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.
2987 Commits
32 Branches
290MB
Tree: 7bf3261289
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '7bf3261289'
${ noResults }
FFmpeg/libavcodec/mpegaudiodec.c

2692 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 "mpegaudio.h"

  28. #include "dsputil.h"

  29. 

  30. /*

  31.  * TODO:

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

  33.  *  - test lsf / mpeg25 extensively.

  34.  */

  35. 

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

  37.    audio decoder */

  38. #ifdef CONFIG_MPEGAUDIO_HP

  39. #define USE_HIGHPRECISION

  40. #endif

  41. 

  42. #ifdef USE_HIGHPRECISION

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

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

  45. #else

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

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

  48. #endif

  49. 

  50. #define FRAC_ONE    (1 << FRAC_BITS)

  51. 

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

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

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

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

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

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

  58. 

  59. #if FRAC_BITS <= 15

  60. typedef int16_t MPA_INT;

  61. #else

  62. typedef int32_t MPA_INT;

  63. #endif

  64. 

  65. /****************/

  66. 

  67. #define HEADER_SIZE 4

  68. #define BACKSTEP_SIZE 512

  69. 

  70. struct GranuleDef;

  71. 

  72. typedef struct MPADecodeContext {

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

  74.     int inbuf_index;

  75.     uint8_t *inbuf_ptr, *inbuf;

  76.     int frame_size;

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

  78.                                    (zero if currently unknown) */

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

  80.     uint32_t free_format_next_header; 

  81.     int error_protection;

  82.     int layer;

  83.     int sample_rate;

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

  85.     int bit_rate;

  86.     int old_frame_size;

  87.     GetBitContext gb;

  88.     int nb_channels;

  89.     int mode;

  90.     int mode_ext;

  91.     int lsf;

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

  93.     int synth_buf_offset[MPA_MAX_CHANNELS];

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

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

  96. #ifdef DEBUG

  97.     int frame_count;

  98. #endif

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

  100. } MPADecodeContext;

  101. 

  102. /* layer 3 "granule" */

  103. typedef struct GranuleDef {

  104.     uint8_t scfsi;

  105.     int part2_3_length;

  106.     int big_values;

  107.     int global_gain;

  108.     int scalefac_compress;

  109.     uint8_t block_type;

  110.     uint8_t switch_point;

  111.     int table_select[3];

  112.     int subblock_gain[3];

  113.     uint8_t scalefac_scale;

  114.     uint8_t count1table_select;

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

  116.     int preflag;

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

  118.     uint8_t scale_factors[40];

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

  120. } GranuleDef;

  121. 

  122. #define MODE_EXT_MS_STEREO 2

  123. #define MODE_EXT_I_STEREO  1

  124. 

  125. /* layer 3 huffman tables */

  126. typedef struct HuffTable {

  127.     int xsize;

  128.     const uint8_t *bits;

  129.     const uint16_t *codes;

  130. } HuffTable;

  131. 

  132. #include "mpegaudiodectab.h"

  133. 

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

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

  136. 

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

  138. static VLC huff_vlc[16]; 

  139. static uint8_t *huff_code_table[16];

  140. static VLC huff_quad_vlc[2];

  141. /* computed from band_size_long */

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

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

  144. #define TABLE_4_3_SIZE (8191 + 16)

  145. static int8_t  *table_4_3_exp;

  146. #if FRAC_BITS <= 15

  147. static uint16_t *table_4_3_value;

  148. #else

  149. static uint32_t *table_4_3_value;

  150. #endif

  151. /* intensity stereo coef table */

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

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

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

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

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

  157. 

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

  159. static uint16_t scale_factor_modshift[64];

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

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

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

  163. 

  164. #define SCALE_GEN(v) \

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

  166. 

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

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

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

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

  171. };

  172. 

  173. /* 2^(n/4) */

  174. static uint32_t scale_factor_mult3[4] = {

  175.     FIXR(1.0),

  176.     FIXR(1.18920711500272106671),

  177.     FIXR(1.41421356237309504880),

  178.     FIXR(1.68179283050742908605),

  179. };

  180. 

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

  182.     

  183. /* layer 1 unscaling */

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

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

  186. {

  187.     int shift, mod;

  188.     int64_t val;

  189. 

  190.     shift = scale_factor_modshift[scale_factor];

  191.     mod = shift & 3;

  192.     shift >>= 2;

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

  194.     shift += n;

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

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

  197. }

  198. 

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

  200. {

  201.     int shift, mod, val;

  202. 

  203.     shift = scale_factor_modshift[scale_factor];

  204.     mod = shift & 3;

  205.     shift >>= 2;

  206. 

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

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

  209.     if (shift > 0)

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

  211.     return val;

  212. }

  213. 

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

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

  216. {

  217. #if FRAC_BITS <= 15    

  218.     unsigned int m;

  219. #else

  220.     uint64_t m;

  221. #endif

  222.     int e;

  223. 

  224.     e = table_4_3_exp[value];

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

  226.     e = FRAC_BITS - e;

  227. #if FRAC_BITS <= 15    

  228.     if (e > 31)

  229.         e = 31;

  230. #endif

  231.     m = table_4_3_value[value];

  232. #if FRAC_BITS <= 15    

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

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

  235.     return m;

  236. #else

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

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

  239.     return m;

  240. #endif

  241. }

  242. 

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

  244. #define DEV_ORDER 13

  245. 

  246. #define POW_FRAC_BITS 24

  247. #define POW_FRAC_ONE    (1 << POW_FRAC_BITS)

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

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

  250. 

  251. static int dev_4_3_coefs[DEV_ORDER];

  252. 

  253. static int pow_mult3[3] = {

  254.     POW_FIX(1.0),

  255.     POW_FIX(1.25992104989487316476),

  256.     POW_FIX(1.58740105196819947474),

  257. };

  258. 

  259. static void int_pow_init(void)

  260. {

  261.     int i, a;

  262. 

  263.     a = POW_FIX(1.0);

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

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

  266.         dev_4_3_coefs[i] = a;

  267.     }

  268. }

  269. 

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

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

  272. {

  273.     int e, er, eq, j;

  274.     int a, a1;

  275.     

  276.     /* renormalize */

  277.     a = i;

  278.     e = POW_FRAC_BITS;

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

  280.         a = a << 1;

  281.         e--;

  282.     }

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

  284.     a1 = 0;

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

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

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

  288.     /* exponent compute (exact) */

  289.     e = e * 4;

  290.     er = e % 3;

  291.     eq = e / 3;

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

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

  294.         a = a >> 1;

  295.         eq++;

  296.     }

  297.     /* convert to float */

  298.     while (a < POW_FRAC_ONE) {

  299.         a = a << 1;

  300.         eq--;

  301.     }

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

  303. #if POW_FRAC_BITS > FRAC_BITS

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

  305.     /* correct overflow */

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

  307.         a = a >> 1;

  308.         eq++;

  309.     }

  310. #endif

  311.     *exp_ptr = eq;

  312.     return a;

  313. }

  314. 

  315. static int decode_init(AVCodecContext * avctx)

  316. {

  317.     MPADecodeContext *s = avctx->priv_data;

  318.     static int init=0;

  319.     int i, j, k;

  320. 

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

  322.         s->compute_antialias= compute_antialias_integer;

  323.     else

  324.         s->compute_antialias= compute_antialias_float;

  325. 

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

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

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

  329.             int shift, mod;

  330.             /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

  331.             shift = (i / 3);

  332.             mod = i % 3;

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

  334.         }

  335. 

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

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

  338.             int n, norm;

  339.             n = i + 2;

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

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

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

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

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

  345.                     i, norm, 

  346.                     scale_factor_mult[i][0],

  347.                     scale_factor_mult[i][1],

  348.                     scale_factor_mult[i][2]);

  349.         }

  350.         

  351.         /* window */

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

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

  354.             int v;

  355.             v = mpa_enwindow[i];

  356. #if WFRAC_BITS < 16

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

  358. #endif

  359.             window[i] = v;

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

  361.                 v = -v;

  362.             if (i != 0)

  363.                 window[512 - i] = v;

  364.         }

  365.         

  366.         /* huffman decode tables */

  367.         huff_code_table[0] = NULL;

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

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

  370. 	    int xsize, x, y;

  371.             unsigned int n;

  372.             uint8_t *code_table;

  373. 

  374.             xsize = h->xsize;

  375.             n = xsize * xsize;

  376.             /* XXX: fail test */

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

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

  379.             

  380.             code_table = av_mallocz(n);

  381.             j = 0;

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

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

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

  385.             }

  386.             huff_code_table[i] = code_table;

  387.         }

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

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

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

  391.         }

  392. 

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

  394.             k = 0;

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

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

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

  398.             }

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

  400.         }

  401. 

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

  403. 	if (!av_mallocz_static(&table_4_3_exp,

  404. 			       TABLE_4_3_SIZE * sizeof(table_4_3_exp[0])))

  405. 	    return -1;

  406. 	if (!av_mallocz_static(&table_4_3_value,

  407. 			       TABLE_4_3_SIZE * sizeof(table_4_3_value[0])))

  408.             return -1;

  409.         

  410.         int_pow_init();

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

  412.             int e, m;

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

  414. #if 0

  415.             /* test code */

  416.             {

  417.                 double f, fm;

  418.                 int e1, m1;

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

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

  421.                 m1 = FIXR(2 * fm);

  422. #if FRAC_BITS <= 15

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

  424.                     m1 = m1 >> 1;

  425.                     e1++;

  426.                 }

  427. #endif

  428.                 e1--;

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

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

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

  432.                 }

  433.             }

  434. #endif

  435.             /* normalized to FRAC_BITS */

  436.             table_4_3_value[i] = m;

  437.             table_4_3_exp[i] = e;

  438.         }

  439.         

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

  441.             float f;

  442.             int v;

  443.             if (i != 6) {

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

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

  446.             } else {

  447.                 v = FIXR(1.0);

  448.             }

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

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

  451.         }

  452.         /* invalid values */

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

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

  455. 

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

  457.             double f;

  458.             int e, k;

  459. 

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

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

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

  463.                 k = i & 1;

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

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

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

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

  468.             }

  469.         }

  470. 

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

  472.             float ci, cs, ca;

  473.             ci = ci_table[i];

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

  475.             ca = cs * ci;

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

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

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

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

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

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

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

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

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

  485.         }

  486. 

  487.         /* compute mdct windows */

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

  489.             int v;

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

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

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

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

  494.         }

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

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

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

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

  499. 

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

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

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

  503.         }

  504. 

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

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

  507.         

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

  509.            the sign of the right window coefs */

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

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

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

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

  514.             }

  515.         }

  516. 

  517. #if defined(DEBUG)

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

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

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

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

  522.             printf("\n");

  523.         }

  524. #endif

  525.         init = 1;

  526.     }

  527. 

  528.     s->inbuf_index = 0;

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

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

  531. #ifdef DEBUG

  532.     s->frame_count = 0;

  533. #endif

  534.     return 0;

  535. }

  536. 

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

  538. 

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

  540. 

  541. #define COS0_0  FIXR(0.50060299823519630134)

  542. #define COS0_1  FIXR(0.50547095989754365998)

  543. #define COS0_2  FIXR(0.51544730992262454697)

  544. #define COS0_3  FIXR(0.53104259108978417447)

  545. #define COS0_4  FIXR(0.55310389603444452782)

  546. #define COS0_5  FIXR(0.58293496820613387367)

  547. #define COS0_6  FIXR(0.62250412303566481615)

  548. #define COS0_7  FIXR(0.67480834145500574602)

  549. #define COS0_8  FIXR(0.74453627100229844977)

  550. #define COS0_9  FIXR(0.83934964541552703873)

  551. #define COS0_10 FIXR(0.97256823786196069369)

  552. #define COS0_11 FIXR(1.16943993343288495515)

  553. #define COS0_12 FIXR(1.48416461631416627724)

  554. #define COS0_13 FIXR(2.05778100995341155085)

  555. #define COS0_14 FIXR(3.40760841846871878570)

  556. #define COS0_15 FIXR(10.19000812354805681150)

  557. 

  558. #define COS1_0 FIXR(0.50241928618815570551)

  559. #define COS1_1 FIXR(0.52249861493968888062)

  560. #define COS1_2 FIXR(0.56694403481635770368)

  561. #define COS1_3 FIXR(0.64682178335999012954)

  562. #define COS1_4 FIXR(0.78815462345125022473)

  563. #define COS1_5 FIXR(1.06067768599034747134)

  564. #define COS1_6 FIXR(1.72244709823833392782)

  565. #define COS1_7 FIXR(5.10114861868916385802)

  566. 

  567. #define COS2_0 FIXR(0.50979557910415916894)

  568. #define COS2_1 FIXR(0.60134488693504528054)

  569. #define COS2_2 FIXR(0.89997622313641570463)

  570. #define COS2_3 FIXR(2.56291544774150617881)

  571. 

  572. #define COS3_0 FIXR(0.54119610014619698439)

  573. #define COS3_1 FIXR(1.30656296487637652785)

  574. 

  575. #define COS4_0 FIXR(0.70710678118654752439)

  576. 

  577. /* butterfly operator */

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

  579. {\

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

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

  582.     tab[a] = tmp0;\

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

  584. }

  585. 

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

  587. {\

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

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

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

  591. }

  592. 

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

  594. {\

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

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

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

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

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

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

  601. }

  602. 

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

  604. 

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

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

  607. {

  608.     int tmp0, tmp1;

  609. 

  610.     /* pass 1 */

  611.     BF(0, 31, COS0_0);

  612.     BF(1, 30, COS0_1);

  613.     BF(2, 29, COS0_2);

  614.     BF(3, 28, COS0_3);

  615.     BF(4, 27, COS0_4);

  616.     BF(5, 26, COS0_5);

  617.     BF(6, 25, COS0_6);

  618.     BF(7, 24, COS0_7);

  619.     BF(8, 23, COS0_8);

  620.     BF(9, 22, COS0_9);

  621.     BF(10, 21, COS0_10);

  622.     BF(11, 20, COS0_11);

  623.     BF(12, 19, COS0_12);

  624.     BF(13, 18, COS0_13);

  625.     BF(14, 17, COS0_14);

  626.     BF(15, 16, COS0_15);

  627. 

  628.     /* pass 2 */

  629.     BF(0, 15, COS1_0);

  630.     BF(1, 14, COS1_1);

  631.     BF(2, 13, COS1_2);

  632.     BF(3, 12, COS1_3);

  633.     BF(4, 11, COS1_4);

  634.     BF(5, 10, COS1_5);

  635.     BF(6,  9, COS1_6);

  636.     BF(7,  8, COS1_7);

  637.     

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

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

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

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

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

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

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

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

  646.     

  647.     /* pass 3 */

  648.     BF(0, 7, COS2_0);

  649.     BF(1, 6, COS2_1);

  650.     BF(2, 5, COS2_2);

  651.     BF(3, 4, COS2_3);

  652.     

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

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

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

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

  657.     

  658.     BF(16, 23, COS2_0);

  659.     BF(17, 22, COS2_1);

  660.     BF(18, 21, COS2_2);

  661.     BF(19, 20, COS2_3);

  662.     

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

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

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

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

  667. 

  668.     /* pass 4 */

  669.     BF(0, 3, COS3_0);

  670.     BF(1, 2, COS3_1);

  671.     

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

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

  674.     

  675.     BF(8, 11, COS3_0);

  676.     BF(9, 10, COS3_1);

  677.     

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

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

  680.     

  681.     BF(16, 19, COS3_0);

  682.     BF(17, 18, COS3_1);

  683.     

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

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

  686.     

  687.     BF(24, 27, COS3_0);

  688.     BF(25, 26, COS3_1);

  689.     

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

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

  692.     

  693.     /* pass 5 */

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

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

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

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

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

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

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

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

  702.     

  703.     /* pass 6 */

  704.     

  705.     ADD( 8, 12);

  706.     ADD(12, 10);

  707.     ADD(10, 14);

  708.     ADD(14,  9);

  709.     ADD( 9, 13);

  710.     ADD(13, 11);

  711.     ADD(11, 15);

  712. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  729.     

  730.     ADD(24, 28);

  731.     ADD(28, 26);

  732.     ADD(26, 30);

  733.     ADD(30, 25);

  734.     ADD(25, 29);

  735.     ADD(29, 27);

  736.     ADD(27, 31);

  737. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  754. }

  755. 

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

  757. 

  758. #if FRAC_BITS <= 15

  759. 

  760. static inline int round_sample(int sum)

  761. {

  762.     int sum1;

  763.     sum1 = (sum + (1 << (OUT_SHIFT - 1))) >> OUT_SHIFT;

  764.     if (sum1 < -32768)

  765.         sum1 = -32768;

  766.     else if (sum1 > 32767)

  767.         sum1 = 32767;

  768.     return sum1;

  769. }

  770. 

  771. #if defined(ARCH_POWERPC_405)

  772. 

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

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

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

  776. 

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

  778. #define MULS(ra, rb) \

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

  780. 

  781. #else

  782. 

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

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

  785. 

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

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

  788. 

  789. #endif

  790. 

  791. #else

  792. 

  793. static inline int round_sample(int64_t sum) 

  794. {

  795.     int sum1;

  796.     sum1 = (int)((sum + (int64_t_C(1) << (OUT_SHIFT - 1))) >> OUT_SHIFT);

  797.     if (sum1 < -32768)

  798.         sum1 = -32768;

  799.     else if (sum1 > 32767)

  800.         sum1 = 32767;

  801.     return sum1;

  802. }

  803. 

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

  805. 

  806. #endif

  807. 

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

  809. {                                               \

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

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

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

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

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

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

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

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

  818. }

  819. 

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

  821. {                                               \

  822.     int tmp;\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  847. }

  848. 

  849. 

  850. /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:

  851.    32 samples. */

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

  853. static void synth_filter(MPADecodeContext *s1,

  854.                          int ch, int16_t *samples, int incr, 

  855.                          int32_t sb_samples[SBLIMIT])

  856. {

  857.     int32_t tmp[32];

  858.     register MPA_INT *synth_buf;

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

  860.     int j, offset, v;

  861.     int16_t *samples2;

  862. #if FRAC_BITS <= 15

  863.     int sum, sum2;

  864. #else

  865.     int64_t sum, sum2;

  866. #endif

  867.     

  868.     dct32(tmp, sb_samples);

  869.     

  870.     offset = s1->synth_buf_offset[ch];

  871.     synth_buf = s1->synth_buf[ch] + offset;

  872. 

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

  874.         v = tmp[j];

  875. #if FRAC_BITS <= 15

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

  877.            sound */

  878.         if (v > 32767)

  879.             v = 32767;

  880.         else if (v < -32768)

  881.             v = -32768;

  882. #endif

  883.         synth_buf[j] = v;

  884.     }

  885.     /* copy to avoid wrap */

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

  887. 

  888.     samples2 = samples + 31 * incr;

  889.     w = window;

  890.     w2 = window + 31;

  891. 

  892.     sum = 0;

  893.     p = synth_buf + 16;

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

  895.     p = synth_buf + 48;

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

  897.     *samples = round_sample(sum);

  898.     samples += incr;

  899.     w++;

  900. 

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

  902.        access per two sample */

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

  904.         sum = 0;

  905.         sum2 = 0;

  906.         p = synth_buf + 16 + j;

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

  908.         p = synth_buf + 48 - j;

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

  910. 

  911.         *samples = round_sample(sum);

  912.         samples += incr;

  913.         *samples2 = round_sample(sum2);

  914.         samples2 -= incr;

  915.         w++;

  916.         w2--;

  917.     }

  918.     

  919.     p = synth_buf + 32;

  920.     sum = 0;

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

  922.     *samples = round_sample(sum);

  923. 

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

  925.     s1->synth_buf_offset[ch] = offset;

  926. }

  927. 

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

  929. #define C1  FIXR(0.99144486137381041114)

  930. #define C3  FIXR(0.92387953251128675612)

  931. #define C5  FIXR(0.79335334029123516458)

  932. #define C7  FIXR(0.60876142900872063941)

  933. #define C9  FIXR(0.38268343236508977173)

  934. #define C11 FIXR(0.13052619222005159154)

  935. 

  936. /* 12 points IMDCT. We compute it "by hand" by factorizing obvious

  937.    cases. */

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

  939. {

  940.     int tmp;

  941.     int64_t in1_3, in1_9, in4_3, in4_9;

  942. 

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

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

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

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

  947.     

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

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

  950.     out[0] = tmp;

  951.     out[5] = -tmp;

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

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

  954.     out[1] = tmp;

  955.     out[4] = -tmp;

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

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

  958.     out[2] = tmp;

  959.     out[3] = -tmp;

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

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

  962.     out[6] = tmp;

  963.     out[11] = tmp;

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

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

  966.     out[7] = tmp;

  967.     out[10] = tmp;

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

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

  970.     out[8] = tmp;

  971.     out[9] = tmp;

  972. }

  973. 

  974. #undef C1

  975. #undef C3

  976. #undef C5

  977. #undef C7

  978. #undef C9

  979. #undef C11

  980. 

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

  982. #define C1 FIXR(0.98480775301220805936)

  983. #define C2 FIXR(0.93969262078590838405)

  984. #define C3 FIXR(0.86602540378443864676)

  985. #define C4 FIXR(0.76604444311897803520)

  986. #define C5 FIXR(0.64278760968653932632)

  987. #define C6 FIXR(0.5)

  988. #define C7 FIXR(0.34202014332566873304)

  989. #define C8 FIXR(0.17364817766693034885)

  990. 

  991. /* 0.5 / cos(pi*(2*i+1)/36) */

  992. static const int icos36[9] = {

  993.     FIXR(0.50190991877167369479),

  994.     FIXR(0.51763809020504152469),

  995.     FIXR(0.55168895948124587824),

  996.     FIXR(0.61038729438072803416),

  997.     FIXR(0.70710678118654752439),

  998.     FIXR(0.87172339781054900991),

  999.     FIXR(1.18310079157624925896),

  1000.     FIXR(1.93185165257813657349),

  1001.     FIXR(5.73685662283492756461),

  1002. };

  1003. 

  1004. static const int icos72[18] = {

  1005.     /* 0.5 / cos(pi*(2*i+19)/72) */

  1006.     FIXR(0.74009361646113053152),

  1007.     FIXR(0.82133981585229078570),

  1008.     FIXR(0.93057949835178895673),

  1009.     FIXR(1.08284028510010010928),

  1010.     FIXR(1.30656296487637652785),

  1011.     FIXR(1.66275476171152078719),

  1012.     FIXR(2.31011315767264929558),

  1013.     FIXR(3.83064878777019433457),

  1014.     FIXR(11.46279281302667383546),

  1015. 

  1016.     /* 0.5 / cos(pi*(2*(i + 18) +19)/72) */

  1017.     FIXR(-0.67817085245462840086),

  1018.     FIXR(-0.63023620700513223342),

  1019.     FIXR(-0.59284452371708034528),

  1020.     FIXR(-0.56369097343317117734),

  1021.     FIXR(-0.54119610014619698439),

  1022.     FIXR(-0.52426456257040533932),

  1023.     FIXR(-0.51213975715725461845),

  1024.     FIXR(-0.50431448029007636036),

  1025.     FIXR(-0.50047634258165998492),

  1026. };

  1027. 

  1028. /* using Lee like decomposition followed by hand coded 9 points DCT */

  1029. static void imdct36(int *out, int *in)

  1030. {

  1031.     int i, j, t0, t1, t2, t3, s0, s1, s2, s3;

  1032.     int tmp[18], *tmp1, *in1;

  1033.     int64_t in3_3, in6_6;

  1034. 

  1035.     for(i=17;i>=1;i--)

  1036.         in[i] += in[i-1];

  1037.     for(i=17;i>=3;i-=2)

  1038.         in[i] += in[i-2];

  1039. 

  1040.     for(j=0;j<2;j++) {

  1041.         tmp1 = tmp + j;

  1042.         in1 = in + j;

  1043. 

  1044.         in3_3 = MUL64(in1[2*3], C3);

  1045.         in6_6 = MUL64(in1[2*6], C6);

  1046. 

  1047.         tmp1[0] = FRAC_RND(MUL64(in1[2*1], C1) + in3_3 + 

  1048.                            MUL64(in1[2*5], C5) + MUL64(in1[2*7], C7));

  1049.         tmp1[2] = in1[2*0] + FRAC_RND(MUL64(in1[2*2], C2) + 

  1050.                                       MUL64(in1[2*4], C4) + in6_6 + 

  1051.                                       MUL64(in1[2*8], C8));

  1052.         tmp1[4] = FRAC_RND(MUL64(in1[2*1] - in1[2*5] - in1[2*7], C3));

  1053.         tmp1[6] = FRAC_RND(MUL64(in1[2*2] - in1[2*4] - in1[2*8], C6)) - 

  1054.             in1[2*6] + in1[2*0];

  1055.         tmp1[8] = FRAC_RND(MUL64(in1[2*1], C5) - in3_3 - 

  1056.                            MUL64(in1[2*5], C7) + MUL64(in1[2*7], C1));

  1057.         tmp1[10] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C8) - 

  1058.                                        MUL64(in1[2*4], C2) + in6_6 + 

  1059.                                        MUL64(in1[2*8], C4));

  1060.         tmp1[12] = FRAC_RND(MUL64(in1[2*1], C7) - in3_3 + 

  1061.                             MUL64(in1[2*5], C1) - 

  1062.                             MUL64(in1[2*7], C5));

  1063.         tmp1[14] = in1[2*0] + FRAC_RND(MUL64(-in1[2*2], C4) + 

  1064.                                        MUL64(in1[2*4], C8) + in6_6 - 

  1065.                                        MUL64(in1[2*8], C2));

  1066.         tmp1[16] = in1[2*0] - in1[2*2] + in1[2*4] - in1[2*6] + in1[2*8];

  1067.     }

  1068. 

  1069.     i = 0;

  1070.     for(j=0;j<4;j++) {

  1071.         t0 = tmp[i];

  1072.         t1 = tmp[i + 2];

  1073.         s0 = t1 + t0;

  1074.         s2 = t1 - t0;

  1075. 

  1076.         t2 = tmp[i + 1];

  1077.         t3 = tmp[i + 3];

  1078.         s1 = MULL(t3 + t2, icos36[j]);

  1079.         s3 = MULL(t3 - t2, icos36[8 - j]);

  1080.         

  1081.         t0 = MULL(s0 + s1, icos72[9 + 8 - j]);

  1082.         t1 = MULL(s0 - s1, icos72[8 - j]);

  1083.         out[18 + 9 + j] = t0;

  1084.         out[18 + 8 - j] = t0;

  1085.         out[9 + j] = -t1;

  1086.         out[8 - j] = t1;

  1087.         

  1088.         t0 = MULL(s2 + s3, icos72[9+j]);

  1089.         t1 = MULL(s2 - s3, icos72[j]);

  1090.         out[18 + 9 + (8 - j)] = t0;

  1091.         out[18 + j] = t0;

  1092.         out[9 + (8 - j)] = -t1;

  1093.         out[j] = t1;

  1094.         i += 4;

  1095.     }

  1096. 

  1097.     s0 = tmp[16];

  1098.     s1 = MULL(tmp[17], icos36[4]);

  1099.     t0 = MULL(s0 + s1, icos72[9 + 4]);

  1100.     t1 = MULL(s0 - s1, icos72[4]);

  1101.     out[18 + 9 + 4] = t0;

  1102.     out[18 + 8 - 4] = t0;

  1103.     out[9 + 4] = -t1;

  1104.     out[8 - 4] = t1;

  1105. }

  1106. 

  1107. /* fast header check for resync */

  1108. static int check_header(uint32_t header)

  1109. {

  1110.     /* header */

  1111.     if ((header & 0xffe00000) != 0xffe00000)

  1112. 	return -1;

  1113.     /* layer check */

  1114.     if (((header >> 17) & 3) == 0)

  1115. 	return -1;

  1116.     /* bit rate */

  1117.     if (((header >> 12) & 0xf) == 0xf)

  1118. 	return -1;

  1119.     /* frequency */

  1120.     if (((header >> 10) & 3) == 3)

  1121. 	return -1;

  1122.     return 0;

  1123. }

  1124. 

  1125. /* header + layer + bitrate + freq + lsf/mpeg25 */

  1126. #define SAME_HEADER_MASK \

  1127.    (0xffe00000 | (3 << 17) | (0xf << 12) | (3 << 10) | (3 << 19))

  1128. 

  1129. /* header decoding. MUST check the header before because no

  1130.    consistency check is done there. Return 1 if free format found and

  1131.    that the frame size must be computed externally */

  1132. static int decode_header(MPADecodeContext *s, uint32_t header)

  1133. {

  1134.     int sample_rate, frame_size, mpeg25, padding;

  1135.     int sample_rate_index, bitrate_index;

  1136.     if (header & (1<<20)) {

  1137.         s->lsf = (header & (1<<19)) ? 0 : 1;

  1138.         mpeg25 = 0;

  1139.     } else {

  1140.         s->lsf = 1;

  1141.         mpeg25 = 1;

  1142.     }

  1143.     

  1144.     s->layer = 4 - ((header >> 17) & 3);

  1145.     /* extract frequency */

  1146.     sample_rate_index = (header >> 10) & 3;

  1147.     sample_rate = mpa_freq_tab[sample_rate_index] >> (s->lsf + mpeg25);

  1148.     sample_rate_index += 3 * (s->lsf + mpeg25);

  1149.     s->sample_rate_index = sample_rate_index;

  1150.     s->error_protection = ((header >> 16) & 1) ^ 1;

  1151.     s->sample_rate = sample_rate;

  1152. 

  1153.     bitrate_index = (header >> 12) & 0xf;

  1154.     padding = (header >> 9) & 1;

  1155.     //extension = (header >> 8) & 1;

  1156.     s->mode = (header >> 6) & 3;

  1157.     s->mode_ext = (header >> 4) & 3;

  1158.     //copyright = (header >> 3) & 1;

  1159.     //original = (header >> 2) & 1;

  1160.     //emphasis = header & 3;

  1161. 

  1162.     if (s->mode == MPA_MONO)

  1163.         s->nb_channels = 1;

  1164.     else

  1165.         s->nb_channels = 2;

  1166.     

  1167.     if (bitrate_index != 0) {

  1168.         frame_size = mpa_bitrate_tab[s->lsf][s->layer - 1][bitrate_index];

  1169.         s->bit_rate = frame_size * 1000;

  1170.         switch(s->layer) {

  1171.         case 1:

  1172.             frame_size = (frame_size * 12000) / sample_rate;

  1173.             frame_size = (frame_size + padding) * 4;

  1174.             break;

  1175.         case 2:

  1176.             frame_size = (frame_size * 144000) / sample_rate;

  1177.             frame_size += padding;

  1178.             break;

  1179.         default:

  1180.         case 3:

  1181.             frame_size = (frame_size * 144000) / (sample_rate << s->lsf);

  1182.             frame_size += padding;

  1183.             break;

  1184.         }

  1185.         s->frame_size = frame_size;

  1186.     } else {

  1187.         /* if no frame size computed, signal it */

  1188.         if (!s->free_format_frame_size)

  1189.             return 1;

  1190.         /* free format: compute bitrate and real frame size from the

  1191.            frame size we extracted by reading the bitstream */

  1192.         s->frame_size = s->free_format_frame_size;

  1193.         switch(s->layer) {

  1194.         case 1:

  1195.             s->frame_size += padding  * 4;

  1196.             s->bit_rate = (s->frame_size * sample_rate) / 48000;

  1197.             break;

  1198.         case 2:

  1199.             s->frame_size += padding;

  1200.             s->bit_rate = (s->frame_size * sample_rate) / 144000;

  1201.             break;

  1202.         default:

  1203.         case 3:

  1204.             s->frame_size += padding;

  1205.             s->bit_rate = (s->frame_size * (sample_rate << s->lsf)) / 144000;

  1206.             break;

  1207.         }

  1208.     }

  1209.     

  1210. #if defined(DEBUG)

  1211.     printf("layer%d, %d Hz, %d kbits/s, ",

  1212.            s->layer, s->sample_rate, s->bit_rate);

  1213.     if (s->nb_channels == 2) {

  1214.         if (s->layer == 3) {

  1215.             if (s->mode_ext & MODE_EXT_MS_STEREO)

  1216.                 printf("ms-");

  1217.             if (s->mode_ext & MODE_EXT_I_STEREO)

  1218.                 printf("i-");

  1219.         }

  1220.         printf("stereo");

  1221.     } else {

  1222.         printf("mono");

  1223.     }

  1224.     printf("\n");

  1225. #endif

  1226.     return 0;

  1227. }

  1228. 

  1229. /* useful helper to get mpeg audio stream infos. Return -1 if error in

  1230.    header, otherwise the coded frame size in bytes */

  1231. int mpa_decode_header(AVCodecContext *avctx, uint32_t head)

  1232. {

  1233.     MPADecodeContext s1, *s = &s1;

  1234. 

  1235.     if (check_header(head) != 0)

  1236.         return -1;

  1237. 

  1238.     if (decode_header(s, head) != 0) {

  1239.         return -1;

  1240.     }

  1241. 

  1242.     switch(s->layer) {

  1243.     case 1:

  1244.         avctx->frame_size = 384;

  1245.         break;

  1246.     case 2:

  1247.         avctx->frame_size = 1152;

  1248.         break;

  1249.     default:

  1250.     case 3:

  1251.         if (s->lsf)

  1252.             avctx->frame_size = 576;

  1253.         else

  1254.             avctx->frame_size = 1152;

  1255.         break;

  1256.     }

  1257. 

  1258.     avctx->sample_rate = s->sample_rate;

  1259.     avctx->channels = s->nb_channels;

  1260.     avctx->bit_rate = s->bit_rate;

  1261.     avctx->sub_id = s->layer;

  1262.     return s->frame_size;

  1263. }

  1264. 

  1265. /* return the number of decoded frames */

  1266. static int mp_decode_layer1(MPADecodeContext *s)

  1267. {

  1268.     int bound, i, v, n, ch, j, mant;

  1269.     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];

  1270.     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

  1271. 

  1272.     if (s->mode == MPA_JSTEREO) 

  1273.         bound = (s->mode_ext + 1) * 4;

  1274.     else

  1275.         bound = SBLIMIT;

  1276. 

  1277.     /* allocation bits */

  1278.     for(i=0;i<bound;i++) {

  1279.         for(ch=0;ch<s->nb_channels;ch++) {

  1280.             allocation[ch][i] = get_bits(&s->gb, 4);

  1281.         }

  1282.     }

  1283.     for(i=bound;i<SBLIMIT;i++) {

  1284.         allocation[0][i] = get_bits(&s->gb, 4);

  1285.     }

  1286. 

  1287.     /* scale factors */

  1288.     for(i=0;i<bound;i++) {

  1289.         for(ch=0;ch<s->nb_channels;ch++) {

  1290.             if (allocation[ch][i])

  1291.                 scale_factors[ch][i] = get_bits(&s->gb, 6);

  1292.         }

  1293.     }

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

  1295.         if (allocation[0][i]) {

  1296.             scale_factors[0][i] = get_bits(&s->gb, 6);

  1297.             scale_factors[1][i] = get_bits(&s->gb, 6);

  1298.         }

  1299.     }

  1300.     

  1301.     /* compute samples */

  1302.     for(j=0;j<12;j++) {

  1303.         for(i=0;i<bound;i++) {

  1304.             for(ch=0;ch<s->nb_channels;ch++) {

  1305.                 n = allocation[ch][i];

  1306.                 if (n) {

  1307.                     mant = get_bits(&s->gb, n + 1);

  1308.                     v = l1_unscale(n, mant, scale_factors[ch][i]);

  1309.                 } else {

  1310.                     v = 0;

  1311.                 }

  1312.                 s->sb_samples[ch][j][i] = v;

  1313.             }

  1314.         }

  1315.         for(i=bound;i<SBLIMIT;i++) {

  1316.             n = allocation[0][i];

  1317.             if (n) {

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

  1319.                 v = l1_unscale(n, mant, scale_factors[0][i]);

  1320.                 s->sb_samples[0][j][i] = v;

  1321.                 v = l1_unscale(n, mant, scale_factors[1][i]);

  1322.                 s->sb_samples[1][j][i] = v;

  1323.             } else {

  1324.                 s->sb_samples[0][j][i] = 0;

  1325.                 s->sb_samples[1][j][i] = 0;

  1326.             }

  1327.         }

  1328.     }

  1329.     return 12;

  1330. }

  1331. 

  1332. /* bitrate is in kb/s */

  1333. int l2_select_table(int bitrate, int nb_channels, int freq, int lsf)

  1334. {

  1335.     int ch_bitrate, table;

  1336.     

  1337.     ch_bitrate = bitrate / nb_channels;

  1338.     if (!lsf) {

  1339.         if ((freq == 48000 && ch_bitrate >= 56) ||

  1340.             (ch_bitrate >= 56 && ch_bitrate <= 80)) 

  1341.             table = 0;

  1342.         else if (freq != 48000 && ch_bitrate >= 96) 

  1343.             table = 1;

  1344.         else if (freq != 32000 && ch_bitrate <= 48) 

  1345.             table = 2;

  1346.         else 

  1347.             table = 3;

  1348.     } else {

  1349.         table = 4;

  1350.     }

  1351.     return table;

  1352. }

  1353. 

  1354. static int mp_decode_layer2(MPADecodeContext *s)

  1355. {

  1356.     int sblimit; /* number of used subbands */

  1357.     const unsigned char *alloc_table;

  1358.     int table, bit_alloc_bits, i, j, ch, bound, v;

  1359.     unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];

  1360.     unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];

  1361.     unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;

  1362.     int scale, qindex, bits, steps, k, l, m, b;

  1363. 

  1364.     /* select decoding table */

  1365.     table = l2_select_table(s->bit_rate / 1000, s->nb_channels, 

  1366.                             s->sample_rate, s->lsf);

  1367.     sblimit = sblimit_table[table];

  1368.     alloc_table = alloc_tables[table];

  1369. 

  1370.     if (s->mode == MPA_JSTEREO) 

  1371.         bound = (s->mode_ext + 1) * 4;

  1372.     else

  1373.         bound = sblimit;

  1374. 

  1375.     dprintf("bound=%d sblimit=%d\n", bound, sblimit);

  1376.     /* parse bit allocation */

  1377.     j = 0;

  1378.     for(i=0;i<bound;i++) {

  1379.         bit_alloc_bits = alloc_table[j];

  1380.         for(ch=0;ch<s->nb_channels;ch++) {

  1381.             bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);

  1382.         }

  1383.         j += 1 << bit_alloc_bits;

  1384.     }

  1385.     for(i=bound;i<sblimit;i++) {

  1386.         bit_alloc_bits = alloc_table[j];

  1387.         v = get_bits(&s->gb, bit_alloc_bits);

  1388.         bit_alloc[0][i] = v;

  1389.         bit_alloc[1][i] = v;

  1390.         j += 1 << bit_alloc_bits;

  1391.     }

  1392. 

  1393. #ifdef DEBUG

  1394.     {

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

  1396.             for(i=0;i<sblimit;i++)

  1397.                 printf(" %d", bit_alloc[ch][i]);

  1398.             printf("\n");

  1399.         }

  1400.     }

  1401. #endif

  1402. 

  1403.     /* scale codes */

  1404.     for(i=0;i<sblimit;i++) {

  1405.         for(ch=0;ch<s->nb_channels;ch++) {

  1406.             if (bit_alloc[ch][i]) 

  1407.                 scale_code[ch][i] = get_bits(&s->gb, 2);

  1408.         }

  1409.     }

  1410.     

  1411.     /* scale factors */

  1412.     for(i=0;i<sblimit;i++) {

  1413.         for(ch=0;ch<s->nb_channels;ch++) {

  1414.             if (bit_alloc[ch][i]) {

  1415.                 sf = scale_factors[ch][i];

  1416.                 switch(scale_code[ch][i]) {

  1417.                 default:

  1418.                 case 0:

  1419.                     sf[0] = get_bits(&s->gb, 6);

  1420.                     sf[1] = get_bits(&s->gb, 6);

  1421.                     sf[2] = get_bits(&s->gb, 6);

  1422.                     break;

  1423.                 case 2:

  1424.                     sf[0] = get_bits(&s->gb, 6);

  1425.                     sf[1] = sf[0];

  1426.                     sf[2] = sf[0];

  1427.                     break;

  1428.                 case 1:

  1429.                     sf[0] = get_bits(&s->gb, 6);

  1430.                     sf[2] = get_bits(&s->gb, 6);

  1431.                     sf[1] = sf[0];

  1432.                     break;

  1433.                 case 3:

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

  1435.                     sf[2] = get_bits(&s->gb, 6);

  1436.                     sf[1] = sf[2];

  1437.                     break;

  1438.                 }

  1439.             }

  1440.         }

  1441.     }

  1442. 

  1443. #ifdef DEBUG

  1444.     for(ch=0;ch<s->nb_channels;ch++) {

  1445.         for(i=0;i<sblimit;i++) {

  1446.             if (bit_alloc[ch][i]) {

  1447.                 sf = scale_factors[ch][i];

  1448.                 printf(" %d %d %d", sf[0], sf[1], sf[2]);

  1449.             } else {

  1450.                 printf(" -");

  1451.             }

  1452.         }

  1453.         printf("\n");

  1454.     }

  1455. #endif

  1456. 

  1457.     /* samples */

  1458.     for(k=0;k<3;k++) {

  1459.         for(l=0;l<12;l+=3) {

  1460.             j = 0;

  1461.             for(i=0;i<bound;i++) {

  1462.                 bit_alloc_bits = alloc_table[j];

  1463.                 for(ch=0;ch<s->nb_channels;ch++) {

  1464.                     b = bit_alloc[ch][i];

  1465.                     if (b) {

  1466.                         scale = scale_factors[ch][i][k];

  1467.                         qindex = alloc_table[j+b];

  1468.                         bits = quant_bits[qindex];

  1469.                         if (bits < 0) {

  1470.                             /* 3 values at the same time */

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

  1472.                             steps = quant_steps[qindex];

  1473.                             s->sb_samples[ch][k * 12 + l + 0][i] = 

  1474.                                 l2_unscale_group(steps, v % steps, scale);

  1475.                             v = v / steps;

  1476.                             s->sb_samples[ch][k * 12 + l + 1][i] = 

  1477.                                 l2_unscale_group(steps, v % steps, scale);

  1478.                             v = v / steps;

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

  1480.                                 l2_unscale_group(steps, v, scale);

  1481.                         } else {

  1482.                             for(m=0;m<3;m++) {

  1483.                                 v = get_bits(&s->gb, bits);

  1484.                                 v = l1_unscale(bits - 1, v, scale);

  1485.                                 s->sb_samples[ch][k * 12 + l + m][i] = v;

  1486.                             }

  1487.                         }

  1488.                     } else {

  1489.                         s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1490.                         s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1491.                         s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1492.                     }

  1493.                 }

  1494.                 /* next subband in alloc table */

  1495.                 j += 1 << bit_alloc_bits; 

  1496.             }

  1497.             /* XXX: find a way to avoid this duplication of code */

  1498.             for(i=bound;i<sblimit;i++) {

  1499.                 bit_alloc_bits = alloc_table[j];

  1500.                 b = bit_alloc[0][i];

  1501.                 if (b) {

  1502.                     int mant, scale0, scale1;

  1503.                     scale0 = scale_factors[0][i][k];

  1504.                     scale1 = scale_factors[1][i][k];

  1505.                     qindex = alloc_table[j+b];

  1506.                     bits = quant_bits[qindex];

  1507.                     if (bits < 0) {

  1508.                         /* 3 values at the same time */

  1509.                         v = get_bits(&s->gb, -bits);

  1510.                         steps = quant_steps[qindex];

  1511.                         mant = v % steps;

  1512.                         v = v / steps;

  1513.                         s->sb_samples[0][k * 12 + l + 0][i] = 

  1514.                             l2_unscale_group(steps, mant, scale0);

  1515.                         s->sb_samples[1][k * 12 + l + 0][i] = 

  1516.                             l2_unscale_group(steps, mant, scale1);

  1517.                         mant = v % steps;

  1518.                         v = v / steps;

  1519.                         s->sb_samples[0][k * 12 + l + 1][i] = 

  1520.                             l2_unscale_group(steps, mant, scale0);

  1521.                         s->sb_samples[1][k * 12 + l + 1][i] = 

  1522.                             l2_unscale_group(steps, mant, scale1);

  1523.                         s->sb_samples[0][k * 12 + l + 2][i] = 

  1524.                             l2_unscale_group(steps, v, scale0);

  1525.                         s->sb_samples[1][k * 12 + l + 2][i] = 

  1526.                             l2_unscale_group(steps, v, scale1);

  1527.                     } else {

  1528.                         for(m=0;m<3;m++) {

  1529.                             mant = get_bits(&s->gb, bits);

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

  1531.                                 l1_unscale(bits - 1, mant, scale0);

  1532.                             s->sb_samples[1][k * 12 + l + m][i] = 

  1533.                                 l1_unscale(bits - 1, mant, scale1);

  1534.                         }

  1535.                     }

  1536.                 } else {

  1537.                     s->sb_samples[0][k * 12 + l + 0][i] = 0;

  1538.                     s->sb_samples[0][k * 12 + l + 1][i] = 0;

  1539.                     s->sb_samples[0][k * 12 + l + 2][i] = 0;

  1540.                     s->sb_samples[1][k * 12 + l + 0][i] = 0;

  1541.                     s->sb_samples[1][k * 12 + l + 1][i] = 0;

  1542.                     s->sb_samples[1][k * 12 + l + 2][i] = 0;

  1543.                 }

  1544.                 /* next subband in alloc table */

  1545.                 j += 1 << bit_alloc_bits; 

  1546.             }

  1547.             /* fill remaining samples to zero */

  1548.             for(i=sblimit;i<SBLIMIT;i++) {

  1549.                 for(ch=0;ch<s->nb_channels;ch++) {

  1550.                     s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1551.                     s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1552.                     s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1553.                 }

  1554.             }

  1555.         }

  1556.     }

  1557.     return 3 * 12;

  1558. }

  1559. 

  1560. /*

  1561.  * Seek back in the stream for backstep bytes (at most 511 bytes)

  1562.  */

  1563. static void seek_to_maindata(MPADecodeContext *s, unsigned int backstep)

  1564. {

  1565.     uint8_t *ptr;

  1566. 

  1567.     /* compute current position in stream */

  1568.     ptr = (uint8_t *)(s->gb.buffer + (get_bits_count(&s->gb)>>3));

  1569. 

  1570.     /* copy old data before current one */

  1571.     ptr -= backstep;

  1572.     memcpy(ptr, s->inbuf1[s->inbuf_index ^ 1] + 

  1573.            BACKSTEP_SIZE + s->old_frame_size - backstep, backstep);

  1574.     /* init get bits again */

  1575.     init_get_bits(&s->gb, ptr, (s->frame_size + backstep)*8);

  1576. 

  1577.     /* prepare next buffer */

  1578.     s->inbuf_index ^= 1;

  1579.     s->inbuf = &s->inbuf1[s->inbuf_index][BACKSTEP_SIZE];

  1580.     s->old_frame_size = s->frame_size;

  1581. }

  1582. 

  1583. static inline void lsf_sf_expand(int *slen,

  1584.                                  int sf, int n1, int n2, int n3)

  1585. {

  1586.     if (n3) {

  1587.         slen[3] = sf % n3;

  1588.         sf /= n3;

  1589.     } else {

  1590.         slen[3] = 0;

  1591.     }

  1592.     if (n2) {

  1593.         slen[2] = sf % n2;

  1594.         sf /= n2;

  1595.     } else {

  1596.         slen[2] = 0;

  1597.     }

  1598.     slen[1] = sf % n1;

  1599.     sf /= n1;

  1600.     slen[0] = sf;

  1601. }

  1602. 

  1603. static void exponents_from_scale_factors(MPADecodeContext *s, 

  1604.                                          GranuleDef *g,

  1605.                                          int16_t *exponents)

  1606. {

  1607.     const uint8_t *bstab, *pretab;

  1608.     int len, i, j, k, l, v0, shift, gain, gains[3];

  1609.     int16_t *exp_ptr;

  1610. 

  1611.     exp_ptr = exponents;

  1612.     gain = g->global_gain - 210;

  1613.     shift = g->scalefac_scale + 1;

  1614. 

  1615.     bstab = band_size_long[s->sample_rate_index];

  1616.     pretab = mpa_pretab[g->preflag];

  1617.     for(i=0;i<g->long_end;i++) {

  1618.         v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift);

  1619.         len = bstab[i];

  1620.         for(j=len;j>0;j--)

  1621.             *exp_ptr++ = v0;

  1622.     }

  1623. 

  1624.     if (g->short_start < 13) {

  1625.         bstab = band_size_short[s->sample_rate_index];

  1626.         gains[0] = gain - (g->subblock_gain[0] << 3);

  1627.         gains[1] = gain - (g->subblock_gain[1] << 3);

  1628.         gains[2] = gain - (g->subblock_gain[2] << 3);

  1629.         k = g->long_end;

  1630.         for(i=g->short_start;i<13;i++) {

  1631.             len = bstab[i];

  1632.             for(l=0;l<3;l++) {

  1633.                 v0 = gains[l] - (g->scale_factors[k++] << shift);

  1634.                 for(j=len;j>0;j--)

  1635.                 *exp_ptr++ = v0;

  1636.             }

  1637.         }

  1638.     }

  1639. }

  1640. 

  1641. /* handle n = 0 too */

  1642. static inline int get_bitsz(GetBitContext *s, int n)

  1643. {

  1644.     if (n == 0)

  1645.         return 0;

  1646.     else

  1647.         return get_bits(s, n);

  1648. }

  1649. 

  1650. static int huffman_decode(MPADecodeContext *s, GranuleDef *g,

  1651.                           int16_t *exponents, int end_pos)

  1652. {

  1653.     int s_index;

  1654.     int linbits, code, x, y, l, v, i, j, k, pos;

  1655.     GetBitContext last_gb;

  1656.     VLC *vlc;

  1657.     uint8_t *code_table;

  1658. 

  1659.     /* low frequencies (called big values) */

  1660.     s_index = 0;

  1661.     for(i=0;i<3;i++) {

  1662.         j = g->region_size[i];

  1663.         if (j == 0)

  1664.             continue;

  1665.         /* select vlc table */

  1666.         k = g->table_select[i];

  1667.         l = mpa_huff_data[k][0];

  1668.         linbits = mpa_huff_data[k][1];

  1669.         vlc = &huff_vlc[l];

  1670.         code_table = huff_code_table[l];

  1671. 

  1672.         /* read huffcode and compute each couple */

  1673.         for(;j>0;j--) {

  1674.             if (get_bits_count(&s->gb) >= end_pos)

  1675.                 break;

  1676.             if (code_table) {

  1677.                 code = get_vlc(&s->gb, vlc);

  1678.                 if (code < 0)

  1679.                     return -1;

  1680.                 y = code_table[code];

  1681.                 x = y >> 4;

  1682.                 y = y & 0x0f;

  1683.             } else {

  1684.                 x = 0;

  1685.                 y = 0;

  1686.             }

  1687.             dprintf("region=%d n=%d x=%d y=%d exp=%d\n", 

  1688.                     i, g->region_size[i] - j, x, y, exponents[s_index]);

  1689.             if (x) {

  1690.                 if (x == 15)

  1691.                     x += get_bitsz(&s->gb, linbits);

  1692.                 v = l3_unscale(x, exponents[s_index]);

  1693.                 if (get_bits1(&s->gb))

  1694.                     v = -v;

  1695.             } else {

  1696.                 v = 0;

  1697.             }

  1698.             g->sb_hybrid[s_index++] = v;

  1699.             if (y) {

  1700.                 if (y == 15)

  1701.                     y += get_bitsz(&s->gb, linbits);

  1702.                 v = l3_unscale(y, exponents[s_index]);

  1703.                 if (get_bits1(&s->gb))

  1704.                     v = -v;

  1705.             } else {

  1706.                 v = 0;

  1707.             }

  1708.             g->sb_hybrid[s_index++] = v;

  1709.         }

  1710.     }

  1711.             

  1712.     /* high frequencies */

  1713.     vlc = &huff_quad_vlc[g->count1table_select];

  1714.     last_gb.buffer = NULL;

  1715.     while (s_index <= 572) {

  1716.         pos = get_bits_count(&s->gb);

  1717.         if (pos >= end_pos) {

  1718.             if (pos > end_pos && last_gb.buffer != NULL) {

  1719.                 /* some encoders generate an incorrect size for this

  1720.                    part. We must go back into the data */

  1721.                 s_index -= 4;

  1722.                 s->gb = last_gb;

  1723.             }

  1724.             break;

  1725.         }

  1726.         last_gb= s->gb;

  1727. 

  1728.         code = get_vlc(&s->gb, vlc);

  1729.         dprintf("t=%d code=%d\n", g->count1table_select, code);

  1730.         if (code < 0)

  1731.             return -1;

  1732.         for(i=0;i<4;i++) {

  1733.             if (code & (8 >> i)) {

  1734.                 /* non zero value. Could use a hand coded function for

  1735.                    'one' value */

  1736.                 v = l3_unscale(1, exponents[s_index]);

  1737.                 if(get_bits1(&s->gb))

  1738.                     v = -v;

  1739.             } else {

  1740.                 v = 0;

  1741.             }

  1742.             g->sb_hybrid[s_index++] = v;

  1743.         }

  1744.     }

  1745.     while (s_index < 576)

  1746.         g->sb_hybrid[s_index++] = 0;

  1747.     return 0;

  1748. }

  1749. 

  1750. /* Reorder short blocks from bitstream order to interleaved order. It

  1751.    would be faster to do it in parsing, but the code would be far more

  1752.    complicated */

  1753. static void reorder_block(MPADecodeContext *s, GranuleDef *g)

  1754. {

  1755.     int i, j, k, len;

  1756.     int32_t *ptr, *dst, *ptr1;

  1757.     int32_t tmp[576];

  1758. 

  1759.     if (g->block_type != 2)

  1760.         return;

  1761. 

  1762.     if (g->switch_point) {

  1763.         if (s->sample_rate_index != 8) {

  1764.             ptr = g->sb_hybrid + 36;

  1765.         } else {

  1766.             ptr = g->sb_hybrid + 48;

  1767.         }

  1768.     } else {

  1769.         ptr = g->sb_hybrid;

  1770.     }

  1771.     

  1772.     for(i=g->short_start;i<13;i++) {

  1773.         len = band_size_short[s->sample_rate_index][i];

  1774.         ptr1 = ptr;

  1775.         for(k=0;k<3;k++) {

  1776.             dst = tmp + k;

  1777.             for(j=len;j>0;j--) {

  1778.                 *dst = *ptr++;

  1779.                 dst += 3;

  1780.             }

  1781.         }

  1782.         memcpy(ptr1, tmp, len * 3 * sizeof(int32_t));

  1783.     }

  1784. }

  1785. 

  1786. #define ISQRT2 FIXR(0.70710678118654752440)

  1787. 

  1788. static void compute_stereo(MPADecodeContext *s,

  1789.                            GranuleDef *g0, GranuleDef *g1)

  1790. {

  1791.     int i, j, k, l;

  1792.     int32_t v1, v2;

  1793.     int sf_max, tmp0, tmp1, sf, len, non_zero_found;

  1794.     int32_t (*is_tab)[16];

  1795.     int32_t *tab0, *tab1;

  1796.     int non_zero_found_short[3];

  1797. 

  1798.     /* intensity stereo */

  1799.     if (s->mode_ext & MODE_EXT_I_STEREO) {

  1800.         if (!s->lsf) {

  1801.             is_tab = is_table;

  1802.             sf_max = 7;

  1803.         } else {

  1804.             is_tab = is_table_lsf[g1->scalefac_compress & 1];

  1805.             sf_max = 16;

  1806.         }

  1807.             

  1808.         tab0 = g0->sb_hybrid + 576;

  1809.         tab1 = g1->sb_hybrid + 576;

  1810. 

  1811.         non_zero_found_short[0] = 0;

  1812.         non_zero_found_short[1] = 0;

  1813.         non_zero_found_short[2] = 0;

  1814.         k = (13 - g1->short_start) * 3 + g1->long_end - 3;

  1815.         for(i = 12;i >= g1->short_start;i--) {

  1816.             /* for last band, use previous scale factor */

  1817.             if (i != 11)

  1818.                 k -= 3;

  1819.             len = band_size_short[s->sample_rate_index][i];

  1820.             for(l=2;l>=0;l--) {

  1821.                 tab0 -= len;

  1822.                 tab1 -= len;

  1823.                 if (!non_zero_found_short[l]) {

  1824.                     /* test if non zero band. if so, stop doing i-stereo */

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

  1826.                         if (tab1[j] != 0) {

  1827.                             non_zero_found_short[l] = 1;

  1828.                             goto found1;

  1829.                         }

  1830.                     }

  1831.                     sf = g1->scale_factors[k + l];

  1832.                     if (sf >= sf_max)

  1833.                         goto found1;

  1834. 

  1835.                     v1 = is_tab[0][sf];

  1836.                     v2 = is_tab[1][sf];

  1837.                     for(j=0;j<len;j++) {

  1838.                         tmp0 = tab0[j];

  1839.                         tab0[j] = MULL(tmp0, v1);

  1840.                         tab1[j] = MULL(tmp0, v2);

  1841.                     }

  1842.                 } else {

  1843.                 found1:

  1844.                     if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1845.                         /* lower part of the spectrum : do ms stereo

  1846.                            if enabled */

  1847.                         for(j=0;j<len;j++) {

  1848.                             tmp0 = tab0[j];

  1849.                             tmp1 = tab1[j];

  1850.                             tab0[j] = MULL(tmp0 + tmp1, ISQRT2);

  1851.                             tab1[j] = MULL(tmp0 - tmp1, ISQRT2);

  1852.                         }

  1853.                     }

  1854.                 }

  1855.             }

  1856.         }

  1857. 

  1858.         non_zero_found = non_zero_found_short[0] | 

  1859.             non_zero_found_short[1] | 

  1860.             non_zero_found_short[2];

  1861. 

  1862.         for(i = g1->long_end - 1;i >= 0;i--) {

  1863.             len = band_size_long[s->sample_rate_index][i];

  1864.             tab0 -= len;

  1865.             tab1 -= len;

  1866.             /* test if non zero band. if so, stop doing i-stereo */

  1867.             if (!non_zero_found) {

  1868.                 for(j=0;j<len;j++) {

  1869.                     if (tab1[j] != 0) {

  1870.                         non_zero_found = 1;

  1871.                         goto found2;

  1872.                     }

  1873.                 }

  1874.                 /* for last band, use previous scale factor */

  1875.                 k = (i == 21) ? 20 : i;

  1876.                 sf = g1->scale_factors[k];

  1877.                 if (sf >= sf_max)

  1878.                     goto found2;

  1879.                 v1 = is_tab[0][sf];

  1880.                 v2 = is_tab[1][sf];

  1881.                 for(j=0;j<len;j++) {

  1882.                     tmp0 = tab0[j];

  1883.                     tab0[j] = MULL(tmp0, v1);

  1884.                     tab1[j] = MULL(tmp0, v2);

  1885.                 }

  1886.             } else {

  1887.             found2:

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

  1889.                     /* lower part of the spectrum : do ms stereo

  1890.                        if enabled */

  1891.                     for(j=0;j<len;j++) {

  1892.                         tmp0 = tab0[j];

  1893.                         tmp1 = tab1[j];

  1894.                         tab0[j] = MULL(tmp0 + tmp1, ISQRT2);

  1895.                         tab1[j] = MULL(tmp0 - tmp1, ISQRT2);

  1896.                     }

  1897.                 }

  1898.             }

  1899.         }

  1900.     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1901.         /* ms stereo ONLY */

  1902.         /* NOTE: the 1/sqrt(2) normalization factor is included in the

  1903.            global gain */

  1904.         tab0 = g0->sb_hybrid;

  1905.         tab1 = g1->sb_hybrid;

  1906.         for(i=0;i<576;i++) {

  1907.             tmp0 = tab0[i];

  1908.             tmp1 = tab1[i];

  1909.             tab0[i] = tmp0 + tmp1;

  1910.             tab1[i] = tmp0 - tmp1;

  1911.         }

  1912.     }

  1913. }

  1914. 

  1915. static void compute_antialias_integer(MPADecodeContext *s,

  1916.                               GranuleDef *g)

  1917. {

  1918.     int32_t *ptr, *p0, *p1, *csa;

  1919.     int n, i, j;

  1920. 

  1921.     /* we antialias only "long" bands */

  1922.     if (g->block_type == 2) {

  1923.         if (!g->switch_point)

  1924.             return;

  1925.         /* XXX: check this for 8000Hz case */

  1926.         n = 1;

  1927.     } else {

  1928.         n = SBLIMIT - 1;

  1929.     }

  1930.     

  1931.     ptr = g->sb_hybrid + 18;

  1932.     for(i = n;i > 0;i--) {

  1933.         p0 = ptr - 1;

  1934.         p1 = ptr;

  1935.         csa = &csa_table[0][0];       

  1936.         for(j=0;j<4;j++) {

  1937.             int tmp0 = *p0;

  1938.             int tmp1 = *p1;

  1939. #if 0

  1940.             *p0 = FRAC_RND(MUL64(tmp0, csa[0]) - MUL64(tmp1, csa[1]));

  1941.             *p1 = FRAC_RND(MUL64(tmp0, csa[1]) + MUL64(tmp1, csa[0]));

  1942. #else

  1943.             int64_t tmp2= MUL64(tmp0 + tmp1, csa[0]);

  1944.             *p0 = FRAC_RND(tmp2 - MUL64(tmp1, csa[2]));

  1945.             *p1 = FRAC_RND(tmp2 + MUL64(tmp0, csa[3]));

  1946. #endif

  1947.             p0--; p1++;

  1948.             csa += 4;

  1949.             tmp0 = *p0;

  1950.             tmp1 = *p1;

  1951. #if 0

  1952.             *p0 = FRAC_RND(MUL64(tmp0, csa[0]) - MUL64(tmp1, csa[1]));

  1953.             *p1 = FRAC_RND(MUL64(tmp0, csa[1]) + MUL64(tmp1, csa[0]));

  1954. #else

  1955.             tmp2= MUL64(tmp0 + tmp1, csa[0]);

  1956.             *p0 = FRAC_RND(tmp2 - MUL64(tmp1, csa[2]));

  1957.             *p1 = FRAC_RND(tmp2 + MUL64(tmp0, csa[3]));

  1958. #endif

  1959.             p0--; p1++;

  1960.             csa += 4;

  1961.         }

  1962.         ptr += 18;       

  1963.     }

  1964. }

  1965. 

  1966. static void compute_antialias_float(MPADecodeContext *s,

  1967.                               GranuleDef *g)

  1968. {

  1969.     int32_t *ptr, *p0, *p1;

  1970.     int n, i, j;

  1971. 

  1972.     /* we antialias only "long" bands */

  1973.     if (g->block_type == 2) {

  1974.         if (!g->switch_point)

  1975.             return;

  1976.         /* XXX: check this for 8000Hz case */

  1977.         n = 1;

  1978.     } else {

  1979.         n = SBLIMIT - 1;

  1980.     }

  1981.     

  1982.     ptr = g->sb_hybrid + 18;

  1983.     for(i = n;i > 0;i--) {

  1984.         float *csa = &csa_table_float[0][0];       

  1985.         p0 = ptr - 1;

  1986.         p1 = ptr;

  1987.         for(j=0;j<4;j++) {

  1988.             float tmp0 = *p0;

  1989.             float tmp1 = *p1;

  1990. #if 1

  1991.             *p0 = lrintf(tmp0 * csa[0] - tmp1 * csa[1]);

  1992.             *p1 = lrintf(tmp0 * csa[1] + tmp1 * csa[0]);

  1993. #else

  1994.             float tmp2= (tmp0 + tmp1) * csa[0];

  1995.             *p0 = lrintf(tmp2 - tmp1 * csa[2]);

  1996.             *p1 = lrintf(tmp2 + tmp0 * csa[3]);

  1997. #endif

  1998.             p0--; p1++;

  1999.             csa += 4;

  2000.             tmp0 = *p0;

  2001.             tmp1 = *p1;

  2002. #if 1

  2003.             *p0 = lrintf(tmp0 * csa[0] - tmp1 * csa[1]);

  2004.             *p1 = lrintf(tmp0 * csa[1] + tmp1 * csa[0]);

  2005. #else

  2006.             tmp2= (tmp0 + tmp1) * csa[0];

  2007.             *p0 = lrintf(tmp2 - tmp1 * csa[2]);

  2008.             *p1 = lrintf(tmp2 + tmp0 * csa[3]);

  2009. #endif

  2010.             p0--; p1++;

  2011.             csa += 4;

  2012.         }

  2013.         ptr += 18;       

  2014.     }

  2015. }

  2016. 

  2017. static void compute_imdct(MPADecodeContext *s,

  2018.                           GranuleDef *g, 

  2019.                           int32_t *sb_samples,

  2020.                           int32_t *mdct_buf)

  2021. {

  2022.     int32_t *ptr, *win, *win1, *buf, *buf2, *out_ptr, *ptr1;

  2023.     int32_t in[6];

  2024.     int32_t out[36];

  2025.     int32_t out2[12];

  2026.     int i, j, k, mdct_long_end, v, sblimit;

  2027. 

  2028.     /* find last non zero block */

  2029.     ptr = g->sb_hybrid + 576;

  2030.     ptr1 = g->sb_hybrid + 2 * 18;

  2031.     while (ptr >= ptr1) {

  2032.         ptr -= 6;

  2033.         v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];

  2034.         if (v != 0)

  2035.             break;

  2036.     }

  2037.     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

  2038. 

  2039.     if (g->block_type == 2) {

  2040.         /* XXX: check for 8000 Hz */

  2041.         if (g->switch_point)

  2042.             mdct_long_end = 2;

  2043.         else

  2044.             mdct_long_end = 0;

  2045.     } else {

  2046.         mdct_long_end = sblimit;

  2047.     }

  2048. 

  2049.     buf = mdct_buf;

  2050.     ptr = g->sb_hybrid;

  2051.     for(j=0;j<mdct_long_end;j++) {

  2052.         imdct36(out, ptr);

  2053.         /* apply window & overlap with previous buffer */

  2054.         out_ptr = sb_samples + j;

  2055.         /* select window */

  2056.         if (g->switch_point && j < 2)

  2057.             win1 = mdct_win[0];

  2058.         else

  2059.             win1 = mdct_win[g->block_type];

  2060.         /* select frequency inversion */

  2061.         win = win1 + ((4 * 36) & -(j & 1));

  2062.         for(i=0;i<18;i++) {

  2063.             *out_ptr = MULL(out[i], win[i]) + buf[i];

  2064.             buf[i] = MULL(out[i + 18], win[i + 18]);

  2065.             out_ptr += SBLIMIT;

  2066.         }

  2067.         ptr += 18;

  2068.         buf += 18;

  2069.     }

  2070.     for(j=mdct_long_end;j<sblimit;j++) {

  2071.         for(i=0;i<6;i++) {

  2072.             out[i] = 0;

  2073.             out[6 + i] = 0;

  2074.             out[30+i] = 0;

  2075.         }

  2076.         /* select frequency inversion */

  2077.         win = mdct_win[2] + ((4 * 36) & -(j & 1));

  2078.         buf2 = out + 6;

  2079.         for(k=0;k<3;k++) {

  2080.             /* reorder input for short mdct */

  2081.             ptr1 = ptr + k;

  2082.             for(i=0;i<6;i++) {

  2083.                 in[i] = *ptr1;

  2084.                 ptr1 += 3;

  2085.             }

  2086.             imdct12(out2, in);

  2087.             /* apply 12 point window and do small overlap */

  2088.             for(i=0;i<6;i++) {

  2089.                 buf2[i] = MULL(out2[i], win[i]) + buf2[i];

  2090.                 buf2[i + 6] = MULL(out2[i + 6], win[i + 6]);

  2091.             }

  2092.             buf2 += 6;

  2093.         }

  2094.         /* overlap */

  2095.         out_ptr = sb_samples + j;

  2096.         for(i=0;i<18;i++) {

  2097.             *out_ptr = out[i] + buf[i];

  2098.             buf[i] = out[i + 18];

  2099.             out_ptr += SBLIMIT;

  2100.         }

  2101.         ptr += 18;

  2102.         buf += 18;

  2103.     }

  2104.     /* zero bands */

  2105.     for(j=sblimit;j<SBLIMIT;j++) {

  2106.         /* overlap */

  2107.         out_ptr = sb_samples + j;

  2108.         for(i=0;i<18;i++) {

  2109.             *out_ptr = buf[i];

  2110.             buf[i] = 0;

  2111.             out_ptr += SBLIMIT;

  2112.         }

  2113.         buf += 18;

  2114.     }

  2115. }

  2116. 

  2117. #if defined(DEBUG)

  2118. void sample_dump(int fnum, int32_t *tab, int n)

  2119. {

  2120.     static FILE *files[16], *f;

  2121.     char buf[512];

  2122.     int i;

  2123.     int32_t v;

  2124.     

  2125.     f = files[fnum];

  2126.     if (!f) {

  2127.         sprintf(buf, "/tmp/out%d.%s.pcm", 

  2128.                 fnum, 

  2129. #ifdef USE_HIGHPRECISION

  2130.                 "hp"

  2131. #else

  2132.                 "lp"

  2133. #endif

  2134.                 );

  2135.         f = fopen(buf, "w");

  2136.         if (!f)

  2137.             return;

  2138.         files[fnum] = f;

  2139.     }

  2140.     

  2141.     if (fnum == 0) {

  2142.         static int pos = 0;

  2143.         printf("pos=%d\n", pos);

  2144.         for(i=0;i<n;i++) {

  2145.             printf(" %0.4f", (double)tab[i] / FRAC_ONE);

  2146.             if ((i % 18) == 17)

  2147.                 printf("\n");

  2148.         }

  2149.         pos += n;

  2150.     }

  2151.     for(i=0;i<n;i++) {

  2152.         /* normalize to 23 frac bits */

  2153.         v = tab[i] << (23 - FRAC_BITS);

  2154.         fwrite(&v, 1, sizeof(int32_t), f);

  2155.     }

  2156. }

  2157. #endif

  2158. 

  2159. 

  2160. /* main layer3 decoding function */

  2161. static int mp_decode_layer3(MPADecodeContext *s)

  2162. {

  2163.     int nb_granules, main_data_begin, private_bits;

  2164.     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos, bits_left;

  2165.     GranuleDef granules[2][2], *g;

  2166.     int16_t exponents[576];

  2167. 

  2168.     /* read side info */

  2169.     if (s->lsf) {

  2170.         main_data_begin = get_bits(&s->gb, 8);

  2171.         if (s->nb_channels == 2)

  2172.             private_bits = get_bits(&s->gb, 2);

  2173.         else

  2174.             private_bits = get_bits(&s->gb, 1);

  2175.         nb_granules = 1;

  2176.     } else {

  2177.         main_data_begin = get_bits(&s->gb, 9);

  2178.         if (s->nb_channels == 2)

  2179.             private_bits = get_bits(&s->gb, 3);

  2180.         else

  2181.             private_bits = get_bits(&s->gb, 5);

  2182.         nb_granules = 2;

  2183.         for(ch=0;ch<s->nb_channels;ch++) {

  2184.             granules[ch][0].scfsi = 0; /* all scale factors are transmitted */

  2185.             granules[ch][1].scfsi = get_bits(&s->gb, 4);

  2186.         }

  2187.     }

  2188.     

  2189.     for(gr=0;gr<nb_granules;gr++) {

  2190.         for(ch=0;ch<s->nb_channels;ch++) {

  2191.             dprintf("gr=%d ch=%d: side_info\n", gr, ch);

  2192.             g = &granules[ch][gr];

  2193.             g->part2_3_length = get_bits(&s->gb, 12);

  2194.             g->big_values = get_bits(&s->gb, 9);

  2195.             g->global_gain = get_bits(&s->gb, 8);

  2196.             /* if MS stereo only is selected, we precompute the

  2197.                1/sqrt(2) renormalization factor */

  2198.             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) == 

  2199.                 MODE_EXT_MS_STEREO)

  2200.                 g->global_gain -= 2;

  2201.             if (s->lsf)

  2202.                 g->scalefac_compress = get_bits(&s->gb, 9);

  2203.             else

  2204.                 g->scalefac_compress = get_bits(&s->gb, 4);

  2205.             blocksplit_flag = get_bits(&s->gb, 1);

  2206.             if (blocksplit_flag) {

  2207.                 g->block_type = get_bits(&s->gb, 2);

  2208.                 if (g->block_type == 0)

  2209.                     return -1;

  2210.                 g->switch_point = get_bits(&s->gb, 1);

  2211.                 for(i=0;i<2;i++)

  2212.                     g->table_select[i] = get_bits(&s->gb, 5);

  2213.                 for(i=0;i<3;i++) 

  2214.                     g->subblock_gain[i] = get_bits(&s->gb, 3);

  2215.                 /* compute huffman coded region sizes */

  2216.                 if (g->block_type == 2)

  2217.                     g->region_size[0] = (36 / 2);

  2218.                 else {

  2219.                     if (s->sample_rate_index <= 2) 

  2220.                         g->region_size[0] = (36 / 2);

  2221.                     else if (s->sample_rate_index != 8) 

  2222.                         g->region_size[0] = (54 / 2);

  2223.                     else

  2224.                         g->region_size[0] = (108 / 2);

  2225.                 }

  2226.                 g->region_size[1] = (576 / 2);

  2227.             } else {

  2228.                 int region_address1, region_address2, l;

  2229.                 g->block_type = 0;

  2230.                 g->switch_point = 0;

  2231.                 for(i=0;i<3;i++)

  2232.                     g->table_select[i] = get_bits(&s->gb, 5);

  2233.                 /* compute huffman coded region sizes */

  2234.                 region_address1 = get_bits(&s->gb, 4);

  2235.                 region_address2 = get_bits(&s->gb, 3);

  2236.                 dprintf("region1=%d region2=%d\n", 

  2237.                         region_address1, region_address2);

  2238.                 g->region_size[0] = 

  2239.                     band_index_long[s->sample_rate_index][region_address1 + 1] >> 1;

  2240.                 l = region_address1 + region_address2 + 2;

  2241.                 /* should not overflow */

  2242.                 if (l > 22)

  2243.                     l = 22;

  2244.                 g->region_size[1] = 

  2245.                     band_index_long[s->sample_rate_index][l] >> 1;

  2246.             }

  2247.             /* convert region offsets to region sizes and truncate

  2248.                size to big_values */

  2249.             g->region_size[2] = (576 / 2);

  2250.             j = 0;

  2251.             for(i=0;i<3;i++) {

  2252.                 k = g->region_size[i];

  2253.                 if (k > g->big_values)

  2254.                     k = g->big_values;

  2255.                 g->region_size[i] = k - j;

  2256.                 j = k;

  2257.             }

  2258. 

  2259.             /* compute band indexes */

  2260.             if (g->block_type == 2) {

  2261.                 if (g->switch_point) {

  2262.                     /* if switched mode, we handle the 36 first samples as

  2263.                        long blocks.  For 8000Hz, we handle the 48 first

  2264.                        exponents as long blocks (XXX: check this!) */

  2265.                     if (s->sample_rate_index <= 2)

  2266.                         g->long_end = 8;

  2267.                     else if (s->sample_rate_index != 8)

  2268.                         g->long_end = 6;

  2269.                     else

  2270.                         g->long_end = 4; /* 8000 Hz */

  2271.                     

  2272.                     if (s->sample_rate_index != 8)

  2273.                         g->short_start = 3;

  2274.                     else

  2275.                         g->short_start = 2; 

  2276.                 } else {

  2277.                     g->long_end = 0;

  2278.                     g->short_start = 0;

  2279.                 }

  2280.             } else {

  2281.                 g->short_start = 13;

  2282.                 g->long_end = 22;

  2283.             }

  2284.             

  2285.             g->preflag = 0;

  2286.             if (!s->lsf)

  2287.                 g->preflag = get_bits(&s->gb, 1);

  2288.             g->scalefac_scale = get_bits(&s->gb, 1);

  2289.             g->count1table_select = get_bits(&s->gb, 1);

  2290.             dprintf("block_type=%d switch_point=%d\n",

  2291.                     g->block_type, g->switch_point);

  2292.         }

  2293.     }

  2294. 

  2295.     /* now we get bits from the main_data_begin offset */

  2296.     dprintf("seekback: %d\n", main_data_begin);

  2297.     seek_to_maindata(s, main_data_begin);

  2298. 

  2299.     for(gr=0;gr<nb_granules;gr++) {

  2300.         for(ch=0;ch<s->nb_channels;ch++) {

  2301.             g = &granules[ch][gr];

  2302.             

  2303.             bits_pos = get_bits_count(&s->gb);

  2304.             

  2305.             if (!s->lsf) {

  2306.                 uint8_t *sc;

  2307.                 int slen, slen1, slen2;

  2308. 

  2309.                 /* MPEG1 scale factors */

  2310.                 slen1 = slen_table[0][g->scalefac_compress];

  2311.                 slen2 = slen_table[1][g->scalefac_compress];

  2312.                 dprintf("slen1=%d slen2=%d\n", slen1, slen2);

  2313.                 if (g->block_type == 2) {

  2314.                     n = g->switch_point ? 17 : 18;

  2315.                     j = 0;

  2316.                     for(i=0;i<n;i++)

  2317.                         g->scale_factors[j++] = get_bitsz(&s->gb, slen1);

  2318.                     for(i=0;i<18;i++)

  2319.                         g->scale_factors[j++] = get_bitsz(&s->gb, slen2);

  2320.                     for(i=0;i<3;i++)

  2321.                         g->scale_factors[j++] = 0;

  2322.                 } else {

  2323.                     sc = granules[ch][0].scale_factors;

  2324.                     j = 0;

  2325.                     for(k=0;k<4;k++) {

  2326.                         n = (k == 0 ? 6 : 5);

  2327.                         if ((g->scfsi & (0x8 >> k)) == 0) {

  2328.                             slen = (k < 2) ? slen1 : slen2;

  2329.                             for(i=0;i<n;i++)

  2330.                                 g->scale_factors[j++] = get_bitsz(&s->gb, slen);

  2331.                         } else {

  2332.                             /* simply copy from last granule */

  2333.                             for(i=0;i<n;i++) {

  2334.                                 g->scale_factors[j] = sc[j];

  2335.                                 j++;

  2336.                             }

  2337.                         }

  2338.                     }

  2339.                     g->scale_factors[j++] = 0;

  2340.                 }

  2341. #if defined(DEBUG)

  2342.                 {

  2343.                     printf("scfsi=%x gr=%d ch=%d scale_factors:\n", 

  2344.                            g->scfsi, gr, ch);

  2345.                     for(i=0;i<j;i++)

  2346.                         printf(" %d", g->scale_factors[i]);

  2347.                     printf("\n");

  2348.                 }

  2349. #endif

  2350.             } else {

  2351.                 int tindex, tindex2, slen[4], sl, sf;

  2352. 

  2353.                 /* LSF scale factors */

  2354.                 if (g->block_type == 2) {

  2355.                     tindex = g->switch_point ? 2 : 1;

  2356.                 } else {

  2357.                     tindex = 0;

  2358.                 }

  2359.                 sf = g->scalefac_compress;

  2360.                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {

  2361.                     /* intensity stereo case */

  2362.                     sf >>= 1;

  2363.                     if (sf < 180) {

  2364.                         lsf_sf_expand(slen, sf, 6, 6, 0);

  2365.                         tindex2 = 3;

  2366.                     } else if (sf < 244) {

  2367.                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);

  2368.                         tindex2 = 4;

  2369.                     } else {

  2370.                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);

  2371.                         tindex2 = 5;

  2372.                     }

  2373.                 } else {

  2374.                     /* normal case */

  2375.                     if (sf < 400) {

  2376.                         lsf_sf_expand(slen, sf, 5, 4, 4);

  2377.                         tindex2 = 0;

  2378.                     } else if (sf < 500) {

  2379.                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);

  2380.                         tindex2 = 1;

  2381.                     } else {

  2382.                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);

  2383.                         tindex2 = 2;

  2384.                         g->preflag = 1;

  2385.                     }

  2386.                 }

  2387. 

  2388.                 j = 0;

  2389.                 for(k=0;k<4;k++) {

  2390.                     n = lsf_nsf_table[tindex2][tindex][k];

  2391.                     sl = slen[k];

  2392.                     for(i=0;i<n;i++)

  2393.                         g->scale_factors[j++] = get_bitsz(&s->gb, sl);

  2394.                 }

  2395.                 /* XXX: should compute exact size */

  2396.                 for(;j<40;j++)

  2397.                     g->scale_factors[j] = 0;

  2398. #if defined(DEBUG)

  2399.                 {

  2400.                     printf("gr=%d ch=%d scale_factors:\n", 

  2401.                            gr, ch);

  2402.                     for(i=0;i<40;i++)

  2403.                         printf(" %d", g->scale_factors[i]);

  2404.                     printf("\n");

  2405.                 }

  2406. #endif

  2407.             }

  2408. 

  2409.             exponents_from_scale_factors(s, g, exponents);

  2410. 

  2411.             /* read Huffman coded residue */

  2412.             if (huffman_decode(s, g, exponents,

  2413.                                bits_pos + g->part2_3_length) < 0)

  2414.                 return -1;

  2415. #if defined(DEBUG)

  2416.             sample_dump(0, g->sb_hybrid, 576);

  2417. #endif

  2418. 

  2419.             /* skip extension bits */

  2420.             bits_left = g->part2_3_length - (get_bits_count(&s->gb) - bits_pos);

  2421.             if (bits_left < 0) {

  2422.                 dprintf("bits_left=%d\n", bits_left);

  2423.                 return -1;

  2424.             }

  2425.             while (bits_left >= 16) {

  2426.                 skip_bits(&s->gb, 16);

  2427.                 bits_left -= 16;

  2428.             }

  2429.             if (bits_left > 0)

  2430.                 skip_bits(&s->gb, bits_left);

  2431.         } /* ch */

  2432. 

  2433.         if (s->nb_channels == 2)

  2434.             compute_stereo(s, &granules[0][gr], &granules[1][gr]);

  2435. 

  2436.         for(ch=0;ch<s->nb_channels;ch++) {

  2437.             g = &granules[ch][gr];

  2438. 

  2439.             reorder_block(s, g);

  2440. #if defined(DEBUG)

  2441.             sample_dump(0, g->sb_hybrid, 576);

  2442. #endif

  2443.             s->compute_antialias(s, g);

  2444. #if defined(DEBUG)

  2445.             sample_dump(1, g->sb_hybrid, 576);

  2446. #endif

  2447.             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]); 

  2448. #if defined(DEBUG)

  2449.             sample_dump(2, &s->sb_samples[ch][18 * gr][0], 576);

  2450. #endif

  2451.         }

  2452.     } /* gr */

  2453.     return nb_granules * 18;

  2454. }

  2455. 

  2456. static int mp_decode_frame(MPADecodeContext *s, 

  2457.                            short *samples)

  2458. {

  2459.     int i, nb_frames, ch;

  2460.     short *samples_ptr;

  2461. 

  2462.     init_get_bits(&s->gb, s->inbuf + HEADER_SIZE, 

  2463.                   (s->inbuf_ptr - s->inbuf - HEADER_SIZE)*8);

  2464.     

  2465.     /* skip error protection field */

  2466.     if (s->error_protection)

  2467.         get_bits(&s->gb, 16);

  2468. 

  2469.     dprintf("frame %d:\n", s->frame_count);

  2470.     switch(s->layer) {

  2471.     case 1:

  2472.         nb_frames = mp_decode_layer1(s);

  2473.         break;

  2474.     case 2:

  2475.         nb_frames = mp_decode_layer2(s);

  2476.         break;

  2477.     case 3:

  2478.     default:

  2479.         nb_frames = mp_decode_layer3(s);

  2480.         break;

  2481.     }

  2482. #if defined(DEBUG)

  2483.     for(i=0;i<nb_frames;i++) {

  2484.         for(ch=0;ch<s->nb_channels;ch++) {

  2485.             int j;

  2486.             printf("%d-%d:", i, ch);

  2487.             for(j=0;j<SBLIMIT;j++)

  2488.                 printf(" %0.6f", (double)s->sb_samples[ch][i][j] / FRAC_ONE);

  2489.             printf("\n");

  2490.         }

  2491.     }

  2492. #endif

  2493.     /* apply the synthesis filter */

  2494.     for(ch=0;ch<s->nb_channels;ch++) {

  2495.         samples_ptr = samples + ch;

  2496.         for(i=0;i<nb_frames;i++) {

  2497.             synth_filter(s, ch, samples_ptr, s->nb_channels,

  2498.                          s->sb_samples[ch][i]);

  2499.             samples_ptr += 32 * s->nb_channels;

  2500.         }

  2501.     }

  2502. #ifdef DEBUG

  2503.     s->frame_count++;        

  2504. #endif

  2505.     return nb_frames * 32 * sizeof(short) * s->nb_channels;

  2506. }

  2507. 

  2508. static int decode_frame(AVCodecContext * avctx,

  2509. 			void *data, int *data_size,

  2510. 			uint8_t * buf, int buf_size)

  2511. {

  2512.     MPADecodeContext *s = avctx->priv_data;

  2513.     uint32_t header;

  2514.     uint8_t *buf_ptr;

  2515.     int len, out_size;

  2516.     short *out_samples = data;

  2517. 

  2518.     *data_size = 0;

  2519.     buf_ptr = buf;

  2520.     while (buf_size > 0) {

  2521. 	len = s->inbuf_ptr - s->inbuf;

  2522. 	if (s->frame_size == 0) {

  2523.             /* special case for next header for first frame in free

  2524.                format case (XXX: find a simpler method) */

  2525.             if (s->free_format_next_header != 0) {

  2526.                 s->inbuf[0] = s->free_format_next_header >> 24;

  2527.                 s->inbuf[1] = s->free_format_next_header >> 16;

  2528.                 s->inbuf[2] = s->free_format_next_header >> 8;

  2529.                 s->inbuf[3] = s->free_format_next_header;

  2530.                 s->inbuf_ptr = s->inbuf + 4;

  2531.                 s->free_format_next_header = 0;

  2532.                 goto got_header;

  2533.             }

  2534. 	    /* no header seen : find one. We need at least HEADER_SIZE

  2535.                bytes to parse it */

  2536. 	    len = HEADER_SIZE - len;

  2537. 	    if (len > buf_size)

  2538. 		len = buf_size;

  2539. 	    if (len > 0) {

  2540. 		memcpy(s->inbuf_ptr, buf_ptr, len);

  2541. 		buf_ptr += len;

  2542. 		buf_size -= len;

  2543. 		s->inbuf_ptr += len;

  2544. 	    }

  2545. 	    if ((s->inbuf_ptr - s->inbuf) >= HEADER_SIZE) {

  2546.             got_header:

  2547. 		header = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |

  2548. 		    (s->inbuf[2] << 8) | s->inbuf[3];

  2549. 

  2550. 		if (check_header(header) < 0) {

  2551. 		    /* no sync found : move by one byte (inefficient, but simple!) */

  2552. 		    memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);

  2553. 		    s->inbuf_ptr--;

  2554.                     dprintf("skip %x\n", header);

  2555.                     /* reset free format frame size to give a chance

  2556.                        to get a new bitrate */

  2557.                     s->free_format_frame_size = 0;

  2558. 		} else {

  2559. 		    if (decode_header(s, header) == 1) {

  2560.                         /* free format: prepare to compute frame size */

  2561. 			s->frame_size = -1;

  2562.                     }

  2563.                     /* update codec info */

  2564.                     avctx->sample_rate = s->sample_rate;

  2565.                     avctx->channels = s->nb_channels;

  2566.                     avctx->bit_rate = s->bit_rate;

  2567.                     avctx->sub_id = s->layer;

  2568.                     switch(s->layer) {

  2569.                     case 1:

  2570.                         avctx->frame_size = 384;

  2571.                         break;

  2572.                     case 2:

  2573.                         avctx->frame_size = 1152;

  2574.                         break;

  2575.                     case 3:

  2576.                         if (s->lsf)

  2577.                             avctx->frame_size = 576;

  2578.                         else

  2579.                             avctx->frame_size = 1152;

  2580.                         break;

  2581.                     }

  2582. 		}

  2583. 	    }

  2584.         } else if (s->frame_size == -1) {

  2585.             /* free format : find next sync to compute frame size */

  2586. 	    len = MPA_MAX_CODED_FRAME_SIZE - len;

  2587. 	    if (len > buf_size)

  2588. 		len = buf_size;

  2589.             if (len == 0) {

  2590. 		/* frame too long: resync */

  2591.                 s->frame_size = 0;

  2592. 		memmove(s->inbuf, s->inbuf + 1, s->inbuf_ptr - s->inbuf - 1);

  2593. 		s->inbuf_ptr--;

  2594.             } else {

  2595.                 uint8_t *p, *pend;

  2596.                 uint32_t header1;

  2597.                 int padding;

  2598. 

  2599.                 memcpy(s->inbuf_ptr, buf_ptr, len);

  2600.                 /* check for header */

  2601.                 p = s->inbuf_ptr - 3;

  2602.                 pend = s->inbuf_ptr + len - 4;

  2603.                 while (p <= pend) {

  2604.                     header = (p[0] << 24) | (p[1] << 16) |

  2605.                         (p[2] << 8) | p[3];

  2606.                     header1 = (s->inbuf[0] << 24) | (s->inbuf[1] << 16) |

  2607.                         (s->inbuf[2] << 8) | s->inbuf[3];

  2608.                     /* check with high probability that we have a

  2609.                        valid header */

  2610.                     if ((header & SAME_HEADER_MASK) ==

  2611.                         (header1 & SAME_HEADER_MASK)) {

  2612.                         /* header found: update pointers */

  2613.                         len = (p + 4) - s->inbuf_ptr;

  2614.                         buf_ptr += len;

  2615.                         buf_size -= len;

  2616.                         s->inbuf_ptr = p;

  2617.                         /* compute frame size */

  2618.                         s->free_format_next_header = header;

  2619.                         s->free_format_frame_size = s->inbuf_ptr - s->inbuf;

  2620.                         padding = (header1 >> 9) & 1;

  2621.                         if (s->layer == 1)

  2622.                             s->free_format_frame_size -= padding * 4;

  2623.                         else

  2624.                             s->free_format_frame_size -= padding;

  2625.                         dprintf("free frame size=%d padding=%d\n", 

  2626.                                 s->free_format_frame_size, padding);

  2627.                         decode_header(s, header1);

  2628.                         goto next_data;

  2629.                     }

  2630.                     p++;

  2631.                 }

  2632.                 /* not found: simply increase pointers */

  2633.                 buf_ptr += len;

  2634.                 s->inbuf_ptr += len;

  2635.                 buf_size -= len;

  2636.             }

  2637. 	} else if (len < s->frame_size) {

  2638.             if (s->frame_size > MPA_MAX_CODED_FRAME_SIZE)

  2639.                 s->frame_size = MPA_MAX_CODED_FRAME_SIZE;

  2640. 	    len = s->frame_size - len;

  2641. 	    if (len > buf_size)

  2642. 		len = buf_size;

  2643. 	    memcpy(s->inbuf_ptr, buf_ptr, len);

  2644. 	    buf_ptr += len;

  2645. 	    s->inbuf_ptr += len;

  2646. 	    buf_size -= len;

  2647. 	}

  2648.     next_data:

  2649.         if (s->frame_size > 0 && 

  2650.             (s->inbuf_ptr - s->inbuf) >= s->frame_size) {

  2651.             if (avctx->parse_only) {

  2652.                 /* simply return the frame data */

  2653.                 *(uint8_t **)data = s->inbuf;

  2654.                 out_size = s->inbuf_ptr - s->inbuf;

  2655.             } else {

  2656.                 out_size = mp_decode_frame(s, out_samples);

  2657.             }

  2658. 	    s->inbuf_ptr = s->inbuf;

  2659. 	    s->frame_size = 0;

  2660. 	    *data_size = out_size;

  2661. 	    break;

  2662. 	}

  2663.     }

  2664.     return buf_ptr - buf;

  2665. }

  2666. 

  2667. AVCodec mp2_decoder =

  2668. {

  2669.     "mp2",

  2670.     CODEC_TYPE_AUDIO,

  2671.     CODEC_ID_MP2,

  2672.     sizeof(MPADecodeContext),

  2673.     decode_init,

  2674.     NULL,

  2675.     NULL,

  2676.     decode_frame,

  2677.     CODEC_CAP_PARSE_ONLY,

  2678. };

  2679. 

  2680. AVCodec mp3_decoder =

  2681. {

  2682.     "mp3",

  2683.     CODEC_TYPE_AUDIO,

  2684.     CODEC_ID_MP3,

  2685.     sizeof(MPADecodeContext),

  2686.     decode_init,

  2687.     NULL,

  2688.     NULL,

  2689.     decode_frame,

  2690.     CODEC_CAP_PARSE_ONLY,

  2691. };