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.

3484 lines
125KB

  1. /*
  2. * The simplest mpeg encoder (well, it was the simplest!)
  3. * Copyright (c) 2000,2001 Fabrice Bellard
  4. * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
  5. *
  6. * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
  7. *
  8. * This file is part of FFmpeg.
  9. *
  10. * FFmpeg is free software; you can redistribute it and/or
  11. * modify it under the terms of the GNU Lesser General Public
  12. * License as published by the Free Software Foundation; either
  13. * version 2.1 of the License, or (at your option) any later version.
  14. *
  15. * FFmpeg is distributed in the hope that it will be useful,
  16. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  18. * Lesser General Public License for more details.
  19. *
  20. * You should have received a copy of the GNU Lesser General Public
  21. * License along with FFmpeg; if not, write to the Free Software
  22. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23. */
  24. /**
  25. * @file
  26. * The simplest mpeg encoder (well, it was the simplest!).
  27. */
  28. #include "libavutil/attributes.h"
  29. #include "libavutil/avassert.h"
  30. #include "libavutil/imgutils.h"
  31. #include "libavutil/internal.h"
  32. #include "libavutil/motion_vector.h"
  33. #include "libavutil/timer.h"
  34. #include "avcodec.h"
  35. #include "blockdsp.h"
  36. #include "h264chroma.h"
  37. #include "idctdsp.h"
  38. #include "internal.h"
  39. #include "mathops.h"
  40. #include "mpegutils.h"
  41. #include "mpegvideo.h"
  42. #include "mjpegenc.h"
  43. #include "msmpeg4.h"
  44. #include "qpeldsp.h"
  45. #include "thread.h"
  46. #include <limits.h>
  47. static const uint8_t ff_default_chroma_qscale_table[32] = {
  48. // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
  49. 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
  50. 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31
  51. };
  52. const uint8_t ff_mpeg1_dc_scale_table[128] = {
  53. // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
  54. 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
  55. 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
  56. 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
  57. 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
  58. 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
  59. 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
  60. 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
  61. 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
  62. };
  63. static const uint8_t mpeg2_dc_scale_table1[128] = {
  64. // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
  65. 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
  66. 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
  67. 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
  68. 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
  69. 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
  70. 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
  71. 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
  72. 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
  73. };
  74. static const uint8_t mpeg2_dc_scale_table2[128] = {
  75. // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
  76. 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
  77. 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
  78. 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
  79. 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
  80. 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
  81. 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
  82. 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
  83. 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
  84. };
  85. static const uint8_t mpeg2_dc_scale_table3[128] = {
  86. // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
  87. 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
  88. 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
  89. 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
  90. 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
  91. 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
  92. 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
  93. 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
  94. 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
  95. };
  96. const uint8_t *const ff_mpeg2_dc_scale_table[4] = {
  97. ff_mpeg1_dc_scale_table,
  98. mpeg2_dc_scale_table1,
  99. mpeg2_dc_scale_table2,
  100. mpeg2_dc_scale_table3,
  101. };
  102. const uint8_t ff_alternate_horizontal_scan[64] = {
  103. 0, 1, 2, 3, 8, 9, 16, 17,
  104. 10, 11, 4, 5, 6, 7, 15, 14,
  105. 13, 12, 19, 18, 24, 25, 32, 33,
  106. 26, 27, 20, 21, 22, 23, 28, 29,
  107. 30, 31, 34, 35, 40, 41, 48, 49,
  108. 42, 43, 36, 37, 38, 39, 44, 45,
  109. 46, 47, 50, 51, 56, 57, 58, 59,
  110. 52, 53, 54, 55, 60, 61, 62, 63,
  111. };
  112. const uint8_t ff_alternate_vertical_scan[64] = {
  113. 0, 8, 16, 24, 1, 9, 2, 10,
  114. 17, 25, 32, 40, 48, 56, 57, 49,
  115. 41, 33, 26, 18, 3, 11, 4, 12,
  116. 19, 27, 34, 42, 50, 58, 35, 43,
  117. 51, 59, 20, 28, 5, 13, 6, 14,
  118. 21, 29, 36, 44, 52, 60, 37, 45,
  119. 53, 61, 22, 30, 7, 15, 23, 31,
  120. 38, 46, 54, 62, 39, 47, 55, 63,
  121. };
  122. static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
  123. int16_t *block, int n, int qscale)
  124. {
  125. int i, level, nCoeffs;
  126. const uint16_t *quant_matrix;
  127. nCoeffs= s->block_last_index[n];
  128. block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
  129. /* XXX: only mpeg1 */
  130. quant_matrix = s->intra_matrix;
  131. for(i=1;i<=nCoeffs;i++) {
  132. int j= s->intra_scantable.permutated[i];
  133. level = block[j];
  134. if (level) {
  135. if (level < 0) {
  136. level = -level;
  137. level = (int)(level * qscale * quant_matrix[j]) >> 3;
  138. level = (level - 1) | 1;
  139. level = -level;
  140. } else {
  141. level = (int)(level * qscale * quant_matrix[j]) >> 3;
  142. level = (level - 1) | 1;
  143. }
  144. block[j] = level;
  145. }
  146. }
  147. }
  148. static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
  149. int16_t *block, int n, int qscale)
  150. {
  151. int i, level, nCoeffs;
  152. const uint16_t *quant_matrix;
  153. nCoeffs= s->block_last_index[n];
  154. quant_matrix = s->inter_matrix;
  155. for(i=0; i<=nCoeffs; i++) {
  156. int j= s->intra_scantable.permutated[i];
  157. level = block[j];
  158. if (level) {
  159. if (level < 0) {
  160. level = -level;
  161. level = (((level << 1) + 1) * qscale *
  162. ((int) (quant_matrix[j]))) >> 4;
  163. level = (level - 1) | 1;
  164. level = -level;
  165. } else {
  166. level = (((level << 1) + 1) * qscale *
  167. ((int) (quant_matrix[j]))) >> 4;
  168. level = (level - 1) | 1;
  169. }
  170. block[j] = level;
  171. }
  172. }
  173. }
  174. static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
  175. int16_t *block, int n, int qscale)
  176. {
  177. int i, level, nCoeffs;
  178. const uint16_t *quant_matrix;
  179. if(s->alternate_scan) nCoeffs= 63;
  180. else nCoeffs= s->block_last_index[n];
  181. block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
  182. quant_matrix = s->intra_matrix;
  183. for(i=1;i<=nCoeffs;i++) {
  184. int j= s->intra_scantable.permutated[i];
  185. level = block[j];
  186. if (level) {
  187. if (level < 0) {
  188. level = -level;
  189. level = (int)(level * qscale * quant_matrix[j]) >> 3;
  190. level = -level;
  191. } else {
  192. level = (int)(level * qscale * quant_matrix[j]) >> 3;
  193. }
  194. block[j] = level;
  195. }
  196. }
  197. }
  198. static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
  199. int16_t *block, int n, int qscale)
  200. {
  201. int i, level, nCoeffs;
  202. const uint16_t *quant_matrix;
  203. int sum=-1;
  204. if(s->alternate_scan) nCoeffs= 63;
  205. else nCoeffs= s->block_last_index[n];
  206. block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
  207. sum += block[0];
  208. quant_matrix = s->intra_matrix;
  209. for(i=1;i<=nCoeffs;i++) {
  210. int j= s->intra_scantable.permutated[i];
  211. level = block[j];
  212. if (level) {
  213. if (level < 0) {
  214. level = -level;
  215. level = (int)(level * qscale * quant_matrix[j]) >> 3;
  216. level = -level;
  217. } else {
  218. level = (int)(level * qscale * quant_matrix[j]) >> 3;
  219. }
  220. block[j] = level;
  221. sum+=level;
  222. }
  223. }
  224. block[63]^=sum&1;
  225. }
  226. static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
  227. int16_t *block, int n, int qscale)
  228. {
  229. int i, level, nCoeffs;
  230. const uint16_t *quant_matrix;
  231. int sum=-1;
  232. if(s->alternate_scan) nCoeffs= 63;
  233. else nCoeffs= s->block_last_index[n];
  234. quant_matrix = s->inter_matrix;
  235. for(i=0; i<=nCoeffs; i++) {
  236. int j= s->intra_scantable.permutated[i];
  237. level = block[j];
  238. if (level) {
  239. if (level < 0) {
  240. level = -level;
  241. level = (((level << 1) + 1) * qscale *
  242. ((int) (quant_matrix[j]))) >> 4;
  243. level = -level;
  244. } else {
  245. level = (((level << 1) + 1) * qscale *
  246. ((int) (quant_matrix[j]))) >> 4;
  247. }
  248. block[j] = level;
  249. sum+=level;
  250. }
  251. }
  252. block[63]^=sum&1;
  253. }
  254. static void dct_unquantize_h263_intra_c(MpegEncContext *s,
  255. int16_t *block, int n, int qscale)
  256. {
  257. int i, level, qmul, qadd;
  258. int nCoeffs;
  259. av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
  260. qmul = qscale << 1;
  261. if (!s->h263_aic) {
  262. block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
  263. qadd = (qscale - 1) | 1;
  264. }else{
  265. qadd = 0;
  266. }
  267. if(s->ac_pred)
  268. nCoeffs=63;
  269. else
  270. nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
  271. for(i=1; i<=nCoeffs; i++) {
  272. level = block[i];
  273. if (level) {
  274. if (level < 0) {
  275. level = level * qmul - qadd;
  276. } else {
  277. level = level * qmul + qadd;
  278. }
  279. block[i] = level;
  280. }
  281. }
  282. }
  283. static void dct_unquantize_h263_inter_c(MpegEncContext *s,
  284. int16_t *block, int n, int qscale)
  285. {
  286. int i, level, qmul, qadd;
  287. int nCoeffs;
  288. av_assert2(s->block_last_index[n]>=0);
  289. qadd = (qscale - 1) | 1;
  290. qmul = qscale << 1;
  291. nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
  292. for(i=0; i<=nCoeffs; i++) {
  293. level = block[i];
  294. if (level) {
  295. if (level < 0) {
  296. level = level * qmul - qadd;
  297. } else {
  298. level = level * qmul + qadd;
  299. }
  300. block[i] = level;
  301. }
  302. }
  303. }
  304. static void mpeg_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
  305. int (*mv)[2][4][2],
  306. int mb_x, int mb_y, int mb_intra, int mb_skipped)
  307. {
  308. MpegEncContext *s = opaque;
  309. s->mv_dir = mv_dir;
  310. s->mv_type = mv_type;
  311. s->mb_intra = mb_intra;
  312. s->mb_skipped = mb_skipped;
  313. s->mb_x = mb_x;
  314. s->mb_y = mb_y;
  315. memcpy(s->mv, mv, sizeof(*mv));
  316. ff_init_block_index(s);
  317. ff_update_block_index(s);
  318. s->bdsp.clear_blocks(s->block[0]);
  319. s->dest[0] = s->current_picture.f->data[0] + (s->mb_y * 16 * s->linesize) + s->mb_x * 16;
  320. s->dest[1] = s->current_picture.f->data[1] + (s->mb_y * (16 >> s->chroma_y_shift) * s->uvlinesize) + s->mb_x * (16 >> s->chroma_x_shift);
  321. s->dest[2] = s->current_picture.f->data[2] + (s->mb_y * (16 >> s->chroma_y_shift) * s->uvlinesize) + s->mb_x * (16 >> s->chroma_x_shift);
  322. if (ref)
  323. av_log(s->avctx, AV_LOG_DEBUG, "Interlaced error concealment is not fully implemented\n");
  324. ff_mpv_decode_mb(s, s->block);
  325. }
  326. static void gray16(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
  327. {
  328. while(h--)
  329. memset(dst + h*linesize, 128, 16);
  330. }
  331. static void gray8(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
  332. {
  333. while(h--)
  334. memset(dst + h*linesize, 128, 8);
  335. }
  336. /* init common dct for both encoder and decoder */
  337. static av_cold int dct_init(MpegEncContext *s)
  338. {
  339. ff_blockdsp_init(&s->bdsp, s->avctx);
  340. ff_h264chroma_init(&s->h264chroma, 8); //for lowres
  341. ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
  342. ff_me_cmp_init(&s->mecc, s->avctx);
  343. ff_mpegvideodsp_init(&s->mdsp);
  344. ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
  345. if (s->avctx->debug & FF_DEBUG_NOMC) {
  346. int i;
  347. for (i=0; i<4; i++) {
  348. s->hdsp.avg_pixels_tab[0][i] = gray16;
  349. s->hdsp.put_pixels_tab[0][i] = gray16;
  350. s->hdsp.put_no_rnd_pixels_tab[0][i] = gray16;
  351. s->hdsp.avg_pixels_tab[1][i] = gray8;
  352. s->hdsp.put_pixels_tab[1][i] = gray8;
  353. s->hdsp.put_no_rnd_pixels_tab[1][i] = gray8;
  354. }
  355. }
  356. s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
  357. s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
  358. s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
  359. s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
  360. s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
  361. if (s->flags & CODEC_FLAG_BITEXACT)
  362. s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
  363. s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
  364. if (HAVE_INTRINSICS_NEON)
  365. ff_mpv_common_init_neon(s);
  366. if (ARCH_ALPHA)
  367. ff_mpv_common_init_axp(s);
  368. if (ARCH_ARM)
  369. ff_mpv_common_init_arm(s);
  370. if (ARCH_PPC)
  371. ff_mpv_common_init_ppc(s);
  372. if (ARCH_X86)
  373. ff_mpv_common_init_x86(s);
  374. return 0;
  375. }
  376. av_cold void ff_mpv_idct_init(MpegEncContext *s)
  377. {
  378. ff_idctdsp_init(&s->idsp, s->avctx);
  379. /* load & permutate scantables
  380. * note: only wmv uses different ones
  381. */
  382. if (s->alternate_scan) {
  383. ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_alternate_vertical_scan);
  384. ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_alternate_vertical_scan);
  385. } else {
  386. ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_zigzag_direct);
  387. ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_zigzag_direct);
  388. }
  389. ff_init_scantable(s->idsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
  390. ff_init_scantable(s->idsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);
  391. }
  392. static int frame_size_alloc(MpegEncContext *s, int linesize)
  393. {
  394. int alloc_size = FFALIGN(FFABS(linesize) + 64, 32);
  395. if (s->avctx->hwaccel || s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
  396. return 0;
  397. if (linesize < 24) {
  398. av_log(s->avctx, AV_LOG_ERROR, "Image too small, temporary buffers cannot function\n");
  399. return AVERROR_PATCHWELCOME;
  400. }
  401. // edge emu needs blocksize + filter length - 1
  402. // (= 17x17 for halfpel / 21x21 for h264)
  403. // VC1 computes luma and chroma simultaneously and needs 19X19 + 9x9
  404. // at uvlinesize. It supports only YUV420 so 24x24 is enough
  405. // linesize * interlaced * MBsize
  406. // we also use this buffer for encoding in encode_mb_internal() needig an additional 32 lines
  407. FF_ALLOCZ_ARRAY_OR_GOTO(s->avctx, s->edge_emu_buffer, alloc_size, 4 * 68,
  408. fail);
  409. FF_ALLOCZ_ARRAY_OR_GOTO(s->avctx, s->me.scratchpad, alloc_size, 4 * 16 * 2,
  410. fail)
  411. s->me.temp = s->me.scratchpad;
  412. s->rd_scratchpad = s->me.scratchpad;
  413. s->b_scratchpad = s->me.scratchpad;
  414. s->obmc_scratchpad = s->me.scratchpad + 16;
  415. return 0;
  416. fail:
  417. av_freep(&s->edge_emu_buffer);
  418. return AVERROR(ENOMEM);
  419. }
  420. /**
  421. * Allocate a frame buffer
  422. */
  423. static int alloc_frame_buffer(MpegEncContext *s, Picture *pic)
  424. {
  425. int edges_needed = av_codec_is_encoder(s->avctx->codec);
  426. int r, ret;
  427. pic->tf.f = pic->f;
  428. if (s->codec_id != AV_CODEC_ID_WMV3IMAGE &&
  429. s->codec_id != AV_CODEC_ID_VC1IMAGE &&
  430. s->codec_id != AV_CODEC_ID_MSS2) {
  431. if (edges_needed) {
  432. pic->f->width = s->avctx->width + 2 * EDGE_WIDTH;
  433. pic->f->height = s->avctx->height + 2 * EDGE_WIDTH;
  434. }
  435. r = ff_thread_get_buffer(s->avctx, &pic->tf,
  436. pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);
  437. } else {
  438. pic->f->width = s->avctx->width;
  439. pic->f->height = s->avctx->height;
  440. pic->f->format = s->avctx->pix_fmt;
  441. r = avcodec_default_get_buffer2(s->avctx, pic->f, 0);
  442. }
  443. if (r < 0 || !pic->f->buf[0]) {
  444. av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %p)\n",
  445. r, pic->f->data[0]);
  446. return -1;
  447. }
  448. if (edges_needed) {
  449. int i;
  450. for (i = 0; pic->f->data[i]; i++) {
  451. int offset = (EDGE_WIDTH >> (i ? s->chroma_y_shift : 0)) *
  452. pic->f->linesize[i] +
  453. (EDGE_WIDTH >> (i ? s->chroma_x_shift : 0));
  454. pic->f->data[i] += offset;
  455. }
  456. pic->f->width = s->avctx->width;
  457. pic->f->height = s->avctx->height;
  458. }
  459. if (s->avctx->hwaccel) {
  460. assert(!pic->hwaccel_picture_private);
  461. if (s->avctx->hwaccel->frame_priv_data_size) {
  462. pic->hwaccel_priv_buf = av_buffer_allocz(s->avctx->hwaccel->frame_priv_data_size);
  463. if (!pic->hwaccel_priv_buf) {
  464. av_log(s->avctx, AV_LOG_ERROR, "alloc_frame_buffer() failed (hwaccel private data allocation)\n");
  465. return -1;
  466. }
  467. pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
  468. }
  469. }
  470. if (s->linesize && (s->linesize != pic->f->linesize[0] ||
  471. s->uvlinesize != pic->f->linesize[1])) {
  472. av_log(s->avctx, AV_LOG_ERROR,
  473. "get_buffer() failed (stride changed)\n");
  474. ff_mpeg_unref_picture(s, pic);
  475. return -1;
  476. }
  477. if (pic->f->linesize[1] != pic->f->linesize[2]) {
  478. av_log(s->avctx, AV_LOG_ERROR,
  479. "get_buffer() failed (uv stride mismatch)\n");
  480. ff_mpeg_unref_picture(s, pic);
  481. return -1;
  482. }
  483. if (!s->edge_emu_buffer &&
  484. (ret = frame_size_alloc(s, pic->f->linesize[0])) < 0) {
  485. av_log(s->avctx, AV_LOG_ERROR,
  486. "get_buffer() failed to allocate context scratch buffers.\n");
  487. ff_mpeg_unref_picture(s, pic);
  488. return ret;
  489. }
  490. return 0;
  491. }
  492. void ff_free_picture_tables(Picture *pic)
  493. {
  494. int i;
  495. pic->alloc_mb_width =
  496. pic->alloc_mb_height = 0;
  497. av_buffer_unref(&pic->mb_var_buf);
  498. av_buffer_unref(&pic->mc_mb_var_buf);
  499. av_buffer_unref(&pic->mb_mean_buf);
  500. av_buffer_unref(&pic->mbskip_table_buf);
  501. av_buffer_unref(&pic->qscale_table_buf);
  502. av_buffer_unref(&pic->mb_type_buf);
  503. for (i = 0; i < 2; i++) {
  504. av_buffer_unref(&pic->motion_val_buf[i]);
  505. av_buffer_unref(&pic->ref_index_buf[i]);
  506. }
  507. }
  508. static int alloc_picture_tables(MpegEncContext *s, Picture *pic)
  509. {
  510. const int big_mb_num = s->mb_stride * (s->mb_height + 1) + 1;
  511. const int mb_array_size = s->mb_stride * s->mb_height;
  512. const int b8_array_size = s->b8_stride * s->mb_height * 2;
  513. int i;
  514. pic->mbskip_table_buf = av_buffer_allocz(mb_array_size + 2);
  515. pic->qscale_table_buf = av_buffer_allocz(big_mb_num + s->mb_stride);
  516. pic->mb_type_buf = av_buffer_allocz((big_mb_num + s->mb_stride) *
  517. sizeof(uint32_t));
  518. if (!pic->mbskip_table_buf || !pic->qscale_table_buf || !pic->mb_type_buf)
  519. return AVERROR(ENOMEM);
  520. if (s->encoding) {
  521. pic->mb_var_buf = av_buffer_allocz(mb_array_size * sizeof(int16_t));
  522. pic->mc_mb_var_buf = av_buffer_allocz(mb_array_size * sizeof(int16_t));
  523. pic->mb_mean_buf = av_buffer_allocz(mb_array_size);
  524. if (!pic->mb_var_buf || !pic->mc_mb_var_buf || !pic->mb_mean_buf)
  525. return AVERROR(ENOMEM);
  526. }
  527. if (s->out_format == FMT_H263 || s->encoding || s->avctx->debug_mv ||
  528. (s->avctx->flags2 & CODEC_FLAG2_EXPORT_MVS)) {
  529. int mv_size = 2 * (b8_array_size + 4) * sizeof(int16_t);
  530. int ref_index_size = 4 * mb_array_size;
  531. for (i = 0; mv_size && i < 2; i++) {
  532. pic->motion_val_buf[i] = av_buffer_allocz(mv_size);
  533. pic->ref_index_buf[i] = av_buffer_allocz(ref_index_size);
  534. if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
  535. return AVERROR(ENOMEM);
  536. }
  537. }
  538. pic->alloc_mb_width = s->mb_width;
  539. pic->alloc_mb_height = s->mb_height;
  540. return 0;
  541. }
  542. static int make_tables_writable(Picture *pic)
  543. {
  544. int ret, i;
  545. #define MAKE_WRITABLE(table) \
  546. do {\
  547. if (pic->table &&\
  548. (ret = av_buffer_make_writable(&pic->table)) < 0)\
  549. return ret;\
  550. } while (0)
  551. MAKE_WRITABLE(mb_var_buf);
  552. MAKE_WRITABLE(mc_mb_var_buf);
  553. MAKE_WRITABLE(mb_mean_buf);
  554. MAKE_WRITABLE(mbskip_table_buf);
  555. MAKE_WRITABLE(qscale_table_buf);
  556. MAKE_WRITABLE(mb_type_buf);
  557. for (i = 0; i < 2; i++) {
  558. MAKE_WRITABLE(motion_val_buf[i]);
  559. MAKE_WRITABLE(ref_index_buf[i]);
  560. }
  561. return 0;
  562. }
  563. /**
  564. * Allocate a Picture.
  565. * The pixels are allocated/set by calling get_buffer() if shared = 0
  566. */
  567. int ff_alloc_picture(MpegEncContext *s, Picture *pic, int shared)
  568. {
  569. int i, ret;
  570. if (pic->qscale_table_buf)
  571. if ( pic->alloc_mb_width != s->mb_width
  572. || pic->alloc_mb_height != s->mb_height)
  573. ff_free_picture_tables(pic);
  574. if (shared) {
  575. av_assert0(pic->f->data[0]);
  576. pic->shared = 1;
  577. } else {
  578. av_assert0(!pic->f->buf[0]);
  579. if (alloc_frame_buffer(s, pic) < 0)
  580. return -1;
  581. s->linesize = pic->f->linesize[0];
  582. s->uvlinesize = pic->f->linesize[1];
  583. }
  584. if (!pic->qscale_table_buf)
  585. ret = alloc_picture_tables(s, pic);
  586. else
  587. ret = make_tables_writable(pic);
  588. if (ret < 0)
  589. goto fail;
  590. if (s->encoding) {
  591. pic->mb_var = (uint16_t*)pic->mb_var_buf->data;
  592. pic->mc_mb_var = (uint16_t*)pic->mc_mb_var_buf->data;
  593. pic->mb_mean = pic->mb_mean_buf->data;
  594. }
  595. pic->mbskip_table = pic->mbskip_table_buf->data;
  596. pic->qscale_table = pic->qscale_table_buf->data + 2 * s->mb_stride + 1;
  597. pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * s->mb_stride + 1;
  598. if (pic->motion_val_buf[0]) {
  599. for (i = 0; i < 2; i++) {
  600. pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
  601. pic->ref_index[i] = pic->ref_index_buf[i]->data;
  602. }
  603. }
  604. return 0;
  605. fail:
  606. av_log(s->avctx, AV_LOG_ERROR, "Error allocating a picture.\n");
  607. ff_mpeg_unref_picture(s, pic);
  608. ff_free_picture_tables(pic);
  609. return AVERROR(ENOMEM);
  610. }
  611. /**
  612. * Deallocate a picture.
  613. */
  614. void ff_mpeg_unref_picture(MpegEncContext *s, Picture *pic)
  615. {
  616. int off = offsetof(Picture, mb_mean) + sizeof(pic->mb_mean);
  617. pic->tf.f = pic->f;
  618. /* WM Image / Screen codecs allocate internal buffers with different
  619. * dimensions / colorspaces; ignore user-defined callbacks for these. */
  620. if (s->codec_id != AV_CODEC_ID_WMV3IMAGE &&
  621. s->codec_id != AV_CODEC_ID_VC1IMAGE &&
  622. s->codec_id != AV_CODEC_ID_MSS2)
  623. ff_thread_release_buffer(s->avctx, &pic->tf);
  624. else if (pic->f)
  625. av_frame_unref(pic->f);
  626. av_buffer_unref(&pic->hwaccel_priv_buf);
  627. if (pic->needs_realloc)
  628. ff_free_picture_tables(pic);
  629. memset((uint8_t*)pic + off, 0, sizeof(*pic) - off);
  630. }
  631. static int update_picture_tables(Picture *dst, Picture *src)
  632. {
  633. int i;
  634. #define UPDATE_TABLE(table)\
  635. do {\
  636. if (src->table &&\
  637. (!dst->table || dst->table->buffer != src->table->buffer)) {\
  638. av_buffer_unref(&dst->table);\
  639. dst->table = av_buffer_ref(src->table);\
  640. if (!dst->table) {\
  641. ff_free_picture_tables(dst);\
  642. return AVERROR(ENOMEM);\
  643. }\
  644. }\
  645. } while (0)
  646. UPDATE_TABLE(mb_var_buf);
  647. UPDATE_TABLE(mc_mb_var_buf);
  648. UPDATE_TABLE(mb_mean_buf);
  649. UPDATE_TABLE(mbskip_table_buf);
  650. UPDATE_TABLE(qscale_table_buf);
  651. UPDATE_TABLE(mb_type_buf);
  652. for (i = 0; i < 2; i++) {
  653. UPDATE_TABLE(motion_val_buf[i]);
  654. UPDATE_TABLE(ref_index_buf[i]);
  655. }
  656. dst->mb_var = src->mb_var;
  657. dst->mc_mb_var = src->mc_mb_var;
  658. dst->mb_mean = src->mb_mean;
  659. dst->mbskip_table = src->mbskip_table;
  660. dst->qscale_table = src->qscale_table;
  661. dst->mb_type = src->mb_type;
  662. for (i = 0; i < 2; i++) {
  663. dst->motion_val[i] = src->motion_val[i];
  664. dst->ref_index[i] = src->ref_index[i];
  665. }
  666. dst->alloc_mb_width = src->alloc_mb_width;
  667. dst->alloc_mb_height = src->alloc_mb_height;
  668. return 0;
  669. }
  670. int ff_mpeg_ref_picture(MpegEncContext *s, Picture *dst, Picture *src)
  671. {
  672. int ret;
  673. av_assert0(!dst->f->buf[0]);
  674. av_assert0(src->f->buf[0]);
  675. src->tf.f = src->f;
  676. dst->tf.f = dst->f;
  677. ret = ff_thread_ref_frame(&dst->tf, &src->tf);
  678. if (ret < 0)
  679. goto fail;
  680. ret = update_picture_tables(dst, src);
  681. if (ret < 0)
  682. goto fail;
  683. if (src->hwaccel_picture_private) {
  684. dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
  685. if (!dst->hwaccel_priv_buf)
  686. goto fail;
  687. dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
  688. }
  689. dst->field_picture = src->field_picture;
  690. dst->mb_var_sum = src->mb_var_sum;
  691. dst->mc_mb_var_sum = src->mc_mb_var_sum;
  692. dst->b_frame_score = src->b_frame_score;
  693. dst->needs_realloc = src->needs_realloc;
  694. dst->reference = src->reference;
  695. dst->shared = src->shared;
  696. return 0;
  697. fail:
  698. ff_mpeg_unref_picture(s, dst);
  699. return ret;
  700. }
  701. static void exchange_uv(MpegEncContext *s)
  702. {
  703. int16_t (*tmp)[64];
  704. tmp = s->pblocks[4];
  705. s->pblocks[4] = s->pblocks[5];
  706. s->pblocks[5] = tmp;
  707. }
  708. static int init_duplicate_context(MpegEncContext *s)
  709. {
  710. int y_size = s->b8_stride * (2 * s->mb_height + 1);
  711. int c_size = s->mb_stride * (s->mb_height + 1);
  712. int yc_size = y_size + 2 * c_size;
  713. int i;
  714. if (s->mb_height & 1)
  715. yc_size += 2*s->b8_stride + 2*s->mb_stride;
  716. s->edge_emu_buffer =
  717. s->me.scratchpad =
  718. s->me.temp =
  719. s->rd_scratchpad =
  720. s->b_scratchpad =
  721. s->obmc_scratchpad = NULL;
  722. if (s->encoding) {
  723. FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map,
  724. ME_MAP_SIZE * sizeof(uint32_t), fail)
  725. FF_ALLOCZ_OR_GOTO(s->avctx, s->me.score_map,
  726. ME_MAP_SIZE * sizeof(uint32_t), fail)
  727. if (s->avctx->noise_reduction) {
  728. FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_error_sum,
  729. 2 * 64 * sizeof(int), fail)
  730. }
  731. }
  732. FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64 * 12 * 2 * sizeof(int16_t), fail)
  733. s->block = s->blocks[0];
  734. for (i = 0; i < 12; i++) {
  735. s->pblocks[i] = &s->block[i];
  736. }
  737. if (s->avctx->codec_tag == AV_RL32("VCR2"))
  738. exchange_uv(s);
  739. if (s->out_format == FMT_H263) {
  740. /* ac values */
  741. FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_val_base,
  742. yc_size * sizeof(int16_t) * 16, fail);
  743. s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
  744. s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
  745. s->ac_val[2] = s->ac_val[1] + c_size;
  746. }
  747. return 0;
  748. fail:
  749. return -1; // free() through ff_mpv_common_end()
  750. }
  751. static void free_duplicate_context(MpegEncContext *s)
  752. {
  753. if (!s)
  754. return;
  755. av_freep(&s->edge_emu_buffer);
  756. av_freep(&s->me.scratchpad);
  757. s->me.temp =
  758. s->rd_scratchpad =
  759. s->b_scratchpad =
  760. s->obmc_scratchpad = NULL;
  761. av_freep(&s->dct_error_sum);
  762. av_freep(&s->me.map);
  763. av_freep(&s->me.score_map);
  764. av_freep(&s->blocks);
  765. av_freep(&s->ac_val_base);
  766. s->block = NULL;
  767. }
  768. static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src)
  769. {
  770. #define COPY(a) bak->a = src->a
  771. COPY(edge_emu_buffer);
  772. COPY(me.scratchpad);
  773. COPY(me.temp);
  774. COPY(rd_scratchpad);
  775. COPY(b_scratchpad);
  776. COPY(obmc_scratchpad);
  777. COPY(me.map);
  778. COPY(me.score_map);
  779. COPY(blocks);
  780. COPY(block);
  781. COPY(start_mb_y);
  782. COPY(end_mb_y);
  783. COPY(me.map_generation);
  784. COPY(pb);
  785. COPY(dct_error_sum);
  786. COPY(dct_count[0]);
  787. COPY(dct_count[1]);
  788. COPY(ac_val_base);
  789. COPY(ac_val[0]);
  790. COPY(ac_val[1]);
  791. COPY(ac_val[2]);
  792. #undef COPY
  793. }
  794. int ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src)
  795. {
  796. MpegEncContext bak;
  797. int i, ret;
  798. // FIXME copy only needed parts
  799. // START_TIMER
  800. backup_duplicate_context(&bak, dst);
  801. memcpy(dst, src, sizeof(MpegEncContext));
  802. backup_duplicate_context(dst, &bak);
  803. for (i = 0; i < 12; i++) {
  804. dst->pblocks[i] = &dst->block[i];
  805. }
  806. if (dst->avctx->codec_tag == AV_RL32("VCR2"))
  807. exchange_uv(dst);
  808. if (!dst->edge_emu_buffer &&
  809. (ret = frame_size_alloc(dst, dst->linesize)) < 0) {
  810. av_log(dst->avctx, AV_LOG_ERROR, "failed to allocate context "
  811. "scratch buffers.\n");
  812. return ret;
  813. }
  814. // STOP_TIMER("update_duplicate_context")
  815. // about 10k cycles / 0.01 sec for 1000frames on 1ghz with 2 threads
  816. return 0;
  817. }
  818. int ff_mpeg_update_thread_context(AVCodecContext *dst,
  819. const AVCodecContext *src)
  820. {
  821. int i, ret;
  822. MpegEncContext *s = dst->priv_data, *s1 = src->priv_data;
  823. if (dst == src)
  824. return 0;
  825. av_assert0(s != s1);
  826. // FIXME can parameters change on I-frames?
  827. // in that case dst may need a reinit
  828. if (!s->context_initialized) {
  829. memcpy(s, s1, sizeof(MpegEncContext));
  830. s->avctx = dst;
  831. s->bitstream_buffer = NULL;
  832. s->bitstream_buffer_size = s->allocated_bitstream_buffer_size = 0;
  833. if (s1->context_initialized){
  834. // s->picture_range_start += MAX_PICTURE_COUNT;
  835. // s->picture_range_end += MAX_PICTURE_COUNT;
  836. ff_mpv_idct_init(s);
  837. if((ret = ff_mpv_common_init(s)) < 0){
  838. memset(s, 0, sizeof(MpegEncContext));
  839. s->avctx = dst;
  840. return ret;
  841. }
  842. }
  843. }
  844. if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
  845. s->context_reinit = 0;
  846. s->height = s1->height;
  847. s->width = s1->width;
  848. if ((ret = ff_mpv_common_frame_size_change(s)) < 0)
  849. return ret;
  850. }
  851. s->avctx->coded_height = s1->avctx->coded_height;
  852. s->avctx->coded_width = s1->avctx->coded_width;
  853. s->avctx->width = s1->avctx->width;
  854. s->avctx->height = s1->avctx->height;
  855. s->coded_picture_number = s1->coded_picture_number;
  856. s->picture_number = s1->picture_number;
  857. av_assert0(!s->picture || s->picture != s1->picture);
  858. if(s->picture)
  859. for (i = 0; i < MAX_PICTURE_COUNT; i++) {
  860. ff_mpeg_unref_picture(s, &s->picture[i]);
  861. if (s1->picture[i].f->buf[0] &&
  862. (ret = ff_mpeg_ref_picture(s, &s->picture[i], &s1->picture[i])) < 0)
  863. return ret;
  864. }
  865. #define UPDATE_PICTURE(pic)\
  866. do {\
  867. ff_mpeg_unref_picture(s, &s->pic);\
  868. if (s1->pic.f && s1->pic.f->buf[0])\
  869. ret = ff_mpeg_ref_picture(s, &s->pic, &s1->pic);\
  870. else\
  871. ret = update_picture_tables(&s->pic, &s1->pic);\
  872. if (ret < 0)\
  873. return ret;\
  874. } while (0)
  875. UPDATE_PICTURE(current_picture);
  876. UPDATE_PICTURE(last_picture);
  877. UPDATE_PICTURE(next_picture);
  878. s->last_picture_ptr = REBASE_PICTURE(s1->last_picture_ptr, s, s1);
  879. s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1);
  880. s->next_picture_ptr = REBASE_PICTURE(s1->next_picture_ptr, s, s1);
  881. // Error/bug resilience
  882. s->next_p_frame_damaged = s1->next_p_frame_damaged;
  883. s->workaround_bugs = s1->workaround_bugs;
  884. s->padding_bug_score = s1->padding_bug_score;
  885. // MPEG4 timing info
  886. memcpy(&s->last_time_base, &s1->last_time_base,
  887. (char *) &s1->pb_field_time + sizeof(s1->pb_field_time) -
  888. (char *) &s1->last_time_base);
  889. // B-frame info
  890. s->max_b_frames = s1->max_b_frames;
  891. s->low_delay = s1->low_delay;
  892. s->droppable = s1->droppable;
  893. // DivX handling (doesn't work)
  894. s->divx_packed = s1->divx_packed;
  895. if (s1->bitstream_buffer) {
  896. if (s1->bitstream_buffer_size +
  897. FF_INPUT_BUFFER_PADDING_SIZE > s->allocated_bitstream_buffer_size)
  898. av_fast_malloc(&s->bitstream_buffer,
  899. &s->allocated_bitstream_buffer_size,
  900. s1->allocated_bitstream_buffer_size);
  901. s->bitstream_buffer_size = s1->bitstream_buffer_size;
  902. memcpy(s->bitstream_buffer, s1->bitstream_buffer,
  903. s1->bitstream_buffer_size);
  904. memset(s->bitstream_buffer + s->bitstream_buffer_size, 0,
  905. FF_INPUT_BUFFER_PADDING_SIZE);
  906. }
  907. // linesize dependend scratch buffer allocation
  908. if (!s->edge_emu_buffer)
  909. if (s1->linesize) {
  910. if (frame_size_alloc(s, s1->linesize) < 0) {
  911. av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate context "
  912. "scratch buffers.\n");
  913. return AVERROR(ENOMEM);
  914. }
  915. } else {
  916. av_log(s->avctx, AV_LOG_ERROR, "Context scratch buffers could not "
  917. "be allocated due to unknown size.\n");
  918. }
  919. // MPEG2/interlacing info
  920. memcpy(&s->progressive_sequence, &s1->progressive_sequence,
  921. (char *) &s1->rtp_mode - (char *) &s1->progressive_sequence);
  922. if (!s1->first_field) {
  923. s->last_pict_type = s1->pict_type;
  924. if (s1->current_picture_ptr)
  925. s->last_lambda_for[s1->pict_type] = s1->current_picture_ptr->f->quality;
  926. }
  927. return 0;
  928. }
  929. /**
  930. * Set the given MpegEncContext to common defaults
  931. * (same for encoding and decoding).
  932. * The changed fields will not depend upon the
  933. * prior state of the MpegEncContext.
  934. */
  935. void ff_mpv_common_defaults(MpegEncContext *s)
  936. {
  937. s->y_dc_scale_table =
  938. s->c_dc_scale_table = ff_mpeg1_dc_scale_table;
  939. s->chroma_qscale_table = ff_default_chroma_qscale_table;
  940. s->progressive_frame = 1;
  941. s->progressive_sequence = 1;
  942. s->picture_structure = PICT_FRAME;
  943. s->coded_picture_number = 0;
  944. s->picture_number = 0;
  945. s->f_code = 1;
  946. s->b_code = 1;
  947. s->slice_context_count = 1;
  948. }
  949. /**
  950. * Set the given MpegEncContext to defaults for decoding.
  951. * the changed fields will not depend upon
  952. * the prior state of the MpegEncContext.
  953. */
  954. void ff_mpv_decode_defaults(MpegEncContext *s)
  955. {
  956. ff_mpv_common_defaults(s);
  957. }
  958. void ff_mpv_decode_init(MpegEncContext *s, AVCodecContext *avctx)
  959. {
  960. s->avctx = avctx;
  961. s->width = avctx->coded_width;
  962. s->height = avctx->coded_height;
  963. s->codec_id = avctx->codec->id;
  964. s->workaround_bugs = avctx->workaround_bugs;
  965. s->flags = avctx->flags;
  966. s->flags2 = avctx->flags2;
  967. /* convert fourcc to upper case */
  968. s->codec_tag = avpriv_toupper4(avctx->codec_tag);
  969. s->stream_codec_tag = avpriv_toupper4(avctx->stream_codec_tag);
  970. }
  971. static int init_er(MpegEncContext *s)
  972. {
  973. ERContext *er = &s->er;
  974. int mb_array_size = s->mb_height * s->mb_stride;
  975. int i;
  976. er->avctx = s->avctx;
  977. er->mecc = &s->mecc;
  978. er->mb_index2xy = s->mb_index2xy;
  979. er->mb_num = s->mb_num;
  980. er->mb_width = s->mb_width;
  981. er->mb_height = s->mb_height;
  982. er->mb_stride = s->mb_stride;
  983. er->b8_stride = s->b8_stride;
  984. er->er_temp_buffer = av_malloc(s->mb_height * s->mb_stride);
  985. er->error_status_table = av_mallocz(mb_array_size);
  986. if (!er->er_temp_buffer || !er->error_status_table)
  987. goto fail;
  988. er->mbskip_table = s->mbskip_table;
  989. er->mbintra_table = s->mbintra_table;
  990. for (i = 0; i < FF_ARRAY_ELEMS(s->dc_val); i++)
  991. er->dc_val[i] = s->dc_val[i];
  992. er->decode_mb = mpeg_er_decode_mb;
  993. er->opaque = s;
  994. return 0;
  995. fail:
  996. av_freep(&er->er_temp_buffer);
  997. av_freep(&er->error_status_table);
  998. return AVERROR(ENOMEM);
  999. }
  1000. /**
  1001. * Initialize and allocates MpegEncContext fields dependent on the resolution.
  1002. */
  1003. static int init_context_frame(MpegEncContext *s)
  1004. {
  1005. int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;
  1006. s->mb_width = (s->width + 15) / 16;
  1007. s->mb_stride = s->mb_width + 1;
  1008. s->b8_stride = s->mb_width * 2 + 1;
  1009. mb_array_size = s->mb_height * s->mb_stride;
  1010. mv_table_size = (s->mb_height + 2) * s->mb_stride + 1;
  1011. /* set default edge pos, will be overridden
  1012. * in decode_header if needed */
  1013. s->h_edge_pos = s->mb_width * 16;
  1014. s->v_edge_pos = s->mb_height * 16;
  1015. s->mb_num = s->mb_width * s->mb_height;
  1016. s->block_wrap[0] =
  1017. s->block_wrap[1] =
  1018. s->block_wrap[2] =
  1019. s->block_wrap[3] = s->b8_stride;
  1020. s->block_wrap[4] =
  1021. s->block_wrap[5] = s->mb_stride;
  1022. y_size = s->b8_stride * (2 * s->mb_height + 1);
  1023. c_size = s->mb_stride * (s->mb_height + 1);
  1024. yc_size = y_size + 2 * c_size;
  1025. if (s->mb_height & 1)
  1026. yc_size += 2*s->b8_stride + 2*s->mb_stride;
  1027. FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int), fail); // error ressilience code looks cleaner with this
  1028. for (y = 0; y < s->mb_height; y++)
  1029. for (x = 0; x < s->mb_width; x++)
  1030. s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
  1031. s->mb_index2xy[s->mb_height * s->mb_width] = (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
  1032. if (s->encoding) {
  1033. /* Allocate MV tables */
  1034. FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
  1035. FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
  1036. FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
  1037. FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
  1038. FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
  1039. FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base, mv_table_size * 2 * sizeof(int16_t), fail)
  1040. s->p_mv_table = s->p_mv_table_base + s->mb_stride + 1;
  1041. s->b_forw_mv_table = s->b_forw_mv_table_base + s->mb_stride + 1;
  1042. s->b_back_mv_table = s->b_back_mv_table_base + s->mb_stride + 1;
  1043. s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
  1044. s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base + s->mb_stride + 1;
  1045. s->b_direct_mv_table = s->b_direct_mv_table_base + s->mb_stride + 1;
  1046. /* Allocate MB type table */
  1047. FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size * sizeof(uint16_t), fail) // needed for encoding
  1048. FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
  1049. FF_ALLOC_OR_GOTO(s->avctx, s->cplx_tab,
  1050. mb_array_size * sizeof(float), fail);
  1051. FF_ALLOC_OR_GOTO(s->avctx, s->bits_tab,
  1052. mb_array_size * sizeof(float), fail);
  1053. }
  1054. if (s->codec_id == AV_CODEC_ID_MPEG4 ||
  1055. (s->flags & CODEC_FLAG_INTERLACED_ME)) {
  1056. /* interlaced direct mode decoding tables */
  1057. for (i = 0; i < 2; i++) {
  1058. int j, k;
  1059. for (j = 0; j < 2; j++) {
  1060. for (k = 0; k < 2; k++) {
  1061. FF_ALLOCZ_OR_GOTO(s->avctx,
  1062. s->b_field_mv_table_base[i][j][k],
  1063. mv_table_size * 2 * sizeof(int16_t),
  1064. fail);
  1065. s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
  1066. s->mb_stride + 1;
  1067. }
  1068. FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
  1069. FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
  1070. s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1;
  1071. }
  1072. FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
  1073. }
  1074. }
  1075. if (s->out_format == FMT_H263) {
  1076. /* cbp values */
  1077. FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size + (s->mb_height&1)*2*s->b8_stride, fail);
  1078. s->coded_block = s->coded_block_base + s->b8_stride + 1;
  1079. /* cbp, ac_pred, pred_dir */
  1080. FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table , mb_array_size * sizeof(uint8_t), fail);
  1081. FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail);
  1082. }
  1083. if (s->h263_pred || s->h263_plus || !s->encoding) {
  1084. /* dc values */
  1085. // MN: we need these for error resilience of intra-frames
  1086. FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
  1087. s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
  1088. s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
  1089. s->dc_val[2] = s->dc_val[1] + c_size;
  1090. for (i = 0; i < yc_size; i++)
  1091. s->dc_val_base[i] = 1024;
  1092. }
  1093. /* which mb is a intra block */
  1094. FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
  1095. memset(s->mbintra_table, 1, mb_array_size);
  1096. /* init macroblock skip table */
  1097. FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail);
  1098. // Note the + 1 is for a quicker mpeg4 slice_end detection
  1099. return init_er(s);
  1100. fail:
  1101. return AVERROR(ENOMEM);
  1102. }
  1103. static void clear_context(MpegEncContext *s)
  1104. {
  1105. int i, j, k;
  1106. memset(&s->next_picture, 0, sizeof(s->next_picture));
  1107. memset(&s->last_picture, 0, sizeof(s->last_picture));
  1108. memset(&s->current_picture, 0, sizeof(s->current_picture));
  1109. memset(&s->new_picture, 0, sizeof(s->new_picture));
  1110. memset(s->thread_context, 0, sizeof(s->thread_context));
  1111. s->me.map = NULL;
  1112. s->me.score_map = NULL;
  1113. s->dct_error_sum = NULL;
  1114. s->block = NULL;
  1115. s->blocks = NULL;
  1116. memset(s->pblocks, 0, sizeof(s->pblocks));
  1117. s->ac_val_base = NULL;
  1118. s->ac_val[0] =
  1119. s->ac_val[1] =
  1120. s->ac_val[2] =NULL;
  1121. s->edge_emu_buffer = NULL;
  1122. s->me.scratchpad = NULL;
  1123. s->me.temp =
  1124. s->rd_scratchpad =
  1125. s->b_scratchpad =
  1126. s->obmc_scratchpad = NULL;
  1127. s->bitstream_buffer = NULL;
  1128. s->allocated_bitstream_buffer_size = 0;
  1129. s->picture = NULL;
  1130. s->mb_type = NULL;
  1131. s->p_mv_table_base = NULL;
  1132. s->b_forw_mv_table_base = NULL;
  1133. s->b_back_mv_table_base = NULL;
  1134. s->b_bidir_forw_mv_table_base = NULL;
  1135. s->b_bidir_back_mv_table_base = NULL;
  1136. s->b_direct_mv_table_base = NULL;
  1137. s->p_mv_table = NULL;
  1138. s->b_forw_mv_table = NULL;
  1139. s->b_back_mv_table = NULL;
  1140. s->b_bidir_forw_mv_table = NULL;
  1141. s->b_bidir_back_mv_table = NULL;
  1142. s->b_direct_mv_table = NULL;
  1143. for (i = 0; i < 2; i++) {
  1144. for (j = 0; j < 2; j++) {
  1145. for (k = 0; k < 2; k++) {
  1146. s->b_field_mv_table_base[i][j][k] = NULL;
  1147. s->b_field_mv_table[i][j][k] = NULL;
  1148. }
  1149. s->b_field_select_table[i][j] = NULL;
  1150. s->p_field_mv_table_base[i][j] = NULL;
  1151. s->p_field_mv_table[i][j] = NULL;
  1152. }
  1153. s->p_field_select_table[i] = NULL;
  1154. }
  1155. s->dc_val_base = NULL;
  1156. s->coded_block_base = NULL;
  1157. s->mbintra_table = NULL;
  1158. s->cbp_table = NULL;
  1159. s->pred_dir_table = NULL;
  1160. s->mbskip_table = NULL;
  1161. s->er.error_status_table = NULL;
  1162. s->er.er_temp_buffer = NULL;
  1163. s->mb_index2xy = NULL;
  1164. s->lambda_table = NULL;
  1165. s->cplx_tab = NULL;
  1166. s->bits_tab = NULL;
  1167. }
  1168. /**
  1169. * init common structure for both encoder and decoder.
  1170. * this assumes that some variables like width/height are already set
  1171. */
  1172. av_cold int ff_mpv_common_init(MpegEncContext *s)
  1173. {
  1174. int i;
  1175. int nb_slices = (HAVE_THREADS &&
  1176. s->avctx->active_thread_type & FF_THREAD_SLICE) ?
  1177. s->avctx->thread_count : 1;
  1178. clear_context(s);
  1179. if (s->encoding && s->avctx->slices)
  1180. nb_slices = s->avctx->slices;
  1181. if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
  1182. s->mb_height = (s->height + 31) / 32 * 2;
  1183. else
  1184. s->mb_height = (s->height + 15) / 16;
  1185. if (s->avctx->pix_fmt == AV_PIX_FMT_NONE) {
  1186. av_log(s->avctx, AV_LOG_ERROR,
  1187. "decoding to AV_PIX_FMT_NONE is not supported.\n");
  1188. return -1;
  1189. }
  1190. if (nb_slices > MAX_THREADS || (nb_slices > s->mb_height && s->mb_height)) {
  1191. int max_slices;
  1192. if (s->mb_height)
  1193. max_slices = FFMIN(MAX_THREADS, s->mb_height);
  1194. else
  1195. max_slices = MAX_THREADS;
  1196. av_log(s->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
  1197. " reducing to %d\n", nb_slices, max_slices);
  1198. nb_slices = max_slices;
  1199. }
  1200. if ((s->width || s->height) &&
  1201. av_image_check_size(s->width, s->height, 0, s->avctx))
  1202. return -1;
  1203. dct_init(s);
  1204. s->flags = s->avctx->flags;
  1205. s->flags2 = s->avctx->flags2;
  1206. /* set chroma shifts */
  1207. avcodec_get_chroma_sub_sample(s->avctx->pix_fmt,
  1208. &s->chroma_x_shift,
  1209. &s->chroma_y_shift);
  1210. FF_ALLOCZ_OR_GOTO(s->avctx, s->picture,
  1211. MAX_PICTURE_COUNT * sizeof(Picture), fail);
  1212. for (i = 0; i < MAX_PICTURE_COUNT; i++) {
  1213. s->picture[i].f = av_frame_alloc();
  1214. if (!s->picture[i].f)
  1215. goto fail;
  1216. }
  1217. s->next_picture.f = av_frame_alloc();
  1218. if (!s->next_picture.f)
  1219. goto fail;
  1220. s->last_picture.f = av_frame_alloc();
  1221. if (!s->last_picture.f)
  1222. goto fail;
  1223. s->current_picture.f = av_frame_alloc();
  1224. if (!s->current_picture.f)
  1225. goto fail;
  1226. s->new_picture.f = av_frame_alloc();
  1227. if (!s->new_picture.f)
  1228. goto fail;
  1229. if (init_context_frame(s))
  1230. goto fail;
  1231. s->parse_context.state = -1;
  1232. s->context_initialized = 1;
  1233. s->thread_context[0] = s;
  1234. // if (s->width && s->height) {
  1235. if (nb_slices > 1) {
  1236. for (i = 1; i < nb_slices; i++) {
  1237. s->thread_context[i] = av_malloc(sizeof(MpegEncContext));
  1238. memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
  1239. }
  1240. for (i = 0; i < nb_slices; i++) {
  1241. if (init_duplicate_context(s->thread_context[i]) < 0)
  1242. goto fail;
  1243. s->thread_context[i]->start_mb_y =
  1244. (s->mb_height * (i) + nb_slices / 2) / nb_slices;
  1245. s->thread_context[i]->end_mb_y =
  1246. (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
  1247. }
  1248. } else {
  1249. if (init_duplicate_context(s) < 0)
  1250. goto fail;
  1251. s->start_mb_y = 0;
  1252. s->end_mb_y = s->mb_height;
  1253. }
  1254. s->slice_context_count = nb_slices;
  1255. // }
  1256. return 0;
  1257. fail:
  1258. ff_mpv_common_end(s);
  1259. return -1;
  1260. }
  1261. /**
  1262. * Frees and resets MpegEncContext fields depending on the resolution.
  1263. * Is used during resolution changes to avoid a full reinitialization of the
  1264. * codec.
  1265. */
  1266. static void free_context_frame(MpegEncContext *s)
  1267. {
  1268. int i, j, k;
  1269. av_freep(&s->mb_type);
  1270. av_freep(&s->p_mv_table_base);
  1271. av_freep(&s->b_forw_mv_table_base);
  1272. av_freep(&s->b_back_mv_table_base);
  1273. av_freep(&s->b_bidir_forw_mv_table_base);
  1274. av_freep(&s->b_bidir_back_mv_table_base);
  1275. av_freep(&s->b_direct_mv_table_base);
  1276. s->p_mv_table = NULL;
  1277. s->b_forw_mv_table = NULL;
  1278. s->b_back_mv_table = NULL;
  1279. s->b_bidir_forw_mv_table = NULL;
  1280. s->b_bidir_back_mv_table = NULL;
  1281. s->b_direct_mv_table = NULL;
  1282. for (i = 0; i < 2; i++) {
  1283. for (j = 0; j < 2; j++) {
  1284. for (k = 0; k < 2; k++) {
  1285. av_freep(&s->b_field_mv_table_base[i][j][k]);
  1286. s->b_field_mv_table[i][j][k] = NULL;
  1287. }
  1288. av_freep(&s->b_field_select_table[i][j]);
  1289. av_freep(&s->p_field_mv_table_base[i][j]);
  1290. s->p_field_mv_table[i][j] = NULL;
  1291. }
  1292. av_freep(&s->p_field_select_table[i]);
  1293. }
  1294. av_freep(&s->dc_val_base);
  1295. av_freep(&s->coded_block_base);
  1296. av_freep(&s->mbintra_table);
  1297. av_freep(&s->cbp_table);
  1298. av_freep(&s->pred_dir_table);
  1299. av_freep(&s->mbskip_table);
  1300. av_freep(&s->er.error_status_table);
  1301. av_freep(&s->er.er_temp_buffer);
  1302. av_freep(&s->mb_index2xy);
  1303. av_freep(&s->lambda_table);
  1304. av_freep(&s->cplx_tab);
  1305. av_freep(&s->bits_tab);
  1306. s->linesize = s->uvlinesize = 0;
  1307. }
  1308. int ff_mpv_common_frame_size_change(MpegEncContext *s)
  1309. {
  1310. int i, err = 0;
  1311. if (!s->context_initialized)
  1312. return AVERROR(EINVAL);
  1313. if (s->slice_context_count > 1) {
  1314. for (i = 0; i < s->slice_context_count; i++) {
  1315. free_duplicate_context(s->thread_context[i]);
  1316. }
  1317. for (i = 1; i < s->slice_context_count; i++) {
  1318. av_freep(&s->thread_context[i]);
  1319. }
  1320. } else
  1321. free_duplicate_context(s);
  1322. free_context_frame(s);
  1323. if (s->picture)
  1324. for (i = 0; i < MAX_PICTURE_COUNT; i++) {
  1325. s->picture[i].needs_realloc = 1;
  1326. }
  1327. s->last_picture_ptr =
  1328. s->next_picture_ptr =
  1329. s->current_picture_ptr = NULL;
  1330. // init
  1331. if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
  1332. s->mb_height = (s->height + 31) / 32 * 2;
  1333. else
  1334. s->mb_height = (s->height + 15) / 16;
  1335. if ((s->width || s->height) &&
  1336. (err = av_image_check_size(s->width, s->height, 0, s->avctx)) < 0)
  1337. goto fail;
  1338. if ((err = init_context_frame(s)))
  1339. goto fail;
  1340. s->thread_context[0] = s;
  1341. if (s->width && s->height) {
  1342. int nb_slices = s->slice_context_count;
  1343. if (nb_slices > 1) {
  1344. for (i = 1; i < nb_slices; i++) {
  1345. s->thread_context[i] = av_malloc(sizeof(MpegEncContext));
  1346. memcpy(s->thread_context[i], s, sizeof(MpegEncContext));
  1347. }
  1348. for (i = 0; i < nb_slices; i++) {
  1349. if ((err = init_duplicate_context(s->thread_context[i])) < 0)
  1350. goto fail;
  1351. s->thread_context[i]->start_mb_y =
  1352. (s->mb_height * (i) + nb_slices / 2) / nb_slices;
  1353. s->thread_context[i]->end_mb_y =
  1354. (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
  1355. }
  1356. } else {
  1357. err = init_duplicate_context(s);
  1358. if (err < 0)
  1359. goto fail;
  1360. s->start_mb_y = 0;
  1361. s->end_mb_y = s->mb_height;
  1362. }
  1363. s->slice_context_count = nb_slices;
  1364. }
  1365. return 0;
  1366. fail:
  1367. ff_mpv_common_end(s);
  1368. return err;
  1369. }
  1370. /* init common structure for both encoder and decoder */
  1371. void ff_mpv_common_end(MpegEncContext *s)
  1372. {
  1373. int i;
  1374. if (s->slice_context_count > 1) {
  1375. for (i = 0; i < s->slice_context_count; i++) {
  1376. free_duplicate_context(s->thread_context[i]);
  1377. }
  1378. for (i = 1; i < s->slice_context_count; i++) {
  1379. av_freep(&s->thread_context[i]);
  1380. }
  1381. s->slice_context_count = 1;
  1382. } else free_duplicate_context(s);
  1383. av_freep(&s->parse_context.buffer);
  1384. s->parse_context.buffer_size = 0;
  1385. av_freep(&s->bitstream_buffer);
  1386. s->allocated_bitstream_buffer_size = 0;
  1387. if (s->picture) {
  1388. for (i = 0; i < MAX_PICTURE_COUNT; i++) {
  1389. ff_free_picture_tables(&s->picture[i]);
  1390. ff_mpeg_unref_picture(s, &s->picture[i]);
  1391. av_frame_free(&s->picture[i].f);
  1392. }
  1393. }
  1394. av_freep(&s->picture);
  1395. ff_free_picture_tables(&s->last_picture);
  1396. ff_mpeg_unref_picture(s, &s->last_picture);
  1397. av_frame_free(&s->last_picture.f);
  1398. ff_free_picture_tables(&s->current_picture);
  1399. ff_mpeg_unref_picture(s, &s->current_picture);
  1400. av_frame_free(&s->current_picture.f);
  1401. ff_free_picture_tables(&s->next_picture);
  1402. ff_mpeg_unref_picture(s, &s->next_picture);
  1403. av_frame_free(&s->next_picture.f);
  1404. ff_free_picture_tables(&s->new_picture);
  1405. ff_mpeg_unref_picture(s, &s->new_picture);
  1406. av_frame_free(&s->new_picture.f);
  1407. free_context_frame(s);
  1408. s->context_initialized = 0;
  1409. s->last_picture_ptr =
  1410. s->next_picture_ptr =
  1411. s->current_picture_ptr = NULL;
  1412. s->linesize = s->uvlinesize = 0;
  1413. }
  1414. av_cold void ff_init_rl(RLTable *rl,
  1415. uint8_t static_store[2][2 * MAX_RUN + MAX_LEVEL + 3])
  1416. {
  1417. int8_t max_level[MAX_RUN + 1], max_run[MAX_LEVEL + 1];
  1418. uint8_t index_run[MAX_RUN + 1];
  1419. int last, run, level, start, end, i;
  1420. /* If table is static, we can quit if rl->max_level[0] is not NULL */
  1421. if (static_store && rl->max_level[0])
  1422. return;
  1423. /* compute max_level[], max_run[] and index_run[] */
  1424. for (last = 0; last < 2; last++) {
  1425. if (last == 0) {
  1426. start = 0;
  1427. end = rl->last;
  1428. } else {
  1429. start = rl->last;
  1430. end = rl->n;
  1431. }
  1432. memset(max_level, 0, MAX_RUN + 1);
  1433. memset(max_run, 0, MAX_LEVEL + 1);
  1434. memset(index_run, rl->n, MAX_RUN + 1);
  1435. for (i = start; i < end; i++) {
  1436. run = rl->table_run[i];
  1437. level = rl->table_level[i];
  1438. if (index_run[run] == rl->n)
  1439. index_run[run] = i;
  1440. if (level > max_level[run])
  1441. max_level[run] = level;
  1442. if (run > max_run[level])
  1443. max_run[level] = run;
  1444. }
  1445. if (static_store)
  1446. rl->max_level[last] = static_store[last];
  1447. else
  1448. rl->max_level[last] = av_malloc(MAX_RUN + 1);
  1449. memcpy(rl->max_level[last], max_level, MAX_RUN + 1);
  1450. if (static_store)
  1451. rl->max_run[last] = static_store[last] + MAX_RUN + 1;
  1452. else
  1453. rl->max_run[last] = av_malloc(MAX_LEVEL + 1);
  1454. memcpy(rl->max_run[last], max_run, MAX_LEVEL + 1);
  1455. if (static_store)
  1456. rl->index_run[last] = static_store[last] + MAX_RUN + MAX_LEVEL + 2;
  1457. else
  1458. rl->index_run[last] = av_malloc(MAX_RUN + 1);
  1459. memcpy(rl->index_run[last], index_run, MAX_RUN + 1);
  1460. }
  1461. }
  1462. av_cold void ff_init_vlc_rl(RLTable *rl, unsigned static_size)
  1463. {
  1464. int i, q;
  1465. VLC_TYPE table[1500][2] = {{0}};
  1466. VLC vlc = { .table = table, .table_allocated = static_size };
  1467. av_assert0(static_size <= FF_ARRAY_ELEMS(table));
  1468. init_vlc(&vlc, 9, rl->n + 1, &rl->table_vlc[0][1], 4, 2, &rl->table_vlc[0][0], 4, 2, INIT_VLC_USE_NEW_STATIC);
  1469. for (q = 0; q < 32; q++) {
  1470. int qmul = q * 2;
  1471. int qadd = (q - 1) | 1;
  1472. if (q == 0) {
  1473. qmul = 1;
  1474. qadd = 0;
  1475. }
  1476. for (i = 0; i < vlc.table_size; i++) {
  1477. int code = vlc.table[i][0];
  1478. int len = vlc.table[i][1];
  1479. int level, run;
  1480. if (len == 0) { // illegal code
  1481. run = 66;
  1482. level = MAX_LEVEL;
  1483. } else if (len < 0) { // more bits needed
  1484. run = 0;
  1485. level = code;
  1486. } else {
  1487. if (code == rl->n) { // esc
  1488. run = 66;
  1489. level = 0;
  1490. } else {
  1491. run = rl->table_run[code] + 1;
  1492. level = rl->table_level[code] * qmul + qadd;
  1493. if (code >= rl->last) run += 192;
  1494. }
  1495. }
  1496. rl->rl_vlc[q][i].len = len;
  1497. rl->rl_vlc[q][i].level = level;
  1498. rl->rl_vlc[q][i].run = run;
  1499. }
  1500. }
  1501. }
  1502. static void release_unused_pictures(MpegEncContext *s)
  1503. {
  1504. int i;
  1505. /* release non reference frames */
  1506. for (i = 0; i < MAX_PICTURE_COUNT; i++) {
  1507. if (!s->picture[i].reference)
  1508. ff_mpeg_unref_picture(s, &s->picture[i]);
  1509. }
  1510. }
  1511. static inline int pic_is_unused(MpegEncContext *s, Picture *pic)
  1512. {
  1513. if (pic == s->last_picture_ptr)
  1514. return 0;
  1515. if (!pic->f->buf[0])
  1516. return 1;
  1517. if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
  1518. return 1;
  1519. return 0;
  1520. }
  1521. static int find_unused_picture(MpegEncContext *s, int shared)
  1522. {
  1523. int i;
  1524. if (shared) {
  1525. for (i = 0; i < MAX_PICTURE_COUNT; i++) {
  1526. if (!s->picture[i].f->buf[0] && &s->picture[i] != s->last_picture_ptr)
  1527. return i;
  1528. }
  1529. } else {
  1530. for (i = 0; i < MAX_PICTURE_COUNT; i++) {
  1531. if (pic_is_unused(s, &s->picture[i]))
  1532. return i;
  1533. }
  1534. }
  1535. av_log(s->avctx, AV_LOG_FATAL,
  1536. "Internal error, picture buffer overflow\n");
  1537. /* We could return -1, but the codec would crash trying to draw into a
  1538. * non-existing frame anyway. This is safer than waiting for a random crash.
  1539. * Also the return of this is never useful, an encoder must only allocate
  1540. * as much as allowed in the specification. This has no relationship to how
  1541. * much libavcodec could allocate (and MAX_PICTURE_COUNT is always large
  1542. * enough for such valid streams).
  1543. * Plus, a decoder has to check stream validity and remove frames if too
  1544. * many reference frames are around. Waiting for "OOM" is not correct at
  1545. * all. Similarly, missing reference frames have to be replaced by
  1546. * interpolated/MC frames, anything else is a bug in the codec ...
  1547. */
  1548. abort();
  1549. return -1;
  1550. }
  1551. int ff_find_unused_picture(MpegEncContext *s, int shared)
  1552. {
  1553. int ret = find_unused_picture(s, shared);
  1554. if (ret >= 0 && ret < MAX_PICTURE_COUNT) {
  1555. if (s->picture[ret].needs_realloc) {
  1556. s->picture[ret].needs_realloc = 0;
  1557. ff_free_picture_tables(&s->picture[ret]);
  1558. ff_mpeg_unref_picture(s, &s->picture[ret]);
  1559. }
  1560. }
  1561. return ret;
  1562. }
  1563. static void gray_frame(AVFrame *frame)
  1564. {
  1565. int i, h_chroma_shift, v_chroma_shift;
  1566. av_pix_fmt_get_chroma_sub_sample(frame->format, &h_chroma_shift, &v_chroma_shift);
  1567. for(i=0; i<frame->height; i++)
  1568. memset(frame->data[0] + frame->linesize[0]*i, 0x80, frame->width);
  1569. for(i=0; i<FF_CEIL_RSHIFT(frame->height, v_chroma_shift); i++) {
  1570. memset(frame->data[1] + frame->linesize[1]*i,
  1571. 0x80, FF_CEIL_RSHIFT(frame->width, h_chroma_shift));
  1572. memset(frame->data[2] + frame->linesize[2]*i,
  1573. 0x80, FF_CEIL_RSHIFT(frame->width, h_chroma_shift));
  1574. }
  1575. }
  1576. /**
  1577. * generic function called after decoding
  1578. * the header and before a frame is decoded.
  1579. */
  1580. int ff_mpv_frame_start(MpegEncContext *s, AVCodecContext *avctx)
  1581. {
  1582. int i, ret;
  1583. Picture *pic;
  1584. s->mb_skipped = 0;
  1585. if (!ff_thread_can_start_frame(avctx)) {
  1586. av_log(avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
  1587. return -1;
  1588. }
  1589. /* mark & release old frames */
  1590. if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
  1591. s->last_picture_ptr != s->next_picture_ptr &&
  1592. s->last_picture_ptr->f->buf[0]) {
  1593. ff_mpeg_unref_picture(s, s->last_picture_ptr);
  1594. }
  1595. /* release forgotten pictures */
  1596. /* if (mpeg124/h263) */
  1597. for (i = 0; i < MAX_PICTURE_COUNT; i++) {
  1598. if (&s->picture[i] != s->last_picture_ptr &&
  1599. &s->picture[i] != s->next_picture_ptr &&
  1600. s->picture[i].reference && !s->picture[i].needs_realloc) {
  1601. if (!(avctx->active_thread_type & FF_THREAD_FRAME))
  1602. av_log(avctx, AV_LOG_ERROR,
  1603. "releasing zombie picture\n");
  1604. ff_mpeg_unref_picture(s, &s->picture[i]);
  1605. }
  1606. }
  1607. ff_mpeg_unref_picture(s, &s->current_picture);
  1608. release_unused_pictures(s);
  1609. if (s->current_picture_ptr && !s->current_picture_ptr->f->buf[0]) {
  1610. // we already have a unused image
  1611. // (maybe it was set before reading the header)
  1612. pic = s->current_picture_ptr;
  1613. } else {
  1614. i = ff_find_unused_picture(s, 0);
  1615. if (i < 0) {
  1616. av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
  1617. return i;
  1618. }
  1619. pic = &s->picture[i];
  1620. }
  1621. pic->reference = 0;
  1622. if (!s->droppable) {
  1623. if (s->pict_type != AV_PICTURE_TYPE_B)
  1624. pic->reference = 3;
  1625. }
  1626. pic->f->coded_picture_number = s->coded_picture_number++;
  1627. if (ff_alloc_picture(s, pic, 0) < 0)
  1628. return -1;
  1629. s->current_picture_ptr = pic;
  1630. // FIXME use only the vars from current_pic
  1631. s->current_picture_ptr->f->top_field_first = s->top_field_first;
  1632. if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
  1633. s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
  1634. if (s->picture_structure != PICT_FRAME)
  1635. s->current_picture_ptr->f->top_field_first =
  1636. (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
  1637. }
  1638. s->current_picture_ptr->f->interlaced_frame = !s->progressive_frame &&
  1639. !s->progressive_sequence;
  1640. s->current_picture_ptr->field_picture = s->picture_structure != PICT_FRAME;
  1641. s->current_picture_ptr->f->pict_type = s->pict_type;
  1642. // if (s->flags && CODEC_FLAG_QSCALE)
  1643. // s->current_picture_ptr->quality = s->new_picture_ptr->quality;
  1644. s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
  1645. if ((ret = ff_mpeg_ref_picture(s, &s->current_picture,
  1646. s->current_picture_ptr)) < 0)
  1647. return ret;
  1648. if (s->pict_type != AV_PICTURE_TYPE_B) {
  1649. s->last_picture_ptr = s->next_picture_ptr;
  1650. if (!s->droppable)
  1651. s->next_picture_ptr = s->current_picture_ptr;
  1652. }
  1653. av_dlog(s->avctx, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n",
  1654. s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
  1655. s->last_picture_ptr ? s->last_picture_ptr->f->data[0] : NULL,
  1656. s->next_picture_ptr ? s->next_picture_ptr->f->data[0] : NULL,
  1657. s->current_picture_ptr ? s->current_picture_ptr->f->data[0] : NULL,
  1658. s->pict_type, s->droppable);
  1659. if ((!s->last_picture_ptr || !s->last_picture_ptr->f->buf[0]) &&
  1660. (s->pict_type != AV_PICTURE_TYPE_I ||
  1661. s->picture_structure != PICT_FRAME)) {
  1662. int h_chroma_shift, v_chroma_shift;
  1663. av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
  1664. &h_chroma_shift, &v_chroma_shift);
  1665. if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f->buf[0])
  1666. av_log(avctx, AV_LOG_DEBUG,
  1667. "allocating dummy last picture for B frame\n");
  1668. else if (s->pict_type != AV_PICTURE_TYPE_I)
  1669. av_log(avctx, AV_LOG_ERROR,
  1670. "warning: first frame is no keyframe\n");
  1671. else if (s->picture_structure != PICT_FRAME)
  1672. av_log(avctx, AV_LOG_DEBUG,
  1673. "allocate dummy last picture for field based first keyframe\n");
  1674. /* Allocate a dummy frame */
  1675. i = ff_find_unused_picture(s, 0);
  1676. if (i < 0) {
  1677. av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
  1678. return i;
  1679. }
  1680. s->last_picture_ptr = &s->picture[i];
  1681. s->last_picture_ptr->reference = 3;
  1682. s->last_picture_ptr->f->key_frame = 0;
  1683. s->last_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
  1684. if (ff_alloc_picture(s, s->last_picture_ptr, 0) < 0) {
  1685. s->last_picture_ptr = NULL;
  1686. return -1;
  1687. }
  1688. if (!avctx->hwaccel && !(avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)) {
  1689. for(i=0; i<avctx->height; i++)
  1690. memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i,
  1691. 0x80, avctx->width);
  1692. for(i=0; i<FF_CEIL_RSHIFT(avctx->height, v_chroma_shift); i++) {
  1693. memset(s->last_picture_ptr->f->data[1] + s->last_picture_ptr->f->linesize[1]*i,
  1694. 0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift));
  1695. memset(s->last_picture_ptr->f->data[2] + s->last_picture_ptr->f->linesize[2]*i,
  1696. 0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift));
  1697. }
  1698. if(s->codec_id == AV_CODEC_ID_FLV1 || s->codec_id == AV_CODEC_ID_H263){
  1699. for(i=0; i<avctx->height; i++)
  1700. memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i, 16, avctx->width);
  1701. }
  1702. }
  1703. ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 0);
  1704. ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 1);
  1705. }
  1706. if ((!s->next_picture_ptr || !s->next_picture_ptr->f->buf[0]) &&
  1707. s->pict_type == AV_PICTURE_TYPE_B) {
  1708. /* Allocate a dummy frame */
  1709. i = ff_find_unused_picture(s, 0);
  1710. if (i < 0) {
  1711. av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
  1712. return i;
  1713. }
  1714. s->next_picture_ptr = &s->picture[i];
  1715. s->next_picture_ptr->reference = 3;
  1716. s->next_picture_ptr->f->key_frame = 0;
  1717. s->next_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
  1718. if (ff_alloc_picture(s, s->next_picture_ptr, 0) < 0) {
  1719. s->next_picture_ptr = NULL;
  1720. return -1;
  1721. }
  1722. ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 0);
  1723. ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 1);
  1724. }
  1725. #if 0 // BUFREF-FIXME
  1726. memset(s->last_picture.f->data, 0, sizeof(s->last_picture.f->data));
  1727. memset(s->next_picture.f->data, 0, sizeof(s->next_picture.f->data));
  1728. #endif
  1729. if (s->last_picture_ptr) {
  1730. ff_mpeg_unref_picture(s, &s->last_picture);
  1731. if (s->last_picture_ptr->f->buf[0] &&
  1732. (ret = ff_mpeg_ref_picture(s, &s->last_picture,
  1733. s->last_picture_ptr)) < 0)
  1734. return ret;
  1735. }
  1736. if (s->next_picture_ptr) {
  1737. ff_mpeg_unref_picture(s, &s->next_picture);
  1738. if (s->next_picture_ptr->f->buf[0] &&
  1739. (ret = ff_mpeg_ref_picture(s, &s->next_picture,
  1740. s->next_picture_ptr)) < 0)
  1741. return ret;
  1742. }
  1743. av_assert0(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr &&
  1744. s->last_picture_ptr->f->buf[0]));
  1745. if (s->picture_structure!= PICT_FRAME) {
  1746. int i;
  1747. for (i = 0; i < 4; i++) {
  1748. if (s->picture_structure == PICT_BOTTOM_FIELD) {
  1749. s->current_picture.f->data[i] +=
  1750. s->current_picture.f->linesize[i];
  1751. }
  1752. s->current_picture.f->linesize[i] *= 2;
  1753. s->last_picture.f->linesize[i] *= 2;
  1754. s->next_picture.f->linesize[i] *= 2;
  1755. }
  1756. }
  1757. s->err_recognition = avctx->err_recognition;
  1758. /* set dequantizer, we can't do it during init as
  1759. * it might change for mpeg4 and we can't do it in the header
  1760. * decode as init is not called for mpeg4 there yet */
  1761. if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
  1762. s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
  1763. s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
  1764. } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
  1765. s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
  1766. s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
  1767. } else {
  1768. s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
  1769. s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
  1770. }
  1771. if (s->avctx->debug & FF_DEBUG_NOMC) {
  1772. gray_frame(s->current_picture_ptr->f);
  1773. }
  1774. return 0;
  1775. }
  1776. /* called after a frame has been decoded. */
  1777. void ff_mpv_frame_end(MpegEncContext *s)
  1778. {
  1779. emms_c();
  1780. if (s->current_picture.reference)
  1781. ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
  1782. }
  1783. #if FF_API_VISMV
  1784. static int clip_line(int *sx, int *sy, int *ex, int *ey, int maxx)
  1785. {
  1786. if(*sx > *ex)
  1787. return clip_line(ex, ey, sx, sy, maxx);
  1788. if (*sx < 0) {
  1789. if (*ex < 0)
  1790. return 1;
  1791. *sy = *ey + (*sy - *ey) * (int64_t)*ex / (*ex - *sx);
  1792. *sx = 0;
  1793. }
  1794. if (*ex > maxx) {
  1795. if (*sx > maxx)
  1796. return 1;
  1797. *ey = *sy + (*ey - *sy) * (int64_t)(maxx - *sx) / (*ex - *sx);
  1798. *ex = maxx;
  1799. }
  1800. return 0;
  1801. }
  1802. /**
  1803. * Draw a line from (ex, ey) -> (sx, sy).
  1804. * @param w width of the image
  1805. * @param h height of the image
  1806. * @param stride stride/linesize of the image
  1807. * @param color color of the arrow
  1808. */
  1809. static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey,
  1810. int w, int h, int stride, int color)
  1811. {
  1812. int x, y, fr, f;
  1813. if (clip_line(&sx, &sy, &ex, &ey, w - 1))
  1814. return;
  1815. if (clip_line(&sy, &sx, &ey, &ex, h - 1))
  1816. return;
  1817. sx = av_clip(sx, 0, w - 1);
  1818. sy = av_clip(sy, 0, h - 1);
  1819. ex = av_clip(ex, 0, w - 1);
  1820. ey = av_clip(ey, 0, h - 1);
  1821. buf[sy * stride + sx] += color;
  1822. if (FFABS(ex - sx) > FFABS(ey - sy)) {
  1823. if (sx > ex) {
  1824. FFSWAP(int, sx, ex);
  1825. FFSWAP(int, sy, ey);
  1826. }
  1827. buf += sx + sy * stride;
  1828. ex -= sx;
  1829. f = ((ey - sy) << 16) / ex;
  1830. for (x = 0; x <= ex; x++) {
  1831. y = (x * f) >> 16;
  1832. fr = (x * f) & 0xFFFF;
  1833. buf[y * stride + x] += (color * (0x10000 - fr)) >> 16;
  1834. if(fr) buf[(y + 1) * stride + x] += (color * fr ) >> 16;
  1835. }
  1836. } else {
  1837. if (sy > ey) {
  1838. FFSWAP(int, sx, ex);
  1839. FFSWAP(int, sy, ey);
  1840. }
  1841. buf += sx + sy * stride;
  1842. ey -= sy;
  1843. if (ey)
  1844. f = ((ex - sx) << 16) / ey;
  1845. else
  1846. f = 0;
  1847. for(y= 0; y <= ey; y++){
  1848. x = (y*f) >> 16;
  1849. fr = (y*f) & 0xFFFF;
  1850. buf[y * stride + x] += (color * (0x10000 - fr)) >> 16;
  1851. if(fr) buf[y * stride + x + 1] += (color * fr ) >> 16;
  1852. }
  1853. }
  1854. }
  1855. /**
  1856. * Draw an arrow from (ex, ey) -> (sx, sy).
  1857. * @param w width of the image
  1858. * @param h height of the image
  1859. * @param stride stride/linesize of the image
  1860. * @param color color of the arrow
  1861. */
  1862. static void draw_arrow(uint8_t *buf, int sx, int sy, int ex,
  1863. int ey, int w, int h, int stride, int color, int tail, int direction)
  1864. {
  1865. int dx,dy;
  1866. if (direction) {
  1867. FFSWAP(int, sx, ex);
  1868. FFSWAP(int, sy, ey);
  1869. }
  1870. sx = av_clip(sx, -100, w + 100);
  1871. sy = av_clip(sy, -100, h + 100);
  1872. ex = av_clip(ex, -100, w + 100);
  1873. ey = av_clip(ey, -100, h + 100);
  1874. dx = ex - sx;
  1875. dy = ey - sy;
  1876. if (dx * dx + dy * dy > 3 * 3) {
  1877. int rx = dx + dy;
  1878. int ry = -dx + dy;
  1879. int length = ff_sqrt((rx * rx + ry * ry) << 8);
  1880. // FIXME subpixel accuracy
  1881. rx = ROUNDED_DIV(rx * 3 << 4, length);
  1882. ry = ROUNDED_DIV(ry * 3 << 4, length);
  1883. if (tail) {
  1884. rx = -rx;
  1885. ry = -ry;
  1886. }
  1887. draw_line(buf, sx, sy, sx + rx, sy + ry, w, h, stride, color);
  1888. draw_line(buf, sx, sy, sx - ry, sy + rx, w, h, stride, color);
  1889. }
  1890. draw_line(buf, sx, sy, ex, ey, w, h, stride, color);
  1891. }
  1892. #endif
  1893. static int add_mb(AVMotionVector *mb, uint32_t mb_type,
  1894. int dst_x, int dst_y,
  1895. int src_x, int src_y,
  1896. int direction)
  1897. {
  1898. if (dst_x == src_x && dst_y == src_y)
  1899. return 0;
  1900. mb->w = IS_8X8(mb_type) || IS_8X16(mb_type) ? 8 : 16;
  1901. mb->h = IS_8X8(mb_type) || IS_16X8(mb_type) ? 8 : 16;
  1902. mb->src_x = src_x;
  1903. mb->src_y = src_y;
  1904. mb->dst_x = dst_x;
  1905. mb->dst_y = dst_y;
  1906. mb->source = direction ? 1 : -1;
  1907. mb->flags = 0; // XXX: does mb_type contain extra information that could be exported here?
  1908. return 1;
  1909. }
  1910. /**
  1911. * Print debugging info for the given picture.
  1912. */
  1913. void ff_print_debug_info2(AVCodecContext *avctx, AVFrame *pict, uint8_t *mbskip_table,
  1914. uint32_t *mbtype_table, int8_t *qscale_table, int16_t (*motion_val[2])[2],
  1915. int *low_delay,
  1916. int mb_width, int mb_height, int mb_stride, int quarter_sample)
  1917. {
  1918. if ((avctx->flags2 & CODEC_FLAG2_EXPORT_MVS) && mbtype_table && motion_val[0]) {
  1919. const int shift = 1 + quarter_sample;
  1920. const int mv_sample_log2 = avctx->codec_id == AV_CODEC_ID_H264 || avctx->codec_id == AV_CODEC_ID_SVQ3 ? 2 : 1;
  1921. const int mv_stride = (mb_width << mv_sample_log2) +
  1922. (avctx->codec->id == AV_CODEC_ID_H264 ? 0 : 1);
  1923. int mb_x, mb_y, mbcount = 0;
  1924. /* size is width * height * 2 * 4 where 2 is for directions and 4 is
  1925. * for the maximum number of MB (4 MB in case of IS_8x8) */
  1926. AVMotionVector *mvs = av_malloc_array(mb_width * mb_height, 2 * 4 * sizeof(AVMotionVector));
  1927. if (!mvs)
  1928. return;
  1929. for (mb_y = 0; mb_y < mb_height; mb_y++) {
  1930. for (mb_x = 0; mb_x < mb_width; mb_x++) {
  1931. int i, direction, mb_type = mbtype_table[mb_x + mb_y * mb_stride];
  1932. for (direction = 0; direction < 2; direction++) {
  1933. if (!USES_LIST(mb_type, direction))
  1934. continue;
  1935. if (IS_8X8(mb_type)) {
  1936. for (i = 0; i < 4; i++) {
  1937. int sx = mb_x * 16 + 4 + 8 * (i & 1);
  1938. int sy = mb_y * 16 + 4 + 8 * (i >> 1);
  1939. int xy = (mb_x * 2 + (i & 1) +
  1940. (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
  1941. int mx = (motion_val[direction][xy][0] >> shift) + sx;
  1942. int my = (motion_val[direction][xy][1] >> shift) + sy;
  1943. mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx, my, direction);
  1944. }
  1945. } else if (IS_16X8(mb_type)) {
  1946. for (i = 0; i < 2; i++) {
  1947. int sx = mb_x * 16 + 8;
  1948. int sy = mb_y * 16 + 4 + 8 * i;
  1949. int xy = (mb_x * 2 + (mb_y * 2 + i) * mv_stride) << (mv_sample_log2 - 1);
  1950. int mx = (motion_val[direction][xy][0] >> shift);
  1951. int my = (motion_val[direction][xy][1] >> shift);
  1952. if (IS_INTERLACED(mb_type))
  1953. my *= 2;
  1954. mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx + sx, my + sy, direction);
  1955. }
  1956. } else if (IS_8X16(mb_type)) {
  1957. for (i = 0; i < 2; i++) {
  1958. int sx = mb_x * 16 + 4 + 8 * i;
  1959. int sy = mb_y * 16 + 8;
  1960. int xy = (mb_x * 2 + i + mb_y * 2 * mv_stride) << (mv_sample_log2 - 1);
  1961. int mx = motion_val[direction][xy][0] >> shift;
  1962. int my = motion_val[direction][xy][1] >> shift;
  1963. if (IS_INTERLACED(mb_type))
  1964. my *= 2;
  1965. mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx + sx, my + sy, direction);
  1966. }
  1967. } else {
  1968. int sx = mb_x * 16 + 8;
  1969. int sy = mb_y * 16 + 8;
  1970. int xy = (mb_x + mb_y * mv_stride) << mv_sample_log2;
  1971. int mx = (motion_val[direction][xy][0]>>shift) + sx;
  1972. int my = (motion_val[direction][xy][1]>>shift) + sy;
  1973. mbcount += add_mb(mvs + mbcount, mb_type, sx, sy, mx, my, direction);
  1974. }
  1975. }
  1976. }
  1977. }
  1978. if (mbcount) {
  1979. AVFrameSideData *sd;
  1980. av_log(avctx, AV_LOG_DEBUG, "Adding %d MVs info to frame %d\n", mbcount, avctx->frame_number);
  1981. sd = av_frame_new_side_data(pict, AV_FRAME_DATA_MOTION_VECTORS, mbcount * sizeof(AVMotionVector));
  1982. if (!sd)
  1983. return;
  1984. memcpy(sd->data, mvs, mbcount * sizeof(AVMotionVector));
  1985. }
  1986. av_freep(&mvs);
  1987. }
  1988. /* TODO: export all the following to make them accessible for users (and filters) */
  1989. if (avctx->hwaccel || !mbtype_table
  1990. || (avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU))
  1991. return;
  1992. if (avctx->debug & (FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)) {
  1993. int x,y;
  1994. av_log(avctx, AV_LOG_DEBUG, "New frame, type: %c\n",
  1995. av_get_picture_type_char(pict->pict_type));
  1996. for (y = 0; y < mb_height; y++) {
  1997. for (x = 0; x < mb_width; x++) {
  1998. if (avctx->debug & FF_DEBUG_SKIP) {
  1999. int count = mbskip_table[x + y * mb_stride];
  2000. if (count > 9)
  2001. count = 9;
  2002. av_log(avctx, AV_LOG_DEBUG, "%1d", count);
  2003. }
  2004. if (avctx->debug & FF_DEBUG_QP) {
  2005. av_log(avctx, AV_LOG_DEBUG, "%2d",
  2006. qscale_table[x + y * mb_stride]);
  2007. }
  2008. if (avctx->debug & FF_DEBUG_MB_TYPE) {
  2009. int mb_type = mbtype_table[x + y * mb_stride];
  2010. // Type & MV direction
  2011. if (IS_PCM(mb_type))
  2012. av_log(avctx, AV_LOG_DEBUG, "P");
  2013. else if (IS_INTRA(mb_type) && IS_ACPRED(mb_type))
  2014. av_log(avctx, AV_LOG_DEBUG, "A");
  2015. else if (IS_INTRA4x4(mb_type))
  2016. av_log(avctx, AV_LOG_DEBUG, "i");
  2017. else if (IS_INTRA16x16(mb_type))
  2018. av_log(avctx, AV_LOG_DEBUG, "I");
  2019. else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type))
  2020. av_log(avctx, AV_LOG_DEBUG, "d");
  2021. else if (IS_DIRECT(mb_type))
  2022. av_log(avctx, AV_LOG_DEBUG, "D");
  2023. else if (IS_GMC(mb_type) && IS_SKIP(mb_type))
  2024. av_log(avctx, AV_LOG_DEBUG, "g");
  2025. else if (IS_GMC(mb_type))
  2026. av_log(avctx, AV_LOG_DEBUG, "G");
  2027. else if (IS_SKIP(mb_type))
  2028. av_log(avctx, AV_LOG_DEBUG, "S");
  2029. else if (!USES_LIST(mb_type, 1))
  2030. av_log(avctx, AV_LOG_DEBUG, ">");
  2031. else if (!USES_LIST(mb_type, 0))
  2032. av_log(avctx, AV_LOG_DEBUG, "<");
  2033. else {
  2034. av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
  2035. av_log(avctx, AV_LOG_DEBUG, "X");
  2036. }
  2037. // segmentation
  2038. if (IS_8X8(mb_type))
  2039. av_log(avctx, AV_LOG_DEBUG, "+");
  2040. else if (IS_16X8(mb_type))
  2041. av_log(avctx, AV_LOG_DEBUG, "-");
  2042. else if (IS_8X16(mb_type))
  2043. av_log(avctx, AV_LOG_DEBUG, "|");
  2044. else if (IS_INTRA(mb_type) || IS_16X16(mb_type))
  2045. av_log(avctx, AV_LOG_DEBUG, " ");
  2046. else
  2047. av_log(avctx, AV_LOG_DEBUG, "?");
  2048. if (IS_INTERLACED(mb_type))
  2049. av_log(avctx, AV_LOG_DEBUG, "=");
  2050. else
  2051. av_log(avctx, AV_LOG_DEBUG, " ");
  2052. }
  2053. }
  2054. av_log(avctx, AV_LOG_DEBUG, "\n");
  2055. }
  2056. }
  2057. if ((avctx->debug & (FF_DEBUG_VIS_QP | FF_DEBUG_VIS_MB_TYPE)) ||
  2058. (avctx->debug_mv)) {
  2059. int mb_y;
  2060. int i;
  2061. int h_chroma_shift, v_chroma_shift, block_height;
  2062. #if FF_API_VISMV
  2063. const int shift = 1 + quarter_sample;
  2064. uint8_t *ptr;
  2065. const int width = avctx->width;
  2066. const int height = avctx->height;
  2067. #endif
  2068. const int mv_sample_log2 = avctx->codec_id == AV_CODEC_ID_H264 || avctx->codec_id == AV_CODEC_ID_SVQ3 ? 2 : 1;
  2069. const int mv_stride = (mb_width << mv_sample_log2) +
  2070. (avctx->codec->id == AV_CODEC_ID_H264 ? 0 : 1);
  2071. *low_delay = 0; // needed to see the vectors without trashing the buffers
  2072. avcodec_get_chroma_sub_sample(avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
  2073. av_frame_make_writable(pict);
  2074. pict->opaque = NULL;
  2075. #if FF_API_VISMV
  2076. ptr = pict->data[0];
  2077. #endif
  2078. block_height = 16 >> v_chroma_shift;
  2079. for (mb_y = 0; mb_y < mb_height; mb_y++) {
  2080. int mb_x;
  2081. for (mb_x = 0; mb_x < mb_width; mb_x++) {
  2082. const int mb_index = mb_x + mb_y * mb_stride;
  2083. #if FF_API_VISMV
  2084. if ((avctx->debug_mv) && motion_val[0]) {
  2085. int type;
  2086. for (type = 0; type < 3; type++) {
  2087. int direction = 0;
  2088. switch (type) {
  2089. case 0:
  2090. if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_P_FOR)) ||
  2091. (pict->pict_type!= AV_PICTURE_TYPE_P))
  2092. continue;
  2093. direction = 0;
  2094. break;
  2095. case 1:
  2096. if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_FOR)) ||
  2097. (pict->pict_type!= AV_PICTURE_TYPE_B))
  2098. continue;
  2099. direction = 0;
  2100. break;
  2101. case 2:
  2102. if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_BACK)) ||
  2103. (pict->pict_type!= AV_PICTURE_TYPE_B))
  2104. continue;
  2105. direction = 1;
  2106. break;
  2107. }
  2108. if (!USES_LIST(mbtype_table[mb_index], direction))
  2109. continue;
  2110. if (IS_8X8(mbtype_table[mb_index])) {
  2111. int i;
  2112. for (i = 0; i < 4; i++) {
  2113. int sx = mb_x * 16 + 4 + 8 * (i & 1);
  2114. int sy = mb_y * 16 + 4 + 8 * (i >> 1);
  2115. int xy = (mb_x * 2 + (i & 1) +
  2116. (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
  2117. int mx = (motion_val[direction][xy][0] >> shift) + sx;
  2118. int my = (motion_val[direction][xy][1] >> shift) + sy;
  2119. draw_arrow(ptr, sx, sy, mx, my, width,
  2120. height, pict->linesize[0], 100, 0, direction);
  2121. }
  2122. } else if (IS_16X8(mbtype_table[mb_index])) {
  2123. int i;
  2124. for (i = 0; i < 2; i++) {
  2125. int sx = mb_x * 16 + 8;
  2126. int sy = mb_y * 16 + 4 + 8 * i;
  2127. int xy = (mb_x * 2 + (mb_y * 2 + i) * mv_stride) << (mv_sample_log2 - 1);
  2128. int mx = (motion_val[direction][xy][0] >> shift);
  2129. int my = (motion_val[direction][xy][1] >> shift);
  2130. if (IS_INTERLACED(mbtype_table[mb_index]))
  2131. my *= 2;
  2132. draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
  2133. height, pict->linesize[0], 100, 0, direction);
  2134. }
  2135. } else if (IS_8X16(mbtype_table[mb_index])) {
  2136. int i;
  2137. for (i = 0; i < 2; i++) {
  2138. int sx = mb_x * 16 + 4 + 8 * i;
  2139. int sy = mb_y * 16 + 8;
  2140. int xy = (mb_x * 2 + i + mb_y * 2 * mv_stride) << (mv_sample_log2 - 1);
  2141. int mx = motion_val[direction][xy][0] >> shift;
  2142. int my = motion_val[direction][xy][1] >> shift;
  2143. if (IS_INTERLACED(mbtype_table[mb_index]))
  2144. my *= 2;
  2145. draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
  2146. height, pict->linesize[0], 100, 0, direction);
  2147. }
  2148. } else {
  2149. int sx= mb_x * 16 + 8;
  2150. int sy= mb_y * 16 + 8;
  2151. int xy= (mb_x + mb_y * mv_stride) << mv_sample_log2;
  2152. int mx= (motion_val[direction][xy][0]>>shift) + sx;
  2153. int my= (motion_val[direction][xy][1]>>shift) + sy;
  2154. draw_arrow(ptr, sx, sy, mx, my, width, height, pict->linesize[0], 100, 0, direction);
  2155. }
  2156. }
  2157. }
  2158. #endif
  2159. if ((avctx->debug & FF_DEBUG_VIS_QP)) {
  2160. uint64_t c = (qscale_table[mb_index] * 128 / 31) *
  2161. 0x0101010101010101ULL;
  2162. int y;
  2163. for (y = 0; y < block_height; y++) {
  2164. *(uint64_t *)(pict->data[1] + 8 * mb_x +
  2165. (block_height * mb_y + y) *
  2166. pict->linesize[1]) = c;
  2167. *(uint64_t *)(pict->data[2] + 8 * mb_x +
  2168. (block_height * mb_y + y) *
  2169. pict->linesize[2]) = c;
  2170. }
  2171. }
  2172. if ((avctx->debug & FF_DEBUG_VIS_MB_TYPE) &&
  2173. motion_val[0]) {
  2174. int mb_type = mbtype_table[mb_index];
  2175. uint64_t u,v;
  2176. int y;
  2177. #define COLOR(theta, r) \
  2178. u = (int)(128 + r * cos(theta * 3.141592 / 180)); \
  2179. v = (int)(128 + r * sin(theta * 3.141592 / 180));
  2180. u = v = 128;
  2181. if (IS_PCM(mb_type)) {
  2182. COLOR(120, 48)
  2183. } else if ((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) ||
  2184. IS_INTRA16x16(mb_type)) {
  2185. COLOR(30, 48)
  2186. } else if (IS_INTRA4x4(mb_type)) {
  2187. COLOR(90, 48)
  2188. } else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type)) {
  2189. // COLOR(120, 48)
  2190. } else if (IS_DIRECT(mb_type)) {
  2191. COLOR(150, 48)
  2192. } else if (IS_GMC(mb_type) && IS_SKIP(mb_type)) {
  2193. COLOR(170, 48)
  2194. } else if (IS_GMC(mb_type)) {
  2195. COLOR(190, 48)
  2196. } else if (IS_SKIP(mb_type)) {
  2197. // COLOR(180, 48)
  2198. } else if (!USES_LIST(mb_type, 1)) {
  2199. COLOR(240, 48)
  2200. } else if (!USES_LIST(mb_type, 0)) {
  2201. COLOR(0, 48)
  2202. } else {
  2203. av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
  2204. COLOR(300,48)
  2205. }
  2206. u *= 0x0101010101010101ULL;
  2207. v *= 0x0101010101010101ULL;
  2208. for (y = 0; y < block_height; y++) {
  2209. *(uint64_t *)(pict->data[1] + 8 * mb_x +
  2210. (block_height * mb_y + y) * pict->linesize[1]) = u;
  2211. *(uint64_t *)(pict->data[2] + 8 * mb_x +
  2212. (block_height * mb_y + y) * pict->linesize[2]) = v;
  2213. }
  2214. // segmentation
  2215. if (IS_8X8(mb_type) || IS_16X8(mb_type)) {
  2216. *(uint64_t *)(pict->data[0] + 16 * mb_x + 0 +
  2217. (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
  2218. *(uint64_t *)(pict->data[0] + 16 * mb_x + 8 +
  2219. (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
  2220. }
  2221. if (IS_8X8(mb_type) || IS_8X16(mb_type)) {
  2222. for (y = 0; y < 16; y++)
  2223. pict->data[0][16 * mb_x + 8 + (16 * mb_y + y) *
  2224. pict->linesize[0]] ^= 0x80;
  2225. }
  2226. if (IS_8X8(mb_type) && mv_sample_log2 >= 2) {
  2227. int dm = 1 << (mv_sample_log2 - 2);
  2228. for (i = 0; i < 4; i++) {
  2229. int sx = mb_x * 16 + 8 * (i & 1);
  2230. int sy = mb_y * 16 + 8 * (i >> 1);
  2231. int xy = (mb_x * 2 + (i & 1) +
  2232. (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
  2233. // FIXME bidir
  2234. int32_t *mv = (int32_t *) &motion_val[0][xy];
  2235. if (mv[0] != mv[dm] ||
  2236. mv[dm * mv_stride] != mv[dm * (mv_stride + 1)])
  2237. for (y = 0; y < 8; y++)
  2238. pict->data[0][sx + 4 + (sy + y) * pict->linesize[0]] ^= 0x80;
  2239. if (mv[0] != mv[dm * mv_stride] || mv[dm] != mv[dm * (mv_stride + 1)])
  2240. *(uint64_t *)(pict->data[0] + sx + (sy + 4) *
  2241. pict->linesize[0]) ^= 0x8080808080808080ULL;
  2242. }
  2243. }
  2244. if (IS_INTERLACED(mb_type) &&
  2245. avctx->codec->id == AV_CODEC_ID_H264) {
  2246. // hmm
  2247. }
  2248. }
  2249. mbskip_table[mb_index] = 0;
  2250. }
  2251. }
  2252. }
  2253. }
  2254. void ff_print_debug_info(MpegEncContext *s, Picture *p, AVFrame *pict)
  2255. {
  2256. ff_print_debug_info2(s->avctx, pict, s->mbskip_table, p->mb_type,
  2257. p->qscale_table, p->motion_val, &s->low_delay,
  2258. s->mb_width, s->mb_height, s->mb_stride, s->quarter_sample);
  2259. }
  2260. int ff_mpv_export_qp_table(MpegEncContext *s, AVFrame *f, Picture *p, int qp_type)
  2261. {
  2262. AVBufferRef *ref = av_buffer_ref(p->qscale_table_buf);
  2263. int offset = 2*s->mb_stride + 1;
  2264. if(!ref)
  2265. return AVERROR(ENOMEM);
  2266. av_assert0(ref->size >= offset + s->mb_stride * ((f->height+15)/16));
  2267. ref->size -= offset;
  2268. ref->data += offset;
  2269. return av_frame_set_qp_table(f, ref, s->mb_stride, qp_type);
  2270. }
  2271. static inline int hpel_motion_lowres(MpegEncContext *s,
  2272. uint8_t *dest, uint8_t *src,
  2273. int field_based, int field_select,
  2274. int src_x, int src_y,
  2275. int width, int height, ptrdiff_t stride,
  2276. int h_edge_pos, int v_edge_pos,
  2277. int w, int h, h264_chroma_mc_func *pix_op,
  2278. int motion_x, int motion_y)
  2279. {
  2280. const int lowres = s->avctx->lowres;
  2281. const int op_index = FFMIN(lowres, 3);
  2282. const int s_mask = (2 << lowres) - 1;
  2283. int emu = 0;
  2284. int sx, sy;
  2285. if (s->quarter_sample) {
  2286. motion_x /= 2;
  2287. motion_y /= 2;
  2288. }
  2289. sx = motion_x & s_mask;
  2290. sy = motion_y & s_mask;
  2291. src_x += motion_x >> lowres + 1;
  2292. src_y += motion_y >> lowres + 1;
  2293. src += src_y * stride + src_x;
  2294. if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w, 0) ||
  2295. (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
  2296. s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,
  2297. s->linesize, s->linesize,
  2298. w + 1, (h + 1) << field_based,
  2299. src_x, src_y << field_based,
  2300. h_edge_pos, v_edge_pos);
  2301. src = s->edge_emu_buffer;
  2302. emu = 1;
  2303. }
  2304. sx = (sx << 2) >> lowres;
  2305. sy = (sy << 2) >> lowres;
  2306. if (field_select)
  2307. src += s->linesize;
  2308. pix_op[op_index](dest, src, stride, h, sx, sy);
  2309. return emu;
  2310. }
  2311. /* apply one mpeg motion vector to the three components */
  2312. static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
  2313. uint8_t *dest_y,
  2314. uint8_t *dest_cb,
  2315. uint8_t *dest_cr,
  2316. int field_based,
  2317. int bottom_field,
  2318. int field_select,
  2319. uint8_t **ref_picture,
  2320. h264_chroma_mc_func *pix_op,
  2321. int motion_x, int motion_y,
  2322. int h, int mb_y)
  2323. {
  2324. uint8_t *ptr_y, *ptr_cb, *ptr_cr;
  2325. int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, sx, sy, uvsx, uvsy;
  2326. ptrdiff_t uvlinesize, linesize;
  2327. const int lowres = s->avctx->lowres;
  2328. const int op_index = FFMIN(lowres-1+s->chroma_x_shift, 3);
  2329. const int block_s = 8>>lowres;
  2330. const int s_mask = (2 << lowres) - 1;
  2331. const int h_edge_pos = s->h_edge_pos >> lowres;
  2332. const int v_edge_pos = s->v_edge_pos >> lowres;
  2333. linesize = s->current_picture.f->linesize[0] << field_based;
  2334. uvlinesize = s->current_picture.f->linesize[1] << field_based;
  2335. // FIXME obviously not perfect but qpel will not work in lowres anyway
  2336. if (s->quarter_sample) {
  2337. motion_x /= 2;
  2338. motion_y /= 2;
  2339. }
  2340. if(field_based){
  2341. motion_y += (bottom_field - field_select)*((1 << lowres)-1);
  2342. }
  2343. sx = motion_x & s_mask;
  2344. sy = motion_y & s_mask;
  2345. src_x = s->mb_x * 2 * block_s + (motion_x >> lowres + 1);
  2346. src_y = (mb_y * 2 * block_s >> field_based) + (motion_y >> lowres + 1);
  2347. if (s->out_format == FMT_H263) {
  2348. uvsx = ((motion_x >> 1) & s_mask) | (sx & 1);
  2349. uvsy = ((motion_y >> 1) & s_mask) | (sy & 1);
  2350. uvsrc_x = src_x >> 1;
  2351. uvsrc_y = src_y >> 1;
  2352. } else if (s->out_format == FMT_H261) {
  2353. // even chroma mv's are full pel in H261
  2354. mx = motion_x / 4;
  2355. my = motion_y / 4;
  2356. uvsx = (2 * mx) & s_mask;
  2357. uvsy = (2 * my) & s_mask;
  2358. uvsrc_x = s->mb_x * block_s + (mx >> lowres);
  2359. uvsrc_y = mb_y * block_s + (my >> lowres);
  2360. } else {
  2361. if(s->chroma_y_shift){
  2362. mx = motion_x / 2;
  2363. my = motion_y / 2;
  2364. uvsx = mx & s_mask;
  2365. uvsy = my & s_mask;
  2366. uvsrc_x = s->mb_x * block_s + (mx >> lowres + 1);
  2367. uvsrc_y = (mb_y * block_s >> field_based) + (my >> lowres + 1);
  2368. } else {
  2369. if(s->chroma_x_shift){
  2370. //Chroma422
  2371. mx = motion_x / 2;
  2372. uvsx = mx & s_mask;
  2373. uvsy = motion_y & s_mask;
  2374. uvsrc_y = src_y;
  2375. uvsrc_x = s->mb_x*block_s + (mx >> (lowres+1));
  2376. } else {
  2377. //Chroma444
  2378. uvsx = motion_x & s_mask;
  2379. uvsy = motion_y & s_mask;
  2380. uvsrc_x = src_x;
  2381. uvsrc_y = src_y;
  2382. }
  2383. }
  2384. }
  2385. ptr_y = ref_picture[0] + src_y * linesize + src_x;
  2386. ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
  2387. ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
  2388. if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s, 0) || uvsrc_y<0 ||
  2389. (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
  2390. s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
  2391. linesize >> field_based, linesize >> field_based,
  2392. 17, 17 + field_based,
  2393. src_x, src_y << field_based, h_edge_pos,
  2394. v_edge_pos);
  2395. ptr_y = s->edge_emu_buffer;
  2396. if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
  2397. uint8_t *ubuf = s->edge_emu_buffer + 18 * s->linesize;
  2398. uint8_t *vbuf =ubuf + 9 * s->uvlinesize;
  2399. s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
  2400. uvlinesize >> field_based, uvlinesize >> field_based,
  2401. 9, 9 + field_based,
  2402. uvsrc_x, uvsrc_y << field_based,
  2403. h_edge_pos >> 1, v_edge_pos >> 1);
  2404. s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
  2405. uvlinesize >> field_based,uvlinesize >> field_based,
  2406. 9, 9 + field_based,
  2407. uvsrc_x, uvsrc_y << field_based,
  2408. h_edge_pos >> 1, v_edge_pos >> 1);
  2409. ptr_cb = ubuf;
  2410. ptr_cr = vbuf;
  2411. }
  2412. }
  2413. // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f->data
  2414. if (bottom_field) {
  2415. dest_y += s->linesize;
  2416. dest_cb += s->uvlinesize;
  2417. dest_cr += s->uvlinesize;
  2418. }
  2419. if (field_select) {
  2420. ptr_y += s->linesize;
  2421. ptr_cb += s->uvlinesize;
  2422. ptr_cr += s->uvlinesize;
  2423. }
  2424. sx = (sx << 2) >> lowres;
  2425. sy = (sy << 2) >> lowres;
  2426. pix_op[lowres - 1](dest_y, ptr_y, linesize, h, sx, sy);
  2427. if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
  2428. int hc = s->chroma_y_shift ? (h+1-bottom_field)>>1 : h;
  2429. uvsx = (uvsx << 2) >> lowres;
  2430. uvsy = (uvsy << 2) >> lowres;
  2431. if (hc) {
  2432. pix_op[op_index](dest_cb, ptr_cb, uvlinesize, hc, uvsx, uvsy);
  2433. pix_op[op_index](dest_cr, ptr_cr, uvlinesize, hc, uvsx, uvsy);
  2434. }
  2435. }
  2436. // FIXME h261 lowres loop filter
  2437. }
  2438. static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
  2439. uint8_t *dest_cb, uint8_t *dest_cr,
  2440. uint8_t **ref_picture,
  2441. h264_chroma_mc_func * pix_op,
  2442. int mx, int my)
  2443. {
  2444. const int lowres = s->avctx->lowres;
  2445. const int op_index = FFMIN(lowres, 3);
  2446. const int block_s = 8 >> lowres;
  2447. const int s_mask = (2 << lowres) - 1;
  2448. const int h_edge_pos = s->h_edge_pos >> lowres + 1;
  2449. const int v_edge_pos = s->v_edge_pos >> lowres + 1;
  2450. int emu = 0, src_x, src_y, sx, sy;
  2451. ptrdiff_t offset;
  2452. uint8_t *ptr;
  2453. if (s->quarter_sample) {
  2454. mx /= 2;
  2455. my /= 2;
  2456. }
  2457. /* In case of 8X8, we construct a single chroma motion vector
  2458. with a special rounding */
  2459. mx = ff_h263_round_chroma(mx);
  2460. my = ff_h263_round_chroma(my);
  2461. sx = mx & s_mask;
  2462. sy = my & s_mask;
  2463. src_x = s->mb_x * block_s + (mx >> lowres + 1);
  2464. src_y = s->mb_y * block_s + (my >> lowres + 1);
  2465. offset = src_y * s->uvlinesize + src_x;
  2466. ptr = ref_picture[1] + offset;
  2467. if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) ||
  2468. (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) {
  2469. s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
  2470. s->uvlinesize, s->uvlinesize,
  2471. 9, 9,
  2472. src_x, src_y, h_edge_pos, v_edge_pos);
  2473. ptr = s->edge_emu_buffer;
  2474. emu = 1;
  2475. }
  2476. sx = (sx << 2) >> lowres;
  2477. sy = (sy << 2) >> lowres;
  2478. pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
  2479. ptr = ref_picture[2] + offset;
  2480. if (emu) {
  2481. s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
  2482. s->uvlinesize, s->uvlinesize,
  2483. 9, 9,
  2484. src_x, src_y, h_edge_pos, v_edge_pos);
  2485. ptr = s->edge_emu_buffer;
  2486. }
  2487. pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
  2488. }
  2489. /**
  2490. * motion compensation of a single macroblock
  2491. * @param s context
  2492. * @param dest_y luma destination pointer
  2493. * @param dest_cb chroma cb/u destination pointer
  2494. * @param dest_cr chroma cr/v destination pointer
  2495. * @param dir direction (0->forward, 1->backward)
  2496. * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
  2497. * @param pix_op halfpel motion compensation function (average or put normally)
  2498. * the motion vectors are taken from s->mv and the MV type from s->mv_type
  2499. */
  2500. static inline void MPV_motion_lowres(MpegEncContext *s,
  2501. uint8_t *dest_y, uint8_t *dest_cb,
  2502. uint8_t *dest_cr,
  2503. int dir, uint8_t **ref_picture,
  2504. h264_chroma_mc_func *pix_op)
  2505. {
  2506. int mx, my;
  2507. int mb_x, mb_y, i;
  2508. const int lowres = s->avctx->lowres;
  2509. const int block_s = 8 >>lowres;
  2510. mb_x = s->mb_x;
  2511. mb_y = s->mb_y;
  2512. switch (s->mv_type) {
  2513. case MV_TYPE_16X16:
  2514. mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
  2515. 0, 0, 0,
  2516. ref_picture, pix_op,
  2517. s->mv[dir][0][0], s->mv[dir][0][1],
  2518. 2 * block_s, mb_y);
  2519. break;
  2520. case MV_TYPE_8X8:
  2521. mx = 0;
  2522. my = 0;
  2523. for (i = 0; i < 4; i++) {
  2524. hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) *
  2525. s->linesize) * block_s,
  2526. ref_picture[0], 0, 0,
  2527. (2 * mb_x + (i & 1)) * block_s,
  2528. (2 * mb_y + (i >> 1)) * block_s,
  2529. s->width, s->height, s->linesize,
  2530. s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
  2531. block_s, block_s, pix_op,
  2532. s->mv[dir][i][0], s->mv[dir][i][1]);
  2533. mx += s->mv[dir][i][0];
  2534. my += s->mv[dir][i][1];
  2535. }
  2536. if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
  2537. chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture,
  2538. pix_op, mx, my);
  2539. break;
  2540. case MV_TYPE_FIELD:
  2541. if (s->picture_structure == PICT_FRAME) {
  2542. /* top field */
  2543. mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
  2544. 1, 0, s->field_select[dir][0],
  2545. ref_picture, pix_op,
  2546. s->mv[dir][0][0], s->mv[dir][0][1],
  2547. block_s, mb_y);
  2548. /* bottom field */
  2549. mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
  2550. 1, 1, s->field_select[dir][1],
  2551. ref_picture, pix_op,
  2552. s->mv[dir][1][0], s->mv[dir][1][1],
  2553. block_s, mb_y);
  2554. } else {
  2555. if (s->picture_structure != s->field_select[dir][0] + 1 &&
  2556. s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
  2557. ref_picture = s->current_picture_ptr->f->data;
  2558. }
  2559. mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
  2560. 0, 0, s->field_select[dir][0],
  2561. ref_picture, pix_op,
  2562. s->mv[dir][0][0],
  2563. s->mv[dir][0][1], 2 * block_s, mb_y >> 1);
  2564. }
  2565. break;
  2566. case MV_TYPE_16X8:
  2567. for (i = 0; i < 2; i++) {
  2568. uint8_t **ref2picture;
  2569. if (s->picture_structure == s->field_select[dir][i] + 1 ||
  2570. s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
  2571. ref2picture = ref_picture;
  2572. } else {
  2573. ref2picture = s->current_picture_ptr->f->data;
  2574. }
  2575. mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
  2576. 0, 0, s->field_select[dir][i],
  2577. ref2picture, pix_op,
  2578. s->mv[dir][i][0], s->mv[dir][i][1] +
  2579. 2 * block_s * i, block_s, mb_y >> 1);
  2580. dest_y += 2 * block_s * s->linesize;
  2581. dest_cb += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
  2582. dest_cr += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
  2583. }
  2584. break;
  2585. case MV_TYPE_DMV:
  2586. if (s->picture_structure == PICT_FRAME) {
  2587. for (i = 0; i < 2; i++) {
  2588. int j;
  2589. for (j = 0; j < 2; j++) {
  2590. mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
  2591. 1, j, j ^ i,
  2592. ref_picture, pix_op,
  2593. s->mv[dir][2 * i + j][0],
  2594. s->mv[dir][2 * i + j][1],
  2595. block_s, mb_y);
  2596. }
  2597. pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
  2598. }
  2599. } else {
  2600. for (i = 0; i < 2; i++) {
  2601. mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
  2602. 0, 0, s->picture_structure != i + 1,
  2603. ref_picture, pix_op,
  2604. s->mv[dir][2 * i][0],s->mv[dir][2 * i][1],
  2605. 2 * block_s, mb_y >> 1);
  2606. // after put we make avg of the same block
  2607. pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
  2608. // opposite parity is always in the same
  2609. // frame if this is second field
  2610. if (!s->first_field) {
  2611. ref_picture = s->current_picture_ptr->f->data;
  2612. }
  2613. }
  2614. }
  2615. break;
  2616. default:
  2617. av_assert2(0);
  2618. }
  2619. }
  2620. /**
  2621. * find the lowest MB row referenced in the MVs
  2622. */
  2623. int ff_mpv_lowest_referenced_row(MpegEncContext *s, int dir)
  2624. {
  2625. int my_max = INT_MIN, my_min = INT_MAX, qpel_shift = !s->quarter_sample;
  2626. int my, off, i, mvs;
  2627. if (s->picture_structure != PICT_FRAME || s->mcsel)
  2628. goto unhandled;
  2629. switch (s->mv_type) {
  2630. case MV_TYPE_16X16:
  2631. mvs = 1;
  2632. break;
  2633. case MV_TYPE_16X8:
  2634. mvs = 2;
  2635. break;
  2636. case MV_TYPE_8X8:
  2637. mvs = 4;
  2638. break;
  2639. default:
  2640. goto unhandled;
  2641. }
  2642. for (i = 0; i < mvs; i++) {
  2643. my = s->mv[dir][i][1]<<qpel_shift;
  2644. my_max = FFMAX(my_max, my);
  2645. my_min = FFMIN(my_min, my);
  2646. }
  2647. off = (FFMAX(-my_min, my_max) + 63) >> 6;
  2648. return FFMIN(FFMAX(s->mb_y + off, 0), s->mb_height-1);
  2649. unhandled:
  2650. return s->mb_height-1;
  2651. }
  2652. /* put block[] to dest[] */
  2653. static inline void put_dct(MpegEncContext *s,
  2654. int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
  2655. {
  2656. s->dct_unquantize_intra(s, block, i, qscale);
  2657. s->idsp.idct_put(dest, line_size, block);
  2658. }
  2659. /* add block[] to dest[] */
  2660. static inline void add_dct(MpegEncContext *s,
  2661. int16_t *block, int i, uint8_t *dest, int line_size)
  2662. {
  2663. if (s->block_last_index[i] >= 0) {
  2664. s->idsp.idct_add(dest, line_size, block);
  2665. }
  2666. }
  2667. static inline void add_dequant_dct(MpegEncContext *s,
  2668. int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
  2669. {
  2670. if (s->block_last_index[i] >= 0) {
  2671. s->dct_unquantize_inter(s, block, i, qscale);
  2672. s->idsp.idct_add(dest, line_size, block);
  2673. }
  2674. }
  2675. /**
  2676. * Clean dc, ac, coded_block for the current non-intra MB.
  2677. */
  2678. void ff_clean_intra_table_entries(MpegEncContext *s)
  2679. {
  2680. int wrap = s->b8_stride;
  2681. int xy = s->block_index[0];
  2682. s->dc_val[0][xy ] =
  2683. s->dc_val[0][xy + 1 ] =
  2684. s->dc_val[0][xy + wrap] =
  2685. s->dc_val[0][xy + 1 + wrap] = 1024;
  2686. /* ac pred */
  2687. memset(s->ac_val[0][xy ], 0, 32 * sizeof(int16_t));
  2688. memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
  2689. if (s->msmpeg4_version>=3) {
  2690. s->coded_block[xy ] =
  2691. s->coded_block[xy + 1 ] =
  2692. s->coded_block[xy + wrap] =
  2693. s->coded_block[xy + 1 + wrap] = 0;
  2694. }
  2695. /* chroma */
  2696. wrap = s->mb_stride;
  2697. xy = s->mb_x + s->mb_y * wrap;
  2698. s->dc_val[1][xy] =
  2699. s->dc_val[2][xy] = 1024;
  2700. /* ac pred */
  2701. memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
  2702. memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
  2703. s->mbintra_table[xy]= 0;
  2704. }
  2705. /* generic function called after a macroblock has been parsed by the
  2706. decoder or after it has been encoded by the encoder.
  2707. Important variables used:
  2708. s->mb_intra : true if intra macroblock
  2709. s->mv_dir : motion vector direction
  2710. s->mv_type : motion vector type
  2711. s->mv : motion vector
  2712. s->interlaced_dct : true if interlaced dct used (mpeg2)
  2713. */
  2714. static av_always_inline
  2715. void mpv_decode_mb_internal(MpegEncContext *s, int16_t block[12][64],
  2716. int lowres_flag, int is_mpeg12)
  2717. {
  2718. const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
  2719. if (CONFIG_XVMC &&
  2720. s->avctx->hwaccel && s->avctx->hwaccel->decode_mb) {
  2721. s->avctx->hwaccel->decode_mb(s);//xvmc uses pblocks
  2722. return;
  2723. }
  2724. if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
  2725. /* print DCT coefficients */
  2726. int i,j;
  2727. av_log(s->avctx, AV_LOG_DEBUG, "DCT coeffs of MB at %dx%d:\n", s->mb_x, s->mb_y);
  2728. for(i=0; i<6; i++){
  2729. for(j=0; j<64; j++){
  2730. av_log(s->avctx, AV_LOG_DEBUG, "%5d",
  2731. block[i][s->idsp.idct_permutation[j]]);
  2732. }
  2733. av_log(s->avctx, AV_LOG_DEBUG, "\n");
  2734. }
  2735. }
  2736. s->current_picture.qscale_table[mb_xy] = s->qscale;
  2737. /* update DC predictors for P macroblocks */
  2738. if (!s->mb_intra) {
  2739. if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
  2740. if(s->mbintra_table[mb_xy])
  2741. ff_clean_intra_table_entries(s);
  2742. } else {
  2743. s->last_dc[0] =
  2744. s->last_dc[1] =
  2745. s->last_dc[2] = 128 << s->intra_dc_precision;
  2746. }
  2747. }
  2748. else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
  2749. s->mbintra_table[mb_xy]=1;
  2750. if ( (s->flags&CODEC_FLAG_PSNR)
  2751. || s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor
  2752. || !(s->encoding && (s->intra_only || s->pict_type==AV_PICTURE_TYPE_B) && s->avctx->mb_decision != FF_MB_DECISION_RD)) { //FIXME precalc
  2753. uint8_t *dest_y, *dest_cb, *dest_cr;
  2754. int dct_linesize, dct_offset;
  2755. op_pixels_func (*op_pix)[4];
  2756. qpel_mc_func (*op_qpix)[16];
  2757. const int linesize = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
  2758. const int uvlinesize = s->current_picture.f->linesize[1];
  2759. const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
  2760. const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
  2761. /* avoid copy if macroblock skipped in last frame too */
  2762. /* skip only during decoding as we might trash the buffers during encoding a bit */
  2763. if(!s->encoding){
  2764. uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
  2765. if (s->mb_skipped) {
  2766. s->mb_skipped= 0;
  2767. av_assert2(s->pict_type!=AV_PICTURE_TYPE_I);
  2768. *mbskip_ptr = 1;
  2769. } else if(!s->current_picture.reference) {
  2770. *mbskip_ptr = 1;
  2771. } else{
  2772. *mbskip_ptr = 0; /* not skipped */
  2773. }
  2774. }
  2775. dct_linesize = linesize << s->interlaced_dct;
  2776. dct_offset = s->interlaced_dct ? linesize : linesize * block_size;
  2777. if(readable){
  2778. dest_y= s->dest[0];
  2779. dest_cb= s->dest[1];
  2780. dest_cr= s->dest[2];
  2781. }else{
  2782. dest_y = s->b_scratchpad;
  2783. dest_cb= s->b_scratchpad+16*linesize;
  2784. dest_cr= s->b_scratchpad+32*linesize;
  2785. }
  2786. if (!s->mb_intra) {
  2787. /* motion handling */
  2788. /* decoding or more than one mb_type (MC was already done otherwise) */
  2789. if(!s->encoding){
  2790. if(HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
  2791. if (s->mv_dir & MV_DIR_FORWARD) {
  2792. ff_thread_await_progress(&s->last_picture_ptr->tf,
  2793. ff_mpv_lowest_referenced_row(s, 0),
  2794. 0);
  2795. }
  2796. if (s->mv_dir & MV_DIR_BACKWARD) {
  2797. ff_thread_await_progress(&s->next_picture_ptr->tf,
  2798. ff_mpv_lowest_referenced_row(s, 1),
  2799. 0);
  2800. }
  2801. }
  2802. if(lowres_flag){
  2803. h264_chroma_mc_func *op_pix = s->h264chroma.put_h264_chroma_pixels_tab;
  2804. if (s->mv_dir & MV_DIR_FORWARD) {
  2805. MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix);
  2806. op_pix = s->h264chroma.avg_h264_chroma_pixels_tab;
  2807. }
  2808. if (s->mv_dir & MV_DIR_BACKWARD) {
  2809. MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix);
  2810. }
  2811. }else{
  2812. op_qpix = s->me.qpel_put;
  2813. if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
  2814. op_pix = s->hdsp.put_pixels_tab;
  2815. }else{
  2816. op_pix = s->hdsp.put_no_rnd_pixels_tab;
  2817. }
  2818. if (s->mv_dir & MV_DIR_FORWARD) {
  2819. ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix, op_qpix);
  2820. op_pix = s->hdsp.avg_pixels_tab;
  2821. op_qpix= s->me.qpel_avg;
  2822. }
  2823. if (s->mv_dir & MV_DIR_BACKWARD) {
  2824. ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix, op_qpix);
  2825. }
  2826. }
  2827. }
  2828. /* skip dequant / idct if we are really late ;) */
  2829. if(s->avctx->skip_idct){
  2830. if( (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B)
  2831. ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I)
  2832. || s->avctx->skip_idct >= AVDISCARD_ALL)
  2833. goto skip_idct;
  2834. }
  2835. /* add dct residue */
  2836. if(s->encoding || !( s->msmpeg4_version || s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO
  2837. || (s->codec_id==AV_CODEC_ID_MPEG4 && !s->mpeg_quant))){
  2838. add_dequant_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale);
  2839. add_dequant_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale);
  2840. add_dequant_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale);
  2841. add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
  2842. if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
  2843. if (s->chroma_y_shift){
  2844. add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
  2845. add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
  2846. }else{
  2847. dct_linesize >>= 1;
  2848. dct_offset >>=1;
  2849. add_dequant_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale);
  2850. add_dequant_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale);
  2851. add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
  2852. add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
  2853. }
  2854. }
  2855. } else if(is_mpeg12 || (s->codec_id != AV_CODEC_ID_WMV2)){
  2856. add_dct(s, block[0], 0, dest_y , dct_linesize);
  2857. add_dct(s, block[1], 1, dest_y + block_size, dct_linesize);
  2858. add_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize);
  2859. add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
  2860. if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
  2861. if(s->chroma_y_shift){//Chroma420
  2862. add_dct(s, block[4], 4, dest_cb, uvlinesize);
  2863. add_dct(s, block[5], 5, dest_cr, uvlinesize);
  2864. }else{
  2865. //chroma422
  2866. dct_linesize = uvlinesize << s->interlaced_dct;
  2867. dct_offset = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
  2868. add_dct(s, block[4], 4, dest_cb, dct_linesize);
  2869. add_dct(s, block[5], 5, dest_cr, dct_linesize);
  2870. add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
  2871. add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
  2872. if(!s->chroma_x_shift){//Chroma444
  2873. add_dct(s, block[8], 8, dest_cb+block_size, dct_linesize);
  2874. add_dct(s, block[9], 9, dest_cr+block_size, dct_linesize);
  2875. add_dct(s, block[10], 10, dest_cb+block_size+dct_offset, dct_linesize);
  2876. add_dct(s, block[11], 11, dest_cr+block_size+dct_offset, dct_linesize);
  2877. }
  2878. }
  2879. }//fi gray
  2880. }
  2881. else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
  2882. ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
  2883. }
  2884. } else {
  2885. /* dct only in intra block */
  2886. if(s->encoding || !(s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO)){
  2887. put_dct(s, block[0], 0, dest_y , dct_linesize, s->qscale);
  2888. put_dct(s, block[1], 1, dest_y + block_size, dct_linesize, s->qscale);
  2889. put_dct(s, block[2], 2, dest_y + dct_offset , dct_linesize, s->qscale);
  2890. put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
  2891. if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
  2892. if(s->chroma_y_shift){
  2893. put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
  2894. put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
  2895. }else{
  2896. dct_offset >>=1;
  2897. dct_linesize >>=1;
  2898. put_dct(s, block[4], 4, dest_cb, dct_linesize, s->chroma_qscale);
  2899. put_dct(s, block[5], 5, dest_cr, dct_linesize, s->chroma_qscale);
  2900. put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
  2901. put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
  2902. }
  2903. }
  2904. }else{
  2905. s->idsp.idct_put(dest_y, dct_linesize, block[0]);
  2906. s->idsp.idct_put(dest_y + block_size, dct_linesize, block[1]);
  2907. s->idsp.idct_put(dest_y + dct_offset, dct_linesize, block[2]);
  2908. s->idsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
  2909. if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
  2910. if(s->chroma_y_shift){
  2911. s->idsp.idct_put(dest_cb, uvlinesize, block[4]);
  2912. s->idsp.idct_put(dest_cr, uvlinesize, block[5]);
  2913. }else{
  2914. dct_linesize = uvlinesize << s->interlaced_dct;
  2915. dct_offset = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
  2916. s->idsp.idct_put(dest_cb, dct_linesize, block[4]);
  2917. s->idsp.idct_put(dest_cr, dct_linesize, block[5]);
  2918. s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
  2919. s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
  2920. if(!s->chroma_x_shift){//Chroma444
  2921. s->idsp.idct_put(dest_cb + block_size, dct_linesize, block[8]);
  2922. s->idsp.idct_put(dest_cr + block_size, dct_linesize, block[9]);
  2923. s->idsp.idct_put(dest_cb + block_size + dct_offset, dct_linesize, block[10]);
  2924. s->idsp.idct_put(dest_cr + block_size + dct_offset, dct_linesize, block[11]);
  2925. }
  2926. }
  2927. }//gray
  2928. }
  2929. }
  2930. skip_idct:
  2931. if(!readable){
  2932. s->hdsp.put_pixels_tab[0][0](s->dest[0], dest_y , linesize,16);
  2933. s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
  2934. s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
  2935. }
  2936. }
  2937. }
  2938. void ff_mpv_decode_mb(MpegEncContext *s, int16_t block[12][64])
  2939. {
  2940. #if !CONFIG_SMALL
  2941. if(s->out_format == FMT_MPEG1) {
  2942. if(s->avctx->lowres) mpv_decode_mb_internal(s, block, 1, 1);
  2943. else mpv_decode_mb_internal(s, block, 0, 1);
  2944. } else
  2945. #endif
  2946. if(s->avctx->lowres) mpv_decode_mb_internal(s, block, 1, 0);
  2947. else mpv_decode_mb_internal(s, block, 0, 0);
  2948. }
  2949. void ff_mpeg_draw_horiz_band(MpegEncContext *s, int y, int h)
  2950. {
  2951. ff_draw_horiz_band(s->avctx, s->current_picture_ptr->f,
  2952. s->last_picture_ptr ? s->last_picture_ptr->f : NULL, y, h, s->picture_structure,
  2953. s->first_field, s->low_delay);
  2954. }
  2955. void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
  2956. const int linesize = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
  2957. const int uvlinesize = s->current_picture.f->linesize[1];
  2958. const int mb_size= 4 - s->avctx->lowres;
  2959. s->block_index[0]= s->b8_stride*(s->mb_y*2 ) - 2 + s->mb_x*2;
  2960. s->block_index[1]= s->b8_stride*(s->mb_y*2 ) - 1 + s->mb_x*2;
  2961. s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
  2962. s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
  2963. s->block_index[4]= s->mb_stride*(s->mb_y + 1) + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
  2964. s->block_index[5]= s->mb_stride*(s->mb_y + s->mb_height + 2) + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
  2965. //block_index is not used by mpeg2, so it is not affected by chroma_format
  2966. s->dest[0] = s->current_picture.f->data[0] + ((s->mb_x - 1) << mb_size);
  2967. s->dest[1] = s->current_picture.f->data[1] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
  2968. s->dest[2] = s->current_picture.f->data[2] + ((s->mb_x - 1) << (mb_size - s->chroma_x_shift));
  2969. if(!(s->pict_type==AV_PICTURE_TYPE_B && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
  2970. {
  2971. if(s->picture_structure==PICT_FRAME){
  2972. s->dest[0] += s->mb_y * linesize << mb_size;
  2973. s->dest[1] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
  2974. s->dest[2] += s->mb_y * uvlinesize << (mb_size - s->chroma_y_shift);
  2975. }else{
  2976. s->dest[0] += (s->mb_y>>1) * linesize << mb_size;
  2977. s->dest[1] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
  2978. s->dest[2] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
  2979. av_assert1((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
  2980. }
  2981. }
  2982. }
  2983. /**
  2984. * Permute an 8x8 block.
  2985. * @param block the block which will be permuted according to the given permutation vector
  2986. * @param permutation the permutation vector
  2987. * @param last the last non zero coefficient in scantable order, used to speed the permutation up
  2988. * @param scantable the used scantable, this is only used to speed the permutation up, the block is not
  2989. * (inverse) permutated to scantable order!
  2990. */
  2991. void ff_block_permute(int16_t *block, uint8_t *permutation, const uint8_t *scantable, int last)
  2992. {
  2993. int i;
  2994. int16_t temp[64];
  2995. if(last<=0) return;
  2996. //if(permutation[1]==1) return; //FIXME it is ok but not clean and might fail for some permutations
  2997. for(i=0; i<=last; i++){
  2998. const int j= scantable[i];
  2999. temp[j]= block[j];
  3000. block[j]=0;
  3001. }
  3002. for(i=0; i<=last; i++){
  3003. const int j= scantable[i];
  3004. const int perm_j= permutation[j];
  3005. block[perm_j]= temp[j];
  3006. }
  3007. }
  3008. void ff_mpeg_flush(AVCodecContext *avctx){
  3009. int i;
  3010. MpegEncContext *s = avctx->priv_data;
  3011. if (!s || !s->picture)
  3012. return;
  3013. for (i = 0; i < MAX_PICTURE_COUNT; i++)
  3014. ff_mpeg_unref_picture(s, &s->picture[i]);
  3015. s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
  3016. ff_mpeg_unref_picture(s, &s->current_picture);
  3017. ff_mpeg_unref_picture(s, &s->last_picture);
  3018. ff_mpeg_unref_picture(s, &s->next_picture);
  3019. s->mb_x= s->mb_y= 0;
  3020. s->closed_gop= 0;
  3021. s->parse_context.state= -1;
  3022. s->parse_context.frame_start_found= 0;
  3023. s->parse_context.overread= 0;
  3024. s->parse_context.overread_index= 0;
  3025. s->parse_context.index= 0;
  3026. s->parse_context.last_index= 0;
  3027. s->bitstream_buffer_size=0;
  3028. s->pp_time=0;
  3029. }
  3030. /**
  3031. * set qscale and update qscale dependent variables.
  3032. */
  3033. void ff_set_qscale(MpegEncContext * s, int qscale)
  3034. {
  3035. if (qscale < 1)
  3036. qscale = 1;
  3037. else if (qscale > 31)
  3038. qscale = 31;
  3039. s->qscale = qscale;
  3040. s->chroma_qscale= s->chroma_qscale_table[qscale];
  3041. s->y_dc_scale= s->y_dc_scale_table[ qscale ];
  3042. s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
  3043. }
  3044. void ff_mpv_report_decode_progress(MpegEncContext *s)
  3045. {
  3046. if (s->pict_type != AV_PICTURE_TYPE_B && !s->partitioned_frame && !s->er.error_occurred)
  3047. ff_thread_report_progress(&s->current_picture_ptr->tf, s->mb_y, 0);
  3048. }