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.

421 lines
15KB

  1. /*
  2. * DV decoder
  3. * Copyright (c) 2002 Fabrice Bellard
  4. * Copyright (c) 2004 Roman Shaposhnik
  5. *
  6. * 50 Mbps (DVCPRO50) support
  7. * Copyright (c) 2006 Daniel Maas <dmaas@maasdigital.com>
  8. *
  9. * 100 Mbps (DVCPRO HD) support
  10. * Initial code by Daniel Maas <dmaas@maasdigital.com> (funded by BBC R&D)
  11. * Final code by Roman Shaposhnik
  12. *
  13. * Many thanks to Dan Dennedy <dan@dennedy.org> for providing wealth
  14. * of DV technical info.
  15. *
  16. * This file is part of Libav.
  17. *
  18. * Libav is free software; you can redistribute it and/or
  19. * modify it under the terms of the GNU Lesser General Public
  20. * License as published by the Free Software Foundation; either
  21. * version 2.1 of the License, or (at your option) any later version.
  22. *
  23. * Libav is distributed in the hope that it will be useful,
  24. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  25. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  26. * Lesser General Public License for more details.
  27. *
  28. * You should have received a copy of the GNU Lesser General Public
  29. * License along with Libav; if not, write to the Free Software
  30. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  31. */
  32. /**
  33. * @file
  34. * DV decoder
  35. */
  36. #include "libavutil/imgutils.h"
  37. #include "libavutil/internal.h"
  38. #include "libavutil/pixdesc.h"
  39. #include "avcodec.h"
  40. #include "dv.h"
  41. #include "dvdata.h"
  42. #include "get_bits.h"
  43. #include "idctdsp.h"
  44. #include "internal.h"
  45. #include "put_bits.h"
  46. #include "simple_idct.h"
  47. typedef struct BlockInfo {
  48. const uint32_t *factor_table;
  49. const uint8_t *scan_table;
  50. uint8_t pos; /* position in block */
  51. void (*idct_put)(uint8_t *dest, int line_size, int16_t *block);
  52. uint8_t partial_bit_count;
  53. uint32_t partial_bit_buffer;
  54. int shift_offset;
  55. } BlockInfo;
  56. static const int dv_iweight_bits = 14;
  57. static av_cold int dvvideo_decode_init(AVCodecContext *avctx)
  58. {
  59. DVVideoContext *s = avctx->priv_data;
  60. IDCTDSPContext idsp;
  61. int i;
  62. ff_idctdsp_init(&idsp, avctx);
  63. for (i = 0; i < 64; i++)
  64. s->dv_zigzag[0][i] = idsp.idct_permutation[ff_zigzag_direct[i]];
  65. memcpy(s->dv_zigzag[1], ff_dv_zigzag248_direct, sizeof(s->dv_zigzag[1]));
  66. s->idct_put[0] = idsp.idct_put;
  67. s->idct_put[1] = ff_simple_idct248_put;
  68. return ff_dvvideo_init(avctx);
  69. }
  70. /* decode AC coefficients */
  71. static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, int16_t *block)
  72. {
  73. int last_index = gb->size_in_bits;
  74. const uint8_t *scan_table = mb->scan_table;
  75. const uint32_t *factor_table = mb->factor_table;
  76. int pos = mb->pos;
  77. int partial_bit_count = mb->partial_bit_count;
  78. int level, run, vlc_len, index;
  79. OPEN_READER_NOSIZE(re, gb);
  80. UPDATE_CACHE(re, gb);
  81. /* if we must parse a partial VLC, we do it here */
  82. if (partial_bit_count > 0) {
  83. re_cache = re_cache >> partial_bit_count |
  84. mb->partial_bit_buffer;
  85. re_index -= partial_bit_count;
  86. mb->partial_bit_count = 0;
  87. }
  88. /* get the AC coefficients until last_index is reached */
  89. for (;;) {
  90. ff_dlog(NULL, "%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16),
  91. re_index);
  92. /* our own optimized GET_RL_VLC */
  93. index = NEG_USR32(re_cache, TEX_VLC_BITS);
  94. vlc_len = ff_dv_rl_vlc[index].len;
  95. if (vlc_len < 0) {
  96. index = NEG_USR32((unsigned) re_cache << TEX_VLC_BITS, -vlc_len) +
  97. ff_dv_rl_vlc[index].level;
  98. vlc_len = TEX_VLC_BITS - vlc_len;
  99. }
  100. level = ff_dv_rl_vlc[index].level;
  101. run = ff_dv_rl_vlc[index].run;
  102. /* gotta check if we're still within gb boundaries */
  103. if (re_index + vlc_len > last_index) {
  104. /* should be < 16 bits otherwise a codeword could have been parsed */
  105. mb->partial_bit_count = last_index - re_index;
  106. mb->partial_bit_buffer = re_cache & ~(-1u >> mb->partial_bit_count);
  107. re_index = last_index;
  108. break;
  109. }
  110. re_index += vlc_len;
  111. ff_dlog(NULL, "run=%d level=%d\n", run, level);
  112. pos += run;
  113. if (pos >= 64)
  114. break;
  115. level = (level * factor_table[pos] + (1 << (dv_iweight_bits - 1))) >>
  116. dv_iweight_bits;
  117. block[scan_table[pos]] = level;
  118. UPDATE_CACHE(re, gb);
  119. }
  120. CLOSE_READER(re, gb);
  121. mb->pos = pos;
  122. }
  123. static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
  124. {
  125. int bits_left = get_bits_left(gb);
  126. while (bits_left >= MIN_CACHE_BITS) {
  127. put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS));
  128. bits_left -= MIN_CACHE_BITS;
  129. }
  130. if (bits_left > 0)
  131. put_bits(pb, bits_left, get_bits(gb, bits_left));
  132. }
  133. /* mb_x and mb_y are in units of 8 pixels */
  134. static int dv_decode_video_segment(AVCodecContext *avctx, void *arg)
  135. {
  136. DVVideoContext *s = avctx->priv_data;
  137. DVwork_chunk *work_chunk = arg;
  138. int quant, dc, dct_mode, class1, j;
  139. int mb_index, mb_x, mb_y, last_index;
  140. int y_stride, linesize;
  141. int16_t *block, *block1;
  142. int c_offset;
  143. uint8_t *y_ptr;
  144. const uint8_t *buf_ptr;
  145. PutBitContext pb, vs_pb;
  146. GetBitContext gb;
  147. BlockInfo mb_data[5 * DV_MAX_BPM], *mb, *mb1;
  148. LOCAL_ALIGNED_16(int16_t, sblock, [5 * DV_MAX_BPM], [64]);
  149. LOCAL_ALIGNED_16(uint8_t, mb_bit_buffer, [80 + AV_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
  150. LOCAL_ALIGNED_16(uint8_t, vs_bit_buffer, [80 * 5 + AV_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
  151. const int log2_blocksize = 3;
  152. int is_field_mode[5];
  153. assert((((int) mb_bit_buffer) & 7) == 0);
  154. assert((((int) vs_bit_buffer) & 7) == 0);
  155. memset(sblock, 0, 5 * DV_MAX_BPM * sizeof(*sblock));
  156. /* pass 1: read DC and AC coefficients in blocks */
  157. buf_ptr = &s->buf[work_chunk->buf_offset * 80];
  158. block1 = &sblock[0][0];
  159. mb1 = mb_data;
  160. init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
  161. for (mb_index = 0; mb_index < 5; mb_index++, mb1 += s->sys->bpm, block1 += s->sys->bpm * 64) {
  162. /* skip header */
  163. quant = buf_ptr[3] & 0x0f;
  164. buf_ptr += 4;
  165. init_put_bits(&pb, mb_bit_buffer, 80);
  166. mb = mb1;
  167. block = block1;
  168. is_field_mode[mb_index] = 0;
  169. for (j = 0; j < s->sys->bpm; j++) {
  170. last_index = s->sys->block_sizes[j];
  171. init_get_bits(&gb, buf_ptr, last_index);
  172. /* get the DC */
  173. dc = get_sbits(&gb, 9);
  174. dct_mode = get_bits1(&gb);
  175. class1 = get_bits(&gb, 2);
  176. if (DV_PROFILE_IS_HD(s->sys)) {
  177. mb->idct_put = s->idct_put[0];
  178. mb->scan_table = s->dv_zigzag[0];
  179. mb->factor_table = &s->idct_factor[(j >= 4) * 4 * 16 * 64 +
  180. class1 * 16 * 64 +
  181. quant * 64];
  182. is_field_mode[mb_index] |= !j && dct_mode;
  183. } else {
  184. mb->idct_put = s->idct_put[dct_mode && log2_blocksize == 3];
  185. mb->scan_table = s->dv_zigzag[dct_mode];
  186. mb->factor_table =
  187. &s->idct_factor[(class1 == 3) * 2 * 22 * 64 +
  188. dct_mode * 22 * 64 +
  189. (quant + ff_dv_quant_offset[class1]) * 64];
  190. }
  191. dc = dc << 2;
  192. /* convert to unsigned because 128 is not added in the
  193. * standard IDCT */
  194. dc += 1024;
  195. block[0] = dc;
  196. buf_ptr += last_index >> 3;
  197. mb->pos = 0;
  198. mb->partial_bit_count = 0;
  199. ff_dlog(avctx, "MB block: %d, %d ", mb_index, j);
  200. dv_decode_ac(&gb, mb, block);
  201. /* write the remaining bits in a new buffer only if the
  202. * block is finished */
  203. if (mb->pos >= 64)
  204. bit_copy(&pb, &gb);
  205. block += 64;
  206. mb++;
  207. }
  208. /* pass 2: we can do it just after */
  209. ff_dlog(avctx, "***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
  210. block = block1;
  211. mb = mb1;
  212. init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));
  213. put_bits32(&pb, 0); // padding must be zeroed
  214. flush_put_bits(&pb);
  215. for (j = 0; j < s->sys->bpm; j++, block += 64, mb++) {
  216. if (mb->pos < 64 && get_bits_left(&gb) > 0) {
  217. dv_decode_ac(&gb, mb, block);
  218. /* if still not finished, no need to parse other blocks */
  219. if (mb->pos < 64)
  220. break;
  221. }
  222. }
  223. /* all blocks are finished, so the extra bytes can be used at
  224. * the video segment level */
  225. if (j >= s->sys->bpm)
  226. bit_copy(&vs_pb, &gb);
  227. }
  228. /* we need a pass over the whole video segment */
  229. ff_dlog(avctx, "***pass 3 size=%d\n", put_bits_count(&vs_pb));
  230. block = &sblock[0][0];
  231. mb = mb_data;
  232. init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));
  233. put_bits32(&vs_pb, 0); // padding must be zeroed
  234. flush_put_bits(&vs_pb);
  235. for (mb_index = 0; mb_index < 5; mb_index++) {
  236. for (j = 0; j < s->sys->bpm; j++) {
  237. if (mb->pos < 64) {
  238. ff_dlog(avctx, "start %d:%d\n", mb_index, j);
  239. dv_decode_ac(&gb, mb, block);
  240. }
  241. if (mb->pos >= 64 && mb->pos < 127)
  242. av_log(avctx, AV_LOG_ERROR,
  243. "AC EOB marker is absent pos=%d\n", mb->pos);
  244. block += 64;
  245. mb++;
  246. }
  247. }
  248. /* compute idct and place blocks */
  249. block = &sblock[0][0];
  250. mb = mb_data;
  251. for (mb_index = 0; mb_index < 5; mb_index++) {
  252. dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y);
  253. /* idct_put'ting luminance */
  254. if ((s->sys->pix_fmt == AV_PIX_FMT_YUV420P) ||
  255. (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) ||
  256. (s->sys->height >= 720 && mb_y != 134)) {
  257. y_stride = (s->frame->linesize[0] <<
  258. ((!is_field_mode[mb_index]) * log2_blocksize));
  259. } else {
  260. y_stride = (2 << log2_blocksize);
  261. }
  262. y_ptr = s->frame->data[0] +
  263. ((mb_y * s->frame->linesize[0] + mb_x) << log2_blocksize);
  264. linesize = s->frame->linesize[0] << is_field_mode[mb_index];
  265. mb[0].idct_put(y_ptr, linesize, block + 0 * 64);
  266. if (s->sys->video_stype == 4) { /* SD 422 */
  267. mb[2].idct_put(y_ptr + (1 << log2_blocksize), linesize, block + 2 * 64);
  268. } else {
  269. mb[1].idct_put(y_ptr + (1 << log2_blocksize), linesize, block + 1 * 64);
  270. mb[2].idct_put(y_ptr + y_stride, linesize, block + 2 * 64);
  271. mb[3].idct_put(y_ptr + (1 << log2_blocksize) + y_stride, linesize, block + 3 * 64);
  272. }
  273. mb += 4;
  274. block += 4 * 64;
  275. /* idct_put'ting chrominance */
  276. c_offset = (((mb_y >> (s->sys->pix_fmt == AV_PIX_FMT_YUV420P)) * s->frame->linesize[1] +
  277. (mb_x >> ((s->sys->pix_fmt == AV_PIX_FMT_YUV411P) ? 2 : 1))) << log2_blocksize);
  278. for (j = 2; j; j--) {
  279. uint8_t *c_ptr = s->frame->data[j] + c_offset;
  280. if (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
  281. uint64_t aligned_pixels[64 / 8];
  282. uint8_t *pixels = (uint8_t *) aligned_pixels;
  283. uint8_t *c_ptr1, *ptr1;
  284. int x, y;
  285. mb->idct_put(pixels, 8, block);
  286. for (y = 0; y < (1 << log2_blocksize); y++, c_ptr += s->frame->linesize[j], pixels += 8) {
  287. ptr1 = pixels + (1 << (log2_blocksize - 1));
  288. c_ptr1 = c_ptr + (s->frame->linesize[j] << log2_blocksize);
  289. for (x = 0; x < (1 << (log2_blocksize - 1)); x++) {
  290. c_ptr[x] = pixels[x];
  291. c_ptr1[x] = ptr1[x];
  292. }
  293. }
  294. block += 64;
  295. mb++;
  296. } else {
  297. y_stride = (mb_y == 134) ? (1 << log2_blocksize) :
  298. s->frame->linesize[j] << ((!is_field_mode[mb_index]) * log2_blocksize);
  299. linesize = s->frame->linesize[j] << is_field_mode[mb_index];
  300. (mb++)->idct_put(c_ptr, linesize, block);
  301. block += 64;
  302. if (s->sys->bpm == 8) {
  303. (mb++)->idct_put(c_ptr + y_stride, linesize, block);
  304. block += 64;
  305. }
  306. }
  307. }
  308. }
  309. return 0;
  310. }
  311. /* NOTE: exactly one frame must be given (120000 bytes for NTSC,
  312. * 144000 bytes for PAL - or twice those for 50Mbps) */
  313. static int dvvideo_decode_frame(AVCodecContext *avctx, void *data,
  314. int *got_frame, AVPacket *avpkt)
  315. {
  316. uint8_t *buf = avpkt->data;
  317. int buf_size = avpkt->size;
  318. DVVideoContext *s = avctx->priv_data;
  319. AVFrame *frame = data;
  320. const uint8_t *vsc_pack;
  321. int apt, is16_9, ret;
  322. const AVDVProfile *sys;
  323. sys = av_dv_frame_profile(s->sys, buf, buf_size);
  324. if (!sys || buf_size < sys->frame_size) {
  325. av_log(avctx, AV_LOG_ERROR, "could not find dv frame profile\n");
  326. return -1; /* NOTE: we only accept several full frames */
  327. }
  328. if (sys != s->sys) {
  329. ret = ff_dv_init_dynamic_tables(s, sys);
  330. if (ret < 0) {
  331. av_log(avctx, AV_LOG_ERROR, "Error initializing the work tables.\n");
  332. return ret;
  333. }
  334. s->sys = sys;
  335. }
  336. s->frame = frame;
  337. frame->key_frame = 1;
  338. frame->pict_type = AV_PICTURE_TYPE_I;
  339. avctx->pix_fmt = s->sys->pix_fmt;
  340. avctx->framerate = av_inv_q(s->sys->time_base);
  341. ret = ff_set_dimensions(avctx, s->sys->width, s->sys->height);
  342. if (ret < 0)
  343. return ret;
  344. /* Determine the codec's sample_aspect ratio from the packet */
  345. vsc_pack = buf + 80 * 5 + 48 + 5;
  346. if (*vsc_pack == dv_video_control) {
  347. apt = buf[4] & 0x07;
  348. is16_9 = (vsc_pack[2] & 0x07) == 0x02 ||
  349. (!apt && (vsc_pack[2] & 0x07) == 0x07);
  350. ff_set_sar(avctx, s->sys->sar[is16_9]);
  351. }
  352. if (ff_get_buffer(avctx, frame, 0) < 0) {
  353. av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
  354. return -1;
  355. }
  356. frame->interlaced_frame = 1;
  357. frame->top_field_first = 0;
  358. s->buf = buf;
  359. avctx->execute(avctx, dv_decode_video_segment, s->work_chunks, NULL,
  360. dv_work_pool_size(s->sys), sizeof(DVwork_chunk));
  361. emms_c();
  362. /* return image */
  363. *got_frame = 1;
  364. return s->sys->frame_size;
  365. }
  366. AVCodec ff_dvvideo_decoder = {
  367. .name = "dvvideo",
  368. .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"),
  369. .type = AVMEDIA_TYPE_VIDEO,
  370. .id = AV_CODEC_ID_DVVIDEO,
  371. .priv_data_size = sizeof(DVVideoContext),
  372. .init = dvvideo_decode_init,
  373. .decode = dvvideo_decode_frame,
  374. .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_SLICE_THREADS,
  375. };