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.
4190 Commits
32 Branches
522MB
Tree: 2ce151f898
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '2ce151f898'
${ noResults }
FFmpeg/libavcodec/mpegaudiodec.c

2779 lines
82KB
Raw Blame History 

  1. /*

  2.  * MPEG Audio decoder

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

  4.  *

  5.  * This library is free software; you can redistribute it and/or

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

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

  8.  * version 2 of the License, or (at your option) any later version.

  9.  *

  10.  * This library is distributed in the hope that it will be useful,

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

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

  13.  * Lesser General Public License for more details.

  14.  *

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

  16.  * License along with this library; if not, write to the Free Software

  17.  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

  18.  */

  19. 

  20. /**

  21.  * @file mpegaudiodec.c

  22.  * MPEG Audio decoder.

  23.  */ 

  24. 

  25. //#define DEBUG

  26. #include "avcodec.h"

  27. #include "bitstream.h"

  28. #include "mpegaudio.h"

  29. #include "dsputil.h"

  30. 

  31. /*

  32.  * TODO:

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

  34.  *  - test lsf / mpeg25 extensively.

  35.  */

  36. 

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

  38.    audio decoder */

  39. #ifdef CONFIG_MPEGAUDIO_HP

  40. #define USE_HIGHPRECISION

  41. #endif

  42. 

  43. #ifdef USE_HIGHPRECISION

  44. #define FRAC_BITS   23   /* fractional bits for sb_samples and dct */

  45. #define WFRAC_BITS  16   /* fractional bits for window */

  46. #else

  47. #define FRAC_BITS   15   /* fractional bits for sb_samples and dct */

  48. #define WFRAC_BITS  14   /* fractional bits for window */

  49. #endif

  50. 

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

  52. typedef int32_t OUT_INT;

  53. #define OUT_MAX INT32_MAX

  54. #define OUT_MIN INT32_MIN

  55. #define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 31)

  56. #else

  57. typedef int16_t OUT_INT;

  58. #define OUT_MAX INT16_MAX

  59. #define OUT_MIN INT16_MIN

  60. #define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)

  61. #endif

  62. 

  63. #define FRAC_ONE    (1 << FRAC_BITS)

  64. 

  65. #define MULL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)

  66. #define MUL64(a,b) ((int64_t)(a) * (int64_t)(b))

  67. #define FIX(a)   ((int)((a) * FRAC_ONE))

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

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

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

  71. 

  72. #if FRAC_BITS <= 15

  73. typedef int16_t MPA_INT;

  74. #else

  75. typedef int32_t MPA_INT;

  76. #endif

  77. 

  78. /****************/

  79. 

  80. #define HEADER_SIZE 4

  81. #define BACKSTEP_SIZE 512

  82. 

  83. struct GranuleDef;

  84. 

  85. typedef struct MPADecodeContext {

  86.     uint8_t inbuf1[2][MPA_MAX_CODED_FRAME_SIZE + BACKSTEP_SIZE];	/* input buffer */

  87.     int inbuf_index;

  88.     uint8_t *inbuf_ptr, *inbuf;

  89.     int frame_size;

  90.     int free_format_frame_size; /* frame size in case of free format

  91.                                    (zero if currently unknown) */

  92.     /* next header (used in free format parsing) */

  93.     uint32_t free_format_next_header; 

  94.     int error_protection;

  95.     int layer;

  96.     int sample_rate;

  97.     int sample_rate_index; /* between 0 and 8 */

  98.     int bit_rate;

  99.     int old_frame_size;

  100.     GetBitContext gb;

  101.     int nb_channels;

  102.     int mode;

  103.     int mode_ext;

  104.     int lsf;

  105.     MPA_INT synth_buf[MPA_MAX_CHANNELS][512 * 2] __attribute__((aligned(16)));

  106.     int synth_buf_offset[MPA_MAX_CHANNELS];

  107.     int32_t sb_samples[MPA_MAX_CHANNELS][36][SBLIMIT] __attribute__((aligned(16)));

  108.     int32_t mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */

  109. #ifdef DEBUG

  110.     int frame_count;

  111. #endif

  112.     void (*compute_antialias)(struct MPADecodeContext *s, struct GranuleDef *g);

  113.     int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3

  114.     unsigned int dither_state;

  115. } MPADecodeContext;

  116. 

  117. /* layer 3 "granule" */

  118. typedef struct GranuleDef {

  119.     uint8_t scfsi;

  120.     int part2_3_length;

  121.     int big_values;

  122.     int global_gain;

  123.     int scalefac_compress;

  124.     uint8_t block_type;

  125.     uint8_t switch_point;

  126.     int table_select[3];

  127.     int subblock_gain[3];

  128.     uint8_t scalefac_scale;

  129.     uint8_t count1table_select;

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

  131.     int preflag;

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

  133.     uint8_t scale_factors[40];

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

  135. } GranuleDef;

  136. 

  137. #define MODE_EXT_MS_STEREO 2

  138. #define MODE_EXT_I_STEREO  1

  139. 

  140. /* layer 3 huffman tables */

  141. typedef struct HuffTable {

  142.     int xsize;

  143.     const uint8_t *bits;

  144.     const uint16_t *codes;

  145. } HuffTable;

  146. 

  147. #include "mpegaudiodectab.h"

  148. 

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

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

  151. 

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

  153. static VLC huff_vlc[16]; 

  154. static uint8_t *huff_code_table[16];

  155. static VLC huff_quad_vlc[2];

  156. /* computed from band_size_long */

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

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

  159. #define TABLE_4_3_SIZE (8191 + 16)

  160. static int8_t  *table_4_3_exp;

  161. #if FRAC_BITS <= 15

  162. static uint16_t *table_4_3_value;

  163. #else

  164. static uint32_t *table_4_3_value;

  165. #endif

  166. /* intensity stereo coef table */

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

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

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

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

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

  172. 

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

  174. static uint16_t scale_factor_modshift[64];

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

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

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

  178. 

  179. #define SCALE_GEN(v) \

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

  181. 

  182. static int32_t scale_factor_mult2[3][3] = {

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

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

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

  186. };

  187. 

  188. /* 2^(n/4) */

  189. static uint32_t scale_factor_mult3[4] = {

  190.     FIXR(1.0),

  191.     FIXR(1.18920711500272106671),

  192.     FIXR(1.41421356237309504880),

  193.     FIXR(1.68179283050742908605),

  194. };

  195. 

  196. void ff_mpa_synth_init(MPA_INT *window);

  197. static MPA_INT window[512] __attribute__((aligned(16)));

  198.     

  199. /* layer 1 unscaling */

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

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

  202. {

  203.     int shift, mod;

  204.     int64_t val;

  205. 

  206.     shift = scale_factor_modshift[scale_factor];

  207.     mod = shift & 3;

  208.     shift >>= 2;

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

  210.     shift += n;

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

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

  213. }

  214. 

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

  216. {

  217.     int shift, mod, val;

  218. 

  219.     shift = scale_factor_modshift[scale_factor];

  220.     mod = shift & 3;

  221.     shift >>= 2;

  222. 

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

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

  225.     if (shift > 0)

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

  227.     return val;

  228. }

  229. 

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

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

  232. {

  233. #if FRAC_BITS <= 15    

  234.     unsigned int m;

  235. #else

  236.     uint64_t m;

  237. #endif

  238.     int e;

  239. 

  240.     e = table_4_3_exp[value];

  241.     e += (exponent >> 2);

  242.     e = FRAC_BITS - e;

  243. #if FRAC_BITS <= 15    

  244.     if (e > 31)

  245. #else

  246.     if (e > 63)

  247. #endif

  248.         return 0;

  249.     m = table_4_3_value[value];

  250. #if FRAC_BITS <= 15    

  251.     m = (m * scale_factor_mult3[exponent & 3]);

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

  253.     return m;

  254. #else

  255.     m = MUL64(m, scale_factor_mult3[exponent & 3]);

  256.     m = (m + (uint64_t_C(1) << (e-1))) >> e;

  257.     return m;

  258. #endif

  259. }

  260. 

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

  262. #define DEV_ORDER 13

  263. 

  264. #define POW_FRAC_BITS 24

  265. #define POW_FRAC_ONE    (1 << POW_FRAC_BITS)

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

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

  268. 

  269. static int dev_4_3_coefs[DEV_ORDER];

  270. 

  271. static int pow_mult3[3] = {

  272.     POW_FIX(1.0),

  273.     POW_FIX(1.25992104989487316476),

  274.     POW_FIX(1.58740105196819947474),

  275. };

  276. 

  277. static void int_pow_init(void)

  278. {

  279.     int i, a;

  280. 

  281.     a = POW_FIX(1.0);

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

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

  284.         dev_4_3_coefs[i] = a;

  285.     }

  286. }

  287. 

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

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

  290. {

  291.     int e, er, eq, j;

  292.     int a, a1;

  293.     

  294.     /* renormalize */

  295.     a = i;

  296.     e = POW_FRAC_BITS;

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

  298.         a = a << 1;

  299.         e--;

  300.     }

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

  302.     a1 = 0;

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

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

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

  306.     /* exponent compute (exact) */

  307.     e = e * 4;

  308.     er = e % 3;

  309.     eq = e / 3;

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

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

  312.         a = a >> 1;

  313.         eq++;

  314.     }

  315.     /* convert to float */

  316.     while (a < POW_FRAC_ONE) {

  317.         a = a << 1;

  318.         eq--;

  319.     }

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

  321. #if POW_FRAC_BITS > FRAC_BITS

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

  323.     /* correct overflow */

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

  325.         a = a >> 1;

  326.         eq++;

  327.     }

  328. #endif

  329.     *exp_ptr = eq;

  330.     return a;

  331. }

  332. 

  333. static int decode_init(AVCodecContext * avctx)

  334. {

  335.     MPADecodeContext *s = avctx->priv_data;

  336.     static int init=0;

  337.     int i, j, k;

  338. 

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

  340.     avctx->sample_fmt= SAMPLE_FMT_S32;

  341. #else

  342.     avctx->sample_fmt= SAMPLE_FMT_S16;

  343. #endif    

  344.     

  345.     if(avctx->antialias_algo == FF_AA_INT)

  346.         s->compute_antialias= compute_antialias_integer;

  347.     else

  348.         s->compute_antialias= compute_antialias_float;

  349. 

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

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

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

  353.             int shift, mod;

  354.             /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

  355.             shift = (i / 3);

  356.             mod = i % 3;

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

  358.         }

  359. 

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

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

  362.             int n, norm;

  363.             n = i + 2;

  364.             norm = ((int64_t_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);

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

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

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

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

  369.                     i, norm, 

  370.                     scale_factor_mult[i][0],

  371.                     scale_factor_mult[i][1],

  372.                     scale_factor_mult[i][2]);

  373.         }

  374.         

  375. 	ff_mpa_synth_init(window);

  376.         

  377.         /* huffman decode tables */

  378.         huff_code_table[0] = NULL;

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

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

  381. 	    int xsize, x, y;

  382.             unsigned int n;

  383.             uint8_t *code_table;

  384. 

  385.             xsize = h->xsize;

  386.             n = xsize * xsize;

  387.             /* XXX: fail test */

  388.             init_vlc(&huff_vlc[i], 8, n, 

  389.                      h->bits, 1, 1, h->codes, 2, 2, 1);

  390.             

  391.             code_table = av_mallocz(n);

  392.             j = 0;

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

  394.                 for(y=0;y<xsize;y++)

  395.                     code_table[j++] = (x << 4) | y;

  396.             }

  397.             huff_code_table[i] = code_table;

  398.         }

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

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

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

  402.         }

  403. 

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

  405.             k = 0;

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

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

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

  409.             }

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

  411.         }

  412. 

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

  414. 	table_4_3_exp= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_exp[0]));

  415.         if(!table_4_3_exp)

  416. 	    return -1;

  417. 	table_4_3_value= av_mallocz_static(TABLE_4_3_SIZE * sizeof(table_4_3_value[0]));

  418.         if(!table_4_3_value)

  419.             return -1;

  420.         

  421.         int_pow_init();

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

  423.             int e, m;

  424.             m = int_pow(i, &e);

  425. #if 0

  426.             /* test code */

  427.             {

  428.                 double f, fm;

  429.                 int e1, m1;

  430.                 f = pow((double)i, 4.0 / 3.0);

  431.                 fm = frexp(f, &e1);

  432.                 m1 = FIXR(2 * fm);

  433. #if FRAC_BITS <= 15

  434.                 if ((unsigned short)m1 != m1) {

  435.                     m1 = m1 >> 1;

  436.                     e1++;

  437.                 }

  438. #endif

  439.                 e1--;

  440.                 if (m != m1 || e != e1) {

  441.                     printf("%4d: m=%x m1=%x e=%d e1=%d\n",

  442.                            i, m, m1, e, e1);

  443.                 }

  444.             }

  445. #endif

  446.             /* normalized to FRAC_BITS */

  447.             table_4_3_value[i] = m;

  448.             table_4_3_exp[i] = e;

  449.         }

  450.         

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

  452.             float f;

  453.             int v;

  454.             if (i != 6) {

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

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

  457.             } else {

  458.                 v = FIXR(1.0);

  459.             }

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

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

  462.         }

  463.         /* invalid values */

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

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

  466. 

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

  468.             double f;

  469.             int e, k;

  470. 

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

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

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

  474.                 k = i & 1;

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

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

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

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

  479.             }

  480.         }

  481. 

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

  483.             float ci, cs, ca;

  484.             ci = ci_table[i];

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

  486.             ca = cs * ci;

  487.             csa_table[i][0] = FIX(cs);

  488.             csa_table[i][1] = FIX(ca);

  489.             csa_table[i][2] = FIX(ca) + FIX(cs);

  490.             csa_table[i][3] = FIX(ca) - FIX(cs); 

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

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

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

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

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

  496.         }

  497. 

  498.         /* compute mdct windows */

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

  500.             int v;

  501.             v = FIXR(sin(M_PI * (i + 0.5) / 36.0));

  502.             mdct_win[0][i] = v;

  503.             mdct_win[1][i] = v;

  504.             mdct_win[3][i] = v;

  505.         }

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

  507.             mdct_win[1][18 + i] = FIXR(1.0);

  508.             mdct_win[1][24 + i] = FIXR(sin(M_PI * ((i + 6) + 0.5) / 12.0));

  509.             mdct_win[1][30 + i] = FIXR(0.0);

  510. 

  511.             mdct_win[3][i] = FIXR(0.0);

  512.             mdct_win[3][6 + i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));

  513.             mdct_win[3][12 + i] = FIXR(1.0);

  514.         }

  515. 

  516.         for(i=0;i<12;i++)

  517.             mdct_win[2][i] = FIXR(sin(M_PI * (i + 0.5) / 12.0));

  518.         

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

  520.            the sign of the right window coefs */

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

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

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

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

  525.             }

  526.         }

  527. 

  528. #if defined(DEBUG)

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

  530.             printf("win%d=\n", j);

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

  532.                 printf("%f, ", (double)mdct_win[j][i] / FRAC_ONE);

  533.             printf("\n");

  534.         }

  535. #endif

  536.         init = 1;

  537.     }

  538. 

  539.     s->inbuf_index = 0;

  540.     s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];

  541.     s->inbuf_ptr = s->inbuf;

  542. #ifdef DEBUG

  543.     s->frame_count = 0;

  544. #endif

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

  546.         s->adu_mode = 1;

  547.     return 0;

  548. }

  549. 

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

  551. 

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

  553. 

  554. #define COS0_0  FIXR(0.50060299823519630134)

  555. #define COS0_1  FIXR(0.50547095989754365998)

  556. #define COS0_2  FIXR(0.51544730992262454697)

  557. #define COS0_3  FIXR(0.53104259108978417447)

  558. #define COS0_4  FIXR(0.55310389603444452782)

  559. #define COS0_5  FIXR(0.58293496820613387367)

  560. #define COS0_6  FIXR(0.62250412303566481615)

  561. #define COS0_7  FIXR(0.67480834145500574602)

  562. #define COS0_8  FIXR(0.74453627100229844977)

  563. #define COS0_9  FIXR(0.83934964541552703873)

  564. #define COS0_10 FIXR(0.97256823786196069369)

  565. #define COS0_11 FIXR(1.16943993343288495515)

  566. #define COS0_12 FIXR(1.48416461631416627724)

  567. #define COS0_13 FIXR(2.05778100995341155085)

  568. #define COS0_14 FIXR(3.40760841846871878570)

  569. #define COS0_15 FIXR(10.19000812354805681150)

  570. 

  571. #define COS1_0 FIXR(0.50241928618815570551)

  572. #define COS1_1 FIXR(0.52249861493968888062)

  573. #define COS1_2 FIXR(0.56694403481635770368)

  574. #define COS1_3 FIXR(0.64682178335999012954)

  575. #define COS1_4 FIXR(0.78815462345125022473)

  576. #define COS1_5 FIXR(1.06067768599034747134)

  577. #define COS1_6 FIXR(1.72244709823833392782)

  578. #define COS1_7 FIXR(5.10114861868916385802)

  579. 

  580. #define COS2_0 FIXR(0.50979557910415916894)

  581. #define COS2_1 FIXR(0.60134488693504528054)

  582. #define COS2_2 FIXR(0.89997622313641570463)

  583. #define COS2_3 FIXR(2.56291544774150617881)

  584. 

  585. #define COS3_0 FIXR(0.54119610014619698439)

  586. #define COS3_1 FIXR(1.30656296487637652785)

  587. 

  588. #define COS4_0 FIXR(0.70710678118654752439)

  589. 

  590. /* butterfly operator */

  591. #define BF(a, b, c)\

  592. {\

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

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

  595.     tab[a] = tmp0;\

  596.     tab[b] = MULL(tmp1, c);\

  597. }

  598. 

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

  600. {\

  601.     BF(a, b, COS4_0);\

  602.     BF(c, d, -COS4_0);\

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

  604. }

  605. 

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

  607. {\

  608.     BF(a, b, COS4_0);\

  609.     BF(c, d, -COS4_0);\

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

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

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

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

  614. }

  615. 

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

  617. 

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

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

  620. {

  621.     int tmp0, tmp1;

  622. 

  623.     /* pass 1 */

  624.     BF(0, 31, COS0_0);

  625.     BF(1, 30, COS0_1);

  626.     BF(2, 29, COS0_2);

  627.     BF(3, 28, COS0_3);

  628.     BF(4, 27, COS0_4);

  629.     BF(5, 26, COS0_5);

  630.     BF(6, 25, COS0_6);

  631.     BF(7, 24, COS0_7);

  632.     BF(8, 23, COS0_8);

  633.     BF(9, 22, COS0_9);

  634.     BF(10, 21, COS0_10);

  635.     BF(11, 20, COS0_11);

  636.     BF(12, 19, COS0_12);

  637.     BF(13, 18, COS0_13);

  638.     BF(14, 17, COS0_14);

  639.     BF(15, 16, COS0_15);

  640. 

  641.     /* pass 2 */

  642.     BF(0, 15, COS1_0);

  643.     BF(1, 14, COS1_1);

  644.     BF(2, 13, COS1_2);

  645.     BF(3, 12, COS1_3);

  646.     BF(4, 11, COS1_4);

  647.     BF(5, 10, COS1_5);

  648.     BF(6,  9, COS1_6);

  649.     BF(7,  8, COS1_7);

  650.     

  651.     BF(16, 31, -COS1_0);

  652.     BF(17, 30, -COS1_1);

  653.     BF(18, 29, -COS1_2);

  654.     BF(19, 28, -COS1_3);

  655.     BF(20, 27, -COS1_4);

  656.     BF(21, 26, -COS1_5);

  657.     BF(22, 25, -COS1_6);

  658.     BF(23, 24, -COS1_7);

  659.     

  660.     /* pass 3 */

  661.     BF(0, 7, COS2_0);

  662.     BF(1, 6, COS2_1);

  663.     BF(2, 5, COS2_2);

  664.     BF(3, 4, COS2_3);

  665.     

  666.     BF(8, 15, -COS2_0);

  667.     BF(9, 14, -COS2_1);

  668.     BF(10, 13, -COS2_2);

  669.     BF(11, 12, -COS2_3);

  670.     

  671.     BF(16, 23, COS2_0);

  672.     BF(17, 22, COS2_1);

  673.     BF(18, 21, COS2_2);

  674.     BF(19, 20, COS2_3);

  675.     

  676.     BF(24, 31, -COS2_0);

  677.     BF(25, 30, -COS2_1);

  678.     BF(26, 29, -COS2_2);

  679.     BF(27, 28, -COS2_3);

  680. 

  681.     /* pass 4 */

  682.     BF(0, 3, COS3_0);

  683.     BF(1, 2, COS3_1);

  684.     

  685.     BF(4, 7, -COS3_0);

  686.     BF(5, 6, -COS3_1);

  687.     

  688.     BF(8, 11, COS3_0);

  689.     BF(9, 10, COS3_1);

  690.     

  691.     BF(12, 15, -COS3_0);

  692.     BF(13, 14, -COS3_1);

  693.     

  694.     BF(16, 19, COS3_0);

  695.     BF(17, 18, COS3_1);

  696.     

  697.     BF(20, 23, -COS3_0);

  698.     BF(21, 22, -COS3_1);

  699.     

  700.     BF(24, 27, COS3_0);

  701.     BF(25, 26, COS3_1);

  702.     

  703.     BF(28, 31, -COS3_0);

  704.     BF(29, 30, -COS3_1);

  705.     

  706.     /* pass 5 */

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

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

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

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

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

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

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

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

  715.     

  716.     /* pass 6 */

  717.     

  718.     ADD( 8, 12);

  719.     ADD(12, 10);

  720.     ADD(10, 14);

  721.     ADD(14,  9);

  722.     ADD( 9, 13);

  723.     ADD(13, 11);

  724.     ADD(11, 15);

  725. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  742.     

  743.     ADD(24, 28);

  744.     ADD(28, 26);

  745.     ADD(26, 30);

  746.     ADD(30, 25);

  747.     ADD(25, 29);

  748.     ADD(29, 27);

  749.     ADD(27, 31);

  750. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  767. }

  768. 

  769. #if FRAC_BITS <= 15

  770. 

  771. static inline int round_sample(int *sum)

  772. {

  773.     int sum1;

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

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

  776.     if (sum1 < OUT_MIN)

  777.         sum1 = OUT_MIN;

  778.     else if (sum1 > OUT_MAX)

  779.         sum1 = OUT_MAX;

  780.     return sum1;

  781. }

  782. 

  783. #if defined(ARCH_POWERPC_405)

  784. 

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

  786. #define MACS(rt, ra, rb) \

  787.     asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));

  788. 

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

  790. #define MULS(ra, rb) \

  791.     ({ int __rt; asm ("mullhw %0, %1, %2" : "=r" (__rt) : "r" (ra), "r" (rb)); __rt; })

  792. 

  793. #else

  794. 

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

  796. #define MACS(rt, ra, rb) rt += (ra) * (rb)

  797. 

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

  799. #define MULS(ra, rb) ((ra) * (rb))

  800. 

  801. #endif

  802. 

  803. #else

  804. 

  805. static inline int round_sample(int64_t *sum) 

  806. {

  807.     int sum1;

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

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

  810.     if (sum1 < OUT_MIN)

  811.         sum1 = OUT_MIN;

  812.     else if (sum1 > OUT_MAX)

  813.         sum1 = OUT_MAX;

  814.     return sum1;

  815. }

  816. 

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

  818. 

  819. #endif

  820. 

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

  822. {                                               \

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

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

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

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

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

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

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

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

  831. }

  832. 

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

  834. {                                               \

  835.     int tmp;\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  860. }

  861. 

  862. void ff_mpa_synth_init(MPA_INT *window)

  863. {

  864.     int i;

  865. 

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

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

  868.         int v;

  869.         v = mpa_enwindow[i];

  870. #if WFRAC_BITS < 16

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

  872. #endif

  873.         window[i] = v;

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

  875.             v = -v;

  876.         if (i != 0)

  877.             window[512 - i] = v;

  878.     }	

  879. }

  880. 

  881. /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:

  882.    32 samples. */

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

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

  885. 			 MPA_INT *window, int *dither_state,

  886.                          OUT_INT *samples, int incr, 

  887.                          int32_t sb_samples[SBLIMIT])

  888. {

  889.     int32_t tmp[32];

  890.     register MPA_INT *synth_buf;

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

  892.     int j, offset, v;

  893.     OUT_INT *samples2;

  894. #if FRAC_BITS <= 15

  895.     int sum, sum2;

  896. #else

  897.     int64_t sum, sum2;

  898. #endif

  899. 

  900.     dct32(tmp, sb_samples);

  901.     

  902.     offset = *synth_buf_offset;

  903.     synth_buf = synth_buf_ptr + offset;

  904. 

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

  906.         v = tmp[j];

  907. #if FRAC_BITS <= 15

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

  909.            sound */

  910.         if (v > 32767)

  911.             v = 32767;

  912.         else if (v < -32768)

  913.             v = -32768;

  914. #endif

  915.         synth_buf[j] = v;

  916.     }

  917.     /* copy to avoid wrap */

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

  919. 

  920.     samples2 = samples + 31 * incr;

  921.     w = window;

  922.     w2 = window + 31;

  923. 

  924.     sum = *dither_state;

  925.     p = synth_buf + 16;

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

  927.     p = synth_buf + 48;

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

  929.     *samples = round_sample(&sum);

  930.     samples += incr;

  931.     w++;

  932. 

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

  934.        access per two sample */

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

  936.         sum2 = 0;

  937.         p = synth_buf + 16 + j;

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

  939.         p = synth_buf + 48 - j;

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

  941. 

  942.         *samples = round_sample(&sum);

  943.         samples += incr;

  944.         sum += sum2;

  945.         *samples2 = round_sample(&sum);

  946.         samples2 -= incr;

  947.         w++;

  948.         w2--;

  949.     }

  950.     

  951.     p = synth_buf + 32;

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

  953.     *samples = round_sample(&sum);

  954.     *dither_state= sum;

  955. 

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

  957.     *synth_buf_offset = offset;

  958. }

  959. 

  960. /* cos(pi*i/24) */

  961. #define C1  FIXR(0.99144486137381041114)

  962. #define C3  FIXR(0.92387953251128675612)

  963. #define C5  FIXR(0.79335334029123516458)

  964. #define C7  FIXR(0.60876142900872063941)

  965. #define C9  FIXR(0.38268343236508977173)

  966. #define C11 FIXR(0.13052619222005159154)

  967. 

  968. /* 12 points IMDCT. We compute it "by hand" by factorizing obvious

  969.    cases. */

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

  971. {

  972.     int tmp;

  973.     int64_t in1_3, in1_9, in4_3, in4_9;

  974. 

  975.     in1_3 = MUL64(in[1], C3);

  976.     in1_9 = MUL64(in[1], C9);

  977.     in4_3 = MUL64(in[4], C3);

  978.     in4_9 = MUL64(in[4], C9);

  979.     

  980.     tmp = FRAC_RND(MUL64(in[0], C7) - in1_3 - MUL64(in[2], C11) + 

  981.                    MUL64(in[3], C1) - in4_9 - MUL64(in[5], C5));

  982.     out[0] = tmp;

  983.     out[5] = -tmp;

  984.     tmp = FRAC_RND(MUL64(in[0] - in[3], C9) - in1_3 + 

  985.                    MUL64(in[2] + in[5], C3) - in4_9);

  986.     out[1] = tmp;

  987.     out[4] = -tmp;

  988.     tmp = FRAC_RND(MUL64(in[0], C11) - in1_9 + MUL64(in[2], C7) -

  989.                    MUL64(in[3], C5) + in4_3 - MUL64(in[5], C1));

  990.     out[2] = tmp;

  991.     out[3] = -tmp;

  992.     tmp = FRAC_RND(MUL64(-in[0], C5) + in1_9 + MUL64(in[2], C1) + 

  993.                    MUL64(in[3], C11) - in4_3 - MUL64(in[5], C7));

  994.     out[6] = tmp;

  995.     out[11] = tmp;

  996.     tmp = FRAC_RND(MUL64(-in[0] + in[3], C3) - in1_9 + 

  997.                    MUL64(in[2] + in[5], C9) + in4_3);

  998.     out[7] = tmp;

  999.     out[10] = tmp;

  1000.     tmp = FRAC_RND(-MUL64(in[0], C1) - in1_3 - MUL64(in[2], C5) -

  1001.                    MUL64(in[3], C7) - in4_9 - MUL64(in[5], C11));

  1002.     out[8] = tmp;

  1003.     out[9] = tmp;

  1004. }

  1005. 

  1006. #undef C1

  1007. #undef C3

  1008. #undef C5

  1009. #undef C7

  1010. #undef C9

  1011. #undef C11

  1012. 

  1013. /* cos(pi*i/18) */

  1014. #define C1 FIXR(0.98480775301220805936)

  1015. #define C2 FIXR(0.93969262078590838405)

  1016. #define C3 FIXR(0.86602540378443864676)

  1017. #define C4 FIXR(0.76604444311897803520)

  1018. #define C5 FIXR(0.64278760968653932632)

  1019. #define C6 FIXR(0.5)

  1020. #define C7 FIXR(0.34202014332566873304)

  1021. #define C8 FIXR(0.17364817766693034885)

  1022. 

  1023. /* 0.5 / cos(pi*(2*i+1)/36) */

  1024. static const int icos36[9] = {

  1025.     FIXR(0.50190991877167369479),

  1026.     FIXR(0.51763809020504152469),

  1027.     FIXR(0.55168895948124587824),

  1028.     FIXR(0.61038729438072803416),

  1029.     FIXR(0.70710678118654752439),

  1030.     FIXR(0.87172339781054900991),

  1031.     FIXR(1.18310079157624925896),

  1032.     FIXR(1.93185165257813657349),

  1033.     FIXR(5.73685662283492756461),

  1034. };

  1035. 

  1036. static const int icos72[18] = {

  1037.     /* 0.5 / cos(pi*(2*i+19)/72) */

  1038.     FIXR(0.74009361646113053152),

  1039.     FIXR(0.82133981585229078570),

  1040.     FIXR(0.93057949835178895673),

  1041.     FIXR(1.08284028510010010928),

  1042.     FIXR(1.30656296487637652785),

  1043.     FIXR(1.66275476171152078719),

  1044.     FIXR(2.31011315767264929558),

  1045.     FIXR(3.83064878777019433457),

  1046.     FIXR(11.46279281302667383546),

  1047. 

  1048.     /* 0.5 / cos(pi*(2*(i + 18) +19)/72) */

  1049.     FIXR(-0.67817085245462840086),

  1050.     FIXR(-0.63023620700513223342),

  1051.     FIXR(-0.59284452371708034528),

  1052.     FIXR(-0.56369097343317117734),

  1053.     FIXR(-0.54119610014619698439),

  1054.     FIXR(-0.52426456257040533932),

  1055.     FIXR(-0.51213975715725461845),

  1056.     FIXR(-0.50431448029007636036),

  1057.     FIXR(-0.50047634258165998492),

  1058. };

  1059. 

  1060. /* using Lee like decomposition followed by hand coded 9 points DCT */

  1061. static void imdct36(int *out, int *in)

  1062. {

  1063.     int i, j, t0, t1, t2, t3, s0, s1, s2, s3;

  1064.     int tmp[18], *tmp1, *in1;

  1065.     int64_t in3_3, in6_6;

  1066. 

  1067.     for(i=17;i>=1;i--)

  1068.         in[i] += in[i-1];

  1069.     for(i=17;i>=3;i-=2)

  1070.         in[i] += in[i-2];

  1071. 

  1072.     for(j=0;j<2;j++) {

  1073.         tmp1 = tmp + j;

  1074.         in1 = in + j;

  1075. 

  1076.         in3_3 = MUL64(in1[2*3], C3);

  1077.         in6_6 = MUL64(in1[2*6], C6);

  1078. 

  1079.         tmp1[0] = FRAC_RND(MUL64(in1[2*1], C1) + in3_3 + 

  1080.                            MUL64(in1[2*5], C5) + MUL64(in1[2*7], C7));

  1081.         tmp1[2] = in1[2*0] + FRAC_RND(MUL64(in1[2*2], C2) + 

  1082.                                       MUL64(in1[2*4], C4) + in6_6 + 

  1083.                                       MUL64(in1[2*8], C8));

  1084.         tmp1[4] = FRAC_RND(MUL64(in1[2*1] - in1[2*5] - in1[2*7], C3));

  1085.         tmp1[6] = FRAC_RND(MUL64(in1[2*2] - in1[2*4] - in1[2*8], C6)) - 

  1086.             in1[2*6] + in1[2*0];

  1087.         tmp1[8] = FRAC_RND(MUL64(in1[2*1], C5) - in3_3 - 

  1088.                            MUL64(in1[2*5], C7) + MUL64(in1[2*7], C1));

  1089.         tmp1[10] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C8) - 

  1090.                                        MUL64(in1[2*4], C2) + in6_6 + 

  1091.                                        MUL64(in1[2*8], C4));

  1092.         tmp1[12] = FRAC_RND(MUL64(in1[2*1], C7) - in3_3 + 

  1093.                             MUL64(in1[2*5], C1) - 

  1094.                             MUL64(in1[2*7], C5));

  1095.         tmp1[14] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C4) + 

  1096.                                        MUL64(in1[2*4], C8) + in6_6 - 

  1097.                                        MUL64(in1[2*8], C2));

  1098.         tmp1[16] = in1[2*0] - in1[2*2] + in1[2*4] - in1[2*6] + in1[2*8];

  1099.     }

  1100. 

  1101.     i = 0;

  1102.     for(j=0;j<4;j++) {

  1103.         t0 = tmp[i];

  1104.         t1 = tmp[i + 2];

  1105.         s0 = t1 + t0;

  1106.         s2 = t1 - t0;

  1107. 

  1108.         t2 = tmp[i + 1];

  1109.         t3 = tmp[i + 3];

  1110.         s1 = MULL(t3 + t2, icos36[j]);

  1111.         s3 = MULL(t3 - t2, icos36[8 - j]);

  1112.         

  1113.         t0 = MULL(s0 + s1, icos72[9 + 8 - j]);

  1114.         t1 = MULL(s0 - s1, icos72[8 - j]);

  1115.         out[18 + 9 + j] = t0;

  1116.         out[18 + 8 - j] = t0;

  1117.         out[9 + j] = -t1;

  1118.         out[8 - j] = t1;

  1119.         

  1120.         t0 = MULL(s2 + s3, icos72[9+j]);

  1121.         t1 = MULL(s2 - s3, icos72[j]);

  1122.         out[18 + 9 + (8 - j)] = t0;

  1123.         out[18 + j] = t0;

  1124.         out[9 + (8 - j)] = -t1;

  1125.         out[j] = t1;

  1126.         i += 4;

  1127.     }

  1128. 

  1129.     s0 = tmp[16];

  1130.     s1 = MULL(tmp[17], icos36[4]);

  1131.     t0 = MULL(s0 + s1, icos72[9 + 4]);

  1132.     t1 = MULL(s0 - s1, icos72[4]);

  1133.     out[18 + 9 + 4] = t0;

  1134.     out[18 + 8 - 4] = t0;

  1135.     out[9 + 4] = -t1;

  1136.     out[8 - 4] = t1;

  1137. }

  1138. 

  1139. /* header decoding. MUST check the header before because no

  1140.    consistency check is done there. Return 1 if free format found and

  1141.    that the frame size must be computed externally */

  1142. static int decode_header(MPADecodeContext *s, uint32_t header)

  1143. {

  1144.     int sample_rate, frame_size, mpeg25, padding;

  1145.     int sample_rate_index, bitrate_index;

  1146.     if (header & (1<<20)) {

  1147.         s->lsf = (header & (1<<19)) ? 0 : 1;

  1148.         mpeg25 = 0;

  1149.     } else {

  1150.         s->lsf = 1;

  1151.         mpeg25 = 1;

  1152.     }

  1153.     

  1154.     s->layer = 4 - ((header >> 17) & 3);

  1155.     /* extract frequency */

  1156.     sample_rate_index = (header >> 10) & 3;

  1157.     sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);

  1158.     sample_rate_index += 3 * (s->lsf + mpeg25);

  1159.     s->sample_rate_index = sample_rate_index;

  1160.     s->error_protection = ((header >> 16) & 1) ^ 1;

  1161.     s->sample_rate = sample_rate;

  1162. 

  1163.     bitrate_index = (header >> 12) & 0xf;

  1164.     padding = (header >> 9) & 1;

  1165.     //extension = (header >> 8) & 1;

  1166.     s->mode = (header >> 6) & 3;

  1167.     s->mode_ext = (header >> 4) & 3;

  1168.     //copyright = (header >> 3) & 1;

  1169.     //original = (header >> 2) & 1;

  1170.     //emphasis = header & 3;

  1171. 

  1172.     if (s->mode == MPA_MONO)

  1173.         s->nb_channels = 1;

  1174.     else

  1175.         s->nb_channels = 2;

  1176.     

  1177.     if (bitrate_index != 0) {

  1178.         frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];

  1179.         s->bit_rate = frame_size * 1000;

  1180.         switch(s->layer) {

  1181.         case 1:

  1182.             frame_size = (frame_size * 12000) / sample_rate;

  1183.             frame_size = (frame_size + padding) * 4;

  1184.             break;

  1185.         case 2:

  1186.             frame_size = (frame_size * 144000) / sample_rate;

  1187.             frame_size += padding;

  1188.             break;

  1189.         default:

  1190.         case 3:

  1191.             frame_size = (frame_size * 144000) / (sample_rate << s->lsf);

  1192.             frame_size += padding;

  1193.             break;

  1194.         }

  1195.         s->frame_size = frame_size;

  1196.     } else {

  1197.         /* if no frame size computed, signal it */

  1198.         if (!s->free_format_frame_size)

  1199.             return 1;

  1200.         /* free format: compute bitrate and real frame size from the

  1201.            frame size we extracted by reading the bitstream */

  1202.         s->frame_size = s->free_format_frame_size;

  1203.         switch(s->layer) {

  1204.         case 1:

  1205.             s->frame_size += padding  * 4;

  1206.             s->bit_rate = (s->frame_size * sample_rate) / 48000;

  1207.             break;

  1208.         case 2:

  1209.             s->frame_size += padding;

  1210.             s->bit_rate = (s->frame_size * sample_rate) / 144000;

  1211.             break;

  1212.         default:

  1213.         case 3:

  1214.             s->frame_size += padding;

  1215.             s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;

  1216.             break;

  1217.         }

  1218.     }

  1219.     

  1220. #if defined(DEBUG)

  1221.     printf("layer%d, %d Hz, %d kbits/s, ",

  1222.            s->layer, s->sample_rate, s->bit_rate);

  1223.     if (s->nb_channels == 2) {

  1224.         if (s->layer == 3) {

  1225.             if (s->mode_ext & MODE_EXT_MS_STEREO)

  1226.                 printf("ms-");

  1227.             if (s->mode_ext & MODE_EXT_I_STEREO)

  1228.                 printf("i-");

  1229.         }

  1230.         printf("stereo");

  1231.     } else {

  1232.         printf("mono");

  1233.     }

  1234.     printf("\n");

  1235. #endif

  1236.     return 0;

  1237. }

  1238. 

  1239. /* useful helper to get mpeg audio stream infos. Return -1 if error in

  1240.    header, otherwise the coded frame size in bytes */

  1241. int mpa_decode_header(AVCodecContext *avctx, uint32_t head)

  1242. {

  1243.     MPADecodeContext s1, *s = &s1;

  1244.     memset( s, 0, sizeof(MPADecodeContext) );

  1245. 

  1246.     if (ff_mpa_check_header(head) != 0)

  1247.         return -1;

  1248. 

  1249.     if (decode_header(s, head) != 0) {

  1250.         return -1;

  1251.     }

  1252. 

  1253.     switch(s->layer) {

  1254.     case 1:

  1255.         avctx->frame_size = 384;

  1256.         break;

  1257.     case 2:

  1258.         avctx->frame_size = 1152;

  1259.         break;

  1260.     default:

  1261.     case 3:

  1262.         if (s->lsf)

  1263.             avctx->frame_size = 576;

  1264.         else

  1265.             avctx->frame_size = 1152;

  1266.         break;

  1267.     }

  1268. 

  1269.     avctx->sample_rate = s->sample_rate;

  1270.     avctx->channels = s->nb_channels;

  1271.     avctx->bit_rate = s->bit_rate;

  1272.     avctx->sub_id = s->layer;

  1273.     return s->frame_size;

  1274. }

  1275. 

  1276. /* return the number of decoded frames */

  1277. static int mp_decode_layer1(MPADecodeContext *s)

  1278. {

  1279.     int bound, i, v, n, ch, j, mant;

  1280.     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];

  1281.     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

  1282. 

  1283.     if (s->mode == MPA_JSTEREO) 

  1284.         bound = (s->mode_ext + 1) * 4;

  1285.     else

  1286.         bound = SBLIMIT;

  1287. 

  1288.     /* allocation bits */

  1289.     for(i=0;i<bound;i++) {

  1290.         for(ch=0;ch<s->nb_channels;ch++) {

  1291.             allocation[ch][i] = get_bits(&s->gb, 4);

  1292.         }

  1293.     }

  1294.     for(i=bound;i<SBLIMIT;i++) {

  1295.         allocation[0][i] = get_bits(&s->gb, 4);

  1296.     }

  1297. 

  1298.     /* scale factors */

  1299.     for(i=0;i<bound;i++) {

  1300.         for(ch=0;ch<s->nb_channels;ch++) {

  1301.             if (allocation[ch][i])

  1302.                 scale_factors[ch][i] = get_bits(&s->gb, 6);

  1303.         }

  1304.     }

  1305.     for(i=bound;i<SBLIMIT;i++) {

  1306.         if (allocation[0][i]) {

  1307.             scale_factors[0][i] = get_bits(&s->gb, 6);

  1308.             scale_factors[1][i] = get_bits(&s->gb, 6);

  1309.         }

  1310.     }

  1311.     

  1312.     /* compute samples */

  1313.     for(j=0;j<12;j++) {

  1314.         for(i=0;i<bound;i++) {

  1315.             for(ch=0;ch<s->nb_channels;ch++) {

  1316.                 n = allocation[ch][i];

  1317.                 if (n) {

  1318.                     mant = get_bits(&s->gb, n + 1);

  1319.                     v = l1_unscale(n, mant, scale_factors[ch][i]);

  1320.                 } else {

  1321.                     v = 0;

  1322.                 }

  1323.                 s->sb_samples[ch][j][i] = v;

  1324.             }

  1325.         }

  1326.         for(i=bound;i<SBLIMIT;i++) {

  1327.             n = allocation[0][i];

  1328.             if (n) {

  1329.                 mant = get_bits(&s->gb, n + 1);

  1330.                 v = l1_unscale(n, mant, scale_factors[0][i]);

  1331.                 s->sb_samples[0][j][i] = v;

  1332.                 v = l1_unscale(n, mant, scale_factors[1][i]);

  1333.                 s->sb_samples[1][j][i] = v;

  1334.             } else {

  1335.                 s->sb_samples[0][j][i] = 0;

  1336.                 s->sb_samples[1][j][i] = 0;

  1337.             }

  1338.         }

  1339.     }

  1340.     return 12;

  1341. }

  1342. 

  1343. /* bitrate is in kb/s */

  1344. int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)

  1345. {

  1346.     int ch_bitrate, table;

  1347.     

  1348.     ch_bitrate = bitrate / nb_channels;

  1349.     if (!lsf) {

  1350.         if ((freq == 48000 && ch_bitrate >= 56) ||

  1351.             (ch_bitrate >= 56 && ch_bitrate <= 80)) 

  1352.             table = 0;

  1353.         else if (freq != 48000 && ch_bitrate >= 96) 

  1354.             table = 1;

  1355.         else if (freq != 32000 && ch_bitrate <= 48) 

  1356.             table = 2;

  1357.         else 

  1358.             table = 3;

  1359.     } else {

  1360.         table = 4;

  1361.     }

  1362.     return table;

  1363. }

  1364. 

  1365. static int mp_decode_layer2(MPADecodeContext *s)

  1366. {

  1367.     int sblimit; /* number of used subbands */

  1368.     const unsigned char *alloc_table;

  1369.     int table, bit_alloc_bits, i, j, ch, bound, v;

  1370.     unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];

  1371.     unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];

  1372.     unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;

  1373.     int scale, qindex, bits, steps, k, l, m, b;

  1374. 

  1375.     /* select decoding table */

  1376.     table = l2_select_table(s->bit_rate / 1000, s->nb_channels, 

  1377.                             s->sample_rate, s->lsf);

  1378.     sblimit = sblimit_table[table];

  1379.     alloc_table = alloc_tables[table];

  1380. 

  1381.     if (s->mode == MPA_JSTEREO) 

  1382.         bound = (s->mode_ext + 1) * 4;

  1383.     else

  1384.         bound = sblimit;

  1385. 

  1386.     dprintf("bound=%d sblimit=%d\n", bound, sblimit);

  1387. 

  1388.     /* sanity check */

  1389.     if( bound > sblimit ) bound = sblimit;

  1390. 

  1391.     /* parse bit allocation */

  1392.     j = 0;

  1393.     for(i=0;i<bound;i++) {

  1394.         bit_alloc_bits = alloc_table[j];

  1395.         for(ch=0;ch<s->nb_channels;ch++) {

  1396.             bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);

  1397.         }

  1398.         j += 1 << bit_alloc_bits;

  1399.     }

  1400.     for(i=bound;i<sblimit;i++) {

  1401.         bit_alloc_bits = alloc_table[j];

  1402.         v = get_bits(&s->gb, bit_alloc_bits);

  1403.         bit_alloc[0][i] = v;

  1404.         bit_alloc[1][i] = v;

  1405.         j += 1 << bit_alloc_bits;

  1406.     }

  1407. 

  1408. #ifdef DEBUG

  1409.     {

  1410.         for(ch=0;ch<s->nb_channels;ch++) {

  1411.             for(i=0;i<sblimit;i++)

  1412.                 printf(" %d", bit_alloc[ch][i]);

  1413.             printf("\n");

  1414.         }

  1415.     }

  1416. #endif

  1417. 

  1418.     /* scale codes */

  1419.     for(i=0;i<sblimit;i++) {

  1420.         for(ch=0;ch<s->nb_channels;ch++) {

  1421.             if (bit_alloc[ch][i]) 

  1422.                 scale_code[ch][i] = get_bits(&s->gb, 2);

  1423.         }

  1424.     }

  1425.     

  1426.     /* scale factors */

  1427.     for(i=0;i<sblimit;i++) {

  1428.         for(ch=0;ch<s->nb_channels;ch++) {

  1429.             if (bit_alloc[ch][i]) {

  1430.                 sf = scale_factors[ch][i];

  1431.                 switch(scale_code[ch][i]) {

  1432.                 default:

  1433.                 case 0:

  1434.                     sf[0] = get_bits(&s->gb, 6);

  1435.                     sf[1] = get_bits(&s->gb, 6);

  1436.                     sf[2] = get_bits(&s->gb, 6);

  1437.                     break;

  1438.                 case 2:

  1439.                     sf[0] = get_bits(&s->gb, 6);

  1440.                     sf[1] = sf[0];

  1441.                     sf[2] = sf[0];

  1442.                     break;

  1443.                 case 1:

  1444.                     sf[0] = get_bits(&s->gb, 6);

  1445.                     sf[2] = get_bits(&s->gb, 6);

  1446.                     sf[1] = sf[0];

  1447.                     break;

  1448.                 case 3:

  1449.                     sf[0] = get_bits(&s->gb, 6);

  1450.                     sf[2] = get_bits(&s->gb, 6);

  1451.                     sf[1] = sf[2];

  1452.                     break;

  1453.                 }

  1454.             }

  1455.         }

  1456.     }

  1457. 

  1458. #ifdef DEBUG

  1459.     for(ch=0;ch<s->nb_channels;ch++) {

  1460.         for(i=0;i<sblimit;i++) {

  1461.             if (bit_alloc[ch][i]) {

  1462.                 sf = scale_factors[ch][i];

  1463.                 printf(" %d %d %d", sf[0], sf[1], sf[2]);

  1464.             } else {

  1465.                 printf(" -");

  1466.             }

  1467.         }

  1468.         printf("\n");

  1469.     }

  1470. #endif

  1471. 

  1472.     /* samples */

  1473.     for(k=0;k<3;k++) {

  1474.         for(l=0;l<12;l+=3) {

  1475.             j = 0;

  1476.             for(i=0;i<bound;i++) {

  1477.                 bit_alloc_bits = alloc_table[j];

  1478.                 for(ch=0;ch<s->nb_channels;ch++) {

  1479.                     b = bit_alloc[ch][i];

  1480.                     if (b) {

  1481.                         scale = scale_factors[ch][i][k];

  1482.                         qindex = alloc_table[j+b];

  1483.                         bits = quant_bits[qindex];

  1484.                         if (bits < 0) {

  1485.                             /* 3 values at the same time */

  1486.                             v = get_bits(&s->gb, -bits);

  1487.                             steps = quant_steps[qindex];

  1488.                             s->sb_samples[ch][k * 12 + l + 0][i] = 

  1489.                                 l2_unscale_group(steps, v % steps, scale);

  1490.                             v = v / steps;

  1491.                             s->sb_samples[ch][k * 12 + l + 1][i] = 

  1492.                                 l2_unscale_group(steps, v % steps, scale);

  1493.                             v = v / steps;

  1494.                             s->sb_samples[ch][k * 12 + l + 2][i] = 

  1495.                                 l2_unscale_group(steps, v, scale);

  1496.                         } else {

  1497.                             for(m=0;m<3;m++) {

  1498.                                 v = get_bits(&s->gb, bits);

  1499.                                 v = l1_unscale(bits - 1, v, scale);

  1500.                                 s->sb_samples[ch][k * 12 + l + m][i] = v;

  1501.                             }

  1502.                         }

  1503.                     } else {

  1504.                         s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1505.                         s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1506.                         s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1507.                     }

  1508.                 }

  1509.                 /* next subband in alloc table */

  1510.                 j += 1 << bit_alloc_bits; 

  1511.             }

  1512.             /* XXX: find a way to avoid this duplication of code */

  1513.             for(i=bound;i<sblimit;i++) {

  1514.                 bit_alloc_bits = alloc_table[j];

  1515.                 b = bit_alloc[0][i];

  1516.                 if (b) {

  1517.                     int mant, scale0, scale1;

  1518.                     scale0 = scale_factors[0][i][k];

  1519.                     scale1 = scale_factors[1][i][k];

  1520.                     qindex = alloc_table[j+b];

  1521.                     bits = quant_bits[qindex];

  1522.                     if (bits < 0) {

  1523.                         /* 3 values at the same time */

  1524.                         v = get_bits(&s->gb, -bits);

  1525.                         steps = quant_steps[qindex];

  1526.                         mant = v % steps;

  1527.                         v = v / steps;

  1528.                         s->sb_samples[0][k * 12 + l + 0][i] = 

  1529.                             l2_unscale_group(steps, mant, scale0);

  1530.                         s->sb_samples[1][k * 12 + l + 0][i] = 

  1531.                             l2_unscale_group(steps, mant, scale1);

  1532.                         mant = v % steps;

  1533.                         v = v / steps;

  1534.                         s->sb_samples[0][k * 12 + l + 1][i] = 

  1535.                             l2_unscale_group(steps, mant, scale0);

  1536.                         s->sb_samples[1][k * 12 + l + 1][i] = 

  1537.                             l2_unscale_group(steps, mant, scale1);

  1538.                         s->sb_samples[0][k * 12 + l + 2][i] = 

  1539.                             l2_unscale_group(steps, v, scale0);

  1540.                         s->sb_samples[1][k * 12 + l + 2][i] = 

  1541.                             l2_unscale_group(steps, v, scale1);

  1542.                     } else {

  1543.                         for(m=0;m<3;m++) {

  1544.                             mant = get_bits(&s->gb, bits);

  1545.                             s->sb_samples[0][k * 12 + l + m][i] = 

  1546.                                 l1_unscale(bits - 1, mant, scale0);

  1547.                             s->sb_samples[1][k * 12 + l + m][i] = 

  1548.                                 l1_unscale(bits - 1, mant, scale1);

  1549.                         }

  1550.                     }

  1551.                 } else {

  1552.                     s->sb_samples[0][k * 12 + l + 0][i] = 0;

  1553.                     s->sb_samples[0][k * 12 + l + 1][i] = 0;

  1554.                     s->sb_samples[0][k * 12 + l + 2][i] = 0;

  1555.                     s->sb_samples[1][k * 12 + l + 0][i] = 0;

  1556.                     s->sb_samples[1][k * 12 + l + 1][i] = 0;

  1557.                     s->sb_samples[1][k * 12 + l + 2][i] = 0;

  1558.                 }

  1559.                 /* next subband in alloc table */

  1560.                 j += 1 << bit_alloc_bits; 

  1561.             }

  1562.             /* fill remaining samples to zero */

  1563.             for(i=sblimit;i<SBLIMIT;i++) {

  1564.                 for(ch=0;ch<s->nb_channels;ch++) {

  1565.                     s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1566.                     s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1567.                     s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1568.                 }

  1569.             }

  1570.         }

  1571.     }

  1572.     return 3 * 12;

  1573. }

  1574. 

  1575. /*

  1576.  * Seek back in the stream for backstep bytes (at most 511 bytes)

  1577.  */

  1578. static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)

  1579. {

  1580.     uint8_t *ptr;

  1581. 

  1582.     /* compute current position in stream */

  1583.     ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));

  1584. 

  1585.     /* copy old data before current one */

  1586.     ptr -= backstep;

  1587.     memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] + 

  1588.            BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);

  1589.     /* init get bits again */

  1590.     init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);

  1591. 

  1592.     /* prepare next buffer */

  1593.     s->inbuf_index ^= 1;

  1594.     s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];

  1595.     s->old_frame_size = s->frame_size;

  1596. }

  1597. 

  1598. static inline void lsf_sf_expand(int *slen,

  1599.                                  int sf, int n1, int n2, int n3)

  1600. {

  1601.     if (n3) {

  1602.         slen[3] = sf % n3;

  1603.         sf /= n3;

  1604.     } else {

  1605.         slen[3] = 0;

  1606.     }

  1607.     if (n2) {

  1608.         slen[2] = sf % n2;

  1609.         sf /= n2;

  1610.     } else {

  1611.         slen[2] = 0;

  1612.     }

  1613.     slen[1] = sf % n1;

  1614.     sf /= n1;

  1615.     slen[0] = sf;

  1616. }

  1617. 

  1618. static void exponents_from_scale_factors(MPADecodeContext *s, 

  1619.                                          GranuleDef *g,

  1620.                                          int16_t *exponents)

  1621. {

  1622.     const uint8_t *bstab, *pretab;

  1623.     int len, i, j, k, l, v0, shift, gain, gains[3];

  1624.     int16_t *exp_ptr;

  1625. 

  1626.     exp_ptr = exponents;

  1627.     gain = g->global_gain - 210;

  1628.     shift = g->scalefac_scale + 1;

  1629. 

  1630.     bstab = band_size_long[s->sample_rate_index];

  1631.     pretab = mpa_pretab[g->preflag];

  1632.     for(i=0;i<g->long_end;i++) {

  1633.         v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);

  1634.         len = bstab[i];

  1635.         for(j=len;j>0;j--)

  1636.             *exp_ptr++ = v0;

  1637.     }

  1638. 

  1639.     if (g->short_start < 13) {

  1640.         bstab = band_size_short[s->sample_rate_index];

  1641.         gains[0] = gain - (g->subblock_gain[0] << 3);

  1642.         gains[1] = gain - (g->subblock_gain[1] << 3);

  1643.         gains[2] = gain - (g->subblock_gain[2] << 3);

  1644.         k = g->long_end;

  1645.         for(i=g->short_start;i<13;i++) {

  1646.             len = bstab[i];

  1647.             for(l=0;l<3;l++) {

  1648.                 v0 = gains[l] - (g->scale_factors[k++] << shift);

  1649.                 for(j=len;j>0;j--)

  1650.                 *exp_ptr++ = v0;

  1651.             }

  1652.         }

  1653.     }

  1654. }

  1655. 

  1656. /* handle n = 0 too */

  1657. static inline int get_bitsz(GetBitContext *s, int n)

  1658. {

  1659.     if (n == 0)

  1660.         return 0;

  1661.     else

  1662.         return get_bits(s, n);

  1663. }

  1664. 

  1665. static int huffman_decode(MPADecodeContext *s, GranuleDef *g,

  1666.                           int16_t *exponents, int end_pos)

  1667. {

  1668.     int s_index;

  1669.     int linbits, code, x, y, l, v, i, j, k, pos;

  1670.     GetBitContext last_gb;

  1671.     VLC *vlc;

  1672.     uint8_t *code_table;

  1673. 

  1674.     /* low frequencies (called big values) */

  1675.     s_index = 0;

  1676.     for(i=0;i<3;i++) {

  1677.         j = g->region_size[i];

  1678.         if (j == 0)

  1679.             continue;

  1680.         /* select vlc table */

  1681.         k = g->table_select[i];

  1682.         l = mpa_huff_data[k][0];

  1683.         linbits = mpa_huff_data[k][1];

  1684.         vlc = &huff_vlc[l];

  1685.         code_table = huff_code_table[l];

  1686. 

  1687.         /* read huffcode and compute each couple */

  1688.         for(;j>0;j--) {

  1689.             if (get_bits_count(&s->gb) >= end_pos)

  1690.                 break;

  1691.             if (code_table) {

  1692.                 code = get_vlc(&s->gb, vlc);

  1693.                 if (code < 0)

  1694.                     return -1;

  1695.                 y = code_table[code];

  1696.                 x = y >> 4;

  1697.                 y = y & 0x0f;

  1698.             } else {

  1699.                 x = 0;

  1700.                 y = 0;

  1701.             }

  1702.             dprintf("region=%d n=%d x=%d y=%d exp=%d\n", 

  1703.                     i, g->region_size[i] - j, x, y, exponents[s_index]);

  1704.             if (x) {

  1705.                 if (x == 15)

  1706.                     x += get_bitsz(&s->gb, linbits);

  1707.                 v = l3_unscale(x, exponents[s_index]);

  1708.                 if (get_bits1(&s->gb))

  1709.                     v = -v;

  1710.             } else {

  1711.                 v = 0;

  1712.             }

  1713.             g->sb_hybrid[s_index++] = v;

  1714.             if (y) {

  1715.                 if (y == 15)

  1716.                     y += get_bitsz(&s->gb, linbits);

  1717.                 v = l3_unscale(y, exponents[s_index]);

  1718.                 if (get_bits1(&s->gb))

  1719.                     v = -v;

  1720.             } else {

  1721.                 v = 0;

  1722.             }

  1723.             g->sb_hybrid[s_index++] = v;

  1724.         }

  1725.     }

  1726.             

  1727.     /* high frequencies */

  1728.     vlc = &huff_quad_vlc[g->count1table_select];

  1729.     last_gb.buffer = NULL;

  1730.     while (s_index <= 572) {

  1731.         pos = get_bits_count(&s->gb);

  1732.         if (pos >= end_pos) {

  1733.             if (pos > end_pos && last_gb.buffer != NULL) {

  1734.                 /* some encoders generate an incorrect size for this

  1735.                    part. We must go back into the data */

  1736.                 s_index -= 4;

  1737.                 s->gb = last_gb;

  1738.             }

  1739.             break;

  1740.         }

  1741.         last_gb= s->gb;

  1742. 

  1743.         code = get_vlc(&s->gb, vlc);

  1744.         dprintf("t=%d code=%d\n", g->count1table_select, code);

  1745.         if (code < 0)

  1746.             return -1;

  1747.         for(i=0;i<4;i++) {

  1748.             if (code & (8 >> i)) {

  1749.                 /* non zero value. Could use a hand coded function for

  1750.                    'one' value */

  1751.                 v = l3_unscale(1, exponents[s_index]);

  1752.                 if(get_bits1(&s->gb))

  1753.                     v = -v;

  1754.             } else {

  1755.                 v = 0;

  1756.             }

  1757.             g->sb_hybrid[s_index++] = v;

  1758.         }

  1759.     }

  1760.     while (s_index < 576)

  1761.         g->sb_hybrid[s_index++] = 0;

  1762.     return 0;

  1763. }

  1764. 

  1765. /* Reorder short blocks from bitstream order to interleaved order. It

  1766.    would be faster to do it in parsing, but the code would be far more

  1767.    complicated */

  1768. static void reorder_block(MPADecodeContext *s, GranuleDef *g)

  1769. {

  1770.     int i, j, k, len;

  1771.     int32_t *ptr, *dst, *ptr1;

  1772.     int32_t tmp[576];

  1773. 

  1774.     if (g->block_type != 2)

  1775.         return;

  1776. 

  1777.     if (g->switch_point) {

  1778.         if (s->sample_rate_index != 8) {

  1779.             ptr = g->sb_hybrid + 36;

  1780.         } else {

  1781.             ptr = g->sb_hybrid + 48;

  1782.         }

  1783.     } else {

  1784.         ptr = g->sb_hybrid;

  1785.     }

  1786.     

  1787.     for(i=g->short_start;i<13;i++) {

  1788.         len = band_size_short[s->sample_rate_index][i];

  1789.         ptr1 = ptr;

  1790.         for(k=0;k<3;k++) {

  1791.             dst = tmp + k;

  1792.             for(j=len;j>0;j--) {

  1793.                 *dst = *ptr++;

  1794.                 dst += 3;

  1795.             }

  1796.         }

  1797.         memcpy(ptr1, tmp, len * 3 * sizeof(int32_t));

  1798.     }

  1799. }

  1800. 

  1801. #define ISQRT2 FIXR(0.70710678118654752440)

  1802. 

  1803. static void compute_stereo(MPADecodeContext *s,

  1804.                            GranuleDef *g0, GranuleDef *g1)

  1805. {

  1806.     int i, j, k, l;

  1807.     int32_t v1, v2;

  1808.     int sf_max, tmp0, tmp1, sf, len, non_zero_found;

  1809.     int32_t (*is_tab)[16];

  1810.     int32_t *tab0, *tab1;

  1811.     int non_zero_found_short[3];

  1812. 

  1813.     /* intensity stereo */

  1814.     if (s->mode_ext & MODE_EXT_I_STEREO) {

  1815.         if (!s->lsf) {

  1816.             is_tab = is_table;

  1817.             sf_max = 7;

  1818.         } else {

  1819.             is_tab = is_table_lsf[g1->scalefac_compress & 1];

  1820.             sf_max = 16;

  1821.         }

  1822.             

  1823.         tab0 = g0->sb_hybrid + 576;

  1824.         tab1 = g1->sb_hybrid + 576;

  1825. 

  1826.         non_zero_found_short[0] = 0;

  1827.         non_zero_found_short[1] = 0;

  1828.         non_zero_found_short[2] = 0;

  1829.         k = (13 - g1->short_start) * 3 + g1->long_end - 3;

  1830.         for(i = 12;i >= g1->short_start;i--) {

  1831.             /* for last band, use previous scale factor */

  1832.             if (i != 11)

  1833.                 k -= 3;

  1834.             len = band_size_short[s->sample_rate_index][i];

  1835.             for(l=2;l>=0;l--) {

  1836.                 tab0 -= len;

  1837.                 tab1 -= len;

  1838.                 if (!non_zero_found_short[l]) {

  1839.                     /* test if non zero band. if so, stop doing i-stereo */

  1840.                     for(j=0;j<len;j++) {

  1841.                         if (tab1[j] != 0) {

  1842.                             non_zero_found_short[l] = 1;

  1843.                             goto found1;

  1844.                         }

  1845.                     }

  1846.                     sf = g1->scale_factors[k + l];

  1847.                     if (sf >= sf_max)

  1848.                         goto found1;

  1849. 

  1850.                     v1 = is_tab[0][sf];

  1851.                     v2 = is_tab[1][sf];

  1852.                     for(j=0;j<len;j++) {

  1853.                         tmp0 = tab0[j];

  1854.                         tab0[j] = MULL(tmp0, v1);

  1855.                         tab1[j] = MULL(tmp0, v2);

  1856.                     }

  1857.                 } else {

  1858.                 found1:

  1859.                     if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1860.                         /* lower part of the spectrum : do ms stereo

  1861.                            if enabled */

  1862.                         for(j=0;j<len;j++) {

  1863.                             tmp0 = tab0[j];

  1864.                             tmp1 = tab1[j];

  1865.                             tab0[j] = MULL(tmp0 + tmp1, ISQRT2);

  1866.                             tab1[j] = MULL(tmp0 - tmp1, ISQRT2);

  1867.                         }

  1868.                     }

  1869.                 }

  1870.             }

  1871.         }

  1872. 

  1873.         non_zero_found = non_zero_found_short[0] | 

  1874.             non_zero_found_short[1] | 

  1875.             non_zero_found_short[2];

  1876. 

  1877.         for(i = g1->long_end - 1;i >= 0;i--) {

  1878.             len = band_size_long[s->sample_rate_index][i];

  1879.             tab0 -= len;

  1880.             tab1 -= len;

  1881.             /* test if non zero band. if so, stop doing i-stereo */

  1882.             if (!non_zero_found) {

  1883.                 for(j=0;j<len;j++) {

  1884.                     if (tab1[j] != 0) {

  1885.                         non_zero_found = 1;

  1886.                         goto found2;

  1887.                     }

  1888.                 }

  1889.                 /* for last band, use previous scale factor */

  1890.                 k = (i == 21) ? 20 : i;

  1891.                 sf = g1->scale_factors[k];

  1892.                 if (sf >= sf_max)

  1893.                     goto found2;

  1894.                 v1 = is_tab[0][sf];

  1895.                 v2 = is_tab[1][sf];

  1896.                 for(j=0;j<len;j++) {

  1897.                     tmp0 = tab0[j];

  1898.                     tab0[j] = MULL(tmp0, v1);

  1899.                     tab1[j] = MULL(tmp0, v2);

  1900.                 }

  1901.             } else {

  1902.             found2:

  1903.                 if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1904.                     /* lower part of the spectrum : do ms stereo

  1905.                        if enabled */

  1906.                     for(j=0;j<len;j++) {

  1907.                         tmp0 = tab0[j];

  1908.                         tmp1 = tab1[j];

  1909.                         tab0[j] = MULL(tmp0 + tmp1, ISQRT2);

  1910.                         tab1[j] = MULL(tmp0 - tmp1, ISQRT2);

  1911.                     }

  1912.                 }

  1913.             }

  1914.         }

  1915.     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1916.         /* ms stereo ONLY */

  1917.         /* NOTE: the 1/sqrt(2) normalization factor is included in the

  1918.            global gain */

  1919.         tab0 = g0->sb_hybrid;

  1920.         tab1 = g1->sb_hybrid;

  1921.         for(i=0;i<576;i++) {

  1922.             tmp0 = tab0[i];

  1923.             tmp1 = tab1[i];

  1924.             tab0[i] = tmp0 + tmp1;

  1925.             tab1[i] = tmp0 - tmp1;

  1926.         }

  1927.     }

  1928. }

  1929. 

  1930. static void compute_antialias_integer(MPADecodeContext *s,

  1931.                               GranuleDef *g)

  1932. {

  1933.     int32_t *ptr, *p0, *p1, *csa;

  1934.     int n, i, j;

  1935. 

  1936.     /* we antialias only "long" bands */

  1937.     if (g->block_type == 2) {

  1938.         if (!g->switch_point)

  1939.             return;

  1940.         /* XXX: check this for 8000Hz case */

  1941.         n = 1;

  1942.     } else {

  1943.         n = SBLIMIT - 1;

  1944.     }

  1945.     

  1946.     ptr = g->sb_hybrid + 18;

  1947.     for(i = n;i > 0;i--) {

  1948.         p0 = ptr - 1;

  1949.         p1 = ptr;

  1950.         csa = &csa_table[0][0];       

  1951.         for(j=0;j<4;j++) {

  1952.             int tmp0 = *p0;

  1953.             int tmp1 = *p1;

  1954. #if 0

  1955.             *p0 = FRAC_RND(MUL64(tmp0, csa[0]) - MUL64(tmp1, csa[1]));

  1956.             *p1 = FRAC_RND(MUL64(tmp0, csa[1]) + MUL64(tmp1, csa[0]));

  1957. #else

  1958.             int64_t tmp2= MUL64(tmp0 + tmp1, csa[0]);

  1959.             *p0 = FRAC_RND(tmp2 - MUL64(tmp1, csa[2]));

  1960.             *p1 = FRAC_RND(tmp2 + MUL64(tmp0, csa[3]));

  1961. #endif

  1962.             p0--; p1++;

  1963.             csa += 4;

  1964.             tmp0 = *p0;

  1965.             tmp1 = *p1;

  1966. #if 0

  1967.             *p0 = FRAC_RND(MUL64(tmp0, csa[0]) - MUL64(tmp1, csa[1]));

  1968.             *p1 = FRAC_RND(MUL64(tmp0, csa[1]) + MUL64(tmp1, csa[0]));

  1969. #else

  1970.             tmp2= MUL64(tmp0 + tmp1, csa[0]);

  1971.             *p0 = FRAC_RND(tmp2 - MUL64(tmp1, csa[2]));

  1972.             *p1 = FRAC_RND(tmp2 + MUL64(tmp0, csa[3]));

  1973. #endif

  1974.             p0--; p1++;

  1975.             csa += 4;

  1976.         }

  1977.         ptr += 18;       

  1978.     }

  1979. }

  1980. 

  1981. static void compute_antialias_float(MPADecodeContext *s,

  1982.                               GranuleDef *g)

  1983. {

  1984.     int32_t *ptr, *p0, *p1;

  1985.     int n, i, j;

  1986. 

  1987.     /* we antialias only "long" bands */

  1988.     if (g->block_type == 2) {

  1989.         if (!g->switch_point)

  1990.             return;

  1991.         /* XXX: check this for 8000Hz case */

  1992.         n = 1;

  1993.     } else {

  1994.         n = SBLIMIT - 1;

  1995.     }

  1996.     

  1997.     ptr = g->sb_hybrid + 18;

  1998.     for(i = n;i > 0;i--) {

  1999.         float *csa = &csa_table_float[0][0];       

  2000.         p0 = ptr - 1;

  2001.         p1 = ptr;

  2002.         for(j=0;j<4;j++) {

  2003.             float tmp0 = *p0;

  2004.             float tmp1 = *p1;

  2005. #if 1

  2006.             *p0 = lrintf(tmp0 * csa[0] - tmp1 * csa[1]);

  2007.             *p1 = lrintf(tmp0 * csa[1] + tmp1 * csa[0]);

  2008. #else

  2009.             float tmp2= (tmp0 + tmp1) * csa[0];

  2010.             *p0 = lrintf(tmp2 - tmp1 * csa[2]);

  2011.             *p1 = lrintf(tmp2 + tmp0 * csa[3]);

  2012. #endif

  2013.             p0--; p1++;

  2014.             csa += 4;

  2015.             tmp0 = *p0;

  2016.             tmp1 = *p1;

  2017. #if 1

  2018.             *p0 = lrintf(tmp0 * csa[0] - tmp1 * csa[1]);

  2019.             *p1 = lrintf(tmp0 * csa[1] + tmp1 * csa[0]);

  2020. #else

  2021.             tmp2= (tmp0 + tmp1) * csa[0];

  2022.             *p0 = lrintf(tmp2 - tmp1 * csa[2]);

  2023.             *p1 = lrintf(tmp2 + tmp0 * csa[3]);

  2024. #endif

  2025.             p0--; p1++;

  2026.             csa += 4;

  2027.         }

  2028.         ptr += 18;       

  2029.     }

  2030. }

  2031. 

  2032. static void compute_imdct(MPADecodeContext *s,

  2033.                           GranuleDef *g, 

  2034.                           int32_t *sb_samples,

  2035.                           int32_t *mdct_buf)

  2036. {

  2037.     int32_t *ptr, *win, *win1, *buf, *buf2, *out_ptr, *ptr1;

  2038.     int32_t in[6];

  2039.     int32_t out[36];

  2040.     int32_t out2[12];

  2041.     int i, j, k, mdct_long_end, v, sblimit;

  2042. 

  2043.     /* find last non zero block */

  2044.     ptr = g->sb_hybrid + 576;

  2045.     ptr1 = g->sb_hybrid + 2 * 18;

  2046.     while (ptr >= ptr1) {

  2047.         ptr -= 6;

  2048.         v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];

  2049.         if (v != 0)

  2050.             break;

  2051.     }

  2052.     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

  2053. 

  2054.     if (g->block_type == 2) {

  2055.         /* XXX: check for 8000 Hz */

  2056.         if (g->switch_point)

  2057.             mdct_long_end = 2;

  2058.         else

  2059.             mdct_long_end = 0;

  2060.     } else {

  2061.         mdct_long_end = sblimit;

  2062.     }

  2063. 

  2064.     buf = mdct_buf;

  2065.     ptr = g->sb_hybrid;

  2066.     for(j=0;j<mdct_long_end;j++) {

  2067.         imdct36(out, ptr);

  2068.         /* apply window & overlap with previous buffer */

  2069.         out_ptr = sb_samples + j;

  2070.         /* select window */

  2071.         if (g->switch_point && j < 2)

  2072.             win1 = mdct_win[0];

  2073.         else

  2074.             win1 = mdct_win[g->block_type];

  2075.         /* select frequency inversion */

  2076.         win = win1 + ((4 * 36) & -(j & 1));

  2077.         for(i=0;i<18;i++) {

  2078.             *out_ptr = MULL(out[i], win[i]) + buf[i];

  2079.             buf[i] = MULL(out[i + 18], win[i + 18]);

  2080.             out_ptr += SBLIMIT;

  2081.         }

  2082.         ptr += 18;

  2083.         buf += 18;

  2084.     }

  2085.     for(j=mdct_long_end;j<sblimit;j++) {

  2086.         for(i=0;i<6;i++) {

  2087.             out[i] = 0;

  2088.             out[6 + i] = 0;

  2089.             out[30+i] = 0;

  2090.         }

  2091.         /* select frequency inversion */

  2092.         win = mdct_win[2] + ((4 * 36) & -(j & 1));

  2093.         buf2 = out + 6;

  2094.         for(k=0;k<3;k++) {

  2095.             /* reorder input for short mdct */

  2096.             ptr1 = ptr + k;

  2097.             for(i=0;i<6;i++) {

  2098.                 in[i] = *ptr1;

  2099.                 ptr1 += 3;

  2100.             }

  2101.             imdct12(out2, in);

  2102.             /* apply 12 point window and do small overlap */

  2103.             for(i=0;i<6;i++) {

  2104.                 buf2[i] = MULL(out2[i], win[i]) + buf2[i];

  2105.                 buf2[i + 6] = MULL(out2[i + 6], win[i + 6]);

  2106.             }

  2107.             buf2 += 6;

  2108.         }

  2109.         /* overlap */

  2110.         out_ptr = sb_samples + j;

  2111.         for(i=0;i<18;i++) {

  2112.             *out_ptr = out[i] + buf[i];

  2113.             buf[i] = out[i + 18];

  2114.             out_ptr += SBLIMIT;

  2115.         }

  2116.         ptr += 18;

  2117.         buf += 18;

  2118.     }

  2119.     /* zero bands */

  2120.     for(j=sblimit;j<SBLIMIT;j++) {

  2121.         /* overlap */

  2122.         out_ptr = sb_samples + j;

  2123.         for(i=0;i<18;i++) {

  2124.             *out_ptr = buf[i];

  2125.             buf[i] = 0;

  2126.             out_ptr += SBLIMIT;

  2127.         }

  2128.         buf += 18;

  2129.     }

  2130. }

  2131. 

  2132. #if defined(DEBUG)

  2133. void sample_dump(int fnum, int32_t *tab, int n)

  2134. {

  2135.     static FILE *files[16], *f;

  2136.     char buf[512];

  2137.     int i;

  2138.     int32_t v;

  2139.     

  2140.     f = files[fnum];

  2141.     if (!f) {

  2142.         snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm", 

  2143.                 fnum, 

  2144. #ifdef USE_HIGHPRECISION

  2145.                 "hp"

  2146. #else

  2147.                 "lp"

  2148. #endif

  2149.                 );

  2150.         f = fopen(buf, "w");

  2151.         if (!f)

  2152.             return;

  2153.         files[fnum] = f;

  2154.     }

  2155.     

  2156.     if (fnum == 0) {

  2157.         static int pos = 0;

  2158.         av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);

  2159.         for(i=0;i<n;i++) {

  2160.             av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);

  2161.             if ((i % 18) == 17)

  2162.                 av_log(NULL, AV_LOG_DEBUG, "\n");

  2163.         }

  2164.         pos += n;

  2165.     }

  2166.     for(i=0;i<n;i++) {

  2167.         /* normalize to 23 frac bits */

  2168.         v = tab[i] << (23 - FRAC_BITS);

  2169.         fwrite(&v, 1, sizeof(int32_t), f);

  2170.     }

  2171. }

  2172. #endif

  2173. 

  2174. 

  2175. /* main layer3 decoding function */

  2176. static int mp_decode_layer3(MPADecodeContext *s)

  2177. {

  2178.     int nb_granules, main_data_begin, private_bits;

  2179.     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;

  2180.     GranuleDef granules[2][2], *g;

  2181.     int16_t exponents[576];

  2182. 

  2183.     /* read side info */

  2184.     if (s->lsf) {

  2185.         main_data_begin = get_bits(&s->gb, 8);

  2186.         if (s->nb_channels == 2)

  2187.             private_bits = get_bits(&s->gb, 2);

  2188.         else

  2189.             private_bits = get_bits(&s->gb, 1);

  2190.         nb_granules = 1;

  2191.     } else {

  2192.         main_data_begin = get_bits(&s->gb, 9);

  2193.         if (s->nb_channels == 2)

  2194.             private_bits = get_bits(&s->gb, 3);

  2195.         else

  2196.             private_bits = get_bits(&s->gb, 5);

  2197.         nb_granules = 2;

  2198.         for(ch=0;ch<s->nb_channels;ch++) {

  2199.             granules[ch][0].scfsi = 0; /* all scale factors are transmitted */

  2200.             granules[ch][1].scfsi = get_bits(&s->gb, 4);

  2201.         }

  2202.     }

  2203.     

  2204.     for(gr=0;gr<nb_granules;gr++) {

  2205.         for(ch=0;ch<s->nb_channels;ch++) {

  2206.             dprintf("gr=%d ch=%d: side_info\n", gr, ch);

  2207.             g = &granules[ch][gr];

  2208.             g->part2_3_length = get_bits(&s->gb, 12);

  2209.             g->big_values = get_bits(&s->gb, 9);

  2210.             g->global_gain = get_bits(&s->gb, 8);

  2211.             /* if MS stereo only is selected, we precompute the

  2212.                1/sqrt(2) renormalization factor */

  2213.             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) == 

  2214.                 MODE_EXT_MS_STEREO)

  2215.                 g->global_gain -= 2;

  2216.             if (s->lsf)

  2217.                 g->scalefac_compress = get_bits(&s->gb, 9);

  2218.             else

  2219.                 g->scalefac_compress = get_bits(&s->gb, 4);

  2220.             blocksplit_flag = get_bits(&s->gb, 1);

  2221.             if (blocksplit_flag) {

  2222.                 g->block_type = get_bits(&s->gb, 2);

  2223.                 if (g->block_type == 0)

  2224.                     return -1;

  2225.                 g->switch_point = get_bits(&s->gb, 1);

  2226.                 for(i=0;i<2;i++)

  2227.                     g->table_select[i] = get_bits(&s->gb, 5);

  2228.                 for(i=0;i<3;i++) 

  2229.                     g->subblock_gain[i] = get_bits(&s->gb, 3);

  2230.                 /* compute huffman coded region sizes */

  2231.                 if (g->block_type == 2)

  2232.                     g->region_size[0] = (36 / 2);

  2233.                 else {

  2234.                     if (s->sample_rate_index <= 2) 

  2235.                         g->region_size[0] = (36 / 2);

  2236.                     else if (s->sample_rate_index != 8) 

  2237.                         g->region_size[0] = (54 / 2);

  2238.                     else

  2239.                         g->region_size[0] = (108 / 2);

  2240.                 }

  2241.                 g->region_size[1] = (576 / 2);

  2242.             } else {

  2243.                 int region_address1, region_address2, l;

  2244.                 g->block_type = 0;

  2245.                 g->switch_point = 0;

  2246.                 for(i=0;i<3;i++)

  2247.                     g->table_select[i] = get_bits(&s->gb, 5);

  2248.                 /* compute huffman coded region sizes */

  2249.                 region_address1 = get_bits(&s->gb, 4);

  2250.                 region_address2 = get_bits(&s->gb, 3);

  2251.                 dprintf("region1=%d region2=%d\n", 

  2252.                         region_address1, region_address2);

  2253.                 g->region_size[0] = 

  2254.                     band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;

  2255.                 l = region_address1 + region_address2 + 2;

  2256.                 /* should not overflow */

  2257.                 if (l > 22)

  2258.                     l = 22;

  2259.                 g->region_size[1] = 

  2260.                     band_index_long[s->sample_rate_index][l] >> 1;

  2261.             }

  2262.             /* convert region offsets to region sizes and truncate

  2263.                size to big_values */

  2264.             g->region_size[2] = (576 / 2);

  2265.             j = 0;

  2266.             for(i=0;i<3;i++) {

  2267.                 k = g->region_size[i];

  2268.                 if (k > g->big_values)

  2269.                     k = g->big_values;

  2270.                 g->region_size[i] = k - j;

  2271.                 j = k;

  2272.             }

  2273. 

  2274.             /* compute band indexes */

  2275.             if (g->block_type == 2) {

  2276.                 if (g->switch_point) {

  2277.                     /* if switched mode, we handle the 36 first samples as

  2278.                        long blocks.  For 8000Hz, we handle the 48 first

  2279.                        exponents as long blocks (XXX: check this!) */

  2280.                     if (s->sample_rate_index <= 2)

  2281.                         g->long_end = 8;

  2282.                     else if (s->sample_rate_index != 8)

  2283.                         g->long_end = 6;

  2284.                     else

  2285.                         g->long_end = 4; /* 8000 Hz */

  2286.                     

  2287.                     if (s->sample_rate_index != 8)

  2288.                         g->short_start = 3;

  2289.                     else

  2290.                         g->short_start = 2; 

  2291.                 } else {

  2292.                     g->long_end = 0;

  2293.                     g->short_start = 0;

  2294.                 }

  2295.             } else {

  2296.                 g->short_start = 13;

  2297.                 g->long_end = 22;

  2298.             }

  2299.             

  2300.             g->preflag = 0;

  2301.             if (!s->lsf)

  2302.                 g->preflag = get_bits(&s->gb, 1);

  2303.             g->scalefac_scale = get_bits(&s->gb, 1);

  2304.             g->count1table_select = get_bits(&s->gb, 1);

  2305.             dprintf("block_type=%d switch_point=%d\n",

  2306.                     g->block_type, g->switch_point);

  2307.         }

  2308.     }

  2309. 

  2310.   if (!s->adu_mode) {

  2311.     /* now we get bits from the main_data_begin offset */

  2312.     dprintf("seekback: %d\n", main_data_begin);

  2313.     seek_to_maindata(s, main_data_begin);

  2314.   }

  2315. 

  2316.     for(gr=0;gr<nb_granules;gr++) {

  2317.         for(ch=0;ch<s->nb_channels;ch++) {

  2318.             g = &granules[ch][gr];

  2319.             

  2320.             bits_pos = get_bits_count(&s->gb);

  2321.             

  2322.             if (!s->lsf) {

  2323.                 uint8_t *sc;

  2324.                 int slen, slen1, slen2;

  2325. 

  2326.                 /* MPEG1 scale factors */

  2327.                 slen1 = slen_table[0][g->scalefac_compress];

  2328.                 slen2 = slen_table[1][g->scalefac_compress];

  2329.                 dprintf("slen1=%d slen2=%d\n", slen1, slen2);

  2330.                 if (g->block_type == 2) {

  2331.                     n = g->switch_point ? 17 : 18;

  2332.                     j = 0;

  2333.                     for(i=0;i<n;i++)

  2334.                         g->scale_factors[j++] = get_bitsz(&s->gb, slen1);

  2335.                     for(i=0;i<18;i++)

  2336.                         g->scale_factors[j++] = get_bitsz(&s->gb, slen2);

  2337.                     for(i=0;i<3;i++)

  2338.                         g->scale_factors[j++] = 0;

  2339.                 } else {

  2340.                     sc = granules[ch][0].scale_factors;

  2341.                     j = 0;

  2342.                     for(k=0;k<4;k++) {

  2343.                         n = (k == 0 ? 6 : 5);

  2344.                         if ((g->scfsi & (0x8 >> k)) == 0) {

  2345.                             slen = (k < 2) ? slen1 : slen2;

  2346.                             for(i=0;i<n;i++)

  2347.                                 g->scale_factors[j++] = get_bitsz(&s->gb, slen);

  2348.                         } else {

  2349.                             /* simply copy from last granule */

  2350.                             for(i=0;i<n;i++) {

  2351.                                 g->scale_factors[j] = sc[j];

  2352.                                 j++;

  2353.                             }

  2354.                         }

  2355.                     }

  2356.                     g->scale_factors[j++] = 0;

  2357.                 }

  2358. #if defined(DEBUG)

  2359.                 {

  2360.                     printf("scfsi=%x gr=%d ch=%d scale_factors:\n", 

  2361.                            g->scfsi, gr, ch);

  2362.                     for(i=0;i<j;i++)

  2363.                         printf(" %d", g->scale_factors[i]);

  2364.                     printf("\n");

  2365.                 }

  2366. #endif

  2367.             } else {

  2368.                 int tindex, tindex2, slen[4], sl, sf;

  2369. 

  2370.                 /* LSF scale factors */

  2371.                 if (g->block_type == 2) {

  2372.                     tindex = g->switch_point ? 2 : 1;

  2373.                 } else {

  2374.                     tindex = 0;

  2375.                 }

  2376.                 sf = g->scalefac_compress;

  2377.                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {

  2378.                     /* intensity stereo case */

  2379.                     sf >>= 1;

  2380.                     if (sf < 180) {

  2381.                         lsf_sf_expand(slen, sf, 6, 6, 0);

  2382.                         tindex2 = 3;

  2383.                     } else if (sf < 244) {

  2384.                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);

  2385.                         tindex2 = 4;

  2386.                     } else {

  2387.                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);

  2388.                         tindex2 = 5;

  2389.                     }

  2390.                 } else {

  2391.                     /* normal case */

  2392.                     if (sf < 400) {

  2393.                         lsf_sf_expand(slen, sf, 5, 4, 4);

  2394.                         tindex2 = 0;

  2395.                     } else if (sf < 500) {

  2396.                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);

  2397.                         tindex2 = 1;

  2398.                     } else {

  2399.                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);

  2400.                         tindex2 = 2;

  2401.                         g->preflag = 1;

  2402.                     }

  2403.                 }

  2404. 

  2405.                 j = 0;

  2406.                 for(k=0;k<4;k++) {

  2407.                     n = lsf_nsf_table[tindex2][tindex][k];

  2408.                     sl = slen[k];

  2409.                     for(i=0;i<n;i++)

  2410.                         g->scale_factors[j++] = get_bitsz(&s->gb, sl);

  2411.                 }

  2412.                 /* XXX: should compute exact size */

  2413.                 for(;j<40;j++)

  2414.                     g->scale_factors[j] = 0;

  2415. #if defined(DEBUG)

  2416.                 {

  2417.                     printf("gr=%d ch=%d scale_factors:\n", 

  2418.                            gr, ch);

  2419.                     for(i=0;i<40;i++)

  2420.                         printf(" %d", g->scale_factors[i]);

  2421.                     printf("\n");

  2422.                 }

  2423. #endif

  2424.             }

  2425. 

  2426.             exponents_from_scale_factors(s, g, exponents);

  2427. 

  2428.             /* read Huffman coded residue */

  2429.             if (huffman_decode(s, g, exponents,

  2430.                                bits_pos + g->part2_3_length) < 0)

  2431.                 return -1;

  2432. #if defined(DEBUG)

  2433.             sample_dump(0, g->sb_hybrid, 576);

  2434. #endif

  2435. 

  2436.             /* skip extension bits */

  2437.             bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);

  2438.             if (bits_left < 0) {

  2439.                 dprintf("bits_left=%d\n", bits_left);

  2440.                 return -1;

  2441.             }

  2442.             while (bits_left >= 16) {

  2443.                 skip_bits(&s->gb, 16);

  2444.                 bits_left -= 16;

  2445.             }

  2446.             if (bits_left > 0)

  2447.                 skip_bits(&s->gb, bits_left);

  2448.         } /* ch */

  2449. 

  2450.         if (s->nb_channels == 2)

  2451.             compute_stereo(s, &granules[0][gr], &granules[1][gr]);

  2452. 

  2453.         for(ch=0;ch<s->nb_channels;ch++) {

  2454.             g = &granules[ch][gr];

  2455. 

  2456.             reorder_block(s, g);

  2457. #if defined(DEBUG)

  2458.             sample_dump(0, g->sb_hybrid, 576);

  2459. #endif

  2460.             s->compute_antialias(s, g);

  2461. #if defined(DEBUG)

  2462.             sample_dump(1, g->sb_hybrid, 576);

  2463. #endif

  2464.             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]); 

  2465. #if defined(DEBUG)

  2466.             sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);

  2467. #endif

  2468.         }

  2469.     } /* gr */

  2470.     return nb_granules * 18;

  2471. }

  2472. 

  2473. static int mp_decode_frame(MPADecodeContext *s, 

  2474.                            OUT_INT *samples)

  2475. {

  2476.     int i, nb_frames, ch;

  2477.     OUT_INT *samples_ptr;

  2478. 

  2479.     init_get_bits(&s->gb, s->inbuf + HEADER_SIZE, 

  2480.                   (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);

  2481.     

  2482.     /* skip error protection field */

  2483.     if (s->error_protection)

  2484.         get_bits(&s->gb, 16);

  2485. 

  2486.     dprintf("frame %d:\n", s->frame_count);

  2487.     switch(s->layer) {

  2488.     case 1:

  2489.         nb_frames = mp_decode_layer1(s);

  2490.         break;

  2491.     case 2:

  2492.         nb_frames = mp_decode_layer2(s);

  2493.         break;

  2494.     case 3:

  2495.     default:

  2496.         nb_frames = mp_decode_layer3(s);

  2497.         break;

  2498.     }

  2499. #if defined(DEBUG)

  2500.     for(i=0;i<nb_frames;i++) {

  2501.         for(ch=0;ch<s->nb_channels;ch++) {

  2502.             int j;

  2503.             printf("%d-%d:", i, ch);

  2504.             for(j=0;j<SBLIMIT;j++)

  2505.                 printf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);

  2506.             printf("\n");

  2507.         }

  2508.     }

  2509. #endif

  2510.     /* apply the synthesis filter */

  2511.     for(ch=0;ch<s->nb_channels;ch++) {

  2512.         samples_ptr = samples + ch;

  2513.         for(i=0;i<nb_frames;i++) {

  2514.             ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),

  2515. 			 window, &s->dither_state,

  2516. 			 samples_ptr, s->nb_channels,

  2517.                          s->sb_samples[ch][i]);

  2518.             samples_ptr += 32 * s->nb_channels;

  2519.         }

  2520.     }

  2521. #ifdef DEBUG

  2522.     s->frame_count++;        

  2523. #endif

  2524.     return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;

  2525. }

  2526. 

  2527. static int decode_frame(AVCodecContext * avctx,

  2528. 			void *data, int *data_size,

  2529. 			uint8_t * buf, int buf_size)

  2530. {

  2531.     MPADecodeContext *s = avctx->priv_data;

  2532.     uint32_t header;

  2533.     uint8_t *buf_ptr;

  2534.     int len, out_size;

  2535.     OUT_INT *out_samples = data;

  2536. 

  2537.     buf_ptr = buf;

  2538.     while (buf_size > 0) {

  2539. 	len = s->inbuf_ptr - s->inbuf;

  2540. 	if (s->frame_size == 0) {

  2541.             /* special case for next header for first frame in free

  2542.                format case (XXX: find a simpler method) */

  2543.             if (s->free_format_next_header != 0) {

  2544.                 s->inbuf[0] = s->free_format_next_header >> 24;

  2545.                 s->inbuf[1] = s->free_format_next_header >> 16;

  2546.                 s->inbuf[2] = s->free_format_next_header >> 8;

  2547.                 s->inbuf[3] = s->free_format_next_header;

  2548.                 s->inbuf_ptr = s->inbuf + 4;

  2549.                 s->free_format_next_header = 0;

  2550.                 goto got_header;

  2551.             }

  2552. 	    /* no header seen : find one. We need at least HEADER_SIZE

  2553.                bytes to parse it */

  2554. 	    len = HEADER_SIZE - len;

  2555. 	    if (len > buf_size)

  2556. 		len = buf_size;

  2557. 	    if (len > 0) {

  2558. 		memcpy(s->inbuf_ptr, buf_ptr, len);

  2559. 		buf_ptr += len;

  2560. 		buf_size -= len;

  2561. 		s->inbuf_ptr += len;

  2562. 	    }

  2563. 	    if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {

  2564.             got_header:

  2565. 		header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |

  2566. 		    (s->inbuf[2] << 8) | s->inbuf[3];

  2567. 

  2568. 		if (ff_mpa_check_header(header) < 0) {

  2569. 		    /* no sync found : move by one byte (inefficient, but simple!) */

  2570. 		    memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);

  2571. 		    s->inbuf_ptr--;

  2572.                     dprintf("skip %x\n", header);

  2573.                     /* reset free format frame size to give a chance

  2574.                        to get a new bitrate */

  2575.                     s->free_format_frame_size = 0;

  2576. 		} else {

  2577. 		    if (decode_header(s, header) == 1) {

  2578.                         /* free format: prepare to compute frame size */

  2579. 			s->frame_size = -1;

  2580.                     }

  2581.                     /* update codec info */

  2582.                     avctx->sample_rate = s->sample_rate;

  2583.                     avctx->channels = s->nb_channels;

  2584.                     avctx->bit_rate = s->bit_rate;

  2585.                     avctx->sub_id = s->layer;

  2586.                     switch(s->layer) {

  2587.                     case 1:

  2588.                         avctx->frame_size = 384;

  2589.                         break;

  2590.                     case 2:

  2591.                         avctx->frame_size = 1152;

  2592.                         break;

  2593.                     case 3:

  2594.                         if (s->lsf)

  2595.                             avctx->frame_size = 576;

  2596.                         else

  2597.                             avctx->frame_size = 1152;

  2598.                         break;

  2599.                     }

  2600. 		}

  2601. 	    }

  2602.         } else if (s->frame_size == -1) {

  2603.             /* free format : find next sync to compute frame size */

  2604. 	    len = MPA_MAX_CODED_FRAME_SIZE - len;

  2605. 	    if (len > buf_size)

  2606. 		len = buf_size;

  2607.             if (len == 0) {

  2608. 		/* frame too long: resync */

  2609.                 s->frame_size = 0;

  2610. 		memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);

  2611. 		s->inbuf_ptr--;

  2612.             } else {

  2613.                 uint8_t *p, *pend;

  2614.                 uint32_t header1;

  2615.                 int padding;

  2616. 

  2617.                 memcpy(s->inbuf_ptr, buf_ptr, len);

  2618.                 /* check for header */

  2619.                 p = s->inbuf_ptr - 3;

  2620.                 pend = s->inbuf_ptr + len - 4;

  2621.                 while (p <= pend) {

  2622.                     header = (p[0] << 24) | (p[1] << 16) |

  2623.                         (p[2] << 8) | p[3];

  2624.                     header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |

  2625.                         (s->inbuf[2] << 8) | s->inbuf[3];

  2626.                     /* check with high probability that we have a

  2627.                        valid header */

  2628.                     if ((header & SAME_HEADER_MASK) ==

  2629.                         (header1 & SAME_HEADER_MASK)) {

  2630.                         /* header found: update pointers */

  2631.                         len = (p + 4) - s->inbuf_ptr;

  2632.                         buf_ptr += len;

  2633.                         buf_size -= len;

  2634.                         s->inbuf_ptr = p;

  2635.                         /* compute frame size */

  2636.                         s->free_format_next_header = header;

  2637.                         s->free_format_frame_size = s->inbuf_ptr - s->inbuf;

  2638.                         padding = (header1 >> 9) & 1;

  2639.                         if (s->layer == 1)

  2640.                             s->free_format_frame_size -= padding * 4;

  2641.                         else

  2642.                             s->free_format_frame_size -= padding;

  2643.                         dprintf("free frame size=%d padding=%d\n", 

  2644.                                 s->free_format_frame_size, padding);

  2645.                         decode_header(s, header1);

  2646.                         goto next_data;

  2647.                     }

  2648.                     p++;

  2649.                 }

  2650.                 /* not found: simply increase pointers */

  2651.                 buf_ptr += len;

  2652.                 s->inbuf_ptr += len;

  2653.                 buf_size -= len;

  2654.             }

  2655. 	} else if (len < s->frame_size) {

  2656.             if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)

  2657.                 s->frame_size = MPA_MAX_CODED_FRAME_SIZE;

  2658. 	    len = s->frame_size - len;

  2659. 	    if (len > buf_size)

  2660. 		len = buf_size;

  2661. 	    memcpy(s->inbuf_ptr, buf_ptr, len);

  2662. 	    buf_ptr += len;

  2663. 	    s->inbuf_ptr += len;

  2664. 	    buf_size -= len;

  2665. 	}

  2666.     next_data:

  2667.         if (s->frame_size > 0 && 

  2668.             (s->inbuf_ptr - s->inbuf) >= s->frame_size) {

  2669.             if (avctx->parse_only) {

  2670.                 /* simply return the frame data */

  2671.                 *(uint8_t **)data = s->inbuf;

  2672.                 out_size = s->inbuf_ptr - s->inbuf;

  2673.             } else {

  2674.                 out_size = mp_decode_frame(s, out_samples);

  2675.             }

  2676. 	    s->inbuf_ptr = s->inbuf;

  2677. 	    s->frame_size = 0;

  2678. 	    *data_size = out_size;

  2679. 	    break;

  2680. 	}

  2681.     }

  2682.     return buf_ptr - buf;

  2683. }

  2684. 

  2685. 

  2686. static int decode_frame_adu(AVCodecContext * avctx,

  2687. 			void *data, int *data_size,

  2688. 			uint8_t * buf, int buf_size)

  2689. {

  2690.     MPADecodeContext *s = avctx->priv_data;

  2691.     uint32_t header;

  2692.     int len, out_size;

  2693.     OUT_INT *out_samples = data;

  2694. 

  2695.     len = buf_size;

  2696. 

  2697.     // Discard too short frames

  2698.     if (buf_size < HEADER_SIZE) {

  2699.         *data_size = 0;

  2700.         return buf_size;

  2701.     }

  2702. 

  2703. 

  2704.     if (len > MPA_MAX_CODED_FRAME_SIZE)

  2705.         len = MPA_MAX_CODED_FRAME_SIZE;

  2706. 

  2707.     memcpy(s->inbuf, buf, len);

  2708.     s->inbuf_ptr = s->inbuf + len;

  2709. 

  2710.     // Get header and restore sync word

  2711.     header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |

  2712.               (s->inbuf[2] << 8) | s->inbuf[3] | 0xffe00000;

  2713. 

  2714.     if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame

  2715.         *data_size = 0;

  2716.         return buf_size;

  2717.     }

  2718. 

  2719.     decode_header(s, header);

  2720.     /* update codec info */

  2721.     avctx->sample_rate = s->sample_rate;

  2722.     avctx->channels = s->nb_channels;

  2723.     avctx->bit_rate = s->bit_rate;

  2724.     avctx->sub_id = s->layer;

  2725. 

  2726.     avctx->frame_size=s->frame_size = len;

  2727. 

  2728.     if (avctx->parse_only) {

  2729.         /* simply return the frame data */

  2730.         *(uint8_t **)data = s->inbuf;

  2731.         out_size = s->inbuf_ptr - s->inbuf;

  2732.     } else {

  2733.         out_size = mp_decode_frame(s, out_samples);

  2734.     }

  2735. 

  2736.     *data_size = out_size;

  2737.     return buf_size;

  2738. }

  2739. 

  2740. 

  2741. AVCodec mp2_decoder =

  2742. {

  2743.     "mp2",

  2744.     CODEC_TYPE_AUDIO,

  2745.     CODEC_ID_MP2,

  2746.     sizeof(MPADecodeContext),

  2747.     decode_init,

  2748.     NULL,

  2749.     NULL,

  2750.     decode_frame,

  2751.     CODEC_CAP_PARSE_ONLY,

  2752. };

  2753. 

  2754. AVCodec mp3_decoder =

  2755. {

  2756.     "mp3",

  2757.     CODEC_TYPE_AUDIO,

  2758.     CODEC_ID_MP3,

  2759.     sizeof(MPADecodeContext),

  2760.     decode_init,

  2761.     NULL,

  2762.     NULL,

  2763.     decode_frame,

  2764.     CODEC_CAP_PARSE_ONLY,

  2765. };

  2766. 

  2767. AVCodec mp3adu_decoder =

  2768. {

  2769.     "mp3adu",

  2770.     CODEC_TYPE_AUDIO,

  2771.     CODEC_ID_MP3ADU,

  2772.     sizeof(MPADecodeContext),

  2773.     decode_init,

  2774.     NULL,

  2775.     NULL,

  2776.     decode_frame_adu,

  2777.     CODEC_CAP_PARSE_ONLY,

  2778. };