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.
42979 Commits
32 Branches
411MB
Tree: 8bc4accc37
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '8bc4accc37'
${ noResults }
FFmpeg/libavcodec/intrax8.c

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

  136.     int table_index;

  137. 

  138.     assert(mode < 4);

  139. 

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

  141.         return;

  142. 

  143.     table_index       = get_bits(&s->gb, 3);

  144.     // 2 modes use same tables

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

  146. 

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

  148. }

  149. 

  150. static inline int x8_get_orient_vlc(IntraX8Context *w)

  151. {

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

  153. 

  154.     if (!w->j_orient_vlc) {

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

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

  157.     }

  158.     assert(w->j_orient_vlc);

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

  160. 

  161.     return get_vlc2(&s->gb, w->j_orient_vlc->table, OR_VLC_BITS, OR_VLC_MTD);

  162. }

  163. 

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

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

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

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

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

  169. static const uint32_t ac_decode_table[] = {

  170.     /* 46 */ extra_bits(3) | extra_run   | run_offset(16) | level_offset(0),

  171.     /* 47 */ extra_bits(3) | extra_run   | run_offset(24) | level_offset(0),

  172.     /* 48 */ extra_bits(2) | extra_run   | run_offset(4)  | level_offset(1),

  173.     /* 49 */ extra_bits(3) | extra_run   | run_offset(8)  | level_offset(1),

  174. 

  175.     /* 50 */ extra_bits(5) | extra_run   | run_offset(32) | level_offset(0),

  176.     /* 51 */ extra_bits(4) | extra_run   | run_offset(16) | level_offset(1),

  177. 

  178.     /* 52 */ extra_bits(2) | extra_level | run_offset(0)  | level_offset(4),

  179.     /* 53 */ extra_bits(2) | extra_level | run_offset(0)  | level_offset(8),

  180.     /* 54 */ extra_bits(2) | extra_level | run_offset(0)  | level_offset(12),

  181.     /* 55 */ extra_bits(3) | extra_level | run_offset(0)  | level_offset(16),

  182.     /* 56 */ extra_bits(3) | extra_level | run_offset(0)  | level_offset(24),

  183. 

  184.     /* 57 */ extra_bits(2) | extra_level | run_offset(1)  | level_offset(3),

  185.     /* 58 */ extra_bits(3) | extra_level | run_offset(1)  | level_offset(7),

  186. 

  187.     /* 59 */ extra_bits(2) | extra_run   | run_offset(16) | level_offset(0),

  188.     /* 60 */ extra_bits(2) | extra_run   | run_offset(20) | level_offset(0),

  189.     /* 61 */ extra_bits(2) | extra_run   | run_offset(24) | level_offset(0),

  190.     /* 62 */ extra_bits(2) | extra_run   | run_offset(28) | level_offset(0),

  191.     /* 63 */ extra_bits(4) | extra_run   | run_offset(32) | level_offset(0),

  192.     /* 64 */ extra_bits(4) | extra_run   | run_offset(48) | level_offset(0),

  193. 

  194.     /* 65 */ extra_bits(2) | extra_run   | run_offset(4)  | level_offset(1),

  195.     /* 66 */ extra_bits(3) | extra_run   | run_offset(8)  | level_offset(1),

  196.     /* 67 */ extra_bits(4) | extra_run   | run_offset(16) | level_offset(1),

  197. 

  198.     /* 68 */ extra_bits(2) | extra_level | run_offset(0)  | level_offset(4),

  199.     /* 69 */ extra_bits(3) | extra_level | run_offset(0)  | level_offset(8),

  200.     /* 70 */ extra_bits(4) | extra_level | run_offset(0)  | level_offset(16),

  201. 

  202.     /* 71 */ extra_bits(2) | extra_level | run_offset(1)  | level_offset(3),

  203.     /* 72 */ extra_bits(3) | extra_level | run_offset(1)  | level_offset(7),

  204. };

  205. #undef extra_bits

  206. #undef extra_run

  207. #undef extra_level

  208. #undef run_offset

  209. #undef level_offset

  210. 

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

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

  213. {

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

  215.     int i, e;

  216. 

  217. //    x8_select_ac_table(w, mode);

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

  219. 

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

  221.         int t, l;

  222.         if (i < 0) {

  223.             (*level) =

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

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

  226.             return;

  227.         }

  228. 

  229.         /*

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

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

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

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

  234.          */

  235. 

  236.         (*final) =

  237.         t        = (i > 22);

  238.         i       -= 23 * t;

  239. 

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

  241.          *     11 10'01 01'00 00'00 00'00 00'00 00 => 0xE50000 */

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

  243. 

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

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

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

  247. 

  248.         (*run)   = i & t;

  249.         (*level) = l;

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

  251.         uint32_t sm;

  252.         uint32_t mask;

  253. 

  254.         i -= 46;

  255.         sm = ac_decode_table[i];

  256. 

  257.         e    = get_bits(&s->gb, sm & 0xF);

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

  259.         mask = sm & 0xff;

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

  261. 

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

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

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

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

  266.         static const uint8_t crazy_mix_runlevel[32] = {

  267.             0x22, 0x32, 0x33, 0x53, 0x23, 0x42, 0x43, 0x63,

  268.             0x24, 0x52, 0x34, 0x73, 0x25, 0x62, 0x44, 0x83,

  269.             0x26, 0x72, 0x35, 0x54, 0x27, 0x82, 0x45, 0x64,

  270.             0x28, 0x92, 0x36, 0x74, 0x29, 0xa2, 0x46, 0x84,

  271.         };

  272. 

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

  274.         e        = get_bits(&s->gb, 5); // get the extra bits

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

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

  277.     } else {

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

  279.         (*run)   = get_bits(&s->gb, 6);

  280.         (*final) = get_bits1(&s->gb);

  281.     }

  282.     return;

  283. }

  284. 

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

  286.  *     0, 1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7,

  287.  * }; */

  288. static const uint8_t dc_index_offset[] = {

  289.     0, 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,

  290. };

  291. 

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

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

  294. {

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

  296.     int i, e, c;

  297. 

  298.     assert(mode < 3);

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

  300.         int table_index = get_bits(&s->gb, 3);

  301.         // 4 modes, same table

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

  303.     }

  304.     assert(w->j_dc_vlc);

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

  306. 

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

  308. 

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

  310.     c        = i > 16;

  311.     (*final) = c;

  312.     i       -= 17 * c;

  313. 

  314.     if (i <= 0) {

  315.         (*level) = 0;

  316.         return -i;

  317.     }

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

  319.     c -= c > 1;

  320. 

  321.     e = get_bits(&s->gb, c); // get the extra bits

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

  323. 

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

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

  326.     return 0;

  327. }

  328. 

  329. // end of huffman

  330. 

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

  332. {

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

  334.     int range;

  335.     int sum;

  336.     int quant;

  337. 

  338.     w->dsp.setup_spatial_compensation(s->dest[chroma], s->sc.edge_emu_buffer,

  339.                                       s->current_picture.f->linesize[chroma > 0],

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

  341.     if (chroma) {

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

  343.         quant     = w->quant_dc_chroma;

  344.     } else {

  345.         quant = w->quant;

  346.     }

  347. 

  348.     w->flat_dc = 0;

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

  350.         w->orient = 0;

  351. 

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

  353.         if (range < 3) {

  354.             w->flat_dc      = 1;

  355.             sum            += 9;

  356.             // ((1 << 17) + 9) / (8 + 8 + 1 + 2) = 6899

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

  358.         }

  359.     }

  360.     if (chroma)

  361.         return 0;

  362. 

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

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

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

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

  367.                 w->orient = 11;

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

  369.                 w->orient = 10;

  370.         } else {

  371.             w->orient = 0;

  372.         }

  373.         w->raw_orient = 0;

  374.     } else {

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

  376.             { 0, 8, 4, 10, 11, 2, 6, 9, 1, 3, 5, 7 },

  377.             { 4, 0, 8, 11, 10, 3, 5, 2, 6, 9, 1, 7 },

  378.             { 8, 0, 4, 10, 11, 1, 7, 2, 6, 9, 3, 5 },

  379.         };

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

  381.         if (w->raw_orient < 0)

  382.             return -1;

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

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

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

  386.     }

  387.     return 0;

  388. }

  389. 

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

  391.                                   const int est_run)

  392. {

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

  394. 

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

  396. /*

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

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

  399.  */

  400. }

  401. 

  402. static void x8_get_prediction_chroma(IntraX8Context *const w)

  403. {

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

  405. 

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

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

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

  409. 

  410.     w->raw_orient = 0;

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

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

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

  414.         return;

  415.     }

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

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

  418. }

  419. 

  420. static void x8_get_prediction(IntraX8Context *const w)

  421. {

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

  423.     int a, b, c, i;

  424. 

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

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

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

  428. 

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

  430.     case 0:

  431.         break;

  432.     case 1:

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

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

  435.         w->orient  = 1;

  436.         return;

  437.     case 2:

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

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

  440.         w->orient  = 2;

  441.         return;

  442.     case 3:

  443.         w->est_run = 16;

  444.         w->orient  = 0;

  445.         return;

  446.     }

  447.     // no edge cases

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

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

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

  451. 

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

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

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

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

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

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

  458.     w->est_run >>= 2;

  459. 

  460.     a &= 3;

  461.     b &= 3;

  462.     c &= 3;

  463. 

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

  465.     if (i != 3)

  466.         w->orient = i;

  467.     else

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

  469. /*

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

  471.  * ->{ 0, 1, 0, pad },

  472.  *   { 0, 1, X, pad },

  473.  *   { 2, 2, 2, pad }

  474.  * }

  475.  * pad 2  2  2;

  476.  * pad X  1  0;

  477.  * pad 0  1  0 <-

  478.  * -> 11 10 '10 10 '11 11'01 00 '11 00'01 00 => 0xEAF4C4

  479.  *

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

  481.  * ->{ 0, 2, 1, pad},

  482.  *   { 2, 2, 2, pad}

  483.  * }

  484.  * pad 2  2  2;

  485.  * pad 1  2  0 <-

  486.  * -> 11 10'10 10 '11 01'10 00 => 0xEAD8

  487.  */

  488. }

  489. 

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

  491.                                const int dc_level)

  492. {

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

  494.     int t;

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

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

  497.     switch (direction) {

  498.     case 0:

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

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

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

  502. 

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

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

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

  506. 

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

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

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

  510. 

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

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

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

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

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

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

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

  518. 

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

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

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

  522. 

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

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

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

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

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

  528. 

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

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

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

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

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

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

  535. 

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

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

  538. 

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

  540.         break;

  541.     case 1:

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

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

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

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

  546. 

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

  548.         break;

  549.     case 2:

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

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

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

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

  554. 

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

  556.         break;

  557.     }

  558. #undef B

  559. #undef T

  560. }

  561. 

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

  563.                                   const int linesize)

  564. {

  565.     int k;

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

  567.         memset(dst, pix, 8);

  568.         dst += linesize;

  569.     }

  570. }

  571. 

  572. static const int16_t quant_table[64] = {

  573.     256, 256, 256, 256, 256, 256, 259, 262,

  574.     265, 269, 272, 275, 278, 282, 285, 288,

  575.     292, 295, 299, 303, 306, 310, 314, 317,

  576.     321, 325, 329, 333, 337, 341, 345, 349,

  577.     353, 358, 362, 366, 371, 375, 379, 384,

  578.     389, 393, 398, 403, 408, 413, 417, 422,

  579.     428, 433, 438, 443, 448, 454, 459, 465,

  580.     470, 476, 482, 488, 493, 499, 505, 511,

  581. };

  582. 

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

  584. {

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

  586. 

  587.     uint8_t *scantable;

  588.     int final, run, level;

  589.     int ac_mode, dc_mode, est_run, dc_level;

  590.     int pos, n;

  591.     int zeros_only;

  592.     int use_quant_matrix;

  593.     int sign;

  594. 

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

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

  597. 

  598.     if (chroma)

  599.         dc_mode = 2;

  600.     else

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

  602. 

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

  604.         return -1;

  605.     n          = 0;

  606.     zeros_only = 0;

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

  608.         use_quant_matrix = w->use_quant_matrix;

  609.         if (chroma) {

  610.             ac_mode = 1;

  611.             est_run = 64; // not used

  612.         } else {

  613.             if (w->raw_orient < 3)

  614.                 use_quant_matrix = 0;

  615. 

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

  617.                 ac_mode = 0;

  618.                 est_run = 64;

  619.             } else {

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

  621.                     ac_mode = 2;

  622.                     est_run = w->est_run;

  623.                 } else {

  624.                     ac_mode = 3;

  625.                     est_run = 64;

  626.                 }

  627.             }

  628.         }

  629.         x8_select_ac_table(w, ac_mode);

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

  631.          * -> 10'01' 00'10' 10'00' 01'01' 01'00' 10'00 => 0x928548 */

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

  633.         pos       = 0;

  634.         do {

  635.             n++;

  636.             if (n >= est_run) {

  637.                 ac_mode = 3;

  638.                 x8_select_ac_table(w, 3);

  639.             }

  640. 

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

  642. 

  643.             pos += run + 1;

  644.             if (pos > 63) {

  645.                 // this also handles vlc error in x8_get_ac_rlf

  646.                 return -1;

  647.             }

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

  649.             level += w->qsum;

  650. 

  651.             sign  = -get_bits1(&s->gb);

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

  653. 

  654.             if (use_quant_matrix)

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

  656. 

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

  658.         } while (!final);

  659. 

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

  661.     } else { // DC only

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

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

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

  665.                                            : w->divide_quant_dc_chroma;

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

  667.                                            : w->quant_dc_chroma;

  668. 

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

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

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

  672. 

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

  674.                                   s->dest[chroma],

  675.                                   s->current_picture.f->linesize[!!chroma]);

  676. 

  677.             goto block_placed;

  678.         }

  679.         zeros_only = (dc_level == 0);

  680.     }

  681.     if (!chroma)

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

  683.     else

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

  685. 

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

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

  688.         int direction;

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

  690.          * -> 01'10' 10'10' 00'00' 00'01' 01'11' 11'00 => 0x6A017C */

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

  692.         if (direction != 3) {

  693.             // modify block_last[]

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

  695.         }

  696.     }

  697. 

  698.     if (w->flat_dc) {

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

  700.                               s->current_picture.f->linesize[!!chroma]);

  701.     } else {

  702.         w->dsp.spatial_compensation[w->orient](s->sc.edge_emu_buffer,

  703.                                                s->dest[chroma],

  704.                                                s->current_picture.f->linesize[!!chroma]);

  705.     }

  706.     if (!zeros_only)

  707.         s->idsp.idct_add(s->dest[chroma],

  708.                          s->current_picture.f->linesize[!!chroma],

  709.                          s->block[0]);

  710. 

  711. block_placed:

  712.     if (!chroma)

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

  714. 

  715.     if (s->loop_filter) {

  716.         uint8_t *ptr = s->dest[chroma];

  717.         int linesize = s->current_picture.f->linesize[!!chroma];

  718. 

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

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

  721. 

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

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

  724.     }

  725.     return 0;

  726. }

  727. 

  728. // FIXME maybe merge with ff_*

  729. static void x8_init_block_index(MpegEncContext *s)

  730. {

  731.     // not s->linesize as this would be wrong for field pics

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

  733.     const int linesize   = s->current_picture.f->linesize[0];

  734.     const int uvlinesize = s->current_picture.f->linesize[1];

  735. 

  736.     s->dest[0] = s->current_picture.f->data[0];

  737.     s->dest[1] = s->current_picture.f->data[1];

  738.     s->dest[2] = s->current_picture.f->data[2];

  739. 

  740.     s->dest[0] +=  s->mb_y         * linesize   << 3;

  741.     // chroma blocks are on add rows

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

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

  744. }

  745. 

  746. av_cold int ff_intrax8_common_init(IntraX8Context *w, MpegEncContext *const s)

  747. {

  748.     int ret = x8_vlc_init();

  749.     if (ret < 0)

  750.         return ret;

  751. 

  752.     w->s = s;

  753. 

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

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

  756.     if (!w->prediction_table)

  757.         return AVERROR(ENOMEM);

  758. 

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

  760.                       ff_wmv1_scantable[0]);

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

  762.                       ff_wmv1_scantable[2]);

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

  764.                       ff_wmv1_scantable[3]);

  765. 

  766.     ff_intrax8dsp_init(&w->dsp);

  767. 

  768.     return 0;

  769. }

  770. 

  771. av_cold void ff_intrax8_common_end(IntraX8Context *w)

  772. {

  773.     av_freep(&w->prediction_table);

  774. }

  775. 

  776. int ff_intrax8_decode_picture(IntraX8Context *const w, int dquant,

  777.                               int quant_offset)

  778. {

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

  780.     int mb_xy;

  781.     assert(s);

  782.     w->use_quant_matrix = get_bits1(&s->gb);

  783. 

  784.     w->dquant = dquant;

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

  786.     w->qsum   = quant_offset;

  787. 

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

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

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

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

  792.     } else {

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

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

  795.     }

  796.     x8_reset_vlc_tables(w);

  797. 

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

  799.         x8_init_block_index(s);

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

  801. 

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

  803.             x8_get_prediction(w);

  804.             if (x8_setup_spatial_predictor(w, 0))

  805.                 goto error;

  806.             if (x8_decode_intra_mb(w, 0))

  807.                 goto error;

  808. 

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

  810.                 x8_get_prediction_chroma(w);

  811. 

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

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

  814.                 x8_setup_spatial_predictor(w, 1);

  815.                 if (x8_decode_intra_mb(w, 1))

  816.                     goto error;

  817. 

  818.                 x8_setup_spatial_predictor(w, 2);

  819.                 if (x8_decode_intra_mb(w, 2))

  820.                     goto error;

  821. 

  822.                 s->dest[1] += 8;

  823.                 s->dest[2] += 8;

  824. 

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

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

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

  828.                 s->current_picture.qscale_table[mb_xy] = w->quant;

  829.                 mb_xy++;

  830.             }

  831.             s->dest[0] += 8;

  832.         }

  833.         if (s->mb_y & 1)

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

  835.     }

  836. 

  837. error:

  838.     return 0;

  839. }