
  439. b[5] = d[1];
  440. b[6] = d[2];
  441. b[7] = d[3];
  442. c_ptr += linesize;
  443. b += 16;
  444. }
  445. c_ptr = scratch;
  446. linesize = 16;
  447. }
  448. vs_bit_size += dv_init_enc_block(enc_blk++, c_ptr, linesize, s, 1);
  449. if (s->sys->bpm == 8)
  450. vs_bit_size += dv_init_enc_block(enc_blk++, c_ptr + y_stride,
  451. linesize, s, 1);
  452. }
  453. }
  454. if (vs_total_ac_bits < vs_bit_size)
  455. dv_guess_qnos(&enc_blks[0], qnosp);
  456. /* DIF encoding process */
  457. for (j = 0; j < 5 * s->sys->bpm;) {
  458. int start_mb = j;
  459. dif[3] = *qnosp++;
  460. dif += 4;
  461. /* First pass over individual cells only */
  462. for (i = 0; i < s->sys->bpm; i++, j++) {
  463. int sz = s->sys->block_sizes[i] >> 3;
  464. init_put_bits(&pbs[j], dif, sz);
  465. put_sbits(&pbs[j], 9, ((enc_blks[j].mb[0] >> 3) - 1024 + 2) >> 2);
  466. put_bits(&pbs[j], 1, enc_blks[j].dct_mode);
  467. put_bits(&pbs[j], 2, enc_blks[j].cno);
  468. dv_encode_ac(&enc_blks[j], &pbs[j], &pbs[j + 1]);
  469. dif += sz;
  470. }
  471. /* Second pass over each MB space */
  472. pb = &pbs[start_mb];
  473. for (i = 0; i < s->sys->bpm; i++)
  474. if (enc_blks[start_mb + i].partial_bit_count)
  475. pb = dv_encode_ac(&enc_blks[start_mb + i], pb,
  476. &pbs[start_mb + s->sys->bpm]);
  477. }
  478. /* Third and final pass over the whole video segment space */
  479. pb = &pbs[0];
  480. for (j = 0; j < 5 * s->sys->bpm; j++) {
  481. if (enc_blks[j].partial_bit_count)
  482. pb = dv_encode_ac(&enc_blks[j], pb, &pbs[s->sys->bpm * 5]);
  483. if (enc_blks[j].partial_bit_count)
  484. av_log(avctx, AV_LOG_ERROR, "ac bitstream overflow\n");
  485. }
  486. for (j = 0; j < 5 * s->sys->bpm; j++) {
  487. int pos;
  488. int size = pbs[j].size_in_bits >> 3;
  489. flush_put_bits(&pbs[j]);
  490. pos = put_bits_count(&pbs[j]) >> 3;
  491. if (pos > size) {
  492. av_log(avctx, AV_LOG_ERROR,
  493. "bitstream written beyond buffer size\n");
  494. return -1;
  495. }
  496. memset(pbs[j].buf + pos, 0xff, size - pos);
  497. }
  498. return 0;
  499. }
  500. static inline int dv_write_pack(enum dv_pack_type pack_id, DVVideoContext *c,
  501. uint8_t *buf)
  502. {
  503. /*
  504. * Here's what SMPTE314M says about these two:
  505. * (page 6) APTn, AP1n, AP2n, AP3n: These data shall be identical
  506. * as track application IDs (APTn = 001, AP1n =
  507. * 001, AP2n = 001, AP3n = 001), if the source signal
  508. * comes from a digital VCR. If the signal source is
  509. * unknown, all bits for these data shall be set to 1.
  510. * (page 12) STYPE: STYPE defines a signal type of video signal
  511. * 00000b = 4:1:1 compression
  512. * 00100b = 4:2:2 compression
  513. * XXXXXX = Reserved
  514. * Now, I've got two problems with these statements:
  515. * 1. it looks like APT == 111b should be a safe bet, but it isn't.
  516. * It seems that for PAL as defined in IEC 61834 we have to set
  517. * APT to 000 and for SMPTE314M to 001.
  518. * 2. It is not at all clear what STYPE is used for 4:2:0 PAL
  519. * compression scheme (if any).
  520. */
  521. int apt = (c->sys->pix_fmt == AV_PIX_FMT_YUV420P ? 0 : 1);
  522. int fs = c->frame->top_field_first ? 0x00 : 0x40;
  523. uint8_t aspect = 0;
  524. if ((int) (av_q2d(c->avctx->sample_aspect_ratio) *
  525. c->avctx->width / c->avctx->height * 10) >= 17) /* 16:9 */
  526. aspect = 0x02;
  527. buf[0] = (uint8_t) pack_id;
  528. switch (pack_id) {
  529. case dv_header525: /* I can't imagine why these two weren't defined as real */
  530. case dv_header625: /* packs in SMPTE314M -- they definitely look like ones */
  531. buf[1] = 0xf8 | /* reserved -- always 1 */
  532. (apt & 0x07); /* APT: Track application ID */
  533. buf[2] = (0 << 7) | /* TF1: audio data is 0 - valid; 1 - invalid */
  534. (0x0f << 3) | /* reserved -- always 1 */
  535. (apt & 0x07); /* AP1: Audio application ID */
  536. buf[3] = (0 << 7) | /* TF2: video data is 0 - valid; 1 - invalid */
  537. (0x0f << 3) | /* reserved -- always 1 */
  538. (apt & 0x07); /* AP2: Video application ID */
  539. buf[4] = (0 << 7) | /* TF3: subcode(SSYB) is 0 - valid; 1 - invalid */
  540. (0x0f << 3) | /* reserved -- always 1 */
  541. (apt & 0x07); /* AP3: Subcode application ID */
  542. break;
  543. case dv_video_source:
  544. buf[1] = 0xff; /* reserved -- always 1 */
  545. buf[2] = (1 << 7) | /* B/W: 0 - b/w, 1 - color */
  546. (1 << 6) | /* following CLF is valid - 0, invalid - 1 */
  547. (3 << 4) | /* CLF: color frames ID (see ITU-R BT.470-4) */
  548. 0xf; /* reserved -- always 1 */
  549. buf[3] = (3 << 6) | /* reserved -- always 1 */
  550. (c->sys->dsf << 5) | /* system: 60fields/50fields */
  551. c->sys->video_stype; /* signal type video compression */
  552. buf[4] = 0xff; /* VISC: 0xff -- no information */
  553. break;
  554. case dv_video_control:
  555. buf[1] = (0 << 6) | /* Copy generation management (CGMS) 0 -- free */
  556. 0x3f; /* reserved -- always 1 */
  557. buf[2] = 0xc8 | /* reserved -- always b11001xxx */
  558. aspect;
  559. buf[3] = (1 << 7) | /* frame/field flag 1 -- frame, 0 -- field */
  560. fs | /* first/second field flag 0 -- field 2, 1 -- field 1 */
  561. (1 << 5) | /* frame change flag 0 -- same picture as before, 1 -- different */
  562. (1 << 4) | /* 1 - interlaced, 0 - noninterlaced */
  563. 0xc; /* reserved -- always b1100 */
  564. buf[4] = 0xff; /* reserved -- always 1 */
  565. break;
  566. default:
  567. buf[1] =
  568. buf[2] =
  569. buf[3] =
  570. buf[4] = 0xff;
  571. }
  572. return 5;
  573. }
  574. static inline int dv_write_dif_id(enum dv_section_type t, uint8_t chan_num,
  575. uint8_t seq_num, uint8_t dif_num,
  576. uint8_t *buf)
  577. {
  578. buf[0] = (uint8_t) t; /* Section type */
  579. buf[1] = (seq_num << 4) | /* DIF seq number 0-9 for 525/60; 0-11 for 625/50 */
  580. (chan_num << 3) | /* FSC: for 50Mb/s 0 - first channel; 1 - second */
  581. 7; /* reserved -- always 1 */
  582. buf[2] = dif_num; /* DIF block number Video: 0-134, Audio: 0-8 */
  583. return 3;
  584. }
  585. static inline int dv_write_ssyb_id(uint8_t syb_num, uint8_t fr, uint8_t *buf)
  586. {
  587. if (syb_num == 0 || syb_num == 6) {
  588. buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */
  589. (0 << 4) | /* AP3 (Subcode application ID) */
  590. 0x0f; /* reserved -- always 1 */
  591. } else if (syb_num == 11) {
  592. buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */
  593. 0x7f; /* reserved -- always 1 */
  594. } else {
  595. buf[0] = (fr << 7) | /* FR ID 1 - first half of each channel; 0 - second */
  596. (0 << 4) | /* APT (Track application ID) */
  597. 0x0f; /* reserved -- always 1 */
  598. }
  599. buf[1] = 0xf0 | /* reserved -- always 1 */
  600. (syb_num & 0x0f); /* SSYB number 0 - 11 */
  601. buf[2] = 0xff; /* reserved -- always 1 */
  602. return 3;
  603. }
  604. static void dv_format_frame(DVVideoContext *c, uint8_t *buf)
  605. {
  606. int chan, i, j, k;
  607. for (chan = 0; chan < c->sys->n_difchan; chan++) {
  608. for (i = 0; i < c->sys->difseg_size; i++) {
  609. memset(buf, 0xff, 80 * 6); /* first 6 DIF blocks are for control data */
  610. /* DV header: 1DIF */
  611. buf += dv_write_dif_id(dv_sect_header, chan, i, 0, buf);
  612. buf += dv_write_pack((c->sys->dsf ? dv_header625 : dv_header525),
  613. c, buf);
  614. buf += 72; /* unused bytes */
  615. /* DV subcode: 2DIFs */
  616. for (j = 0; j < 2; j++) {
  617. buf += dv_write_dif_id(dv_sect_subcode, chan, i, j, buf);
  618. for (k = 0; k < 6; k++)
  619. buf += dv_write_ssyb_id(k, (i < c->sys->difseg_size / 2), buf) + 5;
  620. buf += 29; /* unused bytes */
  621. }
  622. /* DV VAUX: 3DIFS */
  623. for (j = 0; j < 3; j++) {
  624. buf += dv_write_dif_id(dv_sect_vaux, chan, i, j, buf);
  625. buf += dv_write_pack(dv_video_source, c, buf);
  626. buf += dv_write_pack(dv_video_control, c, buf);
  627. buf += 7 * 5;
  628. buf += dv_write_pack(dv_video_source, c, buf);
  629. buf += dv_write_pack(dv_video_control, c, buf);
  630. buf += 4 * 5 + 2; /* unused bytes */
  631. }
  632. /* DV Audio/Video: 135 Video DIFs + 9 Audio DIFs */
  633. for (j = 0; j < 135; j++) {
  634. if (j % 15 == 0) {
  635. memset(buf, 0xff, 80);
  636. buf += dv_write_dif_id(dv_sect_audio, chan, i, j / 15, buf);
  637. buf += 77; /* audio control & shuffled PCM audio */
  638. }
  639. buf += dv_write_dif_id(dv_sect_video, chan, i, j, buf);
  640. buf += 77; /* 1 video macroblock: 1 bytes control
  641. * 4 * 14 bytes Y 8x8 data
  642. * 10 bytes Cr 8x8 data
  643. * 10 bytes Cb 8x8 data */
  644. }
  645. }
  646. }
  647. }
  648. static int dvvideo_encode_frame(AVCodecContext *c, AVPacket *pkt,
  649. const AVFrame *frame, int *got_packet)
  650. {
  651. DVVideoContext *s = c->priv_data;
  652. int ret;
  653. if ((ret = ff_alloc_packet2(c, pkt, s->sys->frame_size)) < 0)
  654. return ret;
  655. c->pix_fmt = s->sys->pix_fmt;
  656. s->frame = frame;
  657. c->coded_frame->key_frame = 1;
  658. c->coded_frame->pict_type = AV_PICTURE_TYPE_I;
  659. s->buf = pkt->data;
  660. c->execute(c, dv_encode_video_segment, s->work_chunks, NULL,
  661. dv_work_pool_size(s->sys), sizeof(DVwork_chunk));
  662. emms_c();
  663. dv_format_frame(s, pkt->data);
  664. pkt->flags |= AV_PKT_FLAG_KEY;
  665. *got_packet = 1;
  666. return 0;
  667. }
  668. static int dvvideo_encode_close(AVCodecContext *avctx)
  669. {
  670. av_frame_free(&avctx->coded_frame);
  671. return 0;
  672. }
  673. AVCodec ff_dvvideo_encoder = {
  674. .name = "dvvideo",
  675. .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"),
  676. .type = AVMEDIA_TYPE_VIDEO,
  677. .id = AV_CODEC_ID_DVVIDEO,
  678. .priv_data_size = sizeof(DVVideoContext),
  679. .init = dvvideo_encode_init,
  680. .encode2 = dvvideo_encode_frame,
  681. .close = dvvideo_encode_close,
  682. .capabilities = CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS | CODEC_CAP_INTRA_ONLY,
  683. .pix_fmts = (const enum AVPixelFormat[]) {
  684. AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV422P,
  685. AV_PIX_FMT_YUV420P, AV_PIX_FMT_NONE
  686. },
  687. };