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.
40183 Commits
32 Branches
428MB
Tree: 12eeced8b4
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '12eeced8b4'
${ noResults }
FFmpeg/libavcodec/rv34.c

1807 lines
63KB
Raw Blame History 

  1. /*

  2.  * RV30/40 decoder common data

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

  4.  *

  5.  * This file is part of 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/internal.h"

  28. 

  29. #include "avcodec.h"

  30. #include "dsputil.h"

  31. #include "mpegvideo.h"

  32. #include "golomb.h"

  33. #include "internal.h"

  34. #include "mathops.h"

  35. #include "rectangle.h"

  36. #include "thread.h"

  37. 

  38. #include "rv34vlc.h"

  39. #include "rv34data.h"

  40. #include "rv34.h"

  41. 

  42. //#define DEBUG

  43. 

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

  45. {

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

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

  48. }

  49. 

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

  51. static const int rv34_mb_type_to_lavc[12] = {

  52.     MB_TYPE_INTRA,

  53.     MB_TYPE_INTRA16x16              | MB_TYPE_SEPARATE_DC,

  54.     MB_TYPE_16x16   | MB_TYPE_L0,

  55.     MB_TYPE_8x8     | MB_TYPE_L0,

  56.     MB_TYPE_16x16   | MB_TYPE_L0,

  57.     MB_TYPE_16x16   | MB_TYPE_L1,

  58.     MB_TYPE_SKIP,

  59.     MB_TYPE_DIRECT2 | MB_TYPE_16x16,

  60.     MB_TYPE_16x8    | MB_TYPE_L0,

  61.     MB_TYPE_8x16    | MB_TYPE_L0,

  62.     MB_TYPE_16x16   | MB_TYPE_L0L1,

  63.     MB_TYPE_16x16   | MB_TYPE_L0    | MB_TYPE_SEPARATE_DC

  64. };

  65. 

  66. 

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

  68. 

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

  70. 

  71. /**

  72.  * @name RV30/40 VLC generating functions

  73.  * @{

  74.  */

  75. 

  76. static const int table_offs[] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  95. };

  96. 

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

  98. 

  99. /**

  100.  * Generate VLC from codeword lengths.

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

  102.  * @param size   length of input data

  103.  * @param vlc    output VLC

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

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

  106.  */

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

  108.                          const int num)

  109. {

  110.     int i;

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

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

  113.     uint8_t bits2[MAX_VLC_SIZE];

  114.     int maxbits = 0, realsize = 0;

  115. 

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

  117.         if(bits[i]){

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

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

  120.             realsize++;

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

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

  123.         }

  124.     }

  125. 

  126.     codes[0] = 0;

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

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

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

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

  131. 

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

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

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

  135.                        bits2, 1, 1,

  136.                        cw,    2, 2,

  137.                        syms,  2, 2, INIT_VLC_USE_NEW_STATIC);

  138. }

  139. 

  140. /**

  141.  * Initialize all tables.

  142.  */

  143. static av_cold void rv34_init_tables(void)

  144. {

  145.     int i, j, k;

  146. 

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

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

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

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

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

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

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

  154.             }

  155.         }

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

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

  158.         }

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

  160.     }

  161. 

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

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

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

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

  166.         }

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

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

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

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

  171.         }

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

  173.     }

  174. }

  175. 

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

  177. 

  178. /**

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

  180.  * @{

  181.  */

  182. 

  183. /**

  184.  * Decode coded block pattern.

  185.  */

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

  187. {

  188.     int pattern, code, cbp=0;

  189.     int ones;

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

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

  192.     const int *curshift = shifts;

  193.     int i, t, mask;

  194. 

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

  196.     pattern = code & 0xF;

  197.     code >>= 4;

  198. 

  199.     ones = rv34_count_ones[pattern];

  200. 

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

  202.         if(pattern & mask)

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

  204.     }

  205. 

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

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

  208.         if(t == 1)

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

  210.         if(t == 2)

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

  212.     }

  213.     return cbp;

  214. }

  215. 

  216. /**

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

  218.  */

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

  220. {

  221.     if(coef){

  222.         if(coef == esc){

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

  224.             if(coef > 23){

  225.                 coef -= 23;

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

  227.             }

  228.             coef += esc;

  229.         }

  230.         if(get_bits1(gb))

  231.             coef = -coef;

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

  233.     }

  234. }

  235. 

  236. /**

  237.  * Decode 2x2 subblock of coefficients.

  238.  */

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

  240. {

  241.     int flags = modulo_three_table[code];

  242. 

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

  244.     if(is_block2){

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

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

  247.     }else{

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

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

  250.     }

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

  252. }

  253. 

  254. /**

  255.  * Decode a single coefficient.

  256.  */

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

  258. {

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

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

  261. }

  262. 

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

  264.                                     int q_dc, int q_ac1, int q_ac2)

  265. {

  266.     int flags = modulo_three_table[code];

  267. 

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

  269.     if(is_block2){

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

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

  272.     }else{

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

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

  275.     }

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

  277. }

  278. 

  279. /**

  280.  * Decode coefficients for 4x4 block.

  281.  *

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

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

  284.  *  o--o

  285.  *    /

  286.  *   /

  287.  *  o--o

  288.  */

  289. 

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

  291. {

  292.     int code, pattern, has_ac = 1;

  293. 

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

  295. 

  296.     pattern = code & 0x7;

  297. 

  298.     code >>= 3;

  299. 

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

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

  302.     } else {

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

  304.         if (!pattern)

  305.             return 0;

  306.         has_ac = 0;

  307.     }

  308. 

  309.     if(pattern & 4){

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

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

  312.     }

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

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

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

  316.     }

  317.     if(pattern & 1){

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

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

  320.     }

  321.     return has_ac || pattern;

  322. }

  323. 

  324. /**

  325.  * @name RV30/40 bitstream parsing

  326.  * @{

  327.  */

  328. 

  329. /**

  330.  * Decode starting slice position.

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

  332.  */

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

  334. {

  335.     int i;

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

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

  338.             break;

  339.     return rv34_mb_bits_sizes[i];

  340. }

  341. 

  342. /**

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

  344.  */

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

  346. {

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

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

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

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

  351. }

  352. 

  353. /**

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

  355.  */

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

  357. {

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

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

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

  361.     int t;

  362. 

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

  364.     if(r->is16){

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

  366.         r->block_type = RV34_MB_TYPE_INTRA16x16;

  367.         t = get_bits(gb, 2);

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

  369.         r->luma_vlc   = 2;

  370.     }else{

  371.         if(!r->rv30){

  372.             if(!get_bits1(gb))

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

  374.         }

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

  376.         r->block_type = RV34_MB_TYPE_INTRA;

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

  378.             return -1;

  379.         r->luma_vlc   = 1;

  380.     }

  381. 

  382.     r->chroma_vlc = 0;

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

  384. 

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

  386. }

  387. 

  388. /**

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

  390.  */

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

  392. {

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

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

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

  396.     int i, t;

  397. 

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

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

  400.         return -1;

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

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

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

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

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

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

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

  408.     }

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

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

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

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

  413.         return 0;

  414.     }

  415.     r->chroma_vlc = 1;

  416.     r->luma_vlc   = 0;

  417. 

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

  419.         if(r->is16){

  420.             t = get_bits(gb, 2);

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

  422.             r->luma_vlc   = 2;

  423.         }else{

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

  425.                 return -1;

  426.             r->luma_vlc   = 1;

  427.         }

  428.         r->chroma_vlc = 0;

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

  430.     }else{

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

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

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

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

  435.             r->is16 = 1;

  436.             r->chroma_vlc = 1;

  437.             r->luma_vlc   = 2;

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

  439.         }

  440.     }

  441. 

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

  443. }

  444. 

  445. /** @} */ //bitstream functions

  446. 

  447. /**

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

  449.  * @{

  450.  */

  451. 

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

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

  454. 

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

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

  457. 

  458. /** availability index for subblocks */

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

  460. 

  461. /**

  462.  * motion vector prediction

  463.  *

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

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

  466.  * some other vectors may be used instead.

  467.  */

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

  469. {

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

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

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

  473.     int i, j;

  474.     int mx, my;

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

  476.     int c_off = part_sizes_w[block_type];

  477. 

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

  479.     if(subblock_no == 3)

  480.         c_off = -1;

  481. 

  482.     if(avail[-1]){

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

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

  485.     }

  486.     if(avail[-4]){

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

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

  489.     }else{

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

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

  492.     }

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

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

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

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

  497.         }else{

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

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

  500.         }

  501.     }else{

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

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

  504.     }

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

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

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

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

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

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

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

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

  513.         }

  514.     }

  515. }

  516. 

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

  518. 

  519. /**

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

  521.  */

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

  523. {

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

  525. 

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

  527. }

  528. 

  529. /**

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

  531.  */

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

  533.                                       int A_avail, int B_avail, int C_avail,

  534.                                       int *mx, int *my)

  535. {

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

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

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

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

  540.             *mx /= 2;

  541.             *my /= 2;

  542.         }

  543.     }else{

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

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

  546.     }

  547. }

  548. 

  549. /**

  550.  * motion vector prediction for B-frames

  551.  */

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

  553. {

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

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

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

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

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

  559.     int mx, my;

  560.     int i, j;

  561.     Picture *cur_pic = s->current_picture_ptr;

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

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

  564. 

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

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

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

  568.         has_A = 1;

  569.     }

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

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

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

  573.         has_B = 1;

  574.     }

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

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

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

  578.         has_C = 1;

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

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

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

  582.         has_C = 1;

  583.     }

  584. 

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

  586. 

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

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

  589. 

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

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

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

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

  594.         }

  595.     }

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

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

  598.     }

  599. }

  600. 

  601. /**

  602.  * motion vector prediction - RV3 version

  603.  */

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

  605. {

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

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

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

  609.     int i, j, k;

  610.     int mx, my;

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

  612. 

  613.     if(avail[-1]){

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

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

  616.     }

  617.     if(avail[-4]){

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

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

  620.     }else{

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

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

  623.     }

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

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

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

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

  628.         }else{

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

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

  631.         }

  632.     }else{

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

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

  635.     }

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

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

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

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

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

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

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

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

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

  645.             }

  646.         }

  647.     }

  648. }

  649. 

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

  651. 

  652. /**

  653.  * generic motion compensation function

  654.  *

  655.  * @param r decoder context

  656.  * @param block_type type of the current block

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

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

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

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

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

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

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

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

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

  666.  */

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

  668.                           const int xoff, const int yoff, int mv_off,

  669.                           const int width, const int height, int dir,

  670.                           const int thirdpel, int weighted,

  671.                           qpel_mc_func (*qpel_mc)[16],

  672.                           h264_chroma_mc_func (*chroma_mc))

  673. {

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

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

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

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

  678.     int is16x16 = 1;

  679. 

  680.     if(thirdpel){

  681.         int chroma_mx, chroma_my;

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

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

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

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

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

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

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

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

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

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

  692.     }else{

  693.         int cx, cy;

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

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

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

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

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

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

  700.         umx = cx >> 2;

  701.         umy = cy >> 2;

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

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

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

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

  706.             uvmx = uvmy = 4;

  707.     }

  708. 

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

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

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

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

  713.         ff_thread_await_progress(f, mb_row, 0);

  714.     }

  715. 

  716.     dxy = ly*4 + lx;

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  731. 

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

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

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

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

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

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

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

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

  740.         srcU = uvbuf;

  741.         srcV = uvbuf + 16;

  742.     }

  743.     if(!weighted){

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

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

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

  747.     }else{

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

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

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

  751.     }

  752. 

  753.     if(block_type == RV34_MB_P_16x8){

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

  755.         Y    += 8;

  756.         srcY += 8;

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

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

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

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

  761.     }

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

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

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

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

  766. }

  767. 

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

  769.                         const int xoff, const int yoff, int mv_off,

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

  771. {

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

  773.             r->rdsp.put_pixels_tab,

  774.             r->rdsp.put_chroma_pixels_tab);

  775. }

  776. 

  777. static void rv4_weight(RV34DecContext *r)

  778. {

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

  780.                                                         r->tmp_b_block_y[0],

  781.                                                         r->tmp_b_block_y[1],

  782.                                                         r->weight1,

  783.                                                         r->weight2,

  784.                                                         r->s.linesize);

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

  786.                                                         r->tmp_b_block_uv[0],

  787.                                                         r->tmp_b_block_uv[2],

  788.                                                         r->weight1,

  789.                                                         r->weight2,

  790.                                                         r->s.uvlinesize);

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

  792.                                                         r->tmp_b_block_uv[1],

  793.                                                         r->tmp_b_block_uv[3],

  794.                                                         r->weight1,

  795.                                                         r->weight2,

  796.                                                         r->s.uvlinesize);

  797. }

  798. 

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

  800. {

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

  802. 

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

  804.             r->rdsp.put_pixels_tab,

  805.             r->rdsp.put_chroma_pixels_tab);

  806.     if(!weighted){

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

  808.                 r->rdsp.avg_pixels_tab,

  809.                 r->rdsp.avg_chroma_pixels_tab);

  810.     }else{

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

  812.                 r->rdsp.put_pixels_tab,

  813.                 r->rdsp.put_chroma_pixels_tab);

  814.         rv4_weight(r);

  815.     }

  816. }

  817. 

  818. static void rv34_mc_2mv_skip(RV34DecContext *r)

  819. {

  820.     int i, j;

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

  822. 

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

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

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

  826.                      weighted,

  827.                      r->rdsp.put_pixels_tab,

  828.                      r->rdsp.put_chroma_pixels_tab);

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

  830.                      weighted,

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

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

  833.         }

  834.     if(weighted)

  835.         rv4_weight(r);

  836. }

  837. 

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

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

  840. 

  841. /**

  842.  * Decode motion vector differences

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

  844.  */

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

  846. {

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

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

  849.     int i, j, k, l;

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

  851.     int next_bt;

  852. 

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

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

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

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

  857.     }

  858.     switch(block_type){

  859.     case RV34_MB_TYPE_INTRA:

  860.     case RV34_MB_TYPE_INTRA16x16:

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

  862.         return 0;

  863.     case RV34_MB_SKIP:

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

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

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

  867.             break;

  868.         }

  869.     case RV34_MB_B_DIRECT:

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

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

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

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

  874. 

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

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

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

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

  879.         }else

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

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

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

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

  884.                             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]);

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

  886.             rv34_mc_2mv(r, block_type);

  887.         else

  888.             rv34_mc_2mv_skip(r);

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

  890.         break;

  891.     case RV34_MB_P_16x16:

  892.     case RV34_MB_P_MIX16x16:

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

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

  895.         break;

  896.     case RV34_MB_B_FORWARD:

  897.     case RV34_MB_B_BACKWARD:

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

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

  900.         if(r->rv30)

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

  902.         else

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

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

  905.         break;

  906.     case RV34_MB_P_16x8:

  907.     case RV34_MB_P_8x16:

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

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

  910.         if(block_type == RV34_MB_P_16x8){

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

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

  913.         }

  914.         if(block_type == RV34_MB_P_8x16){

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

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

  917.         }

  918.         break;

  919.     case RV34_MB_B_BIDIR:

  920.         rv34_pred_mv_b  (r, block_type, 0);

  921.         rv34_pred_mv_b  (r, block_type, 1);

  922.         rv34_mc_2mv     (r, block_type);

  923.         break;

  924.     case RV34_MB_P_8x8:

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

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

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

  928.         }

  929.         break;

  930.     }

  931. 

  932.     return 0;

  933. }

  934. /** @} */ // mv group

  935. 

  936. /**

  937.  * @name Macroblock reconstruction functions

  938.  * @{

  939.  */

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

  941. static const int ittrans[9] = {

  942.  DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,

  943.  VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,

  944. };

  945. 

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

  947. static const int ittrans16[4] = {

  948.  DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,

  949. };

  950. 

  951. /**

  952.  * Perform 4x4 intra prediction.

  953.  */

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

  955. {

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

  957.     uint32_t topleft;

  958. 

  959.     if(!up && !left)

  960.         itype = DC_128_PRED;

  961.     else if(!up){

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

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

  964.     }else if(!left){

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

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

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

  968.     }

  969.     if(!down){

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

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

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

  973.     }

  974.     if(!right && up){

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

  976.         prev = (uint8_t*)&topleft;

  977.     }

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

  979. }

  980. 

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

  982. {

  983.     if(!up && !left)

  984.         itype = DC_128_PRED8x8;

  985.     else if(!up){

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

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

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

  989.     }else if(!left){

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

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

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

  993.     }

  994.     return itype;

  995. }

  996. 

  997. static inline void rv34_process_block(RV34DecContext *r,

  998.                                       uint8_t *pdst, int stride,

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

  1000. {

  1001.     MpegEncContext *s = &r->s;

  1002.     DCTELEM *ptr = s->block[0];

  1003.     int has_ac = rv34_decode_block(ptr, &s->gb, r->cur_vlcs,

  1004.                                    fc, sc, q_dc, q_ac, q_ac);

  1005.     if(has_ac){

  1006.         r->rdsp.rv34_idct_add(pdst, stride, ptr);

  1007.     }else{

  1008.         r->rdsp.rv34_idct_dc_add(pdst, stride, ptr[0]);

  1009.         ptr[0] = 0;

  1010.     }

  1011. }

  1012. 

  1013. static void rv34_output_i16x16(RV34DecContext *r, int8_t *intra_types, int cbp)

  1014. {

  1015.     LOCAL_ALIGNED_16(DCTELEM, block16, [16]);

  1016.     MpegEncContext *s    = &r->s;

  1017.     GetBitContext  *gb   = &s->gb;

  1018.     int             q_dc = rv34_qscale_tab[ r->luma_dc_quant_i[s->qscale] ],

  1019.                     q_ac = rv34_qscale_tab[s->qscale];

  1020.     uint8_t        *dst  = s->dest[0];

  1021.     DCTELEM        *ptr  = s->block[0];

  1022.     int i, j, itype, has_ac;

  1023. 

  1024.     memset(block16, 0, 16 * sizeof(*block16));

  1025. 

  1026.     has_ac = rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac);

  1027.     if(has_ac)

  1028.         r->rdsp.rv34_inv_transform(block16);

  1029.     else

  1030.         r->rdsp.rv34_inv_transform_dc(block16);

  1031. 

  1032.     itype = ittrans16[intra_types[0]];

  1033.     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1034.     r->h.pred16x16[itype](dst, s->linesize);

  1035. 

  1036.     for(j = 0; j < 4; j++){

  1037.         for(i = 0; i < 4; i++, cbp >>= 1){

  1038.             int dc = block16[i + j*4];

  1039. 

  1040.             if(cbp & 1){

  1041.                 has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);

  1042.             }else

  1043.                 has_ac = 0;

  1044. 

  1045.             if(has_ac){

  1046.                 ptr[0] = dc;

  1047.                 r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);

  1048.             }else

  1049.                 r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);

  1050.         }

  1051. 

  1052.         dst += 4*s->linesize;

  1053.     }

  1054. 

  1055.     itype = ittrans16[intra_types[0]];

  1056.     if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;

  1057.     itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1058. 

  1059.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];

  1060.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];

  1061. 

  1062.     for(j = 1; j < 3; j++){

  1063.         dst = s->dest[j];

  1064.         r->h.pred8x8[itype](dst, s->uvlinesize);

  1065.         for(i = 0; i < 4; i++, cbp >>= 1){

  1066.             uint8_t *pdst;

  1067.             if(!(cbp & 1)) continue;

  1068.             pdst   = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;

  1069. 

  1070.             rv34_process_block(r, pdst, s->uvlinesize,

  1071.                                r->chroma_vlc, 1, q_dc, q_ac);

  1072.         }

  1073.     }

  1074. }

  1075. 

  1076. static void rv34_output_intra(RV34DecContext *r, int8_t *intra_types, int cbp)

  1077. {

  1078.     MpegEncContext *s   = &r->s;

  1079.     uint8_t        *dst = s->dest[0];

  1080.     int      avail[6*8] = {0};

  1081.     int i, j, k;

  1082.     int idx, q_ac, q_dc;

  1083. 

  1084.     // Set neighbour information.

  1085.     if(r->avail_cache[1])

  1086.         avail[0] = 1;

  1087.     if(r->avail_cache[2])

  1088.         avail[1] = avail[2] = 1;

  1089.     if(r->avail_cache[3])

  1090.         avail[3] = avail[4] = 1;

  1091.     if(r->avail_cache[4])

  1092.         avail[5] = 1;

  1093.     if(r->avail_cache[5])

  1094.         avail[8] = avail[16] = 1;

  1095.     if(r->avail_cache[9])

  1096.         avail[24] = avail[32] = 1;

  1097. 

  1098.     q_ac = rv34_qscale_tab[s->qscale];

  1099.     for(j = 0; j < 4; j++){

  1100.         idx = 9 + j*8;

  1101.         for(i = 0; i < 4; i++, cbp >>= 1, dst += 4, idx++){

  1102.             rv34_pred_4x4_block(r, dst, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);

  1103.             avail[idx] = 1;

  1104.             if(!(cbp & 1)) continue;

  1105. 

  1106.             rv34_process_block(r, dst, s->linesize,

  1107.                                r->luma_vlc, 0, q_ac, q_ac);

  1108.         }

  1109.         dst += s->linesize * 4 - 4*4;

  1110.         intra_types += r->intra_types_stride;

  1111.     }

  1112. 

  1113.     intra_types -= r->intra_types_stride * 4;

  1114. 

  1115.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];

  1116.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];

  1117. 

  1118.     for(k = 0; k < 2; k++){

  1119.         dst = s->dest[1+k];

  1120.         fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);

  1121. 

  1122.         for(j = 0; j < 2; j++){

  1123.             int* acache = r->avail_cache + 6 + j*4;

  1124.             for(i = 0; i < 2; i++, cbp >>= 1, acache++){

  1125.                 int itype = ittrans[intra_types[i*2+j*2*r->intra_types_stride]];

  1126.                 rv34_pred_4x4_block(r, dst+4*i, s->uvlinesize, itype, acache[-4], acache[-1], !i && !j, acache[-3]);

  1127.                 acache[0] = 1;

  1128. 

  1129.                 if(!(cbp&1)) continue;

  1130. 

  1131.                 rv34_process_block(r, dst + 4*i, s->uvlinesize,

  1132.                                    r->chroma_vlc, 1, q_dc, q_ac);

  1133.             }

  1134. 

  1135.             dst += 4*s->uvlinesize;

  1136.         }

  1137.     }

  1138. }

  1139. 

  1140. static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)

  1141. {

  1142.     int d;

  1143.     d = motion_val[0][0] - motion_val[-step][0];

  1144.     if(d < -3 || d > 3)

  1145.         return 1;

  1146.     d = motion_val[0][1] - motion_val[-step][1];

  1147.     if(d < -3 || d > 3)

  1148.         return 1;

  1149.     return 0;

  1150. }

  1151. 

  1152. static int rv34_set_deblock_coef(RV34DecContext *r)

  1153. {

  1154.     MpegEncContext *s = &r->s;

  1155.     int hmvmask = 0, vmvmask = 0, i, j;

  1156.     int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;

  1157.     int16_t (*motion_val)[2] = &s->current_picture_ptr->f.motion_val[0][midx];

  1158.     for(j = 0; j < 16; j += 8){

  1159.         for(i = 0; i < 2; i++){

  1160.             if(is_mv_diff_gt_3(motion_val + i, 1))

  1161.                 vmvmask |= 0x11 << (j + i*2);

  1162.             if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))

  1163.                 hmvmask |= 0x03 << (j + i*2);

  1164.         }

  1165.         motion_val += s->b8_stride;

  1166.     }

  1167.     if(s->first_slice_line)

  1168.         hmvmask &= ~0x000F;

  1169.     if(!s->mb_x)

  1170.         vmvmask &= ~0x1111;

  1171.     if(r->rv30){ //RV30 marks both subblocks on the edge for filtering

  1172.         vmvmask |= (vmvmask & 0x4444) >> 1;

  1173.         hmvmask |= (hmvmask & 0x0F00) >> 4;

  1174.         if(s->mb_x)

  1175.             r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;

  1176.         if(!s->first_slice_line)

  1177.             r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;

  1178.     }

  1179.     return hmvmask | vmvmask;

  1180. }

  1181. 

  1182. static int rv34_decode_inter_macroblock(RV34DecContext *r, int8_t *intra_types)

  1183. {

  1184.     MpegEncContext *s   = &r->s;

  1185.     GetBitContext  *gb  = &s->gb;

  1186.     uint8_t        *dst = s->dest[0];

  1187.     DCTELEM        *ptr = s->block[0];

  1188.     int          mb_pos = s->mb_x + s->mb_y * s->mb_stride;

  1189.     int cbp, cbp2;

  1190.     int q_dc, q_ac, has_ac;

  1191.     int i, j;

  1192.     int dist;

  1193. 

  1194.     // Calculate which neighbours are available. Maybe it's worth optimizing too.

  1195.     memset(r->avail_cache, 0, sizeof(r->avail_cache));

  1196.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);

  1197.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;

  1198.     if(s->mb_x && dist)

  1199.         r->avail_cache[5] =

  1200.         r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];

  1201.     if(dist >= s->mb_width)

  1202.         r->avail_cache[2] =

  1203.         r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];

  1204.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)

  1205.         r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];

  1206.     if(s->mb_x && dist > s->mb_width)

  1207.         r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];

  1208. 

  1209.     s->qscale = r->si.quant;

  1210.     cbp = cbp2 = rv34_decode_inter_mb_header(r, intra_types);

  1211.     r->cbp_luma  [mb_pos] = cbp;

  1212.     r->cbp_chroma[mb_pos] = cbp >> 16;

  1213.     r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];

  1214.     s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;

  1215. 

  1216.     if(cbp == -1)

  1217.         return -1;

  1218. 

  1219.     if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos])){

  1220.         if(r->is16) rv34_output_i16x16(r, intra_types, cbp);

  1221.         else        rv34_output_intra(r, intra_types, cbp);

  1222.         return 0;

  1223.     }

  1224. 

  1225.     if(r->is16){

  1226.         // Only for RV34_MB_P_MIX16x16

  1227.         LOCAL_ALIGNED_16(DCTELEM, block16, [16]);

  1228.         memset(block16, 0, 16 * sizeof(*block16));

  1229.         q_dc = rv34_qscale_tab[ r->luma_dc_quant_p[s->qscale] ];

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

  1231.         if (rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0, q_dc, q_dc, q_ac))

  1232.             r->rdsp.rv34_inv_transform(block16);

  1233.         else

  1234.             r->rdsp.rv34_inv_transform_dc(block16);

  1235. 

  1236.         q_ac = rv34_qscale_tab[s->qscale];

  1237. 

  1238.         for(j = 0; j < 4; j++){

  1239.             for(i = 0; i < 4; i++, cbp >>= 1){

  1240.                 int      dc   = block16[i + j*4];

  1241. 

  1242.                 if(cbp & 1){

  1243.                     has_ac = rv34_decode_block(ptr, gb, r->cur_vlcs, r->luma_vlc, 0, q_ac, q_ac, q_ac);

  1244.                 }else

  1245.                     has_ac = 0;

  1246. 

  1247.                 if(has_ac){

  1248.                     ptr[0] = dc;

  1249.                     r->rdsp.rv34_idct_add(dst+4*i, s->linesize, ptr);

  1250.                 }else

  1251.                     r->rdsp.rv34_idct_dc_add(dst+4*i, s->linesize, dc);

  1252.             }

  1253. 

  1254.             dst += 4*s->linesize;

  1255.         }

  1256. 

  1257.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);

  1258.     }else{

  1259.         q_ac = rv34_qscale_tab[s->qscale];

  1260. 

  1261.         for(j = 0; j < 4; j++){

  1262.             for(i = 0; i < 4; i++, cbp >>= 1){

  1263.                 if(!(cbp & 1)) continue;

  1264. 

  1265.                 rv34_process_block(r, dst + 4*i, s->linesize,

  1266.                                    r->luma_vlc, 0, q_ac, q_ac);

  1267.             }

  1268.             dst += 4*s->linesize;

  1269.         }

  1270.     }

  1271. 

  1272.     q_dc = rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]];

  1273.     q_ac = rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]];

  1274. 

  1275.     for(j = 1; j < 3; j++){

  1276.         dst = s->dest[j];

  1277.         for(i = 0; i < 4; i++, cbp >>= 1){

  1278.             uint8_t *pdst;

  1279.             if(!(cbp & 1)) continue;

  1280.             pdst = dst + (i&1)*4 + (i&2)*2*s->uvlinesize;

  1281. 

  1282.             rv34_process_block(r, pdst, s->uvlinesize,

  1283.                                r->chroma_vlc, 1, q_dc, q_ac);

  1284.         }

  1285.     }

  1286. 

  1287.     return 0;

  1288. }

  1289. 

  1290. static int rv34_decode_intra_macroblock(RV34DecContext *r, int8_t *intra_types)

  1291. {

  1292.     MpegEncContext *s = &r->s;

  1293.     int cbp, dist;

  1294.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;

  1295. 

  1296.     // Calculate which neighbours are available. Maybe it's worth optimizing too.

  1297.     memset(r->avail_cache, 0, sizeof(r->avail_cache));

  1298.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);

  1299.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;

  1300.     if(s->mb_x && dist)

  1301.         r->avail_cache[5] =

  1302.         r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];

  1303.     if(dist >= s->mb_width)

  1304.         r->avail_cache[2] =

  1305.         r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];

  1306.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)

  1307.         r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];

  1308.     if(s->mb_x && dist > s->mb_width)

  1309.         r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];

  1310. 

  1311.     s->qscale = r->si.quant;

  1312.     cbp = rv34_decode_intra_mb_header(r, intra_types);

  1313.     r->cbp_luma  [mb_pos] = cbp;

  1314.     r->cbp_chroma[mb_pos] = cbp >> 16;

  1315.     r->deblock_coefs[mb_pos] = 0xFFFF;

  1316.     s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;

  1317. 

  1318.     if(cbp == -1)

  1319.         return -1;

  1320. 

  1321.     if(r->is16){

  1322.         rv34_output_i16x16(r, intra_types, cbp);

  1323.         return 0;

  1324.     }

  1325. 

  1326.     rv34_output_intra(r, intra_types, cbp);

  1327.     return 0;

  1328. }

  1329. 

  1330. static int check_slice_end(RV34DecContext *r, MpegEncContext *s)

  1331. {

  1332.     int bits;

  1333.     if(s->mb_y >= s->mb_height)

  1334.         return 1;

  1335.     if(!s->mb_num_left)

  1336.         return 1;

  1337.     if(r->s.mb_skip_run > 1)

  1338.         return 0;

  1339.     bits = get_bits_left(&s->gb);

  1340.     if(bits <= 0 || (bits < 8 && !show_bits(&s->gb, bits)))

  1341.         return 1;

  1342.     return 0;

  1343. }

  1344. 

  1345. 

  1346. static void rv34_decoder_free(RV34DecContext *r)

  1347. {

  1348.     av_freep(&r->intra_types_hist);

  1349.     r->intra_types = NULL;

  1350.     av_freep(&r->tmp_b_block_base);

  1351.     av_freep(&r->mb_type);

  1352.     av_freep(&r->cbp_luma);

  1353.     av_freep(&r->cbp_chroma);

  1354.     av_freep(&r->deblock_coefs);

  1355. }

  1356. 

  1357. 

  1358. static int rv34_decoder_alloc(RV34DecContext *r)

  1359. {

  1360.     r->intra_types_stride = r->s.mb_width * 4 + 4;

  1361. 

  1362.     r->cbp_chroma       = av_malloc(r->s.mb_stride * r->s.mb_height *

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

  1364.     r->cbp_luma         = av_malloc(r->s.mb_stride * r->s.mb_height *

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

  1366.     r->deblock_coefs    = av_malloc(r->s.mb_stride * r->s.mb_height *

  1367.                                     sizeof(*r->deblock_coefs));

  1368.     r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 *

  1369.                                     sizeof(*r->intra_types_hist));

  1370.     r->mb_type          = av_mallocz(r->s.mb_stride * r->s.mb_height *

  1371.                                      sizeof(*r->mb_type));

  1372. 

  1373.     if (!(r->cbp_chroma       && r->cbp_luma && r->deblock_coefs &&

  1374.           r->intra_types_hist && r->mb_type)) {

  1375.         rv34_decoder_free(r);

  1376.         return AVERROR(ENOMEM);

  1377.     }

  1378. 

  1379.     r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;

  1380. 

  1381.     return 0;

  1382. }

  1383. 

  1384. 

  1385. static int rv34_decoder_realloc(RV34DecContext *r)

  1386. {

  1387.     rv34_decoder_free(r);

  1388.     return rv34_decoder_alloc(r);

  1389. }

  1390. 

  1391. 

  1392. static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)

  1393. {

  1394.     MpegEncContext *s = &r->s;

  1395.     GetBitContext *gb = &s->gb;

  1396.     int mb_pos, slice_type;

  1397.     int res;

  1398. 

  1399.     init_get_bits(&r->s.gb, buf, buf_size*8);

  1400.     res = r->parse_slice_header(r, gb, &r->si);

  1401.     if(res < 0){

  1402.         av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");

  1403.         return -1;

  1404.     }

  1405. 

  1406.     slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;

  1407.     if (slice_type != s->pict_type) {

  1408.         av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");

  1409.         return AVERROR_INVALIDDATA;

  1410.     }

  1411.     if (s->width != r->si.width || s->height != r->si.height) {

  1412.         av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");

  1413.         return AVERROR_INVALIDDATA;

  1414.     }

  1415. 

  1416.     r->si.end = end;

  1417.     s->qscale = r->si.quant;

  1418.     s->mb_num_left = r->si.end - r->si.start;

  1419.     r->s.mb_skip_run = 0;

  1420. 

  1421.     mb_pos = s->mb_x + s->mb_y * s->mb_width;

  1422.     if(r->si.start != mb_pos){

  1423.         av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);

  1424.         s->mb_x = r->si.start % s->mb_width;

  1425.         s->mb_y = r->si.start / s->mb_width;

  1426.     }

  1427.     memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));

  1428.     s->first_slice_line = 1;

  1429.     s->resync_mb_x = s->mb_x;

  1430.     s->resync_mb_y = s->mb_y;

  1431. 

  1432.     ff_init_block_index(s);

  1433.     while(!check_slice_end(r, s)) {

  1434.         ff_update_block_index(s);

  1435. 

  1436.         if(r->si.type)

  1437.             res = rv34_decode_inter_macroblock(r, r->intra_types + s->mb_x * 4 + 4);

  1438.         else

  1439.             res = rv34_decode_intra_macroblock(r, r->intra_types + s->mb_x * 4 + 4);

  1440.         if(res < 0){

  1441.             ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_ERROR);

  1442.             return -1;

  1443.         }

  1444.         if (++s->mb_x == s->mb_width) {

  1445.             s->mb_x = 0;

  1446.             s->mb_y++;

  1447.             ff_init_block_index(s);

  1448. 

  1449.             memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1450.             memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1451. 

  1452.             if(r->loop_filter && s->mb_y >= 2)

  1453.                 r->loop_filter(r, s->mb_y - 2);

  1454. 

  1455.             if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))

  1456.                 ff_thread_report_progress(&s->current_picture_ptr->f,

  1457.                                           s->mb_y - 2, 0);

  1458. 

  1459.         }

  1460.         if(s->mb_x == s->resync_mb_x)

  1461.             s->first_slice_line=0;

  1462.         s->mb_num_left--;

  1463.     }

  1464.     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END);

  1465. 

  1466.     return s->mb_y == s->mb_height;

  1467. }

  1468. 

  1469. /** @} */ // recons group end

  1470. 

  1471. /**

  1472.  * Initialize decoder.

  1473.  */

  1474. av_cold int ff_rv34_decode_init(AVCodecContext *avctx)

  1475. {

  1476.     RV34DecContext *r = avctx->priv_data;

  1477.     MpegEncContext *s = &r->s;

  1478.     int ret;

  1479. 

  1480.     ff_MPV_decode_defaults(s);

  1481.     s->avctx      = avctx;

  1482.     s->out_format = FMT_H263;

  1483.     s->codec_id   = avctx->codec_id;

  1484. 

  1485.     s->width  = avctx->width;

  1486.     s->height = avctx->height;

  1487. 

  1488.     r->s.avctx = avctx;

  1489.     avctx->flags |= CODEC_FLAG_EMU_EDGE;

  1490.     r->s.flags |= CODEC_FLAG_EMU_EDGE;

  1491.     avctx->pix_fmt = PIX_FMT_YUV420P;

  1492.     avctx->has_b_frames = 1;

  1493.     s->low_delay = 0;

  1494. 

  1495.     if ((ret = ff_MPV_common_init(s)) < 0)

  1496.         return ret;

  1497. 

  1498.     ff_h264_pred_init(&r->h, CODEC_ID_RV40, 8, 1);

  1499. 

  1500. #if CONFIG_RV30_DECODER

  1501.     if (avctx->codec_id == CODEC_ID_RV30)

  1502.         ff_rv30dsp_init(&r->rdsp, &r->s.dsp);

  1503. #endif

  1504. #if CONFIG_RV40_DECODER

  1505.     if (avctx->codec_id == CODEC_ID_RV40)

  1506.         ff_rv40dsp_init(&r->rdsp, &r->s.dsp);

  1507. #endif

  1508. 

  1509.     if ((ret = rv34_decoder_alloc(r)) < 0)

  1510.         return ret;

  1511. 

  1512.     if(!intra_vlcs[0].cbppattern[0].bits)

  1513.         rv34_init_tables();

  1514. 

  1515.     return 0;

  1516. }

  1517. 

  1518. int ff_rv34_decode_init_thread_copy(AVCodecContext *avctx)

  1519. {

  1520.     int err;

  1521.     RV34DecContext *r = avctx->priv_data;

  1522. 

  1523.     r->s.avctx = avctx;

  1524. 

  1525.     if (avctx->internal->is_copy) {

  1526.         r->tmp_b_block_base = NULL;

  1527.         if ((err = ff_MPV_common_init(&r->s)) < 0)

  1528.             return err;

  1529.         if ((err = rv34_decoder_alloc(r)) < 0)

  1530.             return err;

  1531.     }

  1532.     return 0;

  1533. }

  1534. 

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

  1536. {

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

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

  1539.     int err;

  1540. 

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

  1542.         return 0;

  1543. 

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

  1545.         ff_MPV_common_end(s);

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

  1547.         s->width  = s1->width;

  1548.         if ((err = ff_MPV_common_init(s)) < 0)

  1549.             return err;

  1550.         if ((err = rv34_decoder_realloc(r)) < 0)

  1551.             return err;

  1552.     }

  1553. 

  1554.     if ((err = ff_mpeg_update_thread_context(dst, src)))

  1555.         return err;

  1556. 

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

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

  1559.     r->next_pts = r1->next_pts;

  1560. 

  1561.     memset(&r->si, 0, sizeof(r->si));

  1562. 

  1563.     return 0;

  1564. }

  1565. 

  1566. static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)

  1567. {

  1568.     if(avctx->slice_count) return avctx->slice_offset[n];

  1569.     else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);

  1570. }

  1571. 

  1572. static int finish_frame(AVCodecContext *avctx, AVFrame *pict)

  1573. {

  1574.     RV34DecContext *r = avctx->priv_data;

  1575.     MpegEncContext *s = &r->s;

  1576.     int got_picture = 0;

  1577. 

  1578.     ff_er_frame_end(s);

  1579.     ff_MPV_frame_end(s);

  1580.     s->mb_num_left = 0;

  1581. 

  1582.     if (HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))

  1583.         ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);

  1584. 

  1585.     if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {

  1586.         *pict = s->current_picture_ptr->f;

  1587.         got_picture = 1;

  1588.     } else if (s->last_picture_ptr != NULL) {

  1589.         *pict = s->last_picture_ptr->f;

  1590.         got_picture = 1;

  1591.     }

  1592.     if (got_picture)

  1593.         ff_print_debug_info(s, pict);

  1594. 

  1595.     return got_picture;

  1596. }

  1597. 

  1598. int ff_rv34_decode_frame(AVCodecContext *avctx,

  1599.                             void *data, int *got_picture_ptr,

  1600.                             AVPacket *avpkt)

  1601. {

  1602.     const uint8_t *buf = avpkt->data;

  1603.     int buf_size = avpkt->size;

  1604.     RV34DecContext *r = avctx->priv_data;

  1605.     MpegEncContext *s = &r->s;

  1606.     AVFrame *pict = data;

  1607.     SliceInfo si;

  1608.     int i;

  1609.     int slice_count;

  1610.     const uint8_t *slices_hdr = NULL;

  1611.     int last = 0;

  1612. 

  1613.     /* no supplementary picture */

  1614.     if (buf_size == 0) {

  1615.         /* special case for last picture */

  1616.         if (s->low_delay==0 && s->next_picture_ptr) {

  1617.             *pict = s->next_picture_ptr->f;

  1618.             s->next_picture_ptr = NULL;

  1619. 

  1620.             *got_picture_ptr = 1;

  1621.         }

  1622.         return 0;

  1623.     }

  1624. 

  1625.     if(!avctx->slice_count){

  1626.         slice_count = (*buf++) + 1;

  1627.         slices_hdr = buf + 4;

  1628.         buf += 8 * slice_count;

  1629.         buf_size -= 1 + 8 * slice_count;

  1630.     }else

  1631.         slice_count = avctx->slice_count;

  1632. 

  1633.     //parse first slice header to check whether this frame can be decoded

  1634.     if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||

  1635.        get_slice_offset(avctx, slices_hdr, 0) > buf_size){

  1636.         av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1637.         return AVERROR_INVALIDDATA;

  1638.     }

  1639.     init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);

  1640.     if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){

  1641.         av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");

  1642.         return AVERROR_INVALIDDATA;

  1643.     }

  1644.     if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) &&

  1645.         si.type == AV_PICTURE_TYPE_B) {

  1646.         av_log(avctx, AV_LOG_ERROR, "Invalid decoder state: B-frame without "

  1647.                "reference data.\n");

  1648.         return AVERROR_INVALIDDATA;

  1649.     }

  1650.     if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)

  1651.        || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)

  1652.        ||  avctx->skip_frame >= AVDISCARD_ALL)

  1653.         return avpkt->size;

  1654. 

  1655.     /* first slice */

  1656.     if (si.start == 0) {

  1657.         if (s->mb_num_left > 0) {

  1658.             av_log(avctx, AV_LOG_ERROR, "New frame but still %d MB left.",

  1659.                    s->mb_num_left);

  1660.             ff_er_frame_end(s);

  1661.             ff_MPV_frame_end(s);

  1662.         }

  1663. 

  1664.         if (s->width != si.width || s->height != si.height) {

  1665.             int err;

  1666. 

  1667.             if (HAVE_THREADS &&

  1668.                 (s->avctx->active_thread_type & FF_THREAD_FRAME)) {

  1669.                 av_log_missing_feature(s->avctx, "Width/height changing with "

  1670.                                        "frame threading is", 0);

  1671.                 return AVERROR_PATCHWELCOME;

  1672.             }

  1673. 

  1674.             av_log(s->avctx, AV_LOG_WARNING, "Changing dimensions to %dx%d\n",

  1675.                    si.width, si.height);

  1676.             ff_MPV_common_end(s);

  1677.             s->width  = si.width;

  1678.             s->height = si.height;

  1679.             avcodec_set_dimensions(s->avctx, s->width, s->height);

  1680.             if ((err = ff_MPV_common_init(s)) < 0)

  1681.                 return err;

  1682.             if ((err = rv34_decoder_realloc(r)) < 0)

  1683.                 return err;

  1684.         }

  1685.         s->pict_type = si.type ? si.type : AV_PICTURE_TYPE_I;

  1686.         if (ff_MPV_frame_start(s, s->avctx) < 0)

  1687.             return -1;

  1688.         ff_er_frame_start(s);

  1689.         if (!r->tmp_b_block_base) {

  1690.             int i;

  1691. 

  1692.             r->tmp_b_block_base = av_malloc(s->linesize * 48);

  1693.             for (i = 0; i < 2; i++)

  1694.                 r->tmp_b_block_y[i] = r->tmp_b_block_base

  1695.                                       + i * 16 * s->linesize;

  1696.             for (i = 0; i < 4; i++)

  1697.                 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize

  1698.                                        + (i >> 1) * 8 * s->uvlinesize

  1699.                                        + (i &  1) * 16;

  1700.         }

  1701.         r->cur_pts = si.pts;

  1702.         if (s->pict_type != AV_PICTURE_TYPE_B) {

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

  1704.             r->next_pts = r->cur_pts;

  1705.         } else {

  1706.             int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);

  1707.             int dist0   = GET_PTS_DIFF(r->cur_pts,  r->last_pts);

  1708.             int dist1   = GET_PTS_DIFF(r->next_pts, r->cur_pts);

  1709. 

  1710.             if(!refdist){

  1711.                 r->mv_weight1 = r->mv_weight2 = r->weight1 = r->weight2 = 8192;

  1712.                 r->scaled_weight = 0;

  1713.             }else{

  1714.                 r->mv_weight1 = (dist0 << 14) / refdist;

  1715.                 r->mv_weight2 = (dist1 << 14) / refdist;

  1716.                 if((r->mv_weight1|r->mv_weight2) & 511){

  1717.                     r->weight1 = r->mv_weight1;

  1718.                     r->weight2 = r->mv_weight2;

  1719.                     r->scaled_weight = 0;

  1720.                 }else{

  1721.                     r->weight1 = r->mv_weight1 >> 9;

  1722.                     r->weight2 = r->mv_weight2 >> 9;

  1723.                     r->scaled_weight = 1;

  1724.                 }

  1725.             }

  1726.         }

  1727.         s->mb_x = s->mb_y = 0;

  1728.         ff_thread_finish_setup(s->avctx);

  1729.     } else if (HAVE_THREADS &&

  1730.                (s->avctx->active_thread_type & FF_THREAD_FRAME)) {

  1731.         av_log(s->avctx, AV_LOG_ERROR, "Decoder needs full frames in frame "

  1732.                "multithreading mode (start MB is %d).\n", si.start);

  1733.         return AVERROR_INVALIDDATA;

  1734.     }

  1735. 

  1736.     for(i = 0; i < slice_count; i++){

  1737.         int offset = get_slice_offset(avctx, slices_hdr, i);

  1738.         int size;

  1739.         if(i+1 == slice_count)

  1740.             size = buf_size - offset;

  1741.         else

  1742.             size = get_slice_offset(avctx, slices_hdr, i+1) - offset;

  1743. 

  1744.         if(offset < 0 || offset > buf_size){

  1745.             av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1746.             break;

  1747.         }

  1748. 

  1749.         r->si.end = s->mb_width * s->mb_height;

  1750.         s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;

  1751. 

  1752.         if(i+1 < slice_count){

  1753.             if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||

  1754.                 get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {

  1755.                 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1756.                 break;

  1757.             }

  1758.             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);

  1759.             if(r->parse_slice_header(r, &r->s.gb, &si) < 0){

  1760.                 if(i+2 < slice_count)

  1761.                     size = get_slice_offset(avctx, slices_hdr, i+2) - offset;

  1762.                 else

  1763.                     size = buf_size - offset;

  1764.             }else

  1765.                 r->si.end = si.start;

  1766.         }

  1767.         if (size < 0 || size > buf_size - offset) {

  1768.             av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");

  1769.             break;

  1770.         }

  1771.         last = rv34_decode_slice(r, r->si.end, buf + offset, size);

  1772.         if(last)

  1773.             break;

  1774.     }

  1775. 

  1776.     if (s->current_picture_ptr) {

  1777.         if (last) {

  1778.             if(r->loop_filter)

  1779.                 r->loop_filter(r, s->mb_height - 1);

  1780. 

  1781.             *got_picture_ptr = finish_frame(avctx, pict);

  1782.         } else if (HAVE_THREADS &&

  1783.                    (s->avctx->active_thread_type & FF_THREAD_FRAME)) {

  1784.             av_log(avctx, AV_LOG_INFO, "marking unfished frame as finished\n");

  1785.             /* always mark the current frame as finished, frame-mt supports

  1786.              * only complete frames */

  1787.             ff_er_frame_end(s);

  1788.             ff_MPV_frame_end(s);

  1789.             s->mb_num_left = 0;

  1790.             ff_thread_report_progress(&s->current_picture_ptr->f, INT_MAX, 0);

  1791.             return AVERROR_INVALIDDATA;

  1792.         }

  1793.     }

  1794. 

  1795.     return avpkt->size;

  1796. }

  1797. 

  1798. av_cold int ff_rv34_decode_end(AVCodecContext *avctx)

  1799. {

  1800.     RV34DecContext *r = avctx->priv_data;

  1801. 

  1802.     ff_MPV_common_end(&r->s);

  1803.     rv34_decoder_free(r);

  1804. 

  1805.     return 0;

  1806. }