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.
3170 Commits
32 Branches
449MB
Tree: 8fe2c60e5f
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '8fe2c60e5f'
${ noResults }
FFmpeg/libavcodec/dv.c

1003 lines
30KB
Raw Blame History 

  1. /*

  2.  * DV decoder

  3.  * Copyright (c) 2002 Fabrice Bellard.

  4.  *

  5.  * DV encoder 

  6.  * Copyright (c) 2003 Roman Shaposhnik.

  7.  *

  8.  * Many thanks to Dan Dennedy <dan@dennedy.org> for providing wealth

  9.  * of DV technical info.

  10.  *

  11.  * This library is free software; you can redistribute it and/or

  12.  * modify it under the terms of the GNU Lesser General Public

  13.  * License as published by the Free Software Foundation; either

  14.  * version 2 of the License, or (at your option) any later version.

  15.  *

  16.  * This library is distributed in the hope that it will be useful,

  17.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  18.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  19.  * Lesser General Public License for more details.

  20.  *

  21.  * You should have received a copy of the GNU Lesser General Public

  22.  * License along with this library; if not, write to the Free Software

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

  24.  */

  25. 

  26. /**

  27.  * @file dv.c

  28.  * DV codec.

  29.  */

  30. #include "avcodec.h"

  31. #include "dsputil.h"

  32. #include "mpegvideo.h"

  33. #include "simple_idct.h"

  34. #include "dvdata.h"

  35. 

  36. typedef struct DVVideoContext {

  37.     const DVprofile* sys;

  38.     AVFrame picture;

  39.     uint8_t *buf;

  40.     

  41.     uint8_t dv_zigzag[2][64];

  42.     uint8_t dv_idct_shift[2][22][64];

  43.   

  44.     void (*get_pixels)(DCTELEM *block, const uint8_t *pixels, int line_size);

  45.     void (*fdct[2])(DCTELEM *block);

  46.     void (*idct_put[2])(uint8_t *dest, int line_size, DCTELEM *block);

  47. } DVVideoContext;

  48. 

  49. #define TEX_VLC_BITS 9

  50. 

  51. #ifdef DV_CODEC_TINY_TARGET

  52. #define DV_VLC_MAP_RUN_SIZE 15

  53. #define DV_VLC_MAP_LEV_SIZE 23

  54. #else

  55. #define DV_VLC_MAP_RUN_SIZE  64 

  56. #define DV_VLC_MAP_LEV_SIZE 512

  57. #endif

  58. 

  59. /* MultiThreading */

  60. static uint8_t** dv_anchor;

  61. 

  62. /* XXX: also include quantization */

  63. static RL_VLC_ELEM *dv_rl_vlc;

  64. /* VLC encoding lookup table */

  65. static struct dv_vlc_pair {

  66.    uint32_t vlc;

  67.    uint8_t  size;

  68. } (*dv_vlc_map)[DV_VLC_MAP_LEV_SIZE] = NULL;

  69. 

  70. static void dv_build_unquantize_tables(DVVideoContext *s, uint8_t* perm)

  71. {

  72.     int i, q, j;

  73. 

  74.     /* NOTE: max left shift is 6 */

  75.     for(q = 0; q < 22; q++) {

  76.         /* 88DCT */

  77.         for(i = 1; i < 64; i++) {

  78.             /* 88 table */

  79.             j = perm[i];

  80.             s->dv_idct_shift[0][q][j] =

  81.                 dv_quant_shifts[q][dv_88_areas[i]] + 1;

  82.         }

  83.         

  84.         /* 248DCT */

  85.         for(i = 1; i < 64; i++) {

  86.             /* 248 table */

  87.             s->dv_idct_shift[1][q][i] =  

  88.                 dv_quant_shifts[q][dv_248_areas[i]] + 1;

  89.         }

  90.     }

  91. }

  92. 

  93. static int dvvideo_init(AVCodecContext *avctx)

  94. {

  95.     DVVideoContext *s = avctx->priv_data;

  96.     DSPContext dsp;

  97.     static int done=0;

  98.     int i, j;

  99. 

  100.     if (!done) {

  101.         int i;

  102.         VLC dv_vlc;

  103. 

  104.         done = 1;

  105. 

  106.         dv_vlc_map = av_mallocz(DV_VLC_MAP_LEV_SIZE*DV_VLC_MAP_RUN_SIZE*sizeof(struct dv_vlc_pair));

  107. 	if (!dv_vlc_map)

  108. 	    return -ENOMEM;

  109. 

  110. 	dv_anchor = av_malloc(12*27*sizeof(void*));

  111. 	if (!dv_anchor) {

  112. 	    av_free(dv_vlc_map);

  113. 	    return -ENOMEM;

  114. 	}

  115. 	for (i=0; i<12*27; i++)

  116. 	    (int)dv_anchor[i] = i;

  117. 	

  118.         /* NOTE: as a trick, we use the fact the no codes are unused

  119.            to accelerate the parsing of partial codes */

  120.         init_vlc(&dv_vlc, TEX_VLC_BITS, NB_DV_VLC, 

  121.                  dv_vlc_len, 1, 1, dv_vlc_bits, 2, 2);

  122. 

  123.         dv_rl_vlc = av_malloc(dv_vlc.table_size * sizeof(RL_VLC_ELEM));

  124. 	if (!dv_rl_vlc) {

  125. 	    av_free(dv_anchor);

  126. 	    av_free(dv_vlc_map);

  127. 	    return -ENOMEM;

  128. 	}

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

  130.             int code= dv_vlc.table[i][0];

  131.             int len = dv_vlc.table[i][1];

  132.             int level, run;

  133.         

  134.             if(len<0){ //more bits needed

  135.                 run= 0;

  136.                 level= code;

  137.             } else {

  138.                 run=   dv_vlc_run[code] + 1;

  139.                 level= dv_vlc_level[code];

  140.             }

  141.             dv_rl_vlc[i].len = len;

  142.             dv_rl_vlc[i].level = level;

  143.             dv_rl_vlc[i].run = run;

  144.         }

  145. 	free_vlc(&dv_vlc);

  146. 

  147. 	for (i = 0; i < NB_DV_VLC - 1; i++) {

  148.            if (dv_vlc_run[i] >= DV_VLC_MAP_RUN_SIZE || dv_vlc_level[i] >= DV_VLC_MAP_LEV_SIZE)

  149. 	       continue;

  150. 	   

  151. 	   if (dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size != 0)

  152. 	       continue;

  153. 	       

  154. 	   dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].vlc = dv_vlc_bits[i] << 

  155. 	                                                    (!!dv_vlc_level[i]);

  156. 	   dv_vlc_map[dv_vlc_run[i]][dv_vlc_level[i]].size = dv_vlc_len[i] + 

  157. 	                                                     (!!dv_vlc_level[i]);

  158. 	}

  159. 	for (i = 0; i < DV_VLC_MAP_RUN_SIZE; i++) {

  160. #ifdef DV_CODEC_TINY_TARGET

  161. 	   for (j = 1; j < DV_VLC_MAP_LEV_SIZE; j++) {

  162. 	      if (dv_vlc_map[i][j].size == 0) {

  163. 	          dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |

  164. 		            (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size));

  165. 	          dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size + 

  166. 		                          dv_vlc_map[0][j].size;

  167. 	      }

  168. 	   }

  169. #else

  170. 	   for (j = 1; j < DV_VLC_MAP_LEV_SIZE/2; j++) {

  171. 	      if (dv_vlc_map[i][j].size == 0) {

  172. 	          dv_vlc_map[i][j].vlc = dv_vlc_map[0][j].vlc |

  173. 		            (dv_vlc_map[i-1][0].vlc << (dv_vlc_map[0][j].size));

  174. 	          dv_vlc_map[i][j].size = dv_vlc_map[i-1][0].size + 

  175. 		                          dv_vlc_map[0][j].size;

  176. 	      }

  177. 	      dv_vlc_map[i][((uint16_t)(-j))&0x1ff].vlc = 

  178. 	                                    dv_vlc_map[i][j].vlc | 1;

  179. 	      dv_vlc_map[i][((uint16_t)(-j))&0x1ff].size = 

  180. 	                                    dv_vlc_map[i][j].size;

  181. 	   }

  182. #endif

  183. 	}

  184.     }

  185. 

  186.     /* Generic DSP setup */

  187.     dsputil_init(&dsp, avctx);

  188.     s->get_pixels = dsp.get_pixels;

  189. 

  190.     /* 88DCT setup */

  191.     s->fdct[0] = dsp.fdct;

  192.     s->idct_put[0] = dsp.idct_put;

  193.     for (i=0; i<64; i++)

  194.        s->dv_zigzag[0][i] = dsp.idct_permutation[ff_zigzag_direct[i]];

  195. 

  196.     /* 248DCT setup */

  197.     s->fdct[1] = dsp.fdct248;

  198.     s->idct_put[1] = simple_idct248_put;  // FIXME: need to add it to DSP

  199.     memcpy(s->dv_zigzag[1], ff_zigzag248_direct, 64);

  200. 

  201.     /* XXX: do it only for constant case */

  202.     dv_build_unquantize_tables(s, dsp.idct_permutation);

  203. 

  204.     /* FIXME: I really don't think this should be here */

  205.     if (dv_codec_profile(avctx))

  206. 	avctx->pix_fmt = dv_codec_profile(avctx)->pix_fmt; 

  207.     avctx->coded_frame = &s->picture;

  208.     

  209.     return 0;

  210. }

  211. 

  212. static int dvvideo_end(AVCodecContext *avctx)

  213. {

  214.     avcodec_default_free_buffers(avctx);    

  215.     return 0;

  216. }

  217. 

  218. // #define VLC_DEBUG

  219. 

  220. typedef struct BlockInfo {

  221.     const uint8_t *shift_table;

  222.     const uint8_t *scan_table;

  223.     uint8_t pos; /* position in block */

  224.     uint8_t dct_mode;

  225.     uint8_t partial_bit_count;

  226.     uint16_t partial_bit_buffer;

  227.     int shift_offset;

  228. } BlockInfo;

  229. 

  230. /* block size in bits */

  231. static const uint16_t block_sizes[6] = {

  232.     112, 112, 112, 112, 80, 80

  233. };

  234. /* bit budget for AC only in 5 MBs */

  235. static const int vs_total_ac_bits = (100 * 4 + 68*2) * 5;

  236. /* see dv_88_areas and dv_248_areas for details */

  237. static const int mb_area_start[5] = { 1, 6, 21, 43, 64 }; 

  238. 

  239. #ifndef ALT_BITSTREAM_READER

  240. #warning only works with ALT_BITSTREAM_READER

  241. #endif

  242. 

  243. static inline int get_bits_left(GetBitContext *s)

  244. {

  245.     return s->size_in_bits - get_bits_count(s);

  246. }

  247. 

  248. static inline int get_bits_size(GetBitContext *s)

  249. {

  250.     return s->size_in_bits;

  251. }

  252. 

  253. static inline int put_bits_left(PutBitContext* s)

  254. {

  255.     return (s->buf_end - s->buf) * 8 - put_bits_count(s);

  256. }

  257. 

  258. /* decode ac coefs */

  259. static void dv_decode_ac(GetBitContext *gb, BlockInfo *mb, DCTELEM *block)

  260. {

  261.     int last_index = get_bits_size(gb);

  262.     int last_re_index;

  263.     int shift_offset = mb->shift_offset;

  264.     const uint8_t *scan_table = mb->scan_table;

  265.     const uint8_t *shift_table = mb->shift_table;

  266.     int pos = mb->pos;

  267.     int level, pos1, run;

  268.     int partial_bit_count;

  269.     int sign = 0;

  270. #ifndef ALT_BITSTREAM_READER //FIXME

  271.     int re_index=0; 

  272.     int re1_index=0;

  273. #endif

  274.     OPEN_READER(re, gb);

  275.     

  276. #ifdef VLC_DEBUG

  277.     printf("start\n");

  278. #endif

  279. 

  280.     /* if we must parse a partial vlc, we do it here */

  281.     partial_bit_count = mb->partial_bit_count;

  282.     if (partial_bit_count > 0) {

  283.         uint8_t buf[4];

  284.         uint32_t v;

  285.         int l;

  286.         GetBitContext gb1;

  287. 

  288.         /* build the dummy bit buffer */

  289.         l = 16 - partial_bit_count;

  290.         UPDATE_CACHE(re, gb);

  291. #ifdef VLC_DEBUG

  292.         printf("show=%04x\n", SHOW_UBITS(re, gb, 16));

  293. #endif

  294.         v = (mb->partial_bit_buffer << l) | SHOW_UBITS(re, gb, l);

  295.         buf[0] = v >> 8;

  296.         buf[1] = v;

  297. #ifdef VLC_DEBUG

  298.         printf("v=%04x cnt=%d %04x\n", 

  299.                v, partial_bit_count, (mb->partial_bit_buffer << l));

  300. #endif

  301.         /* try to read the codeword */

  302.         init_get_bits(&gb1, buf, 4*8);

  303.         {

  304.             OPEN_READER(re1, &gb1);

  305.             UPDATE_CACHE(re1, &gb1);

  306.             GET_RL_VLC(level, run, re1, &gb1, dv_rl_vlc, 

  307.                        TEX_VLC_BITS, 2);

  308.             l = re1_index;

  309.             CLOSE_READER(re1, &gb1);

  310.         }

  311. #ifdef VLC_DEBUG

  312.         printf("****run=%d level=%d size=%d\n", run, level, l);

  313. #endif

  314.         /* compute codeword length -- if too long, we cannot parse */

  315.         l -= partial_bit_count;

  316.         if ((re_index + l + (level != 0)) > last_index) {

  317. 	    mb->partial_bit_count += (last_index - re_index);

  318. 	    mb->partial_bit_buffer = v >> (16 - mb->partial_bit_count);

  319.             return;

  320. 	}

  321. 	

  322.         /* skip read bits */

  323.         last_re_index = 0; /* avoid warning */

  324.         re_index += l;

  325.         /* by definition, if we can read the vlc, all partial bits

  326.            will be read (otherwise we could have read the vlc before) */

  327.         mb->partial_bit_count = 0;

  328.         UPDATE_CACHE(re, gb);

  329.         goto handle_vlc;

  330.     }

  331. 

  332.     /* get the AC coefficients until last_index is reached */

  333.     for(;;) {

  334.         UPDATE_CACHE(re, gb);

  335. #ifdef VLC_DEBUG

  336.         printf("%2d: bits=%04x index=%d\n", 

  337.                pos, SHOW_UBITS(re, gb, 16), re_index);

  338. #endif

  339.         last_re_index = re_index;

  340.         GET_RL_VLC(level, run, re, gb, dv_rl_vlc, 

  341.                    TEX_VLC_BITS, 2);

  342.     handle_vlc:

  343. #ifdef VLC_DEBUG

  344.         printf("run=%d level=%d\n", run, level);

  345. #endif

  346. 	if (level) {

  347. 	    sign = SHOW_SBITS(re, gb, 1);

  348. 	    LAST_SKIP_BITS(re, gb, 1);

  349. 	}

  350. 	if (re_index > last_index) {

  351. 	    /* should be < 16 bits otherwise a codeword could have been parsed */

  352. 	    re_index = last_re_index;

  353. 	    UPDATE_CACHE(re, gb);

  354. 	    mb->partial_bit_count = last_index - re_index;

  355. 	    mb->partial_bit_buffer = SHOW_UBITS(re, gb, mb->partial_bit_count);

  356. 	    re_index = last_index;

  357. 	    break;

  358. 	}

  359.         

  360. 	pos += run; 	

  361. 	if (pos >= 64)

  362. 	    break;

  363.         

  364. 	if (level) {

  365.             level = (level ^ sign) - sign;

  366.             pos1 = scan_table[pos];

  367.             level = level << (shift_table[pos1] + shift_offset);

  368.             block[pos1] = level;

  369.             //            printf("run=%d level=%d shift=%d\n", run, level, shift_table[pos1]);

  370.         } 

  371.     }

  372.     CLOSE_READER(re, gb);

  373.     mb->pos = pos;

  374. }

  375. 

  376. static inline void bit_copy(PutBitContext *pb, GetBitContext *gb)

  377. {

  378.     int bits_left = get_bits_left(gb);

  379.     while (bits_left >= 16) {

  380.         put_bits(pb, 16, get_bits(gb, 16));

  381.         bits_left -= 16;

  382.     }

  383.     if (bits_left > 0) {

  384.         put_bits(pb, bits_left, get_bits(gb, bits_left));

  385.     }

  386. }

  387. 

  388. /* mb_x and mb_y are in units of 8 pixels */

  389. static inline void dv_decode_video_segment(DVVideoContext *s, 

  390.                                            uint8_t *buf_ptr1, 

  391.                                            const uint16_t *mb_pos_ptr)

  392. {

  393.     int quant, dc, dct_mode, class1, j;

  394.     int mb_index, mb_x, mb_y, v, last_index;

  395.     DCTELEM *block, *block1;

  396.     int c_offset;

  397.     uint8_t *y_ptr;

  398.     void (*idct_put)(uint8_t *dest, int line_size, DCTELEM *block);

  399.     uint8_t *buf_ptr;

  400.     PutBitContext pb, vs_pb;

  401.     GetBitContext gb;

  402.     BlockInfo mb_data[5 * 6], *mb, *mb1;

  403.     DCTELEM sblock[5*6][64] __align8;

  404.     uint8_t mb_bit_buffer[80 + 4]; /* allow some slack */

  405.     uint8_t vs_bit_buffer[5 * 80 + 4]; /* allow some slack */

  406. 	    

  407.     memset(sblock, 0, sizeof(sblock));

  408. 

  409.     /* pass 1 : read DC and AC coefficients in blocks */

  410.     buf_ptr = buf_ptr1;

  411.     block1 = &sblock[0][0];

  412.     mb1 = mb_data;

  413.     init_put_bits(&vs_pb, vs_bit_buffer, 5 * 80);

  414.     for(mb_index = 0; mb_index < 5; mb_index++, mb1 += 6, block1 += 6 * 64) {

  415.         /* skip header */

  416.         quant = buf_ptr[3] & 0x0f;

  417.         buf_ptr += 4;

  418.         init_put_bits(&pb, mb_bit_buffer, 80);

  419.         mb = mb1;

  420.         block = block1;

  421.         for(j = 0;j < 6; j++) {

  422.             last_index = block_sizes[j];

  423. 	    init_get_bits(&gb, buf_ptr, last_index);

  424.             

  425.             /* get the dc */

  426.             dc = get_bits(&gb, 9);

  427.             dc = (dc << (32 - 9)) >> (32 - 9);

  428.             dct_mode = get_bits1(&gb);

  429.             mb->dct_mode = dct_mode;

  430.             mb->scan_table = s->dv_zigzag[dct_mode];

  431.             class1 = get_bits(&gb, 2);

  432.             mb->shift_offset = (class1 == 3);

  433.             mb->shift_table = s->dv_idct_shift[dct_mode]

  434.                 [quant + dv_quant_offset[class1]];

  435.             dc = dc << 2;

  436.             /* convert to unsigned because 128 is not added in the

  437.                standard IDCT */

  438.             dc += 1024;

  439.             block[0] = dc;

  440.             buf_ptr += last_index >> 3;

  441.             mb->pos = 0;

  442.             mb->partial_bit_count = 0;

  443. 

  444. #ifdef VLC_DEBUG

  445.             printf("MB block: %d, %d ", mb_index, j);

  446. #endif

  447.             dv_decode_ac(&gb, mb, block);

  448. 

  449.             /* write the remaining bits  in a new buffer only if the

  450.                block is finished */

  451.             if (mb->pos >= 64)

  452.                 bit_copy(&pb, &gb);

  453.             

  454.             block += 64;

  455.             mb++;

  456.         }

  457.         

  458.         /* pass 2 : we can do it just after */

  459. #ifdef VLC_DEBUG

  460.         printf("***pass 2 size=%d MB#=%d\n", put_bits_count(&pb), mb_index);

  461. #endif

  462.         block = block1;

  463.         mb = mb1;

  464.         init_get_bits(&gb, mb_bit_buffer, put_bits_count(&pb));

  465. 	flush_put_bits(&pb);

  466.         for(j = 0;j < 6; j++, block += 64, mb++) {

  467.             if (mb->pos < 64 && get_bits_left(&gb) > 0) {

  468.                 dv_decode_ac(&gb, mb, block);

  469.                 /* if still not finished, no need to parse other blocks */

  470.                 if (mb->pos < 64)

  471.                     break;

  472.             }

  473.         }

  474.         /* all blocks are finished, so the extra bytes can be used at

  475.            the video segment level */

  476.         if (j >= 6)

  477. 	    bit_copy(&vs_pb, &gb);

  478.     }

  479. 

  480.     /* we need a pass other the whole video segment */

  481. #ifdef VLC_DEBUG

  482.     printf("***pass 3 size=%d\n", put_bits_count(&vs_pb));

  483. #endif

  484.     block = &sblock[0][0];

  485.     mb = mb_data;

  486.     init_get_bits(&gb, vs_bit_buffer, put_bits_count(&vs_pb));

  487.     flush_put_bits(&vs_pb);

  488.     for(mb_index = 0; mb_index < 5; mb_index++) {

  489.         for(j = 0;j < 6; j++) {

  490.             if (mb->pos < 64) {

  491. #ifdef VLC_DEBUG

  492.                 printf("start %d:%d\n", mb_index, j);

  493. #endif

  494.                 dv_decode_ac(&gb, mb, block);

  495.             }

  496. 	    if (mb->pos >= 64 && mb->pos < 127)

  497. 		av_log(NULL, AV_LOG_ERROR, "AC EOB marker is absent pos=%d\n", mb->pos);

  498.             block += 64;

  499.             mb++;

  500.         }

  501.     }

  502.     

  503.     /* compute idct and place blocks */

  504.     block = &sblock[0][0];

  505.     mb = mb_data;

  506.     for(mb_index = 0; mb_index < 5; mb_index++) {

  507.         v = *mb_pos_ptr++;

  508.         mb_x = v & 0xff;

  509.         mb_y = v >> 8;

  510.         y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8);

  511.         if (s->sys->pix_fmt == PIX_FMT_YUV411P)

  512.             c_offset = (mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8);

  513.         else

  514.             c_offset = ((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8);

  515.         for(j = 0;j < 6; j++) {

  516.             idct_put = s->idct_put[mb->dct_mode];

  517.             if (j < 4) {

  518.                 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {

  519.                     /* NOTE: at end of line, the macroblock is handled as 420 */

  520.                     idct_put(y_ptr + (j * 8), s->picture.linesize[0], block);

  521.                 } else {

  522.                     idct_put(y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]),

  523.                              s->picture.linesize[0], block);

  524.                 }

  525.             } else {

  526.                 if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8)) {

  527.                     uint64_t aligned_pixels[64/8];

  528.                     uint8_t *pixels= (uint8_t*)aligned_pixels;

  529. 		    uint8_t *c_ptr, *c_ptr1, *ptr;

  530.                     int y, linesize;

  531.                     /* NOTE: at end of line, the macroblock is handled as 420 */

  532.                     idct_put(pixels, 8, block);

  533.                     linesize = s->picture.linesize[6 - j];

  534.                     c_ptr = s->picture.data[6 - j] + c_offset;

  535.                     ptr = pixels;

  536.                     for(y = 0;y < 8; y++) {

  537.                         /* convert to 411P */

  538.                         c_ptr1 = c_ptr + 8*linesize;

  539.                         c_ptr[0]= ptr[0]; c_ptr1[0]= ptr[4];

  540.                         c_ptr[1]= ptr[1]; c_ptr1[1]= ptr[5];

  541.                         c_ptr[2]= ptr[2]; c_ptr1[2]= ptr[6];

  542.                         c_ptr[3]= ptr[3]; c_ptr1[3]= ptr[7];

  543.                         c_ptr += linesize;

  544.                         ptr += 8;

  545.                     }

  546.                 } else {

  547.                     /* don't ask me why they inverted Cb and Cr ! */

  548.                     idct_put(s->picture.data[6 - j] + c_offset, 

  549.                              s->picture.linesize[6 - j], block);

  550.                 }

  551.             }

  552.             block += 64;

  553.             mb++;

  554.         }

  555.     }

  556. }

  557. 

  558. #ifdef DV_CODEC_TINY_TARGET

  559. /* Converts run and level (where level != 0) pair into vlc, returning bit size */

  560. static always_inline int dv_rl2vlc(int run, int l, uint32_t* vlc)

  561. {

  562.     int sign = l >> 8;

  563.     int level = (l ^ sign) - sign;

  564.     int size;

  565.     

  566.     sign = (sign & 1);

  567. 

  568.     if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {

  569.         *vlc = dv_vlc_map[run][level].vlc | sign;

  570. 	size = dv_vlc_map[run][level].size;

  571.     }

  572.     else { 

  573.         if (level < DV_VLC_MAP_LEV_SIZE) {

  574. 	    *vlc = dv_vlc_map[0][level].vlc | sign;

  575. 	    size = dv_vlc_map[0][level].size;

  576. 	} else {

  577.             *vlc = 0xfe00 | (level << 1) | sign;

  578. 	    size = 16;

  579. 	}

  580. 	if (run) {

  581. 	    *vlc |= ((run < 16) ? dv_vlc_map[run-1][0].vlc : 

  582. 	                          (0x1f80 | (run - 1))) << size;

  583. 	    size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;

  584. 	}

  585.     }

  586.     

  587.     return size;

  588. }

  589. 

  590. static always_inline int dv_rl2vlc_size(int run, int l)

  591. {

  592.     int level = (l ^ (l >> 8)) - (l >> 8);

  593.     int size;

  594.     

  595.     if (run < DV_VLC_MAP_RUN_SIZE && level < DV_VLC_MAP_LEV_SIZE) {

  596. 	size = dv_vlc_map[run][level].size; 

  597.     }

  598.     else { 

  599. 	size = (level < DV_VLC_MAP_LEV_SIZE) ? dv_vlc_map[0][level].size : 16;

  600. 	if (run) {

  601. 	    size += (run < 16) ? dv_vlc_map[run-1][0].size : 13;

  602. 	}

  603.     }

  604.     return size;

  605. }

  606. #else

  607. static always_inline int dv_rl2vlc(int run, int l, uint32_t* vlc)

  608. {

  609.     *vlc = dv_vlc_map[run][((uint16_t)l)&0x1ff].vlc;

  610.     return dv_vlc_map[run][((uint16_t)l)&0x1ff].size;

  611. }

  612. 

  613. static always_inline int dv_rl2vlc_size(int run, int l)

  614. {

  615.     return dv_vlc_map[run][((uint16_t)l)&0x1ff].size;

  616. }

  617. #endif

  618. 

  619. typedef struct EncBlockInfo {

  620.     int area_q[4];

  621.     int bit_size[4];

  622.     int prev_run[4];

  623.     int cur_ac;

  624.     int cno;

  625.     int dct_mode;

  626.     DCTELEM *mb;

  627.     uint8_t partial_bit_count;

  628.     uint32_t partial_bit_buffer; /* we can't use uint16_t here */

  629. } EncBlockInfo;

  630. 

  631. static always_inline void dv_encode_ac(EncBlockInfo* bi, PutBitContext* pb_pool, 

  632.                                        int pb_size)

  633. {

  634.     int run;

  635.     int bits_left;

  636.     PutBitContext* pb = pb_pool;

  637.     int size = bi->partial_bit_count;

  638.     uint32_t vlc = bi->partial_bit_buffer;

  639.     

  640.     bi->partial_bit_count = bi->partial_bit_buffer = 0;

  641. vlc_loop:

  642.        /* Find suitable storage space */

  643.        for (; size > (bits_left = put_bits_left(pb)); pb++) {

  644.           if (bits_left) {

  645.               size -= bits_left;

  646. 	      put_bits(pb, bits_left, vlc >> size);

  647. 	      vlc = vlc & ((1<<size)-1);

  648. 	  }

  649. 	  if (pb_size == 1) {

  650. 	      bi->partial_bit_count = size;

  651. 	      bi->partial_bit_buffer = vlc;

  652. 	      return;

  653. 	  }

  654. 	  --pb_size;

  655.        }

  656.        

  657.        /* Store VLC */

  658.        put_bits(pb, size, vlc);

  659.        

  660.        /* Construct the next VLC */

  661.        run = 0;

  662.        for (; bi->cur_ac < 64; bi->cur_ac++, run++) {

  663.            if (bi->mb[bi->cur_ac]) {

  664. 	       size = dv_rl2vlc(run, bi->mb[bi->cur_ac], &vlc);

  665. 	       bi->cur_ac++;

  666. 	       goto vlc_loop;

  667. 	   }

  668.        }

  669.    

  670.        if (bi->cur_ac == 64) {

  671.            size = 4; vlc = 6; /* End Of Block stamp */

  672. 	   bi->cur_ac++;

  673. 	   goto vlc_loop;

  674.        }

  675. }

  676. 

  677. static always_inline void dv_set_class_number(DCTELEM* blk, EncBlockInfo* bi, 

  678.                                               const uint8_t* zigzag_scan, int bias)

  679. {

  680.     int i, area;

  681.     int run;

  682.     int classes[] = {12, 24, 36, 0xffff};

  683. 

  684.     run = 0;

  685.     bi->mb[0] = blk[0]; 

  686.     bi->cno = 0;

  687.     for (area = 0; area < 4; area++) {

  688.        bi->prev_run[area] = run;

  689.        bi->bit_size[area] = 0;

  690.        for (i=mb_area_start[area]; i<mb_area_start[area+1]; i++) {

  691.           bi->mb[i] = (blk[zigzag_scan[i]] / 16);

  692.           while ((bi->mb[i] ^ (bi->mb[i] >> 8)) > classes[bi->cno])

  693.               bi->cno++;

  694.        

  695.           if (bi->mb[i]) {

  696.               bi->bit_size[area] += dv_rl2vlc_size(run, bi->mb[i]);

  697. 	      run = 0;

  698.           } else

  699.               ++run;

  700.        }

  701.     }

  702.     bi->bit_size[3] += 4; /* EOB marker */

  703.     bi->cno += bias;

  704.     

  705.     if (bi->cno >= 3) { /* FIXME: we have to recreate bit_size[], prev_run[] */

  706.         bi->cno = 3;

  707. 	for (i=1; i<64; i++)

  708. 	   bi->mb[i] /= 2;

  709.     }

  710. }

  711. 

  712. #define SC(x, y) ((s[x] - s[y]) ^ ((s[x] - s[y]) >> 7))

  713. static always_inline int dv_guess_dct_mode(DCTELEM *blk) {

  714.     DCTELEM *s;

  715.     int score88 = 0;

  716.     int score248 = 0;

  717.     int i;

  718.     

  719.     /* Compute 8-8 score (small values give a better chance for 8-8 DCT) */

  720.     s = blk;

  721.     for(i=0; i<7; i++) {

  722.         score88 += SC(0,  8) + SC(1, 9) + SC(2, 10) + SC(3, 11) + 

  723. 	           SC(4, 12) + SC(5,13) + SC(6, 14) + SC(7, 15);

  724.         s += 8;

  725.     }

  726.     /* Compute 2-4-8 score (small values give a better chance for 2-4-8 DCT) */

  727.     s = blk;

  728.     for(i=0; i<6; i++) {

  729.         score248 += SC(0, 16) + SC(1,17) + SC(2, 18) + SC(3, 19) +

  730. 	            SC(4, 20) + SC(5,21) + SC(6, 22) + SC(7, 23);

  731.         s += 8;

  732.     }

  733. 

  734.     return (score88 - score248 > -10);

  735. }

  736. 

  737. static inline void dv_guess_qnos(EncBlockInfo* blks, int* qnos)

  738. {

  739.     int size[5];

  740.     int i, j, k, a, run;

  741.     EncBlockInfo* b;

  742.     

  743.     do {

  744.        b = blks;

  745.        for (i=0; i<5; i++) {

  746.           if (!qnos[i])

  747. 	      continue;

  748. 	  

  749. 	  qnos[i]--;

  750. 	  size[i] = 0;

  751.           for (j=0; j<6; j++, b++) {

  752. 	     for (a=0; a<4; a++) {

  753. 	        if (b->area_q[a] != dv_quant_shifts[qnos[i] + dv_quant_offset[b->cno]][a]) {

  754. 		    b->bit_size[a] = (a==3)?4:0;

  755. 		    b->area_q[a]++;

  756. 		    run = b->prev_run[a];

  757. 		    for (k=mb_area_start[a]; k<mb_area_start[a+1]; k++) {

  758. 		       b->mb[k] /= 2;

  759. 		       if (b->mb[k]) {

  760.                            b->bit_size[a] += dv_rl2vlc_size(run, b->mb[k]);

  761. 	                   run = 0;

  762.                        } else

  763.                            ++run;

  764. 		    }

  765. 		}

  766. 		size[i] += b->bit_size[a];

  767. 	     }

  768. 	  }

  769.        }

  770.     } while ((vs_total_ac_bits < size[0] + size[1] + size[2] + size[3] + size[4]) && 

  771.              (qnos[0]|qnos[1]|qnos[2]|qnos[3]|qnos[4]));

  772. }

  773. 

  774. /*

  775.  * This is a very rough initial implementaion. The performance is

  776.  * horrible and the weighting is missing. But it's missing from the 

  777.  * decoding step also -- so at least we're on the same page with decoder ;-)

  778.  */

  779. static inline void dv_encode_video_segment(DVVideoContext *s, 

  780.                                            uint8_t *dif, 

  781.                                            const uint16_t *mb_pos_ptr)

  782. {

  783.     int mb_index, i, j, v;

  784.     int mb_x, mb_y, c_offset, linesize; 

  785.     uint8_t*  y_ptr;

  786.     uint8_t*  data;

  787.     uint8_t*  ptr;

  788.     int       do_edge_wrap;

  789.     DCTELEM   block[64] __align8;

  790.     DCTELEM   sblock[5*6][64] __align8;

  791.     EncBlockInfo  enc_blks[5*6];

  792.     PutBitContext pbs[5*6];

  793.     PutBitContext* pb; 

  794.     EncBlockInfo* enc_blk;

  795.     int       vs_bit_size = 0;

  796.     int       qnos[5];

  797.    

  798.     enc_blk = &enc_blks[0];

  799.     pb = &pbs[0];

  800.     for(mb_index = 0; mb_index < 5; mb_index++) {

  801.         v = *mb_pos_ptr++;

  802.         mb_x = v & 0xff;

  803.         mb_y = v >> 8;

  804.         y_ptr = s->picture.data[0] + (mb_y * s->picture.linesize[0] * 8) + (mb_x * 8);

  805. 	c_offset = (s->sys->pix_fmt == PIX_FMT_YUV411P) ?

  806. 	           ((mb_y * s->picture.linesize[1] * 8) + ((mb_x >> 2) * 8)) :

  807. 		   (((mb_y >> 1) * s->picture.linesize[1] * 8) + ((mb_x >> 1) * 8));

  808. 	do_edge_wrap = 0;

  809. 	qnos[mb_index] = 15; /* No quantization */

  810.         ptr = dif + mb_index*80 + 4;

  811.         for(j = 0;j < 6; j++) {

  812.             if (j < 4) {  /* Four Y blocks */

  813. 		/* NOTE: at end of line, the macroblock is handled as 420 */

  814. 		if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x < (704 / 8)) {

  815.                     data = y_ptr + (j * 8);

  816.                 } else {

  817.                     data = y_ptr + ((j & 1) * 8) + ((j >> 1) * 8 * s->picture.linesize[0]);

  818.                 }

  819. 		linesize = s->picture.linesize[0];

  820.             } else {      /* Cr and Cb blocks */

  821. 	        /* don't ask Fabrice why they inverted Cb and Cr ! */

  822. 	        data = s->picture.data[6 - j] + c_offset;

  823. 		linesize = s->picture.linesize[6 - j];

  824. 		if (s->sys->pix_fmt == PIX_FMT_YUV411P && mb_x >= (704 / 8))

  825. 		    do_edge_wrap = 1;

  826. 	    }	

  827.             

  828. 	    /* Everything is set up -- now just copy data -> DCT block */

  829. 	    if (do_edge_wrap) {  /* Edge wrap copy: 4x16 -> 8x8 */

  830. 		uint8_t* d;

  831. 		DCTELEM *b = block;

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

  833. 		   d = data + 8 * linesize;

  834. 		   b[0] = data[0]; b[1] = data[1]; b[2] = data[2]; b[3] = data[3];

  835.                    b[4] =    d[0]; b[5] =    d[1]; b[6] =    d[2]; b[7] =    d[3];

  836. 		   data += linesize;

  837. 		   b += 8;

  838. 		}

  839. 	    } else {             /* Simple copy: 8x8 -> 8x8 */

  840. 	        s->get_pixels(block, data, linesize);

  841. 	    }

  842. 	  

  843.             enc_blk->dct_mode = dv_guess_dct_mode(block);

  844. 	    enc_blk->mb = &sblock[mb_index*6+j][0];

  845. 	    enc_blk->area_q[0] = enc_blk->area_q[1] = enc_blk->area_q[2] = enc_blk->area_q[3] = 0;

  846. 	    enc_blk->partial_bit_count = 0;

  847. 	    enc_blk->partial_bit_buffer = 0;

  848. 	    enc_blk->cur_ac = 1;

  849. 	    

  850. 	    s->fdct[enc_blk->dct_mode](block);

  851. 	    

  852. 	    dv_set_class_number(block, enc_blk, 

  853. 	                        enc_blk->dct_mode ? ff_zigzag248_direct : ff_zigzag_direct,

  854. 				j/4*(j%2));

  855.            

  856.             init_put_bits(pb, ptr, block_sizes[j]/8);

  857. 	    put_bits(pb, 9, (uint16_t)(((enc_blk->mb[0] >> 3) - 1024) >> 2));

  858. 	    put_bits(pb, 1, enc_blk->dct_mode);

  859. 	    put_bits(pb, 2, enc_blk->cno);

  860. 	    

  861. 	    vs_bit_size += enc_blk->bit_size[0] + enc_blk->bit_size[1] +

  862. 	                   enc_blk->bit_size[2] + enc_blk->bit_size[3];

  863. 	    ++enc_blk;

  864. 	    ++pb;

  865. 	    ptr += block_sizes[j]/8;

  866.         }

  867.     }

  868. 

  869.     if (vs_total_ac_bits < vs_bit_size)

  870.         dv_guess_qnos(&enc_blks[0], &qnos[0]);

  871. 

  872.     for (i=0; i<5; i++) {

  873.        dif[i*80 + 3] = qnos[i];

  874.     }

  875. 

  876.     /* First pass over individual cells only */

  877.     for (j=0; j<5*6; j++)

  878.        dv_encode_ac(&enc_blks[j], &pbs[j], 1);

  879. 

  880.     /* Second pass over each MB space */

  881.     for (j=0; j<5*6; j++) {

  882.        if (enc_blks[j].cur_ac < 65 || enc_blks[j].partial_bit_count)

  883.            dv_encode_ac(&enc_blks[j], &pbs[(j/6)*6], 6);

  884.     }

  885. 

  886.     /* Third and final pass over the whole vides segment space */

  887.     for (j=0; j<5*6; j++) {

  888.        if (enc_blks[j].cur_ac < 65 || enc_blks[j].partial_bit_count)

  889.            dv_encode_ac(&enc_blks[j], &pbs[0], 6*5);

  890.     }

  891. 

  892.     for (j=0; j<5*6; j++)

  893.        flush_put_bits(&pbs[j]);

  894. }

  895. 

  896. static int dv_decode_mt(AVCodecContext *avctx, void* sl)

  897. {

  898.     DVVideoContext *s = avctx->priv_data;

  899.     int slice = (int)sl;

  900.     dv_decode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80],

  901. 	                    &s->sys->video_place[slice*5]);

  902.     return 0;

  903. }

  904. 

  905. static int dv_encode_mt(AVCodecContext *avctx, void* sl)

  906. {

  907.     DVVideoContext *s = avctx->priv_data;

  908.     int slice = (int)sl;

  909.     dv_encode_video_segment(s, &s->buf[((slice/27)*6+(slice/3)+slice*5+7)*80],

  910. 	                    &s->sys->video_place[slice*5]);

  911.     return 0;

  912. }

  913. 

  914. /* NOTE: exactly one frame must be given (120000 bytes for NTSC,

  915.    144000 bytes for PAL) */

  916. static int dvvideo_decode_frame(AVCodecContext *avctx, 

  917.                                  void *data, int *data_size,

  918.                                  uint8_t *buf, int buf_size)

  919. {

  920.     DVVideoContext *s = avctx->priv_data;

  921.   

  922.     *data_size=0;

  923.     /* special case for last picture */

  924.     if(buf_size==0)

  925.         return 0;

  926.     

  927.     s->sys = dv_frame_profile(buf);

  928.     if (!s->sys || buf_size < s->sys->frame_size)

  929.         return -1; /* NOTE: we only accept several full frames */

  930. 

  931.     if(s->picture.data[0])

  932.         avctx->release_buffer(avctx, &s->picture);

  933.     

  934.     s->picture.reference = 0;

  935.     avctx->pix_fmt = s->sys->pix_fmt;

  936.     avctx->width = s->sys->width;

  937.     avctx->height = s->sys->height;

  938.     if(avctx->get_buffer(avctx, &s->picture) < 0) {

  939.         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");

  940.         return -1;

  941.     }

  942.     s->picture.interlaced_frame = 1;

  943.     s->picture.top_field_first = 0;

  944. 

  945.     s->buf = buf;

  946.     avctx->execute(avctx, dv_decode_mt, (void**)&dv_anchor[0], NULL, 

  947. 	           s->sys->difseg_size * 27);

  948.     

  949.     emms_c();

  950. 

  951.     /* return image */

  952.     *data_size = sizeof(AVFrame);

  953.     *(AVFrame*)data= s->picture;

  954.     

  955.     return s->sys->frame_size;

  956. }

  957. 

  958. static int dvvideo_encode_frame(AVCodecContext *c, uint8_t *buf, int buf_size, 

  959.                                 void *data)

  960. {

  961.     DVVideoContext *s = c->priv_data;

  962. 

  963.     s->sys = dv_codec_profile(c);

  964.     if (!s->sys)

  965. 	return -1;

  966.     

  967.     c->pix_fmt = s->sys->pix_fmt;

  968.     s->picture = *((AVFrame *)data);

  969. 

  970.     s->buf = buf;

  971.     c->execute(c, dv_encode_mt, (void**)&dv_anchor[0], NULL, 

  972. 	       s->sys->difseg_size * 27);

  973. 

  974.     emms_c();

  975.     return s->sys->frame_size;

  976. }

  977. 

  978. AVCodec dvvideo_encoder = {

  979.     "dvvideo",

  980.     CODEC_TYPE_VIDEO,

  981.     CODEC_ID_DVVIDEO,

  982.     sizeof(DVVideoContext),

  983.     dvvideo_init,

  984.     dvvideo_encode_frame,

  985.     dvvideo_end,

  986.     NULL,

  987.     CODEC_CAP_DR1,

  988.     NULL

  989. };

  990. 

  991. AVCodec dvvideo_decoder = {

  992.     "dvvideo",

  993.     CODEC_TYPE_VIDEO,

  994.     CODEC_ID_DVVIDEO,

  995.     sizeof(DVVideoContext),

  996.     dvvideo_init,

  997.     NULL,

  998.     dvvideo_end,

  999.     dvvideo_decode_frame,

  1000.     CODEC_CAP_DR1,

  1001.     NULL

  1002. };