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.

789 lines
25KB

  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 "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. av_assert2(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. av_assert2(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. return get_vlc2(&s->gb, w->j_orient_vlc->table, OR_VLC_BITS, OR_VLC_MTD);
  127. }
  128. #define extra_bits(eb) (eb)
  129. #define extra_run (0xFF<<8)
  130. #define extra_level (0x00<<8)
  131. #define run_offset(r) ((r)<<16)
  132. #define level_offset(l) ((l)<<24)
  133. static const uint32_t ac_decode_table[]={
  134. /*46*/ extra_bits(3) | extra_run | run_offset(16) | level_offset( 0),
  135. /*47*/ extra_bits(3) | extra_run | run_offset(24) | level_offset( 0),
  136. /*48*/ extra_bits(2) | extra_run | run_offset( 4) | level_offset( 1),
  137. /*49*/ extra_bits(3) | extra_run | run_offset( 8) | level_offset( 1),
  138. /*50*/ extra_bits(5) | extra_run | run_offset(32) | level_offset( 0),
  139. /*51*/ extra_bits(4) | extra_run | run_offset(16) | level_offset( 1),
  140. /*52*/ extra_bits(2) | extra_level | run_offset( 0) | level_offset( 4),
  141. /*53*/ extra_bits(2) | extra_level | run_offset( 0) | level_offset( 8),
  142. /*54*/ extra_bits(2) | extra_level | run_offset( 0) | level_offset(12),
  143. /*55*/ extra_bits(3) | extra_level | run_offset( 0) | level_offset(16),
  144. /*56*/ extra_bits(3) | extra_level | run_offset( 0) | level_offset(24),
  145. /*57*/ extra_bits(2) | extra_level | run_offset( 1) | level_offset( 3),
  146. /*58*/ extra_bits(3) | extra_level | run_offset( 1) | level_offset( 7),
  147. /*59*/ extra_bits(2) | extra_run | run_offset(16) | level_offset( 0),
  148. /*60*/ extra_bits(2) | extra_run | run_offset(20) | level_offset( 0),
  149. /*61*/ extra_bits(2) | extra_run | run_offset(24) | level_offset( 0),
  150. /*62*/ extra_bits(2) | extra_run | run_offset(28) | level_offset( 0),
  151. /*63*/ extra_bits(4) | extra_run | run_offset(32) | level_offset( 0),
  152. /*64*/ extra_bits(4) | extra_run | run_offset(48) | level_offset( 0),
  153. /*65*/ extra_bits(2) | extra_run | run_offset( 4) | level_offset( 1),
  154. /*66*/ extra_bits(3) | extra_run | run_offset( 8) | level_offset( 1),
  155. /*67*/ extra_bits(4) | extra_run | run_offset(16) | level_offset( 1),
  156. /*68*/ extra_bits(2) | extra_level | run_offset( 0) | level_offset( 4),
  157. /*69*/ extra_bits(3) | extra_level | run_offset( 0) | level_offset( 8),
  158. /*70*/ extra_bits(4) | extra_level | run_offset( 0) | level_offset(16),
  159. /*71*/ extra_bits(2) | extra_level | run_offset( 1) | level_offset( 3),
  160. /*72*/ extra_bits(3) | extra_level | run_offset( 1) | level_offset( 7),
  161. };
  162. //extra_bits = 3bits; extra_run/level = 1 bit; run_offset = 6bits; level_offset = 5 bits;
  163. #undef extra_bits
  164. #undef extra_run
  165. #undef extra_level
  166. #undef run_offset
  167. #undef level_offset
  168. static void x8_get_ac_rlf(IntraX8Context * const w, const int mode,
  169. int * const run, int * const level, int * const final){
  170. MpegEncContext * const s= w->s;
  171. int i,e;
  172. // x8_select_ac_table(w,mode);
  173. i = get_vlc2(&s->gb, w->j_ac_vlc[mode]->table, AC_VLC_BITS, AC_VLC_MTD);
  174. if(i<46){ //[0-45]
  175. int t,l;
  176. if(i<0){
  177. (*level)=(*final)=//prevent 'may be used unilitialized'
  178. (*run)=64;//this would cause error exit in the ac loop
  179. return;
  180. }
  181. (*final) = t = (i>22);
  182. i-=23*t;
  183. /*
  184. i== 0-15 r=0-15 l=0 ;r=i& %01111
  185. i==16-19 r=0-3 l=1 ;r=i& %00011
  186. i==20-21 r=0-1 l=2 ;r=i& %00001
  187. i==22 r=0 l=3 ;r=i& %00000
  188. 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
  189. t=lut_mask[l]={0x0f,0x03,0x01,0x00}[l]; as i<256 the higher bits do not matter */
  190. l=(0xE50000>>(i&(0x1E)))&3;/*0x1E or (~1) or ((i>>1)<<1)*/
  191. t=(0x01030F>>(l<<3));
  192. (*run) = i&t;
  193. (*level) = l;
  194. }else if(i<73){//[46-72]
  195. uint32_t sm;
  196. uint32_t mask;
  197. i-=46;
  198. sm=ac_decode_table[i];
  199. e=get_bits(&s->gb,sm&0xF);sm>>=8;//3bits
  200. mask=sm&0xff;sm>>=8; //1bit
  201. (*run) =(sm&0xff) + (e&( mask));//6bits
  202. (*level)=(sm>>8) + (e&(~mask));//5bits
  203. (*final)=i>(58-46);
  204. }else if(i<75){//[73-74]
  205. static const uint8_t crazy_mix_runlevel[32]={
  206. 0x22,0x32,0x33,0x53,0x23,0x42,0x43,0x63,
  207. 0x24,0x52,0x34,0x73,0x25,0x62,0x44,0x83,
  208. 0x26,0x72,0x35,0x54,0x27,0x82,0x45,0x64,
  209. 0x28,0x92,0x36,0x74,0x29,0xa2,0x46,0x84};
  210. (*final)=!(i&1);
  211. e=get_bits(&s->gb,5);//get the extra bits
  212. (*run) =crazy_mix_runlevel[e]>>4;
  213. (*level)=crazy_mix_runlevel[e]&0x0F;
  214. }else{
  215. (*level)=get_bits( &s->gb, 7-3*(i&1));
  216. (*run) =get_bits( &s->gb, 6);
  217. (*final)=get_bits1(&s->gb);
  218. }
  219. return;
  220. }
  221. //static const uint8_t dc_extra_sbits[] ={0, 1,1, 1,1, 2,2, 3,3, 4,4, 5,5, 6,6, 7,7 };
  222. static const uint8_t dc_index_offset[] ={ 0, 1,2, 3,4, 5,7, 9,13, 17,25, 33,49, 65,97, 129,193};
  223. static int x8_get_dc_rlf(IntraX8Context * const w,int const mode, int * const level, int * const final){
  224. MpegEncContext * const s= w->s;
  225. int i,e,c;
  226. av_assert2(mode<3);
  227. if( !w->j_dc_vlc[mode] ) {
  228. int table_index;
  229. table_index = get_bits(&s->gb, 3);
  230. //4 modes, same table
  231. w->j_dc_vlc[mode]= &j_dc_vlc[w->quant<13][table_index];
  232. }
  233. i=get_vlc2(&s->gb, w->j_dc_vlc[mode]->table, DC_VLC_BITS, DC_VLC_MTD);
  234. /*(i>=17) {i-=17;final=1;}*/
  235. c= i>16;
  236. (*final)=c;
  237. i-=17*c;
  238. if(i<=0){
  239. (*level)=0;
  240. return -i;
  241. }
  242. c=(i+1)>>1;//hackish way to calculate dc_extra_sbits[]
  243. c-=c>1;
  244. e=get_bits(&s->gb,c);//get the extra bits
  245. i=dc_index_offset[i]+(e>>1);
  246. e= -(e & 1);//0,0xffffff
  247. (*level)= (i ^ e) - e;// (i^0)-0 , (i^0xff)-(-1)
  248. return 0;
  249. }
  250. //end of huffman
  251. static int x8_setup_spatial_predictor(IntraX8Context * const w, const int chroma){
  252. MpegEncContext * const s= w->s;
  253. int range;
  254. int sum;
  255. int quant;
  256. w->dsp.setup_spatial_compensation(s->dest[chroma], s->edge_emu_buffer,
  257. s->current_picture.f.linesize[chroma>0],
  258. &range, &sum, w->edges);
  259. if(chroma){
  260. w->orient=w->chroma_orient;
  261. quant=w->quant_dc_chroma;
  262. }else{
  263. quant=w->quant;
  264. }
  265. w->flat_dc=0;
  266. if(range < quant || range < 3){
  267. w->orient=0;
  268. if(range < 3){//yep you read right, a +-1 idct error may break decoding!
  269. w->flat_dc=1;
  270. sum+=9;
  271. w->predicted_dc = (sum*6899)>>17;//((1<<17)+9)/(8+8+1+2)=6899
  272. }
  273. }
  274. if(chroma)
  275. return 0;
  276. av_assert2(w->orient < 3);
  277. if(range < 2*w->quant){
  278. if( (w->edges&3) == 0){
  279. if(w->orient==1) w->orient=11;
  280. if(w->orient==2) w->orient=10;
  281. }else{
  282. w->orient=0;
  283. }
  284. w->raw_orient=0;
  285. }else{
  286. static const uint8_t prediction_table[3][12]={
  287. {0,8,4, 10,11, 2,6,9,1,3,5,7},
  288. {4,0,8, 11,10, 3,5,2,6,9,1,7},
  289. {8,0,4, 10,11, 1,7,2,6,9,3,5}
  290. };
  291. w->raw_orient=x8_get_orient_vlc(w);
  292. if(w->raw_orient<0) return -1;
  293. av_assert2(w->raw_orient < 12 );
  294. av_assert2(w->orient<3);
  295. w->orient=prediction_table[w->orient][w->raw_orient];
  296. }
  297. return 0;
  298. }
  299. static void x8_update_predictions(IntraX8Context * const w, const int orient, const int est_run ){
  300. MpegEncContext * const s= w->s;
  301. w->prediction_table[s->mb_x*2+(s->mb_y&1)] = (est_run<<2) + 1*(orient==4) + 2*(orient==8);
  302. /*
  303. y=2n+0 ->//0 2 4
  304. y=2n+1 ->//1 3 5
  305. */
  306. }
  307. static void x8_get_prediction_chroma(IntraX8Context * const w){
  308. MpegEncContext * const s= w->s;
  309. w->edges = 1*( !(s->mb_x>>1) );
  310. w->edges|= 2*( !(s->mb_y>>1) );
  311. w->edges|= 4*( s->mb_x >= (2*s->mb_width-1) );//mb_x for chroma would always be odd
  312. w->raw_orient=0;
  313. if(w->edges&3){//lut_co[8]={inv,4,8,8, inv,4,8,8}<- =>{1,1,0,0;1,1,0,0} => 0xCC
  314. w->chroma_orient=4<<((0xCC>>w->edges)&1);
  315. return;
  316. }
  317. w->chroma_orient = (w->prediction_table[2*s->mb_x-2] & 0x03)<<2;//block[x-1][y|1-1)]
  318. }
  319. static void x8_get_prediction(IntraX8Context * const w){
  320. MpegEncContext * const s= w->s;
  321. int a,b,c,i;
  322. w->edges = 1*( !s->mb_x );
  323. w->edges|= 2*( !s->mb_y );
  324. w->edges|= 4*( s->mb_x >= (2*s->mb_width-1) );
  325. switch(w->edges&3){
  326. case 0:
  327. break;
  328. case 1:
  329. //take the one from the above block[0][y-1]
  330. w->est_run = w->prediction_table[!(s->mb_y&1)]>>2;
  331. w->orient = 1;
  332. return;
  333. case 2:
  334. //take the one from the previous block[x-1][0]
  335. w->est_run = w->prediction_table[2*s->mb_x-2]>>2;
  336. w->orient = 2;
  337. return;
  338. case 3:
  339. w->est_run = 16;
  340. w->orient = 0;
  341. return;
  342. }
  343. //no edge cases
  344. b= w->prediction_table[2*s->mb_x + !(s->mb_y&1) ];//block[x ][y-1]
  345. a= w->prediction_table[2*s->mb_x-2 + (s->mb_y&1) ];//block[x-1][y ]
  346. c= w->prediction_table[2*s->mb_x-2 + !(s->mb_y&1) ];//block[x-1][y-1]
  347. w->est_run = FFMIN(b,a);
  348. /* This condition has nothing to do with w->edges, even if it looks
  349. similar it would trigger if e.g. x=3;y=2;
  350. I guess somebody wrote something wrong and it became standard. */
  351. if( (s->mb_x & s->mb_y) != 0 ) w->est_run=FFMIN(c,w->est_run);
  352. w->est_run>>=2;
  353. a&=3;
  354. b&=3;
  355. c&=3;
  356. i=( 0xFFEAF4C4>>(2*b+8*a) )&3;
  357. if(i!=3) w->orient=i;
  358. else w->orient=( 0xFFEAD8>>(2*c+8*(w->quant>12)) )&3;
  359. /*
  360. lut1[b][a]={
  361. ->{0, 1, 0, pad},
  362. {0, 1, X, pad},
  363. {2, 2, 2, pad}}
  364. pad 2 2 2; pad X 1 0; pad 0 1 0 <-
  365. -> 11 10 '10 10 '11 11'01 00 '11 00'01 00=>0xEAF4C4
  366. lut2[q>12][c]={
  367. ->{0,2,1,pad},
  368. {2,2,2,pad}}
  369. pad 2 2 2; pad 1 2 0 <-
  370. -> 11 10'10 10 '11 01'10 00=>0xEAD8
  371. */
  372. }
  373. static void x8_ac_compensation(IntraX8Context * const w, int const direction, int const dc_level){
  374. MpegEncContext * const s= w->s;
  375. int t;
  376. #define B(x,y) s->block[0][s->dsp.idct_permutation[(x)+(y)*8]]
  377. #define T(x) ((x) * dc_level + 0x8000) >> 16;
  378. switch(direction){
  379. case 0:
  380. t = T(3811);//h
  381. B(1,0) -= t;
  382. B(0,1) -= t;
  383. t = T(487);//e
  384. B(2,0) -= t;
  385. B(0,2) -= t;
  386. t = T(506);//f
  387. B(3,0) -= t;
  388. B(0,3) -= t;
  389. t = T(135);//c
  390. B(4,0) -= t;
  391. B(0,4) -= t;
  392. B(2,1) += t;
  393. B(1,2) += t;
  394. B(3,1) += t;
  395. B(1,3) += t;
  396. t = T(173);//d
  397. B(5,0) -= t;
  398. B(0,5) -= t;
  399. t = T(61);//b
  400. B(6,0) -= t;
  401. B(0,6) -= t;
  402. B(5,1) += t;
  403. B(1,5) += t;
  404. t = T(42); //a
  405. B(7,0) -= t;
  406. B(0,7) -= t;
  407. B(4,1) += t;
  408. B(1,4) += t;
  409. B(4,4) += t;
  410. t = T(1084);//g
  411. B(1,1) += t;
  412. s->block_last_index[0] = FFMAX(s->block_last_index[0], 7*8);
  413. break;
  414. case 1:
  415. B(0,1) -= T(6269);
  416. B(0,3) -= T( 708);
  417. B(0,5) -= T( 172);
  418. B(0,7) -= T( 73);
  419. s->block_last_index[0] = FFMAX(s->block_last_index[0], 7*8);
  420. break;
  421. case 2:
  422. B(1,0) -= T(6269);
  423. B(3,0) -= T( 708);
  424. B(5,0) -= T( 172);
  425. B(7,0) -= T( 73);
  426. s->block_last_index[0] = FFMAX(s->block_last_index[0], 7);
  427. break;
  428. }
  429. #undef B
  430. #undef T
  431. }
  432. static void dsp_x8_put_solidcolor(uint8_t const pix, uint8_t * dst, int const linesize){
  433. int k;
  434. for(k=0;k<8;k++){
  435. memset(dst,pix,8);
  436. dst+=linesize;
  437. }
  438. }
  439. static const int16_t quant_table[64] = {
  440. 256, 256, 256, 256, 256, 256, 259, 262,
  441. 265, 269, 272, 275, 278, 282, 285, 288,
  442. 292, 295, 299, 303, 306, 310, 314, 317,
  443. 321, 325, 329, 333, 337, 341, 345, 349,
  444. 353, 358, 362, 366, 371, 375, 379, 384,
  445. 389, 393, 398, 403, 408, 413, 417, 422,
  446. 428, 433, 438, 443, 448, 454, 459, 465,
  447. 470, 476, 482, 488, 493, 499, 505, 511
  448. };
  449. static int x8_decode_intra_mb(IntraX8Context* const w, const int chroma){
  450. MpegEncContext * const s= w->s;
  451. uint8_t * scantable;
  452. int final,run,level;
  453. int ac_mode,dc_mode,est_run,dc_level;
  454. int pos,n;
  455. int zeros_only;
  456. int use_quant_matrix;
  457. int sign;
  458. av_assert2(w->orient<12);
  459. s->dsp.clear_block(s->block[0]);
  460. if(chroma){
  461. dc_mode=2;
  462. }else{
  463. dc_mode=!!w->est_run;//0,1
  464. }
  465. if(x8_get_dc_rlf(w, dc_mode, &dc_level, &final)) return -1;
  466. n=0;
  467. zeros_only=0;
  468. if(!final){//decode ac
  469. use_quant_matrix=w->use_quant_matrix;
  470. if(chroma){
  471. ac_mode = 1;
  472. est_run = 64;//not used
  473. }else{
  474. if (w->raw_orient < 3){
  475. use_quant_matrix = 0;
  476. }
  477. if(w->raw_orient > 4){
  478. ac_mode = 0;
  479. est_run = 64;
  480. }else{
  481. if(w->est_run > 1){
  482. ac_mode = 2;
  483. est_run=w->est_run;
  484. }else{
  485. ac_mode = 3;
  486. est_run = 64;
  487. }
  488. }
  489. }
  490. x8_select_ac_table(w,ac_mode);
  491. /*scantable_selector[12]={0,2,0,1,1,1,0,2,2,0,1,2};<-
  492. -> 10'01' 00'10' 10'00' 01'01' 01'00' 10'00 =>0x928548 */
  493. scantable = w->scantable[ (0x928548>>(2*w->orient))&3 ].permutated;
  494. pos=0;
  495. do {
  496. n++;
  497. if( n >= est_run ){
  498. ac_mode=3;
  499. x8_select_ac_table(w,3);
  500. }
  501. x8_get_ac_rlf(w,ac_mode,&run,&level,&final);
  502. pos+=run+1;
  503. if(pos>63){
  504. //this also handles vlc error in x8_get_ac_rlf
  505. return -1;
  506. }
  507. level= (level+1) * w->dquant;
  508. level+= w->qsum;
  509. sign = - get_bits1(&s->gb);
  510. level = (level ^ sign) - sign;
  511. if(use_quant_matrix){
  512. level = (level*quant_table[pos])>>8;
  513. }
  514. s->block[0][ scantable[pos] ]=level;
  515. }while(!final);
  516. s->block_last_index[0]=pos;
  517. }else{//DC only
  518. s->block_last_index[0]=0;
  519. if(w->flat_dc && ((unsigned)(dc_level+1)) < 3){//[-1;1]
  520. int32_t divide_quant= !chroma ? w->divide_quant_dc_luma:
  521. w->divide_quant_dc_chroma;
  522. int32_t dc_quant = !chroma ? w->quant:
  523. w->quant_dc_chroma;
  524. //original intent dc_level+=predicted_dc/quant; but it got lost somewhere in the rounding
  525. dc_level+= (w->predicted_dc*divide_quant + (1<<12) )>>13;
  526. dsp_x8_put_solidcolor( av_clip_uint8((dc_level*dc_quant+4)>>3),
  527. s->dest[chroma], s->current_picture.f.linesize[!!chroma]);
  528. goto block_placed;
  529. }
  530. zeros_only = (dc_level == 0);
  531. }
  532. if(!chroma){
  533. s->block[0][0] = dc_level*w->quant;
  534. }else{
  535. s->block[0][0] = dc_level*w->quant_dc_chroma;
  536. }
  537. //there is !zero_only check in the original, but dc_level check is enough
  538. if( (unsigned int)(dc_level+1) >= 3 && (w->edges&3) != 3 ){
  539. int direction;
  540. /*ac_comp_direction[orient] = { 0, 3, 3, 1, 1, 0, 0, 0, 2, 2, 2, 1 };<-
  541. -> 01'10' 10'10' 00'00' 00'01' 01'11' 11'00 =>0x6A017C */
  542. direction= (0x6A017C>>(w->orient*2))&3;
  543. if (direction != 3){
  544. x8_ac_compensation(w, direction, s->block[0][0]);//modify block_last[]
  545. }
  546. }
  547. if(w->flat_dc){
  548. dsp_x8_put_solidcolor(w->predicted_dc, s->dest[chroma], s->current_picture.f.linesize[!!chroma]);
  549. }else{
  550. w->dsp.spatial_compensation[w->orient]( s->edge_emu_buffer,
  551. s->dest[chroma],
  552. s->current_picture.f.linesize[!!chroma] );
  553. }
  554. if(!zeros_only)
  555. s->dsp.idct_add ( s->dest[chroma],
  556. s->current_picture.f.linesize[!!chroma],
  557. s->block[0] );
  558. block_placed:
  559. if(!chroma){
  560. x8_update_predictions(w,w->orient,n);
  561. }
  562. if(s->loop_filter){
  563. uint8_t* ptr = s->dest[chroma];
  564. int linesize = s->current_picture.f.linesize[!!chroma];
  565. if(!( (w->edges&2) || ( zeros_only && (w->orient|4)==4 ) )){
  566. w->dsp.h_loop_filter(ptr, linesize, w->quant);
  567. }
  568. if(!( (w->edges&1) || ( zeros_only && (w->orient|8)==8 ) )){
  569. w->dsp.v_loop_filter(ptr, linesize, w->quant);
  570. }
  571. }
  572. return 0;
  573. }
  574. static void x8_init_block_index(MpegEncContext *s){ //FIXME maybe merge with ff_*
  575. //not s->linesize as this would be wrong for field pics
  576. //not that IntraX8 has interlacing support ;)
  577. const int linesize = s->current_picture.f.linesize[0];
  578. const int uvlinesize = s->current_picture.f.linesize[1];
  579. s->dest[0] = s->current_picture.f.data[0];
  580. s->dest[1] = s->current_picture.f.data[1];
  581. s->dest[2] = s->current_picture.f.data[2];
  582. s->dest[0] += s->mb_y * linesize << 3;
  583. s->dest[1] += ( s->mb_y&(~1) ) * uvlinesize << 2;//chroma blocks are on add rows
  584. s->dest[2] += ( s->mb_y&(~1) ) * uvlinesize << 2;
  585. }
  586. /**
  587. * Initialize IntraX8 frame decoder.
  588. * Requires valid MpegEncContext with valid s->mb_width before calling.
  589. * @param w pointer to IntraX8Context
  590. * @param s pointer to MpegEncContext of the parent codec
  591. */
  592. av_cold void ff_intrax8_common_init(IntraX8Context * w, MpegEncContext * const s){
  593. w->s=s;
  594. x8_vlc_init();
  595. av_assert0(s->mb_width>0);
  596. w->prediction_table=av_mallocz(s->mb_width*2*2);//two rows, 2 blocks per cannon mb
  597. ff_init_scantable(s->dsp.idct_permutation, &w->scantable[0], ff_wmv1_scantable[0]);
  598. ff_init_scantable(s->dsp.idct_permutation, &w->scantable[1], ff_wmv1_scantable[2]);
  599. ff_init_scantable(s->dsp.idct_permutation, &w->scantable[2], ff_wmv1_scantable[3]);
  600. ff_intrax8dsp_init(&w->dsp);
  601. }
  602. /**
  603. * Destroy IntraX8 frame structure.
  604. * @param w pointer to IntraX8Context
  605. */
  606. av_cold void ff_intrax8_common_end(IntraX8Context * w)
  607. {
  608. av_freep(&w->prediction_table);
  609. }
  610. /**
  611. * Decode single IntraX8 frame.
  612. * The parent codec must fill s->loopfilter and s->gb (bitstream).
  613. * The parent codec must call MPV_frame_start(), ff_er_frame_start() before calling this function.
  614. * The parent codec must call ff_er_frame_end(), MPV_frame_end() after calling this function.
  615. * This function does not use MPV_decode_mb().
  616. * lowres decoding is theoretically impossible.
  617. * @param w pointer to IntraX8Context
  618. * @param dquant doubled quantizer, it would be odd in case of VC-1 halfpq==1.
  619. * @param quant_offset offset away from zero
  620. */
  621. //FIXME extern uint8_t ff_wmv3_dc_scale_table[32];
  622. int ff_intrax8_decode_picture(IntraX8Context * const w, int dquant, int quant_offset){
  623. MpegEncContext * const s= w->s;
  624. int mb_xy;
  625. w->use_quant_matrix = get_bits1(&s->gb);
  626. w->dquant = dquant;
  627. w->quant = dquant >> 1;
  628. w->qsum = quant_offset;
  629. w->divide_quant_dc_luma = ((1<<16) + (w->quant>>1)) / w->quant;
  630. if(w->quant < 5){
  631. w->quant_dc_chroma = w->quant;
  632. w->divide_quant_dc_chroma = w->divide_quant_dc_luma;
  633. }else{
  634. w->quant_dc_chroma = w->quant+((w->quant+3)>>3);
  635. w->divide_quant_dc_chroma = ((1<<16) + (w->quant_dc_chroma>>1)) / w->quant_dc_chroma;
  636. }
  637. x8_reset_vlc_tables(w);
  638. s->resync_mb_x=0;
  639. s->resync_mb_y=0;
  640. for(s->mb_y=0; s->mb_y < s->mb_height*2; s->mb_y++){
  641. x8_init_block_index(s);
  642. mb_xy=(s->mb_y>>1)*s->mb_stride;
  643. for(s->mb_x=0; s->mb_x < s->mb_width*2; s->mb_x++){
  644. x8_get_prediction(w);
  645. if(x8_setup_spatial_predictor(w,0)) goto error;
  646. if(x8_decode_intra_mb(w,0)) goto error;
  647. if( s->mb_x & s->mb_y & 1 ){
  648. x8_get_prediction_chroma(w);
  649. /*when setting up chroma, no vlc is read,
  650. so no error condition can be reached*/
  651. x8_setup_spatial_predictor(w,1);
  652. if(x8_decode_intra_mb(w,1)) goto error;
  653. x8_setup_spatial_predictor(w,2);
  654. if(x8_decode_intra_mb(w,2)) goto error;
  655. s->dest[1]+= 8;
  656. s->dest[2]+= 8;
  657. /*emulate MB info in the relevant tables*/
  658. s->mbskip_table [mb_xy]=0;
  659. s->mbintra_table[mb_xy]=1;
  660. s->current_picture.f.qscale_table[mb_xy] = w->quant;
  661. mb_xy++;
  662. }
  663. s->dest[0]+= 8;
  664. }
  665. if(s->mb_y&1){
  666. ff_draw_horiz_band(s, (s->mb_y-1)*8, 16);
  667. }
  668. }
  669. error:
  670. ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y,
  671. (s->mb_x>>1)-1, (s->mb_y>>1)-1,
  672. ER_MB_END );
  673. return 0;
  674. }