Subversion Repositories Kolibri OS

Rev

Go to most recent revision | Blame | Last modification | View Log | RSS feed

  1. /*
  2.  * RV30/40 decoder common data
  3.  * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
  4.  *
  5.  * This file is part of FFmpeg.
  6.  *
  7.  * FFmpeg 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.  * FFmpeg 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 FFmpeg; 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/imgutils.h"
  28. #include "libavutil/internal.h"
  29.  
  30. #include "avcodec.h"
  31. #include "error_resilience.h"
  32. #include "mpegvideo.h"
  33. #include "golomb.h"
  34. #include "internal.h"
  35. #include "mathops.h"
  36. #include "rectangle.h"
  37. #include "thread.h"
  38.  
  39. #include "rv34vlc.h"
  40. #include "rv34data.h"
  41. #include "rv34.h"
  42.  
  43. static inline void ZERO8x2(void* dst, int stride)
  44. {
  45.     fill_rectangle(dst,                 1, 2, stride, 0, 4);
  46.     fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
  47. }
  48.  
  49. /** translation of RV30/40 macroblock types to lavc ones */
  50. static const int rv34_mb_type_to_lavc[12] = {
  51.     MB_TYPE_INTRA,
  52.     MB_TYPE_INTRA16x16              | MB_TYPE_SEPARATE_DC,
  53.     MB_TYPE_16x16   | MB_TYPE_L0,
  54.     MB_TYPE_8x8     | MB_TYPE_L0,
  55.     MB_TYPE_16x16   | MB_TYPE_L0,
  56.     MB_TYPE_16x16   | MB_TYPE_L1,
  57.     MB_TYPE_SKIP,
  58.     MB_TYPE_DIRECT2 | MB_TYPE_16x16,
  59.     MB_TYPE_16x8    | MB_TYPE_L0,
  60.     MB_TYPE_8x16    | MB_TYPE_L0,
  61.     MB_TYPE_16x16   | MB_TYPE_L0L1,
  62.     MB_TYPE_16x16   | MB_TYPE_L0    | MB_TYPE_SEPARATE_DC
  63. };
  64.  
  65.  
  66. static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
  67.  
  68. static int rv34_decode_mv(RV34DecContext *r, int block_type);
  69.  
  70. /**
  71.  * @name RV30/40 VLC generating functions
  72.  * @{
  73.  */
  74.  
  75. static const int table_offs[] = {
  76.       0,   1818,   3622,   4144,   4698,   5234,   5804,   5868,   5900,   5932,
  77.    5996,   6252,   6316,   6348,   6380,   7674,   8944,  10274,  11668,  12250,
  78.   14060,  15846,  16372,  16962,  17512,  18148,  18180,  18212,  18244,  18308,
  79.   18564,  18628,  18660,  18692,  20036,  21314,  22648,  23968,  24614,  26384,
  80.   28190,  28736,  29366,  29938,  30608,  30640,  30672,  30704,  30768,  31024,
  81.   31088,  31120,  31184,  32570,  33898,  35236,  36644,  37286,  39020,  40802,
  82.   41368,  42052,  42692,  43348,  43380,  43412,  43444,  43476,  43604,  43668,
  83.   43700,  43732,  45100,  46430,  47778,  49160,  49802,  51550,  53340,  53972,
  84.   54648,  55348,  55994,  56122,  56154,  56186,  56218,  56346,  56410,  56442,
  85.   56474,  57878,  59290,  60636,  62036,  62682,  64460,  64524,  64588,  64716,
  86.   64844,  66076,  67466,  67978,  68542,  69064,  69648,  70296,  72010,  72074,
  87.   72138,  72202,  72330,  73572,  74936,  75454,  76030,  76566,  77176,  77822,
  88.   79582,  79646,  79678,  79742,  79870,  81180,  82536,  83064,  83672,  84242,
  89.   84934,  85576,  87384,  87448,  87480,  87544,  87672,  88982,  90340,  90902,
  90.   91598,  92182,  92846,  93488,  95246,  95278,  95310,  95374,  95502,  96878,
  91.   98266,  98848,  99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
  92.  103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
  93.  111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
  94. };
  95.  
  96. static VLC_TYPE table_data[117592][2];
  97.  
  98. /**
  99.  * Generate VLC from codeword lengths.
  100.  * @param bits   codeword lengths (zeroes are accepted)
  101.  * @param size   length of input data
  102.  * @param vlc    output VLC
  103.  * @param insyms symbols for input codes (NULL for default ones)
  104.  * @param num    VLC table number (for static initialization)
  105.  */
  106. static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
  107.                          const int num)
  108. {
  109.     int i;
  110.     int counts[17] = {0}, codes[17];
  111.     uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE];
  112.     uint8_t bits2[MAX_VLC_SIZE];
  113.     int maxbits = 0, realsize = 0;
  114.  
  115.     for(i = 0; i < size; i++){
  116.         if(bits[i]){
  117.             bits2[realsize] = bits[i];
  118.             syms[realsize] = insyms ? insyms[i] : i;
  119.             realsize++;
  120.             maxbits = FFMAX(maxbits, bits[i]);
  121.             counts[bits[i]]++;
  122.         }
  123.     }
  124.  
  125.     codes[0] = 0;
  126.     for(i = 0; i < 16; i++)
  127.         codes[i+1] = (codes[i] + counts[i]) << 1;
  128.     for(i = 0; i < realsize; i++)
  129.         cw[i] = codes[bits2[i]]++;
  130.  
  131.     vlc->table = &table_data[table_offs[num]];
  132.     vlc->table_allocated = table_offs[num + 1] - table_offs[num];
  133.     ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
  134.                        bits2, 1, 1,
  135.                        cw,    2, 2,
  136.                        syms,  2, 2, INIT_VLC_USE_NEW_STATIC);
  137. }
  138.  
  139. /**
  140.  * Initialize all tables.
  141.  */
  142. static av_cold void rv34_init_tables(void)
  143. {
  144.     int i, j, k;
  145.  
  146.     for(i = 0; i < NUM_INTRA_TABLES; i++){
  147.         for(j = 0; j < 2; j++){
  148.             rv34_gen_vlc(rv34_table_intra_cbppat   [i][j], CBPPAT_VLC_SIZE,   &intra_vlcs[i].cbppattern[j],     NULL, 19*i + 0 + j);
  149.             rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
  150.             rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j],  NULL, 19*i + 4 + j);
  151.             for(k = 0; k < 4; k++){
  152.                 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);
  153.             }
  154.         }
  155.         for(j = 0; j < 4; j++){
  156.             rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
  157.         }
  158.         rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
  159.     }
  160.  
  161.     for(i = 0; i < NUM_INTER_TABLES; i++){
  162.         rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
  163.         for(j = 0; j < 4; j++){
  164.             rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
  165.         }
  166.         for(j = 0; j < 2; j++){
  167.             rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j],  NULL, i*12 + 100 + j);
  168.             rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
  169.             rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j],  NULL, i*12 + 104 + j);
  170.         }
  171.         rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
  172.     }
  173. }
  174.  
  175. /** @} */ // vlc group
  176.  
  177. /**
  178.  * @name RV30/40 4x4 block decoding functions
  179.  * @{
  180.  */
  181.  
  182. /**
  183.  * Decode coded block pattern.
  184.  */
  185. static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
  186. {
  187.     int pattern, code, cbp=0;
  188.     int ones;
  189.     static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
  190.     static const int shifts[4] = { 0, 2, 8, 10 };
  191.     const int *curshift = shifts;
  192.     int i, t, mask;
  193.  
  194.     code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
  195.     pattern = code & 0xF;
  196.     code >>= 4;
  197.  
  198.     ones = rv34_count_ones[pattern];
  199.  
  200.     for(mask = 8; mask; mask >>= 1, curshift++){
  201.         if(pattern & mask)
  202.             cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
  203.     }
  204.  
  205.     for(i = 0; i < 4; i++){
  206.         t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
  207.         if(t == 1)
  208.             cbp |= cbp_masks[get_bits1(gb)] << i;
  209.         if(t == 2)
  210.             cbp |= cbp_masks[2] << i;
  211.     }
  212.     return cbp;
  213. }
  214.  
  215. /**
  216.  * Get one coefficient value from the bitstream and store it.
  217.  */
  218. static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
  219. {
  220.     if(coef){
  221.         if(coef == esc){
  222.             coef = get_vlc2(gb, vlc->table, 9, 2);
  223.             if(coef > 23){
  224.                 coef -= 23;
  225.                 coef = 22 + ((1 << coef) | get_bits(gb, coef));
  226.             }
  227.             coef += esc;
  228.         }
  229.         if(get_bits1(gb))
  230.             coef = -coef;
  231.         *dst = (coef*q + 8) >> 4;
  232.     }
  233. }
  234.  
  235. /**
  236.  * Decode 2x2 subblock of coefficients.
  237.  */
  238. static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
  239. {
  240.     int flags = modulo_three_table[code];
  241.  
  242.     decode_coeff(    dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q);
  243.     if(is_block2){
  244.         decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
  245.         decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
  246.     }else{
  247.         decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
  248.         decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
  249.     }
  250.     decode_coeff(    dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
  251. }
  252.  
  253. /**
  254.  * Decode a single coefficient.
  255.  */
  256. static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)
  257. {
  258.     int coeff = modulo_three_table[code] >> 6;
  259.     decode_coeff(dst, coeff, 3, gb, vlc, q);
  260. }
  261.  
  262. static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,
  263.                                     int q_dc, int q_ac1, int q_ac2)
  264. {
  265.     int flags = modulo_three_table[code];
  266.  
  267.     decode_coeff(dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q_dc);
  268.     decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
  269.     decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
  270.     decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
  271. }
  272.  
  273. /**
  274.  * Decode coefficients for 4x4 block.
  275.  *
  276.  * This is done by filling 2x2 subblocks with decoded coefficients
  277.  * in this order (the same for subblocks and subblock coefficients):
  278.  *  o--o
  279.  *    /
  280.  *   /
  281.  *  o--o
  282.  */
  283.  
  284. 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)
  285. {
  286.     int code, pattern, has_ac = 1;
  287.  
  288.     code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
  289.  
  290.     pattern = code & 0x7;
  291.  
  292.     code >>= 3;
  293.  
  294.     if (modulo_three_table[code] & 0x3F) {
  295.         decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
  296.     } else {
  297.         decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
  298.         if (!pattern)
  299.             return 0;
  300.         has_ac = 0;
  301.     }
  302.  
  303.     if(pattern & 4){
  304.         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
  305.         decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
  306.     }
  307.     if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
  308.         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
  309.         decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
  310.     }
  311.     if(pattern & 1){
  312.         code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
  313.         decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
  314.     }
  315.     return has_ac | pattern;
  316. }
  317.  
  318. /**
  319.  * @name RV30/40 bitstream parsing
  320.  * @{
  321.  */
  322.  
  323. /**
  324.  * Decode starting slice position.
  325.  * @todo Maybe replace with ff_h263_decode_mba() ?
  326.  */
  327. int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
  328. {
  329.     int i;
  330.     for(i = 0; i < 5; i++)
  331.         if(rv34_mb_max_sizes[i] >= mb_size - 1)
  332.             break;
  333.     return rv34_mb_bits_sizes[i];
  334. }
  335.  
  336. /**
  337.  * Select VLC set for decoding from current quantizer, modifier and frame type.
  338.  */
  339. static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
  340. {
  341.     if(mod == 2 && quant < 19) quant += 10;
  342.     else if(mod && quant < 26) quant += 5;
  343.     return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
  344.                 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
  345. }
  346.  
  347. /**
  348.  * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
  349.  */
  350. static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
  351. {
  352.     MpegEncContext *s = &r->s;
  353.     GetBitContext *gb = &s->gb;
  354.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
  355.     int t;
  356.  
  357.     r->is16 = get_bits1(gb);
  358.     if(r->is16){
  359.         s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA16x16;
  360.         r->block_type = RV34_MB_TYPE_INTRA16x16;
  361.         t = get_bits(gb, 2);
  362.         fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
  363.         r->luma_vlc   = 2;
  364.     }else{
  365.         if(!r->rv30){
  366.             if(!get_bits1(gb))
  367.                 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
  368.         }
  369.         s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA;
  370.         r->block_type = RV34_MB_TYPE_INTRA;
  371.         if(r->decode_intra_types(r, gb, intra_types) < 0)
  372.             return -1;
  373.         r->luma_vlc   = 1;
  374.     }
  375.  
  376.     r->chroma_vlc = 0;
  377.     r->cur_vlcs   = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
  378.  
  379.     return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
  380. }
  381.  
  382. /**
  383.  * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
  384.  */
  385. static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
  386. {
  387.     MpegEncContext *s = &r->s;
  388.     GetBitContext *gb = &s->gb;
  389.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
  390.     int i, t;
  391.  
  392.     r->block_type = r->decode_mb_info(r);
  393.     if(r->block_type == -1)
  394.         return -1;
  395.     s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
  396.     r->mb_type[mb_pos] = r->block_type;
  397.     if(r->block_type == RV34_MB_SKIP){
  398.         if(s->pict_type == AV_PICTURE_TYPE_P)
  399.             r->mb_type[mb_pos] = RV34_MB_P_16x16;
  400.         if(s->pict_type == AV_PICTURE_TYPE_B)
  401.             r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
  402.     }
  403.     r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
  404.     rv34_decode_mv(r, r->block_type);
  405.     if(r->block_type == RV34_MB_SKIP){
  406.         fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
  407.         return 0;
  408.     }
  409.     r->chroma_vlc = 1;
  410.     r->luma_vlc   = 0;
  411.  
  412.     if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
  413.         if(r->is16){
  414.             t = get_bits(gb, 2);
  415.             fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
  416.             r->luma_vlc   = 2;
  417.         }else{
  418.             if(r->decode_intra_types(r, gb, intra_types) < 0)
  419.                 return -1;
  420.             r->luma_vlc   = 1;
  421.         }
  422.         r->chroma_vlc = 0;
  423.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
  424.     }else{
  425.         for(i = 0; i < 16; i++)
  426.             intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
  427.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
  428.         if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
  429.             r->is16 = 1;
  430.             r->chroma_vlc = 1;
  431.             r->luma_vlc   = 2;
  432.             r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
  433.         }
  434.     }
  435.  
  436.     return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
  437. }
  438.  
  439. /** @} */ //bitstream functions
  440.  
  441. /**
  442.  * @name motion vector related code (prediction, reconstruction, motion compensation)
  443.  * @{
  444.  */
  445.  
  446. /** macroblock partition width in 8x8 blocks */
  447. static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
  448.  
  449. /** macroblock partition height in 8x8 blocks */
  450. static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
  451.  
  452. /** availability index for subblocks */
  453. static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
  454.  
  455. /**
  456.  * motion vector prediction
  457.  *
  458.  * Motion prediction performed for the block by using median prediction of
  459.  * motion vectors from the left, top and right top blocks but in corner cases
  460.  * some other vectors may be used instead.
  461.  */
  462. static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
  463. {
  464.     MpegEncContext *s = &r->s;
  465.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
  466.     int A[2] = {0}, B[2], C[2];
  467.     int i, j;
  468.     int mx, my;
  469.     int* avail = r->avail_cache + avail_indexes[subblock_no];
  470.     int c_off = part_sizes_w[block_type];
  471.  
  472.     mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
  473.     if(subblock_no == 3)
  474.         c_off = -1;
  475.  
  476.     if(avail[-1]){
  477.         A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
  478.         A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
  479.     }
  480.     if(avail[-4]){
  481.         B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
  482.         B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
  483.     }else{
  484.         B[0] = A[0];
  485.         B[1] = A[1];
  486.     }
  487.     if(!avail[c_off-4]){
  488.         if(avail[-4] && (avail[-1] || r->rv30)){
  489.             C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
  490.             C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
  491.         }else{
  492.             C[0] = A[0];
  493.             C[1] = A[1];
  494.         }
  495.     }else{
  496.         C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
  497.         C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
  498.     }
  499.     mx = mid_pred(A[0], B[0], C[0]);
  500.     my = mid_pred(A[1], B[1], C[1]);
  501.     mx += r->dmv[dmv_no][0];
  502.     my += r->dmv[dmv_no][1];
  503.     for(j = 0; j < part_sizes_h[block_type]; j++){
  504.         for(i = 0; i < part_sizes_w[block_type]; i++){
  505.             s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
  506.             s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
  507.         }
  508.     }
  509. }
  510.  
  511. #define GET_PTS_DIFF(a, b) ((a - b + 8192) & 0x1FFF)
  512.  
  513. /**
  514.  * Calculate motion vector component that should be added for direct blocks.
  515.  */
  516. static int calc_add_mv(RV34DecContext *r, int dir, int val)
  517. {
  518.     int mul = dir ? -r->mv_weight2 : r->mv_weight1;
  519.  
  520.     return (val * mul + 0x2000) >> 14;
  521. }
  522.  
  523. /**
  524.  * Predict motion vector for B-frame macroblock.
  525.  */
  526. static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
  527.                                       int A_avail, int B_avail, int C_avail,
  528.                                       int *mx, int *my)
  529. {
  530.     if(A_avail + B_avail + C_avail != 3){
  531.         *mx = A[0] + B[0] + C[0];
  532.         *my = A[1] + B[1] + C[1];
  533.         if(A_avail + B_avail + C_avail == 2){
  534.             *mx /= 2;
  535.             *my /= 2;
  536.         }
  537.     }else{
  538.         *mx = mid_pred(A[0], B[0], C[0]);
  539.         *my = mid_pred(A[1], B[1], C[1]);
  540.     }
  541. }
  542.  
  543. /**
  544.  * motion vector prediction for B-frames
  545.  */
  546. static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
  547. {
  548.     MpegEncContext *s = &r->s;
  549.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
  550.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
  551.     int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 };
  552.     int has_A = 0, has_B = 0, has_C = 0;
  553.     int mx, my;
  554.     int i, j;
  555.     Picture *cur_pic = s->current_picture_ptr;
  556.     const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
  557.     int type = cur_pic->mb_type[mb_pos];
  558.  
  559.     if((r->avail_cache[6-1] & type) & mask){
  560.         A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
  561.         A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
  562.         has_A = 1;
  563.     }
  564.     if((r->avail_cache[6-4] & type) & mask){
  565.         B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
  566.         B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
  567.         has_B = 1;
  568.     }
  569.     if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
  570.         C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
  571.         C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
  572.         has_C = 1;
  573.     }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
  574.         C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
  575.         C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
  576.         has_C = 1;
  577.     }
  578.  
  579.     rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
  580.  
  581.     mx += r->dmv[dir][0];
  582.     my += r->dmv[dir][1];
  583.  
  584.     for(j = 0; j < 2; j++){
  585.         for(i = 0; i < 2; i++){
  586.             cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
  587.             cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
  588.         }
  589.     }
  590.     if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
  591.         ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
  592.     }
  593. }
  594.  
  595. /**
  596.  * motion vector prediction - RV3 version
  597.  */
  598. static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
  599. {
  600.     MpegEncContext *s = &r->s;
  601.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
  602.     int A[2] = {0}, B[2], C[2];
  603.     int i, j, k;
  604.     int mx, my;
  605.     int* avail = r->avail_cache + avail_indexes[0];
  606.  
  607.     if(avail[-1]){
  608.         A[0] = s->current_picture_ptr->motion_val[0][mv_pos - 1][0];
  609.         A[1] = s->current_picture_ptr->motion_val[0][mv_pos - 1][1];
  610.     }
  611.     if(avail[-4]){
  612.         B[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][0];
  613.         B[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][1];
  614.     }else{
  615.         B[0] = A[0];
  616.         B[1] = A[1];
  617.     }
  618.     if(!avail[-4 + 2]){
  619.         if(avail[-4] && (avail[-1])){
  620.             C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][0];
  621.             C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][1];
  622.         }else{
  623.             C[0] = A[0];
  624.             C[1] = A[1];
  625.         }
  626.     }else{
  627.         C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][0];
  628.         C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][1];
  629.     }
  630.     mx = mid_pred(A[0], B[0], C[0]);
  631.     my = mid_pred(A[1], B[1], C[1]);
  632.     mx += r->dmv[0][0];
  633.     my += r->dmv[0][1];
  634.     for(j = 0; j < 2; j++){
  635.         for(i = 0; i < 2; i++){
  636.             for(k = 0; k < 2; k++){
  637.                 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
  638.                 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
  639.             }
  640.         }
  641.     }
  642. }
  643.  
  644. static const int chroma_coeffs[3] = { 0, 3, 5 };
  645.  
  646. /**
  647.  * generic motion compensation function
  648.  *
  649.  * @param r decoder context
  650.  * @param block_type type of the current block
  651.  * @param xoff horizontal offset from the start of the current block
  652.  * @param yoff vertical offset from the start of the current block
  653.  * @param mv_off offset to the motion vector information
  654.  * @param width width of the current partition in 8x8 blocks
  655.  * @param height height of the current partition in 8x8 blocks
  656.  * @param dir motion compensation direction (i.e. from the last or the next reference frame)
  657.  * @param thirdpel motion vectors are specified in 1/3 of pixel
  658.  * @param qpel_mc a set of functions used to perform luma motion compensation
  659.  * @param chroma_mc a set of functions used to perform chroma motion compensation
  660.  */
  661. static inline void rv34_mc(RV34DecContext *r, const int block_type,
  662.                           const int xoff, const int yoff, int mv_off,
  663.                           const int width, const int height, int dir,
  664.                           const int thirdpel, int weighted,
  665.                           qpel_mc_func (*qpel_mc)[16],
  666.                           h264_chroma_mc_func (*chroma_mc))
  667. {
  668.     MpegEncContext *s = &r->s;
  669.     uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
  670.     int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
  671.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
  672.     int is16x16 = 1;
  673.  
  674.     if(thirdpel){
  675.         int chroma_mx, chroma_my;
  676.         mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
  677.         my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
  678.         lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
  679.         ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
  680.         chroma_mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
  681.         chroma_my = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
  682.         umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
  683.         umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
  684.         uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
  685.         uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
  686.     }else{
  687.         int cx, cy;
  688.         mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
  689.         my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
  690.         lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
  691.         ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
  692.         cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
  693.         cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
  694.         umx = cx >> 2;
  695.         umy = cy >> 2;
  696.         uvmx = (cx & 3) << 1;
  697.         uvmy = (cy & 3) << 1;
  698.         //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
  699.         if(uvmx == 6 && uvmy == 6)
  700.             uvmx = uvmy = 4;
  701.     }
  702.  
  703.     if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
  704.         /* wait for the referenced mb row to be finished */
  705.         int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
  706.         ThreadFrame *f = dir ? &s->next_picture_ptr->tf : &s->last_picture_ptr->tf;
  707.         ff_thread_await_progress(f, mb_row, 0);
  708.     }
  709.  
  710.     dxy = ly*4 + lx;
  711.     srcY = dir ? s->next_picture_ptr->f.data[0] : s->last_picture_ptr->f.data[0];
  712.     srcU = dir ? s->next_picture_ptr->f.data[1] : s->last_picture_ptr->f.data[1];
  713.     srcV = dir ? s->next_picture_ptr->f.data[2] : s->last_picture_ptr->f.data[2];
  714.     src_x = s->mb_x * 16 + xoff + mx;
  715.     src_y = s->mb_y * 16 + yoff + my;
  716.     uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
  717.     uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
  718.     srcY += src_y * s->linesize + src_x;
  719.     srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
  720.     srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
  721.     if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
  722.        (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||
  723.        (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {
  724.         uint8_t *uvbuf = s->edge_emu_buffer + 22 * s->linesize;
  725.  
  726.         srcY -= 2 + 2*s->linesize;
  727.         s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, srcY, s->linesize,
  728.                                  (width<<3)+6, (height<<3)+6, src_x - 2, src_y - 2,
  729.                                  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, s->uvlinesize, srcU, s->uvlinesize,
  732.                                  (width<<2)+1, (height<<2)+1, uvsrc_x, uvsrc_y,
  733.                                  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
  734.         s->vdsp.emulated_edge_mc(uvbuf + 16, s->uvlinesize, srcV, s->uvlinesize,
  735.                                  (width<<2)+1, (height<<2)+1, uvsrc_x, uvsrc_y,
  736.                                  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
  737.         srcU = uvbuf;
  738.         srcV = uvbuf + 16;
  739.     }
  740.     if(!weighted){
  741.         Y = s->dest[0] + xoff      + yoff     *s->linesize;
  742.         U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
  743.         V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
  744.     }else{
  745.         Y = r->tmp_b_block_y [dir]     +  xoff     +  yoff    *s->linesize;
  746.         U = r->tmp_b_block_uv[dir*2]   + (xoff>>1) + (yoff>>1)*s->uvlinesize;
  747.         V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
  748.     }
  749.  
  750.     if(block_type == RV34_MB_P_16x8){
  751.         qpel_mc[1][dxy](Y, srcY, s->linesize);
  752.         Y    += 8;
  753.         srcY += 8;
  754.     }else if(block_type == RV34_MB_P_8x16){
  755.         qpel_mc[1][dxy](Y, srcY, s->linesize);
  756.         Y    += 8 * s->linesize;
  757.         srcY += 8 * s->linesize;
  758.     }
  759.     is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
  760.     qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
  761.     chroma_mc[2-width]   (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
  762.     chroma_mc[2-width]   (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
  763. }
  764.  
  765. static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
  766.                         const int xoff, const int yoff, int mv_off,
  767.                         const int width, const int height, int dir)
  768. {
  769.     rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
  770.             r->rdsp.put_pixels_tab,
  771.             r->rdsp.put_chroma_pixels_tab);
  772. }
  773.  
  774. static void rv4_weight(RV34DecContext *r)
  775. {
  776.     r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
  777.                                                         r->tmp_b_block_y[0],
  778.                                                         r->tmp_b_block_y[1],
  779.                                                         r->weight1,
  780.                                                         r->weight2,
  781.                                                         r->s.linesize);
  782.     r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
  783.                                                         r->tmp_b_block_uv[0],
  784.                                                         r->tmp_b_block_uv[2],
  785.                                                         r->weight1,
  786.                                                         r->weight2,
  787.                                                         r->s.uvlinesize);
  788.     r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
  789.                                                         r->tmp_b_block_uv[1],
  790.                                                         r->tmp_b_block_uv[3],
  791.                                                         r->weight1,
  792.                                                         r->weight2,
  793.                                                         r->s.uvlinesize);
  794. }
  795.  
  796. static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
  797. {
  798.     int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
  799.  
  800.     rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
  801.             r->rdsp.put_pixels_tab,
  802.             r->rdsp.put_chroma_pixels_tab);
  803.     if(!weighted){
  804.         rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
  805.                 r->rdsp.avg_pixels_tab,
  806.                 r->rdsp.avg_chroma_pixels_tab);
  807.     }else{
  808.         rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
  809.                 r->rdsp.put_pixels_tab,
  810.                 r->rdsp.put_chroma_pixels_tab);
  811.         rv4_weight(r);
  812.     }
  813. }
  814.  
  815. static void rv34_mc_2mv_skip(RV34DecContext *r)
  816. {
  817.     int i, j;
  818.     int weighted = !r->rv30 && r->weight1 != 8192;
  819.  
  820.     for(j = 0; j < 2; j++)
  821.         for(i = 0; i < 2; i++){
  822.              rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
  823.                      weighted,
  824.                      r->rdsp.put_pixels_tab,
  825.                      r->rdsp.put_chroma_pixels_tab);
  826.              rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
  827.                      weighted,
  828.                      weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
  829.                      weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
  830.         }
  831.     if(weighted)
  832.         rv4_weight(r);
  833. }
  834.  
  835. /** number of motion vectors in each macroblock type */
  836. static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
  837.  
  838. /**
  839.  * Decode motion vector differences
  840.  * and perform motion vector reconstruction and motion compensation.
  841.  */
  842. static int rv34_decode_mv(RV34DecContext *r, int block_type)
  843. {
  844.     MpegEncContext *s = &r->s;
  845.     GetBitContext *gb = &s->gb;
  846.     int i, j, k, l;
  847.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
  848.     int next_bt;
  849.  
  850.     memset(r->dmv, 0, sizeof(r->dmv));
  851.     for(i = 0; i < num_mvs[block_type]; i++){
  852.         r->dmv[i][0] = svq3_get_se_golomb(gb);
  853.         r->dmv[i][1] = svq3_get_se_golomb(gb);
  854.     }
  855.     switch(block_type){
  856.     case RV34_MB_TYPE_INTRA:
  857.     case RV34_MB_TYPE_INTRA16x16:
  858.         ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
  859.         return 0;
  860.     case RV34_MB_SKIP:
  861.         if(s->pict_type == AV_PICTURE_TYPE_P){
  862.             ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
  863.             rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
  864.             break;
  865.         }
  866.     case RV34_MB_B_DIRECT:
  867.         //surprisingly, it uses motion scheme from next reference frame
  868.         /* wait for the current mb row to be finished */
  869.         if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
  870.             ff_thread_await_progress(&s->next_picture_ptr->tf, FFMAX(0, s->mb_y-1), 0);
  871.  
  872.         next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
  873.         if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
  874.             ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
  875.             ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
  876.         }else
  877.             for(j = 0; j < 2; j++)
  878.                 for(i = 0; i < 2; i++)
  879.                     for(k = 0; k < 2; k++)
  880.                         for(l = 0; l < 2; l++)
  881.                             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]);
  882.         if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
  883.             rv34_mc_2mv(r, block_type);
  884.         else
  885.             rv34_mc_2mv_skip(r);
  886.         ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
  887.         break;
  888.     case RV34_MB_P_16x16:
  889.     case RV34_MB_P_MIX16x16:
  890.         rv34_pred_mv(r, block_type, 0, 0);
  891.         rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
  892.         break;
  893.     case RV34_MB_B_FORWARD:
  894.     case RV34_MB_B_BACKWARD:
  895.         r->dmv[1][0] = r->dmv[0][0];
  896.         r->dmv[1][1] = r->dmv[0][1];
  897.         if(r->rv30)
  898.             rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
  899.         else
  900.             rv34_pred_mv_b  (r, block_type, block_type == RV34_MB_B_BACKWARD);
  901.         rv34_mc_1mv     (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
  902.         break;
  903.     case RV34_MB_P_16x8:
  904.     case RV34_MB_P_8x16:
  905.         rv34_pred_mv(r, block_type, 0, 0);
  906.         rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
  907.         if(block_type == RV34_MB_P_16x8){
  908.             rv34_mc_1mv(r, block_type, 0, 0, 0,            2, 1, 0);
  909.             rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
  910.         }
  911.         if(block_type == RV34_MB_P_8x16){
  912.             rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
  913.             rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
  914.         }
  915.         break;
  916.     case RV34_MB_B_BIDIR:
  917.         rv34_pred_mv_b  (r, block_type, 0);
  918.         rv34_pred_mv_b  (r, block_type, 1);
  919.         rv34_mc_2mv     (r, block_type);
  920.         break;
  921.     case RV34_MB_P_8x8:
  922.         for(i=0;i< 4;i++){
  923.             rv34_pred_mv(r, block_type, i, i);
  924.             rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
  925.         }
  926.         break;
  927.     }
  928.  
  929.     return 0;
  930. }
  931. /** @} */ // mv group
  932.  
  933. /**
  934.  * @name Macroblock reconstruction functions
  935.  * @{
  936.  */
  937. /** mapping of RV30/40 intra prediction types to standard H.264 types */
  938. static const int ittrans[9] = {
  939.  DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
  940.  VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
  941. };
  942.  
  943. /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
  944. static const int ittrans16[4] = {
  945.  DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
  946. };
  947.  
  948. /**
  949.  * Perform 4x4 intra prediction.
  950.  */
  951. static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
  952. {
  953.     uint8_t *prev = dst - stride + 4;
  954.     uint32_t topleft;
  955.  
  956.     if(!up && !left)
  957.         itype = DC_128_PRED;
  958.     else if(!up){
  959.         if(itype == VERT_PRED) itype = HOR_PRED;
  960.         if(itype == DC_PRED)   itype = LEFT_DC_PRED;
  961.     }else if(!left){
  962.         if(itype == HOR_PRED)  itype = VERT_PRED;
  963.         if(itype == DC_PRED)   itype = TOP_DC_PRED;
  964.         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
  965.     }
  966.     if(!down){
  967.         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
  968.         if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
  969.         if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
  970.     }
  971.     if(!right && up){
  972.         topleft = dst[-stride + 3] * 0x01010101u;
  973.         prev = (uint8_t*)&topleft;
  974.     }
  975.     r->h.pred4x4[itype](dst, prev, stride);
  976. }
  977.  
  978. static inline int adjust_pred16(int itype, int up, int left)
  979. {
  980.     if(!up && !left)
  981.         itype = DC_128_PRED8x8;
  982.     else if(!up){
  983.         if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
  984.         if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
  985.         if(itype == DC_PRED8x8)   itype = LEFT_DC_PRED8x8;
  986.     }else if(!left){
  987.         if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
  988.         if(itype == HOR_PRED8x8)  itype = VERT_PRED8x8;
  989.         if(itype == DC_PRED8x8)   itype = TOP_DC_PRED8x8;
  990.     }
  991.     return itype;
  992. }
  993.  
  994. static inline void rv34_process_block(RV34DecContext *r,
  995.                                       uint8_t *pdst, int stride,
  996.                                       int fc, int sc, int q_dc, int q_ac)
  997. {
  998.     MpegEncContext *s = &r->s;
  999.     int16_t *ptr = s->block[0];
  1000.     int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
  1001.                                    fc, sc, q_dc, q_ac, q_ac);
  1002.     if(has_ac){
  1003.         r->rdsp.rv34_idct_add(pdst, stride, ptr);
  1004.     }else{
  1005.         r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
  1006.         ptr[0] = 0;
  1007.     }
  1008. }
  1009.  
  1010. static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
  1011. {
  1012.     LOCAL_ALIGNED_16(int16_t, block16, [16]);
  1013.     MpegEncContext *s    = &r->s;
  1014.     GetBitContext  *gb   = &s->gb;
  1015.     int             q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
  1016.                     q_ac = rv34_qscale_tab[s->qscale];
  1017.     uint8_t        *dst  = s->dest[0];
  1018.     int16_t        *ptr  = s->block[0];
  1019.     int i, j, itype, has_ac;
  1020.  
  1021.     memset(block16, 0, 16 * sizeof(*block16));
  1022.  
  1023.     has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
  1024.     if(has_ac)
  1025.         r->rdsp.rv34_inv_transform(block16);
  1026.     else
  1027.         r->rdsp.rv34_inv_transform_dc(block16);
  1028.  
  1029.     itype = ittrans16[intra_types[0]];
  1030.     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
  1031.     r->h.pred16x16[itype](dst, s->linesize);
  1032.  
  1033.     for(j = 0; j < 4; j++){
  1034.         for(i = 0; i < 4; i++, cbp >>= 1){
  1035.             int dc = block16[i + j*4];
  1036.  
  1037.             if(cbp & 1){
  1038.                 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
  1039.             }else
  1040.                 has_ac = 0;
  1041.  
  1042.             if(has_ac){
  1043.                 ptr[0] = dc;
  1044.                 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
  1045.             }else
  1046.                 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
  1047.         }
  1048.  
  1049.         dst += 4*s->linesize;
  1050.     }
  1051.  
  1052.     itype = ittrans16[intra_types[0]];
  1053.     if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
  1054.     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
  1055.  
  1056.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
  1057.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
  1058.  
  1059.     for(j = 1; j < 3; j++){
  1060.         dst = s->dest[j];
  1061.         r->h.pred8x8[itype](dst, s->uvlinesize);
  1062.         for(i = 0; i < 4; i++, cbp >>= 1){
  1063.             uint8_t *pdst;
  1064.             if(!(cbp & 1)) continue;
  1065.             pdst   = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
  1066.  
  1067.             rv34_process_block(r, pdst, s->uvlinesize,
  1068.                                r->chroma_vlc, 1, q_dc, q_ac);
  1069.         }
  1070.     }
  1071. }
  1072.  
  1073. static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
  1074. {
  1075.     MpegEncContext *s   = &r->s;
  1076.     uint8_t        *dst = s->dest[0];
  1077.     int      avail[6*8] = {0};
  1078.     int i, j, k;
  1079.     int idx, q_ac, q_dc;
  1080.  
  1081.     // Set neighbour information.
  1082.     if(r->avail_cache[1])
  1083.         avail[0] = 1;
  1084.     if(r->avail_cache[2])
  1085.         avail[1] = avail[2] = 1;
  1086.     if(r->avail_cache[3])
  1087.         avail[3] = avail[4] = 1;
  1088.     if(r->avail_cache[4])
  1089.         avail[5] = 1;
  1090.     if(r->avail_cache[5])
  1091.         avail[8] = avail[16] = 1;
  1092.     if(r->avail_cache[9])
  1093.         avail[24] = avail[32] = 1;
  1094.  
  1095.     q_ac = rv34_qscale_tab[s->qscale];
  1096.     for(j = 0; j < 4; j++){
  1097.         idx = 9 + j*8;
  1098.         for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
  1099.             rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
  1100.             avail[idx] = 1;
  1101.             if(!(cbp & 1)) continue;
  1102.  
  1103.             rv34_process_block(r, dst, s->linesize,
  1104.                                r->luma_vlc, 0, q_ac, q_ac);
  1105.         }
  1106.         dst += s->linesize * 4 - 4*4;
  1107.         intra_types += r->intra_types_stride;
  1108.     }
  1109.  
  1110.     intra_types -= r->intra_types_stride * 4;
  1111.  
  1112.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
  1113.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
  1114.  
  1115.     for(k = 0; k < 2; k++){
  1116.         dst = s->dest[1+k];
  1117.         fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
  1118.  
  1119.         for(j = 0; j < 2; j++){
  1120.             int* acache = r->avail_cache + 6 + j*4;
  1121.             for(i = 0; i < 2; i++, cbp >>= 1, acache++){
  1122.                 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
  1123.                 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
  1124.                 acache[0] = 1;
  1125.  
  1126.                 if(!(cbp&1)) continue;
  1127.  
  1128.                 rv34_process_block(r, dst + 4*i, s->uvlinesize,
  1129.                                    r->chroma_vlc, 1, q_dc, q_ac);
  1130.             }
  1131.  
  1132.             dst += 4*s->uvlinesize;
  1133.         }
  1134.     }
  1135. }
  1136.  
  1137. static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
  1138. {
  1139.     int d;
  1140.     d = motion_val[0][0] - motion_val[-step][0];
  1141.     if(d < -3 || d > 3)
  1142.         return 1;
  1143.     d = motion_val[0][1] - motion_val[-step][1];
  1144.     if(d < -3 || d > 3)
  1145.         return 1;
  1146.     return 0;
  1147. }
  1148.  
  1149. static int rv34_set_deblock_coef(RV34DecContext *r)
  1150. {
  1151.     MpegEncContext *s = &r->s;
  1152.     int hmvmask = 0, vmvmask = 0, i, j;
  1153.     int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
  1154.     int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
  1155.     for(j = 0; j < 16; j += 8){
  1156.         for(i = 0; i < 2; i++){
  1157.             if(is_mv_diff_gt_3(motion_val + i, 1))
  1158.                 vmvmask |= 0x11 << (j + i*2);
  1159.             if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
  1160.                 hmvmask |= 0x03 << (j + i*2);
  1161.         }
  1162.         motion_val += s->b8_stride;
  1163.     }
  1164.     if(s->first_slice_line)
  1165.         hmvmask &= ~0x000F;
  1166.     if(!s->mb_x)
  1167.         vmvmask &= ~0x1111;
  1168.     if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
  1169.         vmvmask |= (vmvmask & 0x4444) >> 1;
  1170.         hmvmask |= (hmvmask & 0x0F00) >> 4;
  1171.         if(s->mb_x)
  1172.             r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
  1173.         if(!s->first_slice_line)
  1174.             r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
  1175.     }
  1176.     return hmvmask | vmvmask;
  1177. }
  1178.  
  1179. static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
  1180. {
  1181.     MpegEncContext *s   = &r->s;
  1182.     GetBitContext  *gb  = &s->gb;
  1183.     uint8_t        *dst = s->dest[0];
  1184.     int16_t        *ptr = s->block[0];
  1185.     int          mb_pos = s->mb_x + s->mb_y * s->mb_stride;
  1186.     int cbp, cbp2;
  1187.     int q_dc, q_ac, has_ac;
  1188.     int i, j;
  1189.     int dist;
  1190.  
  1191.     // Calculate which neighbours are available. Maybe it's worth optimizing too.
  1192.     memset(r->avail_cache, 0, sizeof(r->avail_cache));
  1193.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
  1194.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
  1195.     if(s->mb_x && dist)
  1196.         r->avail_cache[5] =
  1197.         r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
  1198.     if(dist >= s->mb_width)
  1199.         r->avail_cache[2] =
  1200.         r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
  1201.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
  1202.         r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
  1203.     if(s->mb_x && dist > s->mb_width)
  1204.         r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
  1205.  
  1206.     s->qscale = r->si.quant;
  1207.     cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
  1208.     r->cbp_luma  [mb_pos] = cbp;
  1209.     r->cbp_chroma[mb_pos] = cbp >> 16;
  1210.     r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
  1211.     s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
  1212.  
  1213.     if(cbp == -1)
  1214.         return -1;
  1215.  
  1216.     if (IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
  1217.         if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
  1218.         else        rv34_output_intra(r, intra_types, cbp);
  1219.         return 0;
  1220.     }
  1221.  
  1222.     if(r->is16){
  1223.         // Only for RV34_MB_P_MIX16x16
  1224.         LOCAL_ALIGNED_16(int16_t, block16, [16]);
  1225.         memset(block16, 0, 16 * sizeof(*block16));
  1226.         q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
  1227.         q_ac = rv34_qscale_tab[s->qscale];
  1228.         if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
  1229.             r->rdsp.rv34_inv_transform(block16);
  1230.         else
  1231.             r->rdsp.rv34_inv_transform_dc(block16);
  1232.  
  1233.         q_ac = rv34_qscale_tab[s->qscale];
  1234.  
  1235.         for(j = 0; j < 4; j++){
  1236.             for(i = 0; i < 4; i++, cbp >>= 1){
  1237.                 int      dc   = block16[i + j*4];
  1238.  
  1239.                 if(cbp & 1){
  1240.                     has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
  1241.                 }else
  1242.                     has_ac = 0;
  1243.  
  1244.                 if(has_ac){
  1245.                     ptr[0] = dc;
  1246.                     r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
  1247.                 }else
  1248.                     r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
  1249.             }
  1250.  
  1251.             dst += 4*s->linesize;
  1252.         }
  1253.  
  1254.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
  1255.     }else{
  1256.         q_ac = rv34_qscale_tab[s->qscale];
  1257.  
  1258.         for(j = 0; j < 4; j++){
  1259.             for(i = 0; i < 4; i++, cbp >>= 1){
  1260.                 if(!(cbp & 1)) continue;
  1261.  
  1262.                 rv34_process_block(r, dst + 4*i, s->linesize,
  1263.                                    r->luma_vlc, 0, q_ac, q_ac);
  1264.             }
  1265.             dst += 4*s->linesize;
  1266.         }
  1267.     }
  1268.  
  1269.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
  1270.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
  1271.  
  1272.     for(j = 1; j < 3; j++){
  1273.         dst = s->dest[j];
  1274.         for(i = 0; i < 4; i++, cbp >>= 1){
  1275.             uint8_t *pdst;
  1276.             if(!(cbp & 1)) continue;
  1277.             pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
  1278.  
  1279.             rv34_process_block(r, pdst, s->uvlinesize,
  1280.                                r->chroma_vlc, 1, q_dc, q_ac);
  1281.         }
  1282.     }
  1283.  
  1284.     return 0;
  1285. }
  1286.  
  1287. static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
  1288. {
  1289.     MpegEncContext *s = &r->s;
  1290.     int cbp, dist;
  1291.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
  1292.  
  1293.     // Calculate which neighbours are available. Maybe it's worth optimizing too.
  1294.     memset(r->avail_cache, 0, sizeof(r->avail_cache));
  1295.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
  1296.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
  1297.     if(s->mb_x && dist)
  1298.         r->avail_cache[5] =
  1299.         r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
  1300.     if(dist >= s->mb_width)
  1301.         r->avail_cache[2] =
  1302.         r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
  1303.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
  1304.         r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
  1305.     if(s->mb_x && dist > s->mb_width)
  1306.         r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
  1307.  
  1308.     s->qscale = r->si.quant;
  1309.     cbp = rv34_decode_intra_mb_header(r, intra_types);
  1310.     r->cbp_luma  [mb_pos] = cbp;
  1311.     r->cbp_chroma[mb_pos] = cbp >> 16;
  1312.     r->deblock_coefs[mb_pos] = 0xFFFF;
  1313.     s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
  1314.  
  1315.     if(cbp == -1)
  1316.         return -1;
  1317.  
  1318.     if(r->is16){
  1319.         rv34_output_i16x16(r, intra_types, cbp);
  1320.         return 0;
  1321.     }
  1322.  
  1323.     rv34_output_intra(r, intra_types, cbp);
  1324.     return 0;
  1325. }
  1326.  
  1327. static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
  1328. {
  1329.     int bits;
  1330.     if(s->mb_y >= s->mb_height)
  1331.         return 1;
  1332.     if(!s->mb_num_left)
  1333.         return 1;
  1334.     if(r->s.mb_skip_run > 1)
  1335.         return 0;
  1336.     bits = get_bits_left(&s->gb);
  1337.     if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
  1338.         return 1;
  1339.     return 0;
  1340. }
  1341.  
  1342.  
  1343. static void rv34_decoder_free(RV34DecContext *r)
  1344. {
  1345.     av_freep(&r->intra_types_hist);
  1346.     r->intra_types = NULL;
  1347.     av_freep(&r->tmp_b_block_base);
  1348.     av_freep(&r->mb_type);
  1349.     av_freep(&r->cbp_luma);
  1350.     av_freep(&r->cbp_chroma);
  1351.     av_freep(&r->deblock_coefs);
  1352. }
  1353.  
  1354.  
  1355. static int rv34_decoder_alloc(RV34DecContext *r)
  1356. {
  1357.     r->intra_types_stride = r->s.mb_width * 4 + 4;
  1358.  
  1359.     r->cbp_chroma       = av_malloc(r->s.mb_stride * r->s.mb_height *
  1360.                                     sizeof(*r->cbp_chroma));
  1361.     r->cbp_luma         = av_malloc(r->s.mb_stride * r->s.mb_height *
  1362.                                     sizeof(*r->cbp_luma));
  1363.     r->deblock_coefs    = av_malloc(r->s.mb_stride * r->s.mb_height *
  1364.                                     sizeof(*r->deblock_coefs));
  1365.     r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
  1366.                                     sizeof(*r->intra_types_hist));
  1367.     r->mb_type          = av_mallocz(r->s.mb_stride * r->s.mb_height *
  1368.                                      sizeof(*r->mb_type));
  1369.  
  1370.     if (!(r->cbp_chroma       && r->cbp_luma && r->deblock_coefs &&
  1371.           r->intra_types_hist && r->mb_type)) {
  1372.         rv34_decoder_free(r);
  1373.         return AVERROR(ENOMEM);
  1374.     }
  1375.  
  1376.     r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
  1377.  
  1378.     return 0;
  1379. }
  1380.  
  1381.  
  1382. static int rv34_decoder_realloc(RV34DecContext *r)
  1383. {
  1384.     rv34_decoder_free(r);
  1385.     return rv34_decoder_alloc(r);
  1386. }
  1387.  
  1388.  
  1389. static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
  1390. {
  1391.     MpegEncContext *s = &r->s;
  1392.     GetBitContext *gb = &s->gb;
  1393.     int mb_pos, slice_type;
  1394.     int res;
  1395.  
  1396.     init_get_bits(&r->s.gb, buf, buf_size*8);
  1397.     res = r->parse_slice_header(r, gb, &r->si);
  1398.     if(res < 0){
  1399.         av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
  1400.         return -1;
  1401.     }
  1402.  
  1403.     slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
  1404.     if (slice_type != s->pict_type) {
  1405.         av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
  1406.         return AVERROR_INVALIDDATA;
  1407.     }
  1408.     if (s->width != r->si.width || s->height != r->si.height) {
  1409.         av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
  1410.         return AVERROR_INVALIDDATA;
  1411.     }
  1412.  
  1413.     r->si.end = end;
  1414.     s->qscale = r->si.quant;
  1415.     s->mb_num_left = r->si.end - r->si.start;
  1416.     r->s.mb_skip_run = 0;
  1417.  
  1418.     mb_pos = s->mb_x + s->mb_y * s->mb_width;
  1419.     if(r->si.start != mb_pos){
  1420.         av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
  1421.         s->mb_x = r->si.start % s->mb_width;
  1422.         s->mb_y = r->si.start / s->mb_width;
  1423.     }
  1424.     memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
  1425.     s->first_slice_line = 1;
  1426.     s->resync_mb_x = s->mb_x;
  1427.     s->resync_mb_y = s->mb_y;
  1428.  
  1429.     ff_init_block_index(s);
  1430.     while(!check_slice_end(r, s)) {
  1431.         ff_update_block_index(s);
  1432.  
  1433.         if(r->si.type)
  1434.             res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
  1435.         else
  1436.             res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
  1437.         if(res < 0){
  1438.             ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
  1439.             return -1;
  1440.         }
  1441.         if (++s->mb_x == s->mb_width) {
  1442.             s->mb_x = 0;
  1443.             s->mb_y++;
  1444.             ff_init_block_index(s);
  1445.  
  1446.             memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
  1447.             memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
  1448.  
  1449.             if(r->loop_filter && s->mb_y >= 2)
  1450.                 r->loop_filter(r, s->mb_y - 2);
  1451.  
  1452.             if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
  1453.                 ff_thread_report_progress(&s->current_picture_ptr->tf,
  1454.                                           s->mb_y - 2, 0);
  1455.  
  1456.         }
  1457.         if(s->mb_x == s->resync_mb_x)
  1458.             s->first_slice_line=0;
  1459.         s->mb_num_left--;
  1460.     }
  1461.     ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
  1462.  
  1463.     return s->mb_y == s->mb_height;
  1464. }
  1465.  
  1466. /** @} */ // recons group end
  1467.  
  1468. /**
  1469.  * Initialize decoder.
  1470.  */
  1471. av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
  1472. {
  1473.     RV34DecContext *r = avctx->priv_data;
  1474.     MpegEncContext *s = &r->s;
  1475.     int ret;
  1476.  
  1477.     ff_MPV_decode_defaults(s);
  1478.     s->avctx      = avctx;
  1479.     s->out_format = FMT_H263;
  1480.     s->codec_id   = avctx->codec_id;
  1481.  
  1482.     s->width  = avctx->width;
  1483.     s->height = avctx->height;
  1484.  
  1485.     r->s.avctx = avctx;
  1486.     avctx->flags |= CODEC_FLAG_EMU_EDGE;
  1487.     r->s.flags |= CODEC_FLAG_EMU_EDGE;
  1488.     avctx->pix_fmt = AV_PIX_FMT_YUV420P;
  1489.     avctx->has_b_frames = 1;
  1490.     s->low_delay = 0;
  1491.  
  1492.     if ((ret = ff_MPV_common_init(s)) < 0)
  1493.         return ret;
  1494.  
  1495.     ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
  1496.  
  1497. #if CONFIG_RV30_DECODER
  1498.     if (avctx->codec_id == AV_CODEC_ID_RV30)
  1499.         ff_rv30dsp_init(&r->rdsp);
  1500. #endif
  1501. #if CONFIG_RV40_DECODER
  1502.     if (avctx->codec_id == AV_CODEC_ID_RV40)
  1503.         ff_rv40dsp_init(&r->rdsp);
  1504. #endif
  1505.  
  1506.     if ((ret = rv34_decoder_alloc(r)) < 0) {
  1507.         ff_MPV_common_end(&r->s);
  1508.         return ret;
  1509.     }
  1510.  
  1511.     if(!intra_vlcs[0].cbppattern[0].bits)
  1512.         rv34_init_tables();
  1513.  
  1514.     avctx->internal->allocate_progress = 1;
  1515.  
  1516.     return 0;
  1517. }
  1518.  
  1519. int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx)
  1520. {
  1521.     int err;
  1522.     RV34DecContext *r = avctx->priv_data;
  1523.  
  1524.     r->s.avctx = avctx;
  1525.  
  1526.     if (avctx->internal->is_copy) {
  1527.         r->tmp_b_block_base = NULL;
  1528.         if ((err = ff_MPV_common_init(&r->s)) < 0)
  1529.             return err;
  1530.         if ((err = rv34_decoder_alloc(r)) < 0) {
  1531.             ff_MPV_common_end(&r->s);
  1532.             return err;
  1533.         }
  1534.     }
  1535.  
  1536.     return 0;
  1537. }
  1538.  
  1539. int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
  1540. {
  1541.     RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
  1542.     MpegEncContext * const s = &r->s, * const s1 = &r1->s;
  1543.     int err;
  1544.  
  1545.     if (dst == src || !s1->context_initialized)
  1546.         return 0;
  1547.  
  1548.     if (s->height != s1->height || s->width != s1->width) {
  1549.         s->height = s1->height;
  1550.         s->width  = s1->width;
  1551.         if ((err = ff_MPV_common_frame_size_change(s)) < 0)
  1552.             return err;
  1553.         if ((err = rv34_decoder_realloc(r)) < 0)
  1554.             return err;
  1555.     }
  1556.  
  1557.     if ((err = ff_mpeg_update_thread_context(dst, src)))
  1558.         return err;
  1559.  
  1560.     r->cur_pts  = r1->cur_pts;
  1561.     r->last_pts = r1->last_pts;
  1562.     r->next_pts = r1->next_pts;
  1563.  
  1564.     memset(&r->si, 0, sizeof(r->si));
  1565.  
  1566.     return 0;
  1567. }
  1568.  
  1569. static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
  1570. {
  1571.     if(avctx->slice_count) return avctx->slice_offset[n];
  1572.     else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);
  1573. }
  1574.  
  1575. static int finish_frame(AVCodecContext *avctx, AVFrame *pict)
  1576. {
  1577.     RV34DecContext *r = avctx->priv_data;
  1578.     MpegEncContext *s = &r->s;
  1579.     int got_picture = 0, ret;
  1580.  
  1581.     ff_er_frame_end(&s->er);
  1582.     ff_MPV_frame_end(s);
  1583.     s->mb_num_left = 0;
  1584.  
  1585.     if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
  1586.         ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
  1587.  
  1588.     if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
  1589.         if ((ret = av_frame_ref(pict, &s->current_picture_ptr->f)) < 0)
  1590.             return ret;
  1591.         ff_print_debug_info(s, s->current_picture_ptr, pict);
  1592.         ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1);
  1593.         got_picture = 1;
  1594.     } else if (s->last_picture_ptr != NULL) {
  1595.         if ((ret = av_frame_ref(pict, &s->last_picture_ptr->f)) < 0)
  1596.             return ret;
  1597.         ff_print_debug_info(s, s->last_picture_ptr, pict);
  1598.         ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1);
  1599.         got_picture = 1;
  1600.     }
  1601.  
  1602.     return got_picture;
  1603. }
  1604.  
  1605. static AVRational update_sar(int old_w, int old_h, AVRational sar, int new_w, int new_h)
  1606. {
  1607.     // attempt to keep aspect during typical resolution switches
  1608.     if (!sar.num)
  1609.         sar = (AVRational){1, 1};
  1610.  
  1611.     sar = av_mul_q(sar, (AVRational){new_h * old_w, new_w * old_h});
  1612.     return sar;
  1613. }
  1614.  
  1615. int ff_rv34_decode_frame(AVCodecContext *avctx,
  1616.                             void *data, int *got_picture_ptr,
  1617.                             AVPacket *avpkt)
  1618. {
  1619.     const uint8_t *buf = avpkt->data;
  1620.     int buf_size = avpkt->size;
  1621.     RV34DecContext *r = avctx->priv_data;
  1622.     MpegEncContext *s = &r->s;
  1623.     AVFrame *pict = data;
  1624.     SliceInfo si;
  1625.     int i, ret;
  1626.     int slice_count;
  1627.     const uint8_t *slices_hdr = NULL;
  1628.     int last = 0;
  1629.  
  1630.     /* no supplementary picture */
  1631.     if (buf_size == 0) {
  1632.         /* special case for last picture */
  1633.         if (s->low_delay==0 && s->next_picture_ptr) {
  1634.             if ((ret = av_frame_ref(pict, &s->next_picture_ptr->f)) < 0)
  1635.                 return ret;
  1636.             s->next_picture_ptr = NULL;
  1637.  
  1638.             *got_picture_ptr = 1;
  1639.         }
  1640.         return 0;
  1641.     }
  1642.  
  1643.     if(!avctx->slice_count){
  1644.         slice_count = (*buf++) + 1;
  1645.         slices_hdr = buf + 4;
  1646.         buf += 8 * slice_count;
  1647.         buf_size -= 1 + 8 * slice_count;
  1648.     }else
  1649.         slice_count = avctx->slice_count;
  1650.  
  1651.     //parse first slice header to check whether this frame can be decoded
  1652.     if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||
  1653.        get_slice_offset(avctx, slices_hdr, 0) > buf_size){
  1654.         av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
  1655.         return AVERROR_INVALIDDATA;
  1656.     }
  1657.     init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);
  1658.     if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
  1659.         av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
  1660.         return AVERROR_INVALIDDATA;
  1661.     }
  1662.     if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) &&
  1663.         si.type == AV_PICTURE_TYPE_B) {
  1664.         av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
  1665.                "reference data.\n");
  1666.         return AVERROR_INVALIDDATA;
  1667.     }
  1668.     if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
  1669.        || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
  1670.        ||  avctx->skip_frame >= AVDISCARD_ALL)
  1671.         return avpkt->size;
  1672.  
  1673.     /* first slice */
  1674.     if (si.start == 0) {
  1675.         if (s->mb_num_left > 0) {
  1676.             av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n",
  1677.                    s->mb_num_left);
  1678.             ff_er_frame_end(&s->er);
  1679.             ff_MPV_frame_end(s);
  1680.         }
  1681.  
  1682.         if (s->width != si.width || s->height != si.height) {
  1683.             int err;
  1684.  
  1685.             av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
  1686.                    si.width, si.height);
  1687.  
  1688.             if (av_image_check_size(si.width, si.height, 0, s->avctx))
  1689.                 return AVERROR_INVALIDDATA;
  1690.  
  1691.             s->avctx->sample_aspect_ratio = update_sar(
  1692.                 s->width, s->height, s->avctx->sample_aspect_ratio,
  1693.                 si.width, si.height);
  1694.             s->width  = si.width;
  1695.             s->height = si.height;
  1696.             avcodec_set_dimensions(s->avctx, s->width, s->height);
  1697.             if ((err = ff_MPV_common_frame_size_change(s)) < 0)
  1698.                 return err;
  1699.             if ((err = rv34_decoder_realloc(r)) < 0)
  1700.                 return err;
  1701.         }
  1702.         s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
  1703.         if (ff_MPV_frame_start(s, s->avctx) < 0)
  1704.             return -1;
  1705.         ff_mpeg_er_frame_start(s);
  1706.         if (!r->tmp_b_block_base) {
  1707.             int i;
  1708.  
  1709.             r->tmp_b_block_base = av_malloc(s->linesize * 48);
  1710.             for (i = 0; i < 2; i++)
  1711.                 r->tmp_b_block_y[i] = r->tmp_b_block_base
  1712.                                       + i * 16 * s->linesize;
  1713.             for (i = 0; i < 4; i++)
  1714.                 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
  1715.                                        + (i >> 1) * 8 * s->uvlinesize
  1716.                                        + (i &  1) * 16;
  1717.         }
  1718.         r->cur_pts = si.pts;
  1719.         if (s->pict_type != AV_PICTURE_TYPE_B) {
  1720.             r->last_pts = r->next_pts;
  1721.             r->next_pts = r->cur_pts;
  1722.         } else {
  1723.             int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
  1724.             int dist0   = GET_PTS_DIFF(r->cur_pts,  r->last_pts);
  1725.             int dist1   = GET_PTS_DIFF(r->next_pts, r->cur_pts);
  1726.  
  1727.             if(!refdist){
  1728.                 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
  1729.                 r->scaled_weight = 0;
  1730.             }else{
  1731.                 r->mv_weight1 = (dist0 << 14) / refdist;
  1732.                 r->mv_weight2 = (dist1 << 14) / refdist;
  1733.                 if((r->mv_weight1|r->mv_weight2) & 511){
  1734.                     r->weight1 = r->mv_weight1;
  1735.                     r->weight2 = r->mv_weight2;
  1736.                     r->scaled_weight = 0;
  1737.                 }else{
  1738.                     r->weight1 = r->mv_weight1 >> 9;
  1739.                     r->weight2 = r->mv_weight2 >> 9;
  1740.                     r->scaled_weight = 1;
  1741.                 }
  1742.             }
  1743.         }
  1744.         s->mb_x = s->mb_y = 0;
  1745.         ff_thread_finish_setup(s->avctx);
  1746.     } else if (HAVE_THREADS &&
  1747.                (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
  1748.         av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
  1749.                "multithreading mode (start MB is %d).\n", si.start);
  1750.         return AVERROR_INVALIDDATA;
  1751.     }
  1752.  
  1753.     for(i = 0; i < slice_count; i++){
  1754.         int offset = get_slice_offset(avctx, slices_hdr, i);
  1755.         int size;
  1756.         if(i+1 == slice_count)
  1757.             size = buf_size - offset;
  1758.         else
  1759.             size = get_slice_offset(avctx, slices_hdr, i+1) - offset;
  1760.  
  1761.         if(offset < 0 || offset > buf_size){
  1762.             av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
  1763.             break;
  1764.         }
  1765.  
  1766.         r->si.end = s->mb_width * s->mb_height;
  1767.         s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
  1768.  
  1769.         if(i+1 < slice_count){
  1770.             if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||
  1771.                 get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {
  1772.                 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
  1773.                 break;
  1774.             }
  1775.             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);
  1776.             if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
  1777.                 if(i+2 < slice_count)
  1778.                     size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
  1779.                 else
  1780.                     size = buf_size - offset;
  1781.             }else
  1782.                 r->si.end = si.start;
  1783.         }
  1784.         if (size < 0 || size > buf_size - offset) {
  1785.             av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");
  1786.             break;
  1787.         }
  1788.         last = rv34_decode_slice(r, r->si.end, buf + offset, size);
  1789.         if(last)
  1790.             break;
  1791.     }
  1792.  
  1793.     if (s->current_picture_ptr) {
  1794.         if (last) {
  1795.             if(r->loop_filter)
  1796.                 r->loop_filter(r, s->mb_height - 1);
  1797.  
  1798.             ret = finish_frame(avctx, pict);
  1799.             if (ret < 0)
  1800.                 return ret;
  1801.             *got_picture_ptr = ret;
  1802.         } else if (HAVE_THREADS &&
  1803.                    (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
  1804.             av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
  1805.             /* always mark the current frame as finished, frame-mt supports
  1806.              * only complete frames */
  1807.             ff_er_frame_end(&s->er);
  1808.             ff_MPV_frame_end(s);
  1809.             s->mb_num_left = 0;
  1810.             ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
  1811.             return AVERROR_INVALIDDATA;
  1812.         }
  1813.     }
  1814.  
  1815.     return avpkt->size;
  1816. }
  1817.  
  1818. av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
  1819. {
  1820.     RV34DecContext *r = avctx->priv_data;
  1821.  
  1822.     ff_MPV_common_end(&r->s);
  1823.     rv34_decoder_free(r);
  1824.  
  1825.     return 0;
  1826. }
  1827.