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.
4212 Commits
32 Branches
2.1GB
Tree: db2fcbbdb3
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'db2fcbbdb3'
${ noResults }
FFmpeg/libavcodec/mpegaudiodec.c

2719 lines
80KB
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. #define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))

  73. //#define MULH(a,b) (((int64_t)(a) * (int64_t)(b))>>32) //gcc 3.4 creates an incredibly bloated mess out of this

  74. static always_inline int MULH(int a, int b){

  75.     return ((int64_t)(a) * (int64_t)(b))>>32;

  76. }

  77. 

  78. #if FRAC_BITS <= 15

  79. typedef int16_t MPA_INT;

  80. #else

  81. typedef int32_t MPA_INT;

  82. #endif

  83. 

  84. /****************/

  85. 

  86. #define HEADER_SIZE 4

  87. #define BACKSTEP_SIZE 512

  88. 

  89. struct GranuleDef;

  90. 

  91. typedef struct MPADecodeContext {

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

  93.     int inbuf_index;

  94.     uint8_t *inbuf_ptr, *inbuf;

  95.     int frame_size;

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

  97.                                    (zero if currently unknown) */

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

  99.     uint32_t free_format_next_header; 

  100.     int error_protection;

  101.     int layer;

  102.     int sample_rate;

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

  104.     int bit_rate;

  105.     int old_frame_size;

  106.     GetBitContext gb;

  107.     int nb_channels;

  108.     int mode;

  109.     int mode_ext;

  110.     int lsf;

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

  112.     int synth_buf_offset[MPA_MAX_CHANNELS];

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

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

  115. #ifdef DEBUG

  116.     int frame_count;

  117. #endif

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

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

  120.     unsigned int dither_state;

  121. } MPADecodeContext;

  122. 

  123. /* layer 3 "granule" */

  124. typedef struct GranuleDef {

  125.     uint8_t scfsi;

  126.     int part2_3_length;

  127.     int big_values;

  128.     int global_gain;

  129.     int scalefac_compress;

  130.     uint8_t block_type;

  131.     uint8_t switch_point;

  132.     int table_select[3];

  133.     int subblock_gain[3];

  134.     uint8_t scalefac_scale;

  135.     uint8_t count1table_select;

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

  137.     int preflag;

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

  139.     uint8_t scale_factors[40];

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

  141. } GranuleDef;

  142. 

  143. #define MODE_EXT_MS_STEREO 2

  144. #define MODE_EXT_I_STEREO  1

  145. 

  146. /* layer 3 huffman tables */

  147. typedef struct HuffTable {

  148.     int xsize;

  149.     const uint8_t *bits;

  150.     const uint16_t *codes;

  151. } HuffTable;

  152. 

  153. #include "mpegaudiodectab.h"

  154. 

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

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

  157. 

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

  159. static VLC huff_vlc[16]; 

  160. static uint8_t *huff_code_table[16];

  161. static VLC huff_quad_vlc[2];

  162. /* computed from band_size_long */

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

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

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

  166. static int8_t  *table_4_3_exp;

  167. static uint32_t *table_4_3_value;

  168. /* intensity stereo coef table */

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

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

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

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

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

  174. 

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

  176. static uint16_t scale_factor_modshift[64];

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

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

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

  180. 

  181. #define SCALE_GEN(v) \

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

  183. 

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

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

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

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

  188. };

  189. 

  190. void ff_mpa_synth_init(MPA_INT *window);

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

  192.     

  193. /* layer 1 unscaling */

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

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

  196. {

  197.     int shift, mod;

  198.     int64_t val;

  199. 

  200.     shift = scale_factor_modshift[scale_factor];

  201.     mod = shift & 3;

  202.     shift >>= 2;

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

  204.     shift += n;

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

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

  207. }

  208. 

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

  210. {

  211.     int shift, mod, val;

  212. 

  213.     shift = scale_factor_modshift[scale_factor];

  214.     mod = shift & 3;

  215.     shift >>= 2;

  216. 

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

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

  219.     if (shift > 0)

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

  221.     return val;

  222. }

  223. 

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

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

  226. {

  227.     unsigned int m;

  228.     int e;

  229. 

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

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

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

  233.     assert(e>=1);

  234.     if (e > 31)

  235.         return 0;

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

  237. 

  238.     return m;

  239. }

  240. 

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

  242. #define DEV_ORDER 13

  243. 

  244. #define POW_FRAC_BITS 24

  245. #define POW_FRAC_ONE    (1 << POW_FRAC_BITS)

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

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

  248. 

  249. static int dev_4_3_coefs[DEV_ORDER];

  250. 

  251. static int pow_mult3[3] = {

  252.     POW_FIX(1.0),

  253.     POW_FIX(1.25992104989487316476),

  254.     POW_FIX(1.58740105196819947474),

  255. };

  256. 

  257. static void int_pow_init(void)

  258. {

  259.     int i, a;

  260. 

  261.     a = POW_FIX(1.0);

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

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

  264.         dev_4_3_coefs[i] = a;

  265.     }

  266. }

  267. 

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

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

  270. {

  271.     int e, er, eq, j;

  272.     int a, a1;

  273.     

  274.     /* renormalize */

  275.     a = i;

  276.     e = POW_FRAC_BITS;

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

  278.         a = a << 1;

  279.         e--;

  280.     }

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

  282.     a1 = 0;

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

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

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

  286.     /* exponent compute (exact) */

  287.     e = e * 4;

  288.     er = e % 3;

  289.     eq = e / 3;

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

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

  292.         a = a >> 1;

  293.         eq++;

  294.     }

  295.     /* convert to float */

  296.     while (a < POW_FRAC_ONE) {

  297.         a = a << 1;

  298.         eq--;

  299.     }

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

  301. #if POW_FRAC_BITS > FRAC_BITS

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

  303.     /* correct overflow */

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

  305.         a = a >> 1;

  306.         eq++;

  307.     }

  308. #endif

  309.     *exp_ptr = eq;

  310.     return a;

  311. }

  312. 

  313. static int decode_init(AVCodecContext * avctx)

  314. {

  315.     MPADecodeContext *s = avctx->priv_data;

  316.     static int init=0;

  317.     int i, j, k;

  318. 

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

  320.     avctx->sample_fmt= SAMPLE_FMT_S32;

  321. #else

  322.     avctx->sample_fmt= SAMPLE_FMT_S16;

  323. #endif    

  324.     

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

  326.         s->compute_antialias= compute_antialias_integer;

  327.     else

  328.         s->compute_antialias= compute_antialias_float;

  329. 

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

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

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

  333.             int shift, mod;

  334.             /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

  335.             shift = (i / 3);

  336.             mod = i % 3;

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

  338.         }

  339. 

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

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

  342.             int n, norm;

  343.             n = i + 2;

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

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

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

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

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

  349.                     i, norm, 

  350.                     scale_factor_mult[i][0],

  351.                     scale_factor_mult[i][1],

  352.                     scale_factor_mult[i][2]);

  353.         }

  354.         

  355. 	ff_mpa_synth_init(window);

  356.         

  357.         /* huffman decode tables */

  358.         huff_code_table[0] = NULL;

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

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

  361. 	    int xsize, x, y;

  362.             unsigned int n;

  363.             uint8_t *code_table;

  364. 

  365.             xsize = h->xsize;

  366.             n = xsize * xsize;

  367.             /* XXX: fail test */

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

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

  370.             

  371.             code_table = av_mallocz(n);

  372.             j = 0;

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

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

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

  376.             }

  377.             huff_code_table[i] = code_table;

  378.         }

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

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

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

  382.         }

  383. 

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

  385.             k = 0;

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

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

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

  389.             }

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

  391.         }

  392. 

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

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

  395.         if(!table_4_3_exp)

  396. 	    return -1;

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

  398.         if(!table_4_3_value)

  399.             return -1;

  400.         

  401.         int_pow_init();

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

  403.             double f, fm;

  404.             int e, m;

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

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

  407.             m = FIXHR(fm*0.5);

  408.             e+= FRAC_BITS - 31 + 5;

  409. 

  410.             /* normalized to FRAC_BITS */

  411.             table_4_3_value[i] = m;

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

  413.             table_4_3_exp[i] = -e;

  414.         }

  415.         

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

  417.             float f;

  418.             int v;

  419.             if (i != 6) {

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

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

  422.             } else {

  423.                 v = FIXR(1.0);

  424.             }

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

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

  427.         }

  428.         /* invalid values */

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

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

  431. 

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

  433.             double f;

  434.             int e, k;

  435. 

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

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

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

  439.                 k = i & 1;

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

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

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

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

  444.             }

  445.         }

  446. 

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

  448.             float ci, cs, ca;

  449.             ci = ci_table[i];

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

  451.             ca = cs * ci;

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

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

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

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

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

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

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

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

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

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

  462.         }

  463. 

  464.         /* compute mdct windows */

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

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

  467.                 double d;

  468.                 

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

  470.                     continue;

  471.                 

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

  473.                 if(j==1){

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

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

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

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

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

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

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

  481.                 }

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

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

  484. 

  485.                 if(j==2)

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

  487.                 else

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

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

  490.             }

  491.         }

  492. 

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

  494.            the sign of the right window coefs */

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

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

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

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

  499.             }

  500.         }

  501. 

  502. #if defined(DEBUG)

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

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

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

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

  507.             printf("\n");

  508.         }

  509. #endif

  510.         init = 1;

  511.     }

  512. 

  513.     s->inbuf_index = 0;

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

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

  516. #ifdef DEBUG

  517.     s->frame_count = 0;

  518. #endif

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

  520.         s->adu_mode = 1;

  521.     return 0;

  522. }

  523. 

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

  525. 

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

  527. 

  528. #define COS0_0  FIXR(0.50060299823519630134)

  529. #define COS0_1  FIXR(0.50547095989754365998)

  530. #define COS0_2  FIXR(0.51544730992262454697)

  531. #define COS0_3  FIXR(0.53104259108978417447)

  532. #define COS0_4  FIXR(0.55310389603444452782)

  533. #define COS0_5  FIXR(0.58293496820613387367)

  534. #define COS0_6  FIXR(0.62250412303566481615)

  535. #define COS0_7  FIXR(0.67480834145500574602)

  536. #define COS0_8  FIXR(0.74453627100229844977)

  537. #define COS0_9  FIXR(0.83934964541552703873)

  538. #define COS0_10 FIXR(0.97256823786196069369)

  539. #define COS0_11 FIXR(1.16943993343288495515)

  540. #define COS0_12 FIXR(1.48416461631416627724)

  541. #define COS0_13 FIXR(2.05778100995341155085)

  542. #define COS0_14 FIXR(3.40760841846871878570)

  543. #define COS0_15 FIXR(10.19000812354805681150)

  544. 

  545. #define COS1_0 FIXR(0.50241928618815570551)

  546. #define COS1_1 FIXR(0.52249861493968888062)

  547. #define COS1_2 FIXR(0.56694403481635770368)

  548. #define COS1_3 FIXR(0.64682178335999012954)

  549. #define COS1_4 FIXR(0.78815462345125022473)

  550. #define COS1_5 FIXR(1.06067768599034747134)

  551. #define COS1_6 FIXR(1.72244709823833392782)

  552. #define COS1_7 FIXR(5.10114861868916385802)

  553. 

  554. #define COS2_0 FIXR(0.50979557910415916894)

  555. #define COS2_1 FIXR(0.60134488693504528054)

  556. #define COS2_2 FIXR(0.89997622313641570463)

  557. #define COS2_3 FIXR(2.56291544774150617881)

  558. 

  559. #define COS3_0 FIXR(0.54119610014619698439)

  560. #define COS3_1 FIXR(1.30656296487637652785)

  561. 

  562. #define COS4_0 FIXR(0.70710678118654752439)

  563. 

  564. /* butterfly operator */

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

  566. {\

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

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

  569.     tab[a] = tmp0;\

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

  571. }

  572. 

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

  574. {\

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

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

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

  578. }

  579. 

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

  581. {\

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

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

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

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

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

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

  588. }

  589. 

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

  591. 

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

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

  594. {

  595.     int tmp0, tmp1;

  596. 

  597.     /* pass 1 */

  598.     BF(0, 31, COS0_0);

  599.     BF(1, 30, COS0_1);

  600.     BF(2, 29, COS0_2);

  601.     BF(3, 28, COS0_3);

  602.     BF(4, 27, COS0_4);

  603.     BF(5, 26, COS0_5);

  604.     BF(6, 25, COS0_6);

  605.     BF(7, 24, COS0_7);

  606.     BF(8, 23, COS0_8);

  607.     BF(9, 22, COS0_9);

  608.     BF(10, 21, COS0_10);

  609.     BF(11, 20, COS0_11);

  610.     BF(12, 19, COS0_12);

  611.     BF(13, 18, COS0_13);

  612.     BF(14, 17, COS0_14);

  613.     BF(15, 16, COS0_15);

  614. 

  615.     /* pass 2 */

  616.     BF(0, 15, COS1_0);

  617.     BF(1, 14, COS1_1);

  618.     BF(2, 13, COS1_2);

  619.     BF(3, 12, COS1_3);

  620.     BF(4, 11, COS1_4);

  621.     BF(5, 10, COS1_5);

  622.     BF(6,  9, COS1_6);

  623.     BF(7,  8, COS1_7);

  624.     

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

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

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

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

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

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

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

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

  633.     

  634.     /* pass 3 */

  635.     BF(0, 7, COS2_0);

  636.     BF(1, 6, COS2_1);

  637.     BF(2, 5, COS2_2);

  638.     BF(3, 4, COS2_3);

  639.     

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

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

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

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

  644.     

  645.     BF(16, 23, COS2_0);

  646.     BF(17, 22, COS2_1);

  647.     BF(18, 21, COS2_2);

  648.     BF(19, 20, COS2_3);

  649.     

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

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

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

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

  654. 

  655.     /* pass 4 */

  656.     BF(0, 3, COS3_0);

  657.     BF(1, 2, COS3_1);

  658.     

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

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

  661.     

  662.     BF(8, 11, COS3_0);

  663.     BF(9, 10, COS3_1);

  664.     

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

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

  667.     

  668.     BF(16, 19, COS3_0);

  669.     BF(17, 18, COS3_1);

  670.     

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

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

  673.     

  674.     BF(24, 27, COS3_0);

  675.     BF(25, 26, COS3_1);

  676.     

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

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

  679.     

  680.     /* pass 5 */

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

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

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

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

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

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

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

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

  689.     

  690.     /* pass 6 */

  691.     

  692.     ADD( 8, 12);

  693.     ADD(12, 10);

  694.     ADD(10, 14);

  695.     ADD(14,  9);

  696.     ADD( 9, 13);

  697.     ADD(13, 11);

  698.     ADD(11, 15);

  699. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  716.     

  717.     ADD(24, 28);

  718.     ADD(28, 26);

  719.     ADD(26, 30);

  720.     ADD(30, 25);

  721.     ADD(25, 29);

  722.     ADD(29, 27);

  723.     ADD(27, 31);

  724. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  741. }

  742. 

  743. #if FRAC_BITS <= 15

  744. 

  745. static inline int round_sample(int *sum)

  746. {

  747.     int sum1;

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

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

  750.     if (sum1 < OUT_MIN)

  751.         sum1 = OUT_MIN;

  752.     else if (sum1 > OUT_MAX)

  753.         sum1 = OUT_MAX;

  754.     return sum1;

  755. }

  756. 

  757. #if defined(ARCH_POWERPC_405)

  758. 

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

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

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

  762. 

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

  764. #define MULS(ra, rb) \

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

  766. 

  767. #else

  768. 

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

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

  771. 

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

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

  774. 

  775. #endif

  776. 

  777. #else

  778. 

  779. static inline int round_sample(int64_t *sum) 

  780. {

  781.     int sum1;

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

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

  784.     if (sum1 < OUT_MIN)

  785.         sum1 = OUT_MIN;

  786.     else if (sum1 > OUT_MAX)

  787.         sum1 = OUT_MAX;

  788.     return sum1;

  789. }

  790. 

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

  792. 

  793. #endif

  794. 

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

  796. {                                               \

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

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

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

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

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

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

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

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

  805. }

  806. 

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

  808. {                                               \

  809.     int tmp;\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  834. }

  835. 

  836. void ff_mpa_synth_init(MPA_INT *window)

  837. {

  838.     int i;

  839. 

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

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

  842.         int v;

  843.         v = mpa_enwindow[i];

  844. #if WFRAC_BITS < 16

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

  846. #endif

  847.         window[i] = v;

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

  849.             v = -v;

  850.         if (i != 0)

  851.             window[512 - i] = v;

  852.     }	

  853. }

  854. 

  855. /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:

  856.    32 samples. */

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

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

  859. 			 MPA_INT *window, int *dither_state,

  860.                          OUT_INT *samples, int incr, 

  861.                          int32_t sb_samples[SBLIMIT])

  862. {

  863.     int32_t tmp[32];

  864.     register MPA_INT *synth_buf;

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

  866.     int j, offset, v;

  867.     OUT_INT *samples2;

  868. #if FRAC_BITS <= 15

  869.     int sum, sum2;

  870. #else

  871.     int64_t sum, sum2;

  872. #endif

  873. 

  874.     dct32(tmp, sb_samples);

  875.     

  876.     offset = *synth_buf_offset;

  877.     synth_buf = synth_buf_ptr + offset;

  878. 

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

  880.         v = tmp[j];

  881. #if FRAC_BITS <= 15

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

  883.            sound */

  884.         if (v > 32767)

  885.             v = 32767;

  886.         else if (v < -32768)

  887.             v = -32768;

  888. #endif

  889.         synth_buf[j] = v;

  890.     }

  891.     /* copy to avoid wrap */

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

  893. 

  894.     samples2 = samples + 31 * incr;

  895.     w = window;

  896.     w2 = window + 31;

  897. 

  898.     sum = *dither_state;

  899.     p = synth_buf + 16;

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

  901.     p = synth_buf + 48;

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

  903.     *samples = round_sample(&sum);

  904.     samples += incr;

  905.     w++;

  906. 

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

  908.        access per two sample */

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

  910.         sum2 = 0;

  911.         p = synth_buf + 16 + j;

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

  913.         p = synth_buf + 48 - j;

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

  915. 

  916.         *samples = round_sample(&sum);

  917.         samples += incr;

  918.         sum += sum2;

  919.         *samples2 = round_sample(&sum);

  920.         samples2 -= incr;

  921.         w++;

  922.         w2--;

  923.     }

  924.     

  925.     p = synth_buf + 32;

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

  927.     *samples = round_sample(&sum);

  928.     *dither_state= sum;

  929. 

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

  931.     *synth_buf_offset = offset;

  932. }

  933. 

  934. #define C3 FIXHR(0.86602540378443864676/2)

  935. 

  936. /* 0.5 / cos(pi*(2*i+1)/36) */

  937. static const int icos36[9] = {

  938.     FIXR(0.50190991877167369479),

  939.     FIXR(0.51763809020504152469), //0

  940.     FIXR(0.55168895948124587824),

  941.     FIXR(0.61038729438072803416),

  942.     FIXR(0.70710678118654752439), //1

  943.     FIXR(0.87172339781054900991),

  944.     FIXR(1.18310079157624925896),

  945.     FIXR(1.93185165257813657349), //2

  946.     FIXR(5.73685662283492756461),

  947. };

  948. 

  949. /* 12 points IMDCT. We compute it "by hand" by factorizing obvious

  950.    cases. */

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

  952. {

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

  954. 

  955.     in0= in[0*3];

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

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

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

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

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

  961.     in5 += in3;

  962.     in3 += in1;

  963. 

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

  965.     in3= MULH(2*in3, C3);

  966.     

  967.     t1 = in0 - in4;

  968.     t2 = MULL(in1 - in5, icos36[4]);

  969. 

  970.     out[ 7]= 

  971.     out[10]= t1 + t2;

  972.     out[ 1]=

  973.     out[ 4]= t1 - t2;

  974. 

  975.     in0 += in4>>1;

  976.     in4 = in0 + in2;

  977.     in1 += in5>>1;

  978.     in5 = MULL(in1 + in3, icos36[1]);    

  979.     out[ 8]= 

  980.     out[ 9]= in4 + in5;

  981.     out[ 2]=

  982.     out[ 3]= in4 - in5;

  983.     

  984.     in0 -= in2;

  985.     in1 = MULL(in1 - in3, icos36[7]);

  986.     out[ 0]=

  987.     out[ 5]= in0 - in1;

  988.     out[ 6]=

  989.     out[11]= in0 + in1;    

  990. }

  991. 

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

  993. #define C1 FIXHR(0.98480775301220805936/2)

  994. #define C2 FIXHR(0.93969262078590838405/2)

  995. #define C3 FIXHR(0.86602540378443864676/2)

  996. #define C4 FIXHR(0.76604444311897803520/2)

  997. #define C5 FIXHR(0.64278760968653932632/2)

  998. #define C6 FIXHR(0.5/2)

  999. #define C7 FIXHR(0.34202014332566873304/2)

  1000. #define C8 FIXHR(0.17364817766693034885/2)

  1001. 

  1002. 

  1003. /* using Lee like decomposition followed by hand coded 9 points DCT */

  1004. static void imdct36(int *out, int *buf, int *in, int *win)

  1005. {

  1006.     int i, j, t0, t1, t2, t3, s0, s1, s2, s3;

  1007.     int tmp[18], *tmp1, *in1;

  1008. 

  1009.     for(i=17;i>=1;i--)

  1010.         in[i] += in[i-1];

  1011.     for(i=17;i>=3;i-=2)

  1012.         in[i] += in[i-2];

  1013. 

  1014.     for(j=0;j<2;j++) {

  1015.         tmp1 = tmp + j;

  1016.         in1 = in + j;

  1017. #if 0

  1018. //more accurate but slower

  1019.         int64_t t0, t1, t2, t3;

  1020.         t2 = in1[2*4] + in1[2*8] - in1[2*2];

  1021.         

  1022.         t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;

  1023.         t1 = in1[2*0] - in1[2*6];

  1024.         tmp1[ 6] = t1 - (t2>>1);

  1025.         tmp1[16] = t1 + t2;

  1026. 

  1027.         t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);

  1028.         t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);

  1029.         t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);

  1030.         

  1031.         tmp1[10] = (t3 - t0 - t2) >> 32;

  1032.         tmp1[ 2] = (t3 + t0 + t1) >> 32;

  1033.         tmp1[14] = (t3 + t2 - t1) >> 32;

  1034.         

  1035.         tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);

  1036.         t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);

  1037.         t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);

  1038.         t0 = MUL64(2*in1[2*3], C3);

  1039. 

  1040.         t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);

  1041. 

  1042.         tmp1[ 0] = (t2 + t3 + t0) >> 32;

  1043.         tmp1[12] = (t2 + t1 - t0) >> 32;

  1044.         tmp1[ 8] = (t3 - t1 - t0) >> 32;

  1045. #else

  1046.         t2 = in1[2*4] + in1[2*8] - in1[2*2];

  1047.         

  1048.         t3 = in1[2*0] + (in1[2*6]>>1);

  1049.         t1 = in1[2*0] - in1[2*6];

  1050.         tmp1[ 6] = t1 - (t2>>1);

  1051.         tmp1[16] = t1 + t2;

  1052. 

  1053.         t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);

  1054.         t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);

  1055.         t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);

  1056.         

  1057.         tmp1[10] = t3 - t0 - t2;

  1058.         tmp1[ 2] = t3 + t0 + t1;

  1059.         tmp1[14] = t3 + t2 - t1;

  1060.         

  1061.         tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);

  1062.         t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);

  1063.         t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);

  1064.         t0 = MULH(2*in1[2*3], C3);

  1065. 

  1066.         t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);

  1067. 

  1068.         tmp1[ 0] = t2 + t3 + t0;

  1069.         tmp1[12] = t2 + t1 - t0;

  1070.         tmp1[ 8] = t3 - t1 - t0;

  1071. #endif

  1072.     }

  1073. 

  1074.     i = 0;

  1075.     for(j=0;j<4;j++) {

  1076.         t0 = tmp[i];

  1077.         t1 = tmp[i + 2];

  1078.         s0 = t1 + t0;

  1079.         s2 = t1 - t0;

  1080. 

  1081.         t2 = tmp[i + 1];

  1082.         t3 = tmp[i + 3];

  1083.         s1 = MULL(t3 + t2, icos36[j]);

  1084.         s3 = MULL(t3 - t2, icos36[8 - j]);

  1085.         

  1086.         t0 = s0 + s1;

  1087.         t1 = s0 - s1;

  1088.         out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];

  1089.         out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];

  1090.         buf[9 + j] = MULH(t0, win[18 + 9 + j]);

  1091.         buf[8 - j] = MULH(t0, win[18 + 8 - j]);

  1092.         

  1093.         t0 = s2 + s3;

  1094.         t1 = s2 - s3;

  1095.         out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];

  1096.         out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];

  1097.         buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);

  1098.         buf[      + j] = MULH(t0, win[18         + j]);

  1099.         i += 4;

  1100.     }

  1101. 

  1102.     s0 = tmp[16];

  1103.     s1 = MULL(tmp[17], icos36[4]);

  1104.     t0 = s0 + s1;

  1105.     t1 = s0 - s1;

  1106.     out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];

  1107.     out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];

  1108.     buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);

  1109.     buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);

  1110. }

  1111. 

  1112. /* header decoding. MUST check the header before because no

  1113.    consistency check is done there. Return 1 if free format found and

  1114.    that the frame size must be computed externally */

  1115. static int decode_header(MPADecodeContext *s, uint32_t header)

  1116. {

  1117.     int sample_rate, frame_size, mpeg25, padding;

  1118.     int sample_rate_index, bitrate_index;

  1119.     if (header & (1<<20)) {

  1120.         s->lsf = (header & (1<<19)) ? 0 : 1;

  1121.         mpeg25 = 0;

  1122.     } else {

  1123.         s->lsf = 1;

  1124.         mpeg25 = 1;

  1125.     }

  1126.     

  1127.     s->layer = 4 - ((header >> 17) & 3);

  1128.     /* extract frequency */

  1129.     sample_rate_index = (header >> 10) & 3;

  1130.     sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);

  1131.     sample_rate_index += 3 * (s->lsf + mpeg25);

  1132.     s->sample_rate_index = sample_rate_index;

  1133.     s->error_protection = ((header >> 16) & 1) ^ 1;

  1134.     s->sample_rate = sample_rate;

  1135. 

  1136.     bitrate_index = (header >> 12) & 0xf;

  1137.     padding = (header >> 9) & 1;

  1138.     //extension = (header >> 8) & 1;

  1139.     s->mode = (header >> 6) & 3;

  1140.     s->mode_ext = (header >> 4) & 3;

  1141.     //copyright = (header >> 3) & 1;

  1142.     //original = (header >> 2) & 1;

  1143.     //emphasis = header & 3;

  1144. 

  1145.     if (s->mode == MPA_MONO)

  1146.         s->nb_channels = 1;

  1147.     else

  1148.         s->nb_channels = 2;

  1149.     

  1150.     if (bitrate_index != 0) {

  1151.         frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];

  1152.         s->bit_rate = frame_size * 1000;

  1153.         switch(s->layer) {

  1154.         case 1:

  1155.             frame_size = (frame_size * 12000) / sample_rate;

  1156.             frame_size = (frame_size + padding) * 4;

  1157.             break;

  1158.         case 2:

  1159.             frame_size = (frame_size * 144000) / sample_rate;

  1160.             frame_size += padding;

  1161.             break;

  1162.         default:

  1163.         case 3:

  1164.             frame_size = (frame_size * 144000) / (sample_rate << s->lsf);

  1165.             frame_size += padding;

  1166.             break;

  1167.         }

  1168.         s->frame_size = frame_size;

  1169.     } else {

  1170.         /* if no frame size computed, signal it */

  1171.         if (!s->free_format_frame_size)

  1172.             return 1;

  1173.         /* free format: compute bitrate and real frame size from the

  1174.            frame size we extracted by reading the bitstream */

  1175.         s->frame_size = s->free_format_frame_size;

  1176.         switch(s->layer) {

  1177.         case 1:

  1178.             s->frame_size += padding  * 4;

  1179.             s->bit_rate = (s->frame_size * sample_rate) / 48000;

  1180.             break;

  1181.         case 2:

  1182.             s->frame_size += padding;

  1183.             s->bit_rate = (s->frame_size * sample_rate) / 144000;

  1184.             break;

  1185.         default:

  1186.         case 3:

  1187.             s->frame_size += padding;

  1188.             s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;

  1189.             break;

  1190.         }

  1191.     }

  1192.     

  1193. #if defined(DEBUG)

  1194.     printf("layer%d, %d Hz, %d kbits/s, ",

  1195.            s->layer, s->sample_rate, s->bit_rate);

  1196.     if (s->nb_channels == 2) {

  1197.         if (s->layer == 3) {

  1198.             if (s->mode_ext & MODE_EXT_MS_STEREO)

  1199.                 printf("ms-");

  1200.             if (s->mode_ext & MODE_EXT_I_STEREO)

  1201.                 printf("i-");

  1202.         }

  1203.         printf("stereo");

  1204.     } else {

  1205.         printf("mono");

  1206.     }

  1207.     printf("\n");

  1208. #endif

  1209.     return 0;

  1210. }

  1211. 

  1212. /* useful helper to get mpeg audio stream infos. Return -1 if error in

  1213.    header, otherwise the coded frame size in bytes */

  1214. int mpa_decode_header(AVCodecContext *avctx, uint32_t head)

  1215. {

  1216.     MPADecodeContext s1, *s = &s1;

  1217.     memset( s, 0, sizeof(MPADecodeContext) );

  1218. 

  1219.     if (ff_mpa_check_header(head) != 0)

  1220.         return -1;

  1221. 

  1222.     if (decode_header(s, head) != 0) {

  1223.         return -1;

  1224.     }

  1225. 

  1226.     switch(s->layer) {

  1227.     case 1:

  1228.         avctx->frame_size = 384;

  1229.         break;

  1230.     case 2:

  1231.         avctx->frame_size = 1152;

  1232.         break;

  1233.     default:

  1234.     case 3:

  1235.         if (s->lsf)

  1236.             avctx->frame_size = 576;

  1237.         else

  1238.             avctx->frame_size = 1152;

  1239.         break;

  1240.     }

  1241. 

  1242.     avctx->sample_rate = s->sample_rate;

  1243.     avctx->channels = s->nb_channels;

  1244.     avctx->bit_rate = s->bit_rate;

  1245.     avctx->sub_id = s->layer;

  1246.     return s->frame_size;

  1247. }

  1248. 

  1249. /* return the number of decoded frames */

  1250. static int mp_decode_layer1(MPADecodeContext *s)

  1251. {

  1252.     int bound, i, v, n, ch, j, mant;

  1253.     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];

  1254.     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

  1255. 

  1256.     if (s->mode == MPA_JSTEREO) 

  1257.         bound = (s->mode_ext + 1) * 4;

  1258.     else

  1259.         bound = SBLIMIT;

  1260. 

  1261.     /* allocation bits */

  1262.     for(i=0;i<bound;i++) {

  1263.         for(ch=0;ch<s->nb_channels;ch++) {

  1264.             allocation[ch][i] = get_bits(&s->gb, 4);

  1265.         }

  1266.     }

  1267.     for(i=bound;i<SBLIMIT;i++) {

  1268.         allocation[0][i] = get_bits(&s->gb, 4);

  1269.     }

  1270. 

  1271.     /* scale factors */

  1272.     for(i=0;i<bound;i++) {

  1273.         for(ch=0;ch<s->nb_channels;ch++) {

  1274.             if (allocation[ch][i])

  1275.                 scale_factors[ch][i] = get_bits(&s->gb, 6);

  1276.         }

  1277.     }

  1278.     for(i=bound;i<SBLIMIT;i++) {

  1279.         if (allocation[0][i]) {

  1280.             scale_factors[0][i] = get_bits(&s->gb, 6);

  1281.             scale_factors[1][i] = get_bits(&s->gb, 6);

  1282.         }

  1283.     }

  1284.     

  1285.     /* compute samples */

  1286.     for(j=0;j<12;j++) {

  1287.         for(i=0;i<bound;i++) {

  1288.             for(ch=0;ch<s->nb_channels;ch++) {

  1289.                 n = allocation[ch][i];

  1290.                 if (n) {

  1291.                     mant = get_bits(&s->gb, n + 1);

  1292.                     v = l1_unscale(n, mant, scale_factors[ch][i]);

  1293.                 } else {

  1294.                     v = 0;

  1295.                 }

  1296.                 s->sb_samples[ch][j][i] = v;

  1297.             }

  1298.         }

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

  1300.             n = allocation[0][i];

  1301.             if (n) {

  1302.                 mant = get_bits(&s->gb, n + 1);

  1303.                 v = l1_unscale(n, mant, scale_factors[0][i]);

  1304.                 s->sb_samples[0][j][i] = v;

  1305.                 v = l1_unscale(n, mant, scale_factors[1][i]);

  1306.                 s->sb_samples[1][j][i] = v;

  1307.             } else {

  1308.                 s->sb_samples[0][j][i] = 0;

  1309.                 s->sb_samples[1][j][i] = 0;

  1310.             }

  1311.         }

  1312.     }

  1313.     return 12;

  1314. }

  1315. 

  1316. /* bitrate is in kb/s */

  1317. int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)

  1318. {

  1319.     int ch_bitrate, table;

  1320.     

  1321.     ch_bitrate = bitrate / nb_channels;

  1322.     if (!lsf) {

  1323.         if ((freq == 48000 && ch_bitrate >= 56) ||

  1324.             (ch_bitrate >= 56 && ch_bitrate <= 80)) 

  1325.             table = 0;

  1326.         else if (freq != 48000 && ch_bitrate >= 96) 

  1327.             table = 1;

  1328.         else if (freq != 32000 && ch_bitrate <= 48) 

  1329.             table = 2;

  1330.         else 

  1331.             table = 3;

  1332.     } else {

  1333.         table = 4;

  1334.     }

  1335.     return table;

  1336. }

  1337. 

  1338. static int mp_decode_layer2(MPADecodeContext *s)

  1339. {

  1340.     int sblimit; /* number of used subbands */

  1341.     const unsigned char *alloc_table;

  1342.     int table, bit_alloc_bits, i, j, ch, bound, v;

  1343.     unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];

  1344.     unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];

  1345.     unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;

  1346.     int scale, qindex, bits, steps, k, l, m, b;

  1347. 

  1348.     /* select decoding table */

  1349.     table = l2_select_table(s->bit_rate / 1000, s->nb_channels, 

  1350.                             s->sample_rate, s->lsf);

  1351.     sblimit = sblimit_table[table];

  1352.     alloc_table = alloc_tables[table];

  1353. 

  1354.     if (s->mode == MPA_JSTEREO) 

  1355.         bound = (s->mode_ext + 1) * 4;

  1356.     else

  1357.         bound = sblimit;

  1358. 

  1359.     dprintf("bound=%d sblimit=%d\n", bound, sblimit);

  1360. 

  1361.     /* sanity check */

  1362.     if( bound > sblimit ) bound = sblimit;

  1363. 

  1364.     /* parse bit allocation */

  1365.     j = 0;

  1366.     for(i=0;i<bound;i++) {

  1367.         bit_alloc_bits = alloc_table[j];

  1368.         for(ch=0;ch<s->nb_channels;ch++) {

  1369.             bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);

  1370.         }

  1371.         j += 1 << bit_alloc_bits;

  1372.     }

  1373.     for(i=bound;i<sblimit;i++) {

  1374.         bit_alloc_bits = alloc_table[j];

  1375.         v = get_bits(&s->gb, bit_alloc_bits);

  1376.         bit_alloc[0][i] = v;

  1377.         bit_alloc[1][i] = v;

  1378.         j += 1 << bit_alloc_bits;

  1379.     }

  1380. 

  1381. #ifdef DEBUG

  1382.     {

  1383.         for(ch=0;ch<s->nb_channels;ch++) {

  1384.             for(i=0;i<sblimit;i++)

  1385.                 printf(" %d", bit_alloc[ch][i]);

  1386.             printf("\n");

  1387.         }

  1388.     }

  1389. #endif

  1390. 

  1391.     /* scale codes */

  1392.     for(i=0;i<sblimit;i++) {

  1393.         for(ch=0;ch<s->nb_channels;ch++) {

  1394.             if (bit_alloc[ch][i]) 

  1395.                 scale_code[ch][i] = get_bits(&s->gb, 2);

  1396.         }

  1397.     }

  1398.     

  1399.     /* scale factors */

  1400.     for(i=0;i<sblimit;i++) {

  1401.         for(ch=0;ch<s->nb_channels;ch++) {

  1402.             if (bit_alloc[ch][i]) {

  1403.                 sf = scale_factors[ch][i];

  1404.                 switch(scale_code[ch][i]) {

  1405.                 default:

  1406.                 case 0:

  1407.                     sf[0] = get_bits(&s->gb, 6);

  1408.                     sf[1] = get_bits(&s->gb, 6);

  1409.                     sf[2] = get_bits(&s->gb, 6);

  1410.                     break;

  1411.                 case 2:

  1412.                     sf[0] = get_bits(&s->gb, 6);

  1413.                     sf[1] = sf[0];

  1414.                     sf[2] = sf[0];

  1415.                     break;

  1416.                 case 1:

  1417.                     sf[0] = get_bits(&s->gb, 6);

  1418.                     sf[2] = get_bits(&s->gb, 6);

  1419.                     sf[1] = sf[0];

  1420.                     break;

  1421.                 case 3:

  1422.                     sf[0] = get_bits(&s->gb, 6);

  1423.                     sf[2] = get_bits(&s->gb, 6);

  1424.                     sf[1] = sf[2];

  1425.                     break;

  1426.                 }

  1427.             }

  1428.         }

  1429.     }

  1430. 

  1431. #ifdef DEBUG

  1432.     for(ch=0;ch<s->nb_channels;ch++) {

  1433.         for(i=0;i<sblimit;i++) {

  1434.             if (bit_alloc[ch][i]) {

  1435.                 sf = scale_factors[ch][i];

  1436.                 printf(" %d %d %d", sf[0], sf[1], sf[2]);

  1437.             } else {

  1438.                 printf(" -");

  1439.             }

  1440.         }

  1441.         printf("\n");

  1442.     }

  1443. #endif

  1444. 

  1445.     /* samples */

  1446.     for(k=0;k<3;k++) {

  1447.         for(l=0;l<12;l+=3) {

  1448.             j = 0;

  1449.             for(i=0;i<bound;i++) {

  1450.                 bit_alloc_bits = alloc_table[j];

  1451.                 for(ch=0;ch<s->nb_channels;ch++) {

  1452.                     b = bit_alloc[ch][i];

  1453.                     if (b) {

  1454.                         scale = scale_factors[ch][i][k];

  1455.                         qindex = alloc_table[j+b];

  1456.                         bits = quant_bits[qindex];

  1457.                         if (bits < 0) {

  1458.                             /* 3 values at the same time */

  1459.                             v = get_bits(&s->gb, -bits);

  1460.                             steps = quant_steps[qindex];

  1461.                             s->sb_samples[ch][k * 12 + l + 0][i] = 

  1462.                                 l2_unscale_group(steps, v % steps, scale);

  1463.                             v = v / steps;

  1464.                             s->sb_samples[ch][k * 12 + l + 1][i] = 

  1465.                                 l2_unscale_group(steps, v % steps, scale);

  1466.                             v = v / steps;

  1467.                             s->sb_samples[ch][k * 12 + l + 2][i] = 

  1468.                                 l2_unscale_group(steps, v, scale);

  1469.                         } else {

  1470.                             for(m=0;m<3;m++) {

  1471.                                 v = get_bits(&s->gb, bits);

  1472.                                 v = l1_unscale(bits - 1, v, scale);

  1473.                                 s->sb_samples[ch][k * 12 + l + m][i] = v;

  1474.                             }

  1475.                         }

  1476.                     } else {

  1477.                         s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1478.                         s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1479.                         s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1480.                     }

  1481.                 }

  1482.                 /* next subband in alloc table */

  1483.                 j += 1 << bit_alloc_bits; 

  1484.             }

  1485.             /* XXX: find a way to avoid this duplication of code */

  1486.             for(i=bound;i<sblimit;i++) {

  1487.                 bit_alloc_bits = alloc_table[j];

  1488.                 b = bit_alloc[0][i];

  1489.                 if (b) {

  1490.                     int mant, scale0, scale1;

  1491.                     scale0 = scale_factors[0][i][k];

  1492.                     scale1 = scale_factors[1][i][k];

  1493.                     qindex = alloc_table[j+b];

  1494.                     bits = quant_bits[qindex];

  1495.                     if (bits < 0) {

  1496.                         /* 3 values at the same time */

  1497.                         v = get_bits(&s->gb, -bits);

  1498.                         steps = quant_steps[qindex];

  1499.                         mant = v % steps;

  1500.                         v = v / steps;

  1501.                         s->sb_samples[0][k * 12 + l + 0][i] = 

  1502.                             l2_unscale_group(steps, mant, scale0);

  1503.                         s->sb_samples[1][k * 12 + l + 0][i] = 

  1504.                             l2_unscale_group(steps, mant, scale1);

  1505.                         mant = v % steps;

  1506.                         v = v / steps;

  1507.                         s->sb_samples[0][k * 12 + l + 1][i] = 

  1508.                             l2_unscale_group(steps, mant, scale0);

  1509.                         s->sb_samples[1][k * 12 + l + 1][i] = 

  1510.                             l2_unscale_group(steps, mant, scale1);

  1511.                         s->sb_samples[0][k * 12 + l + 2][i] = 

  1512.                             l2_unscale_group(steps, v, scale0);

  1513.                         s->sb_samples[1][k * 12 + l + 2][i] = 

  1514.                             l2_unscale_group(steps, v, scale1);

  1515.                     } else {

  1516.                         for(m=0;m<3;m++) {

  1517.                             mant = get_bits(&s->gb, bits);

  1518.                             s->sb_samples[0][k * 12 + l + m][i] = 

  1519.                                 l1_unscale(bits - 1, mant, scale0);

  1520.                             s->sb_samples[1][k * 12 + l + m][i] = 

  1521.                                 l1_unscale(bits - 1, mant, scale1);

  1522.                         }

  1523.                     }

  1524.                 } else {

  1525.                     s->sb_samples[0][k * 12 + l + 0][i] = 0;

  1526.                     s->sb_samples[0][k * 12 + l + 1][i] = 0;

  1527.                     s->sb_samples[0][k * 12 + l + 2][i] = 0;

  1528.                     s->sb_samples[1][k * 12 + l + 0][i] = 0;

  1529.                     s->sb_samples[1][k * 12 + l + 1][i] = 0;

  1530.                     s->sb_samples[1][k * 12 + l + 2][i] = 0;

  1531.                 }

  1532.                 /* next subband in alloc table */

  1533.                 j += 1 << bit_alloc_bits; 

  1534.             }

  1535.             /* fill remaining samples to zero */

  1536.             for(i=sblimit;i<SBLIMIT;i++) {

  1537.                 for(ch=0;ch<s->nb_channels;ch++) {

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

  1539.                     s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1540.                     s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1541.                 }

  1542.             }

  1543.         }

  1544.     }

  1545.     return 3 * 12;

  1546. }

  1547. 

  1548. /*

  1549.  * Seek back in the stream for backstep bytes (at most 511 bytes)

  1550.  */

  1551. static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)

  1552. {

  1553.     uint8_t *ptr;

  1554. 

  1555.     /* compute current position in stream */

  1556.     ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));

  1557. 

  1558.     /* copy old data before current one */

  1559.     ptr -= backstep;

  1560.     memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] + 

  1561.            BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);

  1562.     /* init get bits again */

  1563.     init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);

  1564. 

  1565.     /* prepare next buffer */

  1566.     s->inbuf_index ^= 1;

  1567.     s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];

  1568.     s->old_frame_size = s->frame_size;

  1569. }

  1570. 

  1571. static inline void lsf_sf_expand(int *slen,

  1572.                                  int sf, int n1, int n2, int n3)

  1573. {

  1574.     if (n3) {

  1575.         slen[3] = sf % n3;

  1576.         sf /= n3;

  1577.     } else {

  1578.         slen[3] = 0;

  1579.     }

  1580.     if (n2) {

  1581.         slen[2] = sf % n2;

  1582.         sf /= n2;

  1583.     } else {

  1584.         slen[2] = 0;

  1585.     }

  1586.     slen[1] = sf % n1;

  1587.     sf /= n1;

  1588.     slen[0] = sf;

  1589. }

  1590. 

  1591. static void exponents_from_scale_factors(MPADecodeContext *s, 

  1592.                                          GranuleDef *g,

  1593.                                          int16_t *exponents)

  1594. {

  1595.     const uint8_t *bstab, *pretab;

  1596.     int len, i, j, k, l, v0, shift, gain, gains[3];

  1597.     int16_t *exp_ptr;

  1598. 

  1599.     exp_ptr = exponents;

  1600.     gain = g->global_gain - 210;

  1601.     shift = g->scalefac_scale + 1;

  1602. 

  1603.     bstab = band_size_long[s->sample_rate_index];

  1604.     pretab = mpa_pretab[g->preflag];

  1605.     for(i=0;i<g->long_end;i++) {

  1606.         v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);

  1607.         len = bstab[i];

  1608.         for(j=len;j>0;j--)

  1609.             *exp_ptr++ = v0;

  1610.     }

  1611. 

  1612.     if (g->short_start < 13) {

  1613.         bstab = band_size_short[s->sample_rate_index];

  1614.         gains[0] = gain - (g->subblock_gain[0] << 3);

  1615.         gains[1] = gain - (g->subblock_gain[1] << 3);

  1616.         gains[2] = gain - (g->subblock_gain[2] << 3);

  1617.         k = g->long_end;

  1618.         for(i=g->short_start;i<13;i++) {

  1619.             len = bstab[i];

  1620.             for(l=0;l<3;l++) {

  1621.                 v0 = gains[l] - (g->scale_factors[k++] << shift);

  1622.                 for(j=len;j>0;j--)

  1623.                 *exp_ptr++ = v0;

  1624.             }

  1625.         }

  1626.     }

  1627. }

  1628. 

  1629. /* handle n = 0 too */

  1630. static inline int get_bitsz(GetBitContext *s, int n)

  1631. {

  1632.     if (n == 0)

  1633.         return 0;

  1634.     else

  1635.         return get_bits(s, n);

  1636. }

  1637. 

  1638. static int huffman_decode(MPADecodeContext *s, GranuleDef *g,

  1639.                           int16_t *exponents, int end_pos)

  1640. {

  1641.     int s_index;

  1642.     int linbits, code, x, y, l, v, i, j, k, pos;

  1643.     GetBitContext last_gb;

  1644.     VLC *vlc;

  1645.     uint8_t *code_table;

  1646. 

  1647.     /* low frequencies (called big values) */

  1648.     s_index = 0;

  1649.     for(i=0;i<3;i++) {

  1650.         j = g->region_size[i];

  1651.         if (j == 0)

  1652.             continue;

  1653.         /* select vlc table */

  1654.         k = g->table_select[i];

  1655.         l = mpa_huff_data[k][0];

  1656.         linbits = mpa_huff_data[k][1];

  1657.         vlc = &huff_vlc[l];

  1658.         code_table = huff_code_table[l];

  1659. 

  1660.         /* read huffcode and compute each couple */

  1661.         for(;j>0;j--) {

  1662.             if (get_bits_count(&s->gb) >= end_pos)

  1663.                 break;

  1664.             if (code_table) {

  1665.                 code = get_vlc(&s->gb, vlc);

  1666.                 if (code < 0)

  1667.                     return -1;

  1668.                 y = code_table[code];

  1669.                 x = y >> 4;

  1670.                 y = y & 0x0f;

  1671.             } else {

  1672.                 x = 0;

  1673.                 y = 0;

  1674.             }

  1675.             dprintf("region=%d n=%d x=%d y=%d exp=%d\n", 

  1676.                     i, g->region_size[i] - j, x, y, exponents[s_index]);

  1677.             if (x) {

  1678.                 if (x == 15)

  1679.                     x += get_bitsz(&s->gb, linbits);

  1680.                 v = l3_unscale(x, exponents[s_index]);

  1681.                 if (get_bits1(&s->gb))

  1682.                     v = -v;

  1683.             } else {

  1684.                 v = 0;

  1685.             }

  1686.             g->sb_hybrid[s_index++] = v;

  1687.             if (y) {

  1688.                 if (y == 15)

  1689.                     y += get_bitsz(&s->gb, linbits);

  1690.                 v = l3_unscale(y, exponents[s_index]);

  1691.                 if (get_bits1(&s->gb))

  1692.                     v = -v;

  1693.             } else {

  1694.                 v = 0;

  1695.             }

  1696.             g->sb_hybrid[s_index++] = v;

  1697.         }

  1698.     }

  1699.             

  1700.     /* high frequencies */

  1701.     vlc = &huff_quad_vlc[g->count1table_select];

  1702.     last_gb.buffer = NULL;

  1703.     while (s_index <= 572) {

  1704.         pos = get_bits_count(&s->gb);

  1705.         if (pos >= end_pos) {

  1706.             if (pos > end_pos && last_gb.buffer != NULL) {

  1707.                 /* some encoders generate an incorrect size for this

  1708.                    part. We must go back into the data */

  1709.                 s_index -= 4;

  1710.                 s->gb = last_gb;

  1711.             }

  1712.             break;

  1713.         }

  1714.         last_gb= s->gb;

  1715. 

  1716.         code = get_vlc(&s->gb, vlc);

  1717.         dprintf("t=%d code=%d\n", g->count1table_select, code);

  1718.         if (code < 0)

  1719.             return -1;

  1720.         for(i=0;i<4;i++) {

  1721.             if (code & (8 >> i)) {

  1722.                 /* non zero value. Could use a hand coded function for

  1723.                    'one' value */

  1724.                 v = l3_unscale(1, exponents[s_index]);

  1725.                 if(get_bits1(&s->gb))

  1726.                     v = -v;

  1727.             } else {

  1728.                 v = 0;

  1729.             }

  1730.             g->sb_hybrid[s_index++] = v;

  1731.         }

  1732.     }

  1733.     while (s_index < 576)

  1734.         g->sb_hybrid[s_index++] = 0;

  1735.     return 0;

  1736. }

  1737. 

  1738. /* Reorder short blocks from bitstream order to interleaved order. It

  1739.    would be faster to do it in parsing, but the code would be far more

  1740.    complicated */

  1741. static void reorder_block(MPADecodeContext *s, GranuleDef *g)

  1742. {

  1743.     int i, j, k, len;

  1744.     int32_t *ptr, *dst, *ptr1;

  1745.     int32_t tmp[576];

  1746. 

  1747.     if (g->block_type != 2)

  1748.         return;

  1749. 

  1750.     if (g->switch_point) {

  1751.         if (s->sample_rate_index != 8) {

  1752.             ptr = g->sb_hybrid + 36;

  1753.         } else {

  1754.             ptr = g->sb_hybrid + 48;

  1755.         }

  1756.     } else {

  1757.         ptr = g->sb_hybrid;

  1758.     }

  1759.     

  1760.     for(i=g->short_start;i<13;i++) {

  1761.         len = band_size_short[s->sample_rate_index][i];

  1762.         ptr1 = ptr;

  1763.         for(k=0;k<3;k++) {

  1764.             dst = tmp + k;

  1765.             for(j=len;j>0;j--) {

  1766.                 *dst = *ptr++;

  1767.                 dst += 3;

  1768.             }

  1769.         }

  1770.         memcpy(ptr1, tmp, len * 3 * sizeof(int32_t));

  1771.     }

  1772. }

  1773. 

  1774. #define ISQRT2 FIXR(0.70710678118654752440)

  1775. 

  1776. static void compute_stereo(MPADecodeContext *s,

  1777.                            GranuleDef *g0, GranuleDef *g1)

  1778. {

  1779.     int i, j, k, l;

  1780.     int32_t v1, v2;

  1781.     int sf_max, tmp0, tmp1, sf, len, non_zero_found;

  1782.     int32_t (*is_tab)[16];

  1783.     int32_t *tab0, *tab1;

  1784.     int non_zero_found_short[3];

  1785. 

  1786.     /* intensity stereo */

  1787.     if (s->mode_ext & MODE_EXT_I_STEREO) {

  1788.         if (!s->lsf) {

  1789.             is_tab = is_table;

  1790.             sf_max = 7;

  1791.         } else {

  1792.             is_tab = is_table_lsf[g1->scalefac_compress & 1];

  1793.             sf_max = 16;

  1794.         }

  1795.             

  1796.         tab0 = g0->sb_hybrid + 576;

  1797.         tab1 = g1->sb_hybrid + 576;

  1798. 

  1799.         non_zero_found_short[0] = 0;

  1800.         non_zero_found_short[1] = 0;

  1801.         non_zero_found_short[2] = 0;

  1802.         k = (13 - g1->short_start) * 3 + g1->long_end - 3;

  1803.         for(i = 12;i >= g1->short_start;i--) {

  1804.             /* for last band, use previous scale factor */

  1805.             if (i != 11)

  1806.                 k -= 3;

  1807.             len = band_size_short[s->sample_rate_index][i];

  1808.             for(l=2;l>=0;l--) {

  1809.                 tab0 -= len;

  1810.                 tab1 -= len;

  1811.                 if (!non_zero_found_short[l]) {

  1812.                     /* test if non zero band. if so, stop doing i-stereo */

  1813.                     for(j=0;j<len;j++) {

  1814.                         if (tab1[j] != 0) {

  1815.                             non_zero_found_short[l] = 1;

  1816.                             goto found1;

  1817.                         }

  1818.                     }

  1819.                     sf = g1->scale_factors[k + l];

  1820.                     if (sf >= sf_max)

  1821.                         goto found1;

  1822. 

  1823.                     v1 = is_tab[0][sf];

  1824.                     v2 = is_tab[1][sf];

  1825.                     for(j=0;j<len;j++) {

  1826.                         tmp0 = tab0[j];

  1827.                         tab0[j] = MULL(tmp0, v1);

  1828.                         tab1[j] = MULL(tmp0, v2);

  1829.                     }

  1830.                 } else {

  1831.                 found1:

  1832.                     if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1833.                         /* lower part of the spectrum : do ms stereo

  1834.                            if enabled */

  1835.                         for(j=0;j<len;j++) {

  1836.                             tmp0 = tab0[j];

  1837.                             tmp1 = tab1[j];

  1838.                             tab0[j] = MULL(tmp0 + tmp1, ISQRT2);

  1839.                             tab1[j] = MULL(tmp0 - tmp1, ISQRT2);

  1840.                         }

  1841.                     }

  1842.                 }

  1843.             }

  1844.         }

  1845. 

  1846.         non_zero_found = non_zero_found_short[0] | 

  1847.             non_zero_found_short[1] | 

  1848.             non_zero_found_short[2];

  1849. 

  1850.         for(i = g1->long_end - 1;i >= 0;i--) {

  1851.             len = band_size_long[s->sample_rate_index][i];

  1852.             tab0 -= len;

  1853.             tab1 -= len;

  1854.             /* test if non zero band. if so, stop doing i-stereo */

  1855.             if (!non_zero_found) {

  1856.                 for(j=0;j<len;j++) {

  1857.                     if (tab1[j] != 0) {

  1858.                         non_zero_found = 1;

  1859.                         goto found2;

  1860.                     }

  1861.                 }

  1862.                 /* for last band, use previous scale factor */

  1863.                 k = (i == 21) ? 20 : i;

  1864.                 sf = g1->scale_factors[k];

  1865.                 if (sf >= sf_max)

  1866.                     goto found2;

  1867.                 v1 = is_tab[0][sf];

  1868.                 v2 = is_tab[1][sf];

  1869.                 for(j=0;j<len;j++) {

  1870.                     tmp0 = tab0[j];

  1871.                     tab0[j] = MULL(tmp0, v1);

  1872.                     tab1[j] = MULL(tmp0, v2);

  1873.                 }

  1874.             } else {

  1875.             found2:

  1876.                 if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1877.                     /* lower part of the spectrum : do ms stereo

  1878.                        if enabled */

  1879.                     for(j=0;j<len;j++) {

  1880.                         tmp0 = tab0[j];

  1881.                         tmp1 = tab1[j];

  1882.                         tab0[j] = MULL(tmp0 + tmp1, ISQRT2);

  1883.                         tab1[j] = MULL(tmp0 - tmp1, ISQRT2);

  1884.                     }

  1885.                 }

  1886.             }

  1887.         }

  1888.     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1889.         /* ms stereo ONLY */

  1890.         /* NOTE: the 1/sqrt(2) normalization factor is included in the

  1891.            global gain */

  1892.         tab0 = g0->sb_hybrid;

  1893.         tab1 = g1->sb_hybrid;

  1894.         for(i=0;i<576;i++) {

  1895.             tmp0 = tab0[i];

  1896.             tmp1 = tab1[i];

  1897.             tab0[i] = tmp0 + tmp1;

  1898.             tab1[i] = tmp0 - tmp1;

  1899.         }

  1900.     }

  1901. }

  1902. 

  1903. static void compute_antialias_integer(MPADecodeContext *s,

  1904.                               GranuleDef *g)

  1905. {

  1906.     int32_t *ptr, *csa;

  1907.     int n, i;

  1908. 

  1909.     /* we antialias only "long" bands */

  1910.     if (g->block_type == 2) {

  1911.         if (!g->switch_point)

  1912.             return;

  1913.         /* XXX: check this for 8000Hz case */

  1914.         n = 1;

  1915.     } else {

  1916.         n = SBLIMIT - 1;

  1917.     }

  1918.     

  1919.     ptr = g->sb_hybrid + 18;

  1920.     for(i = n;i > 0;i--) {

  1921.         int tmp0, tmp1, tmp2;

  1922.         csa = &csa_table[0][0];

  1923. #define INT_AA(j) \

  1924.             tmp0 = ptr[-1-j];\

  1925.             tmp1 = ptr[   j];\

  1926.             tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\

  1927.             ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\

  1928.             ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));

  1929. 

  1930.         INT_AA(0)

  1931.         INT_AA(1)

  1932.         INT_AA(2)

  1933.         INT_AA(3)

  1934.         INT_AA(4)

  1935.         INT_AA(5)

  1936.         INT_AA(6)

  1937.         INT_AA(7)

  1938.             

  1939.         ptr += 18;       

  1940.     }

  1941. }

  1942. 

  1943. static void compute_antialias_float(MPADecodeContext *s,

  1944.                               GranuleDef *g)

  1945. {

  1946.     int32_t *ptr;

  1947.     int n, i;

  1948. 

  1949.     /* we antialias only "long" bands */

  1950.     if (g->block_type == 2) {

  1951.         if (!g->switch_point)

  1952.             return;

  1953.         /* XXX: check this for 8000Hz case */

  1954.         n = 1;

  1955.     } else {

  1956.         n = SBLIMIT - 1;

  1957.     }

  1958.     

  1959.     ptr = g->sb_hybrid + 18;

  1960.     for(i = n;i > 0;i--) {

  1961.         float tmp0, tmp1;

  1962.         float *csa = &csa_table_float[0][0];       

  1963. #define FLOAT_AA(j)\

  1964.         tmp0= ptr[-1-j];\

  1965.         tmp1= ptr[   j];\

  1966.         ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\

  1967.         ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);

  1968.         

  1969.         FLOAT_AA(0)

  1970.         FLOAT_AA(1)

  1971.         FLOAT_AA(2)

  1972.         FLOAT_AA(3)

  1973.         FLOAT_AA(4)

  1974.         FLOAT_AA(5)

  1975.         FLOAT_AA(6)

  1976.         FLOAT_AA(7)

  1977. 

  1978.         ptr += 18;       

  1979.     }

  1980. }

  1981. 

  1982. static void compute_imdct(MPADecodeContext *s,

  1983.                           GranuleDef *g, 

  1984.                           int32_t *sb_samples,

  1985.                           int32_t *mdct_buf)

  1986. {

  1987.     int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;

  1988.     int32_t out2[12];

  1989.     int i, j, mdct_long_end, v, sblimit;

  1990. 

  1991.     /* find last non zero block */

  1992.     ptr = g->sb_hybrid + 576;

  1993.     ptr1 = g->sb_hybrid + 2 * 18;

  1994.     while (ptr >= ptr1) {

  1995.         ptr -= 6;

  1996.         v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];

  1997.         if (v != 0)

  1998.             break;

  1999.     }

  2000.     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

  2001. 

  2002.     if (g->block_type == 2) {

  2003.         /* XXX: check for 8000 Hz */

  2004.         if (g->switch_point)

  2005.             mdct_long_end = 2;

  2006.         else

  2007.             mdct_long_end = 0;

  2008.     } else {

  2009.         mdct_long_end = sblimit;

  2010.     }

  2011. 

  2012.     buf = mdct_buf;

  2013.     ptr = g->sb_hybrid;

  2014.     for(j=0;j<mdct_long_end;j++) {

  2015.         /* apply window & overlap with previous buffer */

  2016.         out_ptr = sb_samples + j;

  2017.         /* select window */

  2018.         if (g->switch_point && j < 2)

  2019.             win1 = mdct_win[0];

  2020.         else

  2021.             win1 = mdct_win[g->block_type];

  2022.         /* select frequency inversion */

  2023.         win = win1 + ((4 * 36) & -(j & 1));

  2024.         imdct36(out_ptr, buf, ptr, win);

  2025.         out_ptr += 18*SBLIMIT;

  2026.         ptr += 18;

  2027.         buf += 18;

  2028.     }

  2029.     for(j=mdct_long_end;j<sblimit;j++) {

  2030.         /* select frequency inversion */

  2031.         win = mdct_win[2] + ((4 * 36) & -(j & 1));

  2032.         out_ptr = sb_samples + j;

  2033.         

  2034.         for(i=0; i<6; i++){

  2035.             *out_ptr = buf[i];

  2036.             out_ptr += SBLIMIT;

  2037.         }

  2038.         imdct12(out2, ptr + 0);

  2039.         for(i=0;i<6;i++) {

  2040.             *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];

  2041.             buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);

  2042.             out_ptr += SBLIMIT;

  2043.         }

  2044.         imdct12(out2, ptr + 1);

  2045.         for(i=0;i<6;i++) {

  2046.             *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];

  2047.             buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);

  2048.             out_ptr += SBLIMIT;

  2049.         }

  2050.         imdct12(out2, ptr + 2);

  2051.         for(i=0;i<6;i++) {

  2052.             buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];

  2053.             buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);

  2054.             buf[i + 6*2] = 0;

  2055.         }

  2056.         ptr += 18;

  2057.         buf += 18;

  2058.     }

  2059.     /* zero bands */

  2060.     for(j=sblimit;j<SBLIMIT;j++) {

  2061.         /* overlap */

  2062.         out_ptr = sb_samples + j;

  2063.         for(i=0;i<18;i++) {

  2064.             *out_ptr = buf[i];

  2065.             buf[i] = 0;

  2066.             out_ptr += SBLIMIT;

  2067.         }

  2068.         buf += 18;

  2069.     }

  2070. }

  2071. 

  2072. #if defined(DEBUG)

  2073. void sample_dump(int fnum, int32_t *tab, int n)

  2074. {

  2075.     static FILE *files[16], *f;

  2076.     char buf[512];

  2077.     int i;

  2078.     int32_t v;

  2079.     

  2080.     f = files[fnum];

  2081.     if (!f) {

  2082.         snprintf(buf, sizeof(buf), "/tmp/out%d.%s.pcm", 

  2083.                 fnum, 

  2084. #ifdef USE_HIGHPRECISION

  2085.                 "hp"

  2086. #else

  2087.                 "lp"

  2088. #endif

  2089.                 );

  2090.         f = fopen(buf, "w");

  2091.         if (!f)

  2092.             return;

  2093.         files[fnum] = f;

  2094.     }

  2095.     

  2096.     if (fnum == 0) {

  2097.         static int pos = 0;

  2098.         av_log(NULL, AV_LOG_DEBUG, "pos=%d\n", pos);

  2099.         for(i=0;i<n;i++) {

  2100.             av_log(NULL, AV_LOG_DEBUG, " %0.4f", (double)tab[i] / FRAC_ONE);

  2101.             if ((i % 18) == 17)

  2102.                 av_log(NULL, AV_LOG_DEBUG, "\n");

  2103.         }

  2104.         pos += n;

  2105.     }

  2106.     for(i=0;i<n;i++) {

  2107.         /* normalize to 23 frac bits */

  2108.         v = tab[i] << (23 - FRAC_BITS);

  2109.         fwrite(&v, 1, sizeof(int32_t), f);

  2110.     }

  2111. }

  2112. #endif

  2113. 

  2114. 

  2115. /* main layer3 decoding function */

  2116. static int mp_decode_layer3(MPADecodeContext *s)

  2117. {

  2118.     int nb_granules, main_data_begin, private_bits;

  2119.     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;

  2120.     GranuleDef granules[2][2], *g;

  2121.     int16_t exponents[576];

  2122. 

  2123.     /* read side info */

  2124.     if (s->lsf) {

  2125.         main_data_begin = get_bits(&s->gb, 8);

  2126.         if (s->nb_channels == 2)

  2127.             private_bits = get_bits(&s->gb, 2);

  2128.         else

  2129.             private_bits = get_bits(&s->gb, 1);

  2130.         nb_granules = 1;

  2131.     } else {

  2132.         main_data_begin = get_bits(&s->gb, 9);

  2133.         if (s->nb_channels == 2)

  2134.             private_bits = get_bits(&s->gb, 3);

  2135.         else

  2136.             private_bits = get_bits(&s->gb, 5);

  2137.         nb_granules = 2;

  2138.         for(ch=0;ch<s->nb_channels;ch++) {

  2139.             granules[ch][0].scfsi = 0; /* all scale factors are transmitted */

  2140.             granules[ch][1].scfsi = get_bits(&s->gb, 4);

  2141.         }

  2142.     }

  2143.     

  2144.     for(gr=0;gr<nb_granules;gr++) {

  2145.         for(ch=0;ch<s->nb_channels;ch++) {

  2146.             dprintf("gr=%d ch=%d: side_info\n", gr, ch);

  2147.             g = &granules[ch][gr];

  2148.             g->part2_3_length = get_bits(&s->gb, 12);

  2149.             g->big_values = get_bits(&s->gb, 9);

  2150.             g->global_gain = get_bits(&s->gb, 8);

  2151.             /* if MS stereo only is selected, we precompute the

  2152.                1/sqrt(2) renormalization factor */

  2153.             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) == 

  2154.                 MODE_EXT_MS_STEREO)

  2155.                 g->global_gain -= 2;

  2156.             if (s->lsf)

  2157.                 g->scalefac_compress = get_bits(&s->gb, 9);

  2158.             else

  2159.                 g->scalefac_compress = get_bits(&s->gb, 4);

  2160.             blocksplit_flag = get_bits(&s->gb, 1);

  2161.             if (blocksplit_flag) {

  2162.                 g->block_type = get_bits(&s->gb, 2);

  2163.                 if (g->block_type == 0)

  2164.                     return -1;

  2165.                 g->switch_point = get_bits(&s->gb, 1);

  2166.                 for(i=0;i<2;i++)

  2167.                     g->table_select[i] = get_bits(&s->gb, 5);

  2168.                 for(i=0;i<3;i++) 

  2169.                     g->subblock_gain[i] = get_bits(&s->gb, 3);

  2170.                 /* compute huffman coded region sizes */

  2171.                 if (g->block_type == 2)

  2172.                     g->region_size[0] = (36 / 2);

  2173.                 else {

  2174.                     if (s->sample_rate_index <= 2) 

  2175.                         g->region_size[0] = (36 / 2);

  2176.                     else if (s->sample_rate_index != 8) 

  2177.                         g->region_size[0] = (54 / 2);

  2178.                     else

  2179.                         g->region_size[0] = (108 / 2);

  2180.                 }

  2181.                 g->region_size[1] = (576 / 2);

  2182.             } else {

  2183.                 int region_address1, region_address2, l;

  2184.                 g->block_type = 0;

  2185.                 g->switch_point = 0;

  2186.                 for(i=0;i<3;i++)

  2187.                     g->table_select[i] = get_bits(&s->gb, 5);

  2188.                 /* compute huffman coded region sizes */

  2189.                 region_address1 = get_bits(&s->gb, 4);

  2190.                 region_address2 = get_bits(&s->gb, 3);

  2191.                 dprintf("region1=%d region2=%d\n", 

  2192.                         region_address1, region_address2);

  2193.                 g->region_size[0] = 

  2194.                     band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;

  2195.                 l = region_address1 + region_address2 + 2;

  2196.                 /* should not overflow */

  2197.                 if (l > 22)

  2198.                     l = 22;

  2199.                 g->region_size[1] = 

  2200.                     band_index_long[s->sample_rate_index][l] >> 1;

  2201.             }

  2202.             /* convert region offsets to region sizes and truncate

  2203.                size to big_values */

  2204.             g->region_size[2] = (576 / 2);

  2205.             j = 0;

  2206.             for(i=0;i<3;i++) {

  2207.                 k = g->region_size[i];

  2208.                 if (k > g->big_values)

  2209.                     k = g->big_values;

  2210.                 g->region_size[i] = k - j;

  2211.                 j = k;

  2212.             }

  2213. 

  2214.             /* compute band indexes */

  2215.             if (g->block_type == 2) {

  2216.                 if (g->switch_point) {

  2217.                     /* if switched mode, we handle the 36 first samples as

  2218.                        long blocks.  For 8000Hz, we handle the 48 first

  2219.                        exponents as long blocks (XXX: check this!) */

  2220.                     if (s->sample_rate_index <= 2)

  2221.                         g->long_end = 8;

  2222.                     else if (s->sample_rate_index != 8)

  2223.                         g->long_end = 6;

  2224.                     else

  2225.                         g->long_end = 4; /* 8000 Hz */

  2226.                     

  2227.                     if (s->sample_rate_index != 8)

  2228.                         g->short_start = 3;

  2229.                     else

  2230.                         g->short_start = 2; 

  2231.                 } else {

  2232.                     g->long_end = 0;

  2233.                     g->short_start = 0;

  2234.                 }

  2235.             } else {

  2236.                 g->short_start = 13;

  2237.                 g->long_end = 22;

  2238.             }

  2239.             

  2240.             g->preflag = 0;

  2241.             if (!s->lsf)

  2242.                 g->preflag = get_bits(&s->gb, 1);

  2243.             g->scalefac_scale = get_bits(&s->gb, 1);

  2244.             g->count1table_select = get_bits(&s->gb, 1);

  2245.             dprintf("block_type=%d switch_point=%d\n",

  2246.                     g->block_type, g->switch_point);

  2247.         }

  2248.     }

  2249. 

  2250.   if (!s->adu_mode) {

  2251.     /* now we get bits from the main_data_begin offset */

  2252.     dprintf("seekback: %d\n", main_data_begin);

  2253.     seek_to_maindata(s, main_data_begin);

  2254.   }

  2255. 

  2256.     for(gr=0;gr<nb_granules;gr++) {

  2257.         for(ch=0;ch<s->nb_channels;ch++) {

  2258.             g = &granules[ch][gr];

  2259.             

  2260.             bits_pos = get_bits_count(&s->gb);

  2261.             

  2262.             if (!s->lsf) {

  2263.                 uint8_t *sc;

  2264.                 int slen, slen1, slen2;

  2265. 

  2266.                 /* MPEG1 scale factors */

  2267.                 slen1 = slen_table[0][g->scalefac_compress];

  2268.                 slen2 = slen_table[1][g->scalefac_compress];

  2269.                 dprintf("slen1=%d slen2=%d\n", slen1, slen2);

  2270.                 if (g->block_type == 2) {

  2271.                     n = g->switch_point ? 17 : 18;

  2272.                     j = 0;

  2273.                     for(i=0;i<n;i++)

  2274.                         g->scale_factors[j++] = get_bitsz(&s->gb, slen1);

  2275.                     for(i=0;i<18;i++)

  2276.                         g->scale_factors[j++] = get_bitsz(&s->gb, slen2);

  2277.                     for(i=0;i<3;i++)

  2278.                         g->scale_factors[j++] = 0;

  2279.                 } else {

  2280.                     sc = granules[ch][0].scale_factors;

  2281.                     j = 0;

  2282.                     for(k=0;k<4;k++) {

  2283.                         n = (k == 0 ? 6 : 5);

  2284.                         if ((g->scfsi & (0x8 >> k)) == 0) {

  2285.                             slen = (k < 2) ? slen1 : slen2;

  2286.                             for(i=0;i<n;i++)

  2287.                                 g->scale_factors[j++] = get_bitsz(&s->gb, slen);

  2288.                         } else {

  2289.                             /* simply copy from last granule */

  2290.                             for(i=0;i<n;i++) {

  2291.                                 g->scale_factors[j] = sc[j];

  2292.                                 j++;

  2293.                             }

  2294.                         }

  2295.                     }

  2296.                     g->scale_factors[j++] = 0;

  2297.                 }

  2298. #if defined(DEBUG)

  2299.                 {

  2300.                     printf("scfsi=%x gr=%d ch=%d scale_factors:\n", 

  2301.                            g->scfsi, gr, ch);

  2302.                     for(i=0;i<j;i++)

  2303.                         printf(" %d", g->scale_factors[i]);

  2304.                     printf("\n");

  2305.                 }

  2306. #endif

  2307.             } else {

  2308.                 int tindex, tindex2, slen[4], sl, sf;

  2309. 

  2310.                 /* LSF scale factors */

  2311.                 if (g->block_type == 2) {

  2312.                     tindex = g->switch_point ? 2 : 1;

  2313.                 } else {

  2314.                     tindex = 0;

  2315.                 }

  2316.                 sf = g->scalefac_compress;

  2317.                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {

  2318.                     /* intensity stereo case */

  2319.                     sf >>= 1;

  2320.                     if (sf < 180) {

  2321.                         lsf_sf_expand(slen, sf, 6, 6, 0);

  2322.                         tindex2 = 3;

  2323.                     } else if (sf < 244) {

  2324.                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);

  2325.                         tindex2 = 4;

  2326.                     } else {

  2327.                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);

  2328.                         tindex2 = 5;

  2329.                     }

  2330.                 } else {

  2331.                     /* normal case */

  2332.                     if (sf < 400) {

  2333.                         lsf_sf_expand(slen, sf, 5, 4, 4);

  2334.                         tindex2 = 0;

  2335.                     } else if (sf < 500) {

  2336.                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);

  2337.                         tindex2 = 1;

  2338.                     } else {

  2339.                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);

  2340.                         tindex2 = 2;

  2341.                         g->preflag = 1;

  2342.                     }

  2343.                 }

  2344. 

  2345.                 j = 0;

  2346.                 for(k=0;k<4;k++) {

  2347.                     n = lsf_nsf_table[tindex2][tindex][k];

  2348.                     sl = slen[k];

  2349.                     for(i=0;i<n;i++)

  2350.                         g->scale_factors[j++] = get_bitsz(&s->gb, sl);

  2351.                 }

  2352.                 /* XXX: should compute exact size */

  2353.                 for(;j<40;j++)

  2354.                     g->scale_factors[j] = 0;

  2355. #if defined(DEBUG)

  2356.                 {

  2357.                     printf("gr=%d ch=%d scale_factors:\n", 

  2358.                            gr, ch);

  2359.                     for(i=0;i<40;i++)

  2360.                         printf(" %d", g->scale_factors[i]);

  2361.                     printf("\n");

  2362.                 }

  2363. #endif

  2364.             }

  2365. 

  2366.             exponents_from_scale_factors(s, g, exponents);

  2367. 

  2368.             /* read Huffman coded residue */

  2369.             if (huffman_decode(s, g, exponents,

  2370.                                bits_pos + g->part2_3_length) < 0)

  2371.                 return -1;

  2372. #if defined(DEBUG)

  2373.             sample_dump(0, g->sb_hybrid, 576);

  2374. #endif

  2375. 

  2376.             /* skip extension bits */

  2377.             bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);

  2378.             if (bits_left < 0) {

  2379.                 dprintf("bits_left=%d\n", bits_left);

  2380.                 return -1;

  2381.             }

  2382.             while (bits_left >= 16) {

  2383.                 skip_bits(&s->gb, 16);

  2384.                 bits_left -= 16;

  2385.             }

  2386.             if (bits_left > 0)

  2387.                 skip_bits(&s->gb, bits_left);

  2388.         } /* ch */

  2389. 

  2390.         if (s->nb_channels == 2)

  2391.             compute_stereo(s, &granules[0][gr], &granules[1][gr]);

  2392. 

  2393.         for(ch=0;ch<s->nb_channels;ch++) {

  2394.             g = &granules[ch][gr];

  2395. 

  2396.             reorder_block(s, g);

  2397. #if defined(DEBUG)

  2398.             sample_dump(0, g->sb_hybrid, 576);

  2399. #endif

  2400.             s->compute_antialias(s, g);

  2401. #if defined(DEBUG)

  2402.             sample_dump(1, g->sb_hybrid, 576);

  2403. #endif

  2404.             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]); 

  2405. #if defined(DEBUG)

  2406.             sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);

  2407. #endif

  2408.         }

  2409.     } /* gr */

  2410.     return nb_granules * 18;

  2411. }

  2412. 

  2413. static int mp_decode_frame(MPADecodeContext *s, 

  2414.                            OUT_INT *samples)

  2415. {

  2416.     int i, nb_frames, ch;

  2417.     OUT_INT *samples_ptr;

  2418. 

  2419.     init_get_bits(&s->gb, s->inbuf + HEADER_SIZE, 

  2420.                   (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);

  2421.     

  2422.     /* skip error protection field */

  2423.     if (s->error_protection)

  2424.         get_bits(&s->gb, 16);

  2425. 

  2426.     dprintf("frame %d:\n", s->frame_count);

  2427.     switch(s->layer) {

  2428.     case 1:

  2429.         nb_frames = mp_decode_layer1(s);

  2430.         break;

  2431.     case 2:

  2432.         nb_frames = mp_decode_layer2(s);

  2433.         break;

  2434.     case 3:

  2435.     default:

  2436.         nb_frames = mp_decode_layer3(s);

  2437.         break;

  2438.     }

  2439. #if defined(DEBUG)

  2440.     for(i=0;i<nb_frames;i++) {

  2441.         for(ch=0;ch<s->nb_channels;ch++) {

  2442.             int j;

  2443.             printf("%d-%d:", i, ch);

  2444.             for(j=0;j<SBLIMIT;j++)

  2445.                 printf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);

  2446.             printf("\n");

  2447.         }

  2448.     }

  2449. #endif

  2450.     /* apply the synthesis filter */

  2451.     for(ch=0;ch<s->nb_channels;ch++) {

  2452.         samples_ptr = samples + ch;

  2453.         for(i=0;i<nb_frames;i++) {

  2454.             ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),

  2455. 			 window, &s->dither_state,

  2456. 			 samples_ptr, s->nb_channels,

  2457.                          s->sb_samples[ch][i]);

  2458.             samples_ptr += 32 * s->nb_channels;

  2459.         }

  2460.     }

  2461. #ifdef DEBUG

  2462.     s->frame_count++;        

  2463. #endif

  2464.     return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;

  2465. }

  2466. 

  2467. static int decode_frame(AVCodecContext * avctx,

  2468. 			void *data, int *data_size,

  2469. 			uint8_t * buf, int buf_size)

  2470. {

  2471.     MPADecodeContext *s = avctx->priv_data;

  2472.     uint32_t header;

  2473.     uint8_t *buf_ptr;

  2474.     int len, out_size;

  2475.     OUT_INT *out_samples = data;

  2476. 

  2477.     buf_ptr = buf;

  2478.     while (buf_size > 0) {

  2479. 	len = s->inbuf_ptr - s->inbuf;

  2480. 	if (s->frame_size == 0) {

  2481.             /* special case for next header for first frame in free

  2482.                format case (XXX: find a simpler method) */

  2483.             if (s->free_format_next_header != 0) {

  2484.                 s->inbuf[0] = s->free_format_next_header >> 24;

  2485.                 s->inbuf[1] = s->free_format_next_header >> 16;

  2486.                 s->inbuf[2] = s->free_format_next_header >> 8;

  2487.                 s->inbuf[3] = s->free_format_next_header;

  2488.                 s->inbuf_ptr = s->inbuf + 4;

  2489.                 s->free_format_next_header = 0;

  2490.                 goto got_header;

  2491.             }

  2492. 	    /* no header seen : find one. We need at least HEADER_SIZE

  2493.                bytes to parse it */

  2494. 	    len = HEADER_SIZE - len;

  2495. 	    if (len > buf_size)

  2496. 		len = buf_size;

  2497. 	    if (len > 0) {

  2498. 		memcpy(s->inbuf_ptr, buf_ptr, len);

  2499. 		buf_ptr += len;

  2500. 		buf_size -= len;

  2501. 		s->inbuf_ptr += len;

  2502. 	    }

  2503. 	    if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {

  2504.             got_header:

  2505. 		header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |

  2506. 		    (s->inbuf[2] << 8) | s->inbuf[3];

  2507. 

  2508. 		if (ff_mpa_check_header(header) < 0) {

  2509. 		    /* no sync found : move by one byte (inefficient, but simple!) */

  2510. 		    memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);

  2511. 		    s->inbuf_ptr--;

  2512.                     dprintf("skip %x\n", header);

  2513.                     /* reset free format frame size to give a chance

  2514.                        to get a new bitrate */

  2515.                     s->free_format_frame_size = 0;

  2516. 		} else {

  2517. 		    if (decode_header(s, header) == 1) {

  2518.                         /* free format: prepare to compute frame size */

  2519. 			s->frame_size = -1;

  2520.                     }

  2521.                     /* update codec info */

  2522.                     avctx->sample_rate = s->sample_rate;

  2523.                     avctx->channels = s->nb_channels;

  2524.                     avctx->bit_rate = s->bit_rate;

  2525.                     avctx->sub_id = s->layer;

  2526.                     switch(s->layer) {

  2527.                     case 1:

  2528.                         avctx->frame_size = 384;

  2529.                         break;

  2530.                     case 2:

  2531.                         avctx->frame_size = 1152;

  2532.                         break;

  2533.                     case 3:

  2534.                         if (s->lsf)

  2535.                             avctx->frame_size = 576;

  2536.                         else

  2537.                             avctx->frame_size = 1152;

  2538.                         break;

  2539.                     }

  2540. 		}

  2541. 	    }

  2542.         } else if (s->frame_size == -1) {

  2543.             /* free format : find next sync to compute frame size */

  2544. 	    len = MPA_MAX_CODED_FRAME_SIZE - len;

  2545. 	    if (len > buf_size)

  2546. 		len = buf_size;

  2547.             if (len == 0) {

  2548. 		/* frame too long: resync */

  2549.                 s->frame_size = 0;

  2550. 		memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);

  2551. 		s->inbuf_ptr--;

  2552.             } else {

  2553.                 uint8_t *p, *pend;

  2554.                 uint32_t header1;

  2555.                 int padding;

  2556. 

  2557.                 memcpy(s->inbuf_ptr, buf_ptr, len);

  2558.                 /* check for header */

  2559.                 p = s->inbuf_ptr - 3;

  2560.                 pend = s->inbuf_ptr + len - 4;

  2561.                 while (p <= pend) {

  2562.                     header = (p[0] << 24) | (p[1] << 16) |

  2563.                         (p[2] << 8) | p[3];

  2564.                     header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |

  2565.                         (s->inbuf[2] << 8) | s->inbuf[3];

  2566.                     /* check with high probability that we have a

  2567.                        valid header */

  2568.                     if ((header & SAME_HEADER_MASK) ==

  2569.                         (header1 & SAME_HEADER_MASK)) {

  2570.                         /* header found: update pointers */

  2571.                         len = (p + 4) - s->inbuf_ptr;

  2572.                         buf_ptr += len;

  2573.                         buf_size -= len;

  2574.                         s->inbuf_ptr = p;

  2575.                         /* compute frame size */

  2576.                         s->free_format_next_header = header;

  2577.                         s->free_format_frame_size = s->inbuf_ptr - s->inbuf;

  2578.                         padding = (header1 >> 9) & 1;

  2579.                         if (s->layer == 1)

  2580.                             s->free_format_frame_size -= padding * 4;

  2581.                         else

  2582.                             s->free_format_frame_size -= padding;

  2583.                         dprintf("free frame size=%d padding=%d\n", 

  2584.                                 s->free_format_frame_size, padding);

  2585.                         decode_header(s, header1);

  2586.                         goto next_data;

  2587.                     }

  2588.                     p++;

  2589.                 }

  2590.                 /* not found: simply increase pointers */

  2591.                 buf_ptr += len;

  2592.                 s->inbuf_ptr += len;

  2593.                 buf_size -= len;

  2594.             }

  2595. 	} else if (len < s->frame_size) {

  2596.             if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)

  2597.                 s->frame_size = MPA_MAX_CODED_FRAME_SIZE;

  2598. 	    len = s->frame_size - len;

  2599. 	    if (len > buf_size)

  2600. 		len = buf_size;

  2601. 	    memcpy(s->inbuf_ptr, buf_ptr, len);

  2602. 	    buf_ptr += len;

  2603. 	    s->inbuf_ptr += len;

  2604. 	    buf_size -= len;

  2605. 	}

  2606.     next_data:

  2607.         if (s->frame_size > 0 && 

  2608.             (s->inbuf_ptr - s->inbuf) >= s->frame_size) {

  2609.             if (avctx->parse_only) {

  2610.                 /* simply return the frame data */

  2611.                 *(uint8_t **)data = s->inbuf;

  2612.                 out_size = s->inbuf_ptr - s->inbuf;

  2613.             } else {

  2614.                 out_size = mp_decode_frame(s, out_samples);

  2615.             }

  2616. 	    s->inbuf_ptr = s->inbuf;

  2617. 	    s->frame_size = 0;

  2618. 	    *data_size = out_size;

  2619. 	    break;

  2620. 	}

  2621.     }

  2622.     return buf_ptr - buf;

  2623. }

  2624. 

  2625. 

  2626. static int decode_frame_adu(AVCodecContext * avctx,

  2627. 			void *data, int *data_size,

  2628. 			uint8_t * buf, int buf_size)

  2629. {

  2630.     MPADecodeContext *s = avctx->priv_data;

  2631.     uint32_t header;

  2632.     int len, out_size;

  2633.     OUT_INT *out_samples = data;

  2634. 

  2635.     len = buf_size;

  2636. 

  2637.     // Discard too short frames

  2638.     if (buf_size < HEADER_SIZE) {

  2639.         *data_size = 0;

  2640.         return buf_size;

  2641.     }

  2642. 

  2643. 

  2644.     if (len > MPA_MAX_CODED_FRAME_SIZE)

  2645.         len = MPA_MAX_CODED_FRAME_SIZE;

  2646. 

  2647.     memcpy(s->inbuf, buf, len);

  2648.     s->inbuf_ptr = s->inbuf + len;

  2649. 

  2650.     // Get header and restore sync word

  2651.     header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |

  2652.               (s->inbuf[2] << 8) | s->inbuf[3] | 0xffe00000;

  2653. 

  2654.     if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame

  2655.         *data_size = 0;

  2656.         return buf_size;

  2657.     }

  2658. 

  2659.     decode_header(s, header);

  2660.     /* update codec info */

  2661.     avctx->sample_rate = s->sample_rate;

  2662.     avctx->channels = s->nb_channels;

  2663.     avctx->bit_rate = s->bit_rate;

  2664.     avctx->sub_id = s->layer;

  2665. 

  2666.     avctx->frame_size=s->frame_size = len;

  2667. 

  2668.     if (avctx->parse_only) {

  2669.         /* simply return the frame data */

  2670.         *(uint8_t **)data = s->inbuf;

  2671.         out_size = s->inbuf_ptr - s->inbuf;

  2672.     } else {

  2673.         out_size = mp_decode_frame(s, out_samples);

  2674.     }

  2675. 

  2676.     *data_size = out_size;

  2677.     return buf_size;

  2678. }

  2679. 

  2680. 

  2681. AVCodec mp2_decoder =

  2682. {

  2683.     "mp2",

  2684.     CODEC_TYPE_AUDIO,

  2685.     CODEC_ID_MP2,

  2686.     sizeof(MPADecodeContext),

  2687.     decode_init,

  2688.     NULL,

  2689.     NULL,

  2690.     decode_frame,

  2691.     CODEC_CAP_PARSE_ONLY,

  2692. };

  2693. 

  2694. AVCodec mp3_decoder =

  2695. {

  2696.     "mp3",

  2697.     CODEC_TYPE_AUDIO,

  2698.     CODEC_ID_MP3,

  2699.     sizeof(MPADecodeContext),

  2700.     decode_init,

  2701.     NULL,

  2702.     NULL,

  2703.     decode_frame,

  2704.     CODEC_CAP_PARSE_ONLY,

  2705. };

  2706. 

  2707. AVCodec mp3adu_decoder =

  2708. {

  2709.     "mp3adu",

  2710.     CODEC_TYPE_AUDIO,

  2711.     CODEC_ID_MP3ADU,

  2712.     sizeof(MPADecodeContext),

  2713.     decode_init,

  2714.     NULL,

  2715.     NULL,

  2716.     decode_frame_adu,

  2717.     CODEC_CAP_PARSE_ONLY,

  2718. };