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.
33469 Commits
32 Branches
283MB
Tree: 4b4a6ebd68
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '4b4a6ebd68'
${ noResults }
FFmpeg/libavcodec/rv34.c

1587 lines
57KB
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 "avcodec.h"

  28. #include "dsputil.h"

  29. #include "mpegvideo.h"

  30. #include "golomb.h"

  31. #include "mathops.h"

  32. #include "rectangle.h"

  33. 

  34. #include "rv34vlc.h"

  35. #include "rv34data.h"

  36. #include "rv34.h"

  37. 

  38. //#define DEBUG

  39. 

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

  41. {

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

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

  44. }

  45. 

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

  47. static const int rv34_mb_type_to_lavc[12] = {

  48.     MB_TYPE_INTRA,

  49.     MB_TYPE_INTRA16x16              | MB_TYPE_SEPARATE_DC,

  50.     MB_TYPE_16x16   | MB_TYPE_L0,

  51.     MB_TYPE_8x8     | MB_TYPE_L0,

  52.     MB_TYPE_16x16   | MB_TYPE_L0,

  53.     MB_TYPE_16x16   | MB_TYPE_L1,

  54.     MB_TYPE_SKIP,

  55.     MB_TYPE_DIRECT2 | MB_TYPE_16x16,

  56.     MB_TYPE_16x8    | MB_TYPE_L0,

  57.     MB_TYPE_8x16    | MB_TYPE_L0,

  58.     MB_TYPE_16x16   | MB_TYPE_L0L1,

  59.     MB_TYPE_16x16   | MB_TYPE_L0    | MB_TYPE_SEPARATE_DC

  60. };

  61. 

  62. 

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

  64. 

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

  66. 

  67. /**

  68.  * @name RV30/40 VLC generating functions

  69.  * @{

  70.  */

  71. 

  72. static const int table_offs[] = {

  73.       0,   1818,   3622,   4144,   4698,   5234,   5804,   5868,   5900,   5932,

  74.    5996,   6252,   6316,   6348,   6380,   7674,   8944,  10274,  11668,  12250,

  75.   14060,  15846,  16372,  16962,  17512,  18148,  18180,  18212,  18244,  18308,

  76.   18564,  18628,  18660,  18692,  20036,  21314,  22648,  23968,  24614,  26384,

  77.   28190,  28736,  29366,  29938,  30608,  30640,  30672,  30704,  30768,  31024,

  78.   31088,  31120,  31184,  32570,  33898,  35236,  36644,  37286,  39020,  40802,

  79.   41368,  42052,  42692,  43348,  43380,  43412,  43444,  43476,  43604,  43668,

  80.   43700,  43732,  45100,  46430,  47778,  49160,  49802,  51550,  53340,  53972,

  81.   54648,  55348,  55994,  56122,  56154,  56186,  56218,  56346,  56410,  56442,

  82.   56474,  57878,  59290,  60636,  62036,  62682,  64460,  64524,  64588,  64716,

  83.   64844,  66076,  67466,  67978,  68542,  69064,  69648,  70296,  72010,  72074,

  84.   72138,  72202,  72330,  73572,  74936,  75454,  76030,  76566,  77176,  77822,

  85.   79582,  79646,  79678,  79742,  79870,  81180,  82536,  83064,  83672,  84242,

  86.   84934,  85576,  87384,  87448,  87480,  87544,  87672,  88982,  90340,  90902,

  87.   91598,  92182,  92846,  93488,  95246,  95278,  95310,  95374,  95502,  96878,

  88.   98266,  98848,  99542, 100234, 100884, 101524, 103320, 103352, 103384, 103416,

  89.  103480, 104874, 106222, 106910, 107584, 108258, 108902, 109544, 111366, 111398,

  90.  111430, 111462, 111494, 112878, 114320, 114988, 115660, 116310, 116950, 117592

  91. };

  92. 

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

  94. 

  95. /**

  96.  * Generate VLC from codeword lengths.

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

  98.  * @param size   length of input data

  99.  * @param vlc    output VLC

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

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

  102.  */

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

  104.                          const int num)

  105. {

  106.     int i;

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

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

  109.     uint8_t bits2[MAX_VLC_SIZE];

  110.     int maxbits = 0, realsize = 0;

  111. 

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

  113.         if(bits[i]){

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

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

  116.             realsize++;

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

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

  119.         }

  120.     }

  121. 

  122.     codes[0] = 0;

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

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

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

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

  127. 

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

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

  130.     init_vlc_sparse(vlc, FFMIN(maxbits, 9), realsize,

  131.                     bits2, 1, 1,

  132.                     cw,    2, 2,

  133.                     syms,  2, 2, INIT_VLC_USE_NEW_STATIC);

  134. }

  135. 

  136. /**

  137.  * Initialize all tables.

  138.  */

  139. static av_cold void rv34_init_tables(void)

  140. {

  141.     int i, j, k;

  142. 

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

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

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

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

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

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

  149.                 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);

  150.             }

  151.         }

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

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

  154.         }

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

  156.     }

  157. 

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

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

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

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

  162.         }

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

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

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

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

  167.         }

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

  169.     }

  170. }

  171. 

  172. /** @} */ // vlc group

  173. 

  174. 

  175. /**

  176.  * @name RV30/40 inverse transform functions

  177.  * @{

  178.  */

  179. 

  180. static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)

  181. {

  182.     int i;

  183. 

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

  185.         const int z0 = 13*(block[i+8*0] +    block[i+8*2]);

  186.         const int z1 = 13*(block[i+8*0] -    block[i+8*2]);

  187.         const int z2 =  7* block[i+8*1] - 17*block[i+8*3];

  188.         const int z3 = 17* block[i+8*1] +  7*block[i+8*3];

  189. 

  190.         temp[4*i+0] = z0 + z3;

  191.         temp[4*i+1] = z1 + z2;

  192.         temp[4*i+2] = z1 - z2;

  193.         temp[4*i+3] = z0 - z3;

  194.     }

  195. }

  196. 

  197. /**

  198.  * Real Video 3.0/4.0 inverse transform

  199.  * Code is almost the same as in SVQ3, only scaling is different.

  200.  */

  201. static void rv34_inv_transform(DCTELEM *block){

  202.     int temp[16];

  203.     int i;

  204. 

  205.     rv34_row_transform(temp, block);

  206. 

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

  208.         const int z0 = 13*(temp[4*0+i] +    temp[4*2+i]) + 0x200;

  209.         const int z1 = 13*(temp[4*0+i] -    temp[4*2+i]) + 0x200;

  210.         const int z2 =  7* temp[4*1+i] - 17*temp[4*3+i];

  211.         const int z3 = 17* temp[4*1+i] +  7*temp[4*3+i];

  212. 

  213.         block[i*8+0] = (z0 + z3) >> 10;

  214.         block[i*8+1] = (z1 + z2) >> 10;

  215.         block[i*8+2] = (z1 - z2) >> 10;

  216.         block[i*8+3] = (z0 - z3) >> 10;

  217.     }

  218. 

  219. }

  220. 

  221. /**

  222.  * RealVideo 3.0/4.0 inverse transform for DC block

  223.  *

  224.  * Code is almost the same as rv34_inv_transform()

  225.  * but final coefficients are multiplied by 1.5 and have no rounding.

  226.  */

  227. static void rv34_inv_transform_noround(DCTELEM *block){

  228.     int temp[16];

  229.     int i;

  230. 

  231.     rv34_row_transform(temp, block);

  232. 

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

  234.         const int z0 = 13*(temp[4*0+i] +    temp[4*2+i]);

  235.         const int z1 = 13*(temp[4*0+i] -    temp[4*2+i]);

  236.         const int z2 =  7* temp[4*1+i] - 17*temp[4*3+i];

  237.         const int z3 = 17* temp[4*1+i] +  7*temp[4*3+i];

  238. 

  239.         block[i*8+0] = ((z0 + z3) * 3) >> 11;

  240.         block[i*8+1] = ((z1 + z2) * 3) >> 11;

  241.         block[i*8+2] = ((z1 - z2) * 3) >> 11;

  242.         block[i*8+3] = ((z0 - z3) * 3) >> 11;

  243.     }

  244. 

  245. }

  246. 

  247. /** @} */ // transform

  248. 

  249. 

  250. /**

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

  252.  * @{

  253.  */

  254. 

  255. /**

  256.  * Decode coded block pattern.

  257.  */

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

  259. {

  260.     int pattern, code, cbp=0;

  261.     int ones;

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

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

  264.     const int *curshift = shifts;

  265.     int i, t, mask;

  266. 

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

  268.     pattern = code & 0xF;

  269.     code >>= 4;

  270. 

  271.     ones = rv34_count_ones[pattern];

  272. 

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

  274.         if(pattern & mask)

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

  276.     }

  277. 

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

  279.         t = modulo_three_table[code][i];

  280.         if(t == 1)

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

  282.         if(t == 2)

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

  284.     }

  285.     return cbp;

  286. }

  287. 

  288. /**

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

  290.  */

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

  292. {

  293.     if(coef){

  294.         if(coef == esc){

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

  296.             if(coef > 23){

  297.                 coef -= 23;

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

  299.             }

  300.             coef += esc;

  301.         }

  302.         if(get_bits1(gb))

  303.             coef = -coef;

  304.         *dst = coef;

  305.     }

  306. }

  307. 

  308. /**

  309.  * Decode 2x2 subblock of coefficients.

  310.  */

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

  312. {

  313.     int coeffs[4];

  314. 

  315.     coeffs[0] = modulo_three_table[code][0];

  316.     coeffs[1] = modulo_three_table[code][1];

  317.     coeffs[2] = modulo_three_table[code][2];

  318.     coeffs[3] = modulo_three_table[code][3];

  319.     decode_coeff(dst  , coeffs[0], 3, gb, vlc);

  320.     if(is_block2){

  321.         decode_coeff(dst+8, coeffs[1], 2, gb, vlc);

  322.         decode_coeff(dst+1, coeffs[2], 2, gb, vlc);

  323.     }else{

  324.         decode_coeff(dst+1, coeffs[1], 2, gb, vlc);

  325.         decode_coeff(dst+8, coeffs[2], 2, gb, vlc);

  326.     }

  327.     decode_coeff(dst+9, coeffs[3], 2, gb, vlc);

  328. }

  329. 

  330. /**

  331.  * Decode coefficients for 4x4 block.

  332.  *

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

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

  335.  *  o--o

  336.  *    /

  337.  *   /

  338.  *  o--o

  339.  */

  340. 

  341. static inline void rv34_decode_block(DCTELEM *dst, GetBitContext *gb, RV34VLC *rvlc, int fc, int sc)

  342. {

  343.     int code, pattern;

  344. 

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

  346. 

  347.     pattern = code & 0x7;

  348. 

  349.     code >>= 3;

  350.     decode_subblock(dst, code, 0, gb, &rvlc->coefficient);

  351. 

  352.     if(pattern & 4){

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

  354.         decode_subblock(dst + 2, code, 0, gb, &rvlc->coefficient);

  355.     }

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

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

  358.         decode_subblock(dst + 8*2, code, 1, gb, &rvlc->coefficient);

  359.     }

  360.     if(pattern & 1){

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

  362.         decode_subblock(dst + 8*2+2, code, 0, gb, &rvlc->coefficient);

  363.     }

  364. 

  365. }

  366. 

  367. /**

  368.  * Dequantize ordinary 4x4 block.

  369.  * @todo optimize

  370.  */

  371. static inline void rv34_dequant4x4(DCTELEM *block, int Qdc, int Q)

  372. {

  373.     int i, j;

  374. 

  375.     block[0] = (block[0] * Qdc + 8) >> 4;

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

  377.         for(j = !i; j < 4; j++)

  378.             block[j + i*8] = (block[j + i*8] * Q + 8) >> 4;

  379. }

  380. 

  381. /**

  382.  * Dequantize 4x4 block of DC values for 16x16 macroblock.

  383.  * @todo optimize

  384.  */

  385. static inline void rv34_dequant4x4_16x16(DCTELEM *block, int Qdc, int Q)

  386. {

  387.     int i;

  388. 

  389.     for(i = 0; i < 3; i++)

  390.          block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Qdc + 8) >> 4;

  391.     for(; i < 16; i++)

  392.          block[rv34_dezigzag[i]] = (block[rv34_dezigzag[i]] * Q + 8) >> 4;

  393. }

  394. /** @} */ //block functions

  395. 

  396. 

  397. /**

  398.  * @name RV30/40 bitstream parsing

  399.  * @{

  400.  */

  401. 

  402. /**

  403.  * Decode starting slice position.

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

  405.  */

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

  407. {

  408.     int i;

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

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

  411.             break;

  412.     return rv34_mb_bits_sizes[i];

  413. }

  414. 

  415. /**

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

  417.  */

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

  419. {

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

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

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

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

  424. }

  425. 

  426. /**

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

  428.  */

  429. static int rv34_decode_mb_header(RV34DecContext *r, int8_t *intra_types)

  430. {

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

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

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

  434.     int i, t;

  435. 

  436.     if(!r->si.type){

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

  438.         if(!r->is16 && !r->rv30){

  439.             if(!get_bits1(gb))

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

  441.         }

  442.         s->current_picture_ptr->f.mb_type[mb_pos] = r->is16 ? MB_TYPE_INTRA16x16 : MB_TYPE_INTRA;

  443.         r->block_type = r->is16 ? RV34_MB_TYPE_INTRA16x16 : RV34_MB_TYPE_INTRA;

  444.     }else{

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

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

  447.             return -1;

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

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

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

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

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

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

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

  455.         }

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

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

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

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

  460.             return 0;

  461.         }

  462.         r->chroma_vlc = 1;

  463.         r->luma_vlc   = 0;

  464.     }

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

  466.         if(r->is16){

  467.             t = get_bits(gb, 2);

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

  469.             r->luma_vlc   = 2;

  470.         }else{

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

  472.                 return -1;

  473.             r->luma_vlc   = 1;

  474.         }

  475.         r->chroma_vlc = 0;

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

  477.     }else{

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

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

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

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

  482.             r->is16 = 1;

  483.             r->chroma_vlc = 1;

  484.             r->luma_vlc   = 2;

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

  486.         }

  487.     }

  488. 

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

  490. }

  491. 

  492. /** @} */ //bitstream functions

  493. 

  494. /**

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

  496.  * @{

  497.  */

  498. 

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

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

  501. 

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

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

  504. 

  505. /** availability index for subblocks */

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

  507. 

  508. /**

  509.  * motion vector prediction

  510.  *

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

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

  513.  * some other vectors may be used instead.

  514.  */

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

  516. {

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

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

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

  520.     int i, j;

  521.     int mx, my;

  522.     int avail_index = avail_indexes[subblock_no];

  523.     int c_off = part_sizes_w[block_type];

  524. 

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

  526.     if(subblock_no == 3)

  527.         c_off = -1;

  528. 

  529.     if(r->avail_cache[avail_index - 1]){

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

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

  532.     }

  533.     if(r->avail_cache[avail_index - 4]){

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

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

  536.     }else{

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

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

  539.     }

  540.     if(!r->avail_cache[avail_index - 4 + c_off]){

  541.         if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1] || r->rv30)){

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

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

  544.         }else{

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

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

  547.         }

  548.     }else{

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

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

  551.     }

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

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

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

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

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

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

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

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

  560.         }

  561.     }

  562. }

  563. 

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

  565. 

  566. /**

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

  568.  */

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

  570. {

  571.     int mul = dir ? -r->weight2 : r->weight1;

  572. 

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

  574. }

  575. 

  576. /**

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

  578.  */

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

  580.                                       int A_avail, int B_avail, int C_avail,

  581.                                       int *mx, int *my)

  582. {

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

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

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

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

  587.             *mx /= 2;

  588.             *my /= 2;

  589.         }

  590.     }else{

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

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

  593.     }

  594. }

  595. 

  596. /**

  597.  * motion vector prediction for B-frames

  598.  */

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

  600. {

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

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

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

  604.     int A[2], B[2], C[2];

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

  606.     int mx, my;

  607.     int i, j;

  608.     Picture *cur_pic = s->current_picture_ptr;

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

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

  611. 

  612.     memset(A, 0, sizeof(A));

  613.     memset(B, 0, sizeof(B));

  614.     memset(C, 0, sizeof(C));

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

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

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

  618.         has_A = 1;

  619.     }

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

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

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

  623.         has_B = 1;

  624.     }

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

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

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

  628.         has_C = 1;

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

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

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

  632.         has_C = 1;

  633.     }

  634. 

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

  636. 

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

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

  639. 

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

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

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

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

  644.         }

  645.     }

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

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

  648.     }

  649. }

  650. 

  651. /**

  652.  * motion vector prediction - RV3 version

  653.  */

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

  655. {

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

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

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

  659.     int i, j, k;

  660.     int mx, my;

  661.     int avail_index = avail_indexes[0];

  662. 

  663.     if(r->avail_cache[avail_index - 1]){

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

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

  666.     }

  667.     if(r->avail_cache[avail_index - 4]){

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

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

  670.     }else{

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

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

  673.     }

  674.     if(!r->avail_cache[avail_index - 4 + 2]){

  675.         if(r->avail_cache[avail_index - 4] && (r->avail_cache[avail_index - 1])){

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

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

  678.         }else{

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

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

  681.         }

  682.     }else{

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

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

  685.     }

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

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

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

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

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

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

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

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

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

  695.             }

  696.         }

  697.     }

  698. }

  699. 

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

  701. 

  702. /**

  703.  * generic motion compensation function

  704.  *

  705.  * @param r decoder context

  706.  * @param block_type type of the current block

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

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

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

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

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

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

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

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

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

  716.  */

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

  718.                           const int xoff, const int yoff, int mv_off,

  719.                           const int width, const int height, int dir,

  720.                           const int thirdpel, int weighted,

  721.                           qpel_mc_func (*qpel_mc)[16],

  722.                           h264_chroma_mc_func (*chroma_mc))

  723. {

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

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

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

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

  728.     int is16x16 = 1;

  729. 

  730.     if(thirdpel){

  731.         int chroma_mx, chroma_my;

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

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

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

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

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

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

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

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

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

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

  742.     }else{

  743.         int cx, cy;

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

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

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

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

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

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

  750.         umx = cx >> 2;

  751.         umy = cy >> 2;

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

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

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

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

  756.             uvmx = uvmy = 4;

  757.     }

  758.     dxy = ly*4 + lx;

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

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

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

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

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

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

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

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

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

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

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

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

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

  772. 

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

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

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

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

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

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

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

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

  781.         srcU = uvbuf;

  782.         srcV = uvbuf + 16;

  783.     }

  784.     if(!weighted){

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

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

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

  788.     }else{

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

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

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

  792.     }

  793. 

  794.     if(block_type == RV34_MB_P_16x8){

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

  796.         Y    += 8;

  797.         srcY += 8;

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

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

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

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

  802.     }

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

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

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

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

  807. }

  808. 

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

  810.                         const int xoff, const int yoff, int mv_off,

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

  812. {

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

  814.             r->rdsp.put_pixels_tab,

  815.             r->rdsp.put_chroma_pixels_tab);

  816. }

  817. 

  818. static void rv4_weight(RV34DecContext *r)

  819. {

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

  821.                                       r->tmp_b_block_y[0],

  822.                                       r->tmp_b_block_y[1],

  823.                                       r->weight1,

  824.                                       r->weight2,

  825.                                       r->s.linesize);

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

  827.                                       r->tmp_b_block_uv[0],

  828.                                       r->tmp_b_block_uv[2],

  829.                                       r->weight1,

  830.                                       r->weight2,

  831.                                       r->s.uvlinesize);

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

  833.                                       r->tmp_b_block_uv[1],

  834.                                       r->tmp_b_block_uv[3],

  835.                                       r->weight1,

  836.                                       r->weight2,

  837.                                       r->s.uvlinesize);

  838. }

  839. 

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

  841. {

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

  843. 

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

  845.             r->rdsp.put_pixels_tab,

  846.             r->rdsp.put_chroma_pixels_tab);

  847.     if(!weighted){

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

  849.                 r->rdsp.avg_pixels_tab,

  850.                 r->rdsp.avg_chroma_pixels_tab);

  851.     }else{

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

  853.                 r->rdsp.put_pixels_tab,

  854.                 r->rdsp.put_chroma_pixels_tab);

  855.         rv4_weight(r);

  856.     }

  857. }

  858. 

  859. static void rv34_mc_2mv_skip(RV34DecContext *r)

  860. {

  861.     int i, j;

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

  863. 

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

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

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

  867.                      weighted,

  868.                      r->rdsp.put_pixels_tab,

  869.                      r->rdsp.put_chroma_pixels_tab);

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

  871.                      weighted,

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

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

  874.         }

  875.     if(weighted)

  876.         rv4_weight(r);

  877. }

  878. 

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

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

  881. 

  882. /**

  883.  * Decode motion vector differences

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

  885.  */

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

  887. {

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

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

  890.     int i, j, k, l;

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

  892.     int next_bt;

  893. 

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

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

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

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

  898.     }

  899.     switch(block_type){

  900.     case RV34_MB_TYPE_INTRA:

  901.     case RV34_MB_TYPE_INTRA16x16:

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

  903.         return 0;

  904.     case RV34_MB_SKIP:

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

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

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

  908.             break;

  909.         }

  910.     case RV34_MB_B_DIRECT:

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

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

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

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

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

  916.         }else

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

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

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

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

  921.                             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]);

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

  923.             rv34_mc_2mv(r, block_type);

  924.         else

  925.             rv34_mc_2mv_skip(r);

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

  927.         break;

  928.     case RV34_MB_P_16x16:

  929.     case RV34_MB_P_MIX16x16:

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

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

  932.         break;

  933.     case RV34_MB_B_FORWARD:

  934.     case RV34_MB_B_BACKWARD:

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

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

  937.         if(r->rv30)

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

  939.         else

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

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

  942.         break;

  943.     case RV34_MB_P_16x8:

  944.     case RV34_MB_P_8x16:

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

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

  947.         if(block_type == RV34_MB_P_16x8){

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

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

  950.         }

  951.         if(block_type == RV34_MB_P_8x16){

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

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

  954.         }

  955.         break;

  956.     case RV34_MB_B_BIDIR:

  957.         rv34_pred_mv_b  (r, block_type, 0);

  958.         rv34_pred_mv_b  (r, block_type, 1);

  959.         rv34_mc_2mv     (r, block_type);

  960.         break;

  961.     case RV34_MB_P_8x8:

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

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

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

  965.         }

  966.         break;

  967.     }

  968. 

  969.     return 0;

  970. }

  971. /** @} */ // mv group

  972. 

  973. /**

  974.  * @name Macroblock reconstruction functions

  975.  * @{

  976.  */

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

  978. static const int ittrans[9] = {

  979.  DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,

  980.  VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,

  981. };

  982. 

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

  984. static const int ittrans16[4] = {

  985.  DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,

  986. };

  987. 

  988. /**

  989.  * Perform 4x4 intra prediction.

  990.  */

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

  992. {

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

  994.     uint32_t topleft;

  995. 

  996.     if(!up && !left)

  997.         itype = DC_128_PRED;

  998.     else if(!up){

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

  1000.         if(itype == DC_PRED)   itype = LEFT_DC_PRED;

  1001.     }else if(!left){

  1002.         if(itype == HOR_PRED)  itype = VERT_PRED;

  1003.         if(itype == DC_PRED)   itype = TOP_DC_PRED;

  1004.         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;

  1005.     }

  1006.     if(!down){

  1007.         if(itype == DIAG_DOWN_LEFT_PRED) itype = DIAG_DOWN_LEFT_PRED_RV40_NODOWN;

  1008.         if(itype == HOR_UP_PRED) itype = HOR_UP_PRED_RV40_NODOWN;

  1009.         if(itype == VERT_LEFT_PRED) itype = VERT_LEFT_PRED_RV40_NODOWN;

  1010.     }

  1011.     if(!right && up){

  1012.         topleft = dst[-stride + 3] * 0x01010101;

  1013.         prev = (uint8_t*)&topleft;

  1014.     }

  1015.     r->h.pred4x4[itype](dst, prev, stride);

  1016. }

  1017. 

  1018. /** add_pixels_clamped for 4x4 block */

  1019. static void rv34_add_4x4_block(uint8_t *dst, int stride, DCTELEM block[64], int off)

  1020. {

  1021.     int x, y;

  1022.     for(y = 0; y < 4; y++)

  1023.         for(x = 0; x < 4; x++)

  1024.             dst[x + y*stride] = av_clip_uint8(dst[x + y*stride] + block[off + x+y*8]);

  1025. }

  1026. 

  1027. static inline int adjust_pred16(int itype, int up, int left)

  1028. {

  1029.     if(!up && !left)

  1030.         itype = DC_128_PRED8x8;

  1031.     else if(!up){

  1032.         if(itype == PLANE_PRED8x8)itype = HOR_PRED8x8;

  1033.         if(itype == VERT_PRED8x8) itype = HOR_PRED8x8;

  1034.         if(itype == DC_PRED8x8)   itype = LEFT_DC_PRED8x8;

  1035.     }else if(!left){

  1036.         if(itype == PLANE_PRED8x8)itype = VERT_PRED8x8;

  1037.         if(itype == HOR_PRED8x8)  itype = VERT_PRED8x8;

  1038.         if(itype == DC_PRED8x8)   itype = TOP_DC_PRED8x8;

  1039.     }

  1040.     return itype;

  1041. }

  1042. 

  1043. static void rv34_output_macroblock(RV34DecContext *r, int8_t *intra_types, int cbp, int is16)

  1044. {

  1045.     MpegEncContext *s = &r->s;

  1046.     DSPContext *dsp = &s->dsp;

  1047.     int i, j;

  1048.     uint8_t *Y, *U, *V;

  1049.     int itype;

  1050.     int avail[6*8] = {0};

  1051.     int idx;

  1052. 

  1053.     // Set neighbour information.

  1054.     if(r->avail_cache[1])

  1055.         avail[0] = 1;

  1056.     if(r->avail_cache[2])

  1057.         avail[1] = avail[2] = 1;

  1058.     if(r->avail_cache[3])

  1059.         avail[3] = avail[4] = 1;

  1060.     if(r->avail_cache[4])

  1061.         avail[5] = 1;

  1062.     if(r->avail_cache[5])

  1063.         avail[8] = avail[16] = 1;

  1064.     if(r->avail_cache[9])

  1065.         avail[24] = avail[32] = 1;

  1066. 

  1067.     Y = s->dest[0];

  1068.     U = s->dest[1];

  1069.     V = s->dest[2];

  1070.     if(!is16){

  1071.         for(j = 0; j < 4; j++){

  1072.             idx = 9 + j*8;

  1073.             for(i = 0; i < 4; i++, cbp >>= 1, Y += 4, idx++){

  1074.                 rv34_pred_4x4_block(r, Y, s->linesize, ittrans[intra_types[i]], avail[idx-8], avail[idx-1], avail[idx+7], avail[idx-7]);

  1075.                 avail[idx] = 1;

  1076.                 if(cbp & 1)

  1077.                     rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);

  1078.             }

  1079.             Y += s->linesize * 4 - 4*4;

  1080.             intra_types += r->intra_types_stride;

  1081.         }

  1082.         intra_types -= r->intra_types_stride * 4;

  1083.         fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);

  1084.         for(j = 0; j < 2; j++){

  1085.             idx = 6 + j*4;

  1086.             for(i = 0; i < 2; i++, cbp >>= 1, idx++){

  1087.                 rv34_pred_4x4_block(r, U + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);

  1088.                 rv34_pred_4x4_block(r, V + i*4 + j*4*s->uvlinesize, s->uvlinesize, ittrans[intra_types[i*2+j*2*r->intra_types_stride]], r->avail_cache[idx-4], r->avail_cache[idx-1], !i && !j, r->avail_cache[idx-3]);

  1089.                 r->avail_cache[idx] = 1;

  1090.                 if(cbp & 0x01)

  1091.                     rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);

  1092.                 if(cbp & 0x10)

  1093.                     rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);

  1094.             }

  1095.         }

  1096.     }else{

  1097.         itype = ittrans16[intra_types[0]];

  1098.         itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1099.         r->h.pred16x16[itype](Y, s->linesize);

  1100.         dsp->add_pixels_clamped(s->block[0], Y,     s->linesize);

  1101.         dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);

  1102.         Y += s->linesize * 8;

  1103.         dsp->add_pixels_clamped(s->block[2], Y,     s->linesize);

  1104.         dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);

  1105. 

  1106.         itype = ittrans16[intra_types[0]];

  1107.         if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;

  1108.         itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1109.         r->h.pred8x8[itype](U, s->uvlinesize);

  1110.         dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);

  1111.         r->h.pred8x8[itype](V, s->uvlinesize);

  1112.         dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);

  1113.     }

  1114. }

  1115. 

  1116. /**

  1117.  * mask for retrieving all bits in coded block pattern

  1118.  * corresponding to one 8x8 block

  1119.  */

  1120. #define LUMA_CBP_BLOCK_MASK 0x33

  1121. 

  1122. #define U_CBP_MASK 0x0F0000

  1123. #define V_CBP_MASK 0xF00000

  1124. 

  1125. /** @} */ // recons group

  1126. 

  1127. 

  1128. static void rv34_apply_differences(RV34DecContext *r, int cbp)

  1129. {

  1130.     static const int shifts[4] = { 0, 2, 8, 10 };

  1131.     MpegEncContext *s = &r->s;

  1132.     int i;

  1133. 

  1134.     for(i = 0; i < 4; i++)

  1135.         if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)

  1136.             s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);

  1137.     if(cbp & U_CBP_MASK)

  1138.         s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);

  1139.     if(cbp & V_CBP_MASK)

  1140.         s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);

  1141. }

  1142. 

  1143. static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)

  1144. {

  1145.     int d;

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

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

  1148.         return 1;

  1149.     d = motion_val[0][1] - motion_val[-step][1];

  1150.     if(d < -3 || d > 3)

  1151.         return 1;

  1152.     return 0;

  1153. }

  1154. 

  1155. static int rv34_set_deblock_coef(RV34DecContext *r)

  1156. {

  1157.     MpegEncContext *s = &r->s;

  1158.     int hmvmask = 0, vmvmask = 0, i, j;

  1159.     int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;

  1160.     int16_t (*motion_val)[2] = &s->current_picture_ptr->f.motion_val[0][midx];

  1161.     for(j = 0; j < 16; j += 8){

  1162.         for(i = 0; i < 2; i++){

  1163.             if(is_mv_diff_gt_3(motion_val + i, 1))

  1164.                 vmvmask |= 0x11 << (j + i*2);

  1165.             if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))

  1166.                 hmvmask |= 0x03 << (j + i*2);

  1167.         }

  1168.         motion_val += s->b8_stride;

  1169.     }

  1170.     if(s->first_slice_line)

  1171.         hmvmask &= ~0x000F;

  1172.     if(!s->mb_x)

  1173.         vmvmask &= ~0x1111;

  1174.     if(r->rv30){ //RV30 marks both subblocks on the edge for filtering

  1175.         vmvmask |= (vmvmask & 0x4444) >> 1;

  1176.         hmvmask |= (hmvmask & 0x0F00) >> 4;

  1177.         if(s->mb_x)

  1178.             r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;

  1179.         if(!s->first_slice_line)

  1180.             r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;

  1181.     }

  1182.     return hmvmask | vmvmask;

  1183. }

  1184. 

  1185. static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)

  1186. {

  1187.     MpegEncContext *s = &r->s;

  1188.     GetBitContext *gb = &s->gb;

  1189.     int cbp, cbp2;

  1190.     int i, blknum, blkoff;

  1191.     DCTELEM block16[64];

  1192.     int luma_dc_quant;

  1193.     int dist;

  1194.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;

  1195. 

  1196.     // Calculate which neighbours are available. Maybe it's worth optimizing too.

  1197.     memset(r->avail_cache, 0, sizeof(r->avail_cache));

  1198.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);

  1199.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;

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

  1201.         r->avail_cache[5] =

  1202.         r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];

  1203.     if(dist >= s->mb_width)

  1204.         r->avail_cache[2] =

  1205.         r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];

  1206.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)

  1207.         r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];

  1208.     if(s->mb_x && dist > s->mb_width)

  1209.         r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];

  1210. 

  1211.     s->qscale = r->si.quant;

  1212.     cbp = cbp2 = rv34_decode_mb_header(r, intra_types);

  1213.     r->cbp_luma  [mb_pos] = cbp;

  1214.     r->cbp_chroma[mb_pos] = cbp >> 16;

  1215.     if(s->pict_type == AV_PICTURE_TYPE_I)

  1216.         r->deblock_coefs[mb_pos] = 0xFFFF;

  1217.     else

  1218.         r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];

  1219.     s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;

  1220. 

  1221.     if(cbp == -1)

  1222.         return -1;

  1223. 

  1224.     luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];

  1225.     if(r->is16){

  1226.         memset(block16, 0, sizeof(block16));

  1227.         rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);

  1228.         rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);

  1229.         rv34_inv_transform_noround(block16);

  1230.     }

  1231. 

  1232.     for(i = 0; i < 16; i++, cbp >>= 1){

  1233.         if(!r->is16 && !(cbp & 1)) continue;

  1234.         blknum = ((i & 2) >> 1) + ((i & 8) >> 2);

  1235.         blkoff = ((i & 1) << 2) + ((i & 4) << 3);

  1236.         if(cbp & 1)

  1237.             rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);

  1238.         rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);

  1239.         if(r->is16) //FIXME: optimize

  1240.             s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];

  1241.         rv34_inv_transform(s->block[blknum] + blkoff);

  1242.     }

  1243.     if(r->block_type == RV34_MB_P_MIX16x16)

  1244.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);

  1245.     for(; i < 24; i++, cbp >>= 1){

  1246.         if(!(cbp & 1)) continue;

  1247.         blknum = ((i & 4) >> 2) + 4;

  1248.         blkoff = ((i & 1) << 2) + ((i & 2) << 4);

  1249.         rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);

  1250.         rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);

  1251.         rv34_inv_transform(s->block[blknum] + blkoff);

  1252.     }

  1253.     if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos]))

  1254.         rv34_output_macroblock(r, intra_types, cbp2, r->is16);

  1255.     else

  1256.         rv34_apply_differences(r, cbp2);

  1257. 

  1258.     return 0;

  1259. }

  1260. 

  1261. static int check_slice_end(RV34DecContext *r, MpegEncContext *s)

  1262. {

  1263.     int bits;

  1264.     if(s->mb_y >= s->mb_height)

  1265.         return 1;

  1266.     if(!s->mb_num_left)

  1267.         return 1;

  1268.     if(r->s.mb_skip_run > 1)

  1269.         return 0;

  1270.     bits = r->bits - get_bits_count(&s->gb);

  1271.     if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))

  1272.         return 1;

  1273.     return 0;

  1274. }

  1275. 

  1276. static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)

  1277. {

  1278.     MpegEncContext *s = &r->s;

  1279.     GetBitContext *gb = &s->gb;

  1280.     int mb_pos;

  1281.     int res;

  1282. 

  1283.     init_get_bits(&r->s.gb, buf, buf_size*8);

  1284.     res = r->parse_slice_header(r, gb, &r->si);

  1285.     if(res < 0){

  1286.         av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");

  1287.         return -1;

  1288.     }

  1289. 

  1290.     if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {

  1291.         if(s->width != r->si.width || s->height != r->si.height){

  1292.             av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);

  1293.             MPV_common_end(s);

  1294.             s->width  = r->si.width;

  1295.             s->height = r->si.height;

  1296.             avcodec_set_dimensions(s->avctx, s->width, s->height);

  1297.             if(MPV_common_init(s) < 0)

  1298.                 return -1;

  1299.             r->intra_types_stride = s->mb_width*4 + 4;

  1300.             r->intra_types_hist = av_realloc(r->intra_types_hist, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));

  1301.             r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;

  1302.             r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));

  1303.             r->cbp_luma   = av_realloc(r->cbp_luma,   r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));

  1304.             r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));

  1305.             r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));

  1306.             av_freep(&r->tmp_b_block_base);

  1307.         }

  1308.         s->pict_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;

  1309.         if(MPV_frame_start(s, s->avctx) < 0)

  1310.             return -1;

  1311.         ff_er_frame_start(s);

  1312.         if (!r->tmp_b_block_base) {

  1313.             int i;

  1314. 

  1315.             r->tmp_b_block_base = av_malloc(s->linesize * 48);

  1316.             for (i = 0; i < 2; i++)

  1317.                 r->tmp_b_block_y[i] = r->tmp_b_block_base + i * 16 * s->linesize;

  1318.             for (i = 0; i < 4; i++)

  1319.                 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize

  1320.                                        + (i >> 1) * 8 * s->uvlinesize + (i & 1) * 16;

  1321.         }

  1322.         r->cur_pts = r->si.pts;

  1323.         if(s->pict_type != AV_PICTURE_TYPE_B){

  1324.             r->last_pts = r->next_pts;

  1325.             r->next_pts = r->cur_pts;

  1326.         }else{

  1327.             int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);

  1328.             int dist0   = GET_PTS_DIFF(r->cur_pts,  r->last_pts);

  1329.             int dist1   = GET_PTS_DIFF(r->next_pts, r->cur_pts);

  1330. 

  1331.             if(!refdist){

  1332.                 r->weight1 = r->weight2 = 8192;

  1333.             }else{

  1334.                 r->weight1 = (dist0 << 14) / refdist;

  1335.                 r->weight2 = (dist1 << 14) / refdist;

  1336.             }

  1337.         }

  1338.         s->mb_x = s->mb_y = 0;

  1339.     } else {

  1340.         int slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;

  1341. 

  1342.         if (slice_type != s->pict_type) {

  1343.             av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");

  1344.             return AVERROR_INVALIDDATA;

  1345.         }

  1346.         if (s->width != r->si.width || s->height != r->si.height) {

  1347.             av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");

  1348.             return AVERROR_INVALIDDATA;

  1349.         }

  1350.     }

  1351. 

  1352.     r->si.end = end;

  1353.     s->qscale = r->si.quant;

  1354.     r->bits = buf_size*8;

  1355.     s->mb_num_left = r->si.end - r->si.start;

  1356.     r->s.mb_skip_run = 0;

  1357. 

  1358.     mb_pos = s->mb_x + s->mb_y * s->mb_width;

  1359.     if(r->si.start != mb_pos){

  1360.         av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);

  1361.         s->mb_x = r->si.start % s->mb_width;

  1362.         s->mb_y = r->si.start / s->mb_width;

  1363.     }

  1364.     memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));

  1365.     s->first_slice_line = 1;

  1366.     s->resync_mb_x = s->mb_x;

  1367.     s->resync_mb_y = s->mb_y;

  1368. 

  1369.     ff_init_block_index(s);

  1370.     while(!check_slice_end(r, s)) {

  1371.         ff_update_block_index(s);

  1372.         s->dsp.clear_blocks(s->block[0]);

  1373. 

  1374.         if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 4) < 0){

  1375.             ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_ERROR|DC_ERROR|MV_ERROR);

  1376.             return -1;

  1377.         }

  1378.         if (++s->mb_x == s->mb_width) {

  1379.             s->mb_x = 0;

  1380.             s->mb_y++;

  1381.             ff_init_block_index(s);

  1382. 

  1383.             memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1384.             memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1385. 

  1386.             if(r->loop_filter && s->mb_y >= 2)

  1387.                 r->loop_filter(r, s->mb_y - 2);

  1388.         }

  1389.         if(s->mb_x == s->resync_mb_x)

  1390.             s->first_slice_line=0;

  1391.         s->mb_num_left--;

  1392.     }

  1393.     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, AC_END|DC_END|MV_END);

  1394. 

  1395.     return s->mb_y == s->mb_height;

  1396. }

  1397. 

  1398. /** @} */ // recons group end

  1399. 

  1400. /**

  1401.  * Initialize decoder.

  1402.  */

  1403. av_cold int ff_rv34_decode_init(AVCodecContext *avctx)

  1404. {

  1405.     RV34DecContext *r = avctx->priv_data;

  1406.     MpegEncContext *s = &r->s;

  1407. 

  1408.     MPV_decode_defaults(s);

  1409.     s->avctx      = avctx;

  1410.     s->out_format = FMT_H263;

  1411.     s->codec_id   = avctx->codec_id;

  1412. 

  1413.     s->width  = avctx->width;

  1414.     s->height = avctx->height;

  1415. 

  1416.     r->s.avctx = avctx;

  1417.     avctx->flags |= CODEC_FLAG_EMU_EDGE;

  1418.     r->s.flags |= CODEC_FLAG_EMU_EDGE;

  1419.     avctx->pix_fmt = PIX_FMT_YUV420P;

  1420.     avctx->has_b_frames = 1;

  1421.     s->low_delay = 0;

  1422. 

  1423.     if (MPV_common_init(s) < 0)

  1424.         return -1;

  1425. 

  1426.     ff_h264_pred_init(&r->h, CODEC_ID_RV40, 8, 1);

  1427. 

  1428. #if CONFIG_RV30_DECODER

  1429.     if (avctx->codec_id == CODEC_ID_RV30)

  1430.         ff_rv30dsp_init(&r->rdsp, &r->s.dsp);

  1431. #endif

  1432. #if CONFIG_RV40_DECODER

  1433.     if (avctx->codec_id == CODEC_ID_RV40)

  1434.         ff_rv40dsp_init(&r->rdsp, &r->s.dsp);

  1435. #endif

  1436. 

  1437.     r->intra_types_stride = 4*s->mb_stride + 4;

  1438.     r->intra_types_hist = av_malloc(r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));

  1439.     r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;

  1440. 

  1441.     r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));

  1442. 

  1443.     r->cbp_luma   = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));

  1444.     r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));

  1445.     r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));

  1446. 

  1447.     if(!intra_vlcs[0].cbppattern[0].bits)

  1448.         rv34_init_tables();

  1449. 

  1450.     return 0;

  1451. }

  1452. 

  1453. static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)

  1454. {

  1455.     if(avctx->slice_count) return avctx->slice_offset[n];

  1456.     else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);

  1457. }

  1458. 

  1459. int ff_rv34_decode_frame(AVCodecContext *avctx,

  1460.                             void *data, int *data_size,

  1461.                             AVPacket *avpkt)

  1462. {

  1463.     const uint8_t *buf = avpkt->data;

  1464.     int buf_size = avpkt->size;

  1465.     RV34DecContext *r = avctx->priv_data;

  1466.     MpegEncContext *s = &r->s;

  1467.     AVFrame *pict = data;

  1468.     SliceInfo si;

  1469.     int i;

  1470.     int slice_count;

  1471.     const uint8_t *slices_hdr = NULL;

  1472.     int last = 0;

  1473. 

  1474.     /* no supplementary picture */

  1475.     if (buf_size == 0) {

  1476.         /* special case for last picture */

  1477.         if (s->low_delay==0 && s->next_picture_ptr) {

  1478.             *pict = *(AVFrame*)s->next_picture_ptr;

  1479.             s->next_picture_ptr = NULL;

  1480. 

  1481.             *data_size = sizeof(AVFrame);

  1482.         }

  1483.         return 0;

  1484.     }

  1485. 

  1486.     if(!avctx->slice_count){

  1487.         slice_count = (*buf++) + 1;

  1488.         slices_hdr = buf + 4;

  1489.         buf += 8 * slice_count;

  1490.         buf_size -= 1 + 8 * slice_count;

  1491.     }else

  1492.         slice_count = avctx->slice_count;

  1493. 

  1494.     //parse first slice header to check whether this frame can be decoded

  1495.     if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||

  1496.        get_slice_offset(avctx, slices_hdr, 0) > buf_size){

  1497.         av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1498.         return -1;

  1499.     }

  1500.     init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);

  1501.     if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){

  1502.         av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");

  1503.         return -1;

  1504.     }

  1505.     if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) && si.type == AV_PICTURE_TYPE_B)

  1506.         return -1;

  1507.     if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)

  1508.        || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)

  1509.        ||  avctx->skip_frame >= AVDISCARD_ALL)

  1510.         return avpkt->size;

  1511. 

  1512.     for(i = 0; i < slice_count; i++){

  1513.         int offset = get_slice_offset(avctx, slices_hdr, i);

  1514.         int size;

  1515.         if(i+1 == slice_count)

  1516.             size = buf_size - offset;

  1517.         else

  1518.             size = get_slice_offset(avctx, slices_hdr, i+1) - offset;

  1519. 

  1520.         if(offset < 0 || offset > buf_size){

  1521.             av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1522.             break;

  1523.         }

  1524. 

  1525.         r->si.end = s->mb_width * s->mb_height;

  1526.         if(i+1 < slice_count){

  1527.             if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||

  1528.                 get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {

  1529.                 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1530.                 break;

  1531.             }

  1532.             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);

  1533.             if(r->parse_slice_header(r, &r->s.gb, &si) < 0){

  1534.                 if(i+2 < slice_count)

  1535.                     size = get_slice_offset(avctx, slices_hdr, i+2) - offset;

  1536.                 else

  1537.                     size = buf_size - offset;

  1538.             }else

  1539.                 r->si.end = si.start;

  1540.         }

  1541.         if (size < 0 || size > buf_size - offset) {

  1542.             av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");

  1543.             break;

  1544.         }

  1545.         last = rv34_decode_slice(r, r->si.end, buf + offset, size);

  1546.         s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;

  1547.         if(last)

  1548.             break;

  1549.     }

  1550. 

  1551.     if(last && s->current_picture_ptr){

  1552.         if(r->loop_filter)

  1553.             r->loop_filter(r, s->mb_height - 1);

  1554.         ff_er_frame_end(s);

  1555.         MPV_frame_end(s);

  1556.         if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {

  1557.             *pict = *(AVFrame*)s->current_picture_ptr;

  1558.         } else if (s->last_picture_ptr != NULL) {

  1559.             *pict = *(AVFrame*)s->last_picture_ptr;

  1560.         }

  1561. 

  1562.         if(s->last_picture_ptr || s->low_delay){

  1563.             *data_size = sizeof(AVFrame);

  1564.             ff_print_debug_info(s, pict);

  1565.         }

  1566.         s->current_picture_ptr = NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)

  1567.     }

  1568.     return avpkt->size;

  1569. }

  1570. 

  1571. av_cold int ff_rv34_decode_end(AVCodecContext *avctx)

  1572. {

  1573.     RV34DecContext *r = avctx->priv_data;

  1574. 

  1575.     MPV_common_end(&r->s);

  1576. 

  1577.     av_freep(&r->intra_types_hist);

  1578.     r->intra_types = NULL;

  1579.     av_freep(&r->tmp_b_block_base);

  1580.     av_freep(&r->mb_type);

  1581.     av_freep(&r->cbp_luma);

  1582.     av_freep(&r->cbp_chroma);

  1583.     av_freep(&r->deblock_coefs);

  1584. 

  1585.     return 0;

  1586. }