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.

419 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 FFmpeg.
  17. *
  18. * FFmpeg 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. * FFmpeg 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 FFmpeg; 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/avassert.h"
  37. #include "libavutil/internal.h"
  38. #include "libavutil/imgutils.h"
  39. #include "libavutil/pixdesc.h"
  40. #include "avcodec.h"
  41. #include "internal.h"
  42. #include "get_bits.h"
  43. #include "put_bits.h"
  44. #include "simple_idct.h"
  45. #include "dvdata.h"
  46. #include "dv.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. DSPContext dsp;
  61. int i;
  62. memset(&dsp,0, sizeof(dsp));
  63. ff_dsputil_init(&dsp, avctx);
  64. for (i = 0; i < 64; i++)
  65. s->dv_zigzag[0][i] = dsp.idct_permutation[ff_zigzag_direct[i]];
  66. if (avctx->lowres){
  67. for (i = 0; i < 64; i++){
  68. int j = ff_dv_zigzag248_direct[i];
  69. s->dv_zigzag[1][i] = dsp.idct_permutation[(j & 7) + (j & 8) * 4 + (j & 48) / 2];
  70. }
  71. }else
  72. memcpy(s->dv_zigzag[1], ff_dv_zigzag248_direct, sizeof(s->dv_zigzag[1]));
  73. s->idct_put[0] = dsp.idct_put;
  74. s->idct_put[1] = ff_simple_idct248_put;
  75. return ff_dvvideo_init(avctx);
  76. }
  77. /* decode AC coefficients */
  78. static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, int16_t *block)
  79. {
  80. int last_index = gb->size_in_bits;
  81. const uint8_t *scan_table = mb->scan_table;
  82. const uint32_t *factor_table = mb->factor_table;
  83. int pos = mb->pos;
  84. int partial_bit_count = mb->partial_bit_count;
  85. int level, run, vlc_len, index;
  86. OPEN_READER(re, gb);
  87. UPDATE_CACHE(re, gb);
  88. /* if we must parse a partial VLC, we do it here */
  89. if (partial_bit_count > 0) {
  90. re_cache = re_cache >> partial_bit_count | mb->partial_bit_buffer;
  91. re_index -= partial_bit_count;
  92. mb->partial_bit_count = 0;
  93. }
  94. /* get the AC coefficients until last_index is reached */
  95. for (;;) {
  96. av_dlog(NULL, "%2d: bits=%04x index=%d\n", pos, SHOW_UBITS(re, gb, 16),
  97. re_index);
  98. /* our own optimized GET_RL_VLC */
  99. index = NEG_USR32(re_cache, TEX_VLC_BITS);
  100. vlc_len = ff_dv_rl_vlc[index].len;
  101. if (vlc_len < 0) {
  102. index = NEG_USR32((unsigned)re_cache << TEX_VLC_BITS, -vlc_len) +
  103. ff_dv_rl_vlc[index].level;
  104. vlc_len = TEX_VLC_BITS - vlc_len;
  105. }
  106. level = ff_dv_rl_vlc[index].level;
  107. run = ff_dv_rl_vlc[index].run;
  108. /* gotta check if we're still within gb boundaries */
  109. if (re_index + vlc_len > last_index) {
  110. /* should be < 16 bits otherwise a codeword could have been parsed */
  111. mb->partial_bit_count = last_index - re_index;
  112. mb->partial_bit_buffer = re_cache & ~(-1u >> mb->partial_bit_count);
  113. re_index = last_index;
  114. break;
  115. }
  116. re_index += vlc_len;
  117. av_dlog(NULL, "run=%d level=%d\n", run, level);
  118. pos += run;
  119. if (pos >= 64)
  120. break;
  121. level = (level * factor_table[pos] + (1 << (dv_iweight_bits - 1))) >> dv_iweight_bits;
  122. block[scan_table[pos]] = level;
  123. UPDATE_CACHE(re, gb);
  124. }
  125. CLOSE_READER(re, gb);
  126. mb->pos = pos;
  127. }
  128. static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)
  129. {
  130. int bits_left = get_bits_left(gb);
  131. while (bits_left >= MIN_CACHE_BITS) {
  132. put_bits(pb, MIN_CACHE_BITS, get_bits(gb, MIN_CACHE_BITS));
  133. bits_left -= MIN_CACHE_BITS;
  134. }
  135. if (bits_left > 0) {
  136. put_bits(pb, bits_left, get_bits(gb, bits_left));
  137. }
  138. }
  139. /* mb_x and mb_y are in units of 8 pixels */
  140. static int dv_decode_video_segment(AVCodecContext *avctx, void *arg)
  141. {
  142. DVVideoContext *s = avctx->priv_data;
  143. DVwork_chunk *work_chunk = arg;
  144. int quant, dc, dct_mode, class1, j;
  145. int mb_index, mb_x, mb_y, last_index;
  146. int y_stride, linesize;
  147. int16_t *block, *block1;
  148. int c_offset;
  149. uint8_t *y_ptr;
  150. const uint8_t *buf_ptr;
  151. PutBitContext pb, vs_pb;
  152. GetBitContext gb;
  153. BlockInfo mb_data[5 * DV_MAX_BPM], *mb, *mb1;
  154. LOCAL_ALIGNED_16(int16_t, sblock, [5*DV_MAX_BPM], [64]);
  155. LOCAL_ALIGNED_16(uint8_t, mb_bit_buffer, [ 80 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
  156. LOCAL_ALIGNED_16(uint8_t, vs_bit_buffer, [5*80 + FF_INPUT_BUFFER_PADDING_SIZE]); /* allow some slack */
  157. const int log2_blocksize = 3-s->avctx->lowres;
  158. int is_field_mode[5];
  159. av_assert1((((int)mb_bit_buffer) & 7) == 0);
  160. av_assert1((((int)vs_bit_buffer) & 7) == 0);
  161. memset(sblock, 0, 5*DV_MAX_BPM*sizeof(*sblock));
  162. /* pass 1: read DC and AC coefficients in blocks */
  163. buf_ptr = &s->buf[work_chunk->buf_offset*80];
  164. block1 = &sblock[0][0];
  165. mb1 = mb_data;
  166. init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);
  167. for (mb_index = 0; mb_index < 5; mb_index++, mb1 += s->sys->bpm, block1 += s->sys->bpm * 64) {
  168. /* skip header */
  169. quant = buf_ptr[3] & 0x0f;
  170. buf_ptr += 4;
  171. init_put_bits(&pb, mb_bit_buffer, 80);
  172. mb = mb1;
  173. block = block1;
  174. is_field_mode[mb_index] = 0;
  175. for (j = 0; j < s->sys->bpm; j++) {
  176. last_index = s->sys->block_sizes[j];
  177. init_get_bits(&gb, buf_ptr, last_index);
  178. /* get the DC */
  179. dc = get_sbits(&gb, 9);
  180. dct_mode = get_bits1(&gb);
  181. class1 = get_bits(&gb, 2);
  182. if (DV_PROFILE_IS_HD(s->sys)) {
  183. mb->idct_put = s->idct_put[0];
  184. mb->scan_table = s->dv_zigzag[0];
  185. mb->factor_table = &s->idct_factor[(j >= 4)*4*16*64 + class1*16*64 + quant*64];
  186. is_field_mode[mb_index] |= !j && dct_mode;
  187. } else {
  188. mb->idct_put = s->idct_put[dct_mode && log2_blocksize == 3];
  189. mb->scan_table = s->dv_zigzag[dct_mode];
  190. mb->factor_table = &s->idct_factor[(class1 == 3)*2*22*64 + dct_mode*22*64 +
  191. (quant + ff_dv_quant_offset[class1])*64];
  192. }
  193. dc = dc << 2;
  194. /* convert to unsigned because 128 is not added in the
  195. standard IDCT */
  196. dc += 1024;
  197. block[0] = dc;
  198. buf_ptr += last_index >> 3;
  199. mb->pos = 0;
  200. mb->partial_bit_count = 0;
  201. av_dlog(avctx, "MB block: %d, %d ", mb_index, j);
  202. dv_decode_ac(&gb, mb, block);
  203. /* write the remaining bits in a new buffer only if the
  204. block is finished */
  205. if (mb->pos >= 64)
  206. bit_copy(&pb, &gb);
  207. block += 64;
  208. mb++;
  209. }
  210. /* pass 2: we can do it just after */
  211. av_dlog(avctx, "***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);
  212. block = block1;
  213. mb = mb1;
  214. init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));
  215. put_bits32(&pb, 0); // padding must be zeroed
  216. flush_put_bits(&pb);
  217. for (j = 0; j < s->sys->bpm; j++, block += 64, mb++) {
  218. if (mb->pos < 64 && get_bits_left(&gb) > 0) {
  219. dv_decode_ac(&gb, mb, block);
  220. /* if still not finished, no need to parse other blocks */
  221. if (mb->pos < 64)
  222. break;
  223. }
  224. }
  225. /* all blocks are finished, so the extra bytes can be used at
  226. the video segment level */
  227. if (j >= s->sys->bpm)
  228. bit_copy(&vs_pb, &gb);
  229. }
  230. /* we need a pass over the whole video segment */
  231. av_dlog(avctx, "***pass 3 size=%d\n", put_bits_count(&vs_pb));
  232. block = &sblock[0][0];
  233. mb = mb_data;
  234. init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));
  235. put_bits32(&vs_pb, 0); // padding must be zeroed
  236. flush_put_bits(&vs_pb);
  237. for (mb_index = 0; mb_index < 5; mb_index++) {
  238. for (j = 0; j < s->sys->bpm; j++) {
  239. if (mb->pos < 64 && get_bits_left(&gb) > 0) {
  240. av_dlog(avctx, "start %d:%d\n", mb_index, j);
  241. dv_decode_ac(&gb, mb, block);
  242. }
  243. if (mb->pos >= 64 && mb->pos < 127)
  244. av_log(avctx, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos);
  245. block += 64;
  246. mb++;
  247. }
  248. }
  249. /* compute idct and place blocks */
  250. block = &sblock[0][0];
  251. mb = mb_data;
  252. for (mb_index = 0; mb_index < 5; mb_index++) {
  253. dv_calculate_mb_xy(s, work_chunk, mb_index, &mb_x, &mb_y);
  254. /* idct_put'ting luminance */
  255. if ((s->sys->pix_fmt == AV_PIX_FMT_YUV420P) ||
  256. (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) ||
  257. (s->sys->height >= 720 && mb_y != 134)) {
  258. y_stride = (s->frame->linesize[0] << ((!is_field_mode[mb_index]) * log2_blocksize));
  259. } else {
  260. y_stride = (2 << log2_blocksize);
  261. }
  262. y_ptr = s->frame->data[0] + ((mb_y * s->frame->linesize[0] + mb_x) << log2_blocksize);
  263. linesize = s->frame->linesize[0] << is_field_mode[mb_index];
  264. mb[0] .idct_put(y_ptr , linesize, block + 0*64);
  265. if (s->sys->video_stype == 4) { /* SD 422 */
  266. mb[2].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 2*64);
  267. } else {
  268. mb[1].idct_put(y_ptr + (1 << log2_blocksize) , linesize, block + 1*64);
  269. mb[2].idct_put(y_ptr + y_stride, linesize, block + 2*64);
  270. mb[3].idct_put(y_ptr + (1 << log2_blocksize) + y_stride, linesize, block + 3*64);
  271. }
  272. mb += 4;
  273. block += 4*64;
  274. /* idct_put'ting chrominance */
  275. c_offset = (((mb_y >> (s->sys->pix_fmt == AV_PIX_FMT_YUV420P)) * s->frame->linesize[1] +
  276. (mb_x >> ((s->sys->pix_fmt == AV_PIX_FMT_YUV411P) ? 2 : 1))) << log2_blocksize);
  277. for (j = 2; j; j--) {
  278. uint8_t *c_ptr = s->frame->data[j] + c_offset;
  279. if (s->sys->pix_fmt == AV_PIX_FMT_YUV411P && mb_x >= (704 / 8)) {
  280. uint64_t aligned_pixels[64/8];
  281. uint8_t *pixels = (uint8_t*)aligned_pixels;
  282. uint8_t *c_ptr1, *ptr1;
  283. int x, y;
  284. mb->idct_put(pixels, 8, block);
  285. for (y = 0; y < (1 << log2_blocksize); y++, c_ptr += s->frame->linesize[j], pixels += 8) {
  286. ptr1 = pixels + ((1 << (log2_blocksize))>>1);
  287. c_ptr1 = c_ptr + (s->frame->linesize[j] << log2_blocksize);
  288. for (x = 0; x < (1 << FFMAX(log2_blocksize - 1, 0)); x++) {
  289. c_ptr[x] = pixels[x];
  290. c_ptr1[x] = ptr1[x];
  291. }
  292. }
  293. block += 64; mb++;
  294. } else {
  295. y_stride = (mb_y == 134) ? (1 << log2_blocksize) :
  296. s->frame->linesize[j] << ((!is_field_mode[mb_index]) * log2_blocksize);
  297. linesize = s->frame->linesize[j] << is_field_mode[mb_index];
  298. (mb++)-> idct_put(c_ptr , linesize, block); block += 64;
  299. if (s->sys->bpm == 8) {
  300. (mb++)->idct_put(c_ptr + y_stride, linesize, block); block += 64;
  301. }
  302. }
  303. }
  304. }
  305. return 0;
  306. }
  307. /* NOTE: exactly one frame must be given (120000 bytes for NTSC,
  308. 144000 bytes for PAL - or twice those for 50Mbps) */
  309. static int dvvideo_decode_frame(AVCodecContext *avctx,
  310. void *data, int *got_frame,
  311. AVPacket *avpkt)
  312. {
  313. uint8_t *buf = avpkt->data;
  314. int buf_size = avpkt->size;
  315. DVVideoContext *s = avctx->priv_data;
  316. const uint8_t* vsc_pack;
  317. int apt, is16_9, ret;
  318. const DVprofile *sys;
  319. sys = avpriv_dv_frame_profile2(avctx, s->sys, buf, buf_size);
  320. if (!sys || buf_size < sys->frame_size) {
  321. av_log(avctx, AV_LOG_ERROR, "could not find dv frame profile\n");
  322. return -1; /* NOTE: we only accept several full frames */
  323. }
  324. if (sys != s->sys) {
  325. ret = ff_dv_init_dynamic_tables(s, sys);
  326. if (ret < 0) {
  327. av_log(avctx, AV_LOG_ERROR, "Error initializing the work tables.\n");
  328. return ret;
  329. }
  330. s->sys = sys;
  331. }
  332. s->frame = data;
  333. s->frame->key_frame = 1;
  334. s->frame->pict_type = AV_PICTURE_TYPE_I;
  335. avctx->pix_fmt = s->sys->pix_fmt;
  336. avctx->time_base = s->sys->time_base;
  337. ret = ff_set_dimensions(avctx, s->sys->width, s->sys->height);
  338. if (ret < 0)
  339. return ret;
  340. /* Determine the codec's sample_aspect ratio from the packet */
  341. vsc_pack = buf + 80*5 + 48 + 5;
  342. if ( *vsc_pack == dv_video_control ) {
  343. apt = buf[4] & 0x07;
  344. is16_9 = (vsc_pack && ((vsc_pack[2] & 0x07) == 0x02 || (!apt && (vsc_pack[2] & 0x07) == 0x07)));
  345. ff_set_sar(avctx, s->sys->sar[is16_9]);
  346. }
  347. if ((ret = ff_get_buffer(avctx, s->frame, 0)) < 0)
  348. return ret;
  349. s->frame->interlaced_frame = 1;
  350. s->frame->top_field_first = 0;
  351. /* Determine the codec's field order from the packet */
  352. if ( *vsc_pack == dv_video_control ) {
  353. s->frame->top_field_first = !(vsc_pack[3] & 0x40);
  354. }
  355. s->buf = buf;
  356. avctx->execute(avctx, dv_decode_video_segment, s->work_chunks, NULL,
  357. dv_work_pool_size(s->sys), sizeof(DVwork_chunk));
  358. emms_c();
  359. /* return image */
  360. *got_frame = 1;
  361. return s->sys->frame_size;
  362. }
  363. AVCodec ff_dvvideo_decoder = {
  364. .name = "dvvideo",
  365. .long_name = NULL_IF_CONFIG_SMALL("DV (Digital Video)"),
  366. .type = AVMEDIA_TYPE_VIDEO,
  367. .id = AV_CODEC_ID_DVVIDEO,
  368. .priv_data_size = sizeof(DVVideoContext),
  369. .init = dvvideo_decode_init,
  370. .decode = dvvideo_decode_frame,
  371. .capabilities = CODEC_CAP_DR1 | CODEC_CAP_SLICE_THREADS,
  372. .max_lowres = 3,
  373. };