falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
36686 Commits
32 Branches
283MB
Tree: fc386f2eea
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'fc386f2eea'
${ noResults }
FFmpeg/libavcodec/rv34.c

1789 lines
62KB
Raw Blame History 

  1. /*

  2.  * RV30/40 decoder common data

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

  4.  *

  5.  * This file is part of Libav.

  6.  *

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

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

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

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

  11.  *

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

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

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

  15.  * Lesser General Public License for more details.

  16.  *

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

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

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

  20.  */

  21. 

  22. /**

  23.  * @file

  24.  * RV30/40 decoder common data

  25.  */

  26. 

  27. #include "libavutil/internal.h"

  28. 

  29. #include "avcodec.h"

  30. #include "mpegvideo.h"

  31. #include "golomb.h"

  32. #include "internal.h"

  33. #include "mathops.h"

  34. #include "rectangle.h"

  35. #include "thread.h"

  36. 

  37. #include "rv34vlc.h"

  38. #include "rv34data.h"

  39. #include "rv34.h"

  40. 

  41. //#define DEBUG

  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 bistream 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->f.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->f.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->f.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->f.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->f.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->f.motion_val[0][mv_pos-1][0];

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

  479.     }

  480.     if(avail[-4]){

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

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

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

  497.         C[1] = s->current_picture_ptr->f.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->f.motion_val[0][mv_pos + i + j*s->b8_stride][0] = mx;

  506.             s->current_picture_ptr->f.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->f.mb_type[mb_pos];

  558. 

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

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

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

  566.         B[1] = cur_pic->f.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->f.motion_val[dir][mv_pos - s->b8_stride + 2][0];

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

  575.         C[1] = cur_pic->f.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->f.motion_val[dir][mv_pos + i + j*s->b8_stride][0] = mx;

  587.             cur_pic->f.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->f.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->f.motion_val[0][mv_pos - 1][0];

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

  610.     }

  611.     if(avail[-4]){

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

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

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

  628.         C[1] = s->current_picture_ptr->f.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->f.motion_val[k][mv_pos + i + j*s->b8_stride][0] = mx;

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

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

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

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

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

  681.         chroma_my = s->current_picture_ptr->f.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->f.motion_val[dir][mv_pos][0] >> 2;

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

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

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

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

  693.         cy = s->current_picture_ptr->f.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.         AVFrame *f = dir ? &s->next_picture_ptr->f : &s->last_picture_ptr->f;

  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, srcY, s->linesize, (width<<3)+6, (height<<3)+6,

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

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

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

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

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

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

  734.         srcU = uvbuf;

  735.         srcV = uvbuf + 16;

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

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

  760. }

  761. 

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

  763.                         const int xoff, const int yoff, int mv_off,

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

  765. {

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

  767.             r->rdsp.put_pixels_tab,

  768.             r->rdsp.put_chroma_pixels_tab);

  769. }

  770. 

  771. static void rv4_weight(RV34DecContext *r)

  772. {

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

  774.                                                         r->tmp_b_block_y[0],

  775.                                                         r->tmp_b_block_y[1],

  776.                                                         r->weight1,

  777.                                                         r->weight2,

  778.                                                         r->s.linesize);

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

  780.                                                         r->tmp_b_block_uv[0],

  781.                                                         r->tmp_b_block_uv[2],

  782.                                                         r->weight1,

  783.                                                         r->weight2,

  784.                                                         r->s.uvlinesize);

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

  786.                                                         r->tmp_b_block_uv[1],

  787.                                                         r->tmp_b_block_uv[3],

  788.                                                         r->weight1,

  789.                                                         r->weight2,

  790.                                                         r->s.uvlinesize);

  791. }

  792. 

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

  794. {

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

  796. 

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

  798.             r->rdsp.put_pixels_tab,

  799.             r->rdsp.put_chroma_pixels_tab);

  800.     if(!weighted){

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

  802.                 r->rdsp.avg_pixels_tab,

  803.                 r->rdsp.avg_chroma_pixels_tab);

  804.     }else{

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

  806.                 r->rdsp.put_pixels_tab,

  807.                 r->rdsp.put_chroma_pixels_tab);

  808.         rv4_weight(r);

  809.     }

  810. }

  811. 

  812. static void rv34_mc_2mv_skip(RV34DecContext *r)

  813. {

  814.     int i, j;

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

  816. 

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

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

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

  820.                      weighted,

  821.                      r->rdsp.put_pixels_tab,

  822.                      r->rdsp.put_chroma_pixels_tab);

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

  824.                      weighted,

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

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

  827.         }

  828.     if(weighted)

  829.         rv4_weight(r);

  830. }

  831. 

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

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

  834. 

  835. /**

  836.  * Decode motion vector differences

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

  838.  */

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

  840. {

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

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

  843.     int i, j, k, l;

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

  845.     int next_bt;

  846. 

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

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

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

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

  851.     }

  852.     switch(block_type){

  853.     case RV34_MB_TYPE_INTRA:

  854.     case RV34_MB_TYPE_INTRA16x16:

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

  856.         return 0;

  857.     case RV34_MB_SKIP:

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

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

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

  861.             break;

  862.         }

  863.     case RV34_MB_B_DIRECT:

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

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

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

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

  868. 

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

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

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

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

  873.         }else

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

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

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

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

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

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

  880.             rv34_mc_2mv(r, block_type);

  881.         else

  882.             rv34_mc_2mv_skip(r);

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

  884.         break;

  885.     case RV34_MB_P_16x16:

  886.     case RV34_MB_P_MIX16x16:

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

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

  889.         break;

  890.     case RV34_MB_B_FORWARD:

  891.     case RV34_MB_B_BACKWARD:

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

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

  894.         if(r->rv30)

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

  896.         else

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

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

  899.         break;

  900.     case RV34_MB_P_16x8:

  901.     case RV34_MB_P_8x16:

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

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

  904.         if(block_type == RV34_MB_P_16x8){

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

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

  907.         }

  908.         if(block_type == RV34_MB_P_8x16){

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

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

  911.         }

  912.         break;

  913.     case RV34_MB_B_BIDIR:

  914.         rv34_pred_mv_b  (r, block_type, 0);

  915.         rv34_pred_mv_b  (r, block_type, 1);

  916.         rv34_mc_2mv     (r, block_type);

  917.         break;

  918.     case RV34_MB_P_8x8:

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

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

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

  922.         }

  923.         break;

  924.     }

  925. 

  926.     return 0;

  927. }

  928. /** @} */ // mv group

  929. 

  930. /**

  931.  * @name Macroblock reconstruction functions

  932.  * @{

  933.  */

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

  935. static const int ittrans[9] = {

  936.  DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,

  937.  VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,

  938. };

  939. 

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

  941. static const int ittrans16[4] = {

  942.  DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,

  943. };

  944. 

  945. /**

  946.  * Perform 4x4 intra prediction.

  947.  */

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

  949. {

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

  951.     uint32_t topleft;

  952. 

  953.     if(!up && !left)

  954.         itype = DC_128_PRED;

  955.     else if(!up){

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

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

  958.     }else if(!left){

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

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

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

  962.     }

  963.     if(!down){

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

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

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

  967.     }

  968.     if(!right && up){

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

  970.         prev = (uint8_t*)&topleft;

  971.     }

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

  973. }

  974. 

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

  976. {

  977.     if(!up && !left)

  978.         itype = DC_128_PRED8x8;

  979.     else if(!up){

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

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

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

  983.     }else if(!left){

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

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

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

  987.     }

  988.     return itype;

  989. }

  990. 

  991. static inline void rv34_process_block(RV34DecContext *r,

  992.                                       uint8_t *pdst, int stride,

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

  994. {

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

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

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

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

  999.     if(has_ac){

  1000.         r->rdsp.rv34_idct_add(pdst, stride, ptr);

  1001.     }else{

  1002.         r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);

  1003.         ptr[0] = 0;

  1004.     }

  1005. }

  1006. 

  1007. static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)

  1008. {

  1009.     LOCAL_ALIGNED_16(int16_t, block16, [16]);

  1010.     MpegEncContext *s    = &r->s;

  1011.     GetBitContext  *gb   = &s->gb;

  1012.     int             q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],

  1013.                     q_ac = rv34_qscale_tab[s->qscale];

  1014.     uint8_t        *dst  = s->dest[0];

  1015.     int16_t        *ptr  = s->block[0];

  1016.     int i, j, itype, has_ac;

  1017. 

  1018.     memset(block16, 0, 16 * sizeof(*block16));

  1019. 

  1020.     has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);

  1021.     if(has_ac)

  1022.         r->rdsp.rv34_inv_transform(block16);

  1023.     else

  1024.         r->rdsp.rv34_inv_transform_dc(block16);

  1025. 

  1026.     itype = ittrans16[intra_types[0]];

  1027.     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1028.     r->h.pred16x16[itype](dst, s->linesize);

  1029. 

  1030.     for(j = 0; j < 4; j++){

  1031.         for(i = 0; i < 4; i++, cbp >>= 1){

  1032.             int dc = block16[i + j*4];

  1033. 

  1034.             if(cbp & 1){

  1035.                 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);

  1036.             }else

  1037.                 has_ac = 0;

  1038. 

  1039.             if(has_ac){

  1040.                 ptr[0] = dc;

  1041.                 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);

  1042.             }else

  1043.                 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);

  1044.         }

  1045. 

  1046.         dst += 4*s->linesize;

  1047.     }

  1048. 

  1049.     itype = ittrans16[intra_types[0]];

  1050.     if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;

  1051.     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1052. 

  1053.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];

  1054.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];

  1055. 

  1056.     for(j = 1; j < 3; j++){

  1057.         dst = s->dest[j];

  1058.         r->h.pred8x8[itype](dst, s->uvlinesize);

  1059.         for(i = 0; i < 4; i++, cbp >>= 1){

  1060.             uint8_t *pdst;

  1061.             if(!(cbp & 1)) continue;

  1062.             pdst   = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;

  1063. 

  1064.             rv34_process_block(r, pdst, s->uvlinesize,

  1065.                                r->chroma_vlc, 1, q_dc, q_ac);

  1066.         }

  1067.     }

  1068. }

  1069. 

  1070. static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)

  1071. {

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

  1073.     uint8_t        *dst = s->dest[0];

  1074.     int      avail[6*8] = {0};

  1075.     int i, j, k;

  1076.     int idx, q_ac, q_dc;

  1077. 

  1078.     // Set neighbour information.

  1079.     if(r->avail_cache[1])

  1080.         avail[0] = 1;

  1081.     if(r->avail_cache[2])

  1082.         avail[1] = avail[2] = 1;

  1083.     if(r->avail_cache[3])

  1084.         avail[3] = avail[4] = 1;

  1085.     if(r->avail_cache[4])

  1086.         avail[5] = 1;

  1087.     if(r->avail_cache[5])

  1088.         avail[8] = avail[16] = 1;

  1089.     if(r->avail_cache[9])

  1090.         avail[24] = avail[32] = 1;

  1091. 

  1092.     q_ac = rv34_qscale_tab[s->qscale];

  1093.     for(j = 0; j < 4; j++){

  1094.         idx = 9 + j*8;

  1095.         for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){

  1096.             rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);

  1097.             avail[idx] = 1;

  1098.             if(!(cbp & 1)) continue;

  1099. 

  1100.             rv34_process_block(r, dst, s->linesize,

  1101.                                r->luma_vlc, 0, q_ac, q_ac);

  1102.         }

  1103.         dst += s->linesize * 4 - 4*4;

  1104.         intra_types += r->intra_types_stride;

  1105.     }

  1106. 

  1107.     intra_types -= r->intra_types_stride * 4;

  1108. 

  1109.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];

  1110.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];

  1111. 

  1112.     for(k = 0; k < 2; k++){

  1113.         dst = s->dest[1+k];

  1114.         fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);

  1115. 

  1116.         for(j = 0; j < 2; j++){

  1117.             int* acache = r->avail_cache + 6 + j*4;

  1118.             for(i = 0; i < 2; i++, cbp >>= 1, acache++){

  1119.                 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];

  1120.                 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);

  1121.                 acache[0] = 1;

  1122. 

  1123.                 if(!(cbp&1)) continue;

  1124. 

  1125.                 rv34_process_block(r, dst + 4*i, s->uvlinesize,

  1126.                                    r->chroma_vlc, 1, q_dc, q_ac);

  1127.             }

  1128. 

  1129.             dst += 4*s->uvlinesize;

  1130.         }

  1131.     }

  1132. }

  1133. 

  1134. static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)

  1135. {

  1136.     int d;

  1137.     d = motion_val[0][0] - motion_val[-step][0];

  1138.     if(d < -3 || d > 3)

  1139.         return 1;

  1140.     d = motion_val[0][1] - motion_val[-step][1];

  1141.     if(d < -3 || d > 3)

  1142.         return 1;

  1143.     return 0;

  1144. }

  1145. 

  1146. static int rv34_set_deblock_coef(RV34DecContext *r)

  1147. {

  1148.     MpegEncContext *s = &r->s;

  1149.     int hmvmask = 0, vmvmask = 0, i, j;

  1150.     int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;

  1151.     int16_t (*motion_val)[2] = &s->current_picture_ptr->f.motion_val[0][midx];

  1152.     for(j = 0; j < 16; j += 8){

  1153.         for(i = 0; i < 2; i++){

  1154.             if(is_mv_diff_gt_3(motion_val + i, 1))

  1155.                 vmvmask |= 0x11 << (j + i*2);

  1156.             if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))

  1157.                 hmvmask |= 0x03 << (j + i*2);

  1158.         }

  1159.         motion_val += s->b8_stride;

  1160.     }

  1161.     if(s->first_slice_line)

  1162.         hmvmask &= ~0x000F;

  1163.     if(!s->mb_x)

  1164.         vmvmask &= ~0x1111;

  1165.     if(r->rv30){ //RV30 marks both subblocks on the edge for filtering

  1166.         vmvmask |= (vmvmask & 0x4444) >> 1;

  1167.         hmvmask |= (hmvmask & 0x0F00) >> 4;

  1168.         if(s->mb_x)

  1169.             r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;

  1170.         if(!s->first_slice_line)

  1171.             r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;

  1172.     }

  1173.     return hmvmask | vmvmask;

  1174. }

  1175. 

  1176. static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)

  1177. {

  1178.     MpegEncContext *s   = &r->s;

  1179.     GetBitContext  *gb  = &s->gb;

  1180.     uint8_t        *dst = s->dest[0];

  1181.     int16_t        *ptr = s->block[0];

  1182.     int          mb_pos = s->mb_x + s->mb_y * s->mb_stride;

  1183.     int cbp, cbp2;

  1184.     int q_dc, q_ac, has_ac;

  1185.     int i, j;

  1186.     int dist;

  1187. 

  1188.     // Calculate which neighbours are available. Maybe it's worth optimizing too.

  1189.     memset(r->avail_cache, 0, sizeof(r->avail_cache));

  1190.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);

  1191.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;

  1192.     if(s->mb_x && dist)

  1193.         r->avail_cache[5] =

  1194.         r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];

  1195.     if(dist >= s->mb_width)

  1196.         r->avail_cache[2] =

  1197.         r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];

  1198.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)

  1199.         r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];

  1200.     if(s->mb_x && dist > s->mb_width)

  1201.         r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];

  1202. 

  1203.     s->qscale = r->si.quant;

  1204.     cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);

  1205.     r->cbp_luma  [mb_pos] = cbp;

  1206.     r->cbp_chroma[mb_pos] = cbp >> 16;

  1207.     r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];

  1208.     s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;

  1209. 

  1210.     if(cbp == -1)

  1211.         return -1;

  1212. 

  1213.     if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){

  1214.         if(r->is16) rv34_output_i16x16(r, intra_types, cbp);

  1215.         else        rv34_output_intra(r, intra_types, cbp);

  1216.         return 0;

  1217.     }

  1218. 

  1219.     if(r->is16){

  1220.         // Only for RV34_MB_P_MIX16x16

  1221.         LOCAL_ALIGNED_16(int16_t, block16, [16]);

  1222.         memset(block16, 0, 16 * sizeof(*block16));

  1223.         q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];

  1224.         q_ac = rv34_qscale_tab[s->qscale];

  1225.         if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))

  1226.             r->rdsp.rv34_inv_transform(block16);

  1227.         else

  1228.             r->rdsp.rv34_inv_transform_dc(block16);

  1229. 

  1230.         q_ac = rv34_qscale_tab[s->qscale];

  1231. 

  1232.         for(j = 0; j < 4; j++){

  1233.             for(i = 0; i < 4; i++, cbp >>= 1){

  1234.                 int      dc   = block16[i + j*4];

  1235. 

  1236.                 if(cbp & 1){

  1237.                     has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);

  1238.                 }else

  1239.                     has_ac = 0;

  1240. 

  1241.                 if(has_ac){

  1242.                     ptr[0] = dc;

  1243.                     r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);

  1244.                 }else

  1245.                     r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);

  1246.             }

  1247. 

  1248.             dst += 4*s->linesize;

  1249.         }

  1250. 

  1251.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);

  1252.     }else{

  1253.         q_ac = rv34_qscale_tab[s->qscale];

  1254. 

  1255.         for(j = 0; j < 4; j++){

  1256.             for(i = 0; i < 4; i++, cbp >>= 1){

  1257.                 if(!(cbp & 1)) continue;

  1258. 

  1259.                 rv34_process_block(r, dst + 4*i, s->linesize,

  1260.                                    r->luma_vlc, 0, q_ac, q_ac);

  1261.             }

  1262.             dst += 4*s->linesize;

  1263.         }

  1264.     }

  1265. 

  1266.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];

  1267.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];

  1268. 

  1269.     for(j = 1; j < 3; j++){

  1270.         dst = s->dest[j];

  1271.         for(i = 0; i < 4; i++, cbp >>= 1){

  1272.             uint8_t *pdst;

  1273.             if(!(cbp & 1)) continue;

  1274.             pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;

  1275. 

  1276.             rv34_process_block(r, pdst, s->uvlinesize,

  1277.                                r->chroma_vlc, 1, q_dc, q_ac);

  1278.         }

  1279.     }

  1280. 

  1281.     return 0;

  1282. }

  1283. 

  1284. static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)

  1285. {

  1286.     MpegEncContext *s = &r->s;

  1287.     int cbp, dist;

  1288.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;

  1289. 

  1290.     // Calculate which neighbours are available. Maybe it's worth optimizing too.

  1291.     memset(r->avail_cache, 0, sizeof(r->avail_cache));

  1292.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);

  1293.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;

  1294.     if(s->mb_x && dist)

  1295.         r->avail_cache[5] =

  1296.         r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];

  1297.     if(dist >= s->mb_width)

  1298.         r->avail_cache[2] =

  1299.         r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];

  1300.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)

  1301.         r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];

  1302.     if(s->mb_x && dist > s->mb_width)

  1303.         r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];

  1304. 

  1305.     s->qscale = r->si.quant;

  1306.     cbp = rv34_decode_intra_mb_header(r, intra_types);

  1307.     r->cbp_luma  [mb_pos] = cbp;

  1308.     r->cbp_chroma[mb_pos] = cbp >> 16;

  1309.     r->deblock_coefs[mb_pos] = 0xFFFF;

  1310.     s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;

  1311. 

  1312.     if(cbp == -1)

  1313.         return -1;

  1314. 

  1315.     if(r->is16){

  1316.         rv34_output_i16x16(r, intra_types, cbp);

  1317.         return 0;

  1318.     }

  1319. 

  1320.     rv34_output_intra(r, intra_types, cbp);

  1321.     return 0;

  1322. }

  1323. 

  1324. static int check_slice_end(RV34DecContext *r, MpegEncContext *s)

  1325. {

  1326.     int bits;

  1327.     if(s->mb_y >= s->mb_height)

  1328.         return 1;

  1329.     if(!s->mb_num_left)

  1330.         return 1;

  1331.     if(r->s.mb_skip_run > 1)

  1332.         return 0;

  1333.     bits = get_bits_left(&s->gb);

  1334.     if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))

  1335.         return 1;

  1336.     return 0;

  1337. }

  1338. 

  1339. 

  1340. static void rv34_decoder_free(RV34DecContext *r)

  1341. {

  1342.     av_freep(&r->intra_types_hist);

  1343.     r->intra_types = NULL;

  1344.     av_freep(&r->tmp_b_block_base);

  1345.     av_freep(&r->mb_type);

  1346.     av_freep(&r->cbp_luma);

  1347.     av_freep(&r->cbp_chroma);

  1348.     av_freep(&r->deblock_coefs);

  1349. }

  1350. 

  1351. 

  1352. static int rv34_decoder_alloc(RV34DecContext *r)

  1353. {

  1354.     r->intra_types_stride = r->s.mb_width * 4 + 4;

  1355. 

  1356.     r->cbp_chroma       = av_malloc(r->s.mb_stride * r->s.mb_height *

  1357.                                     sizeof(*r->cbp_chroma));

  1358.     r->cbp_luma         = av_malloc(r->s.mb_stride * r->s.mb_height *

  1359.                                     sizeof(*r->cbp_luma));

  1360.     r->deblock_coefs    = av_malloc(r->s.mb_stride * r->s.mb_height *

  1361.                                     sizeof(*r->deblock_coefs));

  1362.     r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *

  1363.                                     sizeof(*r->intra_types_hist));

  1364.     r->mb_type          = av_mallocz(r->s.mb_stride * r->s.mb_height *

  1365.                                      sizeof(*r->mb_type));

  1366. 

  1367.     if (!(r->cbp_chroma       && r->cbp_luma && r->deblock_coefs &&

  1368.           r->intra_types_hist && r->mb_type)) {

  1369.         rv34_decoder_free(r);

  1370.         return AVERROR(ENOMEM);

  1371.     }

  1372. 

  1373.     r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;

  1374. 

  1375.     return 0;

  1376. }

  1377. 

  1378. 

  1379. static int rv34_decoder_realloc(RV34DecContext *r)

  1380. {

  1381.     rv34_decoder_free(r);

  1382.     return rv34_decoder_alloc(r);

  1383. }

  1384. 

  1385. 

  1386. static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)

  1387. {

  1388.     MpegEncContext *s = &r->s;

  1389.     GetBitContext *gb = &s->gb;

  1390.     int mb_pos, slice_type;

  1391.     int res;

  1392. 

  1393.     init_get_bits(&r->s.gb, buf, buf_size*8);

  1394.     res = r->parse_slice_header(r, gb, &r->si);

  1395.     if(res < 0){

  1396.         av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");

  1397.         return -1;

  1398.     }

  1399. 

  1400.     slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;

  1401.     if (slice_type != s->pict_type) {

  1402.         av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");

  1403.         return AVERROR_INVALIDDATA;

  1404.     }

  1405. 

  1406.     r->si.end = end;

  1407.     s->qscale = r->si.quant;

  1408.     s->mb_num_left = r->si.end - r->si.start;

  1409.     r->s.mb_skip_run = 0;

  1410. 

  1411.     mb_pos = s->mb_x + s->mb_y * s->mb_width;

  1412.     if(r->si.start != mb_pos){

  1413.         av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);

  1414.         s->mb_x = r->si.start % s->mb_width;

  1415.         s->mb_y = r->si.start / s->mb_width;

  1416.     }

  1417.     memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));

  1418.     s->first_slice_line = 1;

  1419.     s->resync_mb_x = s->mb_x;

  1420.     s->resync_mb_y = s->mb_y;

  1421. 

  1422.     ff_init_block_index(s);

  1423.     while(!check_slice_end(r, s)) {

  1424.         ff_update_block_index(s);

  1425. 

  1426.         if(r->si.type)

  1427.             res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);

  1428.         else

  1429.             res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);

  1430.         if(res < 0){

  1431.             ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);

  1432.             return -1;

  1433.         }

  1434.         if (++s->mb_x == s->mb_width) {

  1435.             s->mb_x = 0;

  1436.             s->mb_y++;

  1437.             ff_init_block_index(s);

  1438. 

  1439.             memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1440.             memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1441. 

  1442.             if(r->loop_filter && s->mb_y >= 2)

  1443.                 r->loop_filter(r, s->mb_y - 2);

  1444. 

  1445.             if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))

  1446.                 ff_thread_report_progress(&s->current_picture_ptr->f,

  1447.                                           s->mb_y - 2, 0);

  1448. 

  1449.         }

  1450.         if(s->mb_x == s->resync_mb_x)

  1451.             s->first_slice_line=0;

  1452.         s->mb_num_left--;

  1453.     }

  1454.     ff_er_add_slice(&s->er, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);

  1455. 

  1456.     return s->mb_y == s->mb_height;

  1457. }

  1458. 

  1459. /** @} */ // recons group end

  1460. 

  1461. /**

  1462.  * Initialize decoder.

  1463.  */

  1464. av_cold int ff_rv34_decode_init(AVCodecContext *avctx)

  1465. {

  1466.     RV34DecContext *r = avctx->priv_data;

  1467.     MpegEncContext *s = &r->s;

  1468.     int ret;

  1469. 

  1470.     ff_MPV_decode_defaults(s);

  1471.     s->avctx      = avctx;

  1472.     s->out_format = FMT_H263;

  1473.     s->codec_id   = avctx->codec_id;

  1474. 

  1475.     s->width  = avctx->width;

  1476.     s->height = avctx->height;

  1477. 

  1478.     r->s.avctx = avctx;

  1479.     avctx->flags |= CODEC_FLAG_EMU_EDGE;

  1480.     r->s.flags |= CODEC_FLAG_EMU_EDGE;

  1481.     avctx->pix_fmt = AV_PIX_FMT_YUV420P;

  1482.     avctx->has_b_frames = 1;

  1483.     s->low_delay = 0;

  1484. 

  1485.     if ((ret = ff_MPV_common_init(s)) < 0)

  1486.         return ret;

  1487. 

  1488.     ff_h264_pred_init(&r->h, AV_CODEC_ID_RV40, 8, 1);

  1489. 

  1490. #if CONFIG_RV30_DECODER

  1491.     if (avctx->codec_id == AV_CODEC_ID_RV30)

  1492.         ff_rv30dsp_init(&r->rdsp);

  1493. #endif

  1494. #if CONFIG_RV40_DECODER

  1495.     if (avctx->codec_id == AV_CODEC_ID_RV40)

  1496.         ff_rv40dsp_init(&r->rdsp);

  1497. #endif

  1498. 

  1499.     if ((ret = rv34_decoder_alloc(r)) < 0)

  1500.         return ret;

  1501. 

  1502.     if(!intra_vlcs[0].cbppattern[0].bits)

  1503.         rv34_init_tables();

  1504. 

  1505.     return 0;

  1506. }

  1507. 

  1508. int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx)

  1509. {

  1510.     int err;

  1511.     RV34DecContext *r = avctx->priv_data;

  1512. 

  1513.     r->s.avctx = avctx;

  1514. 

  1515.     if (avctx->internal->is_copy) {

  1516.         r->tmp_b_block_base = NULL;

  1517.         if ((err = ff_MPV_common_init(&r->s)) < 0)

  1518.             return err;

  1519.         if ((err = rv34_decoder_alloc(r)) < 0)

  1520.             return err;

  1521.     }

  1522.     return 0;

  1523. }

  1524. 

  1525. int ff_rv34_decode_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)

  1526. {

  1527.     RV34DecContext *r = dst->priv_data, *r1 = src->priv_data;

  1528.     MpegEncContext * const s = &r->s, * const s1 = &r1->s;

  1529.     int err;

  1530. 

  1531.     if (dst == src || !s1->context_initialized)

  1532.         return 0;

  1533. 

  1534.     if (s->height != s1->height || s->width != s1->width) {

  1535.         s->height = s1->height;

  1536.         s->width  = s1->width;

  1537.         if ((err = ff_MPV_common_frame_size_change(s)) < 0)

  1538.             return err;

  1539.         if ((err = rv34_decoder_realloc(r)) < 0)

  1540.             return err;

  1541.     }

  1542. 

  1543.     if ((err = ff_mpeg_update_thread_context(dst, src)))

  1544.         return err;

  1545. 

  1546.     r->cur_pts  = r1->cur_pts;

  1547.     r->last_pts = r1->last_pts;

  1548.     r->next_pts = r1->next_pts;

  1549. 

  1550.     memset(&r->si, 0, sizeof(r->si));

  1551. 

  1552.     return 0;

  1553. }

  1554. 

  1555. static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)

  1556. {

  1557.     if(avctx->slice_count) return avctx->slice_offset[n];

  1558.     else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);

  1559. }

  1560. 

  1561. static int finish_frame(AVCodecContext *avctx, AVFrame *pict)

  1562. {

  1563.     RV34DecContext *r = avctx->priv_data;

  1564.     MpegEncContext *s = &r->s;

  1565.     int got_picture = 0;

  1566. 

  1567.     ff_er_frame_end(&s->er);

  1568.     ff_MPV_frame_end(s);

  1569.     s->mb_num_left = 0;

  1570. 

  1571.     if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))

  1572.         ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);

  1573. 

  1574.     if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {

  1575.         *pict = s->current_picture_ptr->f;

  1576.         got_picture = 1;

  1577.     } else if (s->last_picture_ptr != NULL) {

  1578.         *pict = s->last_picture_ptr->f;

  1579.         got_picture = 1;

  1580.     }

  1581.     if (got_picture)

  1582.         ff_print_debug_info(s, pict);

  1583. 

  1584.     return got_picture;

  1585. }

  1586. 

  1587. int ff_rv34_decode_frame(AVCodecContext *avctx,

  1588.                             void *data, int *got_picture_ptr,

  1589.                             AVPacket *avpkt)

  1590. {

  1591.     const uint8_t *buf = avpkt->data;

  1592.     int buf_size = avpkt->size;

  1593.     RV34DecContext *r = avctx->priv_data;

  1594.     MpegEncContext *s = &r->s;

  1595.     AVFrame *pict = data;

  1596.     SliceInfo si;

  1597.     int i;

  1598.     int slice_count;

  1599.     const uint8_t *slices_hdr = NULL;

  1600.     int last = 0;

  1601. 

  1602.     /* no supplementary picture */

  1603.     if (buf_size == 0) {

  1604.         /* special case for last picture */

  1605.         if (s->low_delay==0 && s->next_picture_ptr) {

  1606.             *pict = s->next_picture_ptr->f;

  1607.             s->next_picture_ptr = NULL;

  1608. 

  1609.             *got_picture_ptr = 1;

  1610.         }

  1611.         return 0;

  1612.     }

  1613. 

  1614.     if(!avctx->slice_count){

  1615.         slice_count = (*buf++) + 1;

  1616.         slices_hdr = buf + 4;

  1617.         buf += 8 * slice_count;

  1618.         buf_size -= 1 + 8 * slice_count;

  1619.     }else

  1620.         slice_count = avctx->slice_count;

  1621. 

  1622.     //parse first slice header to check whether this frame can be decoded

  1623.     if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||

  1624.        get_slice_offset(avctx, slices_hdr, 0) > buf_size){

  1625.         av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1626.         return AVERROR_INVALIDDATA;

  1627.     }

  1628.     init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);

  1629.     if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){

  1630.         av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");

  1631.         return AVERROR_INVALIDDATA;

  1632.     }

  1633.     if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) &&

  1634.         si.type == AV_PICTURE_TYPE_B) {

  1635.         av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "

  1636.                "reference data.\n");

  1637.         return AVERROR_INVALIDDATA;

  1638.     }

  1639.     if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)

  1640.        || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)

  1641.        ||  avctx->skip_frame >= AVDISCARD_ALL)

  1642.         return avpkt->size;

  1643. 

  1644.     /* first slice */

  1645.     if (si.start == 0) {

  1646.         if (s->mb_num_left > 0) {

  1647.             av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.",

  1648.                    s->mb_num_left);

  1649.             ff_er_frame_end(&s->er);

  1650.             ff_MPV_frame_end(s);

  1651.         }

  1652. 

  1653.         if (s->width != si.width || s->height != si.height) {

  1654.             int err;

  1655. 

  1656.             av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",

  1657.                    si.width, si.height);

  1658. 

  1659.             s->width  = si.width;

  1660.             s->height = si.height;

  1661.             avcodec_set_dimensions(s->avctx, s->width, s->height);

  1662.             if ((err = ff_MPV_common_frame_size_change(s)) < 0)

  1663.                 return err;

  1664.             if ((err = rv34_decoder_realloc(r)) < 0)

  1665.                 return err;

  1666.         }

  1667.         s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;

  1668.         if (ff_MPV_frame_start(s, s->avctx) < 0)

  1669.             return -1;

  1670.         ff_mpeg_er_frame_start(s);

  1671.         if (!r->tmp_b_block_base) {

  1672.             int i;

  1673. 

  1674.             r->tmp_b_block_base = av_malloc(s->linesize * 48);

  1675.             for (i = 0; i < 2; i++)

  1676.                 r->tmp_b_block_y[i] = r->tmp_b_block_base

  1677.                                       + i * 16 * s->linesize;

  1678.             for (i = 0; i < 4; i++)

  1679.                 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize

  1680.                                        + (i >> 1) * 8 * s->uvlinesize

  1681.                                        + (i &  1) * 16;

  1682.         }

  1683.         r->cur_pts = si.pts;

  1684.         if (s->pict_type != AV_PICTURE_TYPE_B) {

  1685.             r->last_pts = r->next_pts;

  1686.             r->next_pts = r->cur_pts;

  1687.         } else {

  1688.             int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);

  1689.             int dist0   = GET_PTS_DIFF(r->cur_pts,  r->last_pts);

  1690.             int dist1   = GET_PTS_DIFF(r->next_pts, r->cur_pts);

  1691. 

  1692.             if(!refdist){

  1693.                 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;

  1694.                 r->scaled_weight = 0;

  1695.             }else{

  1696.                 r->mv_weight1 = (dist0 << 14) / refdist;

  1697.                 r->mv_weight2 = (dist1 << 14) / refdist;

  1698.                 if((r->mv_weight1|r->mv_weight2) & 511){

  1699.                     r->weight1 = r->mv_weight1;

  1700.                     r->weight2 = r->mv_weight2;

  1701.                     r->scaled_weight = 0;

  1702.                 }else{

  1703.                     r->weight1 = r->mv_weight1 >> 9;

  1704.                     r->weight2 = r->mv_weight2 >> 9;

  1705.                     r->scaled_weight = 1;

  1706.                 }

  1707.             }

  1708.         }

  1709.         s->mb_x = s->mb_y = 0;

  1710.         ff_thread_finish_setup(s->avctx);

  1711.     } else if (HAVE_THREADS &&

  1712.                (s->avctx->active_thread_type & FF_THREAD_FRAME)) {

  1713.         av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "

  1714.                "multithreading mode (start MB is %d).\n", si.start);

  1715.         return AVERROR_INVALIDDATA;

  1716.     }

  1717. 

  1718.     for(i = 0; i < slice_count; i++){

  1719.         int offset = get_slice_offset(avctx, slices_hdr, i);

  1720.         int size;

  1721.         if(i+1 == slice_count)

  1722.             size = buf_size - offset;

  1723.         else

  1724.             size = get_slice_offset(avctx, slices_hdr, i+1) - offset;

  1725. 

  1726.         if(offset < 0 || offset > buf_size){

  1727.             av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1728.             break;

  1729.         }

  1730. 

  1731.         r->si.end = s->mb_width * s->mb_height;

  1732.         s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;

  1733. 

  1734.         if(i+1 < slice_count){

  1735.             if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||

  1736.                 get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {

  1737.                 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1738.                 break;

  1739.             }

  1740.             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);

  1741.             if(r->parse_slice_header(r, &r->s.gb, &si) < 0){

  1742.                 if(i+2 < slice_count)

  1743.                     size = get_slice_offset(avctx, slices_hdr, i+2) - offset;

  1744.                 else

  1745.                     size = buf_size - offset;

  1746.             }else

  1747.                 r->si.end = si.start;

  1748.         }

  1749.         if (size < 0 || size > buf_size - offset) {

  1750.             av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");

  1751.             break;

  1752.         }

  1753.         last = rv34_decode_slice(r, r->si.end, buf + offset, size);

  1754.         if(last)

  1755.             break;

  1756.     }

  1757. 

  1758.     if (s->current_picture_ptr) {

  1759.         if (last) {

  1760.             if(r->loop_filter)

  1761.                 r->loop_filter(r, s->mb_height - 1);

  1762. 

  1763.             *got_picture_ptr = finish_frame(avctx, pict);

  1764.         } else if (HAVE_THREADS &&

  1765.                    (s->avctx->active_thread_type & FF_THREAD_FRAME)) {

  1766.             av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");

  1767.             /* always mark the current frame as finished, frame-mt supports

  1768.              * only complete frames */

  1769.             ff_er_frame_end(&s->er);

  1770.             ff_MPV_frame_end(s);

  1771.             s->mb_num_left = 0;

  1772.             ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);

  1773.             return AVERROR_INVALIDDATA;

  1774.         }

  1775.     }

  1776. 

  1777.     return avpkt->size;

  1778. }

  1779. 

  1780. av_cold int ff_rv34_decode_end(AVCodecContext *avctx)

  1781. {

  1782.     RV34DecContext *r = avctx->priv_data;

  1783. 

  1784.     ff_MPV_common_end(&r->s);

  1785.     rv34_decoder_free(r);

  1786. 

  1787.     return 0;

  1788. }