falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
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.
2126 Commits
32 Branches
296MB
Tree: e67e14d54c
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e67e14d54c'
${ noResults }
FFmpeg/libavcodec/svq3.c

716 lines
21KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2003 The FFmpeg Project.

  3.  *

  4.  * This library 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 of the License, or (at your option) any later version.

  8.  *

  9.  * This library 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 this library; if not, write to the Free Software

  16.  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

  17.  *

  18.  *

  19.  * How to use this decoder:

  20.  * SVQ3 data is transported within Apple Quicktime files. Quicktime files

  21.  * have stsd atoms to describe media trak properties. A stsd atom for a

  22.  * video trak contains 1 or more ImageDescription atoms. These atoms begin

  23.  * with the 4-byte length of the atom followed by the codec fourcc. Some

  24.  * decoders need information in this atom to operate correctly. Such

  25.  * is the case with SVQ3. In order to get the best use out of this decoder,

  26.  * the calling app must make the SVQ3 ImageDescription atom available

  27.  * via the AVCodecContext's extradata[_size] field:

  28.  *

  29.  * AVCodecContext.extradata = pointer to ImageDescription, first characters 

  30.  * are expected to be 'S', 'V', 'Q', and '3', NOT the 4-byte atom length

  31.  * AVCodecContext.extradata_size = size of ImageDescription atom memory 

  32.  * buffer (which will be the same as the ImageDescription atom size field 

  33.  * from the QT file, minus 4 bytes since the length is missing)

  34.  *

  35.  * You will know you have these parameters passed correctly when the decoder

  36.  * correctly decodes this file:

  37.  *  ftp://ftp.mplayerhq.hu/MPlayer/samples/V-codecs/SVQ3/Vertical400kbit.sorenson3.mov

  38.  *

  39.  */

  40.  

  41. /**

  42.  * @file svq3.c

  43.  * svq3 decoder.

  44.  */

  45. 

  46. #define FULLPEL_MODE  1 

  47. #define HALFPEL_MODE  2 

  48. #define THIRDPEL_MODE 3

  49.  

  50. /* dual scan (from some older h264 draft)

  51.  o-->o-->o   o

  52.          |  /|

  53.  o   o   o / o

  54.  | / |   |/  |

  55.  o   o   o   o

  56.    / 

  57.  o-->o-->o-->o

  58. */

  59. static const uint8_t svq3_scan[16]={

  60.  0+0*4, 1+0*4, 2+0*4, 2+1*4,

  61.  2+2*4, 3+0*4, 3+1*4, 3+2*4,

  62.  0+1*4, 0+2*4, 1+1*4, 1+2*4,

  63.  0+3*4, 1+3*4, 2+3*4, 3+3*4,

  64. };

  65. 

  66. static const uint8_t svq3_pred_0[25][2] = {

  67.   { 0, 0 },

  68.   { 1, 0 }, { 0, 1 },

  69.   { 0, 2 }, { 1, 1 }, { 2, 0 },

  70.   { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },

  71.   { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

  72.   { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },

  73.   { 2, 4 }, { 3, 3 }, { 4, 2 },

  74.   { 4, 3 }, { 3, 4 },

  75.   { 4, 4 }

  76. };

  77. 

  78. static const int8_t svq3_pred_1[6][6][5] = {

  79.   { { 2,-1,-1,-1,-1 }, { 2, 1,-1,-1,-1 }, { 1, 2,-1,-1,-1 },

  80.     { 2, 1,-1,-1,-1 }, { 1, 2,-1,-1,-1 }, { 1, 2,-1,-1,-1 } },

  81.   { { 0, 2,-1,-1,-1 }, { 0, 2, 1, 4, 3 }, { 0, 1, 2, 4, 3 },

  82.     { 0, 2, 1, 4, 3 }, { 2, 0, 1, 3, 4 }, { 0, 4, 2, 1, 3 } },

  83.   { { 2, 0,-1,-1,-1 }, { 2, 1, 0, 4, 3 }, { 1, 2, 4, 0, 3 },

  84.     { 2, 1, 0, 4, 3 }, { 2, 1, 4, 3, 0 }, { 1, 2, 4, 0, 3 } },

  85.   { { 2, 0,-1,-1,-1 }, { 2, 0, 1, 4, 3 }, { 1, 2, 0, 4, 3 },

  86.     { 2, 1, 0, 4, 3 }, { 2, 1, 3, 4, 0 }, { 2, 4, 1, 0, 3 } },

  87.   { { 0, 2,-1,-1,-1 }, { 0, 2, 1, 3, 4 }, { 1, 2, 3, 0, 4 },

  88.     { 2, 0, 1, 3, 4 }, { 2, 1, 3, 0, 4 }, { 2, 0, 4, 3, 1 } },

  89.   { { 0, 2,-1,-1,-1 }, { 0, 2, 4, 1, 3 }, { 1, 4, 2, 0, 3 },

  90.     { 4, 2, 0, 1, 3 }, { 2, 0, 1, 4, 3 }, { 4, 2, 1, 0, 3 } },

  91. };

  92. 

  93. static const struct { uint8_t run; uint8_t level; } svq3_dct_tables[2][16] = {

  94.   { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  95.     { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  96.   { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  97.     { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  98. };

  99. 

  100. static const uint32_t svq3_dequant_coeff[32] = {

  101.    3881,  4351,  4890,  5481,  6154,  6914,  7761,  8718,

  102.    9781, 10987, 12339, 13828, 15523, 17435, 19561, 21873,

  103.   24552, 27656, 30847, 34870, 38807, 43747, 49103, 54683,

  104.   61694, 68745, 77615, 89113,100253,109366,126635,141533

  105. };

  106. 

  107. 

  108. static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp){

  109.     const int qmul= svq3_dequant_coeff[qp];

  110. #define stride 16

  111.     int i;

  112.     int temp[16];

  113.     static const int x_offset[4]={0, 1*stride, 4* stride,  5*stride};

  114.     static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};

  115. 

  116.     for(i=0; i<4; i++){

  117.         const int offset= y_offset[i];

  118.         const int z0= 13*(block[offset+stride*0] +    block[offset+stride*4]);

  119.         const int z1= 13*(block[offset+stride*0] -    block[offset+stride*4]);

  120.         const int z2=  7* block[offset+stride*1] - 17*block[offset+stride*5];

  121.         const int z3= 17* block[offset+stride*1] +  7*block[offset+stride*5];

  122. 

  123.         temp[4*i+0]= z0+z3;

  124.         temp[4*i+1]= z1+z2;

  125.         temp[4*i+2]= z1-z2;

  126.         temp[4*i+3]= z0-z3;

  127.     }

  128. 

  129.     for(i=0; i<4; i++){

  130.         const int offset= x_offset[i];

  131.         const int z0= 13*(temp[4*0+i] +    temp[4*2+i]);

  132.         const int z1= 13*(temp[4*0+i] -    temp[4*2+i]);

  133.         const int z2=  7* temp[4*1+i] - 17*temp[4*3+i];

  134.         const int z3= 17* temp[4*1+i] +  7*temp[4*3+i];

  135. 

  136.         block[stride*0 +offset]= ((z0 + z3)*qmul + 0x80000)>>20;

  137.         block[stride*2 +offset]= ((z1 + z2)*qmul + 0x80000)>>20;

  138.         block[stride*8 +offset]= ((z1 - z2)*qmul + 0x80000)>>20;

  139.         block[stride*10+offset]= ((z0 - z3)*qmul + 0x80000)>>20;

  140.     }

  141. }

  142. #undef stride

  143. 

  144. static void svq3_add_idct_c (uint8_t *dst, DCTELEM *block, int stride, int qp, int dc){

  145.     const int qmul= svq3_dequant_coeff[qp];

  146.     int i;

  147.     uint8_t *cm = cropTbl + MAX_NEG_CROP;

  148. 

  149.     if (dc) {

  150.         dc = 13*13*((dc == 1) ? 1538*block[0] : ((qmul*(block[0] >> 3)) / 2));

  151.         block[0] = 0;

  152.     }

  153. 

  154.     for (i=0; i < 4; i++) {

  155.         const int z0= 13*(block[0 + 4*i] +    block[2 + 4*i]);

  156.         const int z1= 13*(block[0 + 4*i] -    block[2 + 4*i]);

  157.         const int z2=  7* block[1 + 4*i] - 17*block[3 + 4*i];

  158.         const int z3= 17* block[1 + 4*i] +  7*block[3 + 4*i];

  159. 

  160.         block[0 + 4*i]= z0 + z3;

  161.         block[1 + 4*i]= z1 + z2;

  162.         block[2 + 4*i]= z1 - z2;

  163.         block[3 + 4*i]= z0 - z3;

  164.     }

  165. 

  166.     for (i=0; i < 4; i++) {

  167.         const int z0= 13*(block[i + 4*0] +    block[i + 4*2]);

  168.         const int z1= 13*(block[i + 4*0] -    block[i + 4*2]);

  169.         const int z2=  7* block[i + 4*1] - 17*block[i + 4*3];

  170.         const int z3= 17* block[i + 4*1] +  7*block[i + 4*3];

  171.         const int rr= (dc + 0x80000);

  172. 

  173.         dst[i + stride*0]= cm[ dst[i + stride*0] + (((z0 + z3)*qmul + rr) >> 20) ];

  174.         dst[i + stride*1]= cm[ dst[i + stride*1] + (((z1 + z2)*qmul + rr) >> 20) ];

  175.         dst[i + stride*2]= cm[ dst[i + stride*2] + (((z1 - z2)*qmul + rr) >> 20) ];

  176.         dst[i + stride*3]= cm[ dst[i + stride*3] + (((z0 - z3)*qmul + rr) >> 20) ];

  177.     }

  178. }

  179. 

  180. static void pred4x4_down_left_svq3_c(uint8_t *src, uint8_t *topright, int stride){

  181.     LOAD_TOP_EDGE    

  182.     LOAD_LEFT_EDGE    

  183.     const __attribute__((unused)) int unu0= t0;

  184.     const __attribute__((unused)) int unu1= l0;

  185. 

  186.     src[0+0*stride]=(l1 + t1)>>1;

  187.     src[1+0*stride]=

  188.     src[0+1*stride]=(l2 + t2)>>1;

  189.     src[2+0*stride]=

  190.     src[1+1*stride]=

  191.     src[0+2*stride]=

  192.     src[3+0*stride]=

  193.     src[2+1*stride]=

  194.     src[1+2*stride]=

  195.     src[0+3*stride]=

  196.     src[3+1*stride]=

  197.     src[2+2*stride]=

  198.     src[1+3*stride]=

  199.     src[3+2*stride]=

  200.     src[2+3*stride]=

  201.     src[3+3*stride]=(l3 + t3)>>1;

  202. };

  203. 

  204. static void pred16x16_plane_svq3_c(uint8_t *src, int stride){

  205.     pred16x16_plane_compat_c(src, stride, 1);

  206. }

  207. 

  208. static inline int svq3_decode_block (GetBitContext *gb, DCTELEM *block,

  209. 				     int index, const int type) {

  210. 

  211.   static const uint8_t *const scan_patterns[4] =

  212.   { luma_dc_zigzag_scan, zigzag_scan, svq3_scan, chroma_dc_scan };

  213. 

  214.   int run, level, sign, vlc, limit;

  215.   const int intra = (3 * type) >> 2;

  216.   const uint8_t *const scan = scan_patterns[type];

  217. 

  218.   for (limit=(16 >> intra); index < 16; index=limit, limit+=8) {

  219.     for (; (vlc = svq3_get_ue_golomb (gb)) != 0; index++) {

  220. 

  221.       if (vlc == INVALID_VLC)

  222. 	return -1;

  223. 

  224.       sign = (vlc & 0x1) - 1;

  225.       vlc  = (vlc + 1) >> 1;

  226. 

  227.       if (type == 3) {

  228. 	if (vlc < 3) {

  229. 	  run   = 0;

  230. 	  level = vlc;

  231. 	} else if (vlc < 4) {

  232. 	  run   = 1;

  233. 	  level = 1;

  234. 	} else {

  235. 	  run   = (vlc & 0x3);

  236. 	  level = ((vlc + 9) >> 2) - run;

  237. 	}

  238.       } else {

  239. 	if (vlc < 16) {

  240. 	  run   = svq3_dct_tables[intra][vlc].run;

  241. 	  level = svq3_dct_tables[intra][vlc].level;

  242. 	} else if (intra) {

  243. 	  run   = (vlc & 0x7);

  244. 	  level = (vlc >> 3) + ((run == 0) ? 8 : ((run < 2) ? 2 : ((run < 5) ? 0 : -1)));

  245. 	} else {

  246. 	  run   = (vlc & 0xF);

  247. 	  level = (vlc >> 4) + ((run == 0) ? 4 : ((run < 3) ? 2 : ((run < 10) ? 1 : 0)));

  248. 	}

  249.       }

  250. 

  251.       if ((index += run) >= limit)

  252. 	return -1;

  253. 

  254.       block[scan[index]] = (level ^ sign) - sign;

  255.     }

  256. 

  257.     if (type != 2) {

  258.       break;

  259.     }

  260.   }

  261. 

  262.   return 0;

  263. }

  264. 

  265. static inline void svq3_mc_dir_part (MpegEncContext *s, int x, int y,

  266. 				     int width, int height, int mx, int my, int dxy, int thirdpel) {

  267.   uint8_t *src, *dest;

  268.   int i, emu = 0;

  269.   int blocksize= 2 - (width>>3); //16->0, 8->1, 4->2

  270. 

  271.   mx += x;

  272.   my += y;

  273.   

  274.   if (mx < 0 || mx >= (s->h_edge_pos - width  - 1) ||

  275.       my < 0 || my >= (s->v_edge_pos - height - 1)) {

  276. 

  277.     if ((s->flags & CODEC_FLAG_EMU_EDGE)) {

  278.       emu = 1;

  279.     }

  280. 

  281.     mx = clip (mx, -16, (s->h_edge_pos - width  + 15));

  282.     my = clip (my, -16, (s->v_edge_pos - height + 15));

  283.   }

  284. 

  285.   /* form component predictions */

  286.   dest = s->current_picture.data[0] + x + y*s->linesize;

  287.   src  = s->last_picture.data[0] + mx + my*s->linesize;

  288. 

  289.   if (emu) {

  290.     ff_emulated_edge_mc (s, src, s->linesize, (width + 1), (height + 1),

  291. 			 mx, my, s->h_edge_pos, s->v_edge_pos);

  292.     src = s->edge_emu_buffer;

  293.   }

  294.   if(thirdpel)

  295.     s->dsp.put_tpel_pixels_tab[dxy](dest, src, s->linesize, width, height);

  296.   else

  297.     s->dsp.put_pixels_tab[blocksize][dxy](dest, src, s->linesize, height);

  298. 

  299.   if (!(s->flags & CODEC_FLAG_GRAY)) {

  300.     mx	   = (mx + (mx < (int) x)) >> 1;

  301.     my	   = (my + (my < (int) y)) >> 1;

  302.     width  = (width  >> 1);

  303.     height = (height >> 1);

  304.     blocksize++;

  305. 

  306.     for (i=1; i < 3; i++) {

  307.       dest = s->current_picture.data[i] + (x >> 1) + (y >> 1)*s->uvlinesize;

  308.       src  = s->last_picture.data[i] + mx + my*s->uvlinesize;

  309. 

  310.       if (emu) {

  311. 	ff_emulated_edge_mc (s, src, s->uvlinesize, (width + 1), (height + 1),

  312. 			     mx, my, (s->h_edge_pos >> 1), (s->v_edge_pos >> 1));

  313. 	src = s->edge_emu_buffer;

  314.       }

  315.       if(thirdpel)

  316.         s->dsp.put_tpel_pixels_tab[dxy](dest, src, s->uvlinesize, width, height);

  317.       else

  318.         s->dsp.put_pixels_tab[blocksize][dxy](dest, src, s->uvlinesize, height);

  319.     }

  320.   }

  321. }

  322. 

  323. static int svq3_decode_mb (H264Context *h, unsigned int mb_type) {

  324.   int cbp, dir, mode, mx, my, dx, dy, x, y, part_width, part_height;

  325.   int i, j, k, l, m;

  326.   uint32_t vlc;

  327.   int8_t *top, *left;

  328.   MpegEncContext *const s = (MpegEncContext *) h;

  329.   const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;

  330.   const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;

  331. 

  332.   h->top_samples_available	= (s->mb_y == 0) ? 0x33FF : 0xFFFF;

  333.   h->left_samples_available	= (s->mb_x == 0) ? 0x5F5F : 0xFFFF;

  334.   h->topright_samples_available	= 0xFFFF;

  335. 

  336.   if (mb_type == 0) {		/* SKIP */

  337.     svq3_mc_dir_part (s, 16*s->mb_x, 16*s->mb_y, 16, 16, 0, 0, 0, 0);

  338. 

  339.     cbp = 0;

  340.     mb_type = MB_TYPE_SKIP;

  341.   } else if (mb_type < 8) {	/* INTER */

  342.     if (h->thirdpel_flag && h->halfpel_flag == !get_bits (&s->gb, 1)) {

  343.       mode = THIRDPEL_MODE;

  344.     } else if (h->halfpel_flag && h->thirdpel_flag == !get_bits (&s->gb, 1)) {

  345.       mode = HALFPEL_MODE;

  346.     } else {

  347.       mode = FULLPEL_MODE;

  348.     }

  349. 

  350.     /* fill caches */

  351.     /* note ref_cache[0] should contain here:

  352.         ????????

  353.         ???11111

  354.         N??11111

  355.         N??11111

  356.         N??11111

  357.         N

  358.     */

  359.     

  360.     if (s->mb_x > 0) {

  361.       for (i=0; i < 4; i++) {

  362. 	*(uint32_t *) h->mv_cache[0][scan8[0] - 1 + i*8] = *(uint32_t *) s->current_picture.motion_val[0][b_xy - 1 + i*h->b_stride];

  363.       }

  364.     } else {

  365.       for (i=0; i < 4; i++) {

  366. 	*(uint32_t *) h->mv_cache[0][scan8[0] - 1 + i*8] = 0;

  367.       }

  368.     }

  369.     if (s->mb_y > 0) {

  370.       memcpy (h->mv_cache[0][scan8[0] - 1*8], s->current_picture.motion_val[0][b_xy - h->b_stride], 4*2*sizeof(int16_t));

  371.       memset (&h->ref_cache[0][scan8[0] - 1*8], 1, 4);

  372. 

  373.       if (s->mb_x < (s->mb_width - 1)) {

  374. 	*(uint32_t *) h->mv_cache[0][scan8[0] + 4 - 1*8] = *(uint32_t *) s->current_picture.motion_val[0][b_xy - h->b_stride + 4];

  375. 	h->ref_cache[0][scan8[0] + 4 - 1*8] = 1;

  376.       }else

  377. 	h->ref_cache[0][scan8[0] + 4 - 1*8] = PART_NOT_AVAILABLE;

  378.       if (s->mb_x > 0) {

  379. 	*(uint32_t *) h->mv_cache[0][scan8[0] - 1 - 1*8] = *(uint32_t *) s->current_picture.motion_val[0][b_xy - h->b_stride - 1];

  380. 	h->ref_cache[0][scan8[0] - 1 - 1*8] = 1;

  381.       }else

  382. 	h->ref_cache[0][scan8[0] - 1 - 1*8] = PART_NOT_AVAILABLE;

  383.     }else

  384.       memset (&h->ref_cache[0][scan8[0] - 1*8 - 1], PART_NOT_AVAILABLE, 8);

  385. 

  386.     /* decode motion vector(s) and form prediction(s) */

  387.     part_width  = ((mb_type & 5) == 5) ? 4 : 8 << (mb_type & 1);

  388.     part_height = 16 >> ((unsigned) mb_type / 3);

  389. 

  390.     for (i=0; i < 16; i+=part_height) {

  391.       for (j=0; j < 16; j+=part_width) {

  392.         int dxy;

  393. 	x = 16*s->mb_x + j;

  394. 	y = 16*s->mb_y + i;

  395. 	k = ((j>>2)&1) + ((i>>1)&2) + ((j>>1)&4) + (i&8);

  396. 

  397. 	pred_motion (h, k, (part_width >> 2), 0, 1, &mx, &my);

  398. 

  399. 	/* clip motion vector prediction to frame border */

  400. 	mx = clip (mx, -6*x, 6*(s->h_edge_pos - part_width  - x));

  401. 	my = clip (my, -6*y, 6*(s->v_edge_pos - part_height - y));

  402. 

  403. 	/* get motion vector differential */

  404. 	dy = svq3_get_se_golomb (&s->gb);

  405. 	dx = svq3_get_se_golomb (&s->gb);

  406. 

  407. 	if (dx == INVALID_VLC || dy == INVALID_VLC) {

  408. 	  return -1;

  409. 	}

  410. 	/* compute motion vector */

  411. 	if (mode == THIRDPEL_MODE) {

  412.           int fx, fy;

  413.           mx = ((mx + 1)>>1) + dx;

  414.           my = ((my + 1)>>1) + dy;

  415.           fx= ((unsigned)(mx + 0x3000))/3 - 0x1000;

  416.           fy= ((unsigned)(my + 0x3000))/3 - 0x1000;

  417.           dxy= (mx - 3*fx) + 4*(my - 3*fy);

  418. 

  419.           svq3_mc_dir_part (s, x, y, part_width, part_height, fx, fy, dxy, 1);

  420.           mx += mx;

  421.           my += my;

  422. 	} else if (mode == HALFPEL_MODE) {

  423. 	  mx = ((unsigned)(mx + 1 + 0x3000))/3 + dx - 0x1000;

  424. 	  my = ((unsigned)(my + 1 + 0x3000))/3 + dy - 0x1000;

  425.           dxy= (mx&1) + 2*(my&1);

  426. 

  427.           svq3_mc_dir_part (s, x, y, part_width, part_height, mx>>1, my>>1, dxy, 0);

  428.           mx *= 3;

  429.           my *= 3;

  430. 	} else {

  431.           assert(mode == FULLPEL_MODE);

  432. 	  mx = ((unsigned)(mx + 3 + 0x6000))/6 + dx - 0x1000;

  433. 	  my = ((unsigned)(my + 3 + 0x6000))/6 + dy - 0x1000;

  434. 

  435. 	  svq3_mc_dir_part (s, x, y, part_width, part_height, mx, my, 0, 0);

  436.           mx *= 6;

  437.           my *= 6;

  438. 	}

  439. 

  440.         /* update mv_cache */

  441.         fill_rectangle(h->mv_cache[0][scan8[k]], part_width>>2, part_height>>2, 8, pack16to32(mx,my), 4);

  442.       }

  443.     }

  444. 

  445.     for (i=0; i < 4; i++) {

  446.       memcpy (s->current_picture.motion_val[0][b_xy + i*h->b_stride], h->mv_cache[0][scan8[0] + 8*i], 4*2*sizeof(int16_t));

  447.     }

  448. 

  449.     if ((vlc = svq3_get_ue_golomb (&s->gb)) >= 48)

  450.       return -1;

  451. 

  452.     cbp = golomb_to_inter_cbp[vlc];

  453.     mb_type = MB_TYPE_16x16;

  454.   } else if (mb_type == 8) {	/* INTRA4x4 */

  455.     memset (h->intra4x4_pred_mode_cache, -1, 8*5*sizeof(int8_t));

  456. 

  457.     if (s->mb_x > 0) {

  458.       for (i=0; i < 4; i++) {

  459. 	h->intra4x4_pred_mode_cache[scan8[0] - 1 + i*8] = h->intra4x4_pred_mode[mb_xy - 1][i];

  460.       }

  461.     }

  462.     if (s->mb_y > 0) {

  463.       h->intra4x4_pred_mode_cache[4+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][4];

  464.       h->intra4x4_pred_mode_cache[5+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][5];

  465.       h->intra4x4_pred_mode_cache[6+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][6];

  466.       h->intra4x4_pred_mode_cache[7+8*0] = h->intra4x4_pred_mode[mb_xy - s->mb_stride][3];

  467.     }

  468. 

  469.     /* decode prediction codes for luma blocks */

  470.     for (i=0; i < 16; i+=2) {

  471.       vlc = svq3_get_ue_golomb (&s->gb);

  472. 

  473.       if (vlc >= 25)

  474. 	return -1;

  475. 

  476.       left    = &h->intra4x4_pred_mode_cache[scan8[i] - 1];

  477.       top     = &h->intra4x4_pred_mode_cache[scan8[i] - 8];

  478. 

  479.       left[1] = svq3_pred_1[top[0] + 1][left[0] + 1][svq3_pred_0[vlc][0]];

  480.       left[2] = svq3_pred_1[top[1] + 1][left[1] + 1][svq3_pred_0[vlc][1]];

  481. 

  482.       if (left[1] == -1 || left[2] == -1)

  483. 	return -1;

  484.     }

  485. 

  486.     write_back_intra_pred_mode (h);

  487.     check_intra4x4_pred_mode (h);

  488. 

  489.     if ((vlc = svq3_get_ue_golomb (&s->gb)) >= 48)

  490.       return -1;

  491. 

  492.     cbp = golomb_to_intra4x4_cbp[vlc];

  493.     mb_type = MB_TYPE_INTRA4x4;

  494.   } else {			/* INTRA16x16 */

  495.     dir = i_mb_type_info[mb_type - 8].pred_mode;

  496.     dir = (dir >> 1) ^ 3*(dir & 1) ^ 1;

  497. 

  498.     if ((h->intra16x16_pred_mode = check_intra_pred_mode (h, dir)) == -1)

  499.       return -1;

  500. 

  501.     cbp = i_mb_type_info[mb_type - 8].cbp;

  502.     mb_type = MB_TYPE_INTRA16x16;

  503.   }

  504. 

  505.   if (!IS_INTER(mb_type) && s->pict_type != I_TYPE) {

  506.     for (i=0; i < 4; i++) {

  507.       memset (s->current_picture.motion_val[0][b_xy + i*h->b_stride], 0, 4*2*sizeof(int16_t));

  508.     }

  509.   }

  510.   if (!IS_INTRA4x4(mb_type)) {

  511.     memset (h->intra4x4_pred_mode[mb_xy], DC_PRED, 8);

  512.   }

  513.   if (!IS_SKIP(mb_type)) {

  514.     memset (h->non_zero_count_cache + 8, 0, 4*9*sizeof(uint8_t));

  515.     s->dsp.clear_blocks(h->mb);

  516.   }

  517. 

  518.   if (IS_INTRA16x16(mb_type) || (s->pict_type != I_TYPE && s->adaptive_quant && cbp)) {

  519.     s->qscale += svq3_get_se_golomb (&s->gb);

  520. 

  521.     if (s->qscale > 31)

  522.       return -1;

  523.   }

  524.   if (IS_INTRA16x16(mb_type)) {

  525.     if (svq3_decode_block (&s->gb, h->mb, 0, 0))

  526.       return -1;

  527.   }

  528. 

  529.   if (!IS_SKIP(mb_type) && cbp) {

  530.     l = IS_INTRA16x16(mb_type) ? 1 : 0;

  531.     m = ((s->qscale < 24 && IS_INTRA4x4(mb_type)) ? 2 : 1);

  532. 

  533.     for (i=0; i < 4; i++) {

  534.       if ((cbp & (1 << i))) {

  535. 	for (j=0; j < 4; j++) {

  536. 	  k = l ? ((j&1) + 2*(i&1) + 2*(j&2) + 4*(i&2)) : (4*i + j);

  537. 	  h->non_zero_count_cache[ scan8[k] ] = 1;

  538. 

  539. 	  if (svq3_decode_block (&s->gb, &h->mb[16*k], l, m))

  540. 	    return -1;

  541. 	}

  542.       }

  543.     }

  544. 

  545.     if ((cbp & 0x30)) {

  546.       for (i=0; i < 2; ++i) {

  547. 	if (svq3_decode_block (&s->gb, &h->mb[16*(16 + 4*i)], 0, 3))

  548. 	  return -1;

  549.       }

  550. 

  551.       if ((cbp & 0x20)) {

  552. 	for (i=0; i < 8; i++) {

  553. 	  h->non_zero_count_cache[ scan8[16+i] ] = 1;

  554. 

  555. 	  if (svq3_decode_block (&s->gb, &h->mb[16*(16 + i)], 1, 1))

  556. 	    return -1;

  557. 	}

  558.       }

  559.     }

  560.   }

  561. 

  562.   s->current_picture.mb_type[mb_xy] = mb_type;

  563. 

  564.   if (IS_INTRA(mb_type)) {

  565.     h->chroma_pred_mode = check_intra_pred_mode (h, DC_PRED8x8);

  566.   }

  567. 

  568.   return 0;

  569. }

  570. 

  571. static int svq3_decode_frame (AVCodecContext *avctx,

  572. 			      void *data, int *data_size,

  573. 			      uint8_t *buf, int buf_size) {

  574.   MpegEncContext *const s = avctx->priv_data;

  575.   H264Context *const h = avctx->priv_data;

  576.   int i;

  577. 

  578.   s->flags = avctx->flags;

  579.   

  580.   if (!s->context_initialized) {

  581.     s->width = avctx->width;

  582.     s->height = avctx->height;

  583.     h->pred4x4[DIAG_DOWN_LEFT_PRED] = pred4x4_down_left_svq3_c;

  584.     h->pred16x16[PLANE_PRED8x8] = pred16x16_plane_svq3_c;

  585.     h->halfpel_flag = 1;

  586.     h->thirdpel_flag = 1;

  587.     h->chroma_qp = 4;

  588. 

  589.     if (MPV_common_init (s) < 0)

  590.       return -1;

  591. 

  592.     h->b_stride = 4*s->mb_width;

  593. 

  594.     alloc_tables (h);

  595.   }

  596.   

  597.   s->low_delay= 1;  

  598.   

  599.   if (avctx->extradata && avctx->extradata_size >= 0x63

  600.       && !memcmp (avctx->extradata, "SVQ3", 4)) {

  601. 

  602.     uint8_t *stsd = (uint8_t *) avctx->extradata + 0x62;

  603. 

  604.     if ((*stsd >> 5) != 7 || avctx->extradata_size >= 0x66) {

  605. 

  606.       if ((*stsd >> 5) == 7) {

  607. 	stsd += 3;	/* skip width, height (12 bits each) */

  608.       }

  609. 

  610.       h->halfpel_flag = (*stsd >> 4) & 1;

  611.       h->thirdpel_flag = (*stsd >> 3) & 1;

  612.     }

  613.   }

  614. 

  615.   if ((buf[0] & 0x9F) != 1) {

  616.     /* TODO: what? */

  617.     fprintf (stderr, "unsupported header (%02X)\n", buf[0]);

  618.     return -1;

  619.   } else {

  620.     int length = (buf[0] >> 5) & 3;

  621.     int offset = 0;

  622. 

  623.     for (i=0; i < length; i++) {

  624.       offset = (offset << 8) | buf[i + 1];

  625.     }

  626. 

  627.     if (buf_size < (offset + length + 1) || length == 0)

  628.       return -1;

  629. 

  630.     memcpy (&buf[2], &buf[offset + 2], (length - 1));

  631.   }

  632. 

  633.   init_get_bits (&s->gb, &buf[2], 8*(buf_size - 2));

  634. 

  635.   if ((i = svq3_get_ue_golomb (&s->gb)) == INVALID_VLC || i >= 3)

  636.     return -1;

  637. 

  638.   s->pict_type = golomb_to_pict_type[i];

  639. 

  640.   /* unknown fields */

  641.   get_bits (&s->gb, 1);

  642.   get_bits (&s->gb, 8);

  643. 

  644.   s->qscale = get_bits (&s->gb, 5);

  645.   s->adaptive_quant = get_bits (&s->gb, 1);

  646. 

  647.   /* unknown fields */

  648.   get_bits (&s->gb, 1);

  649.   get_bits (&s->gb, 1);

  650.   get_bits (&s->gb, 2);

  651. 

  652.   while (get_bits (&s->gb, 1)) {

  653.     get_bits (&s->gb, 8);

  654.   }

  655.   

  656.   if(avctx->debug&FF_DEBUG_PICT_INFO){

  657.       printf("%c hpel:%d, tpel:%d aqp:%d qp:%d\n", 

  658.       av_get_pict_type_char(s->pict_type), h->halfpel_flag, h->thirdpel_flag,

  659.       s->adaptive_quant, s->qscale

  660.       );

  661.   }

  662. 

  663.   /* B-frames are not supported */

  664.   if (s->pict_type == B_TYPE/* && avctx->hurry_up*/)

  665.     return buf_size;

  666. 

  667.   frame_start (h);

  668. 

  669.   for(i=0; i<4; i++){

  670.     int j;

  671.     for(j=-1; j<4; j++)

  672.       h->ref_cache[0][scan8[0] + 8*i + j]= 1;

  673.     h->ref_cache[0][scan8[0] + 8*i + j]= PART_NOT_AVAILABLE;

  674.   }

  675.   

  676.   for (s->mb_y=0; s->mb_y < s->mb_height; s->mb_y++) {

  677.     for (s->mb_x=0; s->mb_x < s->mb_width; s->mb_x++) {

  678.       int mb_type = svq3_get_ue_golomb (&s->gb);

  679. 

  680.       if (s->pict_type == I_TYPE) {

  681. 	mb_type += 8;

  682.       }

  683.       if (mb_type > 32 || svq3_decode_mb (h, mb_type)) {

  684. 	fprintf (stderr, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);

  685. 	return -1;

  686.       }

  687. 

  688.       if (mb_type != 0) {

  689. 	hl_decode_mb (h);

  690.       }

  691.     }

  692. 

  693.     ff_draw_horiz_band(s, 16*s->mb_y, 16);

  694.   }

  695. 

  696.   *(AVFrame *) data = *(AVFrame *) &s->current_picture;

  697.   *data_size = sizeof(AVFrame);

  698. 

  699.   MPV_frame_end(s);

  700.   

  701.   return buf_size;

  702. }

  703. 

  704. 

  705. AVCodec svq3_decoder = {

  706.     "svq3",

  707.     CODEC_TYPE_VIDEO,

  708.     CODEC_ID_SVQ3,

  709.     sizeof(H264Context),

  710.     decode_init,

  711.     NULL,

  712.     decode_end,

  713.     svq3_decode_frame,

  714.     CODEC_CAP_DRAW_HORIZ_BAND | CODEC_CAP_DR1,

  715. };