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.
35015 Commits
32 Branches
649MB
Tree: 90c02ae142
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '90c02ae142'
${ noResults }
FFmpeg/libavcodec/rv34.c

1511 lines
55KB
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.  * @name RV30/40 4x4 block decoding functions

  176.  * @{

  177.  */

  178. 

  179. /**

  180.  * Decode coded block pattern.

  181.  */

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

  183. {

  184.     int pattern, code, cbp=0;

  185.     int ones;

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

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

  188.     const int *curshift = shifts;

  189.     int i, t, mask;

  190. 

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

  192.     pattern = code & 0xF;

  193.     code >>= 4;

  194. 

  195.     ones = rv34_count_ones[pattern];

  196. 

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

  198.         if(pattern & mask)

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

  200.     }

  201. 

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

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

  204.         if(t == 1)

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

  206.         if(t == 2)

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

  208.     }

  209.     return cbp;

  210. }

  211. 

  212. /**

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

  214.  */

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

  216. {

  217.     if(coef){

  218.         if(coef == esc){

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

  220.             if(coef > 23){

  221.                 coef -= 23;

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

  223.             }

  224.             coef += esc;

  225.         }

  226.         if(get_bits1(gb))

  227.             coef = -coef;

  228.         *dst = coef;

  229.     }

  230. }

  231. 

  232. /**

  233.  * Decode 2x2 subblock of coefficients.

  234.  */

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

  236. {

  237.     int coeffs[4];

  238. 

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

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

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

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

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

  244.     if(is_block2){

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

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

  247.     }else{

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

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

  250.     }

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

  252. }

  253. 

  254. /**

  255.  * Decode coefficients for 4x4 block.

  256.  *

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

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

  259.  *  o--o

  260.  *    /

  261.  *   /

  262.  *  o--o

  263.  */

  264. 

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

  266. {

  267.     int code, pattern;

  268. 

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

  270. 

  271.     pattern = code & 0x7;

  272. 

  273.     code >>= 3;

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

  275. 

  276.     if(pattern & 4){

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

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

  279.     }

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

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

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

  283.     }

  284.     if(pattern & 1){

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

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

  287.     }

  288. 

  289. }

  290. 

  291. /**

  292.  * Dequantize ordinary 4x4 block.

  293.  * @todo optimize

  294.  */

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

  296. {

  297.     int i, j;

  298. 

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

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

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

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

  303. }

  304. 

  305. /**

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

  307.  * @todo optimize

  308.  */

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

  310. {

  311.     int i;

  312. 

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

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

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

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

  317. }

  318. /** @} */ //block functions

  319. 

  320. 

  321. /**

  322.  * @name RV30/40 bitstream parsing

  323.  * @{

  324.  */

  325. 

  326. /**

  327.  * Decode starting slice position.

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

  329.  */

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

  331. {

  332.     int i;

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

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

  335.             break;

  336.     return rv34_mb_bits_sizes[i];

  337. }

  338. 

  339. /**

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

  341.  */

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

  343. {

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

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

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

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

  348. }

  349. 

  350. /**

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

  352.  */

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

  354. {

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

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

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

  358.     int i, t;

  359. 

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

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

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

  363.             if(!get_bits1(gb))

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

  365.         }

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

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

  368.     }else{

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

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

  371.             return -1;

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

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

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

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

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

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

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

  379.         }

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

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

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

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

  384.             return 0;

  385.         }

  386.         r->chroma_vlc = 1;

  387.         r->luma_vlc   = 0;

  388.     }

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

  390.         if(r->is16){

  391.             t = get_bits(gb, 2);

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

  393.             r->luma_vlc   = 2;

  394.         }else{

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

  396.                 return -1;

  397.             r->luma_vlc   = 1;

  398.         }

  399.         r->chroma_vlc = 0;

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

  401.     }else{

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

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

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

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

  406.             r->is16 = 1;

  407.             r->chroma_vlc = 1;

  408.             r->luma_vlc   = 2;

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

  410.         }

  411.     }

  412. 

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

  414. }

  415. 

  416. /** @} */ //bitstream functions

  417. 

  418. /**

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

  420.  * @{

  421.  */

  422. 

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

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

  425. 

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

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

  428. 

  429. /** availability index for subblocks */

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

  431. 

  432. /**

  433.  * motion vector prediction

  434.  *

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

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

  437.  * some other vectors may be used instead.

  438.  */

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

  440. {

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

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

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

  444.     int i, j;

  445.     int mx, my;

  446.     int avail_index = avail_indexes[subblock_no];

  447.     int c_off = part_sizes_w[block_type];

  448. 

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

  450.     if(subblock_no == 3)

  451.         c_off = -1;

  452. 

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

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

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

  456.     }

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

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

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

  460.     }else{

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

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

  463.     }

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

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

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

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

  468.         }else{

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

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

  471.         }

  472.     }else{

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

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

  475.     }

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

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

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

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

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

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

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

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

  484.         }

  485.     }

  486. }

  487. 

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

  489. 

  490. /**

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

  492.  */

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

  494. {

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

  496. 

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

  498. }

  499. 

  500. /**

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

  502.  */

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

  504.                                       int A_avail, int B_avail, int C_avail,

  505.                                       int *mx, int *my)

  506. {

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

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

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

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

  511.             *mx /= 2;

  512.             *my /= 2;

  513.         }

  514.     }else{

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

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

  517.     }

  518. }

  519. 

  520. /**

  521.  * motion vector prediction for B-frames

  522.  */

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

  524. {

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

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

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

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

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

  530.     int mx, my;

  531.     int i, j;

  532.     Picture *cur_pic = s->current_picture_ptr;

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

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

  535. 

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

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

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

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

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

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

  542.         has_A = 1;

  543.     }

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

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

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

  547.         has_B = 1;

  548.     }

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

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

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

  552.         has_C = 1;

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

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

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

  556.         has_C = 1;

  557.     }

  558. 

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

  560. 

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

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

  563. 

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

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

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

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

  568.         }

  569.     }

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

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

  572.     }

  573. }

  574. 

  575. /**

  576.  * motion vector prediction - RV3 version

  577.  */

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

  579. {

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

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

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

  583.     int i, j, k;

  584.     int mx, my;

  585.     int avail_index = avail_indexes[0];

  586. 

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

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

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

  590.     }

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

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

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

  594.     }else{

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

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

  597.     }

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

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

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

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

  602.         }else{

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

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

  605.         }

  606.     }else{

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

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

  609.     }

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

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

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

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

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

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

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

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

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

  619.             }

  620.         }

  621.     }

  622. }

  623. 

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

  625. 

  626. /**

  627.  * generic motion compensation function

  628.  *

  629.  * @param r decoder context

  630.  * @param block_type type of the current block

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

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

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

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

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

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

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

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

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

  640.  */

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

  642.                           const int xoff, const int yoff, int mv_off,

  643.                           const int width, const int height, int dir,

  644.                           const int thirdpel, int weighted,

  645.                           qpel_mc_func (*qpel_mc)[16],

  646.                           h264_chroma_mc_func (*chroma_mc))

  647. {

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

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

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

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

  652.     int is16x16 = 1;

  653. 

  654.     if(thirdpel){

  655.         int chroma_mx, chroma_my;

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

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

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

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

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

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

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

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

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

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

  666.     }else{

  667.         int cx, cy;

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

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

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

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

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

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

  674.         umx = cx >> 2;

  675.         umy = cy >> 2;

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

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

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

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

  680.             uvmx = uvmy = 4;

  681.     }

  682.     dxy = ly*4 + lx;

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

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

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

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

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

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

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

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

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

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

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

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

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

  696. 

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

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

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

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

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

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

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

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

  705.         srcU = uvbuf;

  706.         srcV = uvbuf + 16;

  707.     }

  708.     if(!weighted){

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

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

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

  712.     }else{

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

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

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

  716.     }

  717. 

  718.     if(block_type == RV34_MB_P_16x8){

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

  720.         Y    += 8;

  721.         srcY += 8;

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

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

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

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

  726.     }

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

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

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

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

  731. }

  732. 

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

  734.                         const int xoff, const int yoff, int mv_off,

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

  736. {

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

  738.             r->rdsp.put_pixels_tab,

  739.             r->rdsp.put_chroma_pixels_tab);

  740. }

  741. 

  742. static void rv4_weight(RV34DecContext *r)

  743. {

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

  745.                                       r->tmp_b_block_y[0],

  746.                                       r->tmp_b_block_y[1],

  747.                                       r->weight1,

  748.                                       r->weight2,

  749.                                       r->s.linesize);

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

  751.                                       r->tmp_b_block_uv[0],

  752.                                       r->tmp_b_block_uv[2],

  753.                                       r->weight1,

  754.                                       r->weight2,

  755.                                       r->s.uvlinesize);

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

  757.                                       r->tmp_b_block_uv[1],

  758.                                       r->tmp_b_block_uv[3],

  759.                                       r->weight1,

  760.                                       r->weight2,

  761.                                       r->s.uvlinesize);

  762. }

  763. 

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

  765. {

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

  767. 

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

  769.             r->rdsp.put_pixels_tab,

  770.             r->rdsp.put_chroma_pixels_tab);

  771.     if(!weighted){

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

  773.                 r->rdsp.avg_pixels_tab,

  774.                 r->rdsp.avg_chroma_pixels_tab);

  775.     }else{

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

  777.                 r->rdsp.put_pixels_tab,

  778.                 r->rdsp.put_chroma_pixels_tab);

  779.         rv4_weight(r);

  780.     }

  781. }

  782. 

  783. static void rv34_mc_2mv_skip(RV34DecContext *r)

  784. {

  785.     int i, j;

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

  787. 

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

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

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

  791.                      weighted,

  792.                      r->rdsp.put_pixels_tab,

  793.                      r->rdsp.put_chroma_pixels_tab);

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

  795.                      weighted,

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

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

  798.         }

  799.     if(weighted)

  800.         rv4_weight(r);

  801. }

  802. 

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

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

  805. 

  806. /**

  807.  * Decode motion vector differences

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

  809.  */

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

  811. {

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

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

  814.     int i, j, k, l;

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

  816.     int next_bt;

  817. 

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

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

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

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

  822.     }

  823.     switch(block_type){

  824.     case RV34_MB_TYPE_INTRA:

  825.     case RV34_MB_TYPE_INTRA16x16:

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

  827.         return 0;

  828.     case RV34_MB_SKIP:

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

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

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

  832.             break;

  833.         }

  834.     case RV34_MB_B_DIRECT:

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

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

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

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

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

  840.         }else

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

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

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

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

  845.                             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]);

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

  847.             rv34_mc_2mv(r, block_type);

  848.         else

  849.             rv34_mc_2mv_skip(r);

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

  851.         break;

  852.     case RV34_MB_P_16x16:

  853.     case RV34_MB_P_MIX16x16:

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

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

  856.         break;

  857.     case RV34_MB_B_FORWARD:

  858.     case RV34_MB_B_BACKWARD:

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

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

  861.         if(r->rv30)

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

  863.         else

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

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

  866.         break;

  867.     case RV34_MB_P_16x8:

  868.     case RV34_MB_P_8x16:

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

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

  871.         if(block_type == RV34_MB_P_16x8){

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

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

  874.         }

  875.         if(block_type == RV34_MB_P_8x16){

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

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

  878.         }

  879.         break;

  880.     case RV34_MB_B_BIDIR:

  881.         rv34_pred_mv_b  (r, block_type, 0);

  882.         rv34_pred_mv_b  (r, block_type, 1);

  883.         rv34_mc_2mv     (r, block_type);

  884.         break;

  885.     case RV34_MB_P_8x8:

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

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

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

  889.         }

  890.         break;

  891.     }

  892. 

  893.     return 0;

  894. }

  895. /** @} */ // mv group

  896. 

  897. /**

  898.  * @name Macroblock reconstruction functions

  899.  * @{

  900.  */

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

  902. static const int ittrans[9] = {

  903.  DC_PRED, VERT_PRED, HOR_PRED, DIAG_DOWN_RIGHT_PRED, DIAG_DOWN_LEFT_PRED,

  904.  VERT_RIGHT_PRED, VERT_LEFT_PRED, HOR_UP_PRED, HOR_DOWN_PRED,

  905. };

  906. 

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

  908. static const int ittrans16[4] = {

  909.  DC_PRED8x8, VERT_PRED8x8, HOR_PRED8x8, PLANE_PRED8x8,

  910. };

  911. 

  912. /**

  913.  * Perform 4x4 intra prediction.

  914.  */

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

  916. {

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

  918.     uint32_t topleft;

  919. 

  920.     if(!up && !left)

  921.         itype = DC_128_PRED;

  922.     else if(!up){

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

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

  925.     }else if(!left){

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

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

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

  929.     }

  930.     if(!down){

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

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

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

  934.     }

  935.     if(!right && up){

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

  937.         prev = (uint8_t*)&topleft;

  938.     }

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

  940. }

  941. 

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

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

  944. {

  945.     int x, y;

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

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

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

  949. }

  950. 

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

  952. {

  953.     if(!up && !left)

  954.         itype = DC_128_PRED8x8;

  955.     else if(!up){

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

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

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

  959.     }else if(!left){

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

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

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

  963.     }

  964.     return itype;

  965. }

  966. 

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

  968. {

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

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

  971.     int i, j;

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

  973.     int itype;

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

  975.     int idx;

  976. 

  977.     // Set neighbour information.

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

  979.         avail[0] = 1;

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

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

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

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

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

  985.         avail[5] = 1;

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

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

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

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

  990. 

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

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

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

  994.     if(!is16){

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

  996.             idx = 9 + j*8;

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

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

  999.                 avail[idx] = 1;

  1000.                 if(cbp & 1)

  1001.                     rv34_add_4x4_block(Y, s->linesize, s->block[(i>>1)+(j&2)], (i&1)*4+(j&1)*32);

  1002.             }

  1003.             Y += s->linesize * 4 - 4*4;

  1004.             intra_types += r->intra_types_stride;

  1005.         }

  1006.         intra_types -= r->intra_types_stride * 4;

  1007.         fill_rectangle(r->avail_cache + 6, 2, 2, 4, 0, 4);

  1008.         for(j = 0; j < 2; j++){

  1009.             idx = 6 + j*4;

  1010.             for(i = 0; i < 2; i++, cbp >>= 1, idx++){

  1011.                 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]);

  1012.                 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]);

  1013.                 r->avail_cache[idx] = 1;

  1014.                 if(cbp & 0x01)

  1015.                     rv34_add_4x4_block(U + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[4], i*4+j*32);

  1016.                 if(cbp & 0x10)

  1017.                     rv34_add_4x4_block(V + i*4 + j*4*s->uvlinesize, s->uvlinesize, s->block[5], i*4+j*32);

  1018.             }

  1019.         }

  1020.     }else{

  1021.         itype = ittrans16[intra_types[0]];

  1022.         itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1023.         r->h.pred16x16[itype](Y, s->linesize);

  1024.         dsp->add_pixels_clamped(s->block[0], Y,     s->linesize);

  1025.         dsp->add_pixels_clamped(s->block[1], Y + 8, s->linesize);

  1026.         Y += s->linesize * 8;

  1027.         dsp->add_pixels_clamped(s->block[2], Y,     s->linesize);

  1028.         dsp->add_pixels_clamped(s->block[3], Y + 8, s->linesize);

  1029. 

  1030.         itype = ittrans16[intra_types[0]];

  1031.         if(itype == PLANE_PRED8x8) itype = DC_PRED8x8;

  1032.         itype = adjust_pred16(itype, r->avail_cache[6-4], r->avail_cache[6-1]);

  1033.         r->h.pred8x8[itype](U, s->uvlinesize);

  1034.         dsp->add_pixels_clamped(s->block[4], U, s->uvlinesize);

  1035.         r->h.pred8x8[itype](V, s->uvlinesize);

  1036.         dsp->add_pixels_clamped(s->block[5], V, s->uvlinesize);

  1037.     }

  1038. }

  1039. 

  1040. /**

  1041.  * mask for retrieving all bits in coded block pattern

  1042.  * corresponding to one 8x8 block

  1043.  */

  1044. #define LUMA_CBP_BLOCK_MASK 0x33

  1045. 

  1046. #define U_CBP_MASK 0x0F0000

  1047. #define V_CBP_MASK 0xF00000

  1048. 

  1049. /** @} */ // recons group

  1050. 

  1051. 

  1052. static void rv34_apply_differences(RV34DecContext *r, int cbp)

  1053. {

  1054.     static const int shifts[4] = { 0, 2, 8, 10 };

  1055.     MpegEncContext *s = &r->s;

  1056.     int i;

  1057. 

  1058.     for(i = 0; i < 4; i++)

  1059.         if((cbp & (LUMA_CBP_BLOCK_MASK << shifts[i])) || r->block_type == RV34_MB_P_MIX16x16)

  1060.             s->dsp.add_pixels_clamped(s->block[i], s->dest[0] + (i & 1)*8 + (i&2)*4*s->linesize, s->linesize);

  1061.     if(cbp & U_CBP_MASK)

  1062.         s->dsp.add_pixels_clamped(s->block[4], s->dest[1], s->uvlinesize);

  1063.     if(cbp & V_CBP_MASK)

  1064.         s->dsp.add_pixels_clamped(s->block[5], s->dest[2], s->uvlinesize);

  1065. }

  1066. 

  1067. static int is_mv_diff_gt_3(int16_t (*motion_val)[2], int step)

  1068. {

  1069.     int d;

  1070.     d = motion_val[0][0] - motion_val[-step][0];

  1071.     if(d < -3 || d > 3)

  1072.         return 1;

  1073.     d = motion_val[0][1] - motion_val[-step][1];

  1074.     if(d < -3 || d > 3)

  1075.         return 1;

  1076.     return 0;

  1077. }

  1078. 

  1079. static int rv34_set_deblock_coef(RV34DecContext *r)

  1080. {

  1081.     MpegEncContext *s = &r->s;

  1082.     int hmvmask = 0, vmvmask = 0, i, j;

  1083.     int midx = s->mb_x * 2 + s->mb_y * 2 * s->b8_stride;

  1084.     int16_t (*motion_val)[2] = &s->current_picture_ptr->f.motion_val[0][midx];

  1085.     for(j = 0; j < 16; j += 8){

  1086.         for(i = 0; i < 2; i++){

  1087.             if(is_mv_diff_gt_3(motion_val + i, 1))

  1088.                 vmvmask |= 0x11 << (j + i*2);

  1089.             if((j || s->mb_y) && is_mv_diff_gt_3(motion_val + i, s->b8_stride))

  1090.                 hmvmask |= 0x03 << (j + i*2);

  1091.         }

  1092.         motion_val += s->b8_stride;

  1093.     }

  1094.     if(s->first_slice_line)

  1095.         hmvmask &= ~0x000F;

  1096.     if(!s->mb_x)

  1097.         vmvmask &= ~0x1111;

  1098.     if(r->rv30){ //RV30 marks both subblocks on the edge for filtering

  1099.         vmvmask |= (vmvmask & 0x4444) >> 1;

  1100.         hmvmask |= (hmvmask & 0x0F00) >> 4;

  1101.         if(s->mb_x)

  1102.             r->deblock_coefs[s->mb_x - 1 + s->mb_y*s->mb_stride] |= (vmvmask & 0x1111) << 3;

  1103.         if(!s->first_slice_line)

  1104.             r->deblock_coefs[s->mb_x + (s->mb_y - 1)*s->mb_stride] |= (hmvmask & 0xF) << 12;

  1105.     }

  1106.     return hmvmask | vmvmask;

  1107. }

  1108. 

  1109. static int rv34_decode_macroblock(RV34DecContext *r, int8_t *intra_types)

  1110. {

  1111.     MpegEncContext *s = &r->s;

  1112.     GetBitContext *gb = &s->gb;

  1113.     int cbp, cbp2;

  1114.     int i, blknum, blkoff;

  1115.     DCTELEM block16[64];

  1116.     int luma_dc_quant;

  1117.     int dist;

  1118.     int mb_pos = s->mb_x + s->mb_y * s->mb_stride;

  1119. 

  1120.     // Calculate which neighbours are available. Maybe it's worth optimizing too.

  1121.     memset(r->avail_cache, 0, sizeof(r->avail_cache));

  1122.     fill_rectangle(r->avail_cache + 6, 2, 2, 4, 1, 4);

  1123.     dist = (s->mb_x - s->resync_mb_x) + (s->mb_y - s->resync_mb_y) * s->mb_width;

  1124.     if(s->mb_x && dist)

  1125.         r->avail_cache[5] =

  1126.         r->avail_cache[9] = s->current_picture_ptr->f.mb_type[mb_pos - 1];

  1127.     if(dist >= s->mb_width)

  1128.         r->avail_cache[2] =

  1129.         r->avail_cache[3] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride];

  1130.     if(((s->mb_x+1) < s->mb_width) && dist >= s->mb_width - 1)

  1131.         r->avail_cache[4] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride + 1];

  1132.     if(s->mb_x && dist > s->mb_width)

  1133.         r->avail_cache[1] = s->current_picture_ptr->f.mb_type[mb_pos - s->mb_stride - 1];

  1134. 

  1135.     s->qscale = r->si.quant;

  1136.     cbp = cbp2 = rv34_decode_mb_header(r, intra_types);

  1137.     r->cbp_luma  [mb_pos] = cbp;

  1138.     r->cbp_chroma[mb_pos] = cbp >> 16;

  1139.     if(s->pict_type == AV_PICTURE_TYPE_I)

  1140.         r->deblock_coefs[mb_pos] = 0xFFFF;

  1141.     else

  1142.         r->deblock_coefs[mb_pos] = rv34_set_deblock_coef(r) | r->cbp_luma[mb_pos];

  1143.     s->current_picture_ptr->f.qscale_table[mb_pos] = s->qscale;

  1144. 

  1145.     if(cbp == -1)

  1146.         return -1;

  1147. 

  1148.     luma_dc_quant = r->block_type == RV34_MB_P_MIX16x16 ? r->luma_dc_quant_p[s->qscale] : r->luma_dc_quant_i[s->qscale];

  1149.     if(r->is16){

  1150.         memset(block16, 0, sizeof(block16));

  1151.         rv34_decode_block(block16, gb, r->cur_vlcs, 3, 0);

  1152.         rv34_dequant4x4_16x16(block16, rv34_qscale_tab[luma_dc_quant],rv34_qscale_tab[s->qscale]);

  1153.         r->rdsp.rv34_inv_transform_tab[1](block16);

  1154.     }

  1155. 

  1156.     for(i = 0; i < 16; i++, cbp >>= 1){

  1157.         if(!r->is16 && !(cbp & 1)) continue;

  1158.         blknum = ((i & 2) >> 1) + ((i & 8) >> 2);

  1159.         blkoff = ((i & 1) << 2) + ((i & 4) << 3);

  1160.         if(cbp & 1)

  1161.             rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->luma_vlc, 0);

  1162.         rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[s->qscale],rv34_qscale_tab[s->qscale]);

  1163.         if(r->is16) //FIXME: optimize

  1164.             s->block[blknum][blkoff] = block16[(i & 3) | ((i & 0xC) << 1)];

  1165.         r->rdsp.rv34_inv_transform_tab[0](s->block[blknum] + blkoff);

  1166.     }

  1167.     if(r->block_type == RV34_MB_P_MIX16x16)

  1168.         r->cur_vlcs = choose_vlc_set(r->si.quant, r->si.vlc_set, 1);

  1169.     for(; i < 24; i++, cbp >>= 1){

  1170.         if(!(cbp & 1)) continue;

  1171.         blknum = ((i & 4) >> 2) + 4;

  1172.         blkoff = ((i & 1) << 2) + ((i & 2) << 4);

  1173.         rv34_decode_block(s->block[blknum] + blkoff, gb, r->cur_vlcs, r->chroma_vlc, 1);

  1174.         rv34_dequant4x4(s->block[blknum] + blkoff, rv34_qscale_tab[rv34_chroma_quant[1][s->qscale]],rv34_qscale_tab[rv34_chroma_quant[0][s->qscale]]);

  1175.         r->rdsp.rv34_inv_transform_tab[0](s->block[blknum] + blkoff);

  1176.     }

  1177.     if (IS_INTRA(s->current_picture_ptr->f.mb_type[mb_pos]))

  1178.         rv34_output_macroblock(r, intra_types, cbp2, r->is16);

  1179.     else

  1180.         rv34_apply_differences(r, cbp2);

  1181. 

  1182.     return 0;

  1183. }

  1184. 

  1185. static int check_slice_end(RV34DecContext *r, MpegEncContext *s)

  1186. {

  1187.     int bits;

  1188.     if(s->mb_y >= s->mb_height)

  1189.         return 1;

  1190.     if(!s->mb_num_left)

  1191.         return 1;

  1192.     if(r->s.mb_skip_run > 1)

  1193.         return 0;

  1194.     bits = r->bits - get_bits_count(&s->gb);

  1195.     if(bits < 0 || (bits < 8 && !show_bits(&s->gb, bits)))

  1196.         return 1;

  1197.     return 0;

  1198. }

  1199. 

  1200. static int rv34_decode_slice(RV34DecContext *r, int end, const uint8_t* buf, int buf_size)

  1201. {

  1202.     MpegEncContext *s = &r->s;

  1203.     GetBitContext *gb = &s->gb;

  1204.     int mb_pos;

  1205.     int res;

  1206. 

  1207.     init_get_bits(&r->s.gb, buf, buf_size*8);

  1208.     res = r->parse_slice_header(r, gb, &r->si);

  1209.     if(res < 0){

  1210.         av_log(s->avctx, AV_LOG_ERROR, "Incorrect or unknown slice header\n");

  1211.         return -1;

  1212.     }

  1213. 

  1214.     if ((s->mb_x == 0 && s->mb_y == 0) || s->current_picture_ptr==NULL) {

  1215.         if(s->width != r->si.width || s->height != r->si.height){

  1216.             av_log(s->avctx, AV_LOG_DEBUG, "Changing dimensions to %dx%d\n", r->si.width,r->si.height);

  1217.             MPV_common_end(s);

  1218.             s->width  = r->si.width;

  1219.             s->height = r->si.height;

  1220.             avcodec_set_dimensions(s->avctx, s->width, s->height);

  1221.             if(MPV_common_init(s) < 0)

  1222.                 return -1;

  1223.             r->intra_types_stride = s->mb_width*4 + 4;

  1224.             r->intra_types_hist = av_realloc(r->intra_types_hist, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));

  1225.             r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;

  1226.             r->mb_type = av_realloc(r->mb_type, r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));

  1227.             r->cbp_luma   = av_realloc(r->cbp_luma,   r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));

  1228.             r->cbp_chroma = av_realloc(r->cbp_chroma, r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));

  1229.             r->deblock_coefs = av_realloc(r->deblock_coefs, r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));

  1230.             av_freep(&r->tmp_b_block_base);

  1231.         }

  1232.         s->pict_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;

  1233.         if(MPV_frame_start(s, s->avctx) < 0)

  1234.             return -1;

  1235.         ff_er_frame_start(s);

  1236.         if (!r->tmp_b_block_base) {

  1237.             int i;

  1238. 

  1239.             r->tmp_b_block_base = av_malloc(s->linesize * 48);

  1240.             for (i = 0; i < 2; i++)

  1241.                 r->tmp_b_block_y[i] = r->tmp_b_block_base + i * 16 * s->linesize;

  1242.             for (i = 0; i < 4; i++)

  1243.                 r->tmp_b_block_uv[i] = r->tmp_b_block_base + 32 * s->linesize

  1244.                                        + (i >> 1) * 8 * s->uvlinesize + (i & 1) * 16;

  1245.         }

  1246.         r->cur_pts = r->si.pts;

  1247.         if(s->pict_type != AV_PICTURE_TYPE_B){

  1248.             r->last_pts = r->next_pts;

  1249.             r->next_pts = r->cur_pts;

  1250.         }else{

  1251.             int refdist = GET_PTS_DIFF(r->next_pts, r->last_pts);

  1252.             int dist0   = GET_PTS_DIFF(r->cur_pts,  r->last_pts);

  1253.             int dist1   = GET_PTS_DIFF(r->next_pts, r->cur_pts);

  1254. 

  1255.             if(!refdist){

  1256.                 r->weight1 = r->weight2 = 8192;

  1257.             }else{

  1258.                 r->weight1 = (dist0 << 14) / refdist;

  1259.                 r->weight2 = (dist1 << 14) / refdist;

  1260.             }

  1261.         }

  1262.         s->mb_x = s->mb_y = 0;

  1263.     } else {

  1264.         int slice_type = r->si.type ? r->si.type : AV_PICTURE_TYPE_I;

  1265. 

  1266.         if (slice_type != s->pict_type) {

  1267.             av_log(s->avctx, AV_LOG_ERROR, "Slice type mismatch\n");

  1268.             return AVERROR_INVALIDDATA;

  1269.         }

  1270.         if (s->width != r->si.width || s->height != r->si.height) {

  1271.             av_log(s->avctx, AV_LOG_ERROR, "Size mismatch\n");

  1272.             return AVERROR_INVALIDDATA;

  1273.         }

  1274.     }

  1275. 

  1276.     r->si.end = end;

  1277.     s->qscale = r->si.quant;

  1278.     r->bits = buf_size*8;

  1279.     s->mb_num_left = r->si.end - r->si.start;

  1280.     r->s.mb_skip_run = 0;

  1281. 

  1282.     mb_pos = s->mb_x + s->mb_y * s->mb_width;

  1283.     if(r->si.start != mb_pos){

  1284.         av_log(s->avctx, AV_LOG_ERROR, "Slice indicates MB offset %d, got %d\n", r->si.start, mb_pos);

  1285.         s->mb_x = r->si.start % s->mb_width;

  1286.         s->mb_y = r->si.start / s->mb_width;

  1287.     }

  1288.     memset(r->intra_types_hist, -1, r->intra_types_stride * 4 * 2 * sizeof(*r->intra_types_hist));

  1289.     s->first_slice_line = 1;

  1290.     s->resync_mb_x = s->mb_x;

  1291.     s->resync_mb_y = s->mb_y;

  1292. 

  1293.     ff_init_block_index(s);

  1294.     while(!check_slice_end(r, s)) {

  1295.         ff_update_block_index(s);

  1296.         s->dsp.clear_blocks(s->block[0]);

  1297. 

  1298.         if(rv34_decode_macroblock(r, r->intra_types + s->mb_x * 4 + 4) < 0){

  1299.             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);

  1300.             return -1;

  1301.         }

  1302.         if (++s->mb_x == s->mb_width) {

  1303.             s->mb_x = 0;

  1304.             s->mb_y++;

  1305.             ff_init_block_index(s);

  1306. 

  1307.             memmove(r->intra_types_hist, r->intra_types, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1308.             memset(r->intra_types, -1, r->intra_types_stride * 4 * sizeof(*r->intra_types_hist));

  1309. 

  1310.             if(r->loop_filter && s->mb_y >= 2)

  1311.                 r->loop_filter(r, s->mb_y - 2);

  1312.         }

  1313.         if(s->mb_x == s->resync_mb_x)

  1314.             s->first_slice_line=0;

  1315.         s->mb_num_left--;

  1316.     }

  1317.     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);

  1318. 

  1319.     return s->mb_y == s->mb_height;

  1320. }

  1321. 

  1322. /** @} */ // recons group end

  1323. 

  1324. /**

  1325.  * Initialize decoder.

  1326.  */

  1327. av_cold int ff_rv34_decode_init(AVCodecContext *avctx)

  1328. {

  1329.     RV34DecContext *r = avctx->priv_data;

  1330.     MpegEncContext *s = &r->s;

  1331. 

  1332.     MPV_decode_defaults(s);

  1333.     s->avctx      = avctx;

  1334.     s->out_format = FMT_H263;

  1335.     s->codec_id   = avctx->codec_id;

  1336. 

  1337.     s->width  = avctx->width;

  1338.     s->height = avctx->height;

  1339. 

  1340.     r->s.avctx = avctx;

  1341.     avctx->flags |= CODEC_FLAG_EMU_EDGE;

  1342.     r->s.flags |= CODEC_FLAG_EMU_EDGE;

  1343.     avctx->pix_fmt = PIX_FMT_YUV420P;

  1344.     avctx->has_b_frames = 1;

  1345.     s->low_delay = 0;

  1346. 

  1347.     if (MPV_common_init(s) < 0)

  1348.         return -1;

  1349. 

  1350.     ff_h264_pred_init(&r->h, CODEC_ID_RV40, 8, 1);

  1351. 

  1352. #if CONFIG_RV30_DECODER

  1353.     if (avctx->codec_id == CODEC_ID_RV30)

  1354.         ff_rv30dsp_init(&r->rdsp, &r->s.dsp);

  1355. #endif

  1356. #if CONFIG_RV40_DECODER

  1357.     if (avctx->codec_id == CODEC_ID_RV40)

  1358.         ff_rv40dsp_init(&r->rdsp, &r->s.dsp);

  1359. #endif

  1360. 

  1361.     r->intra_types_stride = 4*s->mb_stride + 4;

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

  1363.     r->intra_types = r->intra_types_hist + r->intra_types_stride * 4;

  1364. 

  1365.     r->mb_type = av_mallocz(r->s.mb_stride * r->s.mb_height * sizeof(*r->mb_type));

  1366. 

  1367.     r->cbp_luma   = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_luma));

  1368.     r->cbp_chroma = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->cbp_chroma));

  1369.     r->deblock_coefs = av_malloc(r->s.mb_stride * r->s.mb_height * sizeof(*r->deblock_coefs));

  1370. 

  1371.     if(!intra_vlcs[0].cbppattern[0].bits)

  1372.         rv34_init_tables();

  1373. 

  1374.     return 0;

  1375. }

  1376. 

  1377. static int get_slice_offset(AVCodecContext *avctx, const uint8_t *buf, int n)

  1378. {

  1379.     if(avctx->slice_count) return avctx->slice_offset[n];

  1380.     else                   return AV_RL32(buf + n*8 - 4) == 1 ? AV_RL32(buf + n*8) :  AV_RB32(buf + n*8);

  1381. }

  1382. 

  1383. int ff_rv34_decode_frame(AVCodecContext *avctx,

  1384.                             void *data, int *data_size,

  1385.                             AVPacket *avpkt)

  1386. {

  1387.     const uint8_t *buf = avpkt->data;

  1388.     int buf_size = avpkt->size;

  1389.     RV34DecContext *r = avctx->priv_data;

  1390.     MpegEncContext *s = &r->s;

  1391.     AVFrame *pict = data;

  1392.     SliceInfo si;

  1393.     int i;

  1394.     int slice_count;

  1395.     const uint8_t *slices_hdr = NULL;

  1396.     int last = 0;

  1397. 

  1398.     /* no supplementary picture */

  1399.     if (buf_size == 0) {

  1400.         /* special case for last picture */

  1401.         if (s->low_delay==0 && s->next_picture_ptr) {

  1402.             *pict = *(AVFrame*)s->next_picture_ptr;

  1403.             s->next_picture_ptr = NULL;

  1404. 

  1405.             *data_size = sizeof(AVFrame);

  1406.         }

  1407.         return 0;

  1408.     }

  1409. 

  1410.     if(!avctx->slice_count){

  1411.         slice_count = (*buf++) + 1;

  1412.         slices_hdr = buf + 4;

  1413.         buf += 8 * slice_count;

  1414.         buf_size -= 1 + 8 * slice_count;

  1415.     }else

  1416.         slice_count = avctx->slice_count;

  1417. 

  1418.     //parse first slice header to check whether this frame can be decoded

  1419.     if(get_slice_offset(avctx, slices_hdr, 0) < 0 ||

  1420.        get_slice_offset(avctx, slices_hdr, 0) > buf_size){

  1421.         av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1422.         return -1;

  1423.     }

  1424.     init_get_bits(&s->gb, buf+get_slice_offset(avctx, slices_hdr, 0), (buf_size-get_slice_offset(avctx, slices_hdr, 0))*8);

  1425.     if(r->parse_slice_header(r, &r->s.gb, &si) < 0 || si.start){

  1426.         av_log(avctx, AV_LOG_ERROR, "First slice header is incorrect\n");

  1427.         return -1;

  1428.     }

  1429.     if ((!s->last_picture_ptr || !s->last_picture_ptr->f.data[0]) && si.type == AV_PICTURE_TYPE_B)

  1430.         return -1;

  1431.     if(   (avctx->skip_frame >= AVDISCARD_NONREF && si.type==AV_PICTURE_TYPE_B)

  1432.        || (avctx->skip_frame >= AVDISCARD_NONKEY && si.type!=AV_PICTURE_TYPE_I)

  1433.        ||  avctx->skip_frame >= AVDISCARD_ALL)

  1434.         return avpkt->size;

  1435. 

  1436.     for(i = 0; i < slice_count; i++){

  1437.         int offset = get_slice_offset(avctx, slices_hdr, i);

  1438.         int size;

  1439.         if(i+1 == slice_count)

  1440.             size = buf_size - offset;

  1441.         else

  1442.             size = get_slice_offset(avctx, slices_hdr, i+1) - offset;

  1443. 

  1444.         if(offset < 0 || offset > buf_size){

  1445.             av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1446.             break;

  1447.         }

  1448. 

  1449.         r->si.end = s->mb_width * s->mb_height;

  1450.         if(i+1 < slice_count){

  1451.             if (get_slice_offset(avctx, slices_hdr, i+1) < 0 ||

  1452.                 get_slice_offset(avctx, slices_hdr, i+1) > buf_size) {

  1453.                 av_log(avctx, AV_LOG_ERROR, "Slice offset is invalid\n");

  1454.                 break;

  1455.             }

  1456.             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);

  1457.             if(r->parse_slice_header(r, &r->s.gb, &si) < 0){

  1458.                 if(i+2 < slice_count)

  1459.                     size = get_slice_offset(avctx, slices_hdr, i+2) - offset;

  1460.                 else

  1461.                     size = buf_size - offset;

  1462.             }else

  1463.                 r->si.end = si.start;

  1464.         }

  1465.         if (size < 0 || size > buf_size - offset) {

  1466.             av_log(avctx, AV_LOG_ERROR, "Slice size is invalid\n");

  1467.             break;

  1468.         }

  1469.         last = rv34_decode_slice(r, r->si.end, buf + offset, size);

  1470.         s->mb_num_left = r->s.mb_x + r->s.mb_y*r->s.mb_width - r->si.start;

  1471.         if(last)

  1472.             break;

  1473.     }

  1474. 

  1475.     if(last && s->current_picture_ptr){

  1476.         if(r->loop_filter)

  1477.             r->loop_filter(r, s->mb_height - 1);

  1478.         ff_er_frame_end(s);

  1479.         MPV_frame_end(s);

  1480.         if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {

  1481.             *pict = *(AVFrame*)s->current_picture_ptr;

  1482.         } else if (s->last_picture_ptr != NULL) {

  1483.             *pict = *(AVFrame*)s->last_picture_ptr;

  1484.         }

  1485. 

  1486.         if(s->last_picture_ptr || s->low_delay){

  1487.             *data_size = sizeof(AVFrame);

  1488.             ff_print_debug_info(s, pict);

  1489.         }

  1490.         s->current_picture_ptr = NULL; //so we can detect if frame_end wasnt called (find some nicer solution...)

  1491.     }

  1492.     return avpkt->size;

  1493. }

  1494. 

  1495. av_cold int ff_rv34_decode_end(AVCodecContext *avctx)

  1496. {

  1497.     RV34DecContext *r = avctx->priv_data;

  1498. 

  1499.     MPV_common_end(&r->s);

  1500. 

  1501.     av_freep(&r->intra_types_hist);

  1502.     r->intra_types = NULL;

  1503.     av_freep(&r->tmp_b_block_base);

  1504.     av_freep(&r->mb_type);

  1505.     av_freep(&r->cbp_luma);

  1506.     av_freep(&r->cbp_chroma);

  1507.     av_freep(&r->deblock_coefs);

  1508. 

  1509.     return 0;

  1510. }