Subversion Repositories Kolibri OS

Rev

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 "mpegutils.h"
  33. #include "mpegvideo.h"
  34. #include "golomb.h"
  35. #include "internal.h"
  36. #include "mathops.h"
  37. #include "mpeg_er.h"
  38. #include "qpeldsp.h"
  39. #include "rectangle.h"
  40. #include "thread.h"
  41.  
  42. #include "rv34vlc.h"
  43. #include "rv34data.h"
  44. #include "rv34.h"
  45.  
  46. static inline void ZERO8x2(void* dst, int stride)
  47. {
  48.     fill_rectangle(dst,                 1, 2, stride, 0, 4);
  49.     fill_rectangle(((uint8_t*)(dst))+4, 1, 2, stride, 0, 4);
  50. }
  51.  
  52. /** translation of RV30/40 macroblock types to lavc ones */
  53. static const int rv34_mb_type_to_lavc[12] = {
  54.     MB_TYPE_INTRA,
  55.     MB_TYPE_INTRA16x16              | MB_TYPE_SEPARATE_DC,
  56.     MB_TYPE_16x16   | MB_TYPE_L0,
  57.     MB_TYPE_8x8     | MB_TYPE_L0,
  58.     MB_TYPE_16x16   | MB_TYPE_L0,
  59.     MB_TYPE_16x16   | MB_TYPE_L1,
  60.     MB_TYPE_SKIP,
  61.     MB_TYPE_DIRECT2 | MB_TYPE_16x16,
  62.     MB_TYPE_16x8    | MB_TYPE_L0,
  63.     MB_TYPE_8x16    | MB_TYPE_L0,
  64.     MB_TYPE_16x16   | MB_TYPE_L0L1,
  65.     MB_TYPE_16x16   | MB_TYPE_L0    | MB_TYPE_SEPARATE_DC
  66. };
  67.  
  68.  
  69. static RV34VLC intra_vlcs[NUM_INTRA_TABLES], inter_vlcs[NUM_INTER_TABLES];
  70.  
  71. static int rv34_decode_mv(RV34DecContext *r, int block_type);
  72.  
  73. /**
  74.  * @name RV30/40 VLC generating functions
  75.  * @{
  76.  */
  77.  
  78. static const int table_offs[] = {
  79.       0,   1818,   3622,   4144,   4698,   5234,   5804,   5868,   5900,   5932,
  80.    5996,   6252,   6316,   6348,   6380,   7674,   8944,  10274,  11668,  12250,
  81.   14060,  15846,  16372,  16962,  17512,  18148,  18180,  18212,  18244,  18308,
  82.   18564,  18628,  18660,  18692,  20036,  21314,  22648,  23968,  24614,  26384,
  83.   28190,  28736,  29366,  29938,  30608,  30640,  30672,  30704,  30768,  31024,
  84.   31088,  31120,  31184,  32570,  33898,  35236,  36644,  37286,  39020,  40802,
  85.   41368,  42052,  42692,  43348,  43380,  43412,  43444,  43476,  43604,  43668,
  86.   43700,  43732,  45100,  46430,  47778,  49160,  49802,  51550,  53340,  53972,
  87.   54648,  55348,  55994,  56122,  56154,  56186,  56218,  56346,  56410,  56442,
  88.   56474,  57878,  59290,  60636,  62036,  62682,  64460,  64524,  64588,  64716,
  89.   64844,  66076,  67466,  67978,  68542,  69064,  69648,  70296,  72010,  72074,
  90.   72138,  72202,  72330,  73572,  74936,  75454,  76030,  76566,  77176,  77822,
  91.   79582,  79646,  79678,  79742,  79870,  81180,  82536,  83064,  83672,  84242,
  92.   84934,  85576,  87384,  87448,  87480,  87544,  87672,  88982,  90340,  90902,
  93.   91598,  92182,  92846,  93488,  95246,  95278,  95310,  95374,  95502,  96878,
  94.   98266,  98848,  99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,
  95.  103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,
  96.  111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592
  97. };
  98.  
  99. static VLC_TYPE table_data[117592][2];
  100.  
  101. /**
  102.  * Generate VLC from codeword lengths.
  103.  * @param bits   codeword lengths (zeroes are accepted)
  104.  * @param size   length of input data
  105.  * @param vlc    output VLC
  106.  * @param insyms symbols for input codes (NULL for default ones)
  107.  * @param num    VLC table number (for static initialization)
  108.  */
  109. static void rv34_gen_vlc(const uint8_t *bits, int size, VLC *vlc, const uint8_t *insyms,
  110.                          const int num)
  111. {
  112.     int i;
  113.     int counts[17] = {0}, codes[17];
  114.     uint16_t cw[MAX_VLC_SIZE], syms[MAX_VLC_SIZE];
  115.     uint8_t bits2[MAX_VLC_SIZE];
  116.     int maxbits = 0, realsize = 0;
  117.  
  118.     for(i = 0; i < size; i++){
  119.         if(bits[i]){
  120.             bits2[realsize] = bits[i];
  121.             syms[realsize] = insyms ? insyms[i] : i;
  122.             realsize++;
  123.             maxbits = FFMAX(maxbits, bits[i]);
  124.             counts[bits[i]]++;
  125.         }
  126.     }
  127.  
  128.     codes[0] = 0;
  129.     for(i = 0; i < 16; i++)
  130.         codes[i+1] = (codes[i] + counts[i]) << 1;
  131.     for(i = 0; i < realsize; i++)
  132.         cw[i] = codes[bits2[i]]++;
  133.  
  134.     vlc->table = &table_data[table_offs[num]];
  135.     vlc->table_allocated = table_offs[num + 1] - table_offs[num];
  136.     ff_init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,
  137.                        bits2, 1, 1,
  138.                        cw,    2, 2,
  139.                        syms,  2, 2, INIT_VLC_USE_NEW_STATIC);
  140. }
  141.  
  142. /**
  143.  * Initialize all tables.
  144.  */
  145. static av_cold void rv34_init_tables(void)
  146. {
  147.     int i, j, k;
  148.  
  149.     for(i = 0; i < NUM_INTRA_TABLES; i++){
  150.         for(j = 0; j < 2; j++){
  151.             rv34_gen_vlc(rv34_table_intra_cbppat   [i][j], CBPPAT_VLC_SIZE,   &intra_vlcs[i].cbppattern[j],     NULL, 19*i + 0 + j);
  152.             rv34_gen_vlc(rv34_table_intra_secondpat[i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].second_pattern[j], NULL, 19*i + 2 + j);
  153.             rv34_gen_vlc(rv34_table_intra_thirdpat [i][j], OTHERBLK_VLC_SIZE, &intra_vlcs[i].third_pattern[j],  NULL, 19*i + 4 + j);
  154.             for(k = 0; k < 4; k++){
  155.                 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);
  156.             }
  157.         }
  158.         for(j = 0; j < 4; j++){
  159.             rv34_gen_vlc(rv34_table_intra_firstpat[i][j], FIRSTBLK_VLC_SIZE, &intra_vlcs[i].first_pattern[j], NULL, 19*i + 14 + j);
  160.         }
  161.         rv34_gen_vlc(rv34_intra_coeff[i], COEFF_VLC_SIZE, &intra_vlcs[i].coefficient, NULL, 19*i + 18);
  162.     }
  163.  
  164.     for(i = 0; i < NUM_INTER_TABLES; i++){
  165.         rv34_gen_vlc(rv34_inter_cbppat[i], CBPPAT_VLC_SIZE, &inter_vlcs[i].cbppattern[0], NULL, i*12 + 95);
  166.         for(j = 0; j < 4; j++){
  167.             rv34_gen_vlc(rv34_inter_cbp[i][j], CBP_VLC_SIZE, &inter_vlcs[i].cbp[0][j], rv34_cbp_code, i*12 + 96 + j);
  168.         }
  169.         for(j = 0; j < 2; j++){
  170.             rv34_gen_vlc(rv34_table_inter_firstpat [i][j], FIRSTBLK_VLC_SIZE, &inter_vlcs[i].first_pattern[j],  NULL, i*12 + 100 + j);
  171.             rv34_gen_vlc(rv34_table_inter_secondpat[i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].second_pattern[j], NULL, i*12 + 102 + j);
  172.             rv34_gen_vlc(rv34_table_inter_thirdpat [i][j], OTHERBLK_VLC_SIZE, &inter_vlcs[i].third_pattern[j],  NULL, i*12 + 104 + j);
  173.         }
  174.         rv34_gen_vlc(rv34_inter_coeff[i], COEFF_VLC_SIZE, &inter_vlcs[i].coefficient, NULL, i*12 + 106);
  175.     }
  176. }
  177.  
  178. /** @} */ // vlc group
  179.  
  180. /**
  181.  * @name RV30/40 4x4 block decoding functions
  182.  * @{
  183.  */
  184.  
  185. /**
  186.  * Decode coded block pattern.
  187.  */
  188. static int rv34_decode_cbp(GetBitContext *gb, RV34VLC *vlc, int table)
  189. {
  190.     int pattern, code, cbp=0;
  191.     int ones;
  192.     static const int cbp_masks[3] = {0x100000, 0x010000, 0x110000};
  193.     static const int shifts[4] = { 0, 2, 8, 10 };
  194.     const int *curshift = shifts;
  195.     int i, t, mask;
  196.  
  197.     code = get_vlc2(gb, vlc->cbppattern[table].table, 9, 2);
  198.     pattern = code & 0xF;
  199.     code >>= 4;
  200.  
  201.     ones = rv34_count_ones[pattern];
  202.  
  203.     for(mask = 8; mask; mask >>= 1, curshift++){
  204.         if(pattern & mask)
  205.             cbp |= get_vlc2(gb, vlc->cbp[table][ones].table, vlc->cbp[table][ones].bits, 1) << curshift[0];
  206.     }
  207.  
  208.     for(i = 0; i < 4; i++){
  209.         t = (modulo_three_table[code] >> (6 - 2*i)) & 3;
  210.         if(t == 1)
  211.             cbp |= cbp_masks[get_bits1(gb)] << i;
  212.         if(t == 2)
  213.             cbp |= cbp_masks[2] << i;
  214.     }
  215.     return cbp;
  216. }
  217.  
  218. /**
  219.  * Get one coefficient value from the bitstream and store it.
  220.  */
  221. static inline void decode_coeff(int16_t *dst, int coef, int esc, GetBitContext *gb, VLC* vlc, int q)
  222. {
  223.     if(coef){
  224.         if(coef == esc){
  225.             coef = get_vlc2(gb, vlc->table, 9, 2);
  226.             if(coef > 23){
  227.                 coef -= 23;
  228.                 coef = 22 + ((1 << coef) | get_bits(gb, coef));
  229.             }
  230.             coef += esc;
  231.         }
  232.         if(get_bits1(gb))
  233.             coef = -coef;
  234.         *dst = (coef*q + 8) >> 4;
  235.     }
  236. }
  237.  
  238. /**
  239.  * Decode 2x2 subblock of coefficients.
  240.  */
  241. static inline void decode_subblock(int16_t *dst, int code, const int is_block2, GetBitContext *gb, VLC *vlc, int q)
  242. {
  243.     int flags = modulo_three_table[code];
  244.  
  245.     decode_coeff(    dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q);
  246.     if(is_block2){
  247.         decode_coeff(dst+1*4+0, (flags >> 4) & 3, 2, gb, vlc, q);
  248.         decode_coeff(dst+0*4+1, (flags >> 2) & 3, 2, gb, vlc, q);
  249.     }else{
  250.         decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q);
  251.         decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q);
  252.     }
  253.     decode_coeff(    dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q);
  254. }
  255.  
  256. /**
  257.  * Decode a single coefficient.
  258.  */
  259. static inline void decode_subblock1(int16_t *dst, int code, GetBitContext *gb, VLC *vlc, int q)
  260. {
  261.     int coeff = modulo_three_table[code] >> 6;
  262.     decode_coeff(dst, coeff, 3, gb, vlc, q);
  263. }
  264.  
  265. static inline void decode_subblock3(int16_t *dst, int code, GetBitContext *gb, VLC *vlc,
  266.                                     int q_dc, int q_ac1, int q_ac2)
  267. {
  268.     int flags = modulo_three_table[code];
  269.  
  270.     decode_coeff(dst+0*4+0, (flags >> 6)    , 3, gb, vlc, q_dc);
  271.     decode_coeff(dst+0*4+1, (flags >> 4) & 3, 2, gb, vlc, q_ac1);
  272.     decode_coeff(dst+1*4+0, (flags >> 2) & 3, 2, gb, vlc, q_ac1);
  273.     decode_coeff(dst+1*4+1, (flags >> 0) & 3, 2, gb, vlc, q_ac2);
  274. }
  275.  
  276. /**
  277.  * Decode coefficients for 4x4 block.
  278.  *
  279.  * This is done by filling 2x2 subblocks with decoded coefficients
  280.  * in this order (the same for subblocks and subblock coefficients):
  281.  *  o--o
  282.  *    /
  283.  *   /
  284.  *  o--o
  285.  */
  286.  
  287. 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)
  288. {
  289.     int code, pattern, has_ac = 1;
  290.  
  291.     code = get_vlc2(gb, rvlc->first_pattern[fc].table, 9, 2);
  292.  
  293.     pattern = code & 0x7;
  294.  
  295.     code >>= 3;
  296.  
  297.     if (modulo_three_table[code] & 0x3F) {
  298.         decode_subblock3(dst, code, gb, &rvlc->coefficient, q_dc, q_ac1, q_ac2);
  299.     } else {
  300.         decode_subblock1(dst, code, gb, &rvlc->coefficient, q_dc);
  301.         if (!pattern)
  302.             return 0;
  303.         has_ac = 0;
  304.     }
  305.  
  306.     if(pattern & 4){
  307.         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
  308.         decode_subblock(dst + 4*0+2, code, 0, gb, &rvlc->coefficient, q_ac2);
  309.     }
  310.     if(pattern & 2){ // Looks like coefficients 1 and 2 are swapped for this block
  311.         code = get_vlc2(gb, rvlc->second_pattern[sc].table, 9, 2);
  312.         decode_subblock(dst + 4*2+0, code, 1, gb, &rvlc->coefficient, q_ac2);
  313.     }
  314.     if(pattern & 1){
  315.         code = get_vlc2(gb, rvlc->third_pattern[sc].table, 9, 2);
  316.         decode_subblock(dst + 4*2+2, code, 0, gb, &rvlc->coefficient, q_ac2);
  317.     }
  318.     return has_ac | pattern;
  319. }
  320.  
  321. /**
  322.  * @name RV30/40 bitstream parsing
  323.  * @{
  324.  */
  325.  
  326. /**
  327.  * Decode starting slice position.
  328.  * @todo Maybe replace with ff_h263_decode_mba() ?
  329.  */
  330. int ff_rv34_get_start_offset(GetBitContext *gb, int mb_size)
  331. {
  332.     int i;
  333.     for(i = 0; i < 5; i++)
  334.         if(rv34_mb_max_sizes[i] >= mb_size - 1)
  335.             break;
  336.     return rv34_mb_bits_sizes[i];
  337. }
  338.  
  339. /**
  340.  * Select VLC set for decoding from current quantizer, modifier and frame type.
  341.  */
  342. static inline RV34VLC* choose_vlc_set(int quant, int mod, int type)
  343. {
  344.     if(mod == 2 && quant < 19) quant += 10;
  345.     else if(mod && quant < 26) quant += 5;
  346.     return type ? &inter_vlcs[rv34_quant_to_vlc_set[1][av_clip(quant, 0, 30)]]
  347.                 : &intra_vlcs[rv34_quant_to_vlc_set[0][av_clip(quant, 0, 30)]];
  348. }
  349.  
  350. /**
  351.  * Decode intra macroblock header and return CBP in case of success, -1 otherwise.
  352.  */
  353. static int rv34_decode_intra_mb_header(RV34DecContext *r, int8_t *intra_types)
  354. {
  355.     MpegEncContext *s = &r->s;
  356.     GetBitContext *gb = &s->gb;
  357.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
  358.     int t;
  359.  
  360.     r->is16 = get_bits1(gb);
  361.     if(r->is16){
  362.         s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA16x16;
  363.         r->block_type = RV34_MB_TYPE_INTRA16x16;
  364.         t = get_bits(gb, 2);
  365.         fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
  366.         r->luma_vlc   = 2;
  367.     }else{
  368.         if(!r->rv30){
  369.             if(!get_bits1(gb))
  370.                 av_log(s->avctx, AV_LOG_ERROR, "Need DQUANT\n");
  371.         }
  372.         s->current_picture_ptr->mb_type[mb_pos] = MB_TYPE_INTRA;
  373.         r->block_type = RV34_MB_TYPE_INTRA;
  374.         if(r->decode_intra_types(r, gb, intra_types) < 0)
  375.             return -1;
  376.         r->luma_vlc   = 1;
  377.     }
  378.  
  379.     r->chroma_vlc = 0;
  380.     r->cur_vlcs   = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
  381.  
  382.     return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
  383. }
  384.  
  385. /**
  386.  * Decode inter macroblock header and return CBP in case of success, -1 otherwise.
  387.  */
  388. static int rv34_decode_inter_mb_header(RV34DecContext *r, int8_t *intra_types)
  389. {
  390.     MpegEncContext *s = &r->s;
  391.     GetBitContext *gb = &s->gb;
  392.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
  393.     int i, t;
  394.  
  395.     r->block_type = r->decode_mb_info(r);
  396.     if(r->block_type == -1)
  397.         return -1;
  398.     s->current_picture_ptr->mb_type[mb_pos] = rv34_mb_type_to_lavc[r->block_type];
  399.     r->mb_type[mb_pos] = r->block_type;
  400.     if(r->block_type == RV34_MB_SKIP){
  401.         if(s->pict_type == AV_PICTURE_TYPE_P)
  402.             r->mb_type[mb_pos] = RV34_MB_P_16x16;
  403.         if(s->pict_type == AV_PICTURE_TYPE_B)
  404.             r->mb_type[mb_pos] = RV34_MB_B_DIRECT;
  405.     }
  406.     r->is16 = !!IS_INTRA16x16(s->current_picture_ptr->mb_type[mb_pos]);
  407.     rv34_decode_mv(r, r->block_type);
  408.     if(r->block_type == RV34_MB_SKIP){
  409.         fill_rectangle(intra_types, 4, 4, r->intra_types_stride, 0, sizeof(intra_types[0]));
  410.         return 0;
  411.     }
  412.     r->chroma_vlc = 1;
  413.     r->luma_vlc   = 0;
  414.  
  415.     if(IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
  416.         if(r->is16){
  417.             t = get_bits(gb, 2);
  418.             fill_rectangle(intra_types, 4, 4, r->intra_types_stride, t, sizeof(intra_types[0]));
  419.             r->luma_vlc   = 2;
  420.         }else{
  421.             if(r->decode_intra_types(r, gb, intra_types) < 0)
  422.                 return -1;
  423.             r->luma_vlc   = 1;
  424.         }
  425.         r->chroma_vlc = 0;
  426.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
  427.     }else{
  428.         for(i = 0; i < 16; i++)
  429.             intra_types[(i & 3) + (i>>2) * r->intra_types_stride] = 0;
  430.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
  431.         if(r->mb_type[mb_pos] == RV34_MB_P_MIX16x16){
  432.             r->is16 = 1;
  433.             r->chroma_vlc = 1;
  434.             r->luma_vlc   = 2;
  435.             r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 0);
  436.         }
  437.     }
  438.  
  439.     return rv34_decode_cbp(gb, r->cur_vlcs, r->is16);
  440. }
  441.  
  442. /** @} */ //bitstream functions
  443.  
  444. /**
  445.  * @name motion vector related code (prediction, reconstruction, motion compensation)
  446.  * @{
  447.  */
  448.  
  449. /** macroblock partition width in 8x8 blocks */
  450. static const uint8_t part_sizes_w[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 2, 1, 2, 2 };
  451.  
  452. /** macroblock partition height in 8x8 blocks */
  453. static const uint8_t part_sizes_h[RV34_MB_TYPES] = { 2, 2, 2, 1, 2, 2, 2, 2, 1, 2, 2, 2 };
  454.  
  455. /** availability index for subblocks */
  456. static const uint8_t avail_indexes[4] = { 6, 7, 10, 11 };
  457.  
  458. /**
  459.  * motion vector prediction
  460.  *
  461.  * Motion prediction performed for the block by using median prediction of
  462.  * motion vectors from the left, top and right top blocks but in corner cases
  463.  * some other vectors may be used instead.
  464.  */
  465. static void rv34_pred_mv(RV34DecContext *r, int block_type, int subblock_no, int dmv_no)
  466. {
  467.     MpegEncContext *s = &r->s;
  468.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
  469.     int A[2] = {0}, B[2], C[2];
  470.     int i, j;
  471.     int mx, my;
  472.     int* avail = r->avail_cache + avail_indexes[subblock_no];
  473.     int c_off = part_sizes_w[block_type];
  474.  
  475.     mv_pos += (subblock_no & 1) + (subblock_no >> 1)*s->b8_stride;
  476.     if(subblock_no == 3)
  477.         c_off = -1;
  478.  
  479.     if(avail[-1]){
  480.         A[0] = s->current_picture_ptr->motion_val[0][mv_pos-1][0];
  481.         A[1] = s->current_picture_ptr->motion_val[0][mv_pos-1][1];
  482.     }
  483.     if(avail[-4]){
  484.         B[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][0];
  485.         B[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride][1];
  486.     }else{
  487.         B[0] = A[0];
  488.         B[1] = A[1];
  489.     }
  490.     if(!avail[c_off-4]){
  491.         if(avail[-4] && (avail[-1] || r->rv30)){
  492.             C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][0];
  493.             C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride-1][1];
  494.         }else{
  495.             C[0] = A[0];
  496.             C[1] = A[1];
  497.         }
  498.     }else{
  499.         C[0] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][0];
  500.         C[1] = s->current_picture_ptr->motion_val[0][mv_pos-s->b8_stride+c_off][1];
  501.     }
  502.     mx = mid_pred(A[0], B[0], C[0]);
  503.     my = mid_pred(A[1], B[1], C[1]);
  504.     mx += r->dmv[dmv_no][0];
  505.     my += r->dmv[dmv_no][1];
  506.     for(j = 0; j < part_sizes_h[block_type]; j++){
  507.         for(i = 0; i < part_sizes_w[block_type]; i++){
  508.             s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;
  509.             s->current_picture_ptr->motion_val[0][mv_pos + i + j*s->b8_stride][1] = my;
  510.         }
  511.     }
  512. }
  513.  
  514. #define GET_PTS_DIFF(a, b) (((a) - (b) + 8192) & 0x1FFF)
  515.  
  516. /**
  517.  * Calculate motion vector component that should be added for direct blocks.
  518.  */
  519. static int calc_add_mv(RV34DecContext *r, int dir, int val)
  520. {
  521.     int mul = dir ? -r->mv_weight2 : r->mv_weight1;
  522.  
  523.     return (val * mul + 0x2000) >> 14;
  524. }
  525.  
  526. /**
  527.  * Predict motion vector for B-frame macroblock.
  528.  */
  529. static inline void rv34_pred_b_vector(int A[2], int B[2], int C[2],
  530.                                       int A_avail, int B_avail, int C_avail,
  531.                                       int *mx, int *my)
  532. {
  533.     if(A_avail + B_avail + C_avail != 3){
  534.         *mx = A[0] + B[0] + C[0];
  535.         *my = A[1] + B[1] + C[1];
  536.         if(A_avail + B_avail + C_avail == 2){
  537.             *mx /= 2;
  538.             *my /= 2;
  539.         }
  540.     }else{
  541.         *mx = mid_pred(A[0], B[0], C[0]);
  542.         *my = mid_pred(A[1], B[1], C[1]);
  543.     }
  544. }
  545.  
  546. /**
  547.  * motion vector prediction for B-frames
  548.  */
  549. static void rv34_pred_mv_b(RV34DecContext *r, int block_type, int dir)
  550. {
  551.     MpegEncContext *s = &r->s;
  552.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
  553.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
  554.     int A[2] = { 0 }, B[2] = { 0 }, C[2] = { 0 };
  555.     int has_A = 0, has_B = 0, has_C = 0;
  556.     int mx, my;
  557.     int i, j;
  558.     Picture *cur_pic = s->current_picture_ptr;
  559.     const int mask = dir ? MB_TYPE_L1 : MB_TYPE_L0;
  560.     int type = cur_pic->mb_type[mb_pos];
  561.  
  562.     if((r->avail_cache[6-1] & type) & mask){
  563.         A[0] = cur_pic->motion_val[dir][mv_pos - 1][0];
  564.         A[1] = cur_pic->motion_val[dir][mv_pos - 1][1];
  565.         has_A = 1;
  566.     }
  567.     if((r->avail_cache[6-4] & type) & mask){
  568.         B[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][0];
  569.         B[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride][1];
  570.         has_B = 1;
  571.     }
  572.     if(r->avail_cache[6-4] && (r->avail_cache[6-2] & type) & mask){
  573.         C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][0];
  574.         C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride + 2][1];
  575.         has_C = 1;
  576.     }else if((s->mb_x+1) == s->mb_width && (r->avail_cache[6-5] & type) & mask){
  577.         C[0] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][0];
  578.         C[1] = cur_pic->motion_val[dir][mv_pos - s->b8_stride - 1][1];
  579.         has_C = 1;
  580.     }
  581.  
  582.     rv34_pred_b_vector(A, B, C, has_A, has_B, has_C, &mx, &my);
  583.  
  584.     mx += r->dmv[dir][0];
  585.     my += r->dmv[dir][1];
  586.  
  587.     for(j = 0; j < 2; j++){
  588.         for(i = 0; i < 2; i++){
  589.             cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;
  590.             cur_pic->motion_val[dir][mv_pos + i + j*s->b8_stride][1] = my;
  591.         }
  592.     }
  593.     if(block_type == RV34_MB_B_BACKWARD || block_type == RV34_MB_B_FORWARD){
  594.         ZERO8x2(cur_pic->motion_val[!dir][mv_pos], s->b8_stride);
  595.     }
  596. }
  597.  
  598. /**
  599.  * motion vector prediction - RV3 version
  600.  */
  601. static void rv34_pred_mv_rv3(RV34DecContext *r, int block_type, int dir)
  602. {
  603.     MpegEncContext *s = &r->s;
  604.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
  605.     int A[2] = {0}, B[2], C[2];
  606.     int i, j, k;
  607.     int mx, my;
  608.     int* avail = r->avail_cache + avail_indexes[0];
  609.  
  610.     if(avail[-1]){
  611.         A[0] = s->current_picture_ptr->motion_val[0][mv_pos - 1][0];
  612.         A[1] = s->current_picture_ptr->motion_val[0][mv_pos - 1][1];
  613.     }
  614.     if(avail[-4]){
  615.         B[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][0];
  616.         B[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride][1];
  617.     }else{
  618.         B[0] = A[0];
  619.         B[1] = A[1];
  620.     }
  621.     if(!avail[-4 + 2]){
  622.         if(avail[-4] && (avail[-1])){
  623.             C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][0];
  624.             C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride - 1][1];
  625.         }else{
  626.             C[0] = A[0];
  627.             C[1] = A[1];
  628.         }
  629.     }else{
  630.         C[0] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][0];
  631.         C[1] = s->current_picture_ptr->motion_val[0][mv_pos - s->b8_stride + 2][1];
  632.     }
  633.     mx = mid_pred(A[0], B[0], C[0]);
  634.     my = mid_pred(A[1], B[1], C[1]);
  635.     mx += r->dmv[0][0];
  636.     my += r->dmv[0][1];
  637.     for(j = 0; j < 2; j++){
  638.         for(i = 0; i < 2; i++){
  639.             for(k = 0; k < 2; k++){
  640.                 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;
  641.                 s->current_picture_ptr->motion_val[k][mv_pos + i + j*s->b8_stride][1] = my;
  642.             }
  643.         }
  644.     }
  645. }
  646.  
  647. static const int chroma_coeffs[3] = { 0, 3, 5 };
  648.  
  649. /**
  650.  * generic motion compensation function
  651.  *
  652.  * @param r decoder context
  653.  * @param block_type type of the current block
  654.  * @param xoff horizontal offset from the start of the current block
  655.  * @param yoff vertical offset from the start of the current block
  656.  * @param mv_off offset to the motion vector information
  657.  * @param width width of the current partition in 8x8 blocks
  658.  * @param height height of the current partition in 8x8 blocks
  659.  * @param dir motion compensation direction (i.e. from the last or the next reference frame)
  660.  * @param thirdpel motion vectors are specified in 1/3 of pixel
  661.  * @param qpel_mc a set of functions used to perform luma motion compensation
  662.  * @param chroma_mc a set of functions used to perform chroma motion compensation
  663.  */
  664. static inline void rv34_mc(RV34DecContext *r, const int block_type,
  665.                           const int xoff, const int yoff, int mv_off,
  666.                           const int width, const int height, int dir,
  667.                           const int thirdpel, int weighted,
  668.                           qpel_mc_func (*qpel_mc)[16],
  669.                           h264_chroma_mc_func (*chroma_mc))
  670. {
  671.     MpegEncContext *s = &r->s;
  672.     uint8_t *Y, *U, *V, *srcY, *srcU, *srcV;
  673.     int dxy, mx, my, umx, umy, lx, ly, uvmx, uvmy, src_x, src_y, uvsrc_x, uvsrc_y;
  674.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride + mv_off;
  675.     int is16x16 = 1;
  676.     int emu = 0;
  677.  
  678.     if(thirdpel){
  679.         int chroma_mx, chroma_my;
  680.         mx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) / 3 - (1 << 24);
  681.         my = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) / 3 - (1 << 24);
  682.         lx = (s->current_picture_ptr->motion_val[dir][mv_pos][0] + (3 << 24)) % 3;
  683.         ly = (s->current_picture_ptr->motion_val[dir][mv_pos][1] + (3 << 24)) % 3;
  684.         chroma_mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
  685.         chroma_my = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
  686.         umx = (chroma_mx + (3 << 24)) / 3 - (1 << 24);
  687.         umy = (chroma_my + (3 << 24)) / 3 - (1 << 24);
  688.         uvmx = chroma_coeffs[(chroma_mx + (3 << 24)) % 3];
  689.         uvmy = chroma_coeffs[(chroma_my + (3 << 24)) % 3];
  690.     }else{
  691.         int cx, cy;
  692.         mx = s->current_picture_ptr->motion_val[dir][mv_pos][0] >> 2;
  693.         my = s->current_picture_ptr->motion_val[dir][mv_pos][1] >> 2;
  694.         lx = s->current_picture_ptr->motion_val[dir][mv_pos][0] & 3;
  695.         ly = s->current_picture_ptr->motion_val[dir][mv_pos][1] & 3;
  696.         cx = s->current_picture_ptr->motion_val[dir][mv_pos][0] / 2;
  697.         cy = s->current_picture_ptr->motion_val[dir][mv_pos][1] / 2;
  698.         umx = cx >> 2;
  699.         umy = cy >> 2;
  700.         uvmx = (cx & 3) << 1;
  701.         uvmy = (cy & 3) << 1;
  702.         //due to some flaw RV40 uses the same MC compensation routine for H2V2 and H3V3
  703.         if(uvmx == 6 && uvmy == 6)
  704.             uvmx = uvmy = 4;
  705.     }
  706.  
  707.     if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
  708.         /* wait for the referenced mb row to be finished */
  709.         int mb_row = s->mb_y + ((yoff + my + 5 + 8 * height) >> 4);
  710.         ThreadFrame *f = dir ? &s->next_picture_ptr->tf : &s->last_picture_ptr->tf;
  711.         ff_thread_await_progress(f, mb_row, 0);
  712.     }
  713.  
  714.     dxy = ly*4 + lx;
  715.     srcY = dir ? s->next_picture_ptr->f->data[0] : s->last_picture_ptr->f->data[0];
  716.     srcU = dir ? s->next_picture_ptr->f->data[1] : s->last_picture_ptr->f->data[1];
  717.     srcV = dir ? s->next_picture_ptr->f->data[2] : s->last_picture_ptr->f->data[2];
  718.     src_x = s->mb_x * 16 + xoff + mx;
  719.     src_y = s->mb_y * 16 + yoff + my;
  720.     uvsrc_x = s->mb_x * 8 + (xoff >> 1) + umx;
  721.     uvsrc_y = s->mb_y * 8 + (yoff >> 1) + umy;
  722.     srcY += src_y * s->linesize + src_x;
  723.     srcU += uvsrc_y * s->uvlinesize + uvsrc_x;
  724.     srcV += uvsrc_y * s->uvlinesize + uvsrc_x;
  725.     if(s->h_edge_pos - (width << 3) < 6 || s->v_edge_pos - (height << 3) < 6 ||
  726.        (unsigned)(src_x - !!lx*2) > s->h_edge_pos - !!lx*2 - (width <<3) - 4 ||
  727.        (unsigned)(src_y - !!ly*2) > s->v_edge_pos - !!ly*2 - (height<<3) - 4) {
  728.         srcY -= 2 + 2*s->linesize;
  729.         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, srcY,
  730.                                  s->linesize, s->linesize,
  731.                                  (width << 3) + 6, (height << 3) + 6,
  732.                                  src_x - 2, src_y - 2,
  733.                                  s->h_edge_pos, s->v_edge_pos);
  734.         srcY = s->sc.edge_emu_buffer + 2 + 2*s->linesize;
  735.         emu = 1;
  736.     }
  737.     if(!weighted){
  738.         Y = s->dest[0] + xoff      + yoff     *s->linesize;
  739.         U = s->dest[1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
  740.         V = s->dest[2] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
  741.     }else{
  742.         Y = r->tmp_b_block_y [dir]     +  xoff     +  yoff    *s->linesize;
  743.         U = r->tmp_b_block_uv[dir*2]   + (xoff>>1) + (yoff>>1)*s->uvlinesize;
  744.         V = r->tmp_b_block_uv[dir*2+1] + (xoff>>1) + (yoff>>1)*s->uvlinesize;
  745.     }
  746.  
  747.     if(block_type == RV34_MB_P_16x8){
  748.         qpel_mc[1][dxy](Y, srcY, s->linesize);
  749.         Y    += 8;
  750.         srcY += 8;
  751.     }else if(block_type == RV34_MB_P_8x16){
  752.         qpel_mc[1][dxy](Y, srcY, s->linesize);
  753.         Y    += 8 * s->linesize;
  754.         srcY += 8 * s->linesize;
  755.     }
  756.     is16x16 = (block_type != RV34_MB_P_8x8) && (block_type != RV34_MB_P_16x8) && (block_type != RV34_MB_P_8x16);
  757.     qpel_mc[!is16x16][dxy](Y, srcY, s->linesize);
  758.     if (emu) {
  759.         uint8_t *uvbuf = s->sc.edge_emu_buffer;
  760.  
  761.         s->vdsp.emulated_edge_mc(uvbuf, srcU,
  762.                                  s->uvlinesize, s->uvlinesize,
  763.                                  (width << 2) + 1, (height << 2) + 1,
  764.                                  uvsrc_x, uvsrc_y,
  765.                                  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
  766.         srcU = uvbuf;
  767.         uvbuf += 9*s->uvlinesize;
  768.  
  769.         s->vdsp.emulated_edge_mc(uvbuf, srcV,
  770.                                  s->uvlinesize, s->uvlinesize,
  771.                                  (width << 2) + 1, (height << 2) + 1,
  772.                                  uvsrc_x, uvsrc_y,
  773.                                  s->h_edge_pos >> 1, s->v_edge_pos >> 1);
  774.         srcV = uvbuf;
  775.     }
  776.     chroma_mc[2-width]   (U, srcU, s->uvlinesize, height*4, uvmx, uvmy);
  777.     chroma_mc[2-width]   (V, srcV, s->uvlinesize, height*4, uvmx, uvmy);
  778. }
  779.  
  780. static void rv34_mc_1mv(RV34DecContext *r, const int block_type,
  781.                         const int xoff, const int yoff, int mv_off,
  782.                         const int width, const int height, int dir)
  783. {
  784.     rv34_mc(r, block_type, xoff, yoff, mv_off, width, height, dir, r->rv30, 0,
  785.             r->rdsp.put_pixels_tab,
  786.             r->rdsp.put_chroma_pixels_tab);
  787. }
  788.  
  789. static void rv4_weight(RV34DecContext *r)
  790. {
  791.     r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][0](r->s.dest[0],
  792.                                                         r->tmp_b_block_y[0],
  793.                                                         r->tmp_b_block_y[1],
  794.                                                         r->weight1,
  795.                                                         r->weight2,
  796.                                                         r->s.linesize);
  797.     r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[1],
  798.                                                         r->tmp_b_block_uv[0],
  799.                                                         r->tmp_b_block_uv[2],
  800.                                                         r->weight1,
  801.                                                         r->weight2,
  802.                                                         r->s.uvlinesize);
  803.     r->rdsp.rv40_weight_pixels_tab[r->scaled_weight][1](r->s.dest[2],
  804.                                                         r->tmp_b_block_uv[1],
  805.                                                         r->tmp_b_block_uv[3],
  806.                                                         r->weight1,
  807.                                                         r->weight2,
  808.                                                         r->s.uvlinesize);
  809. }
  810.  
  811. static void rv34_mc_2mv(RV34DecContext *r, const int block_type)
  812. {
  813.     int weighted = !r->rv30 && block_type != RV34_MB_B_BIDIR && r->weight1 != 8192;
  814.  
  815.     rv34_mc(r, block_type, 0, 0, 0, 2, 2, 0, r->rv30, weighted,
  816.             r->rdsp.put_pixels_tab,
  817.             r->rdsp.put_chroma_pixels_tab);
  818.     if(!weighted){
  819.         rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 0,
  820.                 r->rdsp.avg_pixels_tab,
  821.                 r->rdsp.avg_chroma_pixels_tab);
  822.     }else{
  823.         rv34_mc(r, block_type, 0, 0, 0, 2, 2, 1, r->rv30, 1,
  824.                 r->rdsp.put_pixels_tab,
  825.                 r->rdsp.put_chroma_pixels_tab);
  826.         rv4_weight(r);
  827.     }
  828. }
  829.  
  830. static void rv34_mc_2mv_skip(RV34DecContext *r)
  831. {
  832.     int i, j;
  833.     int weighted = !r->rv30 && r->weight1 != 8192;
  834.  
  835.     for(j = 0; j < 2; j++)
  836.         for(i = 0; i < 2; i++){
  837.              rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 0, r->rv30,
  838.                      weighted,
  839.                      r->rdsp.put_pixels_tab,
  840.                      r->rdsp.put_chroma_pixels_tab);
  841.              rv34_mc(r, RV34_MB_P_8x8, i*8, j*8, i+j*r->s.b8_stride, 1, 1, 1, r->rv30,
  842.                      weighted,
  843.                      weighted ? r->rdsp.put_pixels_tab : r->rdsp.avg_pixels_tab,
  844.                      weighted ? r->rdsp.put_chroma_pixels_tab : r->rdsp.avg_chroma_pixels_tab);
  845.         }
  846.     if(weighted)
  847.         rv4_weight(r);
  848. }
  849.  
  850. /** number of motion vectors in each macroblock type */
  851. static const int num_mvs[RV34_MB_TYPES] = { 0, 0, 1, 4, 1, 1, 0, 0, 2, 2, 2, 1 };
  852.  
  853. /**
  854.  * Decode motion vector differences
  855.  * and perform motion vector reconstruction and motion compensation.
  856.  */
  857. static int rv34_decode_mv(RV34DecContext *r, int block_type)
  858. {
  859.     MpegEncContext *s = &r->s;
  860.     GetBitContext *gb = &s->gb;
  861.     int i, j, k, l;
  862.     int mv_pos = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
  863.     int next_bt;
  864.  
  865.     memset(r->dmv, 0, sizeof(r->dmv));
  866.     for(i = 0; i < num_mvs[block_type]; i++){
  867.         r->dmv[i][0] = svq3_get_se_golomb(gb);
  868.         r->dmv[i][1] = svq3_get_se_golomb(gb);
  869.     }
  870.     switch(block_type){
  871.     case RV34_MB_TYPE_INTRA:
  872.     case RV34_MB_TYPE_INTRA16x16:
  873.         ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
  874.         return 0;
  875.     case RV34_MB_SKIP:
  876.         if(s->pict_type == AV_PICTURE_TYPE_P){
  877.             ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
  878.             rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
  879.             break;
  880.         }
  881.     case RV34_MB_B_DIRECT:
  882.         //surprisingly, it uses motion scheme from next reference frame
  883.         /* wait for the current mb row to be finished */
  884.         if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
  885.             ff_thread_await_progress(&s->next_picture_ptr->tf, FFMAX(0, s->mb_y-1), 0);
  886.  
  887.         next_bt = s->next_picture_ptr->mb_type[s->mb_x + s->mb_y * s->mb_stride];
  888.         if(IS_INTRA(next_bt) || IS_SKIP(next_bt)){
  889.             ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
  890.             ZERO8x2(s->current_picture_ptr->motion_val[1][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
  891.         }else
  892.             for(j = 0; j < 2; j++)
  893.                 for(i = 0; i < 2; i++)
  894.                     for(k = 0; k < 2; k++)
  895.                         for(l = 0; l < 2; l++)
  896.                             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]);
  897.         if(!(IS_16X8(next_bt) || IS_8X16(next_bt) || IS_8X8(next_bt))) //we can use whole macroblock MC
  898.             rv34_mc_2mv(r, block_type);
  899.         else
  900.             rv34_mc_2mv_skip(r);
  901.         ZERO8x2(s->current_picture_ptr->motion_val[0][s->mb_x * 2 + s->mb_y * 2 * s->b8_stride], s->b8_stride);
  902.         break;
  903.     case RV34_MB_P_16x16:
  904.     case RV34_MB_P_MIX16x16:
  905.         rv34_pred_mv(r, block_type, 0, 0);
  906.         rv34_mc_1mv (r, block_type, 0, 0, 0, 2, 2, 0);
  907.         break;
  908.     case RV34_MB_B_FORWARD:
  909.     case RV34_MB_B_BACKWARD:
  910.         r->dmv[1][0] = r->dmv[0][0];
  911.         r->dmv[1][1] = r->dmv[0][1];
  912.         if(r->rv30)
  913.             rv34_pred_mv_rv3(r, block_type, block_type == RV34_MB_B_BACKWARD);
  914.         else
  915.             rv34_pred_mv_b  (r, block_type, block_type == RV34_MB_B_BACKWARD);
  916.         rv34_mc_1mv     (r, block_type, 0, 0, 0, 2, 2, block_type == RV34_MB_B_BACKWARD);
  917.         break;
  918.     case RV34_MB_P_16x8:
  919.     case RV34_MB_P_8x16:
  920.         rv34_pred_mv(r, block_type, 0, 0);
  921.         rv34_pred_mv(r, block_type, 1 + (block_type == RV34_MB_P_16x8), 1);
  922.         if(block_type == RV34_MB_P_16x8){
  923.             rv34_mc_1mv(r, block_type, 0, 0, 0,            2, 1, 0);
  924.             rv34_mc_1mv(r, block_type, 0, 8, s->b8_stride, 2, 1, 0);
  925.         }
  926.         if(block_type == RV34_MB_P_8x16){
  927.             rv34_mc_1mv(r, block_type, 0, 0, 0, 1, 2, 0);
  928.             rv34_mc_1mv(r, block_type, 8, 0, 1, 1, 2, 0);
  929.         }
  930.         break;
  931.     case RV34_MB_B_BIDIR:
  932.         rv34_pred_mv_b  (r, block_type, 0);
  933.         rv34_pred_mv_b  (r, block_type, 1);
  934.         rv34_mc_2mv     (r, block_type);
  935.         break;
  936.     case RV34_MB_P_8x8:
  937.         for(i=0;i< 4;i++){
  938.             rv34_pred_mv(r, block_type, i, i);
  939.             rv34_mc_1mv (r, block_type, (i&1)<<3, (i&2)<<2, (i&1)+(i>>1)*s->b8_stride, 1, 1, 0);
  940.         }
  941.         break;
  942.     }
  943.  
  944.     return 0;
  945. }
  946. /** @} */ // mv group
  947.  
  948. /**
  949.  * @name Macroblock reconstruction functions
  950.  * @{
  951.  */
  952. /** mapping of RV30/40 intra prediction types to standard H.264 types */
  953. static const int ittrans[9] = {
  954.  DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,
  955.  VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,
  956. };
  957.  
  958. /** mapping of RV30/40 intra 16x16 prediction types to standard H.264 types */
  959. static const int ittrans16[4] = {
  960.  DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,
  961. };
  962.  
  963. /**
  964.  * Perform 4x4 intra prediction.
  965.  */
  966. static void rv34_pred_4x4_block(RV34DecContext *r, uint8_t *dst, int stride, int itype, int up, int left, int down, int right)
  967. {
  968.     uint8_t *prev = dst - stride + 4;
  969.     uint32_t topleft;
  970.  
  971.     if(!up && !left)
  972.         itype = DC_128_PRED;
  973.     else if(!up){
  974.         if(itype == VERT_PRED) itype = HOR_PRED;
  975.         if(itype == DC_PRED)   itype = LEFT_DC_PRED;
  976.     }else if(!left){
  977.         if(itype == HOR_PRED)  itype = VERT_PRED;
  978.         if(itype == DC_PRED)   itype = TOP_DC_PRED;
  979.         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
  980.     }
  981.     if(!down){
  982.         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;
  983.         if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;
  984.         if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;
  985.     }
  986.     if(!right && up){
  987.         topleft = dst[-stride + 3] * 0x01010101u;
  988.         prev = (uint8_t*)&topleft;
  989.     }
  990.     r->h.pred4x4[itype](dst, prev, stride);
  991. }
  992.  
  993. static inline int adjust_pred16(int itype, int up, int left)
  994. {
  995.     if(!up && !left)
  996.         itype = DC_128_PRED8x8;
  997.     else if(!up){
  998.         if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;
  999.         if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;
  1000.         if(itype == DC_PRED8x8)   itype = LEFT_DC_PRED8x8;
  1001.     }else if(!left){
  1002.         if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;
  1003.         if(itype == HOR_PRED8x8)  itype = VERT_PRED8x8;
  1004.         if(itype == DC_PRED8x8)   itype = TOP_DC_PRED8x8;
  1005.     }
  1006.     return itype;
  1007. }
  1008.  
  1009. static inline void rv34_process_block(RV34DecContext *r,
  1010.                                       uint8_t *pdst, int stride,
  1011.                                       int fc, int sc, int q_dc, int q_ac)
  1012. {
  1013.     MpegEncContext *s = &r->s;
  1014.     int16_t *ptr = s->block[0];
  1015.     int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,
  1016.                                    fc, sc, q_dc, q_ac, q_ac);
  1017.     if(has_ac){
  1018.         r->rdsp.rv34_idct_add(pdst, stride, ptr);
  1019.     }else{
  1020.         r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);
  1021.         ptr[0] = 0;
  1022.     }
  1023. }
  1024.  
  1025. static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)
  1026. {
  1027.     LOCAL_ALIGNED_16(int16_t, block16, [16]);
  1028.     MpegEncContext *s    = &r->s;
  1029.     GetBitContext  *gb   = &s->gb;
  1030.     int             q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],
  1031.                     q_ac = rv34_qscale_tab[s->qscale];
  1032.     uint8_t        *dst  = s->dest[0];
  1033.     int16_t        *ptr  = s->block[0];
  1034.     int i, j, itype, has_ac;
  1035.  
  1036.     memset(block16, 0, 16 * sizeof(*block16));
  1037.  
  1038.     has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);
  1039.     if(has_ac)
  1040.         r->rdsp.rv34_inv_transform(block16);
  1041.     else
  1042.         r->rdsp.rv34_inv_transform_dc(block16);
  1043.  
  1044.     itype = ittrans16[intra_types[0]];
  1045.     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
  1046.     r->h.pred16x16[itype](dst, s->linesize);
  1047.  
  1048.     for(j = 0; j < 4; j++){
  1049.         for(i = 0; i < 4; i++, cbp >>= 1){
  1050.             int dc = block16[i + j*4];
  1051.  
  1052.             if(cbp & 1){
  1053.                 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
  1054.             }else
  1055.                 has_ac = 0;
  1056.  
  1057.             if(has_ac){
  1058.                 ptr[0] = dc;
  1059.                 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
  1060.             }else
  1061.                 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
  1062.         }
  1063.  
  1064.         dst += 4*s->linesize;
  1065.     }
  1066.  
  1067.     itype = ittrans16[intra_types[0]];
  1068.     if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;
  1069.     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);
  1070.  
  1071.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
  1072.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
  1073.  
  1074.     for(j = 1; j < 3; j++){
  1075.         dst = s->dest[j];
  1076.         r->h.pred8x8[itype](dst, s->uvlinesize);
  1077.         for(i = 0; i < 4; i++, cbp >>= 1){
  1078.             uint8_t *pdst;
  1079.             if(!(cbp & 1)) continue;
  1080.             pdst   = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
  1081.  
  1082.             rv34_process_block(r, pdst, s->uvlinesize,
  1083.                                r->chroma_vlc, 1, q_dc, q_ac);
  1084.         }
  1085.     }
  1086. }
  1087.  
  1088. static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)
  1089. {
  1090.     MpegEncContext *s   = &r->s;
  1091.     uint8_t        *dst = s->dest[0];
  1092.     int      avail[6*8] = {0};
  1093.     int i, j, k;
  1094.     int idx, q_ac, q_dc;
  1095.  
  1096.     // Set neighbour information.
  1097.     if(r->avail_cache[1])
  1098.         avail[0] = 1;
  1099.     if(r->avail_cache[2])
  1100.         avail[1] = avail[2] = 1;
  1101.     if(r->avail_cache[3])
  1102.         avail[3] = avail[4] = 1;
  1103.     if(r->avail_cache[4])
  1104.         avail[5] = 1;
  1105.     if(r->avail_cache[5])
  1106.         avail[8] = avail[16] = 1;
  1107.     if(r->avail_cache[9])
  1108.         avail[24] = avail[32] = 1;
  1109.  
  1110.     q_ac = rv34_qscale_tab[s->qscale];
  1111.     for(j = 0; j < 4; j++){
  1112.         idx = 9 + j*8;
  1113.         for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){
  1114.             rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);
  1115.             avail[idx] = 1;
  1116.             if(!(cbp & 1)) continue;
  1117.  
  1118.             rv34_process_block(r, dst, s->linesize,
  1119.                                r->luma_vlc, 0, q_ac, q_ac);
  1120.         }
  1121.         dst += s->linesize * 4 - 4*4;
  1122.         intra_types += r->intra_types_stride;
  1123.     }
  1124.  
  1125.     intra_types -= r->intra_types_stride * 4;
  1126.  
  1127.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
  1128.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
  1129.  
  1130.     for(k = 0; k < 2; k++){
  1131.         dst = s->dest[1+k];
  1132.         fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);
  1133.  
  1134.         for(j = 0; j < 2; j++){
  1135.             int* acache = r->avail_cache + 6 + j*4;
  1136.             for(i = 0; i < 2; i++, cbp >>= 1, acache++){
  1137.                 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];
  1138.                 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);
  1139.                 acache[0] = 1;
  1140.  
  1141.                 if(!(cbp&1)) continue;
  1142.  
  1143.                 rv34_process_block(r, dst + 4*i, s->uvlinesize,
  1144.                                    r->chroma_vlc, 1, q_dc, q_ac);
  1145.             }
  1146.  
  1147.             dst += 4*s->uvlinesize;
  1148.         }
  1149.     }
  1150. }
  1151.  
  1152. static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)
  1153. {
  1154.     int d;
  1155.     d = motion_val[0][0] - motion_val[-step][0];
  1156.     if(d < -3 || d > 3)
  1157.         return 1;
  1158.     d = motion_val[0][1] - motion_val[-step][1];
  1159.     if(d < -3 || d > 3)
  1160.         return 1;
  1161.     return 0;
  1162. }
  1163.  
  1164. static int rv34_set_deblock_coef(RV34DecContext *r)
  1165. {
  1166.     MpegEncContext *s = &r->s;
  1167.     int hmvmask = 0, vmvmask = 0, i, j;
  1168.     int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;
  1169.     int16_t (*motion_val)[2] = &s->current_picture_ptr->motion_val[0][midx];
  1170.     for(j = 0; j < 16; j += 8){
  1171.         for(i = 0; i < 2; i++){
  1172.             if(is_mv_diff_gt_3(motion_val + i, 1))
  1173.                 vmvmask |= 0x11 << (j + i*2);
  1174.             if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))
  1175.                 hmvmask |= 0x03 << (j + i*2);
  1176.         }
  1177.         motion_val += s->b8_stride;
  1178.     }
  1179.     if(s->first_slice_line)
  1180.         hmvmask &= ~0x000F;
  1181.     if(!s->mb_x)
  1182.         vmvmask &= ~0x1111;
  1183.     if(r->rv30){ //RV30 marks both subblocks on the edge for filtering
  1184.         vmvmask |= (vmvmask & 0x4444) >> 1;
  1185.         hmvmask |= (hmvmask & 0x0F00) >> 4;
  1186.         if(s->mb_x)
  1187.             r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;
  1188.         if(!s->first_slice_line)
  1189.             r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;
  1190.     }
  1191.     return hmvmask | vmvmask;
  1192. }
  1193.  
  1194. static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)
  1195. {
  1196.     MpegEncContext *s   = &r->s;
  1197.     GetBitContext  *gb  = &s->gb;
  1198.     uint8_t        *dst = s->dest[0];
  1199.     int16_t        *ptr = s->block[0];
  1200.     int          mb_pos = s->mb_x + s->mb_y * s->mb_stride;
  1201.     int cbp, cbp2;
  1202.     int q_dc, q_ac, has_ac;
  1203.     int i, j;
  1204.     int dist;
  1205.  
  1206.     // Calculate which neighbours are available. Maybe it's worth optimizing too.
  1207.     memset(r->avail_cache, 0, sizeof(r->avail_cache));
  1208.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
  1209.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
  1210.     if(s->mb_x && dist)
  1211.         r->avail_cache[5] =
  1212.         r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
  1213.     if(dist >= s->mb_width)
  1214.         r->avail_cache[2] =
  1215.         r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
  1216.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
  1217.         r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
  1218.     if(s->mb_x && dist > s->mb_width)
  1219.         r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
  1220.  
  1221.     s->qscale = r->si.quant;
  1222.     cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);
  1223.     r->cbp_luma  [mb_pos] = cbp;
  1224.     r->cbp_chroma[mb_pos] = cbp >> 16;
  1225.     r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];
  1226.     s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
  1227.  
  1228.     if(cbp == -1)
  1229.         return -1;
  1230.  
  1231.     if (IS_INTRA(s->current_picture_ptr->mb_type[mb_pos])){
  1232.         if(r->is16) rv34_output_i16x16(r, intra_types, cbp);
  1233.         else        rv34_output_intra(r, intra_types, cbp);
  1234.         return 0;
  1235.     }
  1236.  
  1237.     if(r->is16){
  1238.         // Only for RV34_MB_P_MIX16x16
  1239.         LOCAL_ALIGNED_16(int16_t, block16, [16]);
  1240.         memset(block16, 0, 16 * sizeof(*block16));
  1241.         q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];
  1242.         q_ac = rv34_qscale_tab[s->qscale];
  1243.         if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))
  1244.             r->rdsp.rv34_inv_transform(block16);
  1245.         else
  1246.             r->rdsp.rv34_inv_transform_dc(block16);
  1247.  
  1248.         q_ac = rv34_qscale_tab[s->qscale];
  1249.  
  1250.         for(j = 0; j < 4; j++){
  1251.             for(i = 0; i < 4; i++, cbp >>= 1){
  1252.                 int      dc   = block16[i + j*4];
  1253.  
  1254.                 if(cbp & 1){
  1255.                     has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);
  1256.                 }else
  1257.                     has_ac = 0;
  1258.  
  1259.                 if(has_ac){
  1260.                     ptr[0] = dc;
  1261.                     r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);
  1262.                 }else
  1263.                     r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);
  1264.             }
  1265.  
  1266.             dst += 4*s->linesize;
  1267.         }
  1268.  
  1269.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);
  1270.     }else{
  1271.         q_ac = rv34_qscale_tab[s->qscale];
  1272.  
  1273.         for(j = 0; j < 4; j++){
  1274.             for(i = 0; i < 4; i++, cbp >>= 1){
  1275.                 if(!(cbp & 1)) continue;
  1276.  
  1277.                 rv34_process_block(r, dst + 4*i, s->linesize,
  1278.                                    r->luma_vlc, 0, q_ac, q_ac);
  1279.             }
  1280.             dst += 4*s->linesize;
  1281.         }
  1282.     }
  1283.  
  1284.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];
  1285.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];
  1286.  
  1287.     for(j = 1; j < 3; j++){
  1288.         dst = s->dest[j];
  1289.         for(i = 0; i < 4; i++, cbp >>= 1){
  1290.             uint8_t *pdst;
  1291.             if(!(cbp & 1)) continue;
  1292.             pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;
  1293.  
  1294.             rv34_process_block(r, pdst, s->uvlinesize,
  1295.                                r->chroma_vlc, 1, q_dc, q_ac);
  1296.         }
  1297.     }
  1298.  
  1299.     return 0;
  1300. }
  1301.  
  1302. static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)
  1303. {
  1304.     MpegEncContext *s = &r->s;
  1305.     int cbp, dist;
  1306.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;
  1307.  
  1308.     // Calculate which neighbours are available. Maybe it's worth optimizing too.
  1309.     memset(r->avail_cache, 0, sizeof(r->avail_cache));
  1310.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);
  1311.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;
  1312.     if(s->mb_x && dist)
  1313.         r->avail_cache[5] =
  1314.         r->avail_cache[9] = s->current_picture_ptr->mb_type[mb_pos - 1];
  1315.     if(dist >= s->mb_width)
  1316.         r->avail_cache[2] =
  1317.         r->avail_cache[3] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride];
  1318.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)
  1319.         r->avail_cache[4] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride + 1];
  1320.     if(s->mb_x && dist > s->mb_width)
  1321.         r->avail_cache[1] = s->current_picture_ptr->mb_type[mb_pos - s->mb_stride - 1];
  1322.  
  1323.     s->qscale = r->si.quant;
  1324.     cbp = rv34_decode_intra_mb_header(r, intra_types);
  1325.     r->cbp_luma  [mb_pos] = cbp;
  1326.     r->cbp_chroma[mb_pos] = cbp >> 16;
  1327.     r->deblock_coefs[mb_pos] = 0xFFFF;
  1328.     s->current_picture_ptr->qscale_table[mb_pos] = s->qscale;
  1329.  
  1330.     if(cbp == -1)
  1331.         return -1;
  1332.  
  1333.     if(r->is16){
  1334.         rv34_output_i16x16(r, intra_types, cbp);
  1335.         return 0;
  1336.     }
  1337.  
  1338.     rv34_output_intra(r, intra_types, cbp);
  1339.     return 0;
  1340. }
  1341.  
  1342. static int check_slice_end(RV34DecContext *r, MpegEncContext *s)
  1343. {
  1344.     int bits;
  1345.     if(s->mb_y >= s->mb_height)
  1346.         return 1;
  1347.     if(!s->mb_num_left)
  1348.         return 1;
  1349.     if(r->s.mb_skip_run > 1)
  1350.         return 0;
  1351.     bits = get_bits_left(&s->gb);
  1352.     if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))
  1353.         return 1;
  1354.     return 0;
  1355. }
  1356.  
  1357.  
  1358. static void rv34_decoder_free(RV34DecContext *r)
  1359. {
  1360.     av_freep(&r->intra_types_hist);
  1361.     r->intra_types = NULL;
  1362.     av_freep(&r->tmp_b_block_base);
  1363.     av_freep(&r->mb_type);
  1364.     av_freep(&r->cbp_luma);
  1365.     av_freep(&r->cbp_chroma);
  1366.     av_freep(&r->deblock_coefs);
  1367. }
  1368.  
  1369.  
  1370. static int rv34_decoder_alloc(RV34DecContext *r)
  1371. {
  1372.     r->intra_types_stride = r->s.mb_width * 4 + 4;
  1373.  
  1374.     r->cbp_chroma       = av_mallocz(r->s.mb_stride * r->s.mb_height *
  1375.                                     sizeof(*r->cbp_chroma));
  1376.     r->cbp_luma         = av_mallocz(r->s.mb_stride * r->s.mb_height *
  1377.                                     sizeof(*r->cbp_luma));
  1378.     r->deblock_coefs    = av_mallocz(r->s.mb_stride * r->s.mb_height *
  1379.                                     sizeof(*r->deblock_coefs));
  1380.     r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *
  1381.                                     sizeof(*r->intra_types_hist));
  1382.     r->mb_type          = av_mallocz(r->s.mb_stride * r->s.mb_height *
  1383.                                      sizeof(*r->mb_type));
  1384.  
  1385.     if (!(r->cbp_chroma       && r->cbp_luma && r->deblock_coefs &&
  1386.           r->intra_types_hist && r->mb_type)) {
  1387.         rv34_decoder_free(r);
  1388.         return AVERROR(ENOMEM);
  1389.     }
  1390.  
  1391.     r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;
  1392.  
  1393.     return 0;
  1394. }
  1395.  
  1396.  
  1397. static int rv34_decoder_realloc(RV34DecContext *r)
  1398. {
  1399.     rv34_decoder_free(r);
  1400.     return rv34_decoder_alloc(r);
  1401. }
  1402.  
  1403.  
  1404. static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)
  1405. {
  1406.     MpegEncContext *s = &r->s;
  1407.     GetBitContext *gb = &s->gb;
  1408.     int mb_pos, slice_type;
  1409.     int res;
  1410.  
  1411.     init_get_bits(&r->s.gb, buf, buf_size*8);
  1412.     res = r->parse_slice_header(r, gb, &r->si);
  1413.     if(res < 0){
  1414.         av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");
  1415.         return -1;
  1416.     }
  1417.  
  1418.     slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;
  1419.     if (slice_type != s->pict_type) {
  1420.         av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");
  1421.         return AVERROR_INVALIDDATA;
  1422.     }
  1423.     if (s->width != r->si.width || s->height != r->si.height) {
  1424.         av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");
  1425.         return AVERROR_INVALIDDATA;
  1426.     }
  1427.  
  1428.     r->si.end = end;
  1429.     s->qscale = r->si.quant;
  1430.     s->mb_num_left = r->si.end - r->si.start;
  1431.     r->s.mb_skip_run = 0;
  1432.  
  1433.     mb_pos = s->mb_x + s->mb_y * s->mb_width;
  1434.     if(r->si.start != mb_pos){
  1435.         av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);
  1436.         s->mb_x = r->si.start % s->mb_width;
  1437.         s->mb_y = r->si.start / s->mb_width;
  1438.     }
  1439.     memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));
  1440.     s->first_slice_line = 1;
  1441.     s->resync_mb_x = s->mb_x;
  1442.     s->resync_mb_y = s->mb_y;
  1443.  
  1444.     ff_init_block_index(s);
  1445.     while(!check_slice_end(r, s)) {
  1446.         ff_update_block_index(s);
  1447.  
  1448.         if(r->si.type)
  1449.             res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
  1450.         else
  1451.             res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);
  1452.         if(res < 0){
  1453.             ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);
  1454.             return -1;
  1455.         }
  1456.         if (++s->mb_x == s->mb_width) {
  1457.             s->mb_x = 0;
  1458.             s->mb_y++;
  1459.             ff_init_block_index(s);
  1460.  
  1461.             memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
  1462.             memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));
  1463.  
  1464.             if(r->loop_filter && s->mb_y >= 2)
  1465.                 r->loop_filter(r, s->mb_y - 2);
  1466.  
  1467.             if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
  1468.                 ff_thread_report_progress(&s->current_picture_ptr->tf,
  1469.                                           s->mb_y - 2, 0);
  1470.  
  1471.         }
  1472.         if(s->mb_x == s->resync_mb_x)
  1473.             s->first_slice_line=0;
  1474.         s->mb_num_left--;
  1475.     }
  1476.     ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);
  1477.  
  1478.     return s->mb_y == s->mb_height;
  1479. }
  1480.  
  1481. /** @} */ // recons group end
  1482.  
  1483. /**
  1484.  * Initialize decoder.
  1485.  */
  1486. av_cold int ff_rv34_decode_init(AVCodecContext *avctx)
  1487. {
  1488.     RV34DecContext *r = avctx->priv_data;
  1489.     MpegEncContext *s = &r->s;
  1490.     int ret;
  1491.  
  1492.     ff_mpv_decode_defaults(s);
  1493.     ff_mpv_decode_init(s, avctx);
  1494.     s->out_format = FMT_H263;
  1495.  
  1496.     avctx->pix_fmt = AV_PIX_FMT_YUV420P;
  1497.     avctx->has_b_frames = 1;
  1498.     s->low_delay = 0;
  1499.  
  1500.     ff_mpv_idct_init(s);
  1501.     if ((ret = ff_mpv_common_init(s)) < 0)
  1502.         return ret;
  1503.  
  1504.     ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);
  1505.  
  1506. #if CONFIG_RV30_DECODER
  1507.     if (avctx->codec_id == AV_CODEC_ID_RV30)
  1508.         ff_rv30dsp_init(&r->rdsp);
  1509. #endif
  1510. #if CONFIG_RV40_DECODER
  1511.     if (avctx->codec_id == AV_CODEC_ID_RV40)
  1512.         ff_rv40dsp_init(&r->rdsp);
  1513. #endif
  1514.  
  1515.     if ((ret = rv34_decoder_alloc(r)) < 0) {
  1516.         ff_mpv_common_end(&r->s);
  1517.         return ret;
  1518.     }
  1519.  
  1520.     if(!intra_vlcs[0].cbppattern[0].bits)
  1521.         rv34_init_tables();
  1522.  
  1523.     avctx->internal->allocate_progress = 1;
  1524.  
  1525.     return 0;
  1526. }
  1527.  
  1528. int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx)
  1529. {
  1530.     int err;
  1531.     RV34DecContext *r = avctx->priv_data;
  1532.  
  1533.     r->s.avctx = avctx;
  1534.  
  1535.     if (avctx->internal->is_copy) {
  1536.         r->tmp_b_block_base = NULL;
  1537.         r->cbp_chroma       = NULL;
  1538.         r->cbp_luma         = NULL;
  1539.         r->deblock_coefs    = NULL;
  1540.         r->intra_types_hist = NULL;
  1541.         r->mb_type          = NULL;
  1542.  
  1543.         ff_mpv_idct_init(&r->s);
  1544.  
  1545.         if ((err = ff_mpv_common_init(&r->s)) < 0)
  1546.             return err;
  1547.         if ((err = rv34_decoder_alloc(r)) < 0) {
  1548.             ff_mpv_common_end(&r->s);
  1549.             return err;
  1550.         }
  1551.     }
  1552.  
  1553.     return 0;
  1554. }
  1555.  
  1556. int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
  1557. {
  1558.     RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;
  1559.     MpegEncContext * const s = &r->s, * const s1 = &r1->s;
  1560.     int err;
  1561.  
  1562.     if (dst == src || !s1->context_initialized)
  1563.         return 0;
  1564.  
  1565.     if (s->height != s1->height || s->width != s1->width) {
  1566.         s->height = s1->height;
  1567.         s->width  = s1->width;
  1568.         if ((err = ff_mpv_common_frame_size_change(s)) < 0)
  1569.             return err;
  1570.         if ((err = rv34_decoder_realloc(r)) < 0)
  1571.             return err;
  1572.     }
  1573.  
  1574.     r->cur_pts  = r1->cur_pts;
  1575.     r->last_pts = r1->last_pts;
  1576.     r->next_pts = r1->next_pts;
  1577.  
  1578.     memset(&r->si, 0, sizeof(r->si));
  1579.  
  1580.     // Do no call ff_mpeg_update_thread_context on a partially initialized
  1581.     // decoder context.
  1582.     if (!s1->linesize)
  1583.         return 0;
  1584.  
  1585.     return ff_mpeg_update_thread_context(dst, src);
  1586. }
  1587.  
  1588. static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)
  1589. {
  1590.     if(avctx->slice_count) return avctx->slice_offset[n];
  1591.     else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);
  1592. }
  1593.  
  1594. static int finish_frame(AVCodecContext *avctx, AVFrame *pict)
  1595. {
  1596.     RV34DecContext *r = avctx->priv_data;
  1597.     MpegEncContext *s = &r->s;
  1598.     int got_picture = 0, ret;
  1599.  
  1600.     ff_er_frame_end(&s->er);
  1601.     ff_mpv_frame_end(s);
  1602.     s->mb_num_left = 0;
  1603.  
  1604.     if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
  1605.         ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
  1606.  
  1607.     if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
  1608.         if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
  1609.             return ret;
  1610.         ff_print_debug_info(s, s->current_picture_ptr, pict);
  1611.         ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1);
  1612.         got_picture = 1;
  1613.     } else if (s->last_picture_ptr) {
  1614.         if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
  1615.             return ret;
  1616.         ff_print_debug_info(s, s->last_picture_ptr, pict);
  1617.         ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1);
  1618.         got_picture = 1;
  1619.     }
  1620.  
  1621.     return got_picture;
  1622. }
  1623.  
  1624. static AVRational update_sar(int old_w, int old_h, AVRational sar, int new_w, int new_h)
  1625. {
  1626.     // attempt to keep aspect during typical resolution switches
  1627.     if (!sar.num)
  1628.         sar = (AVRational){1, 1};
  1629.  
  1630.     sar = av_mul_q(sar, (AVRational){new_h * old_w, new_w * old_h});
  1631.     return sar;
  1632. }
  1633.  
  1634. int ff_rv34_decode_frame(AVCodecContext *avctx,
  1635.                             void *data, int *got_picture_ptr,
  1636.                             AVPacket *avpkt)
  1637. {
  1638.     const uint8_t *buf = avpkt->data;
  1639.     int buf_size = avpkt->size;
  1640.     RV34DecContext *r = avctx->priv_data;
  1641.     MpegEncContext *s = &r->s;
  1642.     AVFrame *pict = data;
  1643.     SliceInfo si;
  1644.     int i, ret;
  1645.     int slice_count;
  1646.     const uint8_t *slices_hdr = NULL;
  1647.     int last = 0;
  1648.  
  1649.     /* no supplementary picture */
  1650.     if (buf_size == 0) {
  1651.         /* special case for last picture */
  1652.         if (s->low_delay==0 && s->next_picture_ptr) {
  1653.             if ((ret = av_frame_ref(pict, s->next_picture_ptr->f)) < 0)
  1654.                 return ret;
  1655.             s->next_picture_ptr = NULL;
  1656.  
  1657.             *got_picture_ptr = 1;
  1658.         }
  1659.         return 0;
  1660.     }
  1661.  
  1662.     if(!avctx->slice_count){
  1663.         slice_count = (*buf++) + 1;
  1664.         slices_hdr = buf + 4;
  1665.         buf += 8 * slice_count;
  1666.         buf_size -= 1 + 8 * slice_count;
  1667.     }else
  1668.         slice_count = avctx->slice_count;
  1669.  
  1670.     //parse first slice header to check whether this frame can be decoded
  1671.     if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||
  1672.        get_slice_offset(avctx, slices_hdr, 0) > buf_size){
  1673.         av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
  1674.         return AVERROR_INVALIDDATA;
  1675.     }
  1676.     init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);
  1677.     if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){
  1678.         av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");
  1679.         return AVERROR_INVALIDDATA;
  1680.     }
  1681.     if ((!s->last_picture_ptr || !s->last_picture_ptr->f->data[0]) &&
  1682.         si.type == AV_PICTURE_TYPE_B) {
  1683.         av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "
  1684.                "reference data.\n");
  1685.         return AVERROR_INVALIDDATA;
  1686.     }
  1687.     if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)
  1688.        || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)
  1689.        ||  avctx->skip_frame >= AVDISCARD_ALL)
  1690.         return avpkt->size;
  1691.  
  1692.     /* first slice */
  1693.     if (si.start == 0) {
  1694.         if (s->mb_num_left > 0 && s->current_picture_ptr) {
  1695.             av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.\n",
  1696.                    s->mb_num_left);
  1697.             ff_er_frame_end(&s->er);
  1698.             ff_mpv_frame_end(s);
  1699.         }
  1700.  
  1701.         if (s->width != si.width || s->height != si.height) {
  1702.             int err;
  1703.  
  1704.             av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",
  1705.                    si.width, si.height);
  1706.  
  1707.             if (av_image_check_size(si.width, si.height, 0, s->avctx))
  1708.                 return AVERROR_INVALIDDATA;
  1709.  
  1710.             s->avctx->sample_aspect_ratio = update_sar(
  1711.                 s->width, s->height, s->avctx->sample_aspect_ratio,
  1712.                 si.width, si.height);
  1713.             s->width  = si.width;
  1714.             s->height = si.height;
  1715.  
  1716.             err = ff_set_dimensions(s->avctx, s->width, s->height);
  1717.             if (err < 0)
  1718.                 return err;
  1719.  
  1720.             if ((err = ff_mpv_common_frame_size_change(s)) < 0)
  1721.                 return err;
  1722.             if ((err = rv34_decoder_realloc(r)) < 0)
  1723.                 return err;
  1724.         }
  1725.         s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;
  1726.         if (ff_mpv_frame_start(s, s->avctx) < 0)
  1727.             return -1;
  1728.         ff_mpeg_er_frame_start(s);
  1729.         if (!r->tmp_b_block_base) {
  1730.             int i;
  1731.  
  1732.             r->tmp_b_block_base = av_malloc(s->linesize * 48);
  1733.             for (i = 0; i < 2; i++)
  1734.                 r->tmp_b_block_y[i] = r->tmp_b_block_base
  1735.                                       + i * 16 * s->linesize;
  1736.             for (i = 0; i < 4; i++)
  1737.                 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize
  1738.                                        + (i >> 1) * 8 * s->uvlinesize
  1739.                                        + (i &  1) * 16;
  1740.         }
  1741.         r->cur_pts = si.pts;
  1742.         if (s->pict_type != AV_PICTURE_TYPE_B) {
  1743.             r->last_pts = r->next_pts;
  1744.             r->next_pts = r->cur_pts;
  1745.         } else {
  1746.             int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);
  1747.             int dist0   = GET_PTS_DIFF(r->cur_pts,  r->last_pts);
  1748.             int dist1   = GET_PTS_DIFF(r->next_pts, r->cur_pts);
  1749.  
  1750.             if(!refdist){
  1751.                 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;
  1752.                 r->scaled_weight = 0;
  1753.             }else{
  1754.                 r->mv_weight1 = (dist0 << 14) / refdist;
  1755.                 r->mv_weight2 = (dist1 << 14) / refdist;
  1756.                 if((r->mv_weight1|r->mv_weight2) & 511){
  1757.                     r->weight1 = r->mv_weight1;
  1758.                     r->weight2 = r->mv_weight2;
  1759.                     r->scaled_weight = 0;
  1760.                 }else{
  1761.                     r->weight1 = r->mv_weight1 >> 9;
  1762.                     r->weight2 = r->mv_weight2 >> 9;
  1763.                     r->scaled_weight = 1;
  1764.                 }
  1765.             }
  1766.         }
  1767.         s->mb_x = s->mb_y = 0;
  1768.         ff_thread_finish_setup(s->avctx);
  1769.     } else if (HAVE_THREADS &&
  1770.                (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
  1771.         av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "
  1772.                "multithreading mode (start MB is %d).\n", si.start);
  1773.         return AVERROR_INVALIDDATA;
  1774.     }
  1775.  
  1776.     for(i = 0; i < slice_count; i++){
  1777.         int offset = get_slice_offset(avctx, slices_hdr, i);
  1778.         int size;
  1779.         if(i+1 == slice_count)
  1780.             size = buf_size - offset;
  1781.         else
  1782.             size = get_slice_offset(avctx, slices_hdr, i+1) - offset;
  1783.  
  1784.         if(offset < 0 || offset > buf_size){
  1785.             av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
  1786.             break;
  1787.         }
  1788.  
  1789.         r->si.end = s->mb_width * s->mb_height;
  1790.         s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;
  1791.  
  1792.         if(i+1 < slice_count){
  1793.             if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||
  1794.                 get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {
  1795.                 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");
  1796.                 break;
  1797.             }
  1798.             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);
  1799.             if(r->parse_slice_header(r, &r->s.gb, &si) < 0){
  1800.                 if(i+2 < slice_count)
  1801.                     size = get_slice_offset(avctx, slices_hdr, i+2) - offset;
  1802.                 else
  1803.                     size = buf_size - offset;
  1804.             }else
  1805.                 r->si.end = si.start;
  1806.         }
  1807.         if (size < 0 || size > buf_size - offset) {
  1808.             av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");
  1809.             break;
  1810.         }
  1811.         last = rv34_decode_slice(r, r->si.end, buf + offset, size);
  1812.         if(last)
  1813.             break;
  1814.     }
  1815.  
  1816.     if (s->current_picture_ptr) {
  1817.         if (last) {
  1818.             if(r->loop_filter)
  1819.                 r->loop_filter(r, s->mb_height - 1);
  1820.  
  1821.             ret = finish_frame(avctx, pict);
  1822.             if (ret < 0)
  1823.                 return ret;
  1824.             *got_picture_ptr = ret;
  1825.         } else if (HAVE_THREADS &&
  1826.                    (s->avctx->active_thread_type & FF_THREAD_FRAME)) {
  1827.             av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");
  1828.             /* always mark the current frame as finished, frame-mt supports
  1829.              * only complete frames */
  1830.             ff_er_frame_end(&s->er);
  1831.             ff_mpv_frame_end(s);
  1832.             s->mb_num_left = 0;
  1833.             ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
  1834.             return AVERROR_INVALIDDATA;
  1835.         }
  1836.     }
  1837.  
  1838.     return avpkt->size;
  1839. }
  1840.  
  1841. av_cold int ff_rv34_decode_end(AVCodecContext *avctx)
  1842. {
  1843.     RV34DecContext *r = avctx->priv_data;
  1844.  
  1845.     ff_mpv_common_end(&r->s);
  1846.     rv34_decoder_free(r);
  1847.  
  1848.     return 0;
  1849. }
  1850.