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.

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