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.
39205 Commits
32 Branches
1.1GB
Tree: f8c507f44b
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'f8c507f44b'
${ noResults }
FFmpeg/libavcodec/rv34.c

1809 lines
63KB
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 "error_resilience.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. static inline void ZERO8x2(void* dst, int stride)

  43. {

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

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

  46. }

  47. 

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

  49. static const int rv34_mb_type_to_lavc[12] = {

  50.     MB_TYPE_INTRA,

  51.     MB_TYPE_INTRA16x16              | MB_TYPE_SEPARATE_DC,

  52.     MB_TYPE_16x16   | MB_TYPE_L0,

  53.     MB_TYPE_8x8     | MB_TYPE_L0,

  54.     MB_TYPE_16x16   | MB_TYPE_L0,

  55.     MB_TYPE_16x16   | MB_TYPE_L1,

  56.     MB_TYPE_SKIP,

  57.     MB_TYPE_DIRECT2 | MB_TYPE_16x16,

  58.     MB_TYPE_16x8    | MB_TYPE_L0,

  59.     MB_TYPE_8x16    | MB_TYPE_L0,

  60.     MB_TYPE_16x16   | MB_TYPE_L0L1,

  61.     MB_TYPE_16x16   | MB_TYPE_L0    | MB_TYPE_SEPARATE_DC

  62. };

  63. 

  64. 

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

  66. 

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

  68. 

  69. /**

  70.  * @name RV30/40 VLC generating functions

  71.  * @{

  72.  */

  73. 

  74. static const int table_offs[] = {

  75.       0,   1818,   3622,   4144,   4698,   5234,   5804,   5868,   5900,   5932,

  76.    5996,   6252,   6316,   6348,   6380,   7674,   8944,  10274,  11668,  12250,

  77.   14060,  15846,  16372,  16962,  17512,  18148,  18180,  18212,  18244,  18308,

  78.   18564,  18628,  18660,  18692,  20036,  21314,  22648,  23968,  24614,  26384,

  79.   28190,  28736,  29366,  29938,  30608,  30640,  30672,  30704,  30768,  31024,

  80.   31088,  31120,  31184,  32570,  33898,  35236,  36644,  37286,  39020,  40802,

  81.   41368,  42052,  42692,  43348,  43380,  43412,  43444,  43476,  43604,  43668,

  82.   43700,  43732,  45100,  46430,  47778,  49160,  49802,  51550,  53340,  53972,

  83.   54648,  55348,  55994,  56122,  56154,  56186,  56218,  56346,  56410,  56442,

  84.   56474,  57878,  59290,  60636,  62036,  62682,  64460,  64524,  64588,  64716,

  85.   64844,  66076,  67466,  67978,  68542,  69064,  69648,  70296,  72010,  72074,

  86.   72138,  72202,  72330,  73572,  74936,  75454,  76030,  76566,  77176,  77822,

  87.   79582,  79646,  79678,  79742,  79870,  81180,  82536,  83064,  83672,  84242,

  88.   84934,  85576,  87384,  87448,  87480,  87544,  87672,  88982,  90340,  90902,

  89.   91598,  92182,  92846,  93488,  95246,  95278,  95310,  95374,  95502,  96878,

  90.   98266,  98848,  99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,

  91.  103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,

  92.  111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592

  93. };

  94. 

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

  96. 

  97. /**

  98.  * Generate VLC from codeword lengths.

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

  100.  * @param size   length of input data

  101.  * @param vlc    output VLC

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

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

  104.  */

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

  106.                          const int num)

  107. {

  108.     int i;

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

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

  111.     uint8_t bits2[MAX_VLC_SIZE];

  112.     int maxbits = 0, realsize = 0;

  113. 

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

  115.         if(bits[i]){

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

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

  118.             realsize++;

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

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

  121.         }

  122.     }

  123. 

  124.     codes[0] = 0;

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

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

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

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

  129. 

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

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

  132.     ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,

  133.                        bits2, 1, 1,

  134.                        cw,    2, 2,

  135.                        syms,  2, 2, INIT_VLC_USE_NEW_STATIC);

  136. }

  137. 

  138. /**

  139.  * Initialize all tables.

  140.  */

  141. static av_cold void rv34_init_tables(void)

  142. {

  143.     int i, j, k;

  144. 

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

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

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

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

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

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

  151.                 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);

  152.             }

  153.         }

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

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

  156.         }

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

  158.     }

  159. 

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

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

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

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

  164.         }

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

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

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

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

  169.         }

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

  171.     }

  172. }

  173. 

  174. /** @} */ // vlc group

  175. 

  176. /**

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

  178.  * @{

  179.  */

  180. 

  181. /**

  182.  * Decode coded block pattern.

  183.  */

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

  185. {

  186.     int pattern, code, cbp=0;

  187.     int ones;

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

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

  190.     const int *curshift = shifts;

  191.     int i, t, mask;

  192. 

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

  194.     pattern = code & 0xF;

  195.     code >>= 4;

  196. 

  197.     ones = rv34_count_ones[pattern];

  198. 

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

  200.         if(pattern & mask)

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

  202.     }

  203. 

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

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

  206.         if(t == 1)

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

  208.         if(t == 2)

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

  210.     }

  211.     return cbp;

  212. }

  213. 

  214. /**

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

  216.  */

  217. static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)

  218. {

  219.     if(coef){

  220.         if(coef == esc){

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

  222.             if(coef > 23){

  223.                 coef -= 23;

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

  225.             }

  226.             coef += esc;

  227.         }

  228.         if(get_bits1(gb))

  229.             coef = -coef;

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

  231.     }

  232. }

  233. 

  234. /**

  235.  * Decode 2x2 subblock of coefficients.

  236.  */

  237. static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)

  238. {

  239.     int flags = modulo_three_table[code];

  240. 

  241.     decode_coeff(    dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q);

  242.     if(is_block2){

  243.         decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);

  244.         decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);

  245.     }else{

  246.         decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);

  247.         decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);

  248.     }

  249.     decode_coeff(    dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);

  250. }

  251. 

  252. /**

  253.  * Decode a single coefficient.

  254.  */

  255. static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)

  256. {

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

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

  259. }

  260. 

  261. static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,

  262.                                     int q_dc, int q_ac1, int q_ac2)

  263. {

  264.     int flags = modulo_three_table[code];

  265. 

  266.     decode_coeff(dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q_dc);

  267.     decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);

  268.     decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);

  269.     decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);

  270. }

  271. 

  272. /**

  273.  * Decode coefficients for 4x4 block.

  274.  *

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

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

  277.  *  o--o

  278.  *    /

  279.  *   /

  280.  *  o--o

  281.  */

  282. 

  283. static int rv34_decode_block(int16_t *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc, int q_dc, int q_ac1, int q_ac2)

  284. {

  285.     int code, pattern, has_ac = 1;

  286. 

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

  288. 

  289.     pattern = code & 0x7;

  290. 

  291.     code >>= 3;

  292. 

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

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

  295.     } else {

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

  297.         if (!pattern)

  298.             return 0;

  299.         has_ac = 0;

  300.     }

  301. 

  302.     if(pattern & 4){

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

  304.         decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);

  305.     }

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

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

  308.         decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);

  309.     }

  310.     if(pattern & 1){

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

  312.         decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);

  313.     }

  314.     return has_ac | pattern;

  315. }

  316. 

  317. /**

  318.  * @name RV30/40 bitstream parsing

  319.  * @{

  320.  */

  321. 

  322. /**

  323.  * Decode starting slice position.

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

  325.  */

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

  327. {

  328.     int i;

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

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

  331.             break;

  332.     return rv34_mb_bits_sizes[i];

  333. }

  334. 

  335. /**

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

  337.  */

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

  339. {

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

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

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

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

  344. }

  345. 

  346. /**

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

  348.  */

  349. static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)

  350. {

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

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

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

  354.     int t;

  355. 

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

  357.     if(r->is16){

  358.         s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA16x16;

  359.         r->block_type = RV34_MB_TYPE_INTRA16x16;

  360.         t = get_bits(gb, 2);

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

  362.         r->luma_vlc   = 2;

  363.     }else{

  364.         if(!r->rv30){

  365.             if(!get_bits1(gb))

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

  367.         }

  368.         s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA;

  369.         r->block_type = RV34_MB_TYPE_INTRA;

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

  371.             return -1;

  372.         r->luma_vlc   = 1;

  373.     }

  374. 

  375.     r->chroma_vlc = 0;

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

  377. 

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

  379. }

  380. 

  381. /**

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

  383.  */

  384. static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)

  385. {

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

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

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

  389.     int i, t;

  390. 

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

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

  393.         return -1;

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

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

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

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

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

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

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

  401.     }

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

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

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

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

  406.         return 0;

  407.     }

  408.     r->chroma_vlc = 1;

  409.     r->luma_vlc   = 0;

  410. 

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

  412.         if(r->is16){

  413.             t = get_bits(gb, 2);

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

  415.             r->luma_vlc   = 2;

  416.         }else{

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

  418.                 return -1;

  419.             r->luma_vlc   = 1;

  420.         }

  421.         r->chroma_vlc = 0;

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

  423.     }else{

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

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

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

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

  428.             r->is16 = 1;

  429.             r->chroma_vlc = 1;

  430.             r->luma_vlc   = 2;

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

  432.         }

  433.     }

  434. 

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

  436. }

  437. 

  438. /** @} */ //bitstream functions

  439. 

  440. /**

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

  442.  * @{

  443.  */

  444. 

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

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

  447. 

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

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

  450. 

  451. /** availability index for subblocks */

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

  453. 

  454. /**

  455.  * motion vector prediction

  456.  *

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

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

  459.  * some other vectors may be used instead.

  460.  */

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

  462. {

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

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

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

  466.     int i, j;

  467.     int mx, my;

  468.     int* avail = r->avail_cache + avail_indexes[subblock_no];

  469.     int c_off = part_sizes_w[block_type];

  470. 

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

  472.     if(subblock_no == 3)

  473.         c_off = -1;

  474. 

  475.     if(avail[-1]){

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

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

  478.     }

  479.     if(avail[-4]){

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

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

  482.     }else{

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

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

  485.     }

  486.     if(!avail[c_off-4]){

  487.         if(avail[-4] && (avail[-1] || r->rv30)){

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

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

  490.         }else{

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

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

  493.         }

  494.     }else{

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

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

  497.     }

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

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

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

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

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

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

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

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

  506.         }

  507.     }

  508. }

  509. 

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

  511. 

  512. /**

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

  514.  */

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

  516. {

  517.     int mul = dir ? -r->mv_weight2 : r->mv_weight1;

  518. 

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

  520. }

  521. 

  522. /**

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

  524.  */

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

  526.                                       int A_avail, int B_avail, int C_avail,

  527.                                       int *mx, int *my)

  528. {

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

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

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

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

  533.             *mx /= 2;

  534.             *my /= 2;

  535.         }

  536.     }else{

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

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

  539.     }

  540. }

  541. 

  542. /**

  543.  * motion vector prediction for B-frames

  544.  */

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

  546. {

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

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

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

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

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

  552.     int mx, my;

  553.     int i, j;

  554.     Picture *cur_pic = s->current_picture_ptr;

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

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

  557. 

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

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

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

  561.         has_A = 1;

  562.     }

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

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

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

  566.         has_B = 1;

  567.     }

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

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

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

  571.         has_C = 1;

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

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

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

  575.         has_C = 1;

  576.     }

  577. 

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

  579. 

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

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

  582. 

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

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

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

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

  587.         }

  588.     }

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

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

  591.     }

  592. }

  593. 

  594. /**

  595.  * motion vector prediction - RV3 version

  596.  */

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

  598. {

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

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

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

  602.     int i, j, k;

  603.     int mx, my;

  604.     int* avail = r->avail_cache + avail_indexes[0];

  605. 

  606.     if(avail[-1]){

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

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

  609.     }

  610.     if(avail[-4]){

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

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

  613.     }else{

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

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

  616.     }

  617.     if(!avail[-4 + 2]){

  618.         if(avail[-4] && (avail[-1])){

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

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

  621.         }else{

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

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

  624.         }

  625.     }else{

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

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

  628.     }

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

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

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

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

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

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

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

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

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

  638.             }

  639.         }

  640.     }

  641. }

  642. 

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

  644. 

  645. /**

  646.  * generic motion compensation function

  647.  *

  648.  * @param r decoder context

  649.  * @param block_type type of the current block

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

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

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

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

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

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

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

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

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

  659.  */

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

  661.                           const int xoff, const int yoff, int mv_off,

  662.                           const int width, const int height, int dir,

  663.                           const int thirdpel, int weighted,

  664.                           qpel_mc_func (*qpel_mc)[16],

  665.                           h264_chroma_mc_func (*chroma_mc))

  666. {

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

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

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

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

  671.     int is16x16 = 1;

  672. 

  673.     if(thirdpel){

  674.         int chroma_mx, chroma_my;

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

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

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

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

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

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

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

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

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

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

  685.     }else{

  686.         int cx, cy;

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

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

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

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

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

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

  693.         umx = cx >> 2;

  694.         umy = cy >> 2;

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

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

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

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

  699.             uvmx = uvmy = 4;

  700.     }

  701. 

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

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

  704.         int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);

  705.         ThreadFrame *f = dir ? &s->next_picture_ptr->tf : &s->last_picture_ptr->tf;

  706.         ff_thread_await_progress(f, mb_row, 0);

  707.     }

  708. 

  709.     dxy = ly*4 + lx;

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  724. 

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

  726.         s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY,

  727.                                  s->linesize, s->linesize,

  728.                                  (width << 3) + 6, (height << 3) + 6,

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

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

  731.         s->vdsp.emulated_edge_mc(uvbuf, srcU,

  732.                                  s->uvlinesize,s->uvlinesize,

  733.                                  (width << 2) + 1, (height << 2) + 1,

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

  735.         s->vdsp.emulated_edge_mc(uvbuf + 16, srcV,

  736.                                  s->uvlinesize, s->uvlinesize,

  737.                                  (width << 2) + 1, (height << 2) + 1,

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

  739.         srcU = uvbuf;

  740.         srcV = uvbuf + 16;

  741.     }

  742.     if(!weighted){

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

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

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

  746.     }else{

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

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

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

  750.     }

  751. 

  752.     if(block_type == RV34_MB_P_16x8){

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

  754.         Y    += 8;

  755.         srcY += 8;

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

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

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

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

  760.     }

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

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

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

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

  765. }

  766. 

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

  768.                         const int xoff, const int yoff, int mv_off,

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

  770. {

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

  772.             r->rdsp.put_pixels_tab,

  773.             r->rdsp.put_chroma_pixels_tab);

  774. }

  775. 

  776. static void rv4_weight(RV34DecContext *r)

  777. {

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

  779.                                                         r->tmp_b_block_y[0],

  780.                                                         r->tmp_b_block_y[1],

  781.                                                         r->weight1,

  782.                                                         r->weight2,

  783.                                                         r->s.linesize);

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

  785.                                                         r->tmp_b_block_uv[0],

  786.                                                         r->tmp_b_block_uv[2],

  787.                                                         r->weight1,

  788.                                                         r->weight2,

  789.                                                         r->s.uvlinesize);

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

  791.                                                         r->tmp_b_block_uv[1],

  792.                                                         r->tmp_b_block_uv[3],

  793.                                                         r->weight1,

  794.                                                         r->weight2,

  795.                                                         r->s.uvlinesize);

  796. }

  797. 

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

  799. {

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

  801. 

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

  803.             r->rdsp.put_pixels_tab,

  804.             r->rdsp.put_chroma_pixels_tab);

  805.     if(!weighted){

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

  807.                 r->rdsp.avg_pixels_tab,

  808.                 r->rdsp.avg_chroma_pixels_tab);

  809.     }else{

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

  811.                 r->rdsp.put_pixels_tab,

  812.                 r->rdsp.put_chroma_pixels_tab);

  813.         rv4_weight(r);

  814.     }

  815. }

  816. 

  817. static void rv34_mc_2mv_skip(RV34DecContext *r)

  818. {

  819.     int i, j;

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

  821. 

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

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

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

  825.                      weighted,

  826.                      r->rdsp.put_pixels_tab,

  827.                      r->rdsp.put_chroma_pixels_tab);

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

  829.                      weighted,

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

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

  832.         }

  833.     if(weighted)

  834.         rv4_weight(r);

  835. }

  836. 

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

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

  839. 

  840. /**

  841.  * Decode motion vector differences

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

  843.  */

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

  845. {

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

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

  848.     int i, j, k, l;

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

  850.     int next_bt;

  851. 

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

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

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

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

  856.     }

  857.     switch(block_type){

  858.     case RV34_MB_TYPE_INTRA:

  859.     case RV34_MB_TYPE_INTRA16x16:

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

  861.         return 0;

  862.     case RV34_MB_SKIP:

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

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

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

  866.             break;

  867.         }

  868.     case RV34_MB_B_DIRECT:

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

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

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

  872.             ff_thread_await_progress(&s->next_picture_ptr->tf, FFMAX(0, s->mb_y-1), 0);

  873. 

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

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

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

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

  878.         }else

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

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

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

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

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

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

  885.             rv34_mc_2mv(r, block_type);

  886.         else

  887.             rv34_mc_2mv_skip(r);

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

  889.         break;

  890.     case RV34_MB_P_16x16:

  891.     case RV34_MB_P_MIX16x16:

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

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

  894.         break;

  895.     case RV34_MB_B_FORWARD:

  896.     case RV34_MB_B_BACKWARD:

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

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

  899.         if(r->rv30)

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

  901.         else

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

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

  904.         break;

  905.     case RV34_MB_P_16x8:

  906.     case RV34_MB_P_8x16:

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

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

  909.         if(block_type == RV34_MB_P_16x8){

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

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

  912.         }

  913.         if(block_type == RV34_MB_P_8x16){

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

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

  916.         }

  917.         break;

  918.     case RV34_MB_B_BIDIR:

  919.         rv34_pred_mv_b  (r, block_type, 0);

  920.         rv34_pred_mv_b  (r, block_type, 1);

  921.         rv34_mc_2mv     (r, block_type);

  922.         break;

  923.     case RV34_MB_P_8x8:

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

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

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

  927.         }

  928.         break;

  929.     }

  930. 

  931.     return 0;

  932. }

  933. /** @} */ // mv group

  934. 

  935. /**

  936.  * @name Macroblock reconstruction functions

  937.  * @{

  938.  */

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

  940. static const int ittrans[9] = {

  941.  DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,

  942.  VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,

  943. };

  944. 

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

  946. static const int ittrans16[4] = {

  947.  DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,

  948. };

  949. 

  950. /**

  951.  * Perform 4x4 intra prediction.

  952.  */

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

  954. {

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

  956.     uint32_t topleft;

  957. 

  958.     if(!up && !left)

  959.         itype = DC_128_PRED;

  960.     else if(!up){

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

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

  963.     }else if(!left){

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

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

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

  967.     }

  968.     if(!down){

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

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

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

  972.     }

  973.     if(!right && up){

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

  975.         prev = (uint8_t*)&topleft;

  976.     }

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

  978. }

  979. 

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

  981. {

  982.     if(!up && !left)

  983.         itype = DC_128_PRED8x8;

  984.     else if(!up){

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

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

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

  988.     }else if(!left){

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

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

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

  992.     }

  993.     return itype;

  994. }

  995. 

  996. static inline void rv34_process_block(RV34DecContext *r,

  997.                                       uint8_t *pdst, int stride,

  998.                                       int fc, int sc, int q_dc, int q_ac)

  999. {

  1000.     MpegEncContext *s = &r->s;

  1001.     int16_t *ptr = s->block[0];

  1002.     int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,

  1003.                                    fc, sc, q_dc, q_ac, q_ac);

  1004.     if(has_ac){

  1005.         r->rdsp.rv34_idct_add(pdst, stride, ptr);

  1006.     }else{

  1007.         r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);

  1008.         ptr[0] = 0;

  1009.     }

  1010. }

  1011. 

  1012. static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)

  1013. {

  1014.     LOCAL_ALIGNED_16(int16_t, block16, [16]);

  1015.     MpegEncContext *s    = &r->s;

  1016.     GetBitContext  *gb   = &s->gb;

  1017.     int             q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],

  1018.                     q_ac = rv34_qscale_tab[s->qscale];

  1019.     uint8_t        *dst  = s->dest[0];

  1020.     int16_t        *ptr  = s->block[0];

  1021.     int i, j, itype, has_ac;

  1022. 

  1023.     memset(block16, 0, 16 * sizeof(*block16));

  1024. 

  1025.     has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);

  1026.     if(has_ac)

  1027.         r->rdsp.rv34_inv_transform(block16);

  1028.     else

  1029.         r->rdsp.rv34_inv_transform_dc(block16);

  1030. 

  1031.     itype = ittrans16[intra_types[0]];

  1032.     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1033.     r->h.pred16x16[itype](dst, s->linesize);

  1034. 

  1035.     for(j = 0; j < 4; j++){

  1036.         for(i = 0; i < 4; i++, cbp >>= 1){

  1037.             int dc = block16[i + j*4];

  1038. 

  1039.             if(cbp & 1){

  1040.                 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);

  1041.             }else

  1042.                 has_ac = 0;

  1043. 

  1044.             if(has_ac){

  1045.                 ptr[0] = dc;

  1046.                 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);

  1047.             }else

  1048.                 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);

  1049.         }

  1050. 

  1051.         dst += 4*s->linesize;

  1052.     }

  1053. 

  1054.     itype = ittrans16[intra_types[0]];

  1055.     if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;

  1056.     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1057. 

  1058.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];

  1059.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];

  1060. 

  1061.     for(j = 1; j < 3; j++){

  1062.         dst = s->dest[j];

  1063.         r->h.pred8x8[itype](dst, s->uvlinesize);

  1064.         for(i = 0; i < 4; i++, cbp >>= 1){

  1065.             uint8_t *pdst;

  1066.             if(!(cbp & 1)) continue;

  1067.             pdst   = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;

  1068. 

  1069.             rv34_process_block(r, pdst, s->uvlinesize,

  1070.                                r->chroma_vlc, 1, q_dc, q_ac);

  1071.         }

  1072.     }

  1073. }

  1074. 

  1075. static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)

  1076. {

  1077.     MpegEncContext *s   = &r->s;

  1078.     uint8_t        *dst = s->dest[0];

  1079.     int      avail[6*8] = {0};

  1080.     int i, j, k;

  1081.     int idx, q_ac, q_dc;

  1082. 

  1083.     // Set neighbour information.

  1084.     if(r->avail_cache[1])

  1085.         avail[0] = 1;

  1086.     if(r->avail_cache[2])

  1087.         avail[1] = avail[2] = 1;

  1088.     if(r->avail_cache[3])

  1089.         avail[3] = avail[4] = 1;

  1090.     if(r->avail_cache[4])

  1091.         avail[5] = 1;

  1092.     if(r->avail_cache[5])

  1093.         avail[8] = avail[16] = 1;

  1094.     if(r->avail_cache[9])

  1095.         avail[24] = avail[32] = 1;

  1096. 

  1097.     q_ac = rv34_qscale_tab[s->qscale];

  1098.     for(j = 0; j < 4; j++){

  1099.         idx = 9 + j*8;

  1100.         for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){

  1101.             rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);

  1102.             avail[idx] = 1;

  1103.             if(!(cbp & 1)) continue;

  1104. 

  1105.             rv34_process_block(r, dst, s->linesize,

  1106.                                r->luma_vlc, 0, q_ac, q_ac);

  1107.         }

  1108.         dst += s->linesize * 4 - 4*4;

  1109.         intra_types += r->intra_types_stride;

  1110.     }

  1111. 

  1112.     intra_types -= r->intra_types_stride * 4;

  1113. 

  1114.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];

  1115.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];

  1116. 

  1117.     for(k = 0; k < 2; k++){

  1118.         dst = s->dest[1+k];

  1119.         fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);

  1120. 

  1121.         for(j = 0; j < 2; j++){

  1122.             int* acache = r->avail_cache + 6 + j*4;

  1123.             for(i = 0; i < 2; i++, cbp >>= 1, acache++){

  1124.                 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];

  1125.                 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);

  1126.                 acache[0] = 1;

  1127. 

  1128.                 if(!(cbp&1)) continue;

  1129. 

  1130.                 rv34_process_block(r, dst + 4*i, s->uvlinesize,

  1131.                                    r->chroma_vlc, 1, q_dc, q_ac);

  1132.             }

  1133. 

  1134.             dst += 4*s->uvlinesize;

  1135.         }

  1136.     }

  1137. }

  1138. 

  1139. static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)

  1140. {

  1141.     int d;

  1142.     d = motion_val[0][0] - motion_val[-step][0];

  1143.     if(d < -3 || d > 3)

  1144.         return 1;

  1145.     d = motion_val[0][1] - motion_val[-step][1];

  1146.     if(d < -3 || d > 3)

  1147.         return 1;

  1148.     return 0;

  1149. }

  1150. 

  1151. static int rv34_set_deblock_coef(RV34DecContext *r)

  1152. {

  1153.     MpegEncContext *s = &r->s;

  1154.     int hmvmask = 0, vmvmask = 0, i, j;

  1155.     int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;

  1156.     int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];

  1157.     for(j = 0; j < 16; j += 8){

  1158.         for(i = 0; i < 2; i++){

  1159.             if(is_mv_diff_gt_3(motion_val + i, 1))

  1160.                 vmvmask |= 0x11 << (j + i*2);

  1161.             if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))

  1162.                 hmvmask |= 0x03 << (j + i*2);

  1163.         }

  1164.         motion_val += s->b8_stride;

  1165.     }

  1166.     if(s->first_slice_line)

  1167.         hmvmask &= ~0x000F;

  1168.     if(!s->mb_x)

  1169.         vmvmask &= ~0x1111;

  1170.     if(r->rv30){ //RV30 marks both subblocks on the edge for filtering

  1171.         vmvmask |= (vmvmask & 0x4444) >> 1;

  1172.         hmvmask |= (hmvmask & 0x0F00) >> 4;

  1173.         if(s->mb_x)

  1174.             r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;

  1175.         if(!s->first_slice_line)

  1176.             r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;

  1177.     }

  1178.     return hmvmask | vmvmask;

  1179. }

  1180. 

  1181. static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)

  1182. {

  1183.     MpegEncContext *s   = &r->s;

  1184.     GetBitContext  *gb  = &s->gb;

  1185.     uint8_t        *dst = s->dest[0];

  1186.     int16_t        *ptr = s->block[0];

  1187.     int          mb_pos = s->mb_x + s->mb_y * s->mb_stride;

  1188.     int cbp, cbp2;

  1189.     int q_dc, q_ac, has_ac;

  1190.     int i, j;

  1191.     int dist;

  1192. 

  1193.     // Calculate which neighbours are available. Maybe it's worth optimizing too.

  1194.     memset(r->avail_cache, 0, sizeof(r->avail_cache));

  1195.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);

  1196.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;

  1197.     if(s->mb_x && dist)

  1198.         r->avail_cache[5] =

  1199.         r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];

  1200.     if(dist >= s->mb_width)

  1201.         r->avail_cache[2] =

  1202.         r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];

  1203.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)

  1204.         r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];

  1205.     if(s->mb_x && dist > s->mb_width)

  1206.         r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];

  1207. 

  1208.     s->qscale = r->si.quant;

  1209.     cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);

  1210.     r->cbp_luma  [mb_pos] = cbp;

  1211.     r->cbp_chroma[mb_pos] = cbp >> 16;

  1212.     r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];

  1213.     s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;

  1214. 

  1215.     if(cbp == -1)

  1216.         return -1;

  1217. 

  1218.     if (IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){

  1219.         if(r->is16) rv34_output_i16x16(r, intra_types, cbp);

  1220.         else        rv34_output_intra(r, intra_types, cbp);

  1221.         return 0;

  1222.     }

  1223. 

  1224.     if(r->is16){

  1225.         // Only for RV34_MB_P_MIX16x16

  1226.         LOCAL_ALIGNED_16(int16_t, block16, [16]);

  1227.         memset(block16, 0, 16 * sizeof(*block16));

  1228.         q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];

  1229.         q_ac = rv34_qscale_tab[s->qscale];

  1230.         if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))

  1231.             r->rdsp.rv34_inv_transform(block16);

  1232.         else

  1233.             r->rdsp.rv34_inv_transform_dc(block16);

  1234. 

  1235.         q_ac = rv34_qscale_tab[s->qscale];

  1236. 

  1237.         for(j = 0; j < 4; j++){

  1238.             for(i = 0; i < 4; i++, cbp >>= 1){

  1239.                 int      dc   = block16[i + j*4];

  1240. 

  1241.                 if(cbp & 1){

  1242.                     has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);

  1243.                 }else

  1244.                     has_ac = 0;

  1245. 

  1246.                 if(has_ac){

  1247.                     ptr[0] = dc;

  1248.                     r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);

  1249.                 }else

  1250.                     r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);

  1251.             }

  1252. 

  1253.             dst += 4*s->linesize;

  1254.         }

  1255. 

  1256.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);

  1257.     }else{

  1258.         q_ac = rv34_qscale_tab[s->qscale];

  1259. 

  1260.         for(j = 0; j < 4; j++){

  1261.             for(i = 0; i < 4; i++, cbp >>= 1){

  1262.                 if(!(cbp & 1)) continue;

  1263. 

  1264.                 rv34_process_block(r, dst + 4*i, s->linesize,

  1265.                                    r->luma_vlc, 0, q_ac, q_ac);

  1266.             }

  1267.             dst += 4*s->linesize;

  1268.         }

  1269.     }

  1270. 

  1271.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];

  1272.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];

  1273. 

  1274.     for(j = 1; j < 3; j++){

  1275.         dst = s->dest[j];

  1276.         for(i = 0; i < 4; i++, cbp >>= 1){

  1277.             uint8_t *pdst;

  1278.             if(!(cbp & 1)) continue;

  1279.             pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;

  1280. 

  1281.             rv34_process_block(r, pdst, s->uvlinesize,

  1282.                                r->chroma_vlc, 1, q_dc, q_ac);

  1283.         }

  1284.     }

  1285. 

  1286.     return 0;

  1287. }

  1288. 

  1289. static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)

  1290. {

  1291.     MpegEncContext *s = &r->s;

  1292.     int cbp, dist;

  1293.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;

  1294. 

  1295.     // Calculate which neighbours are available. Maybe it's worth optimizing too.

  1296.     memset(r->avail_cache, 0, sizeof(r->avail_cache));

  1297.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);

  1298.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;

  1299.     if(s->mb_x && dist)

  1300.         r->avail_cache[5] =

  1301.         r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];

  1302.     if(dist >= s->mb_width)

  1303.         r->avail_cache[2] =

  1304.         r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];

  1305.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)

  1306.         r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];

  1307.     if(s->mb_x && dist > s->mb_width)

  1308.         r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];

  1309. 

  1310.     s->qscale = r->si.quant;

  1311.     cbp = rv34_decode_intra_mb_header(r, intra_types);

  1312.     r->cbp_luma  [mb_pos] = cbp;

  1313.     r->cbp_chroma[mb_pos] = cbp >> 16;

  1314.     r->deblock_coefs[mb_pos] = 0xFFFF;

  1315.     s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;

  1316. 

  1317.     if(cbp == -1)

  1318.         return -1;

  1319. 

  1320.     if(r->is16){

  1321.         rv34_output_i16x16(r, intra_types, cbp);

  1322.         return 0;

  1323.     }

  1324. 

  1325.     rv34_output_intra(r, intra_types, cbp);

  1326.     return 0;

  1327. }

  1328. 

  1329. static int check_slice_end(RV34DecContext *r, MpegEncContext *s)

  1330. {

  1331.     int bits;

  1332.     if(s->mb_y >= s->mb_height)

  1333.         return 1;

  1334.     if(!s->mb_num_left)

  1335.         return 1;

  1336.     if(r->s.mb_skip_run > 1)

  1337.         return 0;

  1338.     bits = get_bits_left(&s->gb);

  1339.     if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))

  1340.         return 1;

  1341.     return 0;

  1342. }

  1343. 

  1344. 

  1345. static void rv34_decoder_free(RV34DecContext *r)

  1346. {

  1347.     av_freep(&r->intra_types_hist);

  1348.     r->intra_types = NULL;

  1349.     av_freep(&r->tmp_b_block_base);

  1350.     av_freep(&r->mb_type);

  1351.     av_freep(&r->cbp_luma);

  1352.     av_freep(&r->cbp_chroma);

  1353.     av_freep(&r->deblock_coefs);

  1354. }

  1355. 

  1356. 

  1357. static int rv34_decoder_alloc(RV34DecContext *r)

  1358. {

  1359.     r->intra_types_stride = r->s.mb_width * 4 + 4;

  1360. 

  1361.     r->cbp_chroma       = av_malloc(r->s.mb_stride * r->s.mb_height *

  1362.                                     sizeof(*r->cbp_chroma));

  1363.     r->cbp_luma         = av_malloc(r->s.mb_stride * r->s.mb_height *

  1364.                                     sizeof(*r->cbp_luma));

  1365.     r->deblock_coefs    = av_malloc(r->s.mb_stride * r->s.mb_height *

  1366.                                     sizeof(*r->deblock_coefs));

  1367.     r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *

  1368.                                     sizeof(*r->intra_types_hist));

  1369.     r->mb_type          = av_mallocz(r->s.mb_stride * r->s.mb_height *

  1370.                                      sizeof(*r->mb_type));

  1371. 

  1372.     if (!(r->cbp_chroma       && r->cbp_luma && r->deblock_coefs &&

  1373.           r->intra_types_hist && r->mb_type)) {

  1374.         rv34_decoder_free(r);

  1375.         return AVERROR(ENOMEM);

  1376.     }

  1377. 

  1378.     r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;

  1379. 

  1380.     return 0;

  1381. }

  1382. 

  1383. 

  1384. static int rv34_decoder_realloc(RV34DecContext *r)

  1385. {

  1386.     rv34_decoder_free(r);

  1387.     return rv34_decoder_alloc(r);

  1388. }

  1389. 

  1390. 

  1391. static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)

  1392. {

  1393.     MpegEncContext *s = &r->s;

  1394.     GetBitContext *gb = &s->gb;

  1395.     int mb_pos, slice_type;

  1396.     int res;

  1397. 

  1398.     init_get_bits(&r->s.gb, buf, buf_size*8);

  1399.     res = r->parse_slice_header(r, gb, &r->si);

  1400.     if(res < 0){

  1401.         av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");

  1402.         return -1;

  1403.     }

  1404. 

  1405.     slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;

  1406.     if (slice_type != s->pict_type) {

  1407.         av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");

  1408.         return AVERROR_INVALIDDATA;

  1409.     }

  1410. 

  1411.     r->si.end = end;

  1412.     s->qscale = r->si.quant;

  1413.     s->mb_num_left = r->si.end - r->si.start;

  1414.     r->s.mb_skip_run = 0;

  1415. 

  1416.     mb_pos = s->mb_x + s->mb_y * s->mb_width;

  1417.     if(r->si.start != mb_pos){

  1418.         av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);

  1419.         s->mb_x = r->si.start % s->mb_width;

  1420.         s->mb_y = r->si.start / s->mb_width;

  1421.     }

  1422.     memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));

  1423.     s->first_slice_line = 1;

  1424.     s->resync_mb_x = s->mb_x;

  1425.     s->resync_mb_y = s->mb_y;

  1426. 

  1427.     ff_init_block_index(s);

  1428.     while(!check_slice_end(r, s)) {

  1429.         ff_update_block_index(s);

  1430. 

  1431.         if(r->si.type)

  1432.             res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);

  1433.         else

  1434.             res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);

  1435.         if(res < 0){

  1436.             ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);

  1437.             return -1;

  1438.         }

  1439.         if (++s->mb_x == s->mb_width) {

  1440.             s->mb_x = 0;

  1441.             s->mb_y++;

  1442.             ff_init_block_index(s);

  1443. 

  1444.             memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1445.             memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1446. 

  1447.             if(r->loop_filter && s->mb_y >= 2)

  1448.                 r->loop_filter(r, s->mb_y - 2);

  1449. 

  1450.             if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))

  1451.                 ff_thread_report_progress(&s->current_picture_ptr->tf,

  1452.                                           s->mb_y - 2, 0);

  1453. 

  1454.         }

  1455.         if(s->mb_x == s->resync_mb_x)

  1456.             s->first_slice_line=0;

  1457.         s->mb_num_left--;

  1458.     }

  1459.     ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);

  1460. 

  1461.     return s->mb_y == s->mb_height;

  1462. }

  1463. 

  1464. /** @} */ // recons group end

  1465. 

  1466. /**

  1467.  * Initialize decoder.

  1468.  */

  1469. av_cold int ff_rv34_decode_init(AVCodecContext *avctx)

  1470. {

  1471.     RV34DecContext *r = avctx->priv_data;

  1472.     MpegEncContext *s = &r->s;

  1473.     int ret;

  1474. 

  1475.     ff_MPV_decode_defaults(s);

  1476.     s->avctx      = avctx;

  1477.     s->out_format = FMT_H263;

  1478.     s->codec_id   = avctx->codec_id;

  1479. 

  1480.     s->width  = avctx->width;

  1481.     s->height = avctx->height;

  1482. 

  1483.     r->s.avctx = avctx;

  1484.     avctx->pix_fmt = AV_PIX_FMT_YUV420P;

  1485.     avctx->has_b_frames = 1;

  1486.     s->low_delay = 0;

  1487. 

  1488.     if ((ret = ff_MPV_common_init(s)) < 0)

  1489.         return ret;

  1490. 

  1491.     ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);

  1492. 

  1493. #if CONFIG_RV30_DECODER

  1494.     if (avctx->codec_id == AV_CODEC_ID_RV30)

  1495.         ff_rv30dsp_init(&r->rdsp);

  1496. #endif

  1497. #if CONFIG_RV40_DECODER

  1498.     if (avctx->codec_id == AV_CODEC_ID_RV40)

  1499.         ff_rv40dsp_init(&r->rdsp);

  1500. #endif

  1501. 

  1502.     if ((ret = rv34_decoder_alloc(r)) < 0) {

  1503.         ff_MPV_common_end(&r->s);

  1504.         return ret;

  1505.     }

  1506. 

  1507.     if(!intra_vlcs[0].cbppattern[0].bits)

  1508.         rv34_init_tables();

  1509. 

  1510.     avctx->internal->allocate_progress = 1;

  1511. 

  1512.     return 0;

  1513. }

  1514. 

  1515. int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx)

  1516. {

  1517.     int err;

  1518.     RV34DecContext *r = avctx->priv_data;

  1519. 

  1520.     r->s.avctx = avctx;

  1521. 

  1522.     if (avctx->internal->is_copy) {

  1523.         r->tmp_b_block_base = NULL;

  1524.         if ((err = ff_MPV_common_init(&r->s)) < 0)

  1525.             return err;

  1526.         if ((err = rv34_decoder_alloc(r)) < 0) {

  1527.             ff_MPV_common_end(&r->s);

  1528.             return err;

  1529.         }

  1530.     }

  1531. 

  1532.     return 0;

  1533. }

  1534. 

  1535. int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)

  1536. {

  1537.     RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;

  1538.     MpegEncContext * const s = &r->s, * const s1 = &r1->s;

  1539.     int err;

  1540. 

  1541.     if (dst == src || !s1->context_initialized)

  1542.         return 0;

  1543. 

  1544.     if (s->height != s1->height || s->width != s1->width) {

  1545.         s->height = s1->height;

  1546.         s->width  = s1->width;

  1547.         if ((err = ff_MPV_common_frame_size_change(s)) < 0)

  1548.             return err;

  1549.         if ((err = rv34_decoder_realloc(r)) < 0)

  1550.             return err;

  1551.     }

  1552. 

  1553.     if ((err = ff_mpeg_update_thread_context(dst, src)))

  1554.         return err;

  1555. 

  1556.     r->cur_pts  = r1->cur_pts;

  1557.     r->last_pts = r1->last_pts;

  1558.     r->next_pts = r1->next_pts;

  1559. 

  1560.     memset(&r->si, 0, sizeof(r->si));

  1561. 

  1562.     return 0;

  1563. }

  1564. 

  1565. static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)

  1566. {

  1567.     if(avctx->slice_count) return avctx->slice_offset[n];

  1568.     else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);

  1569. }

  1570. 

  1571. static int finish_frame(AVCodecContext *avctx, AVFrame *pict)

  1572. {

  1573.     RV34DecContext *r = avctx->priv_data;

  1574.     MpegEncContext *s = &r->s;

  1575.     int got_picture = 0, ret;

  1576. 

  1577.     ff_er_frame_end(&s->er);

  1578.     ff_MPV_frame_end(s);

  1579.     s->mb_num_left = 0;

  1580. 

  1581.     if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))

  1582.         ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);

  1583. 

  1584.     if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {

  1585.         if ((ret = av_frame_ref(pict, &s->current_picture_ptr->f)) < 0)

  1586.             return ret;

  1587.         ff_print_debug_info(s, s->current_picture_ptr);

  1588.         got_picture = 1;

  1589.     } else if (s->last_picture_ptr != NULL) {

  1590.         if ((ret = av_frame_ref(pict, &s->last_picture_ptr->f)) < 0)

  1591.             return ret;

  1592.         ff_print_debug_info(s, s->last_picture_ptr);

  1593.         got_picture = 1;

  1594.     }

  1595. 

  1596.     return got_picture;

  1597. }

  1598. 

  1599. int ff_rv34_decode_frame(AVCodecContext *avctx,

  1600.                             void *data, int *got_picture_ptr,

  1601.                             AVPacket *avpkt)

  1602. {

  1603.     const uint8_t *buf = avpkt->data;

  1604.     int buf_size = avpkt->size;

  1605.     RV34DecContext *r = avctx->priv_data;

  1606.     MpegEncContext *s = &r->s;

  1607.     AVFrame *pict = data;

  1608.     SliceInfo si;

  1609.     int i, ret;

  1610.     int slice_count;

  1611.     const uint8_t *slices_hdr = NULL;

  1612.     int last = 0;

  1613. 

  1614.     /* no supplementary picture */

  1615.     if (buf_size == 0) {

  1616.         /* special case for last picture */

  1617.         if (s->low_delay==0 && s->next_picture_ptr) {

  1618.             if ((ret = av_frame_ref(pict, &s->next_picture_ptr->f)) < 0)

  1619.                 return ret;

  1620.             s->next_picture_ptr = NULL;

  1621. 

  1622.             *got_picture_ptr = 1;

  1623.         }

  1624.         return 0;

  1625.     }

  1626. 

  1627.     if(!avctx->slice_count){

  1628.         slice_count = (*buf++) + 1;

  1629.         slices_hdr = buf + 4;

  1630.         buf += 8 * slice_count;

  1631.         buf_size -= 1 + 8 * slice_count;

  1632.     }else

  1633.         slice_count = avctx->slice_count;

  1634. 

  1635.     //parse first slice header to check whether this frame can be decoded

  1636.     if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||

  1637.        get_slice_offset(avctx, slices_hdr, 0) > buf_size){

  1638.         av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1639.         return AVERROR_INVALIDDATA;

  1640.     }

  1641.     init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);

  1642.     if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){

  1643.         av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");

  1644.         return AVERROR_INVALIDDATA;

  1645.     }

  1646.     if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) &&

  1647.         si.type == AV_PICTURE_TYPE_B) {

  1648.         av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "

  1649.                "reference data.\n");

  1650.         return AVERROR_INVALIDDATA;

  1651.     }

  1652.     if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)

  1653.        || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)

  1654.        ||  avctx->skip_frame >= AVDISCARD_ALL)

  1655.         return avpkt->size;

  1656. 

  1657.     /* first slice */

  1658.     if (si.start == 0) {

  1659.         if (s->mb_num_left > 0) {

  1660.             av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.",

  1661.                    s->mb_num_left);

  1662.             ff_er_frame_end(&s->er);

  1663.             ff_MPV_frame_end(s);

  1664.         }

  1665. 

  1666.         if (s->width != si.width || s->height != si.height) {

  1667.             int err;

  1668. 

  1669.             av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",

  1670.                    si.width, si.height);

  1671. 

  1672.             s->width  = si.width;

  1673.             s->height = si.height;

  1674. 

  1675.             err = ff_set_dimensions(s->avctx, s->width, s->height);

  1676.             if (err < 0)

  1677.                 return err;

  1678. 

  1679.             if ((err = ff_MPV_common_frame_size_change(s)) < 0)

  1680.                 return err;

  1681.             if ((err = rv34_decoder_realloc(r)) < 0)

  1682.                 return err;

  1683.         }

  1684.         s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;

  1685.         if (ff_MPV_frame_start(s, s->avctx) < 0)

  1686.             return -1;

  1687.         ff_mpeg_er_frame_start(s);

  1688.         if (!r->tmp_b_block_base) {

  1689.             int i;

  1690. 

  1691.             r->tmp_b_block_base = av_malloc(s->linesize * 48);

  1692.             for (i = 0; i < 2; i++)

  1693.                 r->tmp_b_block_y[i] = r->tmp_b_block_base

  1694.                                       + i * 16 * s->linesize;

  1695.             for (i = 0; i < 4; i++)

  1696.                 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize

  1697.                                        + (i >> 1) * 8 * s->uvlinesize

  1698.                                        + (i &  1) * 16;

  1699.         }

  1700.         r->cur_pts = si.pts;

  1701.         if (s->pict_type != AV_PICTURE_TYPE_B) {

  1702.             r->last_pts = r->next_pts;

  1703.             r->next_pts = r->cur_pts;

  1704.         } else {

  1705.             int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);

  1706.             int dist0   = GET_PTS_DIFF(r->cur_pts,  r->last_pts);

  1707.             int dist1   = GET_PTS_DIFF(r->next_pts, r->cur_pts);

  1708. 

  1709.             if(!refdist){

  1710.                 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;

  1711.                 r->scaled_weight = 0;

  1712.             }else{

  1713.                 r->mv_weight1 = (dist0 << 14) / refdist;

  1714.                 r->mv_weight2 = (dist1 << 14) / refdist;

  1715.                 if((r->mv_weight1|r->mv_weight2) & 511){

  1716.                     r->weight1 = r->mv_weight1;

  1717.                     r->weight2 = r->mv_weight2;

  1718.                     r->scaled_weight = 0;

  1719.                 }else{

  1720.                     r->weight1 = r->mv_weight1 >> 9;

  1721.                     r->weight2 = r->mv_weight2 >> 9;

  1722.                     r->scaled_weight = 1;

  1723.                 }

  1724.             }

  1725.         }

  1726.         s->mb_x = s->mb_y = 0;

  1727.         ff_thread_finish_setup(s->avctx);

  1728.     } else if (HAVE_THREADS &&

  1729.                (s->avctx->active_thread_type & FF_THREAD_FRAME)) {

  1730.         av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "

  1731.                "multithreading mode (start MB is %d).\n", si.start);

  1732.         return AVERROR_INVALIDDATA;

  1733.     }

  1734. 

  1735.     for(i = 0; i < slice_count; i++){

  1736.         int offset = get_slice_offset(avctx, slices_hdr, i);

  1737.         int size;

  1738.         if(i+1 == slice_count)

  1739.             size = buf_size - offset;

  1740.         else

  1741.             size = get_slice_offset(avctx, slices_hdr, i+1) - offset;

  1742. 

  1743.         if(offset < 0 || offset > buf_size){

  1744.             av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1745.             break;

  1746.         }

  1747. 

  1748.         r->si.end = s->mb_width * s->mb_height;

  1749.         s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;

  1750. 

  1751.         if(i+1 < slice_count){

  1752.             if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||

  1753.                 get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {

  1754.                 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1755.                 break;

  1756.             }

  1757.             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);

  1758.             if(r->parse_slice_header(r, &r->s.gb, &si) < 0){

  1759.                 if(i+2 < slice_count)

  1760.                     size = get_slice_offset(avctx, slices_hdr, i+2) - offset;

  1761.                 else

  1762.                     size = buf_size - offset;

  1763.             }else

  1764.                 r->si.end = si.start;

  1765.         }

  1766.         if (size < 0 || size > buf_size - offset) {

  1767.             av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");

  1768.             break;

  1769.         }

  1770.         last = rv34_decode_slice(r, r->si.end, buf + offset, size);

  1771.         if(last)

  1772.             break;

  1773.     }

  1774. 

  1775.     if (s->current_picture_ptr) {

  1776.         if (last) {

  1777.             if(r->loop_filter)

  1778.                 r->loop_filter(r, s->mb_height - 1);

  1779. 

  1780.             ret = finish_frame(avctx, pict);

  1781.             if (ret < 0)

  1782.                 return ret;

  1783.             *got_picture_ptr = ret;

  1784.         } else if (HAVE_THREADS &&

  1785.                    (s->avctx->active_thread_type & FF_THREAD_FRAME)) {

  1786.             av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");

  1787.             /* always mark the current frame as finished, frame-mt supports

  1788.              * only complete frames */

  1789.             ff_er_frame_end(&s->er);

  1790.             ff_MPV_frame_end(s);

  1791.             s->mb_num_left = 0;

  1792.             ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);

  1793.             return AVERROR_INVALIDDATA;

  1794.         }

  1795.     }

  1796. 

  1797.     return avpkt->size;

  1798. }

  1799. 

  1800. av_cold int ff_rv34_decode_end(AVCodecContext *avctx)

  1801. {

  1802.     RV34DecContext *r = avctx->priv_data;

  1803. 

  1804.     ff_MPV_common_end(&r->s);

  1805.     rv34_decoder_free(r);

  1806. 

  1807.     return 0;

  1808. }