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.
2093 Commits
32 Branches
296MB
Tree: 25fa62e13a
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '25fa62e13a'
${ noResults }
FFmpeg/libavcodec/svq3.c

777 lines
23KB
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. static const uint8_t svq3_scan[16]={

  47.  0+0*4, 1+0*4, 2+0*4, 2+1*4,

  48.  2+2*4, 3+0*4, 3+1*4, 3+2*4,

  49.  0+1*4, 0+2*4, 1+1*4, 1+2*4,

  50.  0+3*4, 1+3*4, 2+3*4, 3+3*4,

  51. };

  52. 

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

  54.   { 0, 0 },

  55.   { 1, 0 }, { 0, 1 },

  56.   { 0, 2 }, { 1, 1 }, { 2, 0 },

  57.   { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },

  58.   { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

  59.   { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },

  60.   { 2, 4 }, { 3, 3 }, { 4, 2 },

  61.   { 4, 3 }, { 3, 4 },

  62.   { 4, 4 }

  63. };

  64. 

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

  66.   { { 2,-1,-1,-1,-1 }, { 2, 1,-1,-1,-1 }, { 1, 2,-1,-1,-1 },

  67.     { 2, 1,-1,-1,-1 }, { 1, 2,-1,-1,-1 }, { 1, 2,-1,-1,-1 } },

  68.   { { 0, 2,-1,-1,-1 }, { 0, 2, 1, 4, 3 }, { 0, 1, 2, 4, 3 },

  69.     { 0, 2, 1, 4, 3 }, { 2, 0, 1, 3, 4 }, { 0, 4, 2, 1, 3 } },

  70.   { { 2, 0,-1,-1,-1 }, { 2, 1, 0, 4, 3 }, { 1, 2, 4, 0, 3 },

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

  72.   { { 2, 0,-1,-1,-1 }, { 2, 0, 1, 4, 3 }, { 1, 2, 0, 4, 3 },

  73.     { 2, 1, 0, 4, 3 }, { 2, 1, 3, 4, 0 }, { 2, 4, 1, 0, 3 } },

  74.   { { 0, 2,-1,-1,-1 }, { 0, 2, 1, 3, 4 }, { 1, 2, 3, 0, 4 },

  75.     { 2, 0, 1, 3, 4 }, { 2, 1, 3, 0, 4 }, { 2, 0, 4, 3, 1 } },

  76.   { { 0, 2,-1,-1,-1 }, { 0, 2, 4, 1, 3 }, { 1, 4, 2, 0, 3 },

  77.     { 4, 2, 0, 1, 3 }, { 2, 0, 1, 4, 3 }, { 4, 2, 1, 0, 3 } },

  78. };

  79. 

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

  81.   { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  82.     { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  83.   { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  84.     { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  85. };

  86. 

  87. static const uint32_t svq3_dequant_coeff[32] = {

  88.    3881,  4351,  4890,  5481,  6154,  6914,  7761,  8718,

  89.    9781, 10987, 12339, 13828, 15523, 17435, 19561, 21873,

  90.   24552, 27656, 30847, 34870, 38807, 43747, 49103, 54683,

  91.   61694, 68745, 77615, 89113,100253,109366,126635,141533

  92. };

  93. 

  94. 

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

  96.     const int qmul= svq3_dequant_coeff[qp];

  97. #define stride 16

  98.     int i;

  99.     int temp[16];

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

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

  102. 

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

  104.         const int offset= y_offset[i];

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

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

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

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

  109. 

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

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

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

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

  114.     }

  115. 

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

  117.         const int offset= x_offset[i];

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

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

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

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

  122. 

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

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

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

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

  127.     }

  128. }

  129. #undef stride

  130. 

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

  132.     const int qmul= svq3_dequant_coeff[qp];

  133.     int i;

  134.     uint8_t *cm = cropTbl + MAX_NEG_CROP;

  135. 

  136.     if (dc) {

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

  138.         block[0] = 0;

  139.     }

  140. 

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

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

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

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

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

  146. 

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

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

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

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

  151.     }

  152. 

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

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

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

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

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

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

  159. 

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

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

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

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

  164.     }

  165. }

  166. 

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

  168.     LOAD_TOP_EDGE    

  169.     LOAD_LEFT_EDGE    

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

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

  172. 

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

  174.     src[1+0*stride]=

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

  176.     src[2+0*stride]=

  177.     src[1+1*stride]=

  178.     src[0+2*stride]=

  179.     src[3+0*stride]=

  180.     src[2+1*stride]=

  181.     src[1+2*stride]=

  182.     src[0+3*stride]=

  183.     src[3+1*stride]=

  184.     src[2+2*stride]=

  185.     src[1+3*stride]=

  186.     src[3+2*stride]=

  187.     src[2+3*stride]=

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

  189. };

  190. 

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

  192.     pred16x16_plane_compat_c(src, stride, 1);

  193. }

  194. 

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

  196. 				     int index, const int type) {

  197. 

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

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

  200. 

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

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

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

  204. 

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

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

  207. 

  208.       if (vlc == INVALID_VLC)

  209. 	return -1;

  210. 

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

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

  213. 

  214.       if (type == 3) {

  215. 	if (vlc < 3) {

  216. 	  run   = 0;

  217. 	  level = vlc;

  218. 	} else if (vlc < 4) {

  219. 	  run   = 1;

  220. 	  level = 1;

  221. 	} else {

  222. 	  run   = (vlc & 0x3);

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

  224. 	}

  225.       } else {

  226. 	if (vlc < 16) {

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

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

  229. 	} else if (intra) {

  230. 	  run   = (vlc & 0x7);

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

  232. 	} else {

  233. 	  run   = (vlc & 0xF);

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

  235. 	}

  236.       }

  237. 

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

  239. 	return -1;

  240. 

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

  242.     }

  243. 

  244.     if (type != 2) {

  245.       break;

  246.     }

  247.   }

  248. 

  249.   return 0;

  250. }

  251. 

  252. static void sixpel_mc_put (MpegEncContext *s,

  253. 			   uint8_t *src, uint8_t *dst, int stride,

  254. 			   int dxy, int width, int height) {

  255.   int i, j;

  256. 

  257.   switch (dxy) {

  258.   case 6*0+0:

  259.     for (i=0; i < height; i++) {

  260.       memcpy (dst, src, width);

  261.       src += stride;

  262.       dst += stride;

  263.     }

  264.     break;

  265.   case 6*0+2:

  266.     for (i=0; i < height; i++) {

  267.       for (j=0; j < width; j++) {

  268. 	dst[j] = (683*(2*src[j] + src[j+1] + 1)) >> 11;

  269.       }

  270.       src += stride;

  271.       dst += stride;

  272.     }

  273.     break;

  274.   case 6*0+3:

  275.     for (i=0; i < height; i++) {

  276.       for (j=0; j < width; j++) {

  277. 	dst[j] = (src[j] + src[j+1] + 1) >> 1;

  278.       }

  279.       src += stride;

  280.       dst += stride;

  281.     }

  282.     break;

  283.   case 6*0+4:

  284.     for (i=0; i < height; i++) {

  285.       for (j=0; j < width; j++) {

  286. 	dst[j] = (683*(src[j] + 2*src[j+1] + 1)) >> 11;

  287.       }

  288.       src += stride;

  289.       dst += stride;

  290.     }

  291.     break;

  292.   case 6*2+0:

  293.     for (i=0; i < height; i++) {

  294.       for (j=0; j < width; j++) {

  295. 	dst[j] = (683*(2*src[j] + src[j+stride] + 1)) >> 11;

  296.       }

  297.       src += stride;

  298.       dst += stride;

  299.     }

  300.     break;

  301.   case 6*2+2:

  302.     for (i=0; i < height; i++) {

  303.       for (j=0; j < width; j++) {

  304. 	dst[j] = (2731*(4*src[j] + 3*src[j+1] + 3*src[j+stride] + 2*src[j+stride+1] + 6)) >> 15;

  305.       }

  306.       src += stride;

  307.       dst += stride;

  308.     }

  309.     break;

  310.   case 6*2+4:

  311.     for (i=0; i < height; i++) {

  312.       for (j=0; j < width; j++) {

  313. 	dst[j] = (2731*(3*src[j] + 4*src[j+1] + 2*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15;

  314.       }

  315.       src += stride;

  316.       dst += stride;

  317.     }

  318.     break;

  319.   case 6*3+0:

  320.     for (i=0; i < height; i++) {

  321.       for (j=0; j < width; j++) {

  322. 	dst[j] = (src[j] + src[j+stride]+1) >> 1;

  323.       }

  324.       src += stride;

  325.       dst += stride;

  326.     }

  327.     break;

  328.   case 6*3+3:

  329.     for (i=0; i < height; i++) {

  330.       for (j=0; j < width; j++) {

  331. 	dst[j] = (src[j] + src[j+1] + src[j+stride] + src[j+stride+1] + 2) >> 2;

  332.       }

  333.       src += stride;

  334.       dst += stride;

  335.     }

  336.     break;

  337.   case 6*4+0:

  338.     for (i=0; i < height; i++) {

  339.       for (j=0; j < width; j++) {

  340. 	dst[j] = (683*(src[j] + 2*src[j+stride] + 1)) >> 11;

  341.       }

  342.       src += stride;

  343.       dst += stride;

  344.     }

  345.     break;

  346.   case 6*4+2:

  347.     for (i=0; i < height; i++) {

  348.       for (j=0; j < width; j++) {

  349. 	dst[j] = (2731*(3*src[j] + 2*src[j+1] + 4*src[j+stride] + 3*src[j+stride+1] + 6)) >> 15;

  350.       }

  351.       src += stride;

  352.       dst += stride;

  353.     }

  354.     break;

  355.   case 6*4+4:

  356.     for (i=0; i < height; i++) {

  357.       for (j=0; j < width; j++) {

  358. 	dst[j] = (2731*(2*src[j] + 3*src[j+1] + 3*src[j+stride] + 4*src[j+stride+1] + 6)) >> 15;

  359.       }

  360.       src += stride;

  361.       dst += stride;

  362.     }

  363.     break;

  364.   }

  365. }

  366. 

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

  368. 				     int width, int height, int mx, int my) {

  369.   uint8_t *src, *dest;

  370.   int i, emu = 0;

  371.   const int sx = ((unsigned) (mx + 0x7FFFFFFE)) % 6;

  372.   const int sy = ((unsigned) (my + 0x7FFFFFFE)) % 6;

  373.   const int dxy= 6*sy + sx;

  374. 

  375.   /* decode and clip motion vector to frame border (+16) */

  376.   mx = x + (mx - sx) / 6;

  377.   my = y + (my - sy) / 6;

  378. 

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

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

  381. 

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

  383.       emu = 1;

  384.     }

  385. 

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

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

  388.   }

  389. 

  390.   /* form component predictions */

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

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

  393. 

  394.   if (emu) {

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

  396. 			 mx, my, s->width, s->height);

  397.     src = s->edge_emu_buffer;

  398.   }

  399.   sixpel_mc_put (s, src, dest, s->linesize, dxy, width, height);

  400. 

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

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

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

  404.     width  = (width  >> 1);

  405.     height = (height >> 1);

  406. 

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

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

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

  410. 

  411.       if (emu) {

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

  413. 			     mx, my, (s->width >> 1), (s->height >> 1));

  414. 	src = s->edge_emu_buffer;

  415.       }

  416.       sixpel_mc_put (s, src, dest, s->uvlinesize, dxy, width, height);

  417.     }

  418.   }

  419. }

  420. 

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

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

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

  424.   uint32_t vlc;

  425.   int8_t *top, *left;

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

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

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

  429. 

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

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

  432.   h->topright_samples_available	= 0xFFFF;

  433. 

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

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

  436. 

  437.     cbp = 0;

  438.     mb_type = MB_TYPE_SKIP;

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

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

  441.       mode = 3;	/* thirdpel */

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

  443.       mode = 2;	/* halfpel */

  444.     } else {

  445.       mode = 1;	/* fullpel */

  446.     }

  447. 

  448.     /* fill caches */

  449.     memset (h->ref_cache[0], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));

  450. 

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

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

  453. 	*(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];

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

  455.       }

  456.     } else {

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

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

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

  460.       }

  461.     }

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

  463.       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));

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

  465. 

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

  467. 	*(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];

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

  469.       }

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

  471. 	*(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];

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

  473.       }

  474.     }

  475. 

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

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

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

  479. 

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

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

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

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

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

  485. 

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

  487. 

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

  489. 	mx = clip (mx, -6*x, 6*(s->width  - part_width  - x));

  490. 	my = clip (my, -6*y, 6*(s->height - part_height - y));

  491. 

  492. 	/* get motion vector differential */

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

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

  495. 

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

  497. 	  return -1;

  498. 	}

  499. 

  500. 	/* compute motion vector */

  501. 	if (mode == 3) {

  502. 	  mx = ((mx + 1) & ~0x1) + 2*dx;

  503. 	  my = ((my + 1) & ~0x1) + 2*dy;

  504. 	} else if (mode == 2) {

  505. 	  mx = (mx + 1) - ((unsigned) (0x7FFFFFFF + mx) % 3) + 3*dx;

  506. 	  my = (my + 1) - ((unsigned) (0x7FFFFFFF + my) % 3) + 3*dy;

  507. 	} else if (mode == 1) {

  508. 	  mx = (mx + 3) - ((unsigned) (0x7FFFFFFB + mx) % 6) + 6*dx;

  509. 	  my = (my + 3) - ((unsigned) (0x7FFFFFFB + my) % 6) + 6*dy;

  510. 	}

  511. 

  512. 	/* update mv_cache */

  513. 	for (l=0; l < part_height; l+=4) {

  514. 	  for (m=0; m < part_width; m+=4) {

  515. 	    k = scan8[0] + ((m + j) >> 2) + ((l + i) << 1);

  516. 	    h->mv_cache [0][k][0] = mx;

  517. 	    h->mv_cache [0][k][1] = my;

  518. 	    h->ref_cache[0][k] = 1;

  519. 	  }

  520. 	}

  521. 

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

  523.       }

  524.     }

  525. 

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

  527.       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));

  528.     }

  529. 

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

  531.       return -1;

  532. 

  533.     cbp = golomb_to_inter_cbp[vlc];

  534.     mb_type = MB_TYPE_16x16;

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

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

  537. 

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

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

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

  541.       }

  542.     }

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

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

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

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

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

  548.     }

  549. 

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

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

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

  553. 

  554.       if (vlc >= 25)

  555. 	return -1;

  556. 

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

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

  559. 

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

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

  562. 

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

  564. 	return -1;

  565.     }

  566. 

  567.     write_back_intra_pred_mode (h);

  568.     check_intra4x4_pred_mode (h);

  569. 

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

  571.       return -1;

  572. 

  573.     cbp = golomb_to_intra4x4_cbp[vlc];

  574.     mb_type = MB_TYPE_INTRA4x4;

  575.   } else {			/* INTRA16x16 */

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

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

  578. 

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

  580.       return -1;

  581. 

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

  583.     mb_type = MB_TYPE_INTRA16x16;

  584.   }

  585. 

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

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

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

  589.     }

  590.   }

  591.   if (!IS_INTRA4x4(mb_type)) {

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

  593.   }

  594.   if (!IS_SKIP(mb_type)) {

  595.     memset (h->mb, 0, 24*16*sizeof(DCTELEM));

  596.     memset (h->non_zero_count_cache, 0, 8*6*sizeof(uint8_t));

  597.   }

  598. 

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

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

  601. 

  602.     if (s->qscale > 31)

  603.       return -1;

  604.   }

  605.   if (IS_INTRA16x16(mb_type)) {

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

  607.       return -1;

  608.   }

  609. 

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

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

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

  613. 

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

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

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

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

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

  619. 

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

  621. 	    return -1;

  622. 	}

  623.       }

  624.     }

  625. 

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

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

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

  629. 	  return -1;

  630.       }

  631. 

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

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

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

  635. 

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

  637. 	    return -1;

  638. 	}

  639.       }

  640.     }

  641.   }

  642. 

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

  644. 

  645.   if (IS_INTRA(mb_type)) {

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

  647.   }

  648. 

  649.   return 0;

  650. }

  651. 

  652. static int svq3_decode_frame (AVCodecContext *avctx,

  653. 			      void *data, int *data_size,

  654. 			      uint8_t *buf, int buf_size) {

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

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

  657.   int i;

  658. 

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

  660. 

  661.   if (!s->context_initialized) {

  662.     s->width = (avctx->width + 15) & ~15;

  663.     s->height = (avctx->height + 15) & ~15;

  664.     h->b_stride = (s->width >> 2);

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

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

  667.     h->halfpel_flag = 1;

  668.     h->thirdpel_flag = 1;

  669.     h->chroma_qp = 4;

  670. 

  671.     if (MPV_common_init (s) < 0)

  672.       return -1;

  673. 

  674.     alloc_tables (h);

  675.   }

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

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

  678. 

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

  680. 

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

  682. 

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

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

  685.       }

  686. 

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

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

  689.     }

  690.   }

  691. 

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

  693.     /* TODO: what? */

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

  695.     return -1;

  696.   } else {

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

  698.     int offset = 0;

  699. 

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

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

  702.     }

  703. 

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

  705.       return -1;

  706. 

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

  708.   }

  709. 

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

  711. 

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

  713.     return -1;

  714. 

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

  716. 

  717.   /* unknown fields */

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

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

  720. 

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

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

  723. 

  724.   /* unknown fields */

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

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

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

  728. 

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

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

  731.   }

  732. 

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

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

  735.     return buf_size;

  736. 

  737.   frame_start (h);

  738. 

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

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

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

  742. 

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

  744. 	mb_type += 8;

  745.       }

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

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

  748. 	return -1;

  749.       }

  750. 

  751.       if (mb_type != 0) {

  752. 	hl_decode_mb (h);

  753.       }

  754.     }

  755.   }

  756. 

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

  758.   *data_size = sizeof(AVFrame);

  759. 

  760.   MPV_frame_end(s);

  761.   

  762.   return buf_size;

  763. }

  764. 

  765. 

  766. AVCodec svq3_decoder = {

  767.     "svq3",

  768.     CODEC_TYPE_VIDEO,

  769.     CODEC_ID_SVQ3,

  770.     sizeof(H264Context),

  771.     decode_init,

  772.     NULL,

  773.     decode_end,

  774.     svq3_decode_frame,

  775.     CODEC_CAP_DR1,

  776. };