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.
43040 Commits
32 Branches
369MB
Tree: ca2f19b9cc
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ca2f19b9cc'
${ noResults }
FFmpeg/libavcodec/intrax8.c

843 lines
27KB
Raw Blame History 

  1. /*

  2.  * This file is part of Libav.

  3.  *

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

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

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

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

  8.  *

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

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

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

  12.  * Lesser General Public License for more details.

  13.  *

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

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

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

  17.  */

  18. 

  19. /**

  20.  * @file

  21.  * @brief IntraX8 (J-Frame) subdecoder, used by WMV2 and VC-1

  22.  */

  23. 

  24. #include "avcodec.h"

  25. #include "get_bits.h"

  26. #include "idctdsp.h"

  27. #include "mpegvideo.h"

  28. #include "msmpeg4data.h"

  29. #include "intrax8huf.h"

  30. #include "intrax8.h"

  31. #include "intrax8dsp.h"

  32. 

  33. #define MAX_TABLE_DEPTH(table_bits, max_bits) \

  34.     ((max_bits + table_bits - 1) / table_bits)

  35. 

  36. #define DC_VLC_BITS 9

  37. #define AC_VLC_BITS 9

  38. #define OR_VLC_BITS 7

  39. 

  40. #define DC_VLC_MTD MAX_TABLE_DEPTH(DC_VLC_BITS, MAX_DC_VLC_BITS)

  41. #define AC_VLC_MTD MAX_TABLE_DEPTH(AC_VLC_BITS, MAX_AC_VLC_BITS)

  42. #define OR_VLC_MTD MAX_TABLE_DEPTH(OR_VLC_BITS, MAX_OR_VLC_BITS)

  43. 

  44. static VLC j_ac_vlc[2][2][8];  // [quant < 13], [intra / inter], [select]

  45. static VLC j_dc_vlc[2][8];     // [quant], [select]

  46. static VLC j_orient_vlc[2][4]; // [quant], [select]

  47. 

  48. static av_cold int x8_vlc_init(void)

  49. {

  50.     int i;

  51.     int offset = 0;

  52.     int sizeidx = 0;

  53.     static const uint16_t sizes[8 * 4 + 8 * 2 + 2 + 4] = {

  54.         576, 548, 582, 618, 546, 616, 560, 642,

  55.         584, 582, 704, 664, 512, 544, 656, 640,

  56.         512, 648, 582, 566, 532, 614, 596, 648,

  57.         586, 552, 584, 590, 544, 578, 584, 624,

  58. 

  59.         528, 528, 526, 528, 536, 528, 526, 544,

  60.         544, 512, 512, 528, 528, 544, 512, 544,

  61. 

  62.         128, 128, 128, 128, 128, 128,

  63.     };

  64. 

  65.     static VLC_TYPE table[28150][2];

  66. 

  67. // set ac tables

  68. #define init_ac_vlc(dst, src)                                                 \

  69.     do {                                                                      \

  70.         dst.table           = &table[offset];                                 \

  71.         dst.table_allocated = sizes[sizeidx];                                 \

  72.         offset             += sizes[sizeidx++];                               \

  73.         init_vlc(&dst, AC_VLC_BITS, 77, &src[1], 4, 2, &src[0], 4, 2,         \

  74.                  INIT_VLC_USE_NEW_STATIC);                                    \

  75.     } while(0)

  76. 

  77.     for (i = 0; i < 8; i++) {

  78.         init_ac_vlc(j_ac_vlc[0][0][i], x8_ac0_highquant_table[i][0]);

  79.         init_ac_vlc(j_ac_vlc[0][1][i], x8_ac1_highquant_table[i][0]);

  80.         init_ac_vlc(j_ac_vlc[1][0][i], x8_ac0_lowquant_table[i][0]);

  81.         init_ac_vlc(j_ac_vlc[1][1][i], x8_ac1_lowquant_table[i][0]);

  82.     }

  83. #undef init_ac_vlc

  84. 

  85. // set dc tables

  86. #define init_dc_vlc(dst, src)                                                 \

  87.     do {                                                                      \

  88.         dst.table           = &table[offset];                                 \

  89.         dst.table_allocated = sizes[sizeidx];                                 \

  90.         offset             += sizes[sizeidx++];                               \

  91.         init_vlc(&dst, DC_VLC_BITS, 34, &src[1], 4, 2, &src[0], 4, 2,         \

  92.                  INIT_VLC_USE_NEW_STATIC);                                    \

  93.     } while(0)

  94. 

  95.     for (i = 0; i < 8; i++) {

  96.         init_dc_vlc(j_dc_vlc[0][i], x8_dc_highquant_table[i][0]);

  97.         init_dc_vlc(j_dc_vlc[1][i], x8_dc_lowquant_table[i][0]);

  98.     }

  99. #undef init_dc_vlc

  100. 

  101. // set orient tables

  102. #define init_or_vlc(dst, src)                                                 \

  103.     do {                                                                      \

  104.         dst.table           = &table[offset];                                 \

  105.         dst.table_allocated = sizes[sizeidx];                                 \

  106.         offset             += sizes[sizeidx++];                               \

  107.         init_vlc(&dst, OR_VLC_BITS, 12, &src[1], 4, 2, &src[0], 4, 2,         \

  108.                  INIT_VLC_USE_NEW_STATIC);                                    \

  109.     } while(0)

  110. 

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

  112.         init_or_vlc(j_orient_vlc[0][i], x8_orient_highquant_table[i][0]);

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

  114.         init_or_vlc(j_orient_vlc[1][i], x8_orient_lowquant_table[i][0]);

  115. #undef init_or_vlc

  116. 

  117.     if (offset != sizeof(table) / sizeof(VLC_TYPE) / 2) {

  118.         av_log(NULL, AV_LOG_ERROR, "table size %zd does not match needed %i\n",

  119.                sizeof(table) / sizeof(VLC_TYPE) / 2, offset);

  120.         return AVERROR_INVALIDDATA;

  121.     }

  122. 

  123.     return 0;

  124. }

  125. 

  126. static void x8_reset_vlc_tables(IntraX8Context *w)

  127. {

  128.     memset(w->j_dc_vlc, 0, sizeof(w->j_dc_vlc));

  129.     memset(w->j_ac_vlc, 0, sizeof(w->j_ac_vlc));

  130.     w->j_orient_vlc = NULL;

  131. }

  132. 

  133. static inline void x8_select_ac_table(IntraX8Context *const w, int mode)

  134. {

  135.     int table_index;

  136. 

  137.     assert(mode < 4);

  138. 

  139.     if (w->j_ac_vlc[mode])

  140.         return;

  141. 

  142.     table_index       = get_bits(w->gb, 3);

  143.     // 2 modes use same tables

  144.     w->j_ac_vlc[mode] = &j_ac_vlc[w->quant < 13][mode >> 1][table_index];

  145. 

  146.     assert(w->j_ac_vlc[mode]);

  147. }

  148. 

  149. static inline int x8_get_orient_vlc(IntraX8Context *w)

  150. {

  151.     if (!w->j_orient_vlc) {

  152.         int table_index = get_bits(w->gb, 1 + (w->quant < 13));

  153.         w->j_orient_vlc = &j_orient_vlc[w->quant < 13][table_index];

  154.     }

  155.     assert(w->j_orient_vlc);

  156.     assert(w->j_orient_vlc->table);

  157. 

  158.     return get_vlc2(w->gb, w->j_orient_vlc->table, OR_VLC_BITS, OR_VLC_MTD);

  159. }

  160. 

  161. #define extra_bits(eb)  (eb)        // 3 bits

  162. #define extra_run       (0xFF << 8) // 1 bit

  163. #define extra_level     (0x00 << 8) // 1 bit

  164. #define run_offset(r)   ((r) << 16) // 6 bits

  165. #define level_offset(l) ((l) << 24) // 5 bits

  166. static const uint32_t ac_decode_table[] = {

  167.     /* 46 */ extra_bits(3) | extra_run   | run_offset(16) | level_offset(0),

  168.     /* 47 */ extra_bits(3) | extra_run   | run_offset(24) | level_offset(0),

  169.     /* 48 */ extra_bits(2) | extra_run   | run_offset(4)  | level_offset(1),

  170.     /* 49 */ extra_bits(3) | extra_run   | run_offset(8)  | level_offset(1),

  171. 

  172.     /* 50 */ extra_bits(5) | extra_run   | run_offset(32) | level_offset(0),

  173.     /* 51 */ extra_bits(4) | extra_run   | run_offset(16) | level_offset(1),

  174. 

  175.     /* 52 */ extra_bits(2) | extra_level | run_offset(0)  | level_offset(4),

  176.     /* 53 */ extra_bits(2) | extra_level | run_offset(0)  | level_offset(8),

  177.     /* 54 */ extra_bits(2) | extra_level | run_offset(0)  | level_offset(12),

  178.     /* 55 */ extra_bits(3) | extra_level | run_offset(0)  | level_offset(16),

  179.     /* 56 */ extra_bits(3) | extra_level | run_offset(0)  | level_offset(24),

  180. 

  181.     /* 57 */ extra_bits(2) | extra_level | run_offset(1)  | level_offset(3),

  182.     /* 58 */ extra_bits(3) | extra_level | run_offset(1)  | level_offset(7),

  183. 

  184.     /* 59 */ extra_bits(2) | extra_run   | run_offset(16) | level_offset(0),

  185.     /* 60 */ extra_bits(2) | extra_run   | run_offset(20) | level_offset(0),

  186.     /* 61 */ extra_bits(2) | extra_run   | run_offset(24) | level_offset(0),

  187.     /* 62 */ extra_bits(2) | extra_run   | run_offset(28) | level_offset(0),

  188.     /* 63 */ extra_bits(4) | extra_run   | run_offset(32) | level_offset(0),

  189.     /* 64 */ extra_bits(4) | extra_run   | run_offset(48) | level_offset(0),

  190. 

  191.     /* 65 */ extra_bits(2) | extra_run   | run_offset(4)  | level_offset(1),

  192.     /* 66 */ extra_bits(3) | extra_run   | run_offset(8)  | level_offset(1),

  193.     /* 67 */ extra_bits(4) | extra_run   | run_offset(16) | level_offset(1),

  194. 

  195.     /* 68 */ extra_bits(2) | extra_level | run_offset(0)  | level_offset(4),

  196.     /* 69 */ extra_bits(3) | extra_level | run_offset(0)  | level_offset(8),

  197.     /* 70 */ extra_bits(4) | extra_level | run_offset(0)  | level_offset(16),

  198. 

  199.     /* 71 */ extra_bits(2) | extra_level | run_offset(1)  | level_offset(3),

  200.     /* 72 */ extra_bits(3) | extra_level | run_offset(1)  | level_offset(7),

  201. };

  202. #undef extra_bits

  203. #undef extra_run

  204. #undef extra_level

  205. #undef run_offset

  206. #undef level_offset

  207. 

  208. static void x8_get_ac_rlf(IntraX8Context *const w, const int mode,

  209.                           int *const run, int *const level, int *const final)

  210. {

  211.     int i, e;

  212. 

  213. //    x8_select_ac_table(w, mode);

  214.     i = get_vlc2(w->gb, w->j_ac_vlc[mode]->table, AC_VLC_BITS, AC_VLC_MTD);

  215. 

  216.     if (i < 46) { // [0-45]

  217.         int t, l;

  218.         if (i < 0) {

  219.             (*level) =

  220.             (*final) =      // prevent 'may be used unilitialized'

  221.             (*run)   = 64;  // this would cause error exit in the ac loop

  222.             return;

  223.         }

  224. 

  225.         /*

  226.          * i == 0-15  r = 0-15 l = 0; r = i & %01111

  227.          * i == 16-19 r = 0-3  l = 1; r = i & %00011

  228.          * i == 20-21 r = 0-1  l = 2; r = i & %00001

  229.          * i == 22    r = 0    l = 3; r = i & %00000

  230.          */

  231. 

  232.         (*final) =

  233.         t        = (i > 22);

  234.         i       -= 23 * t;

  235. 

  236.         /* l = lut_l[i / 2] = { 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 3 }[i >> 1];

  237.          *     11 10'01 01'00 00'00 00'00 00'00 00 => 0xE50000 */

  238.         l = (0xE50000 >> (i & (0x1E))) & 3; // 0x1E or (~1) or ((i >> 1) << 1)

  239. 

  240.         /* t = lut_mask[l] = { 0x0f, 0x03, 0x01, 0x00 }[l];

  241.          *     as i < 256 the higher bits do not matter */

  242.         t = (0x01030F >> (l << 3));

  243. 

  244.         (*run)   = i & t;

  245.         (*level) = l;

  246.     } else if (i < 73) { // [46-72]

  247.         uint32_t sm;

  248.         uint32_t mask;

  249. 

  250.         i -= 46;

  251.         sm = ac_decode_table[i];

  252. 

  253.         e    = get_bits(w->gb, sm & 0xF);

  254.         sm >>= 8;                               // 3bits

  255.         mask = sm & 0xff;

  256.         sm >>= 8;                               // 1bit

  257. 

  258.         (*run)   = (sm & 0xff) + (e & (mask));  // 6bits

  259.         (*level) = (sm >> 8)   + (e & (~mask)); // 5bits

  260.         (*final) = i > (58 - 46);

  261.     } else if (i < 75) { // [73-74]

  262.         static const uint8_t crazy_mix_runlevel[32] = {

  263.             0x22, 0x32, 0x33, 0x53, 0x23, 0x42, 0x43, 0x63,

  264.             0x24, 0x52, 0x34, 0x73, 0x25, 0x62, 0x44, 0x83,

  265.             0x26, 0x72, 0x35, 0x54, 0x27, 0x82, 0x45, 0x64,

  266.             0x28, 0x92, 0x36, 0x74, 0x29, 0xa2, 0x46, 0x84,

  267.         };

  268. 

  269.         (*final) = !(i & 1);

  270.         e        = get_bits(w->gb, 5); // get the extra bits

  271.         (*run)   = crazy_mix_runlevel[e] >> 4;

  272.         (*level) = crazy_mix_runlevel[e] & 0x0F;

  273.     } else {

  274.         (*level) = get_bits(w->gb, 7 - 3 * (i & 1));

  275.         (*run)   = get_bits(w->gb, 6);

  276.         (*final) = get_bits1(w->gb);

  277.     }

  278.     return;

  279. }

  280. 

  281. /* static const uint8_t dc_extra_sbits[] = {

  282.  *     0, 1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7,

  283.  * }; */

  284. static const uint8_t dc_index_offset[] = {

  285.     0, 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,

  286. };

  287. 

  288. static int x8_get_dc_rlf(IntraX8Context *const w, const int mode,

  289.                          int *const level, int *const final)

  290. {

  291.     int i, e, c;

  292. 

  293.     assert(mode < 3);

  294.     if (!w->j_dc_vlc[mode]) {

  295.         int table_index = get_bits(w->gb, 3);

  296.         // 4 modes, same table

  297.         w->j_dc_vlc[mode] = &j_dc_vlc[w->quant < 13][table_index];

  298.     }

  299.     assert(w->j_dc_vlc);

  300.     assert(w->j_dc_vlc[mode]->table);

  301. 

  302.     i = get_vlc2(w->gb, w->j_dc_vlc[mode]->table, DC_VLC_BITS, DC_VLC_MTD);

  303. 

  304.     /* (i >= 17) { i -= 17; final =1; } */

  305.     c        = i > 16;

  306.     (*final) = c;

  307.     i       -= 17 * c;

  308. 

  309.     if (i <= 0) {

  310.         (*level) = 0;

  311.         return -i;

  312.     }

  313.     c  = (i + 1) >> 1; // hackish way to calculate dc_extra_sbits[]

  314.     c -= c > 1;

  315. 

  316.     e = get_bits(w->gb, c); // get the extra bits

  317.     i = dc_index_offset[i] + (e >> 1);

  318. 

  319.     e        = -(e & 1); // 0, 0xffffff

  320.     (*level) = (i ^ e) - e; // (i ^ 0) -0  , (i ^ 0xff) - (-1)

  321.     return 0;

  322. }

  323. 

  324. // end of huffman

  325. 

  326. static int x8_setup_spatial_predictor(IntraX8Context *const w, const int chroma)

  327. {

  328.     int range;

  329.     int sum;

  330.     int quant;

  331. 

  332.     w->dsp.setup_spatial_compensation(w->dest[chroma], w->scratchpad,

  333.                                       w->frame->linesize[chroma > 0],

  334.                                       &range, &sum, w->edges);

  335.     if (chroma) {

  336.         w->orient = w->chroma_orient;

  337.         quant     = w->quant_dc_chroma;

  338.     } else {

  339.         quant = w->quant;

  340.     }

  341. 

  342.     w->flat_dc = 0;

  343.     if (range < quant || range < 3) {

  344.         w->orient = 0;

  345. 

  346.         // yep you read right, a +-1 idct error may break decoding!

  347.         if (range < 3) {

  348.             w->flat_dc      = 1;

  349.             sum            += 9;

  350.             // ((1 << 17) + 9) / (8 + 8 + 1 + 2) = 6899

  351.             w->predicted_dc = (sum * 6899) >> 17;

  352.         }

  353.     }

  354.     if (chroma)

  355.         return 0;

  356. 

  357.     assert(w->orient < 3);

  358.     if (range < 2 * w->quant) {

  359.         if ((w->edges & 3) == 0) {

  360.             if (w->orient == 1)

  361.                 w->orient = 11;

  362.             if (w->orient == 2)

  363.                 w->orient = 10;

  364.         } else {

  365.             w->orient = 0;

  366.         }

  367.         w->raw_orient = 0;

  368.     } else {

  369.         static const uint8_t prediction_table[3][12] = {

  370.             { 0, 8, 4, 10, 11, 2, 6, 9, 1, 3, 5, 7 },

  371.             { 4, 0, 8, 11, 10, 3, 5, 2, 6, 9, 1, 7 },

  372.             { 8, 0, 4, 10, 11, 1, 7, 2, 6, 9, 3, 5 },

  373.         };

  374.         w->raw_orient = x8_get_orient_vlc(w);

  375.         if (w->raw_orient < 0)

  376.             return -1;

  377.         assert(w->raw_orient < 12);

  378.         assert(w->orient < 3);

  379.         w->orient = prediction_table[w->orient][w->raw_orient];

  380.     }

  381.     return 0;

  382. }

  383. 

  384. static void x8_update_predictions(IntraX8Context *const w, const int orient,

  385.                                   const int est_run)

  386. {

  387.     MpegEncContext *const s = w->s;

  388. 

  389.     w->prediction_table[s->mb_x * 2 + (s->mb_y & 1)] = (est_run << 2) + 1 * (orient == 4) + 2 * (orient == 8);

  390. /*

  391.  * y = 2n + 0 -> // 0 2 4

  392.  * y = 2n + 1 -> // 1 3 5

  393.  */

  394. }

  395. 

  396. static void x8_get_prediction_chroma(IntraX8Context *const w)

  397. {

  398.     MpegEncContext *const s = w->s;

  399. 

  400.     w->edges  = 1 * (!(s->mb_x >> 1));

  401.     w->edges |= 2 * (!(s->mb_y >> 1));

  402.     w->edges |= 4 * (s->mb_x >= (2 * s->mb_width - 1)); // mb_x for chroma would always be odd

  403. 

  404.     w->raw_orient = 0;

  405.     // lut_co[8] = {inv,4,8,8, inv,4,8,8} <- => {1,1,0,0;1,1,0,0} => 0xCC

  406.     if (w->edges & 3) {

  407.         w->chroma_orient = 4 << ((0xCC >> w->edges) & 1);

  408.         return;

  409.     }

  410.     // block[x - 1][y | 1 - 1)]

  411.     w->chroma_orient = (w->prediction_table[2 * s->mb_x - 2] & 0x03) << 2;

  412. }

  413. 

  414. static void x8_get_prediction(IntraX8Context *const w)

  415. {

  416.     MpegEncContext *const s = w->s;

  417.     int a, b, c, i;

  418. 

  419.     w->edges  = 1 * (!s->mb_x);

  420.     w->edges |= 2 * (!s->mb_y);

  421.     w->edges |= 4 * (s->mb_x >= (2 * s->mb_width - 1));

  422. 

  423.     switch (w->edges & 3) {

  424.     case 0:

  425.         break;

  426.     case 1:

  427.         // take the one from the above block[0][y - 1]

  428.         w->est_run = w->prediction_table[!(s->mb_y & 1)] >> 2;

  429.         w->orient  = 1;

  430.         return;

  431.     case 2:

  432.         // take the one from the previous block[x - 1][0]

  433.         w->est_run = w->prediction_table[2 * s->mb_x - 2] >> 2;

  434.         w->orient  = 2;

  435.         return;

  436.     case 3:

  437.         w->est_run = 16;

  438.         w->orient  = 0;

  439.         return;

  440.     }

  441.     // no edge cases

  442.     b = w->prediction_table[2 * s->mb_x     + !(s->mb_y & 1)]; // block[x    ][y - 1]

  443.     a = w->prediction_table[2 * s->mb_x - 2 +  (s->mb_y & 1)]; // block[x - 1][y    ]

  444.     c = w->prediction_table[2 * s->mb_x - 2 + !(s->mb_y & 1)]; // block[x - 1][y - 1]

  445. 

  446.     w->est_run = FFMIN(b, a);

  447.     /* This condition has nothing to do with w->edges, even if it looks

  448.      * similar it would trigger if e.g. x = 3; y = 2;

  449.      * I guess somebody wrote something wrong and it became standard. */

  450.     if ((s->mb_x & s->mb_y) != 0)

  451.         w->est_run = FFMIN(c, w->est_run);

  452.     w->est_run >>= 2;

  453. 

  454.     a &= 3;

  455.     b &= 3;

  456.     c &= 3;

  457. 

  458.     i = (0xFFEAF4C4 >> (2 * b + 8 * a)) & 3;

  459.     if (i != 3)

  460.         w->orient = i;

  461.     else

  462.         w->orient = (0xFFEAD8 >> (2 * c + 8 * (w->quant > 12))) & 3;

  463. /*

  464.  * lut1[b][a] = {

  465.  * ->{ 0, 1, 0, pad },

  466.  *   { 0, 1, X, pad },

  467.  *   { 2, 2, 2, pad }

  468.  * }

  469.  * pad 2  2  2;

  470.  * pad X  1  0;

  471.  * pad 0  1  0 <-

  472.  * -> 11 10 '10 10 '11 11'01 00 '11 00'01 00 => 0xEAF4C4

  473.  *

  474.  * lut2[q>12][c] = {

  475.  * ->{ 0, 2, 1, pad},

  476.  *   { 2, 2, 2, pad}

  477.  * }

  478.  * pad 2  2  2;

  479.  * pad 1  2  0 <-

  480.  * -> 11 10'10 10 '11 01'10 00 => 0xEAD8

  481.  */

  482. }

  483. 

  484. static void x8_ac_compensation(IntraX8Context *const w, const int direction,

  485.                                const int dc_level)

  486. {

  487.     MpegEncContext *const s = w->s;

  488.     int t;

  489. #define B(x, y) s->block[0][w->idsp.idct_permutation[(x) + (y) * 8]]

  490. #define T(x)  ((x) * dc_level + 0x8000) >> 16;

  491.     switch (direction) {

  492.     case 0:

  493.         t        = T(3811); // h

  494.         B(1, 0) -= t;

  495.         B(0, 1) -= t;

  496. 

  497.         t        = T(487); // e

  498.         B(2, 0) -= t;

  499.         B(0, 2) -= t;

  500. 

  501.         t        = T(506); // f

  502.         B(3, 0) -= t;

  503.         B(0, 3) -= t;

  504. 

  505.         t        = T(135); // c

  506.         B(4, 0) -= t;

  507.         B(0, 4) -= t;

  508.         B(2, 1) += t;

  509.         B(1, 2) += t;

  510.         B(3, 1) += t;

  511.         B(1, 3) += t;

  512. 

  513.         t        = T(173); // d

  514.         B(5, 0) -= t;

  515.         B(0, 5) -= t;

  516. 

  517.         t        = T(61); // b

  518.         B(6, 0) -= t;

  519.         B(0, 6) -= t;

  520.         B(5, 1) += t;

  521.         B(1, 5) += t;

  522. 

  523.         t        = T(42); // a

  524.         B(7, 0) -= t;

  525.         B(0, 7) -= t;

  526.         B(4, 1) += t;

  527.         B(1, 4) += t;

  528.         B(4, 4) += t;

  529. 

  530.         t        = T(1084); // g

  531.         B(1, 1) += t;

  532. 

  533.         s->block_last_index[0] = FFMAX(s->block_last_index[0], 7 * 8);

  534.         break;

  535.     case 1:

  536.         B(0, 1) -= T(6269);

  537.         B(0, 3) -= T(708);

  538.         B(0, 5) -= T(172);

  539.         B(0, 7) -= T(73);

  540. 

  541.         s->block_last_index[0] = FFMAX(s->block_last_index[0], 7 * 8);

  542.         break;

  543.     case 2:

  544.         B(1, 0) -= T(6269);

  545.         B(3, 0) -= T(708);

  546.         B(5, 0) -= T(172);

  547.         B(7, 0) -= T(73);

  548. 

  549.         s->block_last_index[0] = FFMAX(s->block_last_index[0], 7);

  550.         break;

  551.     }

  552. #undef B

  553. #undef T

  554. }

  555. 

  556. static void dsp_x8_put_solidcolor(const uint8_t pix, uint8_t *dst,

  557.                                   const int linesize)

  558. {

  559.     int k;

  560.     for (k = 0; k < 8; k++) {

  561.         memset(dst, pix, 8);

  562.         dst += linesize;

  563.     }

  564. }

  565. 

  566. static const int16_t quant_table[64] = {

  567.     256, 256, 256, 256, 256, 256, 259, 262,

  568.     265, 269, 272, 275, 278, 282, 285, 288,

  569.     292, 295, 299, 303, 306, 310, 314, 317,

  570.     321, 325, 329, 333, 337, 341, 345, 349,

  571.     353, 358, 362, 366, 371, 375, 379, 384,

  572.     389, 393, 398, 403, 408, 413, 417, 422,

  573.     428, 433, 438, 443, 448, 454, 459, 465,

  574.     470, 476, 482, 488, 493, 499, 505, 511,

  575. };

  576. 

  577. static int x8_decode_intra_mb(IntraX8Context *const w, const int chroma)

  578. {

  579.     MpegEncContext *const s = w->s;

  580. 

  581.     uint8_t *scantable;

  582.     int final, run, level;

  583.     int ac_mode, dc_mode, est_run, dc_level;

  584.     int pos, n;

  585.     int zeros_only;

  586.     int use_quant_matrix;

  587.     int sign;

  588. 

  589.     assert(w->orient < 12);

  590.     w->bdsp.clear_block(s->block[0]);

  591. 

  592.     if (chroma)

  593.         dc_mode = 2;

  594.     else

  595.         dc_mode = !!w->est_run; // 0, 1

  596. 

  597.     if (x8_get_dc_rlf(w, dc_mode, &dc_level, &final))

  598.         return -1;

  599.     n          = 0;

  600.     zeros_only = 0;

  601.     if (!final) { // decode ac

  602.         use_quant_matrix = w->use_quant_matrix;

  603.         if (chroma) {

  604.             ac_mode = 1;

  605.             est_run = 64; // not used

  606.         } else {

  607.             if (w->raw_orient < 3)

  608.                 use_quant_matrix = 0;

  609. 

  610.             if (w->raw_orient > 4) {

  611.                 ac_mode = 0;

  612.                 est_run = 64;

  613.             } else {

  614.                 if (w->est_run > 1) {

  615.                     ac_mode = 2;

  616.                     est_run = w->est_run;

  617.                 } else {

  618.                     ac_mode = 3;

  619.                     est_run = 64;

  620.                 }

  621.             }

  622.         }

  623.         x8_select_ac_table(w, ac_mode);

  624.         /* scantable_selector[12] = { 0, 2, 0, 1, 1, 1, 0, 2, 2, 0, 1, 2 }; <-

  625.          * -> 10'01' 00'10' 10'00' 01'01' 01'00' 10'00 => 0x928548 */

  626.         scantable = w->scantable[(0x928548 >> (2 * w->orient)) & 3].permutated;

  627.         pos       = 0;

  628.         do {

  629.             n++;

  630.             if (n >= est_run) {

  631.                 ac_mode = 3;

  632.                 x8_select_ac_table(w, 3);

  633.             }

  634. 

  635.             x8_get_ac_rlf(w, ac_mode, &run, &level, &final);

  636. 

  637.             pos += run + 1;

  638.             if (pos > 63) {

  639.                 // this also handles vlc error in x8_get_ac_rlf

  640.                 return -1;

  641.             }

  642.             level  = (level + 1) * w->dquant;

  643.             level += w->qsum;

  644. 

  645.             sign  = -get_bits1(w->gb);

  646.             level = (level ^ sign) - sign;

  647. 

  648.             if (use_quant_matrix)

  649.                 level = (level * quant_table[pos]) >> 8;

  650. 

  651.             s->block[0][scantable[pos]] = level;

  652.         } while (!final);

  653. 

  654.         s->block_last_index[0] = pos;

  655.     } else { // DC only

  656.         s->block_last_index[0] = 0;

  657.         if (w->flat_dc && ((unsigned) (dc_level + 1)) < 3) { // [-1; 1]

  658.             int32_t divide_quant = !chroma ? w->divide_quant_dc_luma

  659.                                            : w->divide_quant_dc_chroma;

  660.             int32_t dc_quant     = !chroma ? w->quant

  661.                                            : w->quant_dc_chroma;

  662. 

  663.             // original intent dc_level += predicted_dc/quant;

  664.             // but it got lost somewhere in the rounding

  665.             dc_level += (w->predicted_dc * divide_quant + (1 << 12)) >> 13;

  666. 

  667.             dsp_x8_put_solidcolor(av_clip_uint8((dc_level * dc_quant + 4) >> 3),

  668.                                   w->dest[chroma],

  669.                                   w->frame->linesize[!!chroma]);

  670. 

  671.             goto block_placed;

  672.         }

  673.         zeros_only = (dc_level == 0);

  674.     }

  675.     if (!chroma)

  676.         s->block[0][0] = dc_level * w->quant;

  677.     else

  678.         s->block[0][0] = dc_level * w->quant_dc_chroma;

  679. 

  680.     // there is !zero_only check in the original, but dc_level check is enough

  681.     if ((unsigned int) (dc_level + 1) >= 3 && (w->edges & 3) != 3) {

  682.         int direction;

  683.         /* ac_comp_direction[orient] = { 0, 3, 3, 1, 1, 0, 0, 0, 2, 2, 2, 1 }; <-

  684.          * -> 01'10' 10'10' 00'00' 00'01' 01'11' 11'00 => 0x6A017C */

  685.         direction = (0x6A017C >> (w->orient * 2)) & 3;

  686.         if (direction != 3) {

  687.             // modify block_last[]

  688.             x8_ac_compensation(w, direction, s->block[0][0]);

  689.         }

  690.     }

  691. 

  692.     if (w->flat_dc) {

  693.         dsp_x8_put_solidcolor(w->predicted_dc, w->dest[chroma],

  694.                               w->frame->linesize[!!chroma]);

  695.     } else {

  696.         w->dsp.spatial_compensation[w->orient](w->scratchpad,

  697.                                                w->dest[chroma],

  698.                                                w->frame->linesize[!!chroma]);

  699.     }

  700.     if (!zeros_only)

  701.         w->idsp.idct_add(w->dest[chroma],

  702.                          w->frame->linesize[!!chroma],

  703.                          s->block[0]);

  704. 

  705. block_placed:

  706.     if (!chroma)

  707.         x8_update_predictions(w, w->orient, n);

  708. 

  709.     if (w->loopfilter) {

  710.         uint8_t *ptr = w->dest[chroma];

  711.         int linesize = w->frame->linesize[!!chroma];

  712. 

  713.         if (!((w->edges & 2) || (zeros_only && (w->orient | 4) == 4)))

  714.             w->dsp.h_loop_filter(ptr, linesize, w->quant);

  715. 

  716.         if (!((w->edges & 1) || (zeros_only && (w->orient | 8) == 8)))

  717.             w->dsp.v_loop_filter(ptr, linesize, w->quant);

  718.     }

  719.     return 0;

  720. }

  721. 

  722. // FIXME maybe merge with ff_*

  723. static void x8_init_block_index(IntraX8Context *w, AVFrame *frame, int mb_y)

  724. {

  725.     // not parent codec linesize as this would be wrong for field pics

  726.     // not that IntraX8 has interlacing support ;)

  727.     const int linesize   = frame->linesize[0];

  728.     const int uvlinesize = frame->linesize[1];

  729. 

  730.     w->dest[0] = frame->data[0];

  731.     w->dest[1] = frame->data[1];

  732.     w->dest[2] = frame->data[2];

  733. 

  734.     w->dest[0] +=  mb_y         * linesize   << 3;

  735.     // chroma blocks are on add rows

  736.     w->dest[1] += (mb_y & (~1)) * uvlinesize << 2;

  737.     w->dest[2] += (mb_y & (~1)) * uvlinesize << 2;

  738. }

  739. 

  740. av_cold int ff_intrax8_common_init(AVCodecContext *avctx,

  741.                                    IntraX8Context *w, IDCTDSPContext *idsp,

  742.                                    MpegEncContext *const s)

  743. {

  744.     int ret = x8_vlc_init();

  745.     if (ret < 0)

  746.         return ret;

  747. 

  748.     w->avctx = avctx;

  749.     w->idsp = *idsp;

  750.     w->s = s;

  751. 

  752.     // two rows, 2 blocks per cannon mb

  753.     w->prediction_table = av_mallocz(s->mb_width * 2 * 2);

  754.     if (!w->prediction_table)

  755.         return AVERROR(ENOMEM);

  756. 

  757.     ff_init_scantable(w->idsp.idct_permutation, &w->scantable[0],

  758.                       ff_wmv1_scantable[0]);

  759.     ff_init_scantable(w->idsp.idct_permutation, &w->scantable[1],

  760.                       ff_wmv1_scantable[2]);

  761.     ff_init_scantable(w->idsp.idct_permutation, &w->scantable[2],

  762.                       ff_wmv1_scantable[3]);

  763. 

  764.     ff_intrax8dsp_init(&w->dsp);

  765.     ff_blockdsp_init(&w->bdsp, avctx);

  766. 

  767.     return 0;

  768. }

  769. 

  770. av_cold void ff_intrax8_common_end(IntraX8Context *w)

  771. {

  772.     av_freep(&w->prediction_table);

  773. }

  774. 

  775. int ff_intrax8_decode_picture(IntraX8Context *const w, Picture *pict,

  776.                               GetBitContext *gb,

  777.                               int dquant, int quant_offset, int loopfilter)

  778. {

  779.     MpegEncContext *const s = w->s;

  780.     int mb_xy;

  781.     assert(s);

  782. 

  783.     w->gb     = gb;

  784.     w->dquant = dquant;

  785.     w->quant  = dquant >> 1;

  786.     w->qsum   = quant_offset;

  787.     w->frame  = pict->f;

  788.     w->loopfilter = loopfilter;

  789.     w->use_quant_matrix = get_bits1(w->gb);

  790. 

  791.     w->divide_quant_dc_luma = ((1 << 16) + (w->quant >> 1)) / w->quant;

  792.     if (w->quant < 5) {

  793.         w->quant_dc_chroma        = w->quant;

  794.         w->divide_quant_dc_chroma = w->divide_quant_dc_luma;

  795.     } else {

  796.         w->quant_dc_chroma        = w->quant + ((w->quant + 3) >> 3);

  797.         w->divide_quant_dc_chroma = ((1 << 16) + (w->quant_dc_chroma >> 1)) / w->quant_dc_chroma;

  798.     }

  799.     x8_reset_vlc_tables(w);

  800. 

  801.     for (s->mb_y = 0; s->mb_y < s->mb_height * 2; s->mb_y++) {

  802.         x8_init_block_index(w, w->frame, s->mb_y);

  803.         mb_xy = (s->mb_y >> 1) * s->mb_stride;

  804. 

  805.         for (s->mb_x = 0; s->mb_x < s->mb_width * 2; s->mb_x++) {

  806.             x8_get_prediction(w);

  807.             if (x8_setup_spatial_predictor(w, 0))

  808.                 goto error;

  809.             if (x8_decode_intra_mb(w, 0))

  810.                 goto error;

  811. 

  812.             if (s->mb_x & s->mb_y & 1) {

  813.                 x8_get_prediction_chroma(w);

  814. 

  815.                 /* when setting up chroma, no vlc is read,

  816.                  * so no error condition can be reached */

  817.                 x8_setup_spatial_predictor(w, 1);

  818.                 if (x8_decode_intra_mb(w, 1))

  819.                     goto error;

  820. 

  821.                 x8_setup_spatial_predictor(w, 2);

  822.                 if (x8_decode_intra_mb(w, 2))

  823.                     goto error;

  824. 

  825.                 w->dest[1] += 8;

  826.                 w->dest[2] += 8;

  827. 

  828.                 /* emulate MB info in the relevant tables */

  829.                 s->mbskip_table[mb_xy]                 = 0;

  830.                 s->mbintra_table[mb_xy]                = 1;

  831.                 pict->qscale_table[mb_xy] = w->quant;

  832.                 mb_xy++;

  833.             }

  834.             w->dest[0] += 8;

  835.         }

  836.         if (s->mb_y & 1)

  837.             ff_mpeg_draw_horiz_band(s, (s->mb_y - 1) * 8, 16);

  838.     }

  839. 

  840. error:

  841.     return 0;

  842. }