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.

844 lines
27KB

  1. /*
  2. * This file is part of FFmpeg.
  3. *
  4. * FFmpeg is free software; you can redistribute it and/or
  5. * modify it under the terms of the GNU Lesser General Public
  6. * License as published by the Free Software Foundation; either
  7. * version 2.1 of the License, or (at your option) any later version.
  8. *
  9. * FFmpeg is distributed in the hope that it will be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12. * Lesser General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU Lesser General Public
  15. * License along with FFmpeg; if not, write to the Free Software
  16. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. /**
  19. * @file
  20. * @brief IntraX8 (J-Frame) subdecoder, used by WMV2 and VC-1
  21. */
  22. #include "libavutil/avassert.h"
  23. #include "avcodec.h"
  24. #include "get_bits.h"
  25. #include "idctdsp.h"
  26. #include "msmpeg4data.h"
  27. #include "intrax8huf.h"
  28. #include "intrax8.h"
  29. #include "intrax8dsp.h"
  30. #include "mpegutils.h"
  31. #define MAX_TABLE_DEPTH(table_bits, max_bits) \
  32. ((max_bits + table_bits - 1) / table_bits)
  33. #define DC_VLC_BITS 9
  34. #define AC_VLC_BITS 9
  35. #define OR_VLC_BITS 7
  36. #define DC_VLC_MTD MAX_TABLE_DEPTH(DC_VLC_BITS, MAX_DC_VLC_BITS)
  37. #define AC_VLC_MTD MAX_TABLE_DEPTH(AC_VLC_BITS, MAX_AC_VLC_BITS)
  38. #define OR_VLC_MTD MAX_TABLE_DEPTH(OR_VLC_BITS, MAX_OR_VLC_BITS)
  39. static VLC j_ac_vlc[2][2][8]; // [quant < 13], [intra / inter], [select]
  40. static VLC j_dc_vlc[2][8]; // [quant], [select]
  41. static VLC j_orient_vlc[2][4]; // [quant], [select]
  42. static av_cold int x8_vlc_init(void)
  43. {
  44. int i;
  45. int offset = 0;
  46. int sizeidx = 0;
  47. static const uint16_t sizes[8 * 4 + 8 * 2 + 2 + 4] = {
  48. 576, 548, 582, 618, 546, 616, 560, 642,
  49. 584, 582, 704, 664, 512, 544, 656, 640,
  50. 512, 648, 582, 566, 532, 614, 596, 648,
  51. 586, 552, 584, 590, 544, 578, 584, 624,
  52. 528, 528, 526, 528, 536, 528, 526, 544,
  53. 544, 512, 512, 528, 528, 544, 512, 544,
  54. 128, 128, 128, 128, 128, 128,
  55. };
  56. static VLC_TYPE table[28150][2];
  57. // set ac tables
  58. #define init_ac_vlc(dst, src) \
  59. do { \
  60. dst.table = &table[offset]; \
  61. dst.table_allocated = sizes[sizeidx]; \
  62. offset += sizes[sizeidx++]; \
  63. init_vlc(&dst, AC_VLC_BITS, 77, &src[1], 4, 2, &src[0], 4, 2, \
  64. INIT_VLC_USE_NEW_STATIC); \
  65. } while(0)
  66. for (i = 0; i < 8; i++) {
  67. init_ac_vlc(j_ac_vlc[0][0][i], x8_ac0_highquant_table[i][0]);
  68. init_ac_vlc(j_ac_vlc[0][1][i], x8_ac1_highquant_table[i][0]);
  69. init_ac_vlc(j_ac_vlc[1][0][i], x8_ac0_lowquant_table[i][0]);
  70. init_ac_vlc(j_ac_vlc[1][1][i], x8_ac1_lowquant_table[i][0]);
  71. }
  72. #undef init_ac_vlc
  73. // set dc tables
  74. #define init_dc_vlc(dst, src) \
  75. do { \
  76. dst.table = &table[offset]; \
  77. dst.table_allocated = sizes[sizeidx]; \
  78. offset += sizes[sizeidx++]; \
  79. init_vlc(&dst, DC_VLC_BITS, 34, &src[1], 4, 2, &src[0], 4, 2, \
  80. INIT_VLC_USE_NEW_STATIC); \
  81. } while(0)
  82. for (i = 0; i < 8; i++) {
  83. init_dc_vlc(j_dc_vlc[0][i], x8_dc_highquant_table[i][0]);
  84. init_dc_vlc(j_dc_vlc[1][i], x8_dc_lowquant_table[i][0]);
  85. }
  86. #undef init_dc_vlc
  87. // set orient tables
  88. #define init_or_vlc(dst, src) \
  89. do { \
  90. dst.table = &table[offset]; \
  91. dst.table_allocated = sizes[sizeidx]; \
  92. offset += sizes[sizeidx++]; \
  93. init_vlc(&dst, OR_VLC_BITS, 12, &src[1], 4, 2, &src[0], 4, 2, \
  94. INIT_VLC_USE_NEW_STATIC); \
  95. } while(0)
  96. for (i = 0; i < 2; i++)
  97. init_or_vlc(j_orient_vlc[0][i], x8_orient_highquant_table[i][0]);
  98. for (i = 0; i < 4; i++)
  99. init_or_vlc(j_orient_vlc[1][i], x8_orient_lowquant_table[i][0]);
  100. #undef init_or_vlc
  101. if (offset != sizeof(table) / sizeof(VLC_TYPE) / 2) {
  102. av_log(NULL, AV_LOG_ERROR, "table size %zd does not match needed %i\n",
  103. sizeof(table) / sizeof(VLC_TYPE) / 2, offset);
  104. return AVERROR_INVALIDDATA;
  105. }
  106. return 0;
  107. }
  108. static void x8_reset_vlc_tables(IntraX8Context *w)
  109. {
  110. memset(w->j_dc_vlc, 0, sizeof(w->j_dc_vlc));
  111. memset(w->j_ac_vlc, 0, sizeof(w->j_ac_vlc));
  112. w->j_orient_vlc = NULL;
  113. }
  114. static inline void x8_select_ac_table(IntraX8Context *const w, int mode)
  115. {
  116. int table_index;
  117. av_assert2(mode < 4);
  118. if (w->j_ac_vlc[mode])
  119. return;
  120. table_index = get_bits(w->gb, 3);
  121. // 2 modes use same tables
  122. w->j_ac_vlc[mode] = &j_ac_vlc[w->quant < 13][mode >> 1][table_index];
  123. av_assert2(w->j_ac_vlc[mode]);
  124. }
  125. static inline int x8_get_orient_vlc(IntraX8Context *w)
  126. {
  127. if (!w->j_orient_vlc) {
  128. int table_index = get_bits(w->gb, 1 + (w->quant < 13));
  129. w->j_orient_vlc = &j_orient_vlc[w->quant < 13][table_index];
  130. }
  131. return get_vlc2(w->gb, w->j_orient_vlc->table, OR_VLC_BITS, OR_VLC_MTD);
  132. }
  133. #define extra_bits(eb) (eb) // 3 bits
  134. #define extra_run (0xFF << 8) // 1 bit
  135. #define extra_level (0x00 << 8) // 1 bit
  136. #define run_offset(r) ((r) << 16) // 6 bits
  137. #define level_offset(l) ((l) << 24) // 5 bits
  138. static const uint32_t ac_decode_table[] = {
  139. /* 46 */ extra_bits(3) | extra_run | run_offset(16) | level_offset(0),
  140. /* 47 */ extra_bits(3) | extra_run | run_offset(24) | level_offset(0),
  141. /* 48 */ extra_bits(2) | extra_run | run_offset(4) | level_offset(1),
  142. /* 49 */ extra_bits(3) | extra_run | run_offset(8) | level_offset(1),
  143. /* 50 */ extra_bits(5) | extra_run | run_offset(32) | level_offset(0),
  144. /* 51 */ extra_bits(4) | extra_run | run_offset(16) | level_offset(1),
  145. /* 52 */ extra_bits(2) | extra_level | run_offset(0) | level_offset(4),
  146. /* 53 */ extra_bits(2) | extra_level | run_offset(0) | level_offset(8),
  147. /* 54 */ extra_bits(2) | extra_level | run_offset(0) | level_offset(12),
  148. /* 55 */ extra_bits(3) | extra_level | run_offset(0) | level_offset(16),
  149. /* 56 */ extra_bits(3) | extra_level | run_offset(0) | level_offset(24),
  150. /* 57 */ extra_bits(2) | extra_level | run_offset(1) | level_offset(3),
  151. /* 58 */ extra_bits(3) | extra_level | run_offset(1) | level_offset(7),
  152. /* 59 */ extra_bits(2) | extra_run | run_offset(16) | level_offset(0),
  153. /* 60 */ extra_bits(2) | extra_run | run_offset(20) | level_offset(0),
  154. /* 61 */ extra_bits(2) | extra_run | run_offset(24) | level_offset(0),
  155. /* 62 */ extra_bits(2) | extra_run | run_offset(28) | level_offset(0),
  156. /* 63 */ extra_bits(4) | extra_run | run_offset(32) | level_offset(0),
  157. /* 64 */ extra_bits(4) | extra_run | run_offset(48) | level_offset(0),
  158. /* 65 */ extra_bits(2) | extra_run | run_offset(4) | level_offset(1),
  159. /* 66 */ extra_bits(3) | extra_run | run_offset(8) | level_offset(1),
  160. /* 67 */ extra_bits(4) | extra_run | run_offset(16) | level_offset(1),
  161. /* 68 */ extra_bits(2) | extra_level | run_offset(0) | level_offset(4),
  162. /* 69 */ extra_bits(3) | extra_level | run_offset(0) | level_offset(8),
  163. /* 70 */ extra_bits(4) | extra_level | run_offset(0) | level_offset(16),
  164. /* 71 */ extra_bits(2) | extra_level | run_offset(1) | level_offset(3),
  165. /* 72 */ extra_bits(3) | extra_level | run_offset(1) | level_offset(7),
  166. };
  167. #undef extra_bits
  168. #undef extra_run
  169. #undef extra_level
  170. #undef run_offset
  171. #undef level_offset
  172. static void x8_get_ac_rlf(IntraX8Context *const w, const int mode,
  173. int *const run, int *const level, int *const final)
  174. {
  175. int i, e;
  176. // x8_select_ac_table(w, mode);
  177. i = get_vlc2(w->gb, w->j_ac_vlc[mode]->table, AC_VLC_BITS, AC_VLC_MTD);
  178. if (i < 46) { // [0-45]
  179. int t, l;
  180. if (i < 0) {
  181. *level =
  182. *final = // prevent 'may be used uninitialized'
  183. *run = 64; // this would cause error exit in the ac loop
  184. return;
  185. }
  186. /*
  187. * i == 0-15 r = 0-15 l = 0; r = i & %01111
  188. * i == 16-19 r = 0-3 l = 1; r = i & %00011
  189. * i == 20-21 r = 0-1 l = 2; r = i & %00001
  190. * i == 22 r = 0 l = 3; r = i & %00000
  191. */
  192. *final =
  193. t = i > 22;
  194. i -= 23 * t;
  195. /* l = lut_l[i / 2] = { 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 3 }[i >> 1];
  196. * 11 10'01 01'00 00'00 00'00 00'00 00 => 0xE50000 */
  197. l = (0xE50000 >> (i & 0x1E)) & 3; // 0x1E or ~1 or (i >> 1 << 1)
  198. /* t = lut_mask[l] = { 0x0f, 0x03, 0x01, 0x00 }[l];
  199. * as i < 256 the higher bits do not matter */
  200. t = 0x01030F >> (l << 3);
  201. *run = i & t;
  202. *level = l;
  203. } else if (i < 73) { // [46-72]
  204. uint32_t sm;
  205. uint32_t mask;
  206. i -= 46;
  207. sm = ac_decode_table[i];
  208. e = get_bits(w->gb, sm & 0xF);
  209. sm >>= 8; // 3 bits
  210. mask = sm & 0xff;
  211. sm >>= 8; // 1 bit
  212. *run = (sm & 0xff) + (e & mask); // 6 bits
  213. *level = (sm >> 8) + (e & ~mask); // 5 bits
  214. *final = i > (58 - 46);
  215. } else if (i < 75) { // [73-74]
  216. static const uint8_t crazy_mix_runlevel[32] = {
  217. 0x22, 0x32, 0x33, 0x53, 0x23, 0x42, 0x43, 0x63,
  218. 0x24, 0x52, 0x34, 0x73, 0x25, 0x62, 0x44, 0x83,
  219. 0x26, 0x72, 0x35, 0x54, 0x27, 0x82, 0x45, 0x64,
  220. 0x28, 0x92, 0x36, 0x74, 0x29, 0xa2, 0x46, 0x84,
  221. };
  222. *final = !(i & 1);
  223. e = get_bits(w->gb, 5); // get the extra bits
  224. *run = crazy_mix_runlevel[e] >> 4;
  225. *level = crazy_mix_runlevel[e] & 0x0F;
  226. } else {
  227. *level = get_bits(w->gb, 7 - 3 * (i & 1));
  228. *run = get_bits(w->gb, 6);
  229. *final = get_bits1(w->gb);
  230. }
  231. return;
  232. }
  233. /* static const uint8_t dc_extra_sbits[] = {
  234. * 0, 1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7,
  235. * }; */
  236. static const uint8_t dc_index_offset[] = {
  237. 0, 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
  238. };
  239. static int x8_get_dc_rlf(IntraX8Context *const w, const int mode,
  240. int *const level, int *const final)
  241. {
  242. int i, e, c;
  243. av_assert2(mode < 3);
  244. if (!w->j_dc_vlc[mode]) {
  245. int table_index = get_bits(w->gb, 3);
  246. // 4 modes, same table
  247. w->j_dc_vlc[mode] = &j_dc_vlc[w->quant < 13][table_index];
  248. }
  249. i = get_vlc2(w->gb, w->j_dc_vlc[mode]->table, DC_VLC_BITS, DC_VLC_MTD);
  250. /* (i >= 17) { i -= 17; final =1; } */
  251. c = i > 16;
  252. *final = c;
  253. i -= 17 * c;
  254. if (i <= 0) {
  255. *level = 0;
  256. return -i;
  257. }
  258. c = (i + 1) >> 1; // hackish way to calculate dc_extra_sbits[]
  259. c -= c > 1;
  260. e = get_bits(w->gb, c); // get the extra bits
  261. i = dc_index_offset[i] + (e >> 1);
  262. e = -(e & 1); // 0, 0xffffff
  263. *level = (i ^ e) - e; // (i ^ 0) - 0, (i ^ 0xff) - (-1)
  264. return 0;
  265. }
  266. // end of huffman
  267. static int x8_setup_spatial_predictor(IntraX8Context *const w, const int chroma)
  268. {
  269. int range;
  270. int sum;
  271. int quant;
  272. w->dsp.setup_spatial_compensation(w->dest[chroma], w->scratchpad,
  273. w->frame->linesize[chroma > 0],
  274. &range, &sum, w->edges);
  275. if (chroma) {
  276. w->orient = w->chroma_orient;
  277. quant = w->quant_dc_chroma;
  278. } else {
  279. quant = w->quant;
  280. }
  281. w->flat_dc = 0;
  282. if (range < quant || range < 3) {
  283. w->orient = 0;
  284. // yep you read right, a +-1 idct error may break decoding!
  285. if (range < 3) {
  286. w->flat_dc = 1;
  287. sum += 9;
  288. // ((1 << 17) + 9) / (8 + 8 + 1 + 2) = 6899
  289. w->predicted_dc = sum * 6899 >> 17;
  290. }
  291. }
  292. if (chroma)
  293. return 0;
  294. av_assert2(w->orient < 3);
  295. if (range < 2 * w->quant) {
  296. if ((w->edges & 3) == 0) {
  297. if (w->orient == 1)
  298. w->orient = 11;
  299. if (w->orient == 2)
  300. w->orient = 10;
  301. } else {
  302. w->orient = 0;
  303. }
  304. w->raw_orient = 0;
  305. } else {
  306. static const uint8_t prediction_table[3][12] = {
  307. { 0, 8, 4, 10, 11, 2, 6, 9, 1, 3, 5, 7 },
  308. { 4, 0, 8, 11, 10, 3, 5, 2, 6, 9, 1, 7 },
  309. { 8, 0, 4, 10, 11, 1, 7, 2, 6, 9, 3, 5 },
  310. };
  311. w->raw_orient = x8_get_orient_vlc(w);
  312. if (w->raw_orient < 0)
  313. return -1;
  314. av_assert2(w->raw_orient < 12);
  315. av_assert2(w->orient < 3);
  316. w->orient=prediction_table[w->orient][w->raw_orient];
  317. }
  318. return 0;
  319. }
  320. static void x8_update_predictions(IntraX8Context *const w, const int orient,
  321. const int est_run)
  322. {
  323. w->prediction_table[w->mb_x * 2 + (w->mb_y & 1)] = (est_run << 2) + 1 * (orient == 4) + 2 * (orient == 8);
  324. /*
  325. * y = 2n + 0 -> // 0 2 4
  326. * y = 2n + 1 -> // 1 3 5
  327. */
  328. }
  329. static void x8_get_prediction_chroma(IntraX8Context *const w)
  330. {
  331. w->edges = 1 * !(w->mb_x >> 1);
  332. w->edges |= 2 * !(w->mb_y >> 1);
  333. w->edges |= 4 * (w->mb_x >= (2 * w->mb_width - 1)); // mb_x for chroma would always be odd
  334. w->raw_orient = 0;
  335. // lut_co[8] = {inv,4,8,8, inv,4,8,8} <- => {1,1,0,0;1,1,0,0} => 0xCC
  336. if (w->edges & 3) {
  337. w->chroma_orient = 4 << ((0xCC >> w->edges) & 1);
  338. return;
  339. }
  340. // block[x - 1][y | 1 - 1)]
  341. w->chroma_orient = (w->prediction_table[2 * w->mb_x - 2] & 0x03) << 2;
  342. }
  343. static void x8_get_prediction(IntraX8Context *const w)
  344. {
  345. int a, b, c, i;
  346. w->edges = 1 * !w->mb_x;
  347. w->edges |= 2 * !w->mb_y;
  348. w->edges |= 4 * (w->mb_x >= (2 * w->mb_width - 1));
  349. switch (w->edges & 3) {
  350. case 0:
  351. break;
  352. case 1:
  353. // take the one from the above block[0][y - 1]
  354. w->est_run = w->prediction_table[!(w->mb_y & 1)] >> 2;
  355. w->orient = 1;
  356. return;
  357. case 2:
  358. // take the one from the previous block[x - 1][0]
  359. w->est_run = w->prediction_table[2 * w->mb_x - 2] >> 2;
  360. w->orient = 2;
  361. return;
  362. case 3:
  363. w->est_run = 16;
  364. w->orient = 0;
  365. return;
  366. }
  367. // no edge cases
  368. b = w->prediction_table[2 * w->mb_x + !(w->mb_y & 1)]; // block[x ][y - 1]
  369. a = w->prediction_table[2 * w->mb_x - 2 + (w->mb_y & 1)]; // block[x - 1][y ]
  370. c = w->prediction_table[2 * w->mb_x - 2 + !(w->mb_y & 1)]; // block[x - 1][y - 1]
  371. w->est_run = FFMIN(b, a);
  372. /* This condition has nothing to do with w->edges, even if it looks
  373. * similar it would trigger if e.g. x = 3; y = 2;
  374. * I guess somebody wrote something wrong and it became standard. */
  375. if ((w->mb_x & w->mb_y) != 0)
  376. w->est_run = FFMIN(c, w->est_run);
  377. w->est_run >>= 2;
  378. a &= 3;
  379. b &= 3;
  380. c &= 3;
  381. i = (0xFFEAF4C4 >> (2 * b + 8 * a)) & 3;
  382. if (i != 3)
  383. w->orient = i;
  384. else
  385. w->orient = (0xFFEAD8 >> (2 * c + 8 * (w->quant > 12))) & 3;
  386. /*
  387. * lut1[b][a] = {
  388. * ->{ 0, 1, 0, pad },
  389. * { 0, 1, X, pad },
  390. * { 2, 2, 2, pad }
  391. * }
  392. * pad 2 2 2;
  393. * pad X 1 0;
  394. * pad 0 1 0 <-
  395. * -> 11 10 '10 10 '11 11'01 00 '11 00'01 00 => 0xEAF4C4
  396. *
  397. * lut2[q>12][c] = {
  398. * ->{ 0, 2, 1, pad},
  399. * { 2, 2, 2, pad}
  400. * }
  401. * pad 2 2 2;
  402. * pad 1 2 0 <-
  403. * -> 11 10'10 10 '11 01'10 00 => 0xEAD8
  404. */
  405. }
  406. static void x8_ac_compensation(IntraX8Context *const w, const int direction,
  407. const int dc_level)
  408. {
  409. int t;
  410. #define B(x,y) w->block[0][w->idct_permutation[(x) + (y) * 8]]
  411. #define T(x) ((x) * dc_level + 0x8000) >> 16;
  412. switch (direction) {
  413. case 0:
  414. t = T(3811); // h
  415. B(1, 0) -= t;
  416. B(0, 1) -= t;
  417. t = T(487); // e
  418. B(2, 0) -= t;
  419. B(0, 2) -= t;
  420. t = T(506); // f
  421. B(3, 0) -= t;
  422. B(0, 3) -= t;
  423. t = T(135); // c
  424. B(4, 0) -= t;
  425. B(0, 4) -= t;
  426. B(2, 1) += t;
  427. B(1, 2) += t;
  428. B(3, 1) += t;
  429. B(1, 3) += t;
  430. t = T(173); // d
  431. B(5, 0) -= t;
  432. B(0, 5) -= t;
  433. t = T(61); // b
  434. B(6, 0) -= t;
  435. B(0, 6) -= t;
  436. B(5, 1) += t;
  437. B(1, 5) += t;
  438. t = T(42); // a
  439. B(7, 0) -= t;
  440. B(0, 7) -= t;
  441. B(4, 1) += t;
  442. B(1, 4) += t;
  443. B(4, 4) += t;
  444. t = T(1084); // g
  445. B(1, 1) += t;
  446. w->block_last_index[0] = FFMAX(w->block_last_index[0], 7 * 8);
  447. break;
  448. case 1:
  449. B(0, 1) -= T(6269);
  450. B(0, 3) -= T(708);
  451. B(0, 5) -= T(172);
  452. B(0, 7) -= T(73);
  453. w->block_last_index[0] = FFMAX(w->block_last_index[0], 7 * 8);
  454. break;
  455. case 2:
  456. B(1, 0) -= T(6269);
  457. B(3, 0) -= T(708);
  458. B(5, 0) -= T(172);
  459. B(7, 0) -= T(73);
  460. w->block_last_index[0] = FFMAX(w->block_last_index[0], 7);
  461. break;
  462. }
  463. #undef B
  464. #undef T
  465. }
  466. static void dsp_x8_put_solidcolor(const uint8_t pix, uint8_t *dst,
  467. const int linesize)
  468. {
  469. int k;
  470. for (k = 0; k < 8; k++) {
  471. memset(dst, pix, 8);
  472. dst += linesize;
  473. }
  474. }
  475. static const int16_t quant_table[64] = {
  476. 256, 256, 256, 256, 256, 256, 259, 262,
  477. 265, 269, 272, 275, 278, 282, 285, 288,
  478. 292, 295, 299, 303, 306, 310, 314, 317,
  479. 321, 325, 329, 333, 337, 341, 345, 349,
  480. 353, 358, 362, 366, 371, 375, 379, 384,
  481. 389, 393, 398, 403, 408, 413, 417, 422,
  482. 428, 433, 438, 443, 448, 454, 459, 465,
  483. 470, 476, 482, 488, 493, 499, 505, 511,
  484. };
  485. static int x8_decode_intra_mb(IntraX8Context *const w, const int chroma)
  486. {
  487. uint8_t *scantable;
  488. int final, run, level;
  489. int ac_mode, dc_mode, est_run, dc_level;
  490. int pos, n;
  491. int zeros_only;
  492. int use_quant_matrix;
  493. int sign;
  494. av_assert2(w->orient < 12);
  495. w->bdsp.clear_block(w->block[0]);
  496. if (chroma)
  497. dc_mode = 2;
  498. else
  499. dc_mode = !!w->est_run; // 0, 1
  500. if (x8_get_dc_rlf(w, dc_mode, &dc_level, &final))
  501. return -1;
  502. n = 0;
  503. zeros_only = 0;
  504. if (!final) { // decode ac
  505. use_quant_matrix = w->use_quant_matrix;
  506. if (chroma) {
  507. ac_mode = 1;
  508. est_run = 64; // not used
  509. } else {
  510. if (w->raw_orient < 3)
  511. use_quant_matrix = 0;
  512. if (w->raw_orient > 4) {
  513. ac_mode = 0;
  514. est_run = 64;
  515. } else {
  516. if (w->est_run > 1) {
  517. ac_mode = 2;
  518. est_run = w->est_run;
  519. } else {
  520. ac_mode = 3;
  521. est_run = 64;
  522. }
  523. }
  524. }
  525. x8_select_ac_table(w, ac_mode);
  526. /* scantable_selector[12] = { 0, 2, 0, 1, 1, 1, 0, 2, 2, 0, 1, 2 }; <-
  527. * -> 10'01' 00'10' 10'00' 01'01' 01'00' 10'00 => 0x928548 */
  528. scantable = w->scantable[(0x928548 >> (2 * w->orient)) & 3].permutated;
  529. pos = 0;
  530. do {
  531. n++;
  532. if (n >= est_run) {
  533. ac_mode = 3;
  534. x8_select_ac_table(w, 3);
  535. }
  536. x8_get_ac_rlf(w, ac_mode, &run, &level, &final);
  537. pos += run + 1;
  538. if (pos > 63) {
  539. // this also handles vlc error in x8_get_ac_rlf
  540. return -1;
  541. }
  542. level = (level + 1) * w->dquant;
  543. level += w->qsum;
  544. sign = -get_bits1(w->gb);
  545. level = (level ^ sign) - sign;
  546. if (use_quant_matrix)
  547. level = (level * quant_table[pos]) >> 8;
  548. w->block[0][scantable[pos]] = level;
  549. } while (!final);
  550. w->block_last_index[0] = pos;
  551. } else { // DC only
  552. w->block_last_index[0] = 0;
  553. if (w->flat_dc && ((unsigned) (dc_level + 1)) < 3) { // [-1; 1]
  554. int32_t divide_quant = !chroma ? w->divide_quant_dc_luma
  555. : w->divide_quant_dc_chroma;
  556. int32_t dc_quant = !chroma ? w->quant
  557. : w->quant_dc_chroma;
  558. // original intent dc_level += predicted_dc/quant;
  559. // but it got lost somewhere in the rounding
  560. dc_level += (w->predicted_dc * divide_quant + (1 << 12)) >> 13;
  561. dsp_x8_put_solidcolor(av_clip_uint8((dc_level * dc_quant + 4) >> 3),
  562. w->dest[chroma],
  563. w->frame->linesize[!!chroma]);
  564. goto block_placed;
  565. }
  566. zeros_only = dc_level == 0;
  567. }
  568. if (!chroma)
  569. w->block[0][0] = dc_level * w->quant;
  570. else
  571. w->block[0][0] = dc_level * w->quant_dc_chroma;
  572. // there is !zero_only check in the original, but dc_level check is enough
  573. if ((unsigned int) (dc_level + 1) >= 3 && (w->edges & 3) != 3) {
  574. int direction;
  575. /* ac_comp_direction[orient] = { 0, 3, 3, 1, 1, 0, 0, 0, 2, 2, 2, 1 }; <-
  576. * -> 01'10' 10'10' 00'00' 00'01' 01'11' 11'00 => 0x6A017C */
  577. direction = (0x6A017C >> (w->orient * 2)) & 3;
  578. if (direction != 3) {
  579. // modify block_last[]
  580. x8_ac_compensation(w, direction, w->block[0][0]);
  581. }
  582. }
  583. if (w->flat_dc) {
  584. dsp_x8_put_solidcolor(w->predicted_dc, w->dest[chroma],
  585. w->frame->linesize[!!chroma]);
  586. } else {
  587. w->dsp.spatial_compensation[w->orient](w->scratchpad,
  588. w->dest[chroma],
  589. w->frame->linesize[!!chroma]);
  590. }
  591. if (!zeros_only)
  592. w->wdsp.idct_add(w->dest[chroma],
  593. w->frame->linesize[!!chroma],
  594. w->block[0]);
  595. block_placed:
  596. if (!chroma)
  597. x8_update_predictions(w, w->orient, n);
  598. if (w->loopfilter) {
  599. uint8_t *ptr = w->dest[chroma];
  600. int linesize = w->frame->linesize[!!chroma];
  601. if (!((w->edges & 2) || (zeros_only && (w->orient | 4) == 4)))
  602. w->dsp.h_loop_filter(ptr, linesize, w->quant);
  603. if (!((w->edges & 1) || (zeros_only && (w->orient | 8) == 8)))
  604. w->dsp.v_loop_filter(ptr, linesize, w->quant);
  605. }
  606. return 0;
  607. }
  608. // FIXME maybe merge with ff_*
  609. static void x8_init_block_index(IntraX8Context *w, AVFrame *frame)
  610. {
  611. // not parent codec linesize as this would be wrong for field pics
  612. // not that IntraX8 has interlacing support ;)
  613. const int linesize = frame->linesize[0];
  614. const int uvlinesize = frame->linesize[1];
  615. w->dest[0] = frame->data[0];
  616. w->dest[1] = frame->data[1];
  617. w->dest[2] = frame->data[2];
  618. w->dest[0] += w->mb_y * linesize << 3;
  619. // chroma blocks are on add rows
  620. w->dest[1] += (w->mb_y & ~1) * uvlinesize << 2;
  621. w->dest[2] += (w->mb_y & ~1) * uvlinesize << 2;
  622. }
  623. av_cold int ff_intrax8_common_init(AVCodecContext *avctx,
  624. IntraX8Context *w, IDCTDSPContext *idsp,
  625. int16_t (*block)[64],
  626. int block_last_index[12],
  627. int mb_width, int mb_height)
  628. {
  629. int ret = x8_vlc_init();
  630. if (ret < 0)
  631. return ret;
  632. w->avctx = avctx;
  633. w->idsp = *idsp;
  634. w->mb_width = mb_width;
  635. w->mb_height = mb_height;
  636. w->block = block;
  637. w->block_last_index = block_last_index;
  638. // two rows, 2 blocks per cannon mb
  639. w->prediction_table = av_mallocz(w->mb_width * 2 * 2);
  640. if (!w->prediction_table)
  641. return AVERROR(ENOMEM);
  642. ff_wmv2dsp_init(&w->wdsp);
  643. ff_init_scantable_permutation(w->idct_permutation,
  644. w->wdsp.idct_perm);
  645. ff_init_scantable(w->idct_permutation, &w->scantable[0],
  646. ff_wmv1_scantable[0]);
  647. ff_init_scantable(w->idct_permutation, &w->scantable[1],
  648. ff_wmv1_scantable[2]);
  649. ff_init_scantable(w->idct_permutation, &w->scantable[2],
  650. ff_wmv1_scantable[3]);
  651. ff_intrax8dsp_init(&w->dsp);
  652. ff_blockdsp_init(&w->bdsp, avctx);
  653. return 0;
  654. }
  655. av_cold void ff_intrax8_common_end(IntraX8Context *w)
  656. {
  657. av_freep(&w->prediction_table);
  658. }
  659. int ff_intrax8_decode_picture(IntraX8Context *const w, Picture *pict,
  660. GetBitContext *gb, int *mb_x, int *mb_y,
  661. int dquant, int quant_offset,
  662. int loopfilter, int lowdelay)
  663. {
  664. int mb_xy;
  665. w->gb = gb;
  666. w->dquant = dquant;
  667. w->quant = dquant >> 1;
  668. w->qsum = quant_offset;
  669. w->frame = pict->f;
  670. w->loopfilter = loopfilter;
  671. w->use_quant_matrix = get_bits1(w->gb);
  672. w->mb_x = *mb_x;
  673. w->mb_y = *mb_y;
  674. w->divide_quant_dc_luma = ((1 << 16) + (w->quant >> 1)) / w->quant;
  675. if (w->quant < 5) {
  676. w->quant_dc_chroma = w->quant;
  677. w->divide_quant_dc_chroma = w->divide_quant_dc_luma;
  678. } else {
  679. w->quant_dc_chroma = w->quant + ((w->quant + 3) >> 3);
  680. w->divide_quant_dc_chroma = ((1 << 16) + (w->quant_dc_chroma >> 1)) / w->quant_dc_chroma;
  681. }
  682. x8_reset_vlc_tables(w);
  683. for (w->mb_y = 0; w->mb_y < w->mb_height * 2; w->mb_y++) {
  684. x8_init_block_index(w, w->frame);
  685. mb_xy = (w->mb_y >> 1) * (w->mb_width + 1);
  686. for (w->mb_x = 0; w->mb_x < w->mb_width * 2; w->mb_x++) {
  687. x8_get_prediction(w);
  688. if (x8_setup_spatial_predictor(w, 0))
  689. goto error;
  690. if (x8_decode_intra_mb(w, 0))
  691. goto error;
  692. if (w->mb_x & w->mb_y & 1) {
  693. x8_get_prediction_chroma(w);
  694. /* when setting up chroma, no vlc is read,
  695. * so no error condition can be reached */
  696. x8_setup_spatial_predictor(w, 1);
  697. if (x8_decode_intra_mb(w, 1))
  698. goto error;
  699. x8_setup_spatial_predictor(w, 2);
  700. if (x8_decode_intra_mb(w, 2))
  701. goto error;
  702. w->dest[1] += 8;
  703. w->dest[2] += 8;
  704. pict->qscale_table[mb_xy] = w->quant;
  705. mb_xy++;
  706. }
  707. w->dest[0] += 8;
  708. }
  709. if (w->mb_y & 1)
  710. ff_draw_horiz_band(w->avctx, w->frame, w->frame,
  711. (w->mb_y - 1) * 8, 16,
  712. PICT_FRAME, 0, lowdelay);
  713. }
  714. error:
  715. *mb_x = w->mb_x;
  716. *mb_y = w->mb_y;
  717. return 0;
  718. }