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 "get_bits.h"
  24. #include "mpegvideo.h"
  25. #include "msmpeg4data.h"
  26. #include "intrax8huf.h"
  27. #include "intrax8.h"
  28. #include "intrax8dsp.h"
  29. #define MAX_TABLE_DEPTH(table_bits, max_bits) ((max_bits+table_bits-1)/table_bits)
  30. #define DC_VLC_BITS 9
  31. #define AC_VLC_BITS 9
  32. #define OR_VLC_BITS 7
  33. #define DC_VLC_MTD MAX_TABLE_DEPTH(DC_VLC_BITS, MAX_DC_VLC_BITS)
  34. #define AC_VLC_MTD MAX_TABLE_DEPTH(AC_VLC_BITS, MAX_AC_VLC_BITS)
  35. #define OR_VLC_MTD MAX_TABLE_DEPTH(OR_VLC_BITS, MAX_OR_VLC_BITS)
  36. static VLC j_ac_vlc[2][2][8]; //[quant<13],[intra/inter],[select]
  37. static VLC j_dc_vlc[2][8]; //[quant], [select]
  38. static VLC j_orient_vlc[2][4]; //[quant], [select]
  39. static av_cold void x8_vlc_init(void){
  40. int i;
  41. int offset = 0;
  42. int sizeidx = 0;
  43. static const uint16_t sizes[8*4 + 8*2 + 2 + 4] = {
  44. 576, 548, 582, 618, 546, 616, 560, 642,
  45. 584, 582, 704, 664, 512, 544, 656, 640,
  46. 512, 648, 582, 566, 532, 614, 596, 648,
  47. 586, 552, 584, 590, 544, 578, 584, 624,
  48. 528, 528, 526, 528, 536, 528, 526, 544,
  49. 544, 512, 512, 528, 528, 544, 512, 544,
  50. 128, 128, 128, 128, 128, 128};
  51. static VLC_TYPE table[28150][2];
  52. #define init_ac_vlc(dst,src) \
  53. dst.table = &table[offset]; \
  54. dst.table_allocated = sizes[sizeidx]; \
  55. offset += sizes[sizeidx++]; \
  56. init_vlc(&dst, \
  57. AC_VLC_BITS,77, \
  58. &src[1],4,2, \
  59. &src[0],4,2, \
  60. INIT_VLC_USE_NEW_STATIC)
  61. //set ac tables
  62. for(i=0;i<8;i++){
  63. init_ac_vlc( j_ac_vlc[0][0][i], x8_ac0_highquant_table[i][0] );
  64. init_ac_vlc( j_ac_vlc[0][1][i], x8_ac1_highquant_table[i][0] );
  65. init_ac_vlc( j_ac_vlc[1][0][i], x8_ac0_lowquant_table [i][0] );
  66. init_ac_vlc( j_ac_vlc[1][1][i], x8_ac1_lowquant_table [i][0] );
  67. }
  68. #undef init_ac_vlc
  69. //set dc tables
  70. #define init_dc_vlc(dst,src) \
  71. dst.table = &table[offset]; \
  72. dst.table_allocated = sizes[sizeidx]; \
  73. offset += sizes[sizeidx++]; \
  74. init_vlc(&dst, \
  75. DC_VLC_BITS,34, \
  76. &src[1],4,2, \
  77. &src[0],4,2, \
  78. INIT_VLC_USE_NEW_STATIC);
  79. for(i=0;i<8;i++){
  80. init_dc_vlc( j_dc_vlc[0][i], x8_dc_highquant_table[i][0]);
  81. init_dc_vlc( j_dc_vlc[1][i], x8_dc_lowquant_table [i][0]);
  82. }
  83. #undef init_dc_vlc
  84. //set orient tables
  85. #define init_or_vlc(dst,src) \
  86. dst.table = &table[offset]; \
  87. dst.table_allocated = sizes[sizeidx]; \
  88. offset += sizes[sizeidx++]; \
  89. init_vlc(&dst, \
  90. OR_VLC_BITS,12, \
  91. &src[1],4,2, \
  92. &src[0],4,2, \
  93. INIT_VLC_USE_NEW_STATIC);
  94. for(i=0;i<2;i++){
  95. init_or_vlc( j_orient_vlc[0][i], x8_orient_highquant_table[i][0]);
  96. }
  97. for(i=0;i<4;i++){
  98. init_or_vlc( j_orient_vlc[1][i], x8_orient_lowquant_table [i][0])
  99. }
  100. if (offset != sizeof(table)/sizeof(VLC_TYPE)/2)
  101. av_log(NULL, AV_LOG_ERROR, "table size %i does not match needed %i\n", (int)(sizeof(table)/sizeof(VLC_TYPE)/2), offset);
  102. }
  103. #undef init_or_vlc
  104. static void x8_reset_vlc_tables(IntraX8Context * w){
  105. memset(w->j_dc_vlc,0,sizeof(w->j_dc_vlc));
  106. memset(w->j_ac_vlc,0,sizeof(w->j_ac_vlc));
  107. w->j_orient_vlc=NULL;
  108. }
  109. static inline void x8_select_ac_table(IntraX8Context * const w , int mode){
  110. MpegEncContext * const s= w->s;
  111. int table_index;
  112. assert(mode<4);
  113. if( w->j_ac_vlc[mode] ) return;
  114. table_index = get_bits(&s->gb, 3);
  115. w->j_ac_vlc[mode] = &j_ac_vlc[w->quant<13][mode>>1][table_index];//2 modes use same tables
  116. assert(w->j_ac_vlc[mode]);
  117. }
  118. static inline int x8_get_orient_vlc(IntraX8Context * w){
  119. MpegEncContext * const s= w->s;
  120. int table_index;
  121. if(!w->j_orient_vlc ){
  122. table_index = get_bits(&s->gb, 1+(w->quant<13) );
  123. w->j_orient_vlc = &j_orient_vlc[w->quant<13][table_index];
  124. }
  125. assert(w->j_orient_vlc);
  126. assert(w->j_orient_vlc->table);
  127. return get_vlc2(&s->gb, w->j_orient_vlc->table, OR_VLC_BITS, OR_VLC_MTD);
  128. }
  129. #define extra_bits(eb) (eb)
  130. #define extra_run (0xFF<<8)
  131. #define extra_level (0x00<<8)
  132. #define run_offset(r) ((r)<<16)
  133. #define level_offset(l) ((l)<<24)
  134. static const uint32_t ac_decode_table[]={
  135. /*46*/ extra_bits(3) | extra_run | run_offset(16) | level_offset( 0),
  136. /*47*/ extra_bits(3) | extra_run | run_offset(24) | level_offset( 0),
  137. /*48*/ extra_bits(2) | extra_run | run_offset( 4) | level_offset( 1),
  138. /*49*/ extra_bits(3) | extra_run | run_offset( 8) | level_offset( 1),
  139. /*50*/ extra_bits(5) | extra_run | run_offset(32) | level_offset( 0),
  140. /*51*/ extra_bits(4) | extra_run | run_offset(16) | level_offset( 1),
  141. /*52*/ extra_bits(2) | extra_level | run_offset( 0) | level_offset( 4),
  142. /*53*/ extra_bits(2) | extra_level | run_offset( 0) | level_offset( 8),
  143. /*54*/ extra_bits(2) | extra_level | run_offset( 0) | level_offset(12),
  144. /*55*/ extra_bits(3) | extra_level | run_offset( 0) | level_offset(16),
  145. /*56*/ extra_bits(3) | extra_level | run_offset( 0) | level_offset(24),
  146. /*57*/ extra_bits(2) | extra_level | run_offset( 1) | level_offset( 3),
  147. /*58*/ extra_bits(3) | extra_level | run_offset( 1) | level_offset( 7),
  148. /*59*/ extra_bits(2) | extra_run | run_offset(16) | level_offset( 0),
  149. /*60*/ extra_bits(2) | extra_run | run_offset(20) | level_offset( 0),
  150. /*61*/ extra_bits(2) | extra_run | run_offset(24) | level_offset( 0),
  151. /*62*/ extra_bits(2) | extra_run | run_offset(28) | level_offset( 0),
  152. /*63*/ extra_bits(4) | extra_run | run_offset(32) | level_offset( 0),
  153. /*64*/ extra_bits(4) | extra_run | run_offset(48) | level_offset( 0),
  154. /*65*/ extra_bits(2) | extra_run | run_offset( 4) | level_offset( 1),
  155. /*66*/ extra_bits(3) | extra_run | run_offset( 8) | level_offset( 1),
  156. /*67*/ extra_bits(4) | extra_run | run_offset(16) | level_offset( 1),
  157. /*68*/ extra_bits(2) | extra_level | run_offset( 0) | level_offset( 4),
  158. /*69*/ extra_bits(3) | extra_level | run_offset( 0) | level_offset( 8),
  159. /*70*/ extra_bits(4) | extra_level | run_offset( 0) | level_offset(16),
  160. /*71*/ extra_bits(2) | extra_level | run_offset( 1) | level_offset( 3),
  161. /*72*/ extra_bits(3) | extra_level | run_offset( 1) | level_offset( 7),
  162. };
  163. //extra_bits = 3bits; extra_run/level = 1 bit; run_offset = 6bits; level_offset = 5 bits;
  164. #undef extra_bits
  165. #undef extra_run
  166. #undef extra_level
  167. #undef run_offset
  168. #undef level_offset
  169. static void x8_get_ac_rlf(IntraX8Context * const w, const int mode,
  170. int * const run, int * const level, int * const final){
  171. MpegEncContext * const s= w->s;
  172. int i,e;
  173. // x8_select_ac_table(w,mode);
  174. i = get_vlc2(&s->gb, w->j_ac_vlc[mode]->table, AC_VLC_BITS, AC_VLC_MTD);
  175. if(i<46){ //[0-45]
  176. int t,l;
  177. if(i<0){
  178. (*level)=(*final)=//prevent 'may be used unilitialized'
  179. (*run)=64;//this would cause error exit in the ac loop
  180. return;
  181. }
  182. (*final) = t = (i>22);
  183. i-=23*t;
  184. /*
  185. i== 0-15 r=0-15 l=0 ;r=i& %01111
  186. i==16-19 r=0-3 l=1 ;r=i& %00011
  187. i==20-21 r=0-1 l=2 ;r=i& %00001
  188. i==22 r=0 l=3 ;r=i& %00000
  189. 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
  190. t=lut_mask[l]={0x0f,0x03,0x01,0x00}[l]; as i<256 the higher bits do not matter */
  191. l=(0xE50000>>(i&(0x1E)))&3;/*0x1E or (~1) or ((i>>1)<<1)*/
  192. t=(0x01030F>>(l<<3));
  193. (*run) = i&t;
  194. (*level) = l;
  195. }else if(i<73){//[46-72]
  196. uint32_t sm;
  197. uint32_t mask;
  198. i-=46;
  199. sm=ac_decode_table[i];
  200. e=get_bits(&s->gb,sm&0xF);sm>>=8;//3bits
  201. mask=sm&0xff;sm>>=8; //1bit
  202. (*run) =(sm&0xff) + (e&( mask));//6bits
  203. (*level)=(sm>>8) + (e&(~mask));//5bits
  204. (*final)=i>(58-46);
  205. }else if(i<75){//[73-74]
  206. static const uint8_t crazy_mix_runlevel[32]={
  207. 0x22,0x32,0x33,0x53,0x23,0x42,0x43,0x63,
  208. 0x24,0x52,0x34,0x73,0x25,0x62,0x44,0x83,
  209. 0x26,0x72,0x35,0x54,0x27,0x82,0x45,0x64,
  210. 0x28,0x92,0x36,0x74,0x29,0xa2,0x46,0x84};
  211. (*final)=!(i&1);
  212. e=get_bits(&s->gb,5);//get the extra bits
  213. (*run) =crazy_mix_runlevel[e]>>4;
  214. (*level)=crazy_mix_runlevel[e]&0x0F;
  215. }else{
  216. (*level)=get_bits( &s->gb, 7-3*(i&1));
  217. (*run) =get_bits( &s->gb, 6);
  218. (*final)=get_bits1(&s->gb);
  219. }
  220. return;
  221. }
  222. //static const uint8_t dc_extra_sbits[] ={0, 1,1, 1,1, 2,2, 3,3, 4,4, 5,5, 6,6, 7,7 };
  223. static const uint8_t dc_index_offset[] ={ 0, 1,2, 3,4, 5,7, 9,13, 17,25, 33,49, 65,97, 129,193};
  224. static int x8_get_dc_rlf(IntraX8Context * const w,int const mode, int * const level, int * const final){
  225. MpegEncContext * const s= w->s;
  226. int i,e,c;
  227. assert(mode<3);
  228. if( !w->j_dc_vlc[mode] ) {
  229. int table_index;
  230. table_index = get_bits(&s->gb, 3);
  231. //4 modes, same table
  232. w->j_dc_vlc[mode]= &j_dc_vlc[w->quant<13][table_index];
  233. }
  234. assert(w->j_dc_vlc);
  235. assert(w->j_dc_vlc[mode]->table);
  236. i=get_vlc2(&s->gb, w->j_dc_vlc[mode]->table, DC_VLC_BITS, DC_VLC_MTD);
  237. /*(i>=17) {i-=17;final=1;}*/
  238. c= i>16;
  239. (*final)=c;
  240. i-=17*c;
  241. if(i<=0){
  242. (*level)=0;
  243. return -i;
  244. }
  245. c=(i+1)>>1;//hackish way to calculate dc_extra_sbits[]
  246. c-=c>1;
  247. e=get_bits(&s->gb,c);//get the extra bits
  248. i=dc_index_offset[i]+(e>>1);
  249. e= -(e & 1);//0,0xffffff
  250. (*level)= (i ^ e) - e;// (i^0)-0 , (i^0xff)-(-1)
  251. return 0;
  252. }
  253. //end of huffman
  254. static int x8_setup_spatial_predictor(IntraX8Context * const w, const int chroma){
  255. MpegEncContext * const s= w->s;
  256. int range;
  257. int sum;
  258. int quant;
  259. w->dsp.setup_spatial_compensation(s->dest[chroma], s->edge_emu_buffer,
  260. s->current_picture.f.linesize[chroma>0],
  261. &range, &sum, w->edges);
  262. if(chroma){
  263. w->orient=w->chroma_orient;
  264. quant=w->quant_dc_chroma;
  265. }else{
  266. quant=w->quant;
  267. }
  268. w->flat_dc=0;
  269. if(range < quant || range < 3){
  270. w->orient=0;
  271. if(range < 3){//yep you read right, a +-1 idct error may break decoding!
  272. w->flat_dc=1;
  273. sum+=9;
  274. w->predicted_dc = (sum*6899)>>17;//((1<<17)+9)/(8+8+1+2)=6899
  275. }
  276. }
  277. if(chroma)
  278. return 0;
  279. assert(w->orient < 3);
  280. if(range < 2*w->quant){
  281. if( (w->edges&3) == 0){
  282. if(w->orient==1) w->orient=11;
  283. if(w->orient==2) w->orient=10;
  284. }else{
  285. w->orient=0;
  286. }
  287. w->raw_orient=0;
  288. }else{
  289. static const uint8_t prediction_table[3][12]={
  290. {0,8,4, 10,11, 2,6,9,1,3,5,7},
  291. {4,0,8, 11,10, 3,5,2,6,9,1,7},
  292. {8,0,4, 10,11, 1,7,2,6,9,3,5}
  293. };
  294. w->raw_orient=x8_get_orient_vlc(w);
  295. if(w->raw_orient<0) return -1;
  296. assert(w->raw_orient < 12 );
  297. assert(w->orient<3);
  298. w->orient=prediction_table[w->orient][w->raw_orient];
  299. }
  300. return 0;
  301. }
  302. static void x8_update_predictions(IntraX8Context * const w, const int orient, const int est_run ){
  303. MpegEncContext * const s= w->s;
  304. w->prediction_table[s->mb_x*2+(s->mb_y&1)] = (est_run<<2) + 1*(orient==4) + 2*(orient==8);
  305. /*
  306. y=2n+0 ->//0 2 4
  307. y=2n+1 ->//1 3 5
  308. */
  309. }
  310. static void x8_get_prediction_chroma(IntraX8Context * const w){
  311. MpegEncContext * const s= w->s;
  312. w->edges = 1*( !(s->mb_x>>1) );
  313. w->edges|= 2*( !(s->mb_y>>1) );
  314. w->edges|= 4*( s->mb_x >= (2*s->mb_width-1) );//mb_x for chroma would always be odd
  315. w->raw_orient=0;
  316. if(w->edges&3){//lut_co[8]={inv,4,8,8, inv,4,8,8}<- =>{1,1,0,0;1,1,0,0} => 0xCC
  317. w->chroma_orient=4<<((0xCC>>w->edges)&1);
  318. return;
  319. }
  320. w->chroma_orient = (w->prediction_table[2*s->mb_x-2] & 0x03)<<2;//block[x-1][y|1-1)]
  321. }
  322. static void x8_get_prediction(IntraX8Context * const w){
  323. MpegEncContext * const s= w->s;
  324. int a,b,c,i;
  325. w->edges = 1*( !s->mb_x );
  326. w->edges|= 2*( !s->mb_y );
  327. w->edges|= 4*( s->mb_x >= (2*s->mb_width-1) );
  328. switch(w->edges&3){
  329. case 0:
  330. break;
  331. case 1:
  332. //take the one from the above block[0][y-1]
  333. w->est_run = w->prediction_table[!(s->mb_y&1)]>>2;
  334. w->orient = 1;
  335. return;
  336. case 2:
  337. //take the one from the previous block[x-1][0]
  338. w->est_run = w->prediction_table[2*s->mb_x-2]>>2;
  339. w->orient = 2;
  340. return;
  341. case 3:
  342. w->est_run = 16;
  343. w->orient = 0;
  344. return;
  345. }
  346. //no edge cases
  347. b= w->prediction_table[2*s->mb_x + !(s->mb_y&1) ];//block[x ][y-1]
  348. a= w->prediction_table[2*s->mb_x-2 + (s->mb_y&1) ];//block[x-1][y ]
  349. c= w->prediction_table[2*s->mb_x-2 + !(s->mb_y&1) ];//block[x-1][y-1]
  350. w->est_run = FFMIN(b,a);
  351. /* This condition has nothing to do with w->edges, even if it looks
  352. similar it would trigger if e.g. x=3;y=2;
  353. I guess somebody wrote something wrong and it became standard. */
  354. if( (s->mb_x & s->mb_y) != 0 ) w->est_run=FFMIN(c,w->est_run);
  355. w->est_run>>=2;
  356. a&=3;
  357. b&=3;
  358. c&=3;
  359. i=( 0xFFEAF4C4>>(2*b+8*a) )&3;
  360. if(i!=3) w->orient=i;
  361. else w->orient=( 0xFFEAD8>>(2*c+8*(w->quant>12)) )&3;
  362. /*
  363. lut1[b][a]={
  364. ->{0, 1, 0, pad},
  365. {0, 1, X, pad},
  366. {2, 2, 2, pad}}
  367. pad 2 2 2; pad X 1 0; pad 0 1 0 <-
  368. -> 11 10 '10 10 '11 11'01 00 '11 00'01 00=>0xEAF4C4
  369. lut2[q>12][c]={
  370. ->{0,2,1,pad},
  371. {2,2,2,pad}}
  372. pad 2 2 2; pad 1 2 0 <-
  373. -> 11 10'10 10 '11 01'10 00=>0xEAD8
  374. */
  375. }
  376. static void x8_ac_compensation(IntraX8Context * const w, int const direction, int const dc_level){
  377. MpegEncContext * const s= w->s;
  378. int t;
  379. #define B(x,y) s->block[0][s->dsp.idct_permutation[(x)+(y)*8]]
  380. #define T(x) ((x) * dc_level + 0x8000) >> 16;
  381. switch(direction){
  382. case 0:
  383. t = T(3811);//h
  384. B(1,0) -= t;
  385. B(0,1) -= t;
  386. t = T(487);//e
  387. B(2,0) -= t;
  388. B(0,2) -= t;
  389. t = T(506);//f
  390. B(3,0) -= t;
  391. B(0,3) -= t;
  392. t = T(135);//c
  393. B(4,0) -= t;
  394. B(0,4) -= t;
  395. B(2,1) += t;
  396. B(1,2) += t;
  397. B(3,1) += t;
  398. B(1,3) += t;
  399. t = T(173);//d
  400. B(5,0) -= t;
  401. B(0,5) -= t;
  402. t = T(61);//b
  403. B(6,0) -= t;
  404. B(0,6) -= t;
  405. B(5,1) += t;
  406. B(1,5) += t;
  407. t = T(42); //a
  408. B(7,0) -= t;
  409. B(0,7) -= t;
  410. B(4,1) += t;
  411. B(1,4) += t;
  412. B(4,4) += t;
  413. t = T(1084);//g
  414. B(1,1) += t;
  415. s->block_last_index[0] = FFMAX(s->block_last_index[0], 7*8);
  416. break;
  417. case 1:
  418. B(0,1) -= T(6269);
  419. B(0,3) -= T( 708);
  420. B(0,5) -= T( 172);
  421. B(0,7) -= T( 73);
  422. s->block_last_index[0] = FFMAX(s->block_last_index[0], 7*8);
  423. break;
  424. case 2:
  425. B(1,0) -= T(6269);
  426. B(3,0) -= T( 708);
  427. B(5,0) -= T( 172);
  428. B(7,0) -= T( 73);
  429. s->block_last_index[0] = FFMAX(s->block_last_index[0], 7);
  430. break;
  431. }
  432. #undef B
  433. #undef T
  434. }
  435. static void dsp_x8_put_solidcolor(uint8_t const pix, uint8_t * dst, int const linesize){
  436. int k;
  437. for(k=0;k<8;k++){
  438. memset(dst,pix,8);
  439. dst+=linesize;
  440. }
  441. }
  442. static const int16_t quant_table[64] = {
  443. 256, 256, 256, 256, 256, 256, 259, 262,
  444. 265, 269, 272, 275, 278, 282, 285, 288,
  445. 292, 295, 299, 303, 306, 310, 314, 317,
  446. 321, 325, 329, 333, 337, 341, 345, 349,
  447. 353, 358, 362, 366, 371, 375, 379, 384,
  448. 389, 393, 398, 403, 408, 413, 417, 422,
  449. 428, 433, 438, 443, 448, 454, 459, 465,
  450. 470, 476, 482, 488, 493, 499, 505, 511
  451. };
  452. static int x8_decode_intra_mb(IntraX8Context* const w, const int chroma){
  453. MpegEncContext * const s= w->s;
  454. uint8_t * scantable;
  455. int final,run,level;
  456. int ac_mode,dc_mode,est_run,dc_level;
  457. int pos,n;
  458. int zeros_only;
  459. int use_quant_matrix;
  460. int sign;
  461. assert(w->orient<12);
  462. s->dsp.clear_block(s->block[0]);
  463. if(chroma){
  464. dc_mode=2;
  465. }else{
  466. dc_mode=!!w->est_run;//0,1
  467. }
  468. if(x8_get_dc_rlf(w, dc_mode, &dc_level, &final)) return -1;
  469. n=0;
  470. zeros_only=0;
  471. if(!final){//decode ac
  472. use_quant_matrix=w->use_quant_matrix;
  473. if(chroma){
  474. ac_mode = 1;
  475. est_run = 64;//not used
  476. }else{
  477. if (w->raw_orient < 3){
  478. use_quant_matrix = 0;
  479. }
  480. if(w->raw_orient > 4){
  481. ac_mode = 0;
  482. est_run = 64;
  483. }else{
  484. if(w->est_run > 1){
  485. ac_mode = 2;
  486. est_run=w->est_run;
  487. }else{
  488. ac_mode = 3;
  489. est_run = 64;
  490. }
  491. }
  492. }
  493. x8_select_ac_table(w,ac_mode);
  494. /*scantable_selector[12]={0,2,0,1,1,1,0,2,2,0,1,2};<-
  495. -> 10'01' 00'10' 10'00' 01'01' 01'00' 10'00 =>0x928548 */
  496. scantable = w->scantable[ (0x928548>>(2*w->orient))&3 ].permutated;
  497. pos=0;
  498. do {
  499. n++;
  500. if( n >= est_run ){
  501. ac_mode=3;
  502. x8_select_ac_table(w,3);
  503. }
  504. x8_get_ac_rlf(w,ac_mode,&run,&level,&final);
  505. pos+=run+1;
  506. if(pos>63){
  507. //this also handles vlc error in x8_get_ac_rlf
  508. return -1;
  509. }
  510. level= (level+1) * w->dquant;
  511. level+= w->qsum;
  512. sign = - get_bits1(&s->gb);
  513. level = (level ^ sign) - sign;
  514. if(use_quant_matrix){
  515. level = (level*quant_table[pos])>>8;
  516. }
  517. s->block[0][ scantable[pos] ]=level;
  518. }while(!final);
  519. s->block_last_index[0]=pos;
  520. }else{//DC only
  521. s->block_last_index[0]=0;
  522. if(w->flat_dc && ((unsigned)(dc_level+1)) < 3){//[-1;1]
  523. int32_t divide_quant= !chroma ? w->divide_quant_dc_luma:
  524. w->divide_quant_dc_chroma;
  525. int32_t dc_quant = !chroma ? w->quant:
  526. w->quant_dc_chroma;
  527. //original intent dc_level+=predicted_dc/quant; but it got lost somewhere in the rounding
  528. dc_level+= (w->predicted_dc*divide_quant + (1<<12) )>>13;
  529. dsp_x8_put_solidcolor( av_clip_uint8((dc_level*dc_quant+4)>>3),
  530. s->dest[chroma], s->current_picture.f.linesize[!!chroma]);
  531. goto block_placed;
  532. }
  533. zeros_only = (dc_level == 0);
  534. }
  535. if(!chroma){
  536. s->block[0][0] = dc_level*w->quant;
  537. }else{
  538. s->block[0][0] = dc_level*w->quant_dc_chroma;
  539. }
  540. //there is !zero_only check in the original, but dc_level check is enough
  541. if( (unsigned int)(dc_level+1) >= 3 && (w->edges&3) != 3 ){
  542. int direction;
  543. /*ac_comp_direction[orient] = { 0, 3, 3, 1, 1, 0, 0, 0, 2, 2, 2, 1 };<-
  544. -> 01'10' 10'10' 00'00' 00'01' 01'11' 11'00 =>0x6A017C */
  545. direction= (0x6A017C>>(w->orient*2))&3;
  546. if (direction != 3){
  547. x8_ac_compensation(w, direction, s->block[0][0]);//modify block_last[]
  548. }
  549. }
  550. if(w->flat_dc){
  551. dsp_x8_put_solidcolor(w->predicted_dc, s->dest[chroma], s->current_picture.f.linesize[!!chroma]);
  552. }else{
  553. w->dsp.spatial_compensation[w->orient]( s->edge_emu_buffer,
  554. s->dest[chroma],
  555. s->current_picture.f.linesize[!!chroma] );
  556. }
  557. if(!zeros_only)
  558. s->dsp.idct_add ( s->dest[chroma],
  559. s->current_picture.f.linesize[!!chroma],
  560. s->block[0] );
  561. block_placed:
  562. if(!chroma){
  563. x8_update_predictions(w,w->orient,n);
  564. }
  565. if(s->loop_filter){
  566. uint8_t* ptr = s->dest[chroma];
  567. int linesize = s->current_picture.f.linesize[!!chroma];
  568. if(!( (w->edges&2) || ( zeros_only && (w->orient|4)==4 ) )){
  569. w->dsp.h_loop_filter(ptr, linesize, w->quant);
  570. }
  571. if(!( (w->edges&1) || ( zeros_only && (w->orient|8)==8 ) )){
  572. w->dsp.v_loop_filter(ptr, linesize, w->quant);
  573. }
  574. }
  575. return 0;
  576. }
  577. static void x8_init_block_index(MpegEncContext *s){ //FIXME maybe merge with ff_*
  578. //not s->linesize as this would be wrong for field pics
  579. //not that IntraX8 has interlacing support ;)
  580. const int linesize = s->current_picture.f.linesize[0];
  581. const int uvlinesize = s->current_picture.f.linesize[1];
  582. s->dest[0] = s->current_picture.f.data[0];
  583. s->dest[1] = s->current_picture.f.data[1];
  584. s->dest[2] = s->current_picture.f.data[2];
  585. s->dest[0] += s->mb_y * linesize << 3;
  586. s->dest[1] += ( s->mb_y&(~1) ) * uvlinesize << 2;//chroma blocks are on add rows
  587. s->dest[2] += ( s->mb_y&(~1) ) * uvlinesize << 2;
  588. }
  589. /**
  590. * Initialize IntraX8 frame decoder.
  591. * Requires valid MpegEncContext with valid s->mb_width before calling.
  592. * @param w pointer to IntraX8Context
  593. * @param s pointer to MpegEncContext of the parent codec
  594. */
  595. av_cold void ff_intrax8_common_init(IntraX8Context * w, MpegEncContext * const s){
  596. w->s=s;
  597. x8_vlc_init();
  598. assert(s->mb_width>0);
  599. w->prediction_table=av_mallocz(s->mb_width*2*2);//two rows, 2 blocks per cannon mb
  600. ff_init_scantable(s->dsp.idct_permutation, &w->scantable[0], ff_wmv1_scantable[0]);
  601. ff_init_scantable(s->dsp.idct_permutation, &w->scantable[1], ff_wmv1_scantable[2]);
  602. ff_init_scantable(s->dsp.idct_permutation, &w->scantable[2], ff_wmv1_scantable[3]);
  603. ff_intrax8dsp_init(&w->dsp);
  604. }
  605. /**
  606. * Destroy IntraX8 frame structure.
  607. * @param w pointer to IntraX8Context
  608. */
  609. av_cold void ff_intrax8_common_end(IntraX8Context * w)
  610. {
  611. av_freep(&w->prediction_table);
  612. }
  613. /**
  614. * Decode single IntraX8 frame.
  615. * The parent codec must fill s->loopfilter and s->gb (bitstream).
  616. * The parent codec must call MPV_frame_start(), ff_er_frame_start() before calling this function.
  617. * The parent codec must call ff_er_frame_end(), MPV_frame_end() after calling this function.
  618. * This function does not use MPV_decode_mb().
  619. * @param w pointer to IntraX8Context
  620. * @param dquant doubled quantizer, it would be odd in case of VC-1 halfpq==1.
  621. * @param quant_offset offset away from zero
  622. */
  623. //FIXME extern uint8_t ff_wmv3_dc_scale_table[32];
  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.f.qscale_table[mb_xy] = w->quant;
  664. mb_xy++;
  665. }
  666. s->dest[0]+= 8;
  667. }
  668. if(s->mb_y&1){
  669. ff_draw_horiz_band(s, (s->mb_y-1)*8, 16);
  670. }
  671. }
  672. error:
  673. ff_er_add_slice(s, 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. }