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.
31742 Commits
32 Branches
498MB
Tree: 06b0246da0
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '06b0246da0'
${ noResults }
FFmpeg/libavcodec/rv34.c

1613 lines
59KB
Raw Blame History 

  1. /*

  2.  * RV30/40 decoder common data

  3.  * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov

  4.  *

  5.  * This file is part of Libav.

  6.  *

  7.  * Libav is free software; you can redistribute it and/or

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

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

  10.  * version 2.1 of the License, or (at your option) any later version.

  11.  *

  12.  * Libav is distributed in the hope that it will be useful,

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

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

  15.  * Lesser General Public License for more details.

  16.  *

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

  18.  * License along with Libav; if not, write to the Free Software

  19.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  20.  */

  21. 

  22. /**

  23.  * @file

  24.  * RV30/40 decoder common data

  25.  */

  26. 

  27. #include "libavutil/internal.h"

  28. 

  29. #include "avcodec.h"

  30. #include "dsputil.h"

  31. #include "mpegvideo.h"

  32. #include "golomb.h"

  33. #include "internal.h"

  34. #include "mathops.h"

  35. #include "rectangle.h"

  36. #include "thread.h"

  37. 

  38. #include "rv34vlc.h"

  39. #include "rv34data.h"

  40. #include "rv34.h"

  41. 

  42. //#define DEBUG

  43. 

  44. static inline void ZERO8x2(void* dst, int stride)

  45. {

  46.     fill_rectangle(dst,                 1, 2, stride, 0, 4);

  47.     fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);

  48. }

  49. 

  50. /** translation of RV30/40 macroblock types to lavc ones */

  51. static const int rv34_mb_type_to_lavc[12] = {

  52.     MB_TYPE_INTRA,

  53.     MB_TYPE_INTRA16x16              | MB_TYPE_SEPARATE_DC,

  54.     MB_TYPE_16x16   | MB_TYPE_L0,

  55.     MB_TYPE_8x8     | MB_TYPE_L0,

  56.     MB_TYPE_16x16   | MB_TYPE_L0,

  57.     MB_TYPE_16x16   | MB_TYPE_L1,

  58.     MB_TYPE_SKIP,

  59.     MB_TYPE_DIRECT2 | MB_TYPE_16x16,

  60.     MB_TYPE_16x8    | MB_TYPE_L0,

  61.     MB_TYPE_8x16    | MB_TYPE_L0,

  62.     MB_TYPE_16x16   | MB_TYPE_L0L1,

  63.     MB_TYPE_16x16   | MB_TYPE_L0    | MB_TYPE_SEPARATE_DC

  64. };

  65. 

  66. 

  67. static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];

  68. 

  69. static int rv34_decode_mv(RV34DecContext *r, int block_type);

  70. 

  71. /**

  72.  * @name RV30/40 VLC generating functions

  73.  * @{

  74.  */

  75. 

  76. static const int table_offs[] = {

  77.       0,   1818,   3622,   4144,   4698,   5234,   5804,   5868,   5900,   5932,

  78.    5996,   6252,   6316,   6348,   6380,   7674,   8944,  10274,  11668,  12250,

  79.   14060,  15846,  16372,  16962,  17512,  18148,  18180,  18212,  18244,  18308,

  80.   18564,  18628,  18660,  18692,  20036,  21314,  22648,  23968,  24614,  26384,

  81.   28190,  28736,  29366,  29938,  30608,  30640,  30672,  30704,  30768,  31024,

  82.   31088,  31120,  31184,  32570,  33898,  35236,  36644,  37286,  39020,  40802,

  83.   41368,  42052,  42692,  43348,  43380,  43412,  43444,  43476,  43604,  43668,

  84.   43700,  43732,  45100,  46430,  47778,  49160,  49802,  51550,  53340,  53972,

  85.   54648,  55348,  55994,  56122,  56154,  56186,  56218,  56346,  56410,  56442,

  86.   56474,  57878,  59290,  60636,  62036,  62682,  64460,  64524,  64588,  64716,

  87.   64844,  66076,  67466,  67978,  68542,  69064,  69648,  70296,  72010,  72074,

  88.   72138,  72202,  72330,  73572,  74936,  75454,  76030,  76566,  77176,  77822,

  89.   79582,  79646,  79678,  79742,  79870,  81180,  82536,  83064,  83672,  84242,

  90.   84934,  85576,  87384,  87448,  87480,  87544,  87672,  88982,  90340,  90902,

  91.   91598,  92182,  92846,  93488,  95246,  95278,  95310,  95374,  95502,  96878,

  92.   98266,  98848,  99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,

  93.  103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,

  94.  111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592

  95. };

  96. 

  97. static VLC_TYPE table_data[117592][2];

  98. 

  99. /**

  100.  * Generate VLC from codeword lengths.

  101.  * @param bits   codeword lengths (zeroes are accepted)

  102.  * @param size   length of input data

  103.  * @param vlc    output VLC

  104.  * @param insyms symbols for input codes (NULL for default ones)

  105.  * @param num    VLC table number (for static initialization)

  106.  */

  107. static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,

  108.                          const int num)

  109. {

  110.     int i;

  111.     int counts[17] = {0}, codes[17];

  112.     uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE];

  113.     uint8_t bits2[MAX_VLC_SIZE];

  114.     int maxbits = 0, realsize = 0;

  115. 

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

  117.         if(bits[i]){

  118.             bits2[realsize] = bits[i];

  119.             syms[realsize] = insyms ? insyms[i] : i;

  120.             realsize++;

  121.             maxbits = FFMAX(maxbits, bits[i]);

  122.             counts[bits[i]]++;

  123.         }

  124.     }

  125. 

  126.     codes[0] = 0;

  127.     for(i = 0; i < 16; i++)

  128.         codes[i+1] = (codes[i] + counts[i]) << 1;

  129.     for(i = 0; i < realsize; i++)

  130.         cw[i] = codes[bits2[i]]++;

  131. 

  132.     vlc->table = &table_data[table_offs[num]];

  133.     vlc->table_allocated = table_offs[num + 1] - table_offs[num];

  134.     init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,

  135.                     bits2, 1, 1,

  136.                     cw,    2, 2,

  137.                     syms,  2, 2, INIT_VLC_USE_NEW_STATIC);

  138. }

  139. 

  140. /**

  141.  * Initialize all tables.

  142.  */

  143. static av_cold void rv34_init_tables(void)

  144. {

  145.     int i, j, k;

  146. 

  147.     for(i = 0; i < NUM_INTRA_TABLES; i++){

  148.         for(j = 0; j < 2; j++){

  149.             rv34_gen_vlc(rv34_table_intra_cbppat   [i][j], CBPPAT_VLC_SIZE,   &intra_vlcs[i].cbppattern[j],     NULL, 19*i + 0 + j);

  150.             rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);

  151.             rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j],  NULL, 19*i + 4 + j);

  152.             for(k = 0; k < 4; k++){

  153.                 rv34_gen_vlc(rv34_table_intra_cbp[i][j+k*2],  CBP_VLC_SIZE,   &intra_vlcs[i].cbp[j][k],         rv34_cbp_code, 19*i + 6 + j*4 + k);

  154.             }

  155.         }

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

  157.             rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);

  158.         }

  159.         rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);

  160.     }

  161. 

  162.     for(i = 0; i < NUM_INTER_TABLES; i++){

  163.         rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);

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

  165.             rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);

  166.         }

  167.         for(j = 0; j < 2; j++){

  168.             rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j],  NULL, i*12 + 100 + j);

  169.             rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);

  170.             rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j],  NULL, i*12 + 104 + j);

  171.         }

  172.         rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);

  173.     }

  174. }

  175. 

  176. /** @} */ // vlc group

  177. 

  178. /**

  179.  * @name RV30/40 4x4 block decoding functions

  180.  * @{

  181.  */

  182. 

  183. /**

  184.  * Decode coded block pattern.

  185.  */

  186. static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)

  187. {

  188.     int pattern, code, cbp=0;

  189.     int ones;

  190.     static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};

  191.     static const int shifts[4] = { 0, 2, 8, 10 };

  192.     const int *curshift = shifts;

  193.     int i, t, mask;

  194. 

  195.     code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);

  196.     pattern = code & 0xF;

  197.     code >>= 4;

  198. 

  199.     ones = rv34_count_ones[pattern];

  200. 

  201.     for(mask = 8; mask; mask >>= 1, curshift++){

  202.         if(pattern & mask)

  203.             cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];

  204.     }

  205. 

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

  207.         t = (modulo_three_table[code] >> (6 - 2*i)) & 3;

  208.         if(t == 1)

  209.             cbp |= cbp_masks[get_bits1(gb)] << i;

  210.         if(t == 2)

  211.             cbp |= cbp_masks[2] << i;

  212.     }

  213.     return cbp;

  214. }

  215. 

  216. /**

  217.  * Get one coefficient value from the bistream and store it.

  218.  */

  219. static inline void decode_coeff(DCTELEM *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)

  220. {

  221.     if(coef){

  222.         if(coef == esc){

  223.             coef = get_vlc2(gb, vlc->table, 9, 2);

  224.             if(coef > 23){

  225.                 coef -= 23;

  226.                 coef = 22 + ((1 << coef) | get_bits(gb, coef));

  227.             }

  228.             coef += esc;

  229.         }

  230.         if(get_bits1(gb))

  231.             coef = -coef;

  232.         *dst = (coef*q + 8) >> 4;

  233.     }

  234. }

  235. 

  236. /**

  237.  * Decode 2x2 subblock of coefficients.

  238.  */

  239. static inline void decode_subblock(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)

  240. {

  241.     int flags = modulo_three_table[code];

  242. 

  243.     decode_coeff(    dst+0, (flags >> 6)    , 3, gb, vlc, q);

  244.     if(is_block2){

  245.         decode_coeff(dst+8, (flags >> 4) & 3, 2, gb, vlc, q);

  246.         decode_coeff(dst+1, (flags >> 2) & 3, 2, gb, vlc, q);

  247.     }else{

  248.         decode_coeff(dst+1, (flags >> 4) & 3, 2, gb, vlc, q);

  249.         decode_coeff(dst+8, (flags >> 2) & 3, 2, gb, vlc, q);

  250.     }

  251.     decode_coeff(    dst+9, (flags >> 0) & 3, 2, gb, vlc, q);

  252. }

  253. 

  254. /**

  255.  * Decode a single coefficient.

  256.  */

  257. static inline void decode_subblock1(DCTELEM *dst, int code, GetBitContext *gb, VLC *vlc, int q)

  258. {

  259.     int coeff = modulo_three_table[code] >> 6;

  260.     decode_coeff(dst, coeff, 3, gb, vlc, q);

  261. }

  262. 

  263. static inline void decode_subblock3(DCTELEM *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc,

  264.                                     int q_dc, int q_ac1, int q_ac2)

  265. {

  266.     int flags = modulo_three_table[code];

  267. 

  268.     decode_coeff(    dst+0, (flags >> 6)    , 3, gb, vlc, q_dc);

  269.     if(is_block2){

  270.         decode_coeff(dst+8, (flags >> 4) & 3, 2, gb, vlc, q_ac1);

  271.         decode_coeff(dst+1, (flags >> 2) & 3, 2, gb, vlc, q_ac1);

  272.     }else{

  273.         decode_coeff(dst+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);

  274.         decode_coeff(dst+8, (flags >> 2) & 3, 2, gb, vlc, q_ac1);

  275.     }

  276.     decode_coeff(    dst+9, (flags >> 0) & 3, 2, gb, vlc, q_ac2);

  277. }

  278. 

  279. /**

  280.  * Decode coefficients for 4x4 block.

  281.  *

  282.  * This is done by filling 2x2 subblocks with decoded coefficients

  283.  * in this order (the same for subblocks and subblock coefficients):

  284.  *  o--o

  285.  *    /

  286.  *   /

  287.  *  o--o

  288.  */

  289. 

  290. static inline int rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)

  291. {

  292.     int code, pattern, has_ac = 1;

  293. 

  294.     code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);

  295. 

  296.     pattern = code & 0x7;

  297. 

  298.     code >>= 3;

  299. 

  300.     if (modulo_three_table[code] & 0x3F) {

  301.         decode_subblock3(dst, code, 0, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);

  302.     } else {

  303.         decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);

  304.         if (!pattern)

  305.             return 0;

  306.         has_ac = 0;

  307.     }

  308. 

  309.     if(pattern & 4){

  310.         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);

  311.         decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient, q_ac2);

  312.     }

  313.     if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block

  314.         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);

  315.         decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient, q_ac2);

  316.     }

  317.     if(pattern & 1){

  318.         code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);

  319.         decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);

  320.     }

  321.     return has_ac || pattern;

  322. }

  323. 

  324. /**

  325.  * @name RV30/40 bitstream parsing

  326.  * @{

  327.  */

  328. 

  329. /**

  330.  * Decode starting slice position.

  331.  * @todo Maybe replace with ff_h263_decode_mba() ?

  332.  */

  333. int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)

  334. {

  335.     int i;

  336.     for(i = 0; i < 5; i++)

  337.         if(rv34_mb_max_sizes[i] >= mb_size - 1)

  338.             break;

  339.     return rv34_mb_bits_sizes[i];

  340. }

  341. 

  342. /**

  343.  * Select VLC set for decoding from current quantizer, modifier and frame type.

  344.  */

  345. static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)

  346. {

  347.     if(mod == 2 && quant < 19) quant += 10;

  348.     else if(mod && quant < 26) quant += 5;

  349.     return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]

  350.                 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];

  351. }

  352. 

  353. /**

  354.  * Decode macroblock header and return CBP in case of success, -1 otherwise.

  355.  */

  356. static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)

  357. {

  358.     MpegEncContext *s = &r->s;

  359.     GetBitContext *gb = &s->gb;

  360.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;

  361.     int i, t;

  362. 

  363.     if(!r->si.type){

  364.         r->is16 = get_bits1(gb);

  365.         if(!r->is16 && !r->rv30){

  366.             if(!get_bits1(gb))

  367.                 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");

  368.         }

  369.         s->current_picture_ptr->f.mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;

  370.         r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;

  371.     }else{

  372.         r->block_type = r->decode_mb_info(r);

  373.         if(r->block_type == -1)

  374.             return -1;

  375.         s->current_picture_ptr->f.mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];

  376.         r->mb_type[mb_pos] = r->block_type;

  377.         if(r->block_type == RV34_MB_SKIP){

  378.             if(s->pict_type == AV_PICTURE_TYPE_P)

  379.                 r->mb_type[mb_pos] = RV34_MB_P_16x16;

  380.             if(s->pict_type == AV_PICTURE_TYPE_B)

  381.                 r->mb_type[mb_pos] = RV34_MB_B_DIRECT;

  382.         }

  383.         r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->f.mb_type[mb_pos]);

  384.         rv34_decode_mv(r, r->block_type);

  385.         if(r->block_type == RV34_MB_SKIP){

  386.             fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));

  387.             return 0;

  388.         }

  389.         r->chroma_vlc = 1;

  390.         r->luma_vlc   = 0;

  391.     }

  392.     if(IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){

  393.         if(r->is16){

  394.             t = get_bits(gb, 2);

  395.             fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));

  396.             r->luma_vlc   = 2;

  397.         }else{

  398.             if(r->decode_intra_types(r, gb, intra_types) < 0)

  399.                 return -1;

  400.             r->luma_vlc   = 1;

  401.         }

  402.         r->chroma_vlc = 0;

  403.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);

  404.     }else{

  405.         for(i = 0; i < 16; i++)

  406.             intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;

  407.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);

  408.         if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){

  409.             r->is16 = 1;

  410.             r->chroma_vlc = 1;

  411.             r->luma_vlc   = 2;

  412.             r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);

  413.         }

  414.     }

  415. 

  416.     return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);

  417. }

  418. 

  419. /** @} */ //bitstream functions

  420. 

  421. /**

  422.  * @name motion vector related code (prediction, reconstruction, motion compensation)

  423.  * @{

  424.  */

  425. 

  426. /** macroblock partition width in 8x8 blocks */

  427. static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };

  428. 

  429. /** macroblock partition height in 8x8 blocks */

  430. static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };

  431. 

  432. /** availability index for subblocks */

  433. static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };

  434. 

  435. /**

  436.  * motion vector prediction

  437.  *

  438.  * Motion prediction performed for the block by using median prediction of

  439.  * motion vectors from the left, top and right top blocks but in corner cases

  440.  * some other vectors may be used instead.

  441.  */

  442. static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)

  443. {

  444.     MpegEncContext *s = &r->s;

  445.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;

  446.     int A[2] = {0}, B[2], C[2];

  447.     int i, j;

  448.     int mx, my;

  449.     int avail_index = avail_indexes[subblock_no];

  450.     int c_off = part_sizes_w[block_type];

  451. 

  452.     mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;

  453.     if(subblock_no == 3)

  454.         c_off = -1;

  455. 

  456.     if(r->avail_cache[avail_index - 1]){

  457.         A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][0];

  458.         A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-1][1];

  459.     }

  460.     if(r->avail_cache[avail_index - 4]){

  461.         B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][0];

  462.         B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride][1];

  463.     }else{

  464.         B[0] = A[0];

  465.         B[1] = A[1];

  466.     }

  467.     if(!r->avail_cache[avail_index - 4 + c_off]){

  468.         if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){

  469.             C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][0];

  470.             C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride-1][1];

  471.         }else{

  472.             C[0] = A[0];

  473.             C[1] = A[1];

  474.         }

  475.     }else{

  476.         C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][0];

  477.         C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos-s->b8_stride+c_off][1];

  478.     }

  479.     mx = mid_pred(A[0], B[0], C[0]);

  480.     my = mid_pred(A[1], B[1], C[1]);

  481.     mx += r->dmv[dmv_no][0];

  482.     my += r->dmv[dmv_no][1];

  483.     for(j = 0; j < part_sizes_h[block_type]; j++){

  484.         for(i = 0; i < part_sizes_w[block_type]; i++){

  485.             s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;

  486.             s->current_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;

  487.         }

  488.     }

  489. }

  490. 

  491. #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)

  492. 

  493. /**

  494.  * Calculate motion vector component that should be added for direct blocks.

  495.  */

  496. static int calc_add_mv(RV34DecContext *r, int dir, int val)

  497. {

  498.     int mul = dir ? -r->weight2 : r->weight1;

  499. 

  500.     return (val * mul + 0x2000) >> 14;

  501. }

  502. 

  503. /**

  504.  * Predict motion vector for B-frame macroblock.

  505.  */

  506. static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],

  507.                                       int A_avail, int B_avail, int C_avail,

  508.                                       int *mx, int *my)

  509. {

  510.     if(A_avail + B_avail + C_avail != 3){

  511.         *mx = A[0] + B[0] + C[0];

  512.         *my = A[1] + B[1] + C[1];

  513.         if(A_avail + B_avail + C_avail == 2){

  514.             *mx /= 2;

  515.             *my /= 2;

  516.         }

  517.     }else{

  518.         *mx = mid_pred(A[0], B[0], C[0]);

  519.         *my = mid_pred(A[1], B[1], C[1]);

  520.     }

  521. }

  522. 

  523. /**

  524.  * motion vector prediction for B-frames

  525.  */

  526. static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)

  527. {

  528.     MpegEncContext *s = &r->s;

  529.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;

  530.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;

  531.     int A[2], B[2], C[2];

  532.     int has_A = 0, has_B = 0, has_C = 0;

  533.     int mx, my;

  534.     int i, j;

  535.     Picture *cur_pic = s->current_picture_ptr;

  536.     const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;

  537.     int type = cur_pic->f.mb_type[mb_pos];

  538. 

  539.     memset(A, 0, sizeof(A));

  540.     memset(B, 0, sizeof(B));

  541.     memset(C, 0, sizeof(C));

  542.     if((r->avail_cache[6-1] & type) & mask){

  543.         A[0] = cur_pic->f.motion_val[dir][mv_pos - 1][0];

  544.         A[1] = cur_pic->f.motion_val[dir][mv_pos - 1][1];

  545.         has_A = 1;

  546.     }

  547.     if((r->avail_cache[6-4] & type) & mask){

  548.         B[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][0];

  549.         B[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride][1];

  550.         has_B = 1;

  551.     }

  552.     if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){

  553.         C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][0];

  554.         C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride + 2][1];

  555.         has_C = 1;

  556.     }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){

  557.         C[0] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][0];

  558.         C[1] = cur_pic->f.motion_val[dir][mv_pos - s->b8_stride - 1][1];

  559.         has_C = 1;

  560.     }

  561. 

  562.     rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);

  563. 

  564.     mx += r->dmv[dir][0];

  565.     my += r->dmv[dir][1];

  566. 

  567.     for(j = 0; j < 2; j++){

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

  569.             cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;

  570.             cur_pic->f.motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;

  571.         }

  572.     }

  573.     if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){

  574.         ZERO8x2(cur_pic->f.motion_val[!dir][mv_pos], s->b8_stride);

  575.     }

  576. }

  577. 

  578. /**

  579.  * motion vector prediction - RV3 version

  580.  */

  581. static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)

  582. {

  583.     MpegEncContext *s = &r->s;

  584.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;

  585.     int A[2] = {0}, B[2], C[2];

  586.     int i, j, k;

  587.     int mx, my;

  588.     int avail_index = avail_indexes[0];

  589. 

  590.     if(r->avail_cache[avail_index - 1]){

  591.         A[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][0];

  592.         A[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - 1][1];

  593.     }

  594.     if(r->avail_cache[avail_index - 4]){

  595.         B[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][0];

  596.         B[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride][1];

  597.     }else{

  598.         B[0] = A[0];

  599.         B[1] = A[1];

  600.     }

  601.     if(!r->avail_cache[avail_index - 4 + 2]){

  602.         if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){

  603.             C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][0];

  604.             C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride - 1][1];

  605.         }else{

  606.             C[0] = A[0];

  607.             C[1] = A[1];

  608.         }

  609.     }else{

  610.         C[0] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][0];

  611.         C[1] = s->current_picture_ptr->f.motion_val[0][mv_pos - s->b8_stride + 2][1];

  612.     }

  613.     mx = mid_pred(A[0], B[0], C[0]);

  614.     my = mid_pred(A[1], B[1], C[1]);

  615.     mx += r->dmv[0][0];

  616.     my += r->dmv[0][1];

  617.     for(j = 0; j < 2; j++){

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

  619.             for(k = 0; k < 2; k++){

  620.                 s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;

  621.                 s->current_picture_ptr->f.motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;

  622.             }

  623.         }

  624.     }

  625. }

  626. 

  627. static const int chroma_coeffs[3] = { 0, 3, 5 };

  628. 

  629. /**

  630.  * generic motion compensation function

  631.  *

  632.  * @param r decoder context

  633.  * @param block_type type of the current block

  634.  * @param xoff horizontal offset from the start of the current block

  635.  * @param yoff vertical offset from the start of the current block

  636.  * @param mv_off offset to the motion vector information

  637.  * @param width width of the current partition in 8x8 blocks

  638.  * @param height height of the current partition in 8x8 blocks

  639.  * @param dir motion compensation direction (i.e. from the last or the next reference frame)

  640.  * @param thirdpel motion vectors are specified in 1/3 of pixel

  641.  * @param qpel_mc a set of functions used to perform luma motion compensation

  642.  * @param chroma_mc a set of functions used to perform chroma motion compensation

  643.  */

  644. static inline void rv34_mc(RV34DecContext *r, const int block_type,

  645.                           const int xoff, const int yoff, int mv_off,

  646.                           const int width, const int height, int dir,

  647.                           const int thirdpel, int weighted,

  648.                           qpel_mc_func (*qpel_mc)[16],

  649.                           h264_chroma_mc_func (*chroma_mc))

  650. {

  651.     MpegEncContext *s = &r->s;

  652.     uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;

  653.     int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;

  654.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;

  655.     int is16x16 = 1;

  656. 

  657.     if(thirdpel){

  658.         int chroma_mx, chroma_my;

  659.         mx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);

  660.         my = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);

  661.         lx = (s->current_picture_ptr->f.motion_val[dir][mv_pos][0] + (3 << 24)) % 3;

  662.         ly = (s->current_picture_ptr->f.motion_val[dir][mv_pos][1] + (3 << 24)) % 3;

  663.         chroma_mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;

  664.         chroma_my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;

  665.         umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);

  666.         umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);

  667.         uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];

  668.         uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];

  669.     }else{

  670.         int cx, cy;

  671.         mx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] >> 2;

  672.         my = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] >> 2;

  673.         lx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] & 3;

  674.         ly = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] & 3;

  675.         cx = s->current_picture_ptr->f.motion_val[dir][mv_pos][0] / 2;

  676.         cy = s->current_picture_ptr->f.motion_val[dir][mv_pos][1] / 2;

  677.         umx = cx >> 2;

  678.         umy = cy >> 2;

  679.         uvmx = (cx & 3) << 1;

  680.         uvmy = (cy & 3) << 1;

  681.         //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3

  682.         if(uvmx == 6 && uvmy == 6)

  683.             uvmx = uvmy = 4;

  684.     }

  685. 

  686.     if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {

  687.         /* wait for the referenced mb row to be finished */

  688.         int mb_row = FFMIN(s->mb_height - 1,

  689.                            s->mb_y + ((yoff + my + 5 + 8 * height) >> 4));

  690.         AVFrame *f = dir ? &s->next_picture_ptr->f : &s->last_picture_ptr->f;

  691.         ff_thread_await_progress(f, mb_row, 0);

  692.     }

  693. 

  694.     dxy = ly*4 + lx;

  695.     srcY = dir ? s->next_picture_ptr->f.data[0] : s->last_picture_ptr->f.data[0];

  696.     srcU = dir ? s->next_picture_ptr->f.data[1] : s->last_picture_ptr->f.data[1];

  697.     srcV = dir ? s->next_picture_ptr->f.data[2] : s->last_picture_ptr->f.data[2];

  698.     src_x = s->mb_x * 16 + xoff + mx;

  699.     src_y = s->mb_y * 16 + yoff + my;

  700.     uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;

  701.     uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;

  702.     srcY += src_y * s->linesize + src_x;

  703.     srcU += uvsrc_y * s->uvlinesize + uvsrc_x;

  704.     srcV += uvsrc_y * s->uvlinesize + uvsrc_x;

  705.     if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||

  706.        (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||

  707.        (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {

  708.         uint8_t *uvbuf = s->edge_emu_buffer + 22 * s->linesize;

  709. 

  710.         srcY -= 2 + 2*s->linesize;

  711.         s->dsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6,

  712.                             src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos);

  713.         srcY = s->edge_emu_buffer + 2 + 2*s->linesize;

  714.         s->dsp.emulated_edge_mc(uvbuf     , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1,

  715.                             uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);

  716.         s->dsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1,

  717.                             uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1);

  718.         srcU = uvbuf;

  719.         srcV = uvbuf + 16;

  720.     }

  721.     if(!weighted){

  722.         Y = s->dest[0] + xoff      + yoff     *s->linesize;

  723.         U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;

  724.         V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;

  725.     }else{

  726.         Y = r->tmp_b_block_y [dir]     +  xoff     +  yoff    *s->linesize;

  727.         U = r->tmp_b_block_uv[dir*2]   + (xoff>>1) + (yoff>>1)*s->uvlinesize;

  728.         V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;

  729.     }

  730. 

  731.     if(block_type == RV34_MB_P_16x8){

  732.         qpel_mc[1][dxy](Y, srcY, s->linesize);

  733.         Y    += 8;

  734.         srcY += 8;

  735.     }else if(block_type == RV34_MB_P_8x16){

  736.         qpel_mc[1][dxy](Y, srcY, s->linesize);

  737.         Y    += 8 * s->linesize;

  738.         srcY += 8 * s->linesize;

  739.     }

  740.     is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);

  741.     qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);

  742.     chroma_mc[2-width]   (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);

  743.     chroma_mc[2-width]   (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);

  744. }

  745. 

  746. static void rv34_mc_1mv(RV34DecContext *r, const int block_type,

  747.                         const int xoff, const int yoff, int mv_off,

  748.                         const int width, const int height, int dir)

  749. {

  750.     rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,

  751.             r->rdsp.put_pixels_tab,

  752.             r->rdsp.put_chroma_pixels_tab);

  753. }

  754. 

  755. static void rv4_weight(RV34DecContext *r)

  756. {

  757.     r->rdsp.rv40_weight_pixels_tab[0](r->s.dest[0],

  758.                                       r->tmp_b_block_y[0],

  759.                                       r->tmp_b_block_y[1],

  760.                                       r->weight1,

  761.                                       r->weight2,

  762.                                       r->s.linesize);

  763.     r->rdsp.rv40_weight_pixels_tab[1](r->s.dest[1],

  764.                                       r->tmp_b_block_uv[0],

  765.                                       r->tmp_b_block_uv[2],

  766.                                       r->weight1,

  767.                                       r->weight2,

  768.                                       r->s.uvlinesize);

  769.     r->rdsp.rv40_weight_pixels_tab[1](r->s.dest[2],

  770.                                       r->tmp_b_block_uv[1],

  771.                                       r->tmp_b_block_uv[3],

  772.                                       r->weight1,

  773.                                       r->weight2,

  774.                                       r->s.uvlinesize);

  775. }

  776. 

  777. static void rv34_mc_2mv(RV34DecContext *r, const int block_type)

  778. {

  779.     int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;

  780. 

  781.     rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,

  782.             r->rdsp.put_pixels_tab,

  783.             r->rdsp.put_chroma_pixels_tab);

  784.     if(!weighted){

  785.         rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,

  786.                 r->rdsp.avg_pixels_tab,

  787.                 r->rdsp.avg_chroma_pixels_tab);

  788.     }else{

  789.         rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,

  790.                 r->rdsp.put_pixels_tab,

  791.                 r->rdsp.put_chroma_pixels_tab);

  792.         rv4_weight(r);

  793.     }

  794. }

  795. 

  796. static void rv34_mc_2mv_skip(RV34DecContext *r)

  797. {

  798.     int i, j;

  799.     int weighted = !r->rv30 && r->weight1 != 8192;

  800. 

  801.     for(j = 0; j < 2; j++)

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

  803.              rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,

  804.                      weighted,

  805.                      r->rdsp.put_pixels_tab,

  806.                      r->rdsp.put_chroma_pixels_tab);

  807.              rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,

  808.                      weighted,

  809.                      weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,

  810.                      weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);

  811.         }

  812.     if(weighted)

  813.         rv4_weight(r);

  814. }

  815. 

  816. /** number of motion vectors in each macroblock type */

  817. static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };

  818. 

  819. /**

  820.  * Decode motion vector differences

  821.  * and perform motion vector reconstruction and motion compensation.

  822.  */

  823. static int rv34_decode_mv(RV34DecContext *r, int block_type)

  824. {

  825.     MpegEncContext *s = &r->s;

  826.     GetBitContext *gb = &s->gb;

  827.     int i, j, k, l;

  828.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;

  829.     int next_bt;

  830. 

  831.     memset(r->dmv, 0, sizeof(r->dmv));

  832.     for(i = 0; i < num_mvs[block_type]; i++){

  833.         r->dmv[i][0] = svq3_get_se_golomb(gb);

  834.         r->dmv[i][1] = svq3_get_se_golomb(gb);

  835.     }

  836.     switch(block_type){

  837.     case RV34_MB_TYPE_INTRA:

  838.     case RV34_MB_TYPE_INTRA16x16:

  839.         ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);

  840.         return 0;

  841.     case RV34_MB_SKIP:

  842.         if(s->pict_type == AV_PICTURE_TYPE_P){

  843.             ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);

  844.             rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);

  845.             break;

  846.         }

  847.     case RV34_MB_B_DIRECT:

  848.         //surprisingly, it uses motion scheme from next reference frame

  849.         /* wait for the current mb row to be finished */

  850.         if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))

  851.             ff_thread_await_progress(&s->next_picture_ptr->f, FFMAX(0, s->mb_y-1), 0);

  852. 

  853.         next_bt = s->next_picture_ptr->f.mb_type[s->mb_x + s->mb_y * s->mb_stride];

  854.         if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){

  855.             ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);

  856.             ZERO8x2(s->current_picture_ptr->f.motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);

  857.         }else

  858.             for(j = 0; j < 2; j++)

  859.                 for(i = 0; i < 2; i++)

  860.                     for(k = 0; k < 2; k++)

  861.                         for(l = 0; l < 2; l++)

  862.                             s->current_picture_ptr->f.motion_val[l][mv_pos + i + j*s->b8_stride][k] = calc_add_mv(r, l, s->next_picture_ptr->f.motion_val[0][mv_pos + i + j*s->b8_stride][k]);

  863.         if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC

  864.             rv34_mc_2mv(r, block_type);

  865.         else

  866.             rv34_mc_2mv_skip(r);

  867.         ZERO8x2(s->current_picture_ptr->f.motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);

  868.         break;

  869.     case RV34_MB_P_16x16:

  870.     case RV34_MB_P_MIX16x16:

  871.         rv34_pred_mv(r, block_type, 0, 0);

  872.         rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);

  873.         break;

  874.     case RV34_MB_B_FORWARD:

  875.     case RV34_MB_B_BACKWARD:

  876.         r->dmv[1][0] = r->dmv[0][0];

  877.         r->dmv[1][1] = r->dmv[0][1];

  878.         if(r->rv30)

  879.             rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);

  880.         else

  881.             rv34_pred_mv_b  (r, block_type, block_type == RV34_MB_B_BACKWARD);

  882.         rv34_mc_1mv     (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);

  883.         break;

  884.     case RV34_MB_P_16x8:

  885.     case RV34_MB_P_8x16:

  886.         rv34_pred_mv(r, block_type, 0, 0);

  887.         rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);

  888.         if(block_type == RV34_MB_P_16x8){

  889.             rv34_mc_1mv(r, block_type, 0, 0, 0,            2, 1, 0);

  890.             rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);

  891.         }

  892.         if(block_type == RV34_MB_P_8x16){

  893.             rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);

  894.             rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);

  895.         }

  896.         break;

  897.     case RV34_MB_B_BIDIR:

  898.         rv34_pred_mv_b  (r, block_type, 0);

  899.         rv34_pred_mv_b  (r, block_type, 1);

  900.         rv34_mc_2mv     (r, block_type);

  901.         break;

  902.     case RV34_MB_P_8x8:

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

  904.             rv34_pred_mv(r, block_type, i, i);

  905.             rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);

  906.         }

  907.         break;

  908.     }

  909. 

  910.     return 0;

  911. }

  912. /** @} */ // mv group

  913. 

  914. /**

  915.  * @name Macroblock reconstruction functions

  916.  * @{

  917.  */

  918. /** mapping of RV30/40 intra prediction types to standard H.264 types */

  919. static const int ittrans[9] = {

  920.  DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,

  921.  VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,

  922. };

  923. 

  924. /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */

  925. static const int ittrans16[4] = {

  926.  DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,

  927. };

  928. 

  929. /**

  930.  * Perform 4x4 intra prediction.

  931.  */

  932. static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)

  933. {

  934.     uint8_t *prev = dst - stride + 4;

  935.     uint32_t topleft;

  936. 

  937.     if(!up && !left)

  938.         itype = DC_128_PRED;

  939.     else if(!up){

  940.         if(itype == VERT_PRED) itype = HOR_PRED;

  941.         if(itype == DC_PRED)   itype = LEFT_DC_PRED;

  942.     }else if(!left){

  943.         if(itype == HOR_PRED)  itype = VERT_PRED;

  944.         if(itype == DC_PRED)   itype = TOP_DC_PRED;

  945.         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;

  946.     }

  947.     if(!down){

  948.         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;

  949.         if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;

  950.         if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;

  951.     }

  952.     if(!right && up){

  953.         topleft = dst[-stride + 3] * 0x01010101u;

  954.         prev = (uint8_t*)&topleft;

  955.     }

  956.     r->h.pred4x4[itype](dst, prev, stride);

  957. }

  958. 

  959. /** add_pixels_clamped for 4x4 block */

  960. static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)

  961. {

  962.     int x, y;

  963.     for(y = 0; y < 4; y++)

  964.         for(x = 0; x < 4; x++)

  965.             dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);

  966. }

  967. 

  968. static inline int adjust_pred16(int itype, int up, int left)

  969. {

  970.     if(!up && !left)

  971.         itype = DC_128_PRED8x8;

  972.     else if(!up){

  973.         if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;

  974.         if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;

  975.         if(itype == DC_PRED8x8)   itype = LEFT_DC_PRED8x8;

  976.     }else if(!left){

  977.         if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;

  978.         if(itype == HOR_PRED8x8)  itype = VERT_PRED8x8;

  979.         if(itype == DC_PRED8x8)   itype = TOP_DC_PRED8x8;

  980.     }

  981.     return itype;

  982. }

  983. 

  984. static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)

  985. {

  986.     MpegEncContext *s = &r->s;

  987.     DSPContext *dsp = &s->dsp;

  988.     int i, j;

  989.     uint8_t *Y, *U, *V;

  990.     int itype;

  991.     int avail[6*8] = {0};

  992.     int idx;

  993. 

  994.     // Set neighbour information.

  995.     if(r->avail_cache[1])

  996.         avail[0] = 1;

  997.     if(r->avail_cache[2])

  998.         avail[1] = avail[2] = 1;

  999.     if(r->avail_cache[3])

  1000.         avail[3] = avail[4] = 1;

  1001.     if(r->avail_cache[4])

  1002.         avail[5] = 1;

  1003.     if(r->avail_cache[5])

  1004.         avail[8] = avail[16] = 1;

  1005.     if(r->avail_cache[9])

  1006.         avail[24] = avail[32] = 1;

  1007. 

  1008.     Y = s->dest[0];

  1009.     U = s->dest[1];

  1010.     V = s->dest[2];

  1011.     if(!is16){

  1012.         for(j = 0; j < 4; j++){

  1013.             idx = 9 + j*8;

  1014.             for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){

  1015.                 rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);

  1016.                 avail[idx] = 1;

  1017.                 if(cbp & 1)

  1018.                     rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);

  1019.             }

  1020.             Y += s->linesize * 4 - 4*4;

  1021.             intra_types += r->intra_types_stride;

  1022.         }

  1023.         intra_types -= r->intra_types_stride * 4;

  1024.         fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);

  1025.         for(j = 0; j < 2; j++){

  1026.             idx = 6 + j*4;

  1027.             for(i = 0; i < 2; i++, cbp >>= 1, idx++){

  1028.                 rv34_pred_4x4_block(r, U + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);

  1029.                 rv34_pred_4x4_block(r, V + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);

  1030.                 r->avail_cache[idx] = 1;

  1031.                 if(cbp & 0x01)

  1032.                     rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);

  1033.                 if(cbp & 0x10)

  1034.                     rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);

  1035.             }

  1036.         }

  1037.     }else{

  1038.         itype = ittrans16[intra_types[0]];

  1039.         itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1040.         r->h.pred16x16[itype](Y, s->linesize);

  1041.         dsp->add_pixels_clamped(s->block[0], Y,     s->linesize);

  1042.         dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);

  1043.         Y += s->linesize * 8;

  1044.         dsp->add_pixels_clamped(s->block[2], Y,     s->linesize);

  1045.         dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);

  1046. 

  1047.         itype = ittrans16[intra_types[0]];

  1048.         if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;

  1049.         itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1050.         r->h.pred8x8[itype](U, s->uvlinesize);

  1051.         dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);

  1052.         r->h.pred8x8[itype](V, s->uvlinesize);

  1053.         dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);

  1054.     }

  1055. }

  1056. 

  1057. /**

  1058.  * mask for retrieving all bits in coded block pattern

  1059.  * corresponding to one 8x8 block

  1060.  */

  1061. #define LUMA_CBP_BLOCK_MASK 0x33

  1062. 

  1063. #define U_CBP_MASK 0x0F0000

  1064. #define V_CBP_MASK 0xF00000

  1065. 

  1066. /** @} */ // recons group

  1067. 

  1068. 

  1069. static void rv34_apply_differences(RV34DecContext *r, int cbp)

  1070. {

  1071.     static const int shifts[4] = { 0, 2, 8, 10 };

  1072.     MpegEncContext *s = &r->s;

  1073.     int i;

  1074. 

  1075.     for(i = 0; i < 4; i++)

  1076.         if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)

  1077.             s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);

  1078.     if(cbp & U_CBP_MASK)

  1079.         s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);

  1080.     if(cbp & V_CBP_MASK)

  1081.         s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);

  1082. }

  1083. 

  1084. static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)

  1085. {

  1086.     int d;

  1087.     d = motion_val[0][0] - motion_val[-step][0];

  1088.     if(d < -3 || d > 3)

  1089.         return 1;

  1090.     d = motion_val[0][1] - motion_val[-step][1];

  1091.     if(d < -3 || d > 3)

  1092.         return 1;

  1093.     return 0;

  1094. }

  1095. 

  1096. static int rv34_set_deblock_coef(RV34DecContext *r)

  1097. {

  1098.     MpegEncContext *s = &r->s;

  1099.     int hmvmask = 0, vmvmask = 0, i, j;

  1100.     int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;

  1101.     int16_t (*motion_val)[2] = &s->current_picture_ptr->f.motion_val[0][midx];

  1102.     for(j = 0; j < 16; j += 8){

  1103.         for(i = 0; i < 2; i++){

  1104.             if(is_mv_diff_gt_3(motion_val + i, 1))

  1105.                 vmvmask |= 0x11 << (j + i*2);

  1106.             if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))

  1107.                 hmvmask |= 0x03 << (j + i*2);

  1108.         }

  1109.         motion_val += s->b8_stride;

  1110.     }

  1111.     if(s->first_slice_line)

  1112.         hmvmask &= ~0x000F;

  1113.     if(!s->mb_x)

  1114.         vmvmask &= ~0x1111;

  1115.     if(r->rv30){ //RV30 marks both subblocks on the edge for filtering

  1116.         vmvmask |= (vmvmask & 0x4444) >> 1;

  1117.         hmvmask |= (hmvmask & 0x0F00) >> 4;

  1118.         if(s->mb_x)

  1119.             r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;

  1120.         if(!s->first_slice_line)

  1121.             r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;

  1122.     }

  1123.     return hmvmask | vmvmask;

  1124. }

  1125. 

  1126. static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)

  1127. {

  1128.     MpegEncContext *s = &r->s;

  1129.     GetBitContext *gb = &s->gb;

  1130.     int cbp, cbp2;

  1131.     int q_dc, q_ac, has_ac;

  1132.     int i, blknum, blkoff;

  1133.     LOCAL_ALIGNED_16(DCTELEM, block16, [64]);

  1134.     int luma_dc_quant;

  1135.     int dist;

  1136.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;

  1137. 

  1138.     // Calculate which neighbours are available. Maybe it's worth optimizing too.

  1139.     memset(r->avail_cache, 0, sizeof(r->avail_cache));

  1140.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);

  1141.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;

  1142.     if(s->mb_x && dist)

  1143.         r->avail_cache[5] =

  1144.         r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];

  1145.     if(dist >= s->mb_width)

  1146.         r->avail_cache[2] =

  1147.         r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];

  1148.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)

  1149.         r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];

  1150.     if(s->mb_x && dist > s->mb_width)

  1151.         r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];

  1152. 

  1153.     s->qscale = r->si.quant;

  1154.     cbp = cbp2 = rv34_decode_mb_header(r, intra_types);

  1155.     r->cbp_luma  [mb_pos] = cbp;

  1156.     r->cbp_chroma[mb_pos] = cbp >> 16;

  1157.     if(s->pict_type == AV_PICTURE_TYPE_I)

  1158.         r->deblock_coefs[mb_pos] = 0xFFFF;

  1159.     else

  1160.         r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];

  1161.     s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;

  1162. 

  1163.     if(cbp == -1)

  1164.         return -1;

  1165. 

  1166.     luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];

  1167.     if(r->is16){

  1168.         q_dc = rv34_qscale_tab[luma_dc_quant];

  1169.         q_ac = rv34_qscale_tab[s->qscale];

  1170.         s->dsp.clear_block(block16);

  1171.         if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))

  1172.             r->rdsp.rv34_inv_transform_tab[1](block16);

  1173.         else

  1174.             r->rdsp.rv34_inv_transform_dc_tab[1](block16);

  1175.     }

  1176. 

  1177.     q_ac = rv34_qscale_tab[s->qscale];

  1178.     for(i = 0; i < 16; i++, cbp >>= 1){

  1179.         DCTELEM *ptr;

  1180.         if(!r->is16 && !(cbp & 1)) continue;

  1181.         blknum = ((i & 2) >> 1) + ((i & 8) >> 2);

  1182.         blkoff = ((i & 1) << 2) + ((i & 4) << 3);

  1183.         ptr    = s->block[blknum] + blkoff;

  1184.         if(cbp & 1)

  1185.             has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);

  1186.         else

  1187.             has_ac = 0;

  1188.         if(r->is16) //FIXME: optimize

  1189.             ptr[0] = block16[(i & 3) | ((i & 0xC) << 1)];

  1190.         if(has_ac)

  1191.             r->rdsp.rv34_inv_transform_tab[0](ptr);

  1192.         else

  1193.             r->rdsp.rv34_inv_transform_dc_tab[0](ptr);

  1194.     }

  1195.     if(r->block_type == RV34_MB_P_MIX16x16)

  1196.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);

  1197.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];

  1198.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];

  1199.     for(; i < 24; i++, cbp >>= 1){

  1200.         DCTELEM *ptr;

  1201.         if(!(cbp & 1)) continue;

  1202.         blknum = ((i & 4) >> 2) + 4;

  1203.         blkoff = ((i & 1) << 2) + ((i & 2) << 4);

  1204.         ptr    = s->block[blknum] + blkoff;

  1205.         if (rv34_decode_block(ptr, gb, r->cur_vlcs, r->chroma_vlc, 1, q_dc, q_ac, q_ac))

  1206.             r->rdsp.rv34_inv_transform_tab[0](ptr);

  1207.         else

  1208.             r->rdsp.rv34_inv_transform_dc_tab[0](ptr);

  1209.     }

  1210.     if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos]))

  1211.         rv34_output_macroblock(r, intra_types, cbp2, r->is16);

  1212.     else

  1213.         rv34_apply_differences(r, cbp2);

  1214. 

  1215.     return 0;

  1216. }

  1217. 

  1218. static int check_slice_end(RV34DecContext *r, MpegEncContext *s)

  1219. {

  1220.     int bits;

  1221.     if(s->mb_y >= s->mb_height)

  1222.         return 1;

  1223.     if(!s->mb_num_left)

  1224.         return 1;

  1225.     if(r->s.mb_skip_run > 1)

  1226.         return 0;

  1227.     bits = get_bits_left(&s->gb);

  1228.     if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))

  1229.         return 1;

  1230.     return 0;

  1231. }

  1232. 

  1233. static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)

  1234. {

  1235.     MpegEncContext *s = &r->s;

  1236.     GetBitContext *gb = &s->gb;

  1237.     int mb_pos;

  1238.     int res;

  1239. 

  1240.     init_get_bits(&r->s.gb, buf, buf_size*8);

  1241.     res = r->parse_slice_header(r, gb, &r->si);

  1242.     if(res < 0){

  1243.         av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");

  1244.         return -1;

  1245.     }

  1246. 

  1247.     if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {

  1248.         if(s->width != r->si.width || s->height != r->si.height){

  1249.             av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);

  1250.             MPV_common_end(s);

  1251.             s->width  = r->si.width;

  1252.             s->height = r->si.height;

  1253.             avcodec_set_dimensions(s->avctx, s->width, s->height);

  1254.             if(MPV_common_init(s) < 0)

  1255.                 return -1;

  1256.             r->intra_types_stride = s->mb_width*4 + 4;

  1257.             r->intra_types_hist = av_realloc(r->intra_types_hist, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));

  1258.             r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;

  1259.             r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));

  1260.             r->cbp_luma   = av_realloc(r->cbp_luma,   r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));

  1261.             r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));

  1262.             r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));

  1263.             av_freep(&r->tmp_b_block_base);

  1264.         }

  1265.         s->pict_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;

  1266.         if(MPV_frame_start(s, s->avctx) < 0)

  1267.             return -1;

  1268.         ff_er_frame_start(s);

  1269.         if (!r->tmp_b_block_base) {

  1270.             int i;

  1271. 

  1272.             r->tmp_b_block_base = av_malloc(s->linesize * 48);

  1273.             for (i = 0; i < 2; i++)

  1274.                 r->tmp_b_block_y[i] = r->tmp_b_block_base + i * 16 * s->linesize;

  1275.             for (i = 0; i < 4; i++)

  1276.                 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize

  1277.                                        + (i >> 1) * 8 * s->uvlinesize + (i & 1) * 16;

  1278.         }

  1279.         r->cur_pts = r->si.pts;

  1280.         if(s->pict_type != AV_PICTURE_TYPE_B){

  1281.             r->last_pts = r->next_pts;

  1282.             r->next_pts = r->cur_pts;

  1283.         }else{

  1284.             int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);

  1285.             int dist0   = GET_PTS_DIFF(r->cur_pts,  r->last_pts);

  1286.             int dist1   = GET_PTS_DIFF(r->next_pts, r->cur_pts);

  1287. 

  1288.             if(!refdist){

  1289.                 r->weight1 = r->weight2 = 8192;

  1290.             }else{

  1291.                 r->weight1 = (dist0 << 14) / refdist;

  1292.                 r->weight2 = (dist1 << 14) / refdist;

  1293.             }

  1294.         }

  1295.         s->mb_x = s->mb_y = 0;

  1296.         ff_thread_finish_setup(s->avctx);

  1297.     } else {

  1298.         int slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;

  1299. 

  1300.         if (slice_type != s->pict_type) {

  1301.             av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");

  1302.             return AVERROR_INVALIDDATA;

  1303.         }

  1304.     }

  1305. 

  1306.     r->si.end = end;

  1307.     s->qscale = r->si.quant;

  1308.     s->mb_num_left = r->si.end - r->si.start;

  1309.     r->s.mb_skip_run = 0;

  1310. 

  1311.     mb_pos = s->mb_x + s->mb_y * s->mb_width;

  1312.     if(r->si.start != mb_pos){

  1313.         av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);

  1314.         s->mb_x = r->si.start % s->mb_width;

  1315.         s->mb_y = r->si.start / s->mb_width;

  1316.     }

  1317.     memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));

  1318.     s->first_slice_line = 1;

  1319.     s->resync_mb_x = s->mb_x;

  1320.     s->resync_mb_y = s->mb_y;

  1321. 

  1322.     ff_init_block_index(s);

  1323.     while(!check_slice_end(r, s)) {

  1324.         ff_update_block_index(s);

  1325.         s->dsp.clear_blocks(s->block[0]);

  1326. 

  1327.         if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 4) < 0){

  1328.             ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);

  1329.             return -1;

  1330.         }

  1331.         if (++s->mb_x == s->mb_width) {

  1332.             s->mb_x = 0;

  1333.             s->mb_y++;

  1334.             ff_init_block_index(s);

  1335. 

  1336.             memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1337.             memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1338. 

  1339.             if(r->loop_filter && s->mb_y >= 2)

  1340.                 r->loop_filter(r, s->mb_y - 2);

  1341. 

  1342.             if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))

  1343.                 ff_thread_report_progress(&s->current_picture_ptr->f,

  1344.                                           s->mb_y - 2, 0);

  1345. 

  1346.         }

  1347.         if(s->mb_x == s->resync_mb_x)

  1348.             s->first_slice_line=0;

  1349.         s->mb_num_left--;

  1350.     }

  1351.     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);

  1352. 

  1353.     return s->mb_y == s->mb_height;

  1354. }

  1355. 

  1356. /** @} */ // recons group end

  1357. 

  1358. /**

  1359.  * Initialize decoder.

  1360.  */

  1361. av_cold int ff_rv34_decode_init(AVCodecContext *avctx)

  1362. {

  1363.     RV34DecContext *r = avctx->priv_data;

  1364.     MpegEncContext *s = &r->s;

  1365. 

  1366.     MPV_decode_defaults(s);

  1367.     s->avctx      = avctx;

  1368.     s->out_format = FMT_H263;

  1369.     s->codec_id   = avctx->codec_id;

  1370. 

  1371.     s->width  = avctx->width;

  1372.     s->height = avctx->height;

  1373. 

  1374.     r->s.avctx = avctx;

  1375.     avctx->flags |= CODEC_FLAG_EMU_EDGE;

  1376.     r->s.flags |= CODEC_FLAG_EMU_EDGE;

  1377.     avctx->pix_fmt = PIX_FMT_YUV420P;

  1378.     avctx->has_b_frames = 1;

  1379.     s->low_delay = 0;

  1380. 

  1381.     if (MPV_common_init(s) < 0)

  1382.         return -1;

  1383. 

  1384.     ff_h264_pred_init(&r->h, CODEC_ID_RV40, 8, 1);

  1385. 

  1386. #if CONFIG_RV30_DECODER

  1387.     if (avctx->codec_id == CODEC_ID_RV30)

  1388.         ff_rv30dsp_init(&r->rdsp, &r->s.dsp);

  1389. #endif

  1390. #if CONFIG_RV40_DECODER

  1391.     if (avctx->codec_id == CODEC_ID_RV40)

  1392.         ff_rv40dsp_init(&r->rdsp, &r->s.dsp);

  1393. #endif

  1394. 

  1395.     r->intra_types_stride = 4*s->mb_stride + 4;

  1396.     r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));

  1397.     r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;

  1398. 

  1399.     r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));

  1400. 

  1401.     r->cbp_luma   = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));

  1402.     r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));

  1403.     r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));

  1404. 

  1405.     if(!intra_vlcs[0].cbppattern[0].bits)

  1406.         rv34_init_tables();

  1407. 

  1408.     return 0;

  1409. }

  1410. 

  1411. int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx)

  1412. {

  1413.     RV34DecContext *r = avctx->priv_data;

  1414. 

  1415.     r->s.avctx = avctx;

  1416. 

  1417.     if (avctx->internal->is_copy) {

  1418.         r->cbp_chroma       = av_malloc(r->s.mb_stride * r->s.mb_height *

  1419.                                         sizeof(*r->cbp_chroma));

  1420.         r->cbp_luma         = av_malloc(r->s.mb_stride * r->s.mb_height *

  1421.                                         sizeof(*r->cbp_luma));

  1422.         r->deblock_coefs    = av_malloc(r->s.mb_stride * r->s.mb_height *

  1423.                                         sizeof(*r->deblock_coefs));

  1424.         r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *

  1425.                                         sizeof(*r->intra_types_hist));

  1426.         r->mb_type          = av_malloc(r->s.mb_stride * r->s.mb_height *

  1427.                                         sizeof(*r->mb_type));

  1428. 

  1429.         if (!(r->cbp_chroma       && r->cbp_luma && r->deblock_coefs &&

  1430.               r->intra_types_hist && r->mb_type)) {

  1431.             av_freep(&r->cbp_chroma);

  1432.             av_freep(&r->cbp_luma);

  1433.             av_freep(&r->deblock_coefs);

  1434.             av_freep(&r->intra_types_hist);

  1435.             av_freep(&r->mb_type);

  1436.             r->intra_types = NULL;

  1437.             return AVERROR(ENOMEM);

  1438.         }

  1439. 

  1440.         r->intra_types      = r->intra_types_hist + r->intra_types_stride * 4;

  1441.         r->tmp_b_block_base = NULL;

  1442. 

  1443.         memset(r->mb_type, 0,  r->s.mb_stride * r->s.mb_height *

  1444.                sizeof(*r->mb_type));

  1445. 

  1446.         MPV_common_init(&r->s);

  1447.     }

  1448.     return 0;

  1449. }

  1450. 

  1451. int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)

  1452. {

  1453.     RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;

  1454.     MpegEncContext * const s = &r->s, * const s1 = &r1->s;

  1455.     int err;

  1456. 

  1457.     if (dst == src || !s1->context_initialized)

  1458.         return 0;

  1459. 

  1460.     if ((err = ff_mpeg_update_thread_context(dst, src)))

  1461.         return err;

  1462. 

  1463.     r->cur_pts  = r1->cur_pts;

  1464.     r->last_pts = r1->last_pts;

  1465.     r->next_pts = r1->next_pts;

  1466. 

  1467.     memset(&r->si, 0, sizeof(r->si));

  1468. 

  1469.     /* necessary since it is it the condition checked for in decode_slice

  1470.      * to call MPV_frame_start. cmp. comment at the end of decode_frame */

  1471.     s->current_picture_ptr = NULL;

  1472. 

  1473.     return 0;

  1474. }

  1475. 

  1476. static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)

  1477. {

  1478.     if(avctx->slice_count) return avctx->slice_offset[n];

  1479.     else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);

  1480. }

  1481. 

  1482. int ff_rv34_decode_frame(AVCodecContext *avctx,

  1483.                             void *data, int *data_size,

  1484.                             AVPacket *avpkt)

  1485. {

  1486.     const uint8_t *buf = avpkt->data;

  1487.     int buf_size = avpkt->size;

  1488.     RV34DecContext *r = avctx->priv_data;

  1489.     MpegEncContext *s = &r->s;

  1490.     AVFrame *pict = data;

  1491.     SliceInfo si;

  1492.     int i;

  1493.     int slice_count;

  1494.     const uint8_t *slices_hdr = NULL;

  1495.     int last = 0;

  1496. 

  1497.     /* no supplementary picture */

  1498.     if (buf_size == 0) {

  1499.         /* special case for last picture */

  1500.         if (s->low_delay==0 && s->next_picture_ptr) {

  1501.             *pict = *(AVFrame*)s->next_picture_ptr;

  1502.             s->next_picture_ptr = NULL;

  1503. 

  1504.             *data_size = sizeof(AVFrame);

  1505.         }

  1506.         return 0;

  1507.     }

  1508. 

  1509.     if(!avctx->slice_count){

  1510.         slice_count = (*buf++) + 1;

  1511.         slices_hdr = buf + 4;

  1512.         buf += 8 * slice_count;

  1513.         buf_size -= 1 + 8 * slice_count;

  1514.     }else

  1515.         slice_count = avctx->slice_count;

  1516. 

  1517.     //parse first slice header to check whether this frame can be decoded

  1518.     if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||

  1519.        get_slice_offset(avctx, slices_hdr, 0) > buf_size){

  1520.         av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1521.         return -1;

  1522.     }

  1523.     init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);

  1524.     if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){

  1525.         av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");

  1526.         return -1;

  1527.     }

  1528.     if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) && si.type == AV_PICTURE_TYPE_B)

  1529.         return -1;

  1530.     if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)

  1531.        || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)

  1532.        ||  avctx->skip_frame >= AVDISCARD_ALL)

  1533.         return avpkt->size;

  1534. 

  1535.     for(i = 0; i < slice_count; i++){

  1536.         int offset = get_slice_offset(avctx, slices_hdr, i);

  1537.         int size;

  1538.         if(i+1 == slice_count)

  1539.             size = buf_size - offset;

  1540.         else

  1541.             size = get_slice_offset(avctx, slices_hdr, i+1) - offset;

  1542. 

  1543.         if(offset < 0 || offset > buf_size){

  1544.             av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1545.             break;

  1546.         }

  1547. 

  1548.         r->si.end = s->mb_width * s->mb_height;

  1549.         if(i+1 < slice_count){

  1550.             if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||

  1551.                 get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {

  1552.                 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1553.                 break;

  1554.             }

  1555.             init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, i+1), (buf_size-get_slice_offset(avctx, slices_hdr, i+1))*8);

  1556.             if(r->parse_slice_header(r, &r->s.gb, &si) < 0){

  1557.                 if(i+2 < slice_count)

  1558.                     size = get_slice_offset(avctx, slices_hdr, i+2) - offset;

  1559.                 else

  1560.                     size = buf_size - offset;

  1561.             }else

  1562.                 r->si.end = si.start;

  1563.         }

  1564.         if (size < 0 || size > buf_size - offset) {

  1565.             av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");

  1566.             break;

  1567.         }

  1568.         last = rv34_decode_slice(r, r->si.end, buf + offset, size);

  1569.         s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;

  1570.         if(last)

  1571.             break;

  1572.     }

  1573. 

  1574.     if(last && s->current_picture_ptr){

  1575.         if(r->loop_filter)

  1576.             r->loop_filter(r, s->mb_height - 1);

  1577.         if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))

  1578.             ff_thread_report_progress(&s->current_picture_ptr->f,

  1579.                                       s->mb_height - 1, 0);

  1580.         ff_er_frame_end(s);

  1581.         MPV_frame_end(s);

  1582.         if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {

  1583.             *pict = *(AVFrame*)s->current_picture_ptr;

  1584.         } else if (s->last_picture_ptr != NULL) {

  1585.             *pict = *(AVFrame*)s->last_picture_ptr;

  1586.         }

  1587. 

  1588.         if(s->last_picture_ptr || s->low_delay){

  1589.             *data_size = sizeof(AVFrame);

  1590.             ff_print_debug_info(s, pict);

  1591.         }

  1592.         s->current_picture_ptr = NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)

  1593.     }

  1594.     return avpkt->size;

  1595. }

  1596. 

  1597. av_cold int ff_rv34_decode_end(AVCodecContext *avctx)

  1598. {

  1599.     RV34DecContext *r = avctx->priv_data;

  1600. 

  1601.     MPV_common_end(&r->s);

  1602. 

  1603.     av_freep(&r->intra_types_hist);

  1604.     r->intra_types = NULL;

  1605.     av_freep(&r->tmp_b_block_base);

  1606.     av_freep(&r->mb_type);

  1607.     av_freep(&r->cbp_luma);

  1608.     av_freep(&r->cbp_chroma);

  1609.     av_freep(&r->deblock_coefs);

  1610. 

  1611.     return 0;

  1612. }