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.

793 lines
25KB

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