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.
13510 Commits
32 Branches
340MB
Tree: 9f95bfe2b9
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '9f95bfe2b9'
${ noResults }
FFmpeg/libavcodec/mpegaudiodec.c

2696 lines
80KB
Raw Blame History 

  1. /*

  2.  * MPEG Audio decoder

  3.  * Copyright (c) 2001, 2002 Fabrice Bellard.

  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 mpegaudiodec.c

  24.  * MPEG Audio decoder.

  25.  */

  26. 

  27. //#define DEBUG

  28. #include "avcodec.h"

  29. #include "bitstream.h"

  30. #include "dsputil.h"

  31. 

  32. /*

  33.  * TODO:

  34.  *  - in low precision mode, use more 16 bit multiplies in synth filter

  35.  *  - test lsf / mpeg25 extensively.

  36.  */

  37. 

  38. /* define USE_HIGHPRECISION to have a bit exact (but slower) mpeg

  39.    audio decoder */

  40. #ifdef CONFIG_MPEGAUDIO_HP

  41. #   define USE_HIGHPRECISION

  42. #endif

  43. 

  44. #include "mpegaudio.h"

  45. #include "mpegaudiodecheader.h"

  46. 

  47. #include "mathops.h"

  48. 

  49. /* WARNING: only correct for posititive numbers */

  50. #define FIXR(a)   ((int)((a) * FRAC_ONE + 0.5))

  51. #define FRAC_RND(a) (((a) + (FRAC_ONE/2)) >> FRAC_BITS)

  52. 

  53. #define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))

  54. 

  55. /****************/

  56. 

  57. #define HEADER_SIZE 4

  58. 

  59. /**

  60.  * Context for MP3On4 decoder

  61.  */

  62. typedef struct MP3On4DecodeContext {

  63.     int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)

  64.     int chan_cfg; ///< channel config number

  65.     int syncword; ///< syncword patch

  66.     MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance

  67. } MP3On4DecodeContext;

  68. 

  69. /* layer 3 "granule" */

  70. typedef struct GranuleDef {

  71.     uint8_t scfsi;

  72.     int part2_3_length;

  73.     int big_values;

  74.     int global_gain;

  75.     int scalefac_compress;

  76.     uint8_t block_type;

  77.     uint8_t switch_point;

  78.     int table_select[3];

  79.     int subblock_gain[3];

  80.     uint8_t scalefac_scale;

  81.     uint8_t count1table_select;

  82.     int region_size[3]; /* number of huffman codes in each region */

  83.     int preflag;

  84.     int short_start, long_end; /* long/short band indexes */

  85.     uint8_t scale_factors[40];

  86.     int32_t sb_hybrid[SBLIMIT * 18]; /* 576 samples */

  87. } GranuleDef;

  88. 

  89. #include "mpegaudiodata.h"

  90. #include "mpegaudiodectab.h"

  91. 

  92. static void compute_antialias_integer(MPADecodeContext *s, GranuleDef *g);

  93. static void compute_antialias_float(MPADecodeContext *s, GranuleDef *g);

  94. 

  95. /* vlc structure for decoding layer 3 huffman tables */

  96. static VLC huff_vlc[16];

  97. static VLC huff_quad_vlc[2];

  98. /* computed from band_size_long */

  99. static uint16_t band_index_long[9][23];

  100. /* XXX: free when all decoders are closed */

  101. #define TABLE_4_3_SIZE (8191 + 16)*4

  102. static int8_t  table_4_3_exp[TABLE_4_3_SIZE];

  103. static uint32_t table_4_3_value[TABLE_4_3_SIZE];

  104. static uint32_t exp_table[512];

  105. static uint32_t expval_table[512][16];

  106. /* intensity stereo coef table */

  107. static int32_t is_table[2][16];

  108. static int32_t is_table_lsf[2][2][16];

  109. static int32_t csa_table[8][4];

  110. static float csa_table_float[8][4];

  111. static int32_t mdct_win[8][36];

  112. 

  113. /* lower 2 bits: modulo 3, higher bits: shift */

  114. static uint16_t scale_factor_modshift[64];

  115. /* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */

  116. static int32_t scale_factor_mult[15][3];

  117. /* mult table for layer 2 group quantization */

  118. 

  119. #define SCALE_GEN(v) \

  120. { FIXR(1.0 * (v)), FIXR(0.7937005259 * (v)), FIXR(0.6299605249 * (v)) }

  121. 

  122. static const int32_t scale_factor_mult2[3][3] = {

  123.     SCALE_GEN(4.0 / 3.0), /* 3 steps */

  124.     SCALE_GEN(4.0 / 5.0), /* 5 steps */

  125.     SCALE_GEN(4.0 / 9.0), /* 9 steps */

  126. };

  127. 

  128. static DECLARE_ALIGNED_16(MPA_INT, window[512]);

  129. 

  130. /**

  131.  * Convert region offsets to region sizes and truncate

  132.  * size to big_values.

  133.  */

  134. void ff_region_offset2size(GranuleDef *g){

  135.     int i, k, j=0;

  136.     g->region_size[2] = (576 / 2);

  137.     for(i=0;i<3;i++) {

  138.         k = FFMIN(g->region_size[i], g->big_values);

  139.         g->region_size[i] = k - j;

  140.         j = k;

  141.     }

  142. }

  143. 

  144. void ff_init_short_region(MPADecodeContext *s, GranuleDef *g){

  145.     if (g->block_type == 2)

  146.         g->region_size[0] = (36 / 2);

  147.     else {

  148.         if (s->sample_rate_index <= 2)

  149.             g->region_size[0] = (36 / 2);

  150.         else if (s->sample_rate_index != 8)

  151.             g->region_size[0] = (54 / 2);

  152.         else

  153.             g->region_size[0] = (108 / 2);

  154.     }

  155.     g->region_size[1] = (576 / 2);

  156. }

  157. 

  158. void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2){

  159.     int l;

  160.     g->region_size[0] =

  161.         band_index_long[s->sample_rate_index][ra1 + 1] >> 1;

  162.     /* should not overflow */

  163.     l = FFMIN(ra1 + ra2 + 2, 22);

  164.     g->region_size[1] =

  165.         band_index_long[s->sample_rate_index][l] >> 1;

  166. }

  167. 

  168. void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g){

  169.     if (g->block_type == 2) {

  170.         if (g->switch_point) {

  171.             /* if switched mode, we handle the 36 first samples as

  172.                 long blocks.  For 8000Hz, we handle the 48 first

  173.                 exponents as long blocks (XXX: check this!) */

  174.             if (s->sample_rate_index <= 2)

  175.                 g->long_end = 8;

  176.             else if (s->sample_rate_index != 8)

  177.                 g->long_end = 6;

  178.             else

  179.                 g->long_end = 4; /* 8000 Hz */

  180. 

  181.             g->short_start = 2 + (s->sample_rate_index != 8);

  182.         } else {

  183.             g->long_end = 0;

  184.             g->short_start = 0;

  185.         }

  186.     } else {

  187.         g->short_start = 13;

  188.         g->long_end = 22;

  189.     }

  190. }

  191. 

  192. /* layer 1 unscaling */

  193. /* n = number of bits of the mantissa minus 1 */

  194. static inline int l1_unscale(int n, int mant, int scale_factor)

  195. {

  196.     int shift, mod;

  197.     int64_t val;

  198. 

  199.     shift = scale_factor_modshift[scale_factor];

  200.     mod = shift & 3;

  201.     shift >>= 2;

  202.     val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);

  203.     shift += n;

  204.     /* NOTE: at this point, 1 <= shift >= 21 + 15 */

  205.     return (int)((val + (1LL << (shift - 1))) >> shift);

  206. }

  207. 

  208. static inline int l2_unscale_group(int steps, int mant, int scale_factor)

  209. {

  210.     int shift, mod, val;

  211. 

  212.     shift = scale_factor_modshift[scale_factor];

  213.     mod = shift & 3;

  214.     shift >>= 2;

  215. 

  216.     val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];

  217.     /* NOTE: at this point, 0 <= shift <= 21 */

  218.     if (shift > 0)

  219.         val = (val + (1 << (shift - 1))) >> shift;

  220.     return val;

  221. }

  222. 

  223. /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */

  224. static inline int l3_unscale(int value, int exponent)

  225. {

  226.     unsigned int m;

  227.     int e;

  228. 

  229.     e = table_4_3_exp  [4*value + (exponent&3)];

  230.     m = table_4_3_value[4*value + (exponent&3)];

  231.     e -= (exponent >> 2);

  232.     assert(e>=1);

  233.     if (e > 31)

  234.         return 0;

  235.     m = (m + (1 << (e-1))) >> e;

  236. 

  237.     return m;

  238. }

  239. 

  240. /* all integer n^(4/3) computation code */

  241. #define DEV_ORDER 13

  242. 

  243. #define POW_FRAC_BITS 24

  244. #define POW_FRAC_ONE    (1 << POW_FRAC_BITS)

  245. #define POW_FIX(a)   ((int)((a) * POW_FRAC_ONE))

  246. #define POW_MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> POW_FRAC_BITS)

  247. 

  248. static int dev_4_3_coefs[DEV_ORDER];

  249. 

  250. #if 0 /* unused */

  251. static int pow_mult3[3] = {

  252.     POW_FIX(1.0),

  253.     POW_FIX(1.25992104989487316476),

  254.     POW_FIX(1.58740105196819947474),

  255. };

  256. #endif

  257. 

  258. static void int_pow_init(void)

  259. {

  260.     int i, a;

  261. 

  262.     a = POW_FIX(1.0);

  263.     for(i=0;i<DEV_ORDER;i++) {

  264.         a = POW_MULL(a, POW_FIX(4.0 / 3.0) - i * POW_FIX(1.0)) / (i + 1);

  265.         dev_4_3_coefs[i] = a;

  266.     }

  267. }

  268. 

  269. #if 0 /* unused, remove? */

  270. /* return the mantissa and the binary exponent */

  271. static int int_pow(int i, int *exp_ptr)

  272. {

  273.     int e, er, eq, j;

  274.     int a, a1;

  275. 

  276.     /* renormalize */

  277.     a = i;

  278.     e = POW_FRAC_BITS;

  279.     while (a < (1 << (POW_FRAC_BITS - 1))) {

  280.         a = a << 1;

  281.         e--;

  282.     }

  283.     a -= (1 << POW_FRAC_BITS);

  284.     a1 = 0;

  285.     for(j = DEV_ORDER - 1; j >= 0; j--)

  286.         a1 = POW_MULL(a, dev_4_3_coefs[j] + a1);

  287.     a = (1 << POW_FRAC_BITS) + a1;

  288.     /* exponent compute (exact) */

  289.     e = e * 4;

  290.     er = e % 3;

  291.     eq = e / 3;

  292.     a = POW_MULL(a, pow_mult3[er]);

  293.     while (a >= 2 * POW_FRAC_ONE) {

  294.         a = a >> 1;

  295.         eq++;

  296.     }

  297.     /* convert to float */

  298.     while (a < POW_FRAC_ONE) {

  299.         a = a << 1;

  300.         eq--;

  301.     }

  302.     /* now POW_FRAC_ONE <= a < 2 * POW_FRAC_ONE */

  303. #if POW_FRAC_BITS > FRAC_BITS

  304.     a = (a + (1 << (POW_FRAC_BITS - FRAC_BITS - 1))) >> (POW_FRAC_BITS - FRAC_BITS);

  305.     /* correct overflow */

  306.     if (a >= 2 * (1 << FRAC_BITS)) {

  307.         a = a >> 1;

  308.         eq++;

  309.     }

  310. #endif

  311.     *exp_ptr = eq;

  312.     return a;

  313. }

  314. #endif

  315. 

  316. static int decode_init(AVCodecContext * avctx)

  317. {

  318.     MPADecodeContext *s = avctx->priv_data;

  319.     static int init=0;

  320.     int i, j, k;

  321. 

  322.     s->avctx = avctx;

  323. 

  324. #if defined(USE_HIGHPRECISION) && defined(CONFIG_AUDIO_NONSHORT)

  325.     avctx->sample_fmt= SAMPLE_FMT_S32;

  326. #else

  327.     avctx->sample_fmt= SAMPLE_FMT_S16;

  328. #endif

  329.     s->error_resilience= avctx->error_resilience;

  330. 

  331.     if(avctx->antialias_algo != FF_AA_FLOAT)

  332.         s->compute_antialias= compute_antialias_integer;

  333.     else

  334.         s->compute_antialias= compute_antialias_float;

  335. 

  336.     if (!init && !avctx->parse_only) {

  337.         /* scale factors table for layer 1/2 */

  338.         for(i=0;i<64;i++) {

  339.             int shift, mod;

  340.             /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

  341.             shift = (i / 3);

  342.             mod = i % 3;

  343.             scale_factor_modshift[i] = mod | (shift << 2);

  344.         }

  345. 

  346.         /* scale factor multiply for layer 1 */

  347.         for(i=0;i<15;i++) {

  348.             int n, norm;

  349.             n = i + 2;

  350.             norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);

  351.             scale_factor_mult[i][0] = MULL(FIXR(1.0 * 2.0), norm);

  352.             scale_factor_mult[i][1] = MULL(FIXR(0.7937005259 * 2.0), norm);

  353.             scale_factor_mult[i][2] = MULL(FIXR(0.6299605249 * 2.0), norm);

  354.             dprintf(avctx, "%d: norm=%x s=%x %x %x\n",

  355.                     i, norm,

  356.                     scale_factor_mult[i][0],

  357.                     scale_factor_mult[i][1],

  358.                     scale_factor_mult[i][2]);

  359.         }

  360. 

  361.         ff_mpa_synth_init(window);

  362. 

  363.         /* huffman decode tables */

  364.         for(i=1;i<16;i++) {

  365.             const HuffTable *h = &mpa_huff_tables[i];

  366.             int xsize, x, y;

  367.             unsigned int n;

  368.             uint8_t  tmp_bits [512];

  369.             uint16_t tmp_codes[512];

  370. 

  371.             memset(tmp_bits , 0, sizeof(tmp_bits ));

  372.             memset(tmp_codes, 0, sizeof(tmp_codes));

  373. 

  374.             xsize = h->xsize;

  375.             n = xsize * xsize;

  376. 

  377.             j = 0;

  378.             for(x=0;x<xsize;x++) {

  379.                 for(y=0;y<xsize;y++){

  380.                     tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j  ];

  381.                     tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];

  382.                 }

  383.             }

  384. 

  385.             /* XXX: fail test */

  386.             init_vlc(&huff_vlc[i], 7, 512,

  387.                      tmp_bits, 1, 1, tmp_codes, 2, 2, 1);

  388.         }

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

  390.             init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,

  391.                      mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, 1);

  392.         }

  393. 

  394.         for(i=0;i<9;i++) {

  395.             k = 0;

  396.             for(j=0;j<22;j++) {

  397.                 band_index_long[i][j] = k;

  398.                 k += band_size_long[i][j];

  399.             }

  400.             band_index_long[i][22] = k;

  401.         }

  402. 

  403.         /* compute n ^ (4/3) and store it in mantissa/exp format */

  404. 

  405.         int_pow_init();

  406.         for(i=1;i<TABLE_4_3_SIZE;i++) {

  407.             double f, fm;

  408.             int e, m;

  409.             f = pow((double)(i/4), 4.0 / 3.0) * pow(2, (i&3)*0.25);

  410.             fm = frexp(f, &e);

  411.             m = (uint32_t)(fm*(1LL<<31) + 0.5);

  412.             e+= FRAC_BITS - 31 + 5 - 100;

  413. 

  414.             /* normalized to FRAC_BITS */

  415.             table_4_3_value[i] = m;

  416. //            av_log(NULL, AV_LOG_DEBUG, "%d %d %f\n", i, m, pow((double)i, 4.0 / 3.0));

  417.             table_4_3_exp[i] = -e;

  418.         }

  419.         for(i=0; i<512*16; i++){

  420.             int exponent= (i>>4);

  421.             double f= pow(i&15, 4.0 / 3.0) * pow(2, (exponent-400)*0.25 + FRAC_BITS + 5);

  422.             expval_table[exponent][i&15]= llrint(f);

  423.             if((i&15)==1)

  424.                 exp_table[exponent]= llrint(f);

  425.         }

  426. 

  427.         for(i=0;i<7;i++) {

  428.             float f;

  429.             int v;

  430.             if (i != 6) {

  431.                 f = tan((double)i * M_PI / 12.0);

  432.                 v = FIXR(f / (1.0 + f));

  433.             } else {

  434.                 v = FIXR(1.0);

  435.             }

  436.             is_table[0][i] = v;

  437.             is_table[1][6 - i] = v;

  438.         }

  439.         /* invalid values */

  440.         for(i=7;i<16;i++)

  441.             is_table[0][i] = is_table[1][i] = 0.0;

  442. 

  443.         for(i=0;i<16;i++) {

  444.             double f;

  445.             int e, k;

  446. 

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

  448.                 e = -(j + 1) * ((i + 1) >> 1);

  449.                 f = pow(2.0, e / 4.0);

  450.                 k = i & 1;

  451.                 is_table_lsf[j][k ^ 1][i] = FIXR(f);

  452.                 is_table_lsf[j][k][i] = FIXR(1.0);

  453.                 dprintf(avctx, "is_table_lsf %d %d: %x %x\n",

  454.                         i, j, is_table_lsf[j][0][i], is_table_lsf[j][1][i]);

  455.             }

  456.         }

  457. 

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

  459.             float ci, cs, ca;

  460.             ci = ci_table[i];

  461.             cs = 1.0 / sqrt(1.0 + ci * ci);

  462.             ca = cs * ci;

  463.             csa_table[i][0] = FIXHR(cs/4);

  464.             csa_table[i][1] = FIXHR(ca/4);

  465.             csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);

  466.             csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);

  467.             csa_table_float[i][0] = cs;

  468.             csa_table_float[i][1] = ca;

  469.             csa_table_float[i][2] = ca + cs;

  470.             csa_table_float[i][3] = ca - cs;

  471. //            printf("%d %d %d %d\n", FIX(cs), FIX(cs-1), FIX(ca), FIX(cs)-FIX(ca));

  472. //            av_log(NULL, AV_LOG_DEBUG,"%f %f %f %f\n", cs, ca, ca+cs, ca-cs);

  473.         }

  474. 

  475.         /* compute mdct windows */

  476.         for(i=0;i<36;i++) {

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

  478.                 double d;

  479. 

  480.                 if(j==2 && i%3 != 1)

  481.                     continue;

  482. 

  483.                 d= sin(M_PI * (i + 0.5) / 36.0);

  484.                 if(j==1){

  485.                     if     (i>=30) d= 0;

  486.                     else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);

  487.                     else if(i>=18) d= 1;

  488.                 }else if(j==3){

  489.                     if     (i<  6) d= 0;

  490.                     else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);

  491.                     else if(i< 18) d= 1;

  492.                 }

  493.                 //merge last stage of imdct into the window coefficients

  494.                 d*= 0.5 / cos(M_PI*(2*i + 19)/72);

  495. 

  496.                 if(j==2)

  497.                     mdct_win[j][i/3] = FIXHR((d / (1<<5)));

  498.                 else

  499.                     mdct_win[j][i  ] = FIXHR((d / (1<<5)));

  500. //                av_log(NULL, AV_LOG_DEBUG, "%2d %d %f\n", i,j,d / (1<<5));

  501.             }

  502.         }

  503. 

  504.         /* NOTE: we do frequency inversion adter the MDCT by changing

  505.            the sign of the right window coefs */

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

  507.             for(i=0;i<36;i+=2) {

  508.                 mdct_win[j + 4][i] = mdct_win[j][i];

  509.                 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];

  510.             }

  511.         }

  512. 

  513. #if defined(DEBUG)

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

  515.             av_log(avctx, AV_LOG_DEBUG, "win%d=\n", j);

  516.             for(i=0;i<36;i++)

  517.                 av_log(avctx, AV_LOG_DEBUG, "%f, ", (double)mdct_win[j][i] / FRAC_ONE);

  518.             av_log(avctx, AV_LOG_DEBUG, "\n");

  519.         }

  520. #endif

  521.         init = 1;

  522.     }

  523. 

  524. #ifdef DEBUG

  525.     s->frame_count = 0;

  526. #endif

  527.     if (avctx->codec_id == CODEC_ID_MP3ADU)

  528.         s->adu_mode = 1;

  529.     return 0;

  530. }

  531. 

  532. /* tab[i][j] = 1.0 / (2.0 * cos(pi*(2*k+1) / 2^(6 - j))) */

  533. 

  534. /* cos(i*pi/64) */

  535. 

  536. #define COS0_0  FIXHR(0.50060299823519630134/2)

  537. #define COS0_1  FIXHR(0.50547095989754365998/2)

  538. #define COS0_2  FIXHR(0.51544730992262454697/2)

  539. #define COS0_3  FIXHR(0.53104259108978417447/2)

  540. #define COS0_4  FIXHR(0.55310389603444452782/2)

  541. #define COS0_5  FIXHR(0.58293496820613387367/2)

  542. #define COS0_6  FIXHR(0.62250412303566481615/2)

  543. #define COS0_7  FIXHR(0.67480834145500574602/2)

  544. #define COS0_8  FIXHR(0.74453627100229844977/2)

  545. #define COS0_9  FIXHR(0.83934964541552703873/2)

  546. #define COS0_10 FIXHR(0.97256823786196069369/2)

  547. #define COS0_11 FIXHR(1.16943993343288495515/4)

  548. #define COS0_12 FIXHR(1.48416461631416627724/4)

  549. #define COS0_13 FIXHR(2.05778100995341155085/8)

  550. #define COS0_14 FIXHR(3.40760841846871878570/8)

  551. #define COS0_15 FIXHR(10.19000812354805681150/32)

  552. 

  553. #define COS1_0 FIXHR(0.50241928618815570551/2)

  554. #define COS1_1 FIXHR(0.52249861493968888062/2)

  555. #define COS1_2 FIXHR(0.56694403481635770368/2)

  556. #define COS1_3 FIXHR(0.64682178335999012954/2)

  557. #define COS1_4 FIXHR(0.78815462345125022473/2)

  558. #define COS1_5 FIXHR(1.06067768599034747134/4)

  559. #define COS1_6 FIXHR(1.72244709823833392782/4)

  560. #define COS1_7 FIXHR(5.10114861868916385802/16)

  561. 

  562. #define COS2_0 FIXHR(0.50979557910415916894/2)

  563. #define COS2_1 FIXHR(0.60134488693504528054/2)

  564. #define COS2_2 FIXHR(0.89997622313641570463/2)

  565. #define COS2_3 FIXHR(2.56291544774150617881/8)

  566. 

  567. #define COS3_0 FIXHR(0.54119610014619698439/2)

  568. #define COS3_1 FIXHR(1.30656296487637652785/4)

  569. 

  570. #define COS4_0 FIXHR(0.70710678118654752439/2)

  571. 

  572. /* butterfly operator */

  573. #define BF(a, b, c, s)\

  574. {\

  575.     tmp0 = tab[a] + tab[b];\

  576.     tmp1 = tab[a] - tab[b];\

  577.     tab[a] = tmp0;\

  578.     tab[b] = MULH(tmp1<<(s), c);\

  579. }

  580. 

  581. #define BF1(a, b, c, d)\

  582. {\

  583.     BF(a, b, COS4_0, 1);\

  584.     BF(c, d,-COS4_0, 1);\

  585.     tab[c] += tab[d];\

  586. }

  587. 

  588. #define BF2(a, b, c, d)\

  589. {\

  590.     BF(a, b, COS4_0, 1);\

  591.     BF(c, d,-COS4_0, 1);\

  592.     tab[c] += tab[d];\

  593.     tab[a] += tab[c];\

  594.     tab[c] += tab[b];\

  595.     tab[b] += tab[d];\

  596. }

  597. 

  598. #define ADD(a, b) tab[a] += tab[b]

  599. 

  600. /* DCT32 without 1/sqrt(2) coef zero scaling. */

  601. static void dct32(int32_t *out, int32_t *tab)

  602. {

  603.     int tmp0, tmp1;

  604. 

  605.     /* pass 1 */

  606.     BF( 0, 31, COS0_0 , 1);

  607.     BF(15, 16, COS0_15, 5);

  608.     /* pass 2 */

  609.     BF( 0, 15, COS1_0 , 1);

  610.     BF(16, 31,-COS1_0 , 1);

  611.     /* pass 1 */

  612.     BF( 7, 24, COS0_7 , 1);

  613.     BF( 8, 23, COS0_8 , 1);

  614.     /* pass 2 */

  615.     BF( 7,  8, COS1_7 , 4);

  616.     BF(23, 24,-COS1_7 , 4);

  617.     /* pass 3 */

  618.     BF( 0,  7, COS2_0 , 1);

  619.     BF( 8, 15,-COS2_0 , 1);

  620.     BF(16, 23, COS2_0 , 1);

  621.     BF(24, 31,-COS2_0 , 1);

  622.     /* pass 1 */

  623.     BF( 3, 28, COS0_3 , 1);

  624.     BF(12, 19, COS0_12, 2);

  625.     /* pass 2 */

  626.     BF( 3, 12, COS1_3 , 1);

  627.     BF(19, 28,-COS1_3 , 1);

  628.     /* pass 1 */

  629.     BF( 4, 27, COS0_4 , 1);

  630.     BF(11, 20, COS0_11, 2);

  631.     /* pass 2 */

  632.     BF( 4, 11, COS1_4 , 1);

  633.     BF(20, 27,-COS1_4 , 1);

  634.     /* pass 3 */

  635.     BF( 3,  4, COS2_3 , 3);

  636.     BF(11, 12,-COS2_3 , 3);

  637.     BF(19, 20, COS2_3 , 3);

  638.     BF(27, 28,-COS2_3 , 3);

  639.     /* pass 4 */

  640.     BF( 0,  3, COS3_0 , 1);

  641.     BF( 4,  7,-COS3_0 , 1);

  642.     BF( 8, 11, COS3_0 , 1);

  643.     BF(12, 15,-COS3_0 , 1);

  644.     BF(16, 19, COS3_0 , 1);

  645.     BF(20, 23,-COS3_0 , 1);

  646.     BF(24, 27, COS3_0 , 1);

  647.     BF(28, 31,-COS3_0 , 1);

  648. 

  649. 

  650. 

  651.     /* pass 1 */

  652.     BF( 1, 30, COS0_1 , 1);

  653.     BF(14, 17, COS0_14, 3);

  654.     /* pass 2 */

  655.     BF( 1, 14, COS1_1 , 1);

  656.     BF(17, 30,-COS1_1 , 1);

  657.     /* pass 1 */

  658.     BF( 6, 25, COS0_6 , 1);

  659.     BF( 9, 22, COS0_9 , 1);

  660.     /* pass 2 */

  661.     BF( 6,  9, COS1_6 , 2);

  662.     BF(22, 25,-COS1_6 , 2);

  663.     /* pass 3 */

  664.     BF( 1,  6, COS2_1 , 1);

  665.     BF( 9, 14,-COS2_1 , 1);

  666.     BF(17, 22, COS2_1 , 1);

  667.     BF(25, 30,-COS2_1 , 1);

  668. 

  669.     /* pass 1 */

  670.     BF( 2, 29, COS0_2 , 1);

  671.     BF(13, 18, COS0_13, 3);

  672.     /* pass 2 */

  673.     BF( 2, 13, COS1_2 , 1);

  674.     BF(18, 29,-COS1_2 , 1);

  675.     /* pass 1 */

  676.     BF( 5, 26, COS0_5 , 1);

  677.     BF(10, 21, COS0_10, 1);

  678.     /* pass 2 */

  679.     BF( 5, 10, COS1_5 , 2);

  680.     BF(21, 26,-COS1_5 , 2);

  681.     /* pass 3 */

  682.     BF( 2,  5, COS2_2 , 1);

  683.     BF(10, 13,-COS2_2 , 1);

  684.     BF(18, 21, COS2_2 , 1);

  685.     BF(26, 29,-COS2_2 , 1);

  686.     /* pass 4 */

  687.     BF( 1,  2, COS3_1 , 2);

  688.     BF( 5,  6,-COS3_1 , 2);

  689.     BF( 9, 10, COS3_1 , 2);

  690.     BF(13, 14,-COS3_1 , 2);

  691.     BF(17, 18, COS3_1 , 2);

  692.     BF(21, 22,-COS3_1 , 2);

  693.     BF(25, 26, COS3_1 , 2);

  694.     BF(29, 30,-COS3_1 , 2);

  695. 

  696.     /* pass 5 */

  697.     BF1( 0,  1,  2,  3);

  698.     BF2( 4,  5,  6,  7);

  699.     BF1( 8,  9, 10, 11);

  700.     BF2(12, 13, 14, 15);

  701.     BF1(16, 17, 18, 19);

  702.     BF2(20, 21, 22, 23);

  703.     BF1(24, 25, 26, 27);

  704.     BF2(28, 29, 30, 31);

  705. 

  706.     /* pass 6 */

  707. 

  708.     ADD( 8, 12);

  709.     ADD(12, 10);

  710.     ADD(10, 14);

  711.     ADD(14,  9);

  712.     ADD( 9, 13);

  713.     ADD(13, 11);

  714.     ADD(11, 15);

  715. 

  716.     out[ 0] = tab[0];

  717.     out[16] = tab[1];

  718.     out[ 8] = tab[2];

  719.     out[24] = tab[3];

  720.     out[ 4] = tab[4];

  721.     out[20] = tab[5];

  722.     out[12] = tab[6];

  723.     out[28] = tab[7];

  724.     out[ 2] = tab[8];

  725.     out[18] = tab[9];

  726.     out[10] = tab[10];

  727.     out[26] = tab[11];

  728.     out[ 6] = tab[12];

  729.     out[22] = tab[13];

  730.     out[14] = tab[14];

  731.     out[30] = tab[15];

  732. 

  733.     ADD(24, 28);

  734.     ADD(28, 26);

  735.     ADD(26, 30);

  736.     ADD(30, 25);

  737.     ADD(25, 29);

  738.     ADD(29, 27);

  739.     ADD(27, 31);

  740. 

  741.     out[ 1] = tab[16] + tab[24];

  742.     out[17] = tab[17] + tab[25];

  743.     out[ 9] = tab[18] + tab[26];

  744.     out[25] = tab[19] + tab[27];

  745.     out[ 5] = tab[20] + tab[28];

  746.     out[21] = tab[21] + tab[29];

  747.     out[13] = tab[22] + tab[30];

  748.     out[29] = tab[23] + tab[31];

  749.     out[ 3] = tab[24] + tab[20];

  750.     out[19] = tab[25] + tab[21];

  751.     out[11] = tab[26] + tab[22];

  752.     out[27] = tab[27] + tab[23];

  753.     out[ 7] = tab[28] + tab[18];

  754.     out[23] = tab[29] + tab[19];

  755.     out[15] = tab[30] + tab[17];

  756.     out[31] = tab[31];

  757. }

  758. 

  759. #if FRAC_BITS <= 15

  760. 

  761. static inline int round_sample(int *sum)

  762. {

  763.     int sum1;

  764.     sum1 = (*sum) >> OUT_SHIFT;

  765.     *sum &= (1<<OUT_SHIFT)-1;

  766.     if (sum1 < OUT_MIN)

  767.         sum1 = OUT_MIN;

  768.     else if (sum1 > OUT_MAX)

  769.         sum1 = OUT_MAX;

  770.     return sum1;

  771. }

  772. 

  773. /* signed 16x16 -> 32 multiply add accumulate */

  774. #define MACS(rt, ra, rb) MAC16(rt, ra, rb)

  775. 

  776. /* signed 16x16 -> 32 multiply */

  777. #define MULS(ra, rb) MUL16(ra, rb)

  778. 

  779. #else

  780. 

  781. static inline int round_sample(int64_t *sum)

  782. {

  783.     int sum1;

  784.     sum1 = (int)((*sum) >> OUT_SHIFT);

  785.     *sum &= (1<<OUT_SHIFT)-1;

  786.     if (sum1 < OUT_MIN)

  787.         sum1 = OUT_MIN;

  788.     else if (sum1 > OUT_MAX)

  789.         sum1 = OUT_MAX;

  790.     return sum1;

  791. }

  792. 

  793. #   define MULS(ra, rb) MUL64(ra, rb)

  794. #endif

  795. 

  796. #define SUM8(sum, op, w, p) \

  797. {                                               \

  798.     sum op MULS((w)[0 * 64], p[0 * 64]);\

  799.     sum op MULS((w)[1 * 64], p[1 * 64]);\

  800.     sum op MULS((w)[2 * 64], p[2 * 64]);\

  801.     sum op MULS((w)[3 * 64], p[3 * 64]);\

  802.     sum op MULS((w)[4 * 64], p[4 * 64]);\

  803.     sum op MULS((w)[5 * 64], p[5 * 64]);\

  804.     sum op MULS((w)[6 * 64], p[6 * 64]);\

  805.     sum op MULS((w)[7 * 64], p[7 * 64]);\

  806. }

  807. 

  808. #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \

  809. {                                               \

  810.     int tmp;\

  811.     tmp = p[0 * 64];\

  812.     sum1 op1 MULS((w1)[0 * 64], tmp);\

  813.     sum2 op2 MULS((w2)[0 * 64], tmp);\

  814.     tmp = p[1 * 64];\

  815.     sum1 op1 MULS((w1)[1 * 64], tmp);\

  816.     sum2 op2 MULS((w2)[1 * 64], tmp);\

  817.     tmp = p[2 * 64];\

  818.     sum1 op1 MULS((w1)[2 * 64], tmp);\

  819.     sum2 op2 MULS((w2)[2 * 64], tmp);\

  820.     tmp = p[3 * 64];\

  821.     sum1 op1 MULS((w1)[3 * 64], tmp);\

  822.     sum2 op2 MULS((w2)[3 * 64], tmp);\

  823.     tmp = p[4 * 64];\

  824.     sum1 op1 MULS((w1)[4 * 64], tmp);\

  825.     sum2 op2 MULS((w2)[4 * 64], tmp);\

  826.     tmp = p[5 * 64];\

  827.     sum1 op1 MULS((w1)[5 * 64], tmp);\

  828.     sum2 op2 MULS((w2)[5 * 64], tmp);\

  829.     tmp = p[6 * 64];\

  830.     sum1 op1 MULS((w1)[6 * 64], tmp);\

  831.     sum2 op2 MULS((w2)[6 * 64], tmp);\

  832.     tmp = p[7 * 64];\

  833.     sum1 op1 MULS((w1)[7 * 64], tmp);\

  834.     sum2 op2 MULS((w2)[7 * 64], tmp);\

  835. }

  836. 

  837. void ff_mpa_synth_init(MPA_INT *window)

  838. {

  839.     int i;

  840. 

  841.     /* max = 18760, max sum over all 16 coefs : 44736 */

  842.     for(i=0;i<257;i++) {

  843.         int v;

  844.         v = ff_mpa_enwindow[i];

  845. #if WFRAC_BITS < 16

  846.         v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);

  847. #endif

  848.         window[i] = v;

  849.         if ((i & 63) != 0)

  850.             v = -v;

  851.         if (i != 0)

  852.             window[512 - i] = v;

  853.     }

  854. }

  855. 

  856. /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:

  857.    32 samples. */

  858. /* XXX: optimize by avoiding ring buffer usage */

  859. void ff_mpa_synth_filter(MPA_INT *synth_buf_ptr, int *synth_buf_offset,

  860.                          MPA_INT *window, int *dither_state,

  861.                          OUT_INT *samples, int incr,

  862.                          int32_t sb_samples[SBLIMIT])

  863. {

  864.     int32_t tmp[32];

  865.     register MPA_INT *synth_buf;

  866.     register const MPA_INT *w, *w2, *p;

  867.     int j, offset, v;

  868.     OUT_INT *samples2;

  869. #if FRAC_BITS <= 15

  870.     int sum, sum2;

  871. #else

  872.     int64_t sum, sum2;

  873. #endif

  874. 

  875.     dct32(tmp, sb_samples);

  876. 

  877.     offset = *synth_buf_offset;

  878.     synth_buf = synth_buf_ptr + offset;

  879. 

  880.     for(j=0;j<32;j++) {

  881.         v = tmp[j];

  882. #if FRAC_BITS <= 15

  883.         /* NOTE: can cause a loss in precision if very high amplitude

  884.            sound */

  885.         v = av_clip_int16(v);

  886. #endif

  887.         synth_buf[j] = v;

  888.     }

  889.     /* copy to avoid wrap */

  890.     memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));

  891. 

  892.     samples2 = samples + 31 * incr;

  893.     w = window;

  894.     w2 = window + 31;

  895. 

  896.     sum = *dither_state;

  897.     p = synth_buf + 16;

  898.     SUM8(sum, +=, w, p);

  899.     p = synth_buf + 48;

  900.     SUM8(sum, -=, w + 32, p);

  901.     *samples = round_sample(&sum);

  902.     samples += incr;

  903.     w++;

  904. 

  905.     /* we calculate two samples at the same time to avoid one memory

  906.        access per two sample */

  907.     for(j=1;j<16;j++) {

  908.         sum2 = 0;

  909.         p = synth_buf + 16 + j;

  910.         SUM8P2(sum, +=, sum2, -=, w, w2, p);

  911.         p = synth_buf + 48 - j;

  912.         SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);

  913. 

  914.         *samples = round_sample(&sum);

  915.         samples += incr;

  916.         sum += sum2;

  917.         *samples2 = round_sample(&sum);

  918.         samples2 -= incr;

  919.         w++;

  920.         w2--;

  921.     }

  922. 

  923.     p = synth_buf + 32;

  924.     SUM8(sum, -=, w + 32, p);

  925.     *samples = round_sample(&sum);

  926.     *dither_state= sum;

  927. 

  928.     offset = (offset - 32) & 511;

  929.     *synth_buf_offset = offset;

  930. }

  931. 

  932. #define C3 FIXHR(0.86602540378443864676/2)

  933. 

  934. /* 0.5 / cos(pi*(2*i+1)/36) */

  935. static const int icos36[9] = {

  936.     FIXR(0.50190991877167369479),

  937.     FIXR(0.51763809020504152469), //0

  938.     FIXR(0.55168895948124587824),

  939.     FIXR(0.61038729438072803416),

  940.     FIXR(0.70710678118654752439), //1

  941.     FIXR(0.87172339781054900991),

  942.     FIXR(1.18310079157624925896),

  943.     FIXR(1.93185165257813657349), //2

  944.     FIXR(5.73685662283492756461),

  945. };

  946. 

  947. /* 0.5 / cos(pi*(2*i+1)/36) */

  948. static const int icos36h[9] = {

  949.     FIXHR(0.50190991877167369479/2),

  950.     FIXHR(0.51763809020504152469/2), //0

  951.     FIXHR(0.55168895948124587824/2),

  952.     FIXHR(0.61038729438072803416/2),

  953.     FIXHR(0.70710678118654752439/2), //1

  954.     FIXHR(0.87172339781054900991/2),

  955.     FIXHR(1.18310079157624925896/4),

  956.     FIXHR(1.93185165257813657349/4), //2

  957. //    FIXHR(5.73685662283492756461),

  958. };

  959. 

  960. /* 12 points IMDCT. We compute it "by hand" by factorizing obvious

  961.    cases. */

  962. static void imdct12(int *out, int *in)

  963. {

  964.     int in0, in1, in2, in3, in4, in5, t1, t2;

  965. 

  966.     in0= in[0*3];

  967.     in1= in[1*3] + in[0*3];

  968.     in2= in[2*3] + in[1*3];

  969.     in3= in[3*3] + in[2*3];

  970.     in4= in[4*3] + in[3*3];

  971.     in5= in[5*3] + in[4*3];

  972.     in5 += in3;

  973.     in3 += in1;

  974. 

  975.     in2= MULH(2*in2, C3);

  976.     in3= MULH(4*in3, C3);

  977. 

  978.     t1 = in0 - in4;

  979.     t2 = MULH(2*(in1 - in5), icos36h[4]);

  980. 

  981.     out[ 7]=

  982.     out[10]= t1 + t2;

  983.     out[ 1]=

  984.     out[ 4]= t1 - t2;

  985. 

  986.     in0 += in4>>1;

  987.     in4 = in0 + in2;

  988.     in5 += 2*in1;

  989.     in1 = MULH(in5 + in3, icos36h[1]);

  990.     out[ 8]=

  991.     out[ 9]= in4 + in1;

  992.     out[ 2]=

  993.     out[ 3]= in4 - in1;

  994. 

  995.     in0 -= in2;

  996.     in5 = MULH(2*(in5 - in3), icos36h[7]);

  997.     out[ 0]=

  998.     out[ 5]= in0 - in5;

  999.     out[ 6]=

  1000.     out[11]= in0 + in5;

  1001. }

  1002. 

  1003. /* cos(pi*i/18) */

  1004. #define C1 FIXHR(0.98480775301220805936/2)

  1005. #define C2 FIXHR(0.93969262078590838405/2)

  1006. #define C3 FIXHR(0.86602540378443864676/2)

  1007. #define C4 FIXHR(0.76604444311897803520/2)

  1008. #define C5 FIXHR(0.64278760968653932632/2)

  1009. #define C6 FIXHR(0.5/2)

  1010. #define C7 FIXHR(0.34202014332566873304/2)

  1011. #define C8 FIXHR(0.17364817766693034885/2)

  1012. 

  1013. 

  1014. /* using Lee like decomposition followed by hand coded 9 points DCT */

  1015. static void imdct36(int *out, int *buf, int *in, int *win)

  1016. {

  1017.     int i, j, t0, t1, t2, t3, s0, s1, s2, s3;

  1018.     int tmp[18], *tmp1, *in1;

  1019. 

  1020.     for(i=17;i>=1;i--)

  1021.         in[i] += in[i-1];

  1022.     for(i=17;i>=3;i-=2)

  1023.         in[i] += in[i-2];

  1024. 

  1025.     for(j=0;j<2;j++) {

  1026.         tmp1 = tmp + j;

  1027.         in1 = in + j;

  1028. #if 0

  1029. //more accurate but slower

  1030.         int64_t t0, t1, t2, t3;

  1031.         t2 = in1[2*4] + in1[2*8] - in1[2*2];

  1032. 

  1033.         t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;

  1034.         t1 = in1[2*0] - in1[2*6];

  1035.         tmp1[ 6] = t1 - (t2>>1);

  1036.         tmp1[16] = t1 + t2;

  1037. 

  1038.         t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);

  1039.         t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);

  1040.         t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);

  1041. 

  1042.         tmp1[10] = (t3 - t0 - t2) >> 32;

  1043.         tmp1[ 2] = (t3 + t0 + t1) >> 32;

  1044.         tmp1[14] = (t3 + t2 - t1) >> 32;

  1045. 

  1046.         tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);

  1047.         t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);

  1048.         t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);

  1049.         t0 = MUL64(2*in1[2*3], C3);

  1050. 

  1051.         t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);

  1052. 

  1053.         tmp1[ 0] = (t2 + t3 + t0) >> 32;

  1054.         tmp1[12] = (t2 + t1 - t0) >> 32;

  1055.         tmp1[ 8] = (t3 - t1 - t0) >> 32;

  1056. #else

  1057.         t2 = in1[2*4] + in1[2*8] - in1[2*2];

  1058. 

  1059.         t3 = in1[2*0] + (in1[2*6]>>1);

  1060.         t1 = in1[2*0] - in1[2*6];

  1061.         tmp1[ 6] = t1 - (t2>>1);

  1062.         tmp1[16] = t1 + t2;

  1063. 

  1064.         t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);

  1065.         t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);

  1066.         t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);

  1067. 

  1068.         tmp1[10] = t3 - t0 - t2;

  1069.         tmp1[ 2] = t3 + t0 + t1;

  1070.         tmp1[14] = t3 + t2 - t1;

  1071. 

  1072.         tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);

  1073.         t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);

  1074.         t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);

  1075.         t0 = MULH(2*in1[2*3], C3);

  1076. 

  1077.         t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);

  1078. 

  1079.         tmp1[ 0] = t2 + t3 + t0;

  1080.         tmp1[12] = t2 + t1 - t0;

  1081.         tmp1[ 8] = t3 - t1 - t0;

  1082. #endif

  1083.     }

  1084. 

  1085.     i = 0;

  1086.     for(j=0;j<4;j++) {

  1087.         t0 = tmp[i];

  1088.         t1 = tmp[i + 2];

  1089.         s0 = t1 + t0;

  1090.         s2 = t1 - t0;

  1091. 

  1092.         t2 = tmp[i + 1];

  1093.         t3 = tmp[i + 3];

  1094.         s1 = MULH(2*(t3 + t2), icos36h[j]);

  1095.         s3 = MULL(t3 - t2, icos36[8 - j]);

  1096. 

  1097.         t0 = s0 + s1;

  1098.         t1 = s0 - s1;

  1099.         out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];

  1100.         out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];

  1101.         buf[9 + j] = MULH(t0, win[18 + 9 + j]);

  1102.         buf[8 - j] = MULH(t0, win[18 + 8 - j]);

  1103. 

  1104.         t0 = s2 + s3;

  1105.         t1 = s2 - s3;

  1106.         out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];

  1107.         out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];

  1108.         buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);

  1109.         buf[      + j] = MULH(t0, win[18         + j]);

  1110.         i += 4;

  1111.     }

  1112. 

  1113.     s0 = tmp[16];

  1114.     s1 = MULH(2*tmp[17], icos36h[4]);

  1115.     t0 = s0 + s1;

  1116.     t1 = s0 - s1;

  1117.     out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];

  1118.     out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];

  1119.     buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);

  1120.     buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);

  1121. }

  1122. 

  1123. /* return the number of decoded frames */

  1124. static int mp_decode_layer1(MPADecodeContext *s)

  1125. {

  1126.     int bound, i, v, n, ch, j, mant;

  1127.     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];

  1128.     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

  1129. 

  1130.     if (s->mode == MPA_JSTEREO)

  1131.         bound = (s->mode_ext + 1) * 4;

  1132.     else

  1133.         bound = SBLIMIT;

  1134. 

  1135.     /* allocation bits */

  1136.     for(i=0;i<bound;i++) {

  1137.         for(ch=0;ch<s->nb_channels;ch++) {

  1138.             allocation[ch][i] = get_bits(&s->gb, 4);

  1139.         }

  1140.     }

  1141.     for(i=bound;i<SBLIMIT;i++) {

  1142.         allocation[0][i] = get_bits(&s->gb, 4);

  1143.     }

  1144. 

  1145.     /* scale factors */

  1146.     for(i=0;i<bound;i++) {

  1147.         for(ch=0;ch<s->nb_channels;ch++) {

  1148.             if (allocation[ch][i])

  1149.                 scale_factors[ch][i] = get_bits(&s->gb, 6);

  1150.         }

  1151.     }

  1152.     for(i=bound;i<SBLIMIT;i++) {

  1153.         if (allocation[0][i]) {

  1154.             scale_factors[0][i] = get_bits(&s->gb, 6);

  1155.             scale_factors[1][i] = get_bits(&s->gb, 6);

  1156.         }

  1157.     }

  1158. 

  1159.     /* compute samples */

  1160.     for(j=0;j<12;j++) {

  1161.         for(i=0;i<bound;i++) {

  1162.             for(ch=0;ch<s->nb_channels;ch++) {

  1163.                 n = allocation[ch][i];

  1164.                 if (n) {

  1165.                     mant = get_bits(&s->gb, n + 1);

  1166.                     v = l1_unscale(n, mant, scale_factors[ch][i]);

  1167.                 } else {

  1168.                     v = 0;

  1169.                 }

  1170.                 s->sb_samples[ch][j][i] = v;

  1171.             }

  1172.         }

  1173.         for(i=bound;i<SBLIMIT;i++) {

  1174.             n = allocation[0][i];

  1175.             if (n) {

  1176.                 mant = get_bits(&s->gb, n + 1);

  1177.                 v = l1_unscale(n, mant, scale_factors[0][i]);

  1178.                 s->sb_samples[0][j][i] = v;

  1179.                 v = l1_unscale(n, mant, scale_factors[1][i]);

  1180.                 s->sb_samples[1][j][i] = v;

  1181.             } else {

  1182.                 s->sb_samples[0][j][i] = 0;

  1183.                 s->sb_samples[1][j][i] = 0;

  1184.             }

  1185.         }

  1186.     }

  1187.     return 12;

  1188. }

  1189. 

  1190. static int mp_decode_layer2(MPADecodeContext *s)

  1191. {

  1192.     int sblimit; /* number of used subbands */

  1193.     const unsigned char *alloc_table;

  1194.     int table, bit_alloc_bits, i, j, ch, bound, v;

  1195.     unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];

  1196.     unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];

  1197.     unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;

  1198.     int scale, qindex, bits, steps, k, l, m, b;

  1199. 

  1200.     /* select decoding table */

  1201.     table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,

  1202.                             s->sample_rate, s->lsf);

  1203.     sblimit = ff_mpa_sblimit_table[table];

  1204.     alloc_table = ff_mpa_alloc_tables[table];

  1205. 

  1206.     if (s->mode == MPA_JSTEREO)

  1207.         bound = (s->mode_ext + 1) * 4;

  1208.     else

  1209.         bound = sblimit;

  1210. 

  1211.     dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);

  1212. 

  1213.     /* sanity check */

  1214.     if( bound > sblimit ) bound = sblimit;

  1215. 

  1216.     /* parse bit allocation */

  1217.     j = 0;

  1218.     for(i=0;i<bound;i++) {

  1219.         bit_alloc_bits = alloc_table[j];

  1220.         for(ch=0;ch<s->nb_channels;ch++) {

  1221.             bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);

  1222.         }

  1223.         j += 1 << bit_alloc_bits;

  1224.     }

  1225.     for(i=bound;i<sblimit;i++) {

  1226.         bit_alloc_bits = alloc_table[j];

  1227.         v = get_bits(&s->gb, bit_alloc_bits);

  1228.         bit_alloc[0][i] = v;

  1229.         bit_alloc[1][i] = v;

  1230.         j += 1 << bit_alloc_bits;

  1231.     }

  1232. 

  1233. #ifdef DEBUG

  1234.     {

  1235.         for(ch=0;ch<s->nb_channels;ch++) {

  1236.             for(i=0;i<sblimit;i++)

  1237.                 dprintf(s->avctx, " %d", bit_alloc[ch][i]);

  1238.             dprintf(s->avctx, "\n");

  1239.         }

  1240.     }

  1241. #endif

  1242. 

  1243.     /* scale codes */

  1244.     for(i=0;i<sblimit;i++) {

  1245.         for(ch=0;ch<s->nb_channels;ch++) {

  1246.             if (bit_alloc[ch][i])

  1247.                 scale_code[ch][i] = get_bits(&s->gb, 2);

  1248.         }

  1249.     }

  1250. 

  1251.     /* scale factors */

  1252.     for(i=0;i<sblimit;i++) {

  1253.         for(ch=0;ch<s->nb_channels;ch++) {

  1254.             if (bit_alloc[ch][i]) {

  1255.                 sf = scale_factors[ch][i];

  1256.                 switch(scale_code[ch][i]) {

  1257.                 default:

  1258.                 case 0:

  1259.                     sf[0] = get_bits(&s->gb, 6);

  1260.                     sf[1] = get_bits(&s->gb, 6);

  1261.                     sf[2] = get_bits(&s->gb, 6);

  1262.                     break;

  1263.                 case 2:

  1264.                     sf[0] = get_bits(&s->gb, 6);

  1265.                     sf[1] = sf[0];

  1266.                     sf[2] = sf[0];

  1267.                     break;

  1268.                 case 1:

  1269.                     sf[0] = get_bits(&s->gb, 6);

  1270.                     sf[2] = get_bits(&s->gb, 6);

  1271.                     sf[1] = sf[0];

  1272.                     break;

  1273.                 case 3:

  1274.                     sf[0] = get_bits(&s->gb, 6);

  1275.                     sf[2] = get_bits(&s->gb, 6);

  1276.                     sf[1] = sf[2];

  1277.                     break;

  1278.                 }

  1279.             }

  1280.         }

  1281.     }

  1282. 

  1283. #ifdef DEBUG

  1284.     for(ch=0;ch<s->nb_channels;ch++) {

  1285.         for(i=0;i<sblimit;i++) {

  1286.             if (bit_alloc[ch][i]) {

  1287.                 sf = scale_factors[ch][i];

  1288.                 dprintf(s->avctx, " %d %d %d", sf[0], sf[1], sf[2]);

  1289.             } else {

  1290.                 dprintf(s->avctx, " -");

  1291.             }

  1292.         }

  1293.         dprintf(s->avctx, "\n");

  1294.     }

  1295. #endif

  1296. 

  1297.     /* samples */

  1298.     for(k=0;k<3;k++) {

  1299.         for(l=0;l<12;l+=3) {

  1300.             j = 0;

  1301.             for(i=0;i<bound;i++) {

  1302.                 bit_alloc_bits = alloc_table[j];

  1303.                 for(ch=0;ch<s->nb_channels;ch++) {

  1304.                     b = bit_alloc[ch][i];

  1305.                     if (b) {

  1306.                         scale = scale_factors[ch][i][k];

  1307.                         qindex = alloc_table[j+b];

  1308.                         bits = ff_mpa_quant_bits[qindex];

  1309.                         if (bits < 0) {

  1310.                             /* 3 values at the same time */

  1311.                             v = get_bits(&s->gb, -bits);

  1312.                             steps = ff_mpa_quant_steps[qindex];

  1313.                             s->sb_samples[ch][k * 12 + l + 0][i] =

  1314.                                 l2_unscale_group(steps, v % steps, scale);

  1315.                             v = v / steps;

  1316.                             s->sb_samples[ch][k * 12 + l + 1][i] =

  1317.                                 l2_unscale_group(steps, v % steps, scale);

  1318.                             v = v / steps;

  1319.                             s->sb_samples[ch][k * 12 + l + 2][i] =

  1320.                                 l2_unscale_group(steps, v, scale);

  1321.                         } else {

  1322.                             for(m=0;m<3;m++) {

  1323.                                 v = get_bits(&s->gb, bits);

  1324.                                 v = l1_unscale(bits - 1, v, scale);

  1325.                                 s->sb_samples[ch][k * 12 + l + m][i] = v;

  1326.                             }

  1327.                         }

  1328.                     } else {

  1329.                         s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1330.                         s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1331.                         s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1332.                     }

  1333.                 }

  1334.                 /* next subband in alloc table */

  1335.                 j += 1 << bit_alloc_bits;

  1336.             }

  1337.             /* XXX: find a way to avoid this duplication of code */

  1338.             for(i=bound;i<sblimit;i++) {

  1339.                 bit_alloc_bits = alloc_table[j];

  1340.                 b = bit_alloc[0][i];

  1341.                 if (b) {

  1342.                     int mant, scale0, scale1;

  1343.                     scale0 = scale_factors[0][i][k];

  1344.                     scale1 = scale_factors[1][i][k];

  1345.                     qindex = alloc_table[j+b];

  1346.                     bits = ff_mpa_quant_bits[qindex];

  1347.                     if (bits < 0) {

  1348.                         /* 3 values at the same time */

  1349.                         v = get_bits(&s->gb, -bits);

  1350.                         steps = ff_mpa_quant_steps[qindex];

  1351.                         mant = v % steps;

  1352.                         v = v / steps;

  1353.                         s->sb_samples[0][k * 12 + l + 0][i] =

  1354.                             l2_unscale_group(steps, mant, scale0);

  1355.                         s->sb_samples[1][k * 12 + l + 0][i] =

  1356.                             l2_unscale_group(steps, mant, scale1);

  1357.                         mant = v % steps;

  1358.                         v = v / steps;

  1359.                         s->sb_samples[0][k * 12 + l + 1][i] =

  1360.                             l2_unscale_group(steps, mant, scale0);

  1361.                         s->sb_samples[1][k * 12 + l + 1][i] =

  1362.                             l2_unscale_group(steps, mant, scale1);

  1363.                         s->sb_samples[0][k * 12 + l + 2][i] =

  1364.                             l2_unscale_group(steps, v, scale0);

  1365.                         s->sb_samples[1][k * 12 + l + 2][i] =

  1366.                             l2_unscale_group(steps, v, scale1);

  1367.                     } else {

  1368.                         for(m=0;m<3;m++) {

  1369.                             mant = get_bits(&s->gb, bits);

  1370.                             s->sb_samples[0][k * 12 + l + m][i] =

  1371.                                 l1_unscale(bits - 1, mant, scale0);

  1372.                             s->sb_samples[1][k * 12 + l + m][i] =

  1373.                                 l1_unscale(bits - 1, mant, scale1);

  1374.                         }

  1375.                     }

  1376.                 } else {

  1377.                     s->sb_samples[0][k * 12 + l + 0][i] = 0;

  1378.                     s->sb_samples[0][k * 12 + l + 1][i] = 0;

  1379.                     s->sb_samples[0][k * 12 + l + 2][i] = 0;

  1380.                     s->sb_samples[1][k * 12 + l + 0][i] = 0;

  1381.                     s->sb_samples[1][k * 12 + l + 1][i] = 0;

  1382.                     s->sb_samples[1][k * 12 + l + 2][i] = 0;

  1383.                 }

  1384.                 /* next subband in alloc table */

  1385.                 j += 1 << bit_alloc_bits;

  1386.             }

  1387.             /* fill remaining samples to zero */

  1388.             for(i=sblimit;i<SBLIMIT;i++) {

  1389.                 for(ch=0;ch<s->nb_channels;ch++) {

  1390.                     s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1391.                     s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1392.                     s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1393.                 }

  1394.             }

  1395.         }

  1396.     }

  1397.     return 3 * 12;

  1398. }

  1399. 

  1400. static inline void lsf_sf_expand(int *slen,

  1401.                                  int sf, int n1, int n2, int n3)

  1402. {

  1403.     if (n3) {

  1404.         slen[3] = sf % n3;

  1405.         sf /= n3;

  1406.     } else {

  1407.         slen[3] = 0;

  1408.     }

  1409.     if (n2) {

  1410.         slen[2] = sf % n2;

  1411.         sf /= n2;

  1412.     } else {

  1413.         slen[2] = 0;

  1414.     }

  1415.     slen[1] = sf % n1;

  1416.     sf /= n1;

  1417.     slen[0] = sf;

  1418. }

  1419. 

  1420. static void exponents_from_scale_factors(MPADecodeContext *s,

  1421.                                          GranuleDef *g,

  1422.                                          int16_t *exponents)

  1423. {

  1424.     const uint8_t *bstab, *pretab;

  1425.     int len, i, j, k, l, v0, shift, gain, gains[3];

  1426.     int16_t *exp_ptr;

  1427. 

  1428.     exp_ptr = exponents;

  1429.     gain = g->global_gain - 210;

  1430.     shift = g->scalefac_scale + 1;

  1431. 

  1432.     bstab = band_size_long[s->sample_rate_index];

  1433.     pretab = mpa_pretab[g->preflag];

  1434.     for(i=0;i<g->long_end;i++) {

  1435.         v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;

  1436.         len = bstab[i];

  1437.         for(j=len;j>0;j--)

  1438.             *exp_ptr++ = v0;

  1439.     }

  1440. 

  1441.     if (g->short_start < 13) {

  1442.         bstab = band_size_short[s->sample_rate_index];

  1443.         gains[0] = gain - (g->subblock_gain[0] << 3);

  1444.         gains[1] = gain - (g->subblock_gain[1] << 3);

  1445.         gains[2] = gain - (g->subblock_gain[2] << 3);

  1446.         k = g->long_end;

  1447.         for(i=g->short_start;i<13;i++) {

  1448.             len = bstab[i];

  1449.             for(l=0;l<3;l++) {

  1450.                 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;

  1451.                 for(j=len;j>0;j--)

  1452.                 *exp_ptr++ = v0;

  1453.             }

  1454.         }

  1455.     }

  1456. }

  1457. 

  1458. /* handle n = 0 too */

  1459. static inline int get_bitsz(GetBitContext *s, int n)

  1460. {

  1461.     if (n == 0)

  1462.         return 0;

  1463.     else

  1464.         return get_bits(s, n);

  1465. }

  1466. 

  1467. 

  1468. static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){

  1469.     if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){

  1470.         s->gb= s->in_gb;

  1471.         s->in_gb.buffer=NULL;

  1472.         assert((get_bits_count(&s->gb) & 7) == 0);

  1473.         skip_bits_long(&s->gb, *pos - *end_pos);

  1474.         *end_pos2=

  1475.         *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;

  1476.         *pos= get_bits_count(&s->gb);

  1477.     }

  1478. }

  1479. 

  1480. static int huffman_decode(MPADecodeContext *s, GranuleDef *g,

  1481.                           int16_t *exponents, int end_pos2)

  1482. {

  1483.     int s_index;

  1484.     int i;

  1485.     int last_pos, bits_left;

  1486.     VLC *vlc;

  1487.     int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);

  1488. 

  1489.     /* low frequencies (called big values) */

  1490.     s_index = 0;

  1491.     for(i=0;i<3;i++) {

  1492.         int j, k, l, linbits;

  1493.         j = g->region_size[i];

  1494.         if (j == 0)

  1495.             continue;

  1496.         /* select vlc table */

  1497.         k = g->table_select[i];

  1498.         l = mpa_huff_data[k][0];

  1499.         linbits = mpa_huff_data[k][1];

  1500.         vlc = &huff_vlc[l];

  1501. 

  1502.         if(!l){

  1503.             memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);

  1504.             s_index += 2*j;

  1505.             continue;

  1506.         }

  1507. 

  1508.         /* read huffcode and compute each couple */

  1509.         for(;j>0;j--) {

  1510.             int exponent, x, y, v;

  1511.             int pos= get_bits_count(&s->gb);

  1512. 

  1513.             if (pos >= end_pos){

  1514. //                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);

  1515.                 switch_buffer(s, &pos, &end_pos, &end_pos2);

  1516. //                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);

  1517.                 if(pos >= end_pos)

  1518.                     break;

  1519.             }

  1520.             y = get_vlc2(&s->gb, vlc->table, 7, 3);

  1521. 

  1522.             if(!y){

  1523.                 g->sb_hybrid[s_index  ] =

  1524.                 g->sb_hybrid[s_index+1] = 0;

  1525.                 s_index += 2;

  1526.                 continue;

  1527.             }

  1528. 

  1529.             exponent= exponents[s_index];

  1530. 

  1531.             dprintf(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",

  1532.                     i, g->region_size[i] - j, x, y, exponent);

  1533.             if(y&16){

  1534.                 x = y >> 5;

  1535.                 y = y & 0x0f;

  1536.                 if (x < 15){

  1537.                     v = expval_table[ exponent ][ x ];

  1538. //                      v = expval_table[ (exponent&3) ][ x ] >> FFMIN(0 - (exponent>>2), 31);

  1539.                 }else{

  1540.                     x += get_bitsz(&s->gb, linbits);

  1541.                     v = l3_unscale(x, exponent);

  1542.                 }

  1543.                 if (get_bits1(&s->gb))

  1544.                     v = -v;

  1545.                 g->sb_hybrid[s_index] = v;

  1546.                 if (y < 15){

  1547.                     v = expval_table[ exponent ][ y ];

  1548.                 }else{

  1549.                     y += get_bitsz(&s->gb, linbits);

  1550.                     v = l3_unscale(y, exponent);

  1551.                 }

  1552.                 if (get_bits1(&s->gb))

  1553.                     v = -v;

  1554.                 g->sb_hybrid[s_index+1] = v;

  1555.             }else{

  1556.                 x = y >> 5;

  1557.                 y = y & 0x0f;

  1558.                 x += y;

  1559.                 if (x < 15){

  1560.                     v = expval_table[ exponent ][ x ];

  1561.                 }else{

  1562.                     x += get_bitsz(&s->gb, linbits);

  1563.                     v = l3_unscale(x, exponent);

  1564.                 }

  1565.                 if (get_bits1(&s->gb))

  1566.                     v = -v;

  1567.                 g->sb_hybrid[s_index+!!y] = v;

  1568.                 g->sb_hybrid[s_index+ !y] = 0;

  1569.             }

  1570.             s_index+=2;

  1571.         }

  1572.     }

  1573. 

  1574.     /* high frequencies */

  1575.     vlc = &huff_quad_vlc[g->count1table_select];

  1576.     last_pos=0;

  1577.     while (s_index <= 572) {

  1578.         int pos, code;

  1579.         pos = get_bits_count(&s->gb);

  1580.         if (pos >= end_pos) {

  1581.             if (pos > end_pos2 && last_pos){

  1582.                 /* some encoders generate an incorrect size for this

  1583.                    part. We must go back into the data */

  1584.                 s_index -= 4;

  1585.                 skip_bits_long(&s->gb, last_pos - pos);

  1586.                 av_log(NULL, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);

  1587.                 if(s->error_resilience >= FF_ER_COMPLIANT)

  1588.                     s_index=0;

  1589.                 break;

  1590.             }

  1591. //                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);

  1592.             switch_buffer(s, &pos, &end_pos, &end_pos2);

  1593. //                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);

  1594.             if(pos >= end_pos)

  1595.                 break;

  1596.         }

  1597.         last_pos= pos;

  1598. 

  1599.         code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);

  1600.         dprintf(s->avctx, "t=%d code=%d\n", g->count1table_select, code);

  1601.         g->sb_hybrid[s_index+0]=

  1602.         g->sb_hybrid[s_index+1]=

  1603.         g->sb_hybrid[s_index+2]=

  1604.         g->sb_hybrid[s_index+3]= 0;

  1605.         while(code){

  1606.             static const int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};

  1607.             int v;

  1608.             int pos= s_index+idxtab[code];

  1609.             code ^= 8>>idxtab[code];

  1610.             v = exp_table[ exponents[pos] ];

  1611. //            v = exp_table[ (exponents[pos]&3) ] >> FFMIN(0 - (exponents[pos]>>2), 31);

  1612.             if(get_bits1(&s->gb))

  1613.                 v = -v;

  1614.             g->sb_hybrid[pos] = v;

  1615.         }

  1616.         s_index+=4;

  1617.     }

  1618.     /* skip extension bits */

  1619.     bits_left = end_pos2 - get_bits_count(&s->gb);

  1620. //av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);

  1621.     if (bits_left < 0/* || bits_left > 500*/) {

  1622.         av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1623.         s_index=0;

  1624.     }else if(bits_left > 0 && s->error_resilience >= FF_ER_AGGRESSIVE){

  1625.         av_log(NULL, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1626.         s_index=0;

  1627.     }

  1628.     memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));

  1629.     skip_bits_long(&s->gb, bits_left);

  1630. 

  1631.     i= get_bits_count(&s->gb);

  1632.     switch_buffer(s, &i, &end_pos, &end_pos2);

  1633. 

  1634.     return 0;

  1635. }

  1636. 

  1637. /* Reorder short blocks from bitstream order to interleaved order. It

  1638.    would be faster to do it in parsing, but the code would be far more

  1639.    complicated */

  1640. static void reorder_block(MPADecodeContext *s, GranuleDef *g)

  1641. {

  1642.     int i, j, len;

  1643.     int32_t *ptr, *dst, *ptr1;

  1644.     int32_t tmp[576];

  1645. 

  1646.     if (g->block_type != 2)

  1647.         return;

  1648. 

  1649.     if (g->switch_point) {

  1650.         if (s->sample_rate_index != 8) {

  1651.             ptr = g->sb_hybrid + 36;

  1652.         } else {

  1653.             ptr = g->sb_hybrid + 48;

  1654.         }

  1655.     } else {

  1656.         ptr = g->sb_hybrid;

  1657.     }

  1658. 

  1659.     for(i=g->short_start;i<13;i++) {

  1660.         len = band_size_short[s->sample_rate_index][i];

  1661.         ptr1 = ptr;

  1662.         dst = tmp;

  1663.         for(j=len;j>0;j--) {

  1664.             *dst++ = ptr[0*len];

  1665.             *dst++ = ptr[1*len];

  1666.             *dst++ = ptr[2*len];

  1667.             ptr++;

  1668.         }

  1669.         ptr+=2*len;

  1670.         memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));

  1671.     }

  1672. }

  1673. 

  1674. #define ISQRT2 FIXR(0.70710678118654752440)

  1675. 

  1676. static void compute_stereo(MPADecodeContext *s,

  1677.                            GranuleDef *g0, GranuleDef *g1)

  1678. {

  1679.     int i, j, k, l;

  1680.     int32_t v1, v2;

  1681.     int sf_max, tmp0, tmp1, sf, len, non_zero_found;

  1682.     int32_t (*is_tab)[16];

  1683.     int32_t *tab0, *tab1;

  1684.     int non_zero_found_short[3];

  1685. 

  1686.     /* intensity stereo */

  1687.     if (s->mode_ext & MODE_EXT_I_STEREO) {

  1688.         if (!s->lsf) {

  1689.             is_tab = is_table;

  1690.             sf_max = 7;

  1691.         } else {

  1692.             is_tab = is_table_lsf[g1->scalefac_compress & 1];

  1693.             sf_max = 16;

  1694.         }

  1695. 

  1696.         tab0 = g0->sb_hybrid + 576;

  1697.         tab1 = g1->sb_hybrid + 576;

  1698. 

  1699.         non_zero_found_short[0] = 0;

  1700.         non_zero_found_short[1] = 0;

  1701.         non_zero_found_short[2] = 0;

  1702.         k = (13 - g1->short_start) * 3 + g1->long_end - 3;

  1703.         for(i = 12;i >= g1->short_start;i--) {

  1704.             /* for last band, use previous scale factor */

  1705.             if (i != 11)

  1706.                 k -= 3;

  1707.             len = band_size_short[s->sample_rate_index][i];

  1708.             for(l=2;l>=0;l--) {

  1709.                 tab0 -= len;

  1710.                 tab1 -= len;

  1711.                 if (!non_zero_found_short[l]) {

  1712.                     /* test if non zero band. if so, stop doing i-stereo */

  1713.                     for(j=0;j<len;j++) {

  1714.                         if (tab1[j] != 0) {

  1715.                             non_zero_found_short[l] = 1;

  1716.                             goto found1;

  1717.                         }

  1718.                     }

  1719.                     sf = g1->scale_factors[k + l];

  1720.                     if (sf >= sf_max)

  1721.                         goto found1;

  1722. 

  1723.                     v1 = is_tab[0][sf];

  1724.                     v2 = is_tab[1][sf];

  1725.                     for(j=0;j<len;j++) {

  1726.                         tmp0 = tab0[j];

  1727.                         tab0[j] = MULL(tmp0, v1);

  1728.                         tab1[j] = MULL(tmp0, v2);

  1729.                     }

  1730.                 } else {

  1731.                 found1:

  1732.                     if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1733.                         /* lower part of the spectrum : do ms stereo

  1734.                            if enabled */

  1735.                         for(j=0;j<len;j++) {

  1736.                             tmp0 = tab0[j];

  1737.                             tmp1 = tab1[j];

  1738.                             tab0[j] = MULL(tmp0 + tmp1, ISQRT2);

  1739.                             tab1[j] = MULL(tmp0 - tmp1, ISQRT2);

  1740.                         }

  1741.                     }

  1742.                 }

  1743.             }

  1744.         }

  1745. 

  1746.         non_zero_found = non_zero_found_short[0] |

  1747.             non_zero_found_short[1] |

  1748.             non_zero_found_short[2];

  1749. 

  1750.         for(i = g1->long_end - 1;i >= 0;i--) {

  1751.             len = band_size_long[s->sample_rate_index][i];

  1752.             tab0 -= len;

  1753.             tab1 -= len;

  1754.             /* test if non zero band. if so, stop doing i-stereo */

  1755.             if (!non_zero_found) {

  1756.                 for(j=0;j<len;j++) {

  1757.                     if (tab1[j] != 0) {

  1758.                         non_zero_found = 1;

  1759.                         goto found2;

  1760.                     }

  1761.                 }

  1762.                 /* for last band, use previous scale factor */

  1763.                 k = (i == 21) ? 20 : i;

  1764.                 sf = g1->scale_factors[k];

  1765.                 if (sf >= sf_max)

  1766.                     goto found2;

  1767.                 v1 = is_tab[0][sf];

  1768.                 v2 = is_tab[1][sf];

  1769.                 for(j=0;j<len;j++) {

  1770.                     tmp0 = tab0[j];

  1771.                     tab0[j] = MULL(tmp0, v1);

  1772.                     tab1[j] = MULL(tmp0, v2);

  1773.                 }

  1774.             } else {

  1775.             found2:

  1776.                 if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1777.                     /* lower part of the spectrum : do ms stereo

  1778.                        if enabled */

  1779.                     for(j=0;j<len;j++) {

  1780.                         tmp0 = tab0[j];

  1781.                         tmp1 = tab1[j];

  1782.                         tab0[j] = MULL(tmp0 + tmp1, ISQRT2);

  1783.                         tab1[j] = MULL(tmp0 - tmp1, ISQRT2);

  1784.                     }

  1785.                 }

  1786.             }

  1787.         }

  1788.     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1789.         /* ms stereo ONLY */

  1790.         /* NOTE: the 1/sqrt(2) normalization factor is included in the

  1791.            global gain */

  1792.         tab0 = g0->sb_hybrid;

  1793.         tab1 = g1->sb_hybrid;

  1794.         for(i=0;i<576;i++) {

  1795.             tmp0 = tab0[i];

  1796.             tmp1 = tab1[i];

  1797.             tab0[i] = tmp0 + tmp1;

  1798.             tab1[i] = tmp0 - tmp1;

  1799.         }

  1800.     }

  1801. }

  1802. 

  1803. static void compute_antialias_integer(MPADecodeContext *s,

  1804.                               GranuleDef *g)

  1805. {

  1806.     int32_t *ptr, *csa;

  1807.     int n, i;

  1808. 

  1809.     /* we antialias only "long" bands */

  1810.     if (g->block_type == 2) {

  1811.         if (!g->switch_point)

  1812.             return;

  1813.         /* XXX: check this for 8000Hz case */

  1814.         n = 1;

  1815.     } else {

  1816.         n = SBLIMIT - 1;

  1817.     }

  1818. 

  1819.     ptr = g->sb_hybrid + 18;

  1820.     for(i = n;i > 0;i--) {

  1821.         int tmp0, tmp1, tmp2;

  1822.         csa = &csa_table[0][0];

  1823. #define INT_AA(j) \

  1824.             tmp0 = ptr[-1-j];\

  1825.             tmp1 = ptr[   j];\

  1826.             tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\

  1827.             ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\

  1828.             ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));

  1829. 

  1830.         INT_AA(0)

  1831.         INT_AA(1)

  1832.         INT_AA(2)

  1833.         INT_AA(3)

  1834.         INT_AA(4)

  1835.         INT_AA(5)

  1836.         INT_AA(6)

  1837.         INT_AA(7)

  1838. 

  1839.         ptr += 18;

  1840.     }

  1841. }

  1842. 

  1843. static void compute_antialias_float(MPADecodeContext *s,

  1844.                               GranuleDef *g)

  1845. {

  1846.     int32_t *ptr;

  1847.     int n, i;

  1848. 

  1849.     /* we antialias only "long" bands */

  1850.     if (g->block_type == 2) {

  1851.         if (!g->switch_point)

  1852.             return;

  1853.         /* XXX: check this for 8000Hz case */

  1854.         n = 1;

  1855.     } else {

  1856.         n = SBLIMIT - 1;

  1857.     }

  1858. 

  1859.     ptr = g->sb_hybrid + 18;

  1860.     for(i = n;i > 0;i--) {

  1861.         float tmp0, tmp1;

  1862.         float *csa = &csa_table_float[0][0];

  1863. #define FLOAT_AA(j)\

  1864.         tmp0= ptr[-1-j];\

  1865.         tmp1= ptr[   j];\

  1866.         ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\

  1867.         ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);

  1868. 

  1869.         FLOAT_AA(0)

  1870.         FLOAT_AA(1)

  1871.         FLOAT_AA(2)

  1872.         FLOAT_AA(3)

  1873.         FLOAT_AA(4)

  1874.         FLOAT_AA(5)

  1875.         FLOAT_AA(6)

  1876.         FLOAT_AA(7)

  1877. 

  1878.         ptr += 18;

  1879.     }

  1880. }

  1881. 

  1882. static void compute_imdct(MPADecodeContext *s,

  1883.                           GranuleDef *g,

  1884.                           int32_t *sb_samples,

  1885.                           int32_t *mdct_buf)

  1886. {

  1887.     int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;

  1888.     int32_t out2[12];

  1889.     int i, j, mdct_long_end, v, sblimit;

  1890. 

  1891.     /* find last non zero block */

  1892.     ptr = g->sb_hybrid + 576;

  1893.     ptr1 = g->sb_hybrid + 2 * 18;

  1894.     while (ptr >= ptr1) {

  1895.         ptr -= 6;

  1896.         v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];

  1897.         if (v != 0)

  1898.             break;

  1899.     }

  1900.     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

  1901. 

  1902.     if (g->block_type == 2) {

  1903.         /* XXX: check for 8000 Hz */

  1904.         if (g->switch_point)

  1905.             mdct_long_end = 2;

  1906.         else

  1907.             mdct_long_end = 0;

  1908.     } else {

  1909.         mdct_long_end = sblimit;

  1910.     }

  1911. 

  1912.     buf = mdct_buf;

  1913.     ptr = g->sb_hybrid;

  1914.     for(j=0;j<mdct_long_end;j++) {

  1915.         /* apply window & overlap with previous buffer */

  1916.         out_ptr = sb_samples + j;

  1917.         /* select window */

  1918.         if (g->switch_point && j < 2)

  1919.             win1 = mdct_win[0];

  1920.         else

  1921.             win1 = mdct_win[g->block_type];

  1922.         /* select frequency inversion */

  1923.         win = win1 + ((4 * 36) & -(j & 1));

  1924.         imdct36(out_ptr, buf, ptr, win);

  1925.         out_ptr += 18*SBLIMIT;

  1926.         ptr += 18;

  1927.         buf += 18;

  1928.     }

  1929.     for(j=mdct_long_end;j<sblimit;j++) {

  1930.         /* select frequency inversion */

  1931.         win = mdct_win[2] + ((4 * 36) & -(j & 1));

  1932.         out_ptr = sb_samples + j;

  1933. 

  1934.         for(i=0; i<6; i++){

  1935.             *out_ptr = buf[i];

  1936.             out_ptr += SBLIMIT;

  1937.         }

  1938.         imdct12(out2, ptr + 0);

  1939.         for(i=0;i<6;i++) {

  1940.             *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];

  1941.             buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);

  1942.             out_ptr += SBLIMIT;

  1943.         }

  1944.         imdct12(out2, ptr + 1);

  1945.         for(i=0;i<6;i++) {

  1946.             *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];

  1947.             buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);

  1948.             out_ptr += SBLIMIT;

  1949.         }

  1950.         imdct12(out2, ptr + 2);

  1951.         for(i=0;i<6;i++) {

  1952.             buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];

  1953.             buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);

  1954.             buf[i + 6*2] = 0;

  1955.         }

  1956.         ptr += 18;

  1957.         buf += 18;

  1958.     }

  1959.     /* zero bands */

  1960.     for(j=sblimit;j<SBLIMIT;j++) {

  1961.         /* overlap */

  1962.         out_ptr = sb_samples + j;

  1963.         for(i=0;i<18;i++) {

  1964.             *out_ptr = buf[i];

  1965.             buf[i] = 0;

  1966.             out_ptr += SBLIMIT;

  1967.         }

  1968.         buf += 18;

  1969.     }

  1970. }

  1971. 

  1972. #if defined(DEBUG)

  1973. void sample_dump(int fnum, int32_t *tab, int n)

  1974. {

  1975.     static FILE *files[16], *f;

  1976.     char buf[512];

  1977.     int i;

  1978.     int32_t v;

  1979. 

  1980.     f = files[fnum];

  1981.     if (!f) {

  1982.         snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm",

  1983.                 fnum,

  1984. #ifdef USE_HIGHPRECISION

  1985.                 "hp"

  1986. #else

  1987.                 "lp"

  1988. #endif

  1989.                 );

  1990.         f = fopen(buf, "w");

  1991.         if (!f)

  1992.             return;

  1993.         files[fnum] = f;

  1994.     }

  1995. 

  1996.     if (fnum == 0) {

  1997.         static int pos = 0;

  1998.         av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);

  1999.         for(i=0;i<n;i++) {

  2000.             av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);

  2001.             if ((i % 18) == 17)

  2002.                 av_log(NULL, AV_LOG_DEBUG, "\n");

  2003.         }

  2004.         pos += n;

  2005.     }

  2006.     for(i=0;i<n;i++) {

  2007.         /* normalize to 23 frac bits */

  2008.         v = tab[i] << (23 - FRAC_BITS);

  2009.         fwrite(&v, 1, sizeof(int32_t), f);

  2010.     }

  2011. }

  2012. #endif

  2013. 

  2014. 

  2015. /* main layer3 decoding function */

  2016. static int mp_decode_layer3(MPADecodeContext *s)

  2017. {

  2018.     int nb_granules, main_data_begin, private_bits;

  2019.     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;

  2020.     GranuleDef granules[2][2], *g;

  2021.     int16_t exponents[576];

  2022. 

  2023.     /* read side info */

  2024.     if (s->lsf) {

  2025.         main_data_begin = get_bits(&s->gb, 8);

  2026.         private_bits = get_bits(&s->gb, s->nb_channels);

  2027.         nb_granules = 1;

  2028.     } else {

  2029.         main_data_begin = get_bits(&s->gb, 9);

  2030.         if (s->nb_channels == 2)

  2031.             private_bits = get_bits(&s->gb, 3);

  2032.         else

  2033.             private_bits = get_bits(&s->gb, 5);

  2034.         nb_granules = 2;

  2035.         for(ch=0;ch<s->nb_channels;ch++) {

  2036.             granules[ch][0].scfsi = 0; /* all scale factors are transmitted */

  2037.             granules[ch][1].scfsi = get_bits(&s->gb, 4);

  2038.         }

  2039.     }

  2040. 

  2041.     for(gr=0;gr<nb_granules;gr++) {

  2042.         for(ch=0;ch<s->nb_channels;ch++) {

  2043.             dprintf(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);

  2044.             g = &granules[ch][gr];

  2045.             g->part2_3_length = get_bits(&s->gb, 12);

  2046.             g->big_values = get_bits(&s->gb, 9);

  2047.             if(g->big_values > 288){

  2048.                 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");

  2049.                 return -1;

  2050.             }

  2051. 

  2052.             g->global_gain = get_bits(&s->gb, 8);

  2053.             /* if MS stereo only is selected, we precompute the

  2054.                1/sqrt(2) renormalization factor */

  2055.             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==

  2056.                 MODE_EXT_MS_STEREO)

  2057.                 g->global_gain -= 2;

  2058.             if (s->lsf)

  2059.                 g->scalefac_compress = get_bits(&s->gb, 9);

  2060.             else

  2061.                 g->scalefac_compress = get_bits(&s->gb, 4);

  2062.             blocksplit_flag = get_bits1(&s->gb);

  2063.             if (blocksplit_flag) {

  2064.                 g->block_type = get_bits(&s->gb, 2);

  2065.                 if (g->block_type == 0){

  2066.                     av_log(NULL, AV_LOG_ERROR, "invalid block type\n");

  2067.                     return -1;

  2068.                 }

  2069.                 g->switch_point = get_bits1(&s->gb);

  2070.                 for(i=0;i<2;i++)

  2071.                     g->table_select[i] = get_bits(&s->gb, 5);

  2072.                 for(i=0;i<3;i++)

  2073.                     g->subblock_gain[i] = get_bits(&s->gb, 3);

  2074.                 ff_init_short_region(s, g);

  2075.             } else {

  2076.                 int region_address1, region_address2;

  2077.                 g->block_type = 0;

  2078.                 g->switch_point = 0;

  2079.                 for(i=0;i<3;i++)

  2080.                     g->table_select[i] = get_bits(&s->gb, 5);

  2081.                 /* compute huffman coded region sizes */

  2082.                 region_address1 = get_bits(&s->gb, 4);

  2083.                 region_address2 = get_bits(&s->gb, 3);

  2084.                 dprintf(s->avctx, "region1=%d region2=%d\n",

  2085.                         region_address1, region_address2);

  2086.                 ff_init_long_region(s, g, region_address1, region_address2);

  2087.             }

  2088.             ff_region_offset2size(g);

  2089.             ff_compute_band_indexes(s, g);

  2090. 

  2091.             g->preflag = 0;

  2092.             if (!s->lsf)

  2093.                 g->preflag = get_bits1(&s->gb);

  2094.             g->scalefac_scale = get_bits1(&s->gb);

  2095.             g->count1table_select = get_bits1(&s->gb);

  2096.             dprintf(s->avctx, "block_type=%d switch_point=%d\n",

  2097.                     g->block_type, g->switch_point);

  2098.         }

  2099.     }

  2100. 

  2101.   if (!s->adu_mode) {

  2102.     const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);

  2103.     assert((get_bits_count(&s->gb) & 7) == 0);

  2104.     /* now we get bits from the main_data_begin offset */

  2105.     dprintf(s->avctx, "seekback: %d\n", main_data_begin);

  2106. //av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);

  2107. 

  2108.     memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);

  2109.     s->in_gb= s->gb;

  2110.         init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);

  2111.         skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));

  2112.   }

  2113. 

  2114.     for(gr=0;gr<nb_granules;gr++) {

  2115.         for(ch=0;ch<s->nb_channels;ch++) {

  2116.             g = &granules[ch][gr];

  2117.             if(get_bits_count(&s->gb)<0){

  2118.                 av_log(NULL, AV_LOG_ERROR, "mdb:%d, lastbuf:%d skipping granule %d\n",

  2119.                                             main_data_begin, s->last_buf_size, gr);

  2120.                 skip_bits_long(&s->gb, g->part2_3_length);

  2121.                 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));

  2122.                 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){

  2123.                     skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);

  2124.                     s->gb= s->in_gb;

  2125.                     s->in_gb.buffer=NULL;

  2126.                 }

  2127.                 continue;

  2128.             }

  2129. 

  2130.             bits_pos = get_bits_count(&s->gb);

  2131. 

  2132.             if (!s->lsf) {

  2133.                 uint8_t *sc;

  2134.                 int slen, slen1, slen2;

  2135. 

  2136.                 /* MPEG1 scale factors */

  2137.                 slen1 = slen_table[0][g->scalefac_compress];

  2138.                 slen2 = slen_table[1][g->scalefac_compress];

  2139.                 dprintf(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);

  2140.                 if (g->block_type == 2) {

  2141.                     n = g->switch_point ? 17 : 18;

  2142.                     j = 0;

  2143.                     if(slen1){

  2144.                         for(i=0;i<n;i++)

  2145.                             g->scale_factors[j++] = get_bits(&s->gb, slen1);

  2146.                     }else{

  2147.                         for(i=0;i<n;i++)

  2148.                             g->scale_factors[j++] = 0;

  2149.                     }

  2150.                     if(slen2){

  2151.                         for(i=0;i<18;i++)

  2152.                             g->scale_factors[j++] = get_bits(&s->gb, slen2);

  2153.                         for(i=0;i<3;i++)

  2154.                             g->scale_factors[j++] = 0;

  2155.                     }else{

  2156.                         for(i=0;i<21;i++)

  2157.                             g->scale_factors[j++] = 0;

  2158.                     }

  2159.                 } else {

  2160.                     sc = granules[ch][0].scale_factors;

  2161.                     j = 0;

  2162.                     for(k=0;k<4;k++) {

  2163.                         n = (k == 0 ? 6 : 5);

  2164.                         if ((g->scfsi & (0x8 >> k)) == 0) {

  2165.                             slen = (k < 2) ? slen1 : slen2;

  2166.                             if(slen){

  2167.                                 for(i=0;i<n;i++)

  2168.                                     g->scale_factors[j++] = get_bits(&s->gb, slen);

  2169.                             }else{

  2170.                                 for(i=0;i<n;i++)

  2171.                                     g->scale_factors[j++] = 0;

  2172.                             }

  2173.                         } else {

  2174.                             /* simply copy from last granule */

  2175.                             for(i=0;i<n;i++) {

  2176.                                 g->scale_factors[j] = sc[j];

  2177.                                 j++;

  2178.                             }

  2179.                         }

  2180.                     }

  2181.                     g->scale_factors[j++] = 0;

  2182.                 }

  2183. #if defined(DEBUG)

  2184.                 {

  2185.                     dprintf(s->avctx, "scfsi=%x gr=%d ch=%d scale_factors:\n",

  2186.                            g->scfsi, gr, ch);

  2187.                     for(i=0;i<j;i++)

  2188.                         dprintf(s->avctx, " %d", g->scale_factors[i]);

  2189.                     dprintf(s->avctx, "\n");

  2190.                 }

  2191. #endif

  2192.             } else {

  2193.                 int tindex, tindex2, slen[4], sl, sf;

  2194. 

  2195.                 /* LSF scale factors */

  2196.                 if (g->block_type == 2) {

  2197.                     tindex = g->switch_point ? 2 : 1;

  2198.                 } else {

  2199.                     tindex = 0;

  2200.                 }

  2201.                 sf = g->scalefac_compress;

  2202.                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {

  2203.                     /* intensity stereo case */

  2204.                     sf >>= 1;

  2205.                     if (sf < 180) {

  2206.                         lsf_sf_expand(slen, sf, 6, 6, 0);

  2207.                         tindex2 = 3;

  2208.                     } else if (sf < 244) {

  2209.                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);

  2210.                         tindex2 = 4;

  2211.                     } else {

  2212.                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);

  2213.                         tindex2 = 5;

  2214.                     }

  2215.                 } else {

  2216.                     /* normal case */

  2217.                     if (sf < 400) {

  2218.                         lsf_sf_expand(slen, sf, 5, 4, 4);

  2219.                         tindex2 = 0;

  2220.                     } else if (sf < 500) {

  2221.                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);

  2222.                         tindex2 = 1;

  2223.                     } else {

  2224.                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);

  2225.                         tindex2 = 2;

  2226.                         g->preflag = 1;

  2227.                     }

  2228.                 }

  2229. 

  2230.                 j = 0;

  2231.                 for(k=0;k<4;k++) {

  2232.                     n = lsf_nsf_table[tindex2][tindex][k];

  2233.                     sl = slen[k];

  2234.                     if(sl){

  2235.                         for(i=0;i<n;i++)

  2236.                             g->scale_factors[j++] = get_bits(&s->gb, sl);

  2237.                     }else{

  2238.                         for(i=0;i<n;i++)

  2239.                             g->scale_factors[j++] = 0;

  2240.                     }

  2241.                 }

  2242.                 /* XXX: should compute exact size */

  2243.                 for(;j<40;j++)

  2244.                     g->scale_factors[j] = 0;

  2245. #if defined(DEBUG)

  2246.                 {

  2247.                     dprintf(s->avctx, "gr=%d ch=%d scale_factors:\n",

  2248.                            gr, ch);

  2249.                     for(i=0;i<40;i++)

  2250.                         dprintf(s->avctx, " %d", g->scale_factors[i]);

  2251.                     dprintf(s->avctx, "\n");

  2252.                 }

  2253. #endif

  2254.             }

  2255. 

  2256.             exponents_from_scale_factors(s, g, exponents);

  2257. 

  2258.             /* read Huffman coded residue */

  2259.             huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);

  2260. #if defined(DEBUG)

  2261.             sample_dump(0, g->sb_hybrid, 576);

  2262. #endif

  2263.         } /* ch */

  2264. 

  2265.         if (s->nb_channels == 2)

  2266.             compute_stereo(s, &granules[0][gr], &granules[1][gr]);

  2267. 

  2268.         for(ch=0;ch<s->nb_channels;ch++) {

  2269.             g = &granules[ch][gr];

  2270. 

  2271.             reorder_block(s, g);

  2272. #if defined(DEBUG)

  2273.             sample_dump(0, g->sb_hybrid, 576);

  2274. #endif

  2275.             s->compute_antialias(s, g);

  2276. #if defined(DEBUG)

  2277.             sample_dump(1, g->sb_hybrid, 576);

  2278. #endif

  2279.             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);

  2280. #if defined(DEBUG)

  2281.             sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);

  2282. #endif

  2283.         }

  2284.     } /* gr */

  2285.     if(get_bits_count(&s->gb)<0)

  2286.         skip_bits_long(&s->gb, -get_bits_count(&s->gb));

  2287.     return nb_granules * 18;

  2288. }

  2289. 

  2290. static int mp_decode_frame(MPADecodeContext *s,

  2291.                            OUT_INT *samples, const uint8_t *buf, int buf_size)

  2292. {

  2293.     int i, nb_frames, ch;

  2294.     OUT_INT *samples_ptr;

  2295. 

  2296.     init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);

  2297. 

  2298.     /* skip error protection field */

  2299.     if (s->error_protection)

  2300.         skip_bits(&s->gb, 16);

  2301. 

  2302.     dprintf(s->avctx, "frame %d:\n", s->frame_count);

  2303.     switch(s->layer) {

  2304.     case 1:

  2305.         s->avctx->frame_size = 384;

  2306.         nb_frames = mp_decode_layer1(s);

  2307.         break;

  2308.     case 2:

  2309.         s->avctx->frame_size = 1152;

  2310.         nb_frames = mp_decode_layer2(s);

  2311.         break;

  2312.     case 3:

  2313.         s->avctx->frame_size = s->lsf ? 576 : 1152;

  2314.     default:

  2315.         nb_frames = mp_decode_layer3(s);

  2316. 

  2317.         s->last_buf_size=0;

  2318.         if(s->in_gb.buffer){

  2319.             align_get_bits(&s->gb);

  2320.             i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;

  2321.             if(i >= 0 && i <= BACKSTEP_SIZE){

  2322.                 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);

  2323.                 s->last_buf_size=i;

  2324.             }else

  2325.                 av_log(NULL, AV_LOG_ERROR, "invalid old backstep %d\n", i);

  2326.             s->gb= s->in_gb;

  2327.             s->in_gb.buffer= NULL;

  2328.         }

  2329. 

  2330.         align_get_bits(&s->gb);

  2331.         assert((get_bits_count(&s->gb) & 7) == 0);

  2332.         i= (s->gb.size_in_bits - get_bits_count(&s->gb))>>3;

  2333. 

  2334.         if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){

  2335.             av_log(NULL, AV_LOG_ERROR, "invalid new backstep %d\n", i);

  2336.             i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);

  2337.         }

  2338.         assert(i <= buf_size - HEADER_SIZE && i>= 0);

  2339.         memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);

  2340.         s->last_buf_size += i;

  2341. 

  2342.         break;

  2343.     }

  2344. #if defined(DEBUG)

  2345.     for(i=0;i<nb_frames;i++) {

  2346.         for(ch=0;ch<s->nb_channels;ch++) {

  2347.             int j;

  2348.             dprintf(s->avctx, "%d-%d:", i, ch);

  2349.             for(j=0;j<SBLIMIT;j++)

  2350.                 dprintf(s->avctx, " %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);

  2351.             dprintf(s->avctx, "\n");

  2352.         }

  2353.     }

  2354. #endif

  2355.     /* apply the synthesis filter */

  2356.     for(ch=0;ch<s->nb_channels;ch++) {

  2357.         samples_ptr = samples + ch;

  2358.         for(i=0;i<nb_frames;i++) {

  2359.             ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),

  2360.                          window, &s->dither_state,

  2361.                          samples_ptr, s->nb_channels,

  2362.                          s->sb_samples[ch][i]);

  2363.             samples_ptr += 32 * s->nb_channels;

  2364.         }

  2365.     }

  2366. #ifdef DEBUG

  2367.     s->frame_count++;

  2368. #endif

  2369.     return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;

  2370. }

  2371. 

  2372. static int decode_frame(AVCodecContext * avctx,

  2373.                         void *data, int *data_size,

  2374.                         const uint8_t * buf, int buf_size)

  2375. {

  2376.     MPADecodeContext *s = avctx->priv_data;

  2377.     uint32_t header;

  2378.     int out_size;

  2379.     OUT_INT *out_samples = data;

  2380. 

  2381. retry:

  2382.     if(buf_size < HEADER_SIZE)

  2383.         return -1;

  2384. 

  2385.     header = AV_RB32(buf);

  2386.     if(ff_mpa_check_header(header) < 0){

  2387.         buf++;

  2388. //        buf_size--;

  2389.         av_log(avctx, AV_LOG_ERROR, "Header missing skipping one byte.\n");

  2390.         goto retry;

  2391.     }

  2392. 

  2393.     if (ff_mpegaudio_decode_header(s, header) == 1) {

  2394.         /* free format: prepare to compute frame size */

  2395.         s->frame_size = -1;

  2396.         return -1;

  2397.     }

  2398.     /* update codec info */

  2399.     avctx->channels = s->nb_channels;

  2400.     avctx->bit_rate = s->bit_rate;

  2401.     avctx->sub_id = s->layer;

  2402. 

  2403.     if(s->frame_size<=0 || s->frame_size > buf_size){

  2404.         av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");

  2405.         return -1;

  2406.     }else if(s->frame_size < buf_size){

  2407.         av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");

  2408.         buf_size= s->frame_size;

  2409.     }

  2410. 

  2411.     out_size = mp_decode_frame(s, out_samples, buf, buf_size);

  2412.     if(out_size>=0){

  2413.         *data_size = out_size;

  2414.         avctx->sample_rate = s->sample_rate;

  2415.         //FIXME maybe move the other codec info stuff from above here too

  2416.     }else

  2417.         av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed

  2418.     s->frame_size = 0;

  2419.     return buf_size;

  2420. }

  2421. 

  2422. static void flush(AVCodecContext *avctx){

  2423.     MPADecodeContext *s = avctx->priv_data;

  2424.     memset(s->synth_buf, 0, sizeof(s->synth_buf));

  2425.     s->last_buf_size= 0;

  2426. }

  2427. 

  2428. #ifdef CONFIG_MP3ADU_DECODER

  2429. static int decode_frame_adu(AVCodecContext * avctx,

  2430.                         void *data, int *data_size,

  2431.                         const uint8_t * buf, int buf_size)

  2432. {

  2433.     MPADecodeContext *s = avctx->priv_data;

  2434.     uint32_t header;

  2435.     int len, out_size;

  2436.     OUT_INT *out_samples = data;

  2437. 

  2438.     len = buf_size;

  2439. 

  2440.     // Discard too short frames

  2441.     if (buf_size < HEADER_SIZE) {

  2442.         *data_size = 0;

  2443.         return buf_size;

  2444.     }

  2445. 

  2446. 

  2447.     if (len > MPA_MAX_CODED_FRAME_SIZE)

  2448.         len = MPA_MAX_CODED_FRAME_SIZE;

  2449. 

  2450.     // Get header and restore sync word

  2451.     header = AV_RB32(buf) | 0xffe00000;

  2452. 

  2453.     if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame

  2454.         *data_size = 0;

  2455.         return buf_size;

  2456.     }

  2457. 

  2458.     ff_mpegaudio_decode_header(s, header);

  2459.     /* update codec info */

  2460.     avctx->sample_rate = s->sample_rate;

  2461.     avctx->channels = s->nb_channels;

  2462.     avctx->bit_rate = s->bit_rate;

  2463.     avctx->sub_id = s->layer;

  2464. 

  2465.     s->frame_size = len;

  2466. 

  2467.     if (avctx->parse_only) {

  2468.         out_size = buf_size;

  2469.     } else {

  2470.         out_size = mp_decode_frame(s, out_samples, buf, buf_size);

  2471.     }

  2472. 

  2473.     *data_size = out_size;

  2474.     return buf_size;

  2475. }

  2476. #endif /* CONFIG_MP3ADU_DECODER */

  2477. 

  2478. #ifdef CONFIG_MP3ON4_DECODER

  2479. 

  2480. #include "mpeg4audio.h"

  2481. 

  2482. /* Next 3 arrays are indexed by channel config number (passed via codecdata) */

  2483. static const uint8_t mp3Frames[8] = {0,1,1,2,3,3,4,5};   /* number of mp3 decoder instances */

  2484. /* offsets into output buffer, assume output order is FL FR BL BR C LFE */

  2485. static const uint8_t chan_offset[8][5] = {

  2486.     {0},

  2487.     {0},            // C

  2488.     {0},            // FLR

  2489.     {2,0},          // C FLR

  2490.     {2,0,3},        // C FLR BS

  2491.     {4,0,2},        // C FLR BLRS

  2492.     {4,0,2,5},      // C FLR BLRS LFE

  2493.     {4,0,2,6,5},    // C FLR BLRS BLR LFE

  2494. };

  2495. 

  2496. 

  2497. static int decode_init_mp3on4(AVCodecContext * avctx)

  2498. {

  2499.     MP3On4DecodeContext *s = avctx->priv_data;

  2500.     MPEG4AudioConfig cfg;

  2501.     int i;

  2502. 

  2503.     if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {

  2504.         av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");

  2505.         return -1;

  2506.     }

  2507. 

  2508.     ff_mpeg4audio_get_config(&cfg, avctx->extradata, avctx->extradata_size);

  2509.     if (!cfg.chan_config || cfg.chan_config > 7) {

  2510.         av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");

  2511.         return -1;

  2512.     }

  2513.     s->chan_cfg = cfg.chan_config;

  2514.     s->frames = mp3Frames[s->chan_cfg];

  2515.     avctx->channels = ff_mpeg4audio_channels[s->chan_cfg];

  2516. 

  2517.     if (cfg.sample_rate < 16000)

  2518.         s->syncword = 0xffe00000;

  2519.     else

  2520.         s->syncword = 0xfff00000;

  2521. 

  2522.     /* Init the first mp3 decoder in standard way, so that all tables get builded

  2523.      * We replace avctx->priv_data with the context of the first decoder so that

  2524.      * decode_init() does not have to be changed.

  2525.      * Other decoders will be initialized here copying data from the first context

  2526.      */

  2527.     // Allocate zeroed memory for the first decoder context

  2528.     s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));

  2529.     // Put decoder context in place to make init_decode() happy

  2530.     avctx->priv_data = s->mp3decctx[0];

  2531.     decode_init(avctx);

  2532.     // Restore mp3on4 context pointer

  2533.     avctx->priv_data = s;

  2534.     s->mp3decctx[0]->adu_mode = 1; // Set adu mode

  2535. 

  2536.     /* Create a separate codec/context for each frame (first is already ok).

  2537.      * Each frame is 1 or 2 channels - up to 5 frames allowed

  2538.      */

  2539.     for (i = 1; i < s->frames; i++) {

  2540.         s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));

  2541.         s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;

  2542.         s->mp3decctx[i]->adu_mode = 1;

  2543.         s->mp3decctx[i]->avctx = avctx;

  2544.     }

  2545. 

  2546.     return 0;

  2547. }

  2548. 

  2549. 

  2550. static int decode_close_mp3on4(AVCodecContext * avctx)

  2551. {

  2552.     MP3On4DecodeContext *s = avctx->priv_data;

  2553.     int i;

  2554. 

  2555.     for (i = 0; i < s->frames; i++)

  2556.         if (s->mp3decctx[i])

  2557.             av_free(s->mp3decctx[i]);

  2558. 

  2559.     return 0;

  2560. }

  2561. 

  2562. 

  2563. static int decode_frame_mp3on4(AVCodecContext * avctx,

  2564.                         void *data, int *data_size,

  2565.                         const uint8_t * buf, int buf_size)

  2566. {

  2567.     MP3On4DecodeContext *s = avctx->priv_data;

  2568.     MPADecodeContext *m;

  2569.     int len, out_size = 0;

  2570.     uint32_t header;

  2571.     OUT_INT *out_samples = data;

  2572.     OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];

  2573.     OUT_INT *outptr, *bp;

  2574.     int fsize;

  2575.     int fr, i, j, n;

  2576.     int off = avctx->channels;

  2577.     const uint8_t *coff = chan_offset[s->chan_cfg];

  2578. 

  2579.     len = buf_size;

  2580. 

  2581.     *data_size = 0;

  2582.     // Discard too short frames

  2583.     if (buf_size < HEADER_SIZE)

  2584.         return -1;

  2585. 

  2586.     // If only one decoder interleave is not needed

  2587.     outptr = s->frames == 1 ? out_samples : decoded_buf;

  2588. 

  2589.     for (fr = 0; fr < s->frames; fr++) {

  2590.         fsize = AV_RB16(buf) >> 4;

  2591.         fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);

  2592.         m = s->mp3decctx[fr];

  2593.         assert (m != NULL);

  2594. 

  2595.         header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header

  2596. 

  2597.         if (ff_mpa_check_header(header) < 0) { // Bad header, discard block

  2598.             *data_size = 0;

  2599.             return buf_size;

  2600.         }

  2601. 

  2602.         ff_mpegaudio_decode_header(m, header);

  2603.         out_size += mp_decode_frame(m, outptr, buf, fsize);

  2604.         buf += fsize;

  2605.         len -= fsize;

  2606. 

  2607.         if(s->frames > 1) {

  2608.             n = m->avctx->frame_size*m->nb_channels;

  2609.             /* interleave output data */

  2610.             bp = out_samples + coff[fr];

  2611.             if(m->nb_channels == 1) {

  2612.                 for(j = 0; j < n; j++) {

  2613.                     *bp = decoded_buf[j];

  2614.                     bp += off;

  2615.                 }

  2616.             } else {

  2617.                 for(j = 0; j < n; j++) {

  2618.                     bp[0] = decoded_buf[j++];

  2619.                     bp[1] = decoded_buf[j];

  2620.                     bp += off;

  2621.                 }

  2622.             }

  2623.         }

  2624.     }

  2625. 

  2626.     /* update codec info */

  2627.     avctx->sample_rate = s->mp3decctx[0]->sample_rate;

  2628.     avctx->bit_rate = 0;

  2629.     for (i = 0; i < s->frames; i++)

  2630.         avctx->bit_rate += s->mp3decctx[i]->bit_rate;

  2631. 

  2632.     *data_size = out_size;

  2633.     return buf_size;

  2634. }

  2635. #endif /* CONFIG_MP3ON4_DECODER */

  2636. 

  2637. #ifdef CONFIG_MP2_DECODER

  2638. AVCodec mp2_decoder =

  2639. {

  2640.     "mp2",

  2641.     CODEC_TYPE_AUDIO,

  2642.     CODEC_ID_MP2,

  2643.     sizeof(MPADecodeContext),

  2644.     decode_init,

  2645.     NULL,

  2646.     NULL,

  2647.     decode_frame,

  2648.     CODEC_CAP_PARSE_ONLY,

  2649.     .flush= flush,

  2650. };

  2651. #endif

  2652. #ifdef CONFIG_MP3_DECODER

  2653. AVCodec mp3_decoder =

  2654. {

  2655.     "mp3",

  2656.     CODEC_TYPE_AUDIO,

  2657.     CODEC_ID_MP3,

  2658.     sizeof(MPADecodeContext),

  2659.     decode_init,

  2660.     NULL,

  2661.     NULL,

  2662.     decode_frame,

  2663.     CODEC_CAP_PARSE_ONLY,

  2664.     .flush= flush,

  2665. };

  2666. #endif

  2667. #ifdef CONFIG_MP3ADU_DECODER

  2668. AVCodec mp3adu_decoder =

  2669. {

  2670.     "mp3adu",

  2671.     CODEC_TYPE_AUDIO,

  2672.     CODEC_ID_MP3ADU,

  2673.     sizeof(MPADecodeContext),

  2674.     decode_init,

  2675.     NULL,

  2676.     NULL,

  2677.     decode_frame_adu,

  2678.     CODEC_CAP_PARSE_ONLY,

  2679.     .flush= flush,

  2680. };

  2681. #endif

  2682. #ifdef CONFIG_MP3ON4_DECODER

  2683. AVCodec mp3on4_decoder =

  2684. {

  2685.     "mp3on4",

  2686.     CODEC_TYPE_AUDIO,

  2687.     CODEC_ID_MP3ON4,

  2688.     sizeof(MP3On4DecodeContext),

  2689.     decode_init_mp3on4,

  2690.     NULL,

  2691.     decode_close_mp3on4,

  2692.     decode_frame_mp3on4,

  2693.     .flush= flush,

  2694. };

  2695. #endif