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.
24563 Commits
32 Branches
871MB
Tree: 31ed3f1f1d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '31ed3f1f1d'
${ noResults }
FFmpeg/libavcodec/mpegaudiodec.c

2620 lines
78KB
Raw Blame History 

  1. /*

  2.  * MPEG Audio decoder

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

  4.  *

  5.  * This file is part of FFmpeg.

  6.  *

  7.  * FFmpeg is free software; you can redistribute it and/or

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

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

  10.  * version 2.1 of the License, or (at your option) any later version.

  11.  *

  12.  * FFmpeg is distributed in the hope that it will be useful,

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

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

  15.  * Lesser General Public License for more details.

  16.  *

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

  18.  * License along with FFmpeg; if not, write to the Free Software

  19.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  20.  */

  21. 

  22. /**

  23.  * @file

  24.  * MPEG Audio decoder.

  25.  */

  26. 

  27. #include "avcodec.h"

  28. #include "get_bits.h"

  29. #include "dsputil.h"

  30. 

  31. /*

  32.  * TODO:

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

  34.  *  - test lsf / mpeg25 extensively.

  35.  */

  36. 

  37. #include "mpegaudio.h"

  38. #include "mpegaudiodecheader.h"

  39. 

  40. #include "mathops.h"

  41. 

  42. #if CONFIG_FLOAT

  43. #   define SHR(a,b)       ((a)*(1.0f/(1<<(b))))

  44. #   define compute_antialias compute_antialias_float

  45. #   define FIXR_OLD(a)    ((int)((a) * FRAC_ONE + 0.5))

  46. #   define FIXR(x)        ((float)(x))

  47. #   define FIXHR(x)       ((float)(x))

  48. #   define MULH3(x, y, s) ((s)*(y)*(x))

  49. #   define MULLx(x, y, s) ((y)*(x))

  50. #   define RENAME(a) a ## _float

  51. #else

  52. #   define SHR(a,b)       ((a)>>(b))

  53. #   define compute_antialias compute_antialias_integer

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

  55. #   define FIXR_OLD(a)    ((int)((a) * FRAC_ONE + 0.5))

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

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

  58. #   define MULH3(x, y, s) MULH((s)*(x), y)

  59. #   define MULLx(x, y, s) MULL(x,y,s)

  60. #   define RENAME(a)      a

  61. #endif

  62. 

  63. /****************/

  64. 

  65. #define HEADER_SIZE 4

  66. 

  67. #include "mpegaudiodata.h"

  68. #include "mpegaudiodectab.h"

  69. 

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

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

  72. static void apply_window_mp3_c(MPA_INT *synth_buf, MPA_INT *window,

  73.                                int *dither_state, OUT_INT *samples, int incr);

  74. 

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

  76. static VLC huff_vlc[16];

  77. static VLC_TYPE huff_vlc_tables[

  78.   0+128+128+128+130+128+154+166+

  79.   142+204+190+170+542+460+662+414

  80.   ][2];

  81. static const int huff_vlc_tables_sizes[16] = {

  82.   0, 128, 128, 128, 130, 128, 154, 166,

  83.   142, 204, 190, 170, 542, 460, 662, 414

  84. };

  85. static VLC huff_quad_vlc[2];

  86. static VLC_TYPE huff_quad_vlc_tables[128+16][2];

  87. static const int huff_quad_vlc_tables_sizes[2] = {

  88.   128, 16

  89. };

  90. /* computed from band_size_long */

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

  92. #include "mpegaudio_tablegen.h"

  93. /* intensity stereo coef table */

  94. static INTFLOAT is_table[2][16];

  95. static INTFLOAT is_table_lsf[2][2][16];

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

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

  98. static INTFLOAT mdct_win[8][36];

  99. 

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

  101. static uint16_t scale_factor_modshift[64];

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

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

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

  105. 

  106. #define SCALE_GEN(v) \

  107. { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }

  108. 

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

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

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

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

  113. };

  114. 

  115. DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512];

  116. 

  117. /**

  118.  * Convert region offsets to region sizes and truncate

  119.  * size to big_values.

  120.  */

  121. static void ff_region_offset2size(GranuleDef *g){

  122.     int i, k, j=0;

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

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

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

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

  127.         j = k;

  128.     }

  129. }

  130. 

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

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

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

  134.     else {

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

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

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

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

  139.         else

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

  141.     }

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

  143. }

  144. 

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

  146.     int l;

  147.     g->region_size[0] =

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

  149.     /* should not overflow */

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

  151.     g->region_size[1] =

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

  153. }

  154. 

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

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

  157.         if (g->switch_point) {

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

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

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

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

  162.                 g->long_end = 8;

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

  164.                 g->long_end = 6;

  165.             else

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

  167. 

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

  169.         } else {

  170.             g->long_end = 0;

  171.             g->short_start = 0;

  172.         }

  173.     } else {

  174.         g->short_start = 13;

  175.         g->long_end = 22;

  176.     }

  177. }

  178. 

  179. /* layer 1 unscaling */

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

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

  182. {

  183.     int shift, mod;

  184.     int64_t val;

  185. 

  186.     shift = scale_factor_modshift[scale_factor];

  187.     mod = shift & 3;

  188.     shift >>= 2;

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

  190.     shift += n;

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

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

  193. }

  194. 

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

  196. {

  197.     int shift, mod, val;

  198. 

  199.     shift = scale_factor_modshift[scale_factor];

  200.     mod = shift & 3;

  201.     shift >>= 2;

  202. 

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

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

  205.     if (shift > 0)

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

  207.     return val;

  208. }

  209. 

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

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

  212. {

  213.     unsigned int m;

  214.     int e;

  215. 

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

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

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

  219.     assert(e>=1);

  220.     if (e > 31)

  221.         return 0;

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

  223. 

  224.     return m;

  225. }

  226. 

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

  228. #define DEV_ORDER 13

  229. 

  230. #define POW_FRAC_BITS 24

  231. #define POW_FRAC_ONE    (1 << POW_FRAC_BITS)

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

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

  234. 

  235. static int dev_4_3_coefs[DEV_ORDER];

  236. 

  237. #if 0 /* unused */

  238. static int pow_mult3[3] = {

  239.     POW_FIX(1.0),

  240.     POW_FIX(1.25992104989487316476),

  241.     POW_FIX(1.58740105196819947474),

  242. };

  243. #endif

  244. 

  245. static av_cold void int_pow_init(void)

  246. {

  247.     int i, a;

  248. 

  249.     a = POW_FIX(1.0);

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

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

  252.         dev_4_3_coefs[i] = a;

  253.     }

  254. }

  255. 

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

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

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

  259. {

  260.     int e, er, eq, j;

  261.     int a, a1;

  262. 

  263.     /* renormalize */

  264.     a = i;

  265.     e = POW_FRAC_BITS;

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

  267.         a = a << 1;

  268.         e--;

  269.     }

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

  271.     a1 = 0;

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

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

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

  275.     /* exponent compute (exact) */

  276.     e = e * 4;

  277.     er = e % 3;

  278.     eq = e / 3;

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

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

  281.         a = a >> 1;

  282.         eq++;

  283.     }

  284.     /* convert to float */

  285.     while (a < POW_FRAC_ONE) {

  286.         a = a << 1;

  287.         eq--;

  288.     }

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

  290. #if POW_FRAC_BITS > FRAC_BITS

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

  292.     /* correct overflow */

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

  294.         a = a >> 1;

  295.         eq++;

  296.     }

  297. #endif

  298.     *exp_ptr = eq;

  299.     return a;

  300. }

  301. #endif

  302. 

  303. static av_cold int decode_init(AVCodecContext * avctx)

  304. {

  305.     MPADecodeContext *s = avctx->priv_data;

  306.     static int init=0;

  307.     int i, j, k;

  308. 

  309.     s->avctx = avctx;

  310.     s->apply_window_mp3 = apply_window_mp3_c;

  311. 

  312.     avctx->sample_fmt= OUT_FMT;

  313.     s->error_recognition= avctx->error_recognition;

  314. 

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

  316.         int offset;

  317. 

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

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

  320.             int shift, mod;

  321.             /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

  322.             shift = (i / 3);

  323.             mod = i % 3;

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

  325.         }

  326. 

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

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

  329.             int n, norm;

  330.             n = i + 2;

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

  332.             scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0          * 2.0), FRAC_BITS);

  333.             scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);

  334.             scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);

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

  336.                     i, norm,

  337.                     scale_factor_mult[i][0],

  338.                     scale_factor_mult[i][1],

  339.                     scale_factor_mult[i][2]);

  340.         }

  341. 

  342.         RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));

  343. 

  344.         /* huffman decode tables */

  345.         offset = 0;

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

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

  348.             int xsize, x, y;

  349.             uint8_t  tmp_bits [512];

  350.             uint16_t tmp_codes[512];

  351. 

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

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

  354. 

  355.             xsize = h->xsize;

  356. 

  357.             j = 0;

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

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

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

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

  362.                 }

  363.             }

  364. 

  365.             /* XXX: fail test */

  366.             huff_vlc[i].table = huff_vlc_tables+offset;

  367.             huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];

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

  369.                      tmp_bits, 1, 1, tmp_codes, 2, 2,

  370.                      INIT_VLC_USE_NEW_STATIC);

  371.             offset += huff_vlc_tables_sizes[i];

  372.         }

  373.         assert(offset == FF_ARRAY_ELEMS(huff_vlc_tables));

  374. 

  375.         offset = 0;

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

  377.             huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;

  378.             huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];

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

  380.                      mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,

  381.                      INIT_VLC_USE_NEW_STATIC);

  382.             offset += huff_quad_vlc_tables_sizes[i];

  383.         }

  384.         assert(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));

  385. 

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

  387.             k = 0;

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

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

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

  391.             }

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

  393.         }

  394. 

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

  396. 

  397.         int_pow_init();

  398.         mpegaudio_tableinit();

  399. 

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

  401.             float f;

  402.             INTFLOAT v;

  403.             if (i != 6) {

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

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

  406.             } else {

  407.                 v = FIXR(1.0);

  408.             }

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

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

  411.         }

  412.         /* invalid values */

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

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

  415. 

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

  417.             double f;

  418.             int e, k;

  419. 

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

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

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

  423.                 k = i & 1;

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

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

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

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

  428.             }

  429.         }

  430. 

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

  432.             float ci, cs, ca;

  433.             ci = ci_table[i];

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

  435.             ca = cs * ci;

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

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

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

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

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

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

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

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

  444.         }

  445. 

  446.         /* compute mdct windows */

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

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

  449.                 double d;

  450. 

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

  452.                     continue;

  453. 

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

  455.                 if(j==1){

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

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

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

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

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

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

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

  463.                 }

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

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

  466. 

  467.                 if(j==2)

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

  469.                 else

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

  471.             }

  472.         }

  473. 

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

  475.            the sign of the right window coefs */

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

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

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

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

  480.             }

  481.         }

  482. 

  483.         init = 1;

  484.     }

  485. 

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

  487.         s->adu_mode = 1;

  488.     return 0;

  489. }

  490. 

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

  492. 

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

  494. 

  495. #define COS0_0  FIXHR(0.50060299823519630134/2)

  496. #define COS0_1  FIXHR(0.50547095989754365998/2)

  497. #define COS0_2  FIXHR(0.51544730992262454697/2)

  498. #define COS0_3  FIXHR(0.53104259108978417447/2)

  499. #define COS0_4  FIXHR(0.55310389603444452782/2)

  500. #define COS0_5  FIXHR(0.58293496820613387367/2)

  501. #define COS0_6  FIXHR(0.62250412303566481615/2)

  502. #define COS0_7  FIXHR(0.67480834145500574602/2)

  503. #define COS0_8  FIXHR(0.74453627100229844977/2)

  504. #define COS0_9  FIXHR(0.83934964541552703873/2)

  505. #define COS0_10 FIXHR(0.97256823786196069369/2)

  506. #define COS0_11 FIXHR(1.16943993343288495515/4)

  507. #define COS0_12 FIXHR(1.48416461631416627724/4)

  508. #define COS0_13 FIXHR(2.05778100995341155085/8)

  509. #define COS0_14 FIXHR(3.40760841846871878570/8)

  510. #define COS0_15 FIXHR(10.19000812354805681150/32)

  511. 

  512. #define COS1_0 FIXHR(0.50241928618815570551/2)

  513. #define COS1_1 FIXHR(0.52249861493968888062/2)

  514. #define COS1_2 FIXHR(0.56694403481635770368/2)

  515. #define COS1_3 FIXHR(0.64682178335999012954/2)

  516. #define COS1_4 FIXHR(0.78815462345125022473/2)

  517. #define COS1_5 FIXHR(1.06067768599034747134/4)

  518. #define COS1_6 FIXHR(1.72244709823833392782/4)

  519. #define COS1_7 FIXHR(5.10114861868916385802/16)

  520. 

  521. #define COS2_0 FIXHR(0.50979557910415916894/2)

  522. #define COS2_1 FIXHR(0.60134488693504528054/2)

  523. #define COS2_2 FIXHR(0.89997622313641570463/2)

  524. #define COS2_3 FIXHR(2.56291544774150617881/8)

  525. 

  526. #define COS3_0 FIXHR(0.54119610014619698439/2)

  527. #define COS3_1 FIXHR(1.30656296487637652785/4)

  528. 

  529. #define COS4_0 FIXHR(0.70710678118654752439/2)

  530. 

  531. /* butterfly operator */

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

  533. {\

  534.     tmp0 = val##a + val##b;\

  535.     tmp1 = val##a - val##b;\

  536.     val##a = tmp0;\

  537.     val##b = MULH3(tmp1, c, 1<<(s));\

  538. }

  539. 

  540. #define BF0(a, b, c, s)\

  541. {\

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

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

  544.     val##a = tmp0;\

  545.     val##b = MULH3(tmp1, c, 1<<(s));\

  546. }

  547. 

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

  549. {\

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

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

  552.     val##c += val##d;\

  553. }

  554. 

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

  556. {\

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

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

  559.     val##c += val##d;\

  560.     val##a += val##c;\

  561.     val##c += val##b;\

  562.     val##b += val##d;\

  563. }

  564. 

  565. #define ADD(a, b) val##a += val##b

  566. 

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

  568. static void dct32(INTFLOAT *out, const INTFLOAT *tab)

  569. {

  570.     INTFLOAT tmp0, tmp1;

  571. 

  572.     INTFLOAT val0 , val1 , val2 , val3 , val4 , val5 , val6 , val7 ,

  573.              val8 , val9 , val10, val11, val12, val13, val14, val15,

  574.              val16, val17, val18, val19, val20, val21, val22, val23,

  575.              val24, val25, val26, val27, val28, val29, val30, val31;

  576. 

  577.     /* pass 1 */

  578.     BF0( 0, 31, COS0_0 , 1);

  579.     BF0(15, 16, COS0_15, 5);

  580.     /* pass 2 */

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

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

  583.     /* pass 1 */

  584.     BF0( 7, 24, COS0_7 , 1);

  585.     BF0( 8, 23, COS0_8 , 1);

  586.     /* pass 2 */

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

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

  589.     /* pass 3 */

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

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

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

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

  594.     /* pass 1 */

  595.     BF0( 3, 28, COS0_3 , 1);

  596.     BF0(12, 19, COS0_12, 2);

  597.     /* pass 2 */

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

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

  600.     /* pass 1 */

  601.     BF0( 4, 27, COS0_4 , 1);

  602.     BF0(11, 20, COS0_11, 2);

  603.     /* pass 2 */

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

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

  606.     /* pass 3 */

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

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

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

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

  611.     /* pass 4 */

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

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

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

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

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

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

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

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

  620. 

  621. 

  622. 

  623.     /* pass 1 */

  624.     BF0( 1, 30, COS0_1 , 1);

  625.     BF0(14, 17, COS0_14, 3);

  626.     /* pass 2 */

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

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

  629.     /* pass 1 */

  630.     BF0( 6, 25, COS0_6 , 1);

  631.     BF0( 9, 22, COS0_9 , 1);

  632.     /* pass 2 */

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

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

  635.     /* pass 3 */

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

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

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

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

  640. 

  641.     /* pass 1 */

  642.     BF0( 2, 29, COS0_2 , 1);

  643.     BF0(13, 18, COS0_13, 3);

  644.     /* pass 2 */

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

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

  647.     /* pass 1 */

  648.     BF0( 5, 26, COS0_5 , 1);

  649.     BF0(10, 21, COS0_10, 1);

  650.     /* pass 2 */

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

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

  653.     /* pass 3 */

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

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

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

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

  658.     /* pass 4 */

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

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

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

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

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

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

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

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

  667. 

  668.     /* pass 5 */

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

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

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

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

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

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

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

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

  677. 

  678.     /* pass 6 */

  679. 

  680.     ADD( 8, 12);

  681.     ADD(12, 10);

  682.     ADD(10, 14);

  683.     ADD(14,  9);

  684.     ADD( 9, 13);

  685.     ADD(13, 11);

  686.     ADD(11, 15);

  687. 

  688.     out[ 0] = val0;

  689.     out[16] = val1;

  690.     out[ 8] = val2;

  691.     out[24] = val3;

  692.     out[ 4] = val4;

  693.     out[20] = val5;

  694.     out[12] = val6;

  695.     out[28] = val7;

  696.     out[ 2] = val8;

  697.     out[18] = val9;

  698.     out[10] = val10;

  699.     out[26] = val11;

  700.     out[ 6] = val12;

  701.     out[22] = val13;

  702.     out[14] = val14;

  703.     out[30] = val15;

  704. 

  705.     ADD(24, 28);

  706.     ADD(28, 26);

  707.     ADD(26, 30);

  708.     ADD(30, 25);

  709.     ADD(25, 29);

  710.     ADD(29, 27);

  711.     ADD(27, 31);

  712. 

  713.     out[ 1] = val16 + val24;

  714.     out[17] = val17 + val25;

  715.     out[ 9] = val18 + val26;

  716.     out[25] = val19 + val27;

  717.     out[ 5] = val20 + val28;

  718.     out[21] = val21 + val29;

  719.     out[13] = val22 + val30;

  720.     out[29] = val23 + val31;

  721.     out[ 3] = val24 + val20;

  722.     out[19] = val25 + val21;

  723.     out[11] = val26 + val22;

  724.     out[27] = val27 + val23;

  725.     out[ 7] = val28 + val18;

  726.     out[23] = val29 + val19;

  727.     out[15] = val30 + val17;

  728.     out[31] = val31;

  729. }

  730. 

  731. #if CONFIG_FLOAT

  732. static inline float round_sample(float *sum)

  733. {

  734.     float sum1=*sum;

  735.     *sum = 0;

  736.     return sum1;

  737. }

  738. 

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

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

  741. 

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

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

  744. 

  745. #define MLSS(rt, ra, rb) rt-=(ra)*(rb)

  746. 

  747. #elif FRAC_BITS <= 15

  748. 

  749. static inline int round_sample(int *sum)

  750. {

  751.     int sum1;

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

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

  754.     return av_clip(sum1, OUT_MIN, OUT_MAX);

  755. }

  756. 

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

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

  759. 

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

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

  762. 

  763. #define MLSS(rt, ra, rb) MLS16(rt, ra, rb)

  764. 

  765. #else

  766. 

  767. static inline int round_sample(int64_t *sum)

  768. {

  769.     int sum1;

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

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

  772.     return av_clip(sum1, OUT_MIN, OUT_MAX);

  773. }

  774. 

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

  776. #   define MACS(rt, ra, rb) MAC64(rt, ra, rb)

  777. #   define MLSS(rt, ra, rb) MLS64(rt, ra, rb)

  778. #endif

  779. 

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

  781. {                                         \

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

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

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

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

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

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

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

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

  790. }

  791. 

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

  793. {                                               \

  794.     INTFLOAT tmp;\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  819. }

  820. 

  821. void av_cold RENAME(ff_mpa_synth_init)(MPA_INT *window)

  822. {

  823.     int i;

  824. 

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

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

  827.         INTFLOAT v;

  828.         v = ff_mpa_enwindow[i];

  829. #if CONFIG_FLOAT

  830.         v *= 1.0 / (1LL<<(16 + FRAC_BITS));

  831. #elif WFRAC_BITS < 16

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

  833. #endif

  834.         window[i] = v;

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

  836.             v = -v;

  837.         if (i != 0)

  838.             window[512 - i] = v;

  839.     }

  840. }

  841. 

  842. static void apply_window_mp3_c(MPA_INT *synth_buf, MPA_INT *window,

  843.                                int *dither_state, OUT_INT *samples, int incr)

  844. {

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

  846.     int j;

  847.     OUT_INT *samples2;

  848. #if CONFIG_FLOAT

  849.     float sum, sum2;

  850. #elif FRAC_BITS <= 15

  851.     int sum, sum2;

  852. #else

  853.     int64_t sum, sum2;

  854. #endif

  855. 

  856.     /* copy to avoid wrap */

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

  858. 

  859.     samples2 = samples + 31 * incr;

  860.     w = window;

  861.     w2 = window + 31;

  862. 

  863.     sum = *dither_state;

  864.     p = synth_buf + 16;

  865.     SUM8(MACS, sum, w, p);

  866.     p = synth_buf + 48;

  867.     SUM8(MLSS, sum, w + 32, p);

  868.     *samples = round_sample(&sum);

  869.     samples += incr;

  870.     w++;

  871. 

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

  873.        access per two sample */

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

  875.         sum2 = 0;

  876.         p = synth_buf + 16 + j;

  877.         SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);

  878.         p = synth_buf + 48 - j;

  879.         SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);

  880. 

  881.         *samples = round_sample(&sum);

  882.         samples += incr;

  883.         sum += sum2;

  884.         *samples2 = round_sample(&sum);

  885.         samples2 -= incr;

  886.         w++;

  887.         w2--;

  888.     }

  889. 

  890.     p = synth_buf + 32;

  891.     SUM8(MLSS, sum, w + 32, p);

  892.     *samples = round_sample(&sum);

  893.     *dither_state= sum;

  894. }

  895. 

  896. 

  897. /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:

  898.    32 samples. */

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

  900. #if !CONFIG_FLOAT

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

  902.                          MPA_INT *window, int *dither_state,

  903.                          OUT_INT *samples, int incr,

  904.                          INTFLOAT sb_samples[SBLIMIT])

  905. {

  906.     register MPA_INT *synth_buf;

  907.     int offset;

  908. #if FRAC_BITS <= 15

  909.     int32_t tmp[32];

  910.     int j;

  911. #endif

  912. 

  913.     offset = *synth_buf_offset;

  914.     synth_buf = synth_buf_ptr + offset;

  915. 

  916. #if FRAC_BITS <= 15

  917.     dct32(tmp, sb_samples);

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

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

  920.            sound */

  921.         synth_buf[j] = av_clip_int16(tmp[j]);

  922.     }

  923. #else

  924.     dct32(synth_buf, sb_samples);

  925. #endif

  926. 

  927.     apply_window_mp3_c(synth_buf, window, dither_state, samples, incr);

  928. 

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

  930.     *synth_buf_offset = offset;

  931. }

  932. #endif

  933. 

  934. #define C3 FIXHR(0.86602540378443864676/2)

  935. 

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

  937. static const INTFLOAT icos36[9] = {

  938.     FIXR(0.50190991877167369479),

  939.     FIXR(0.51763809020504152469), //0

  940.     FIXR(0.55168895948124587824),

  941.     FIXR(0.61038729438072803416),

  942.     FIXR(0.70710678118654752439), //1

  943.     FIXR(0.87172339781054900991),

  944.     FIXR(1.18310079157624925896),

  945.     FIXR(1.93185165257813657349), //2

  946.     FIXR(5.73685662283492756461),

  947. };

  948. 

  949. /* 0.5 / cos(pi*(2*i+1)/36) */

  950. static const INTFLOAT icos36h[9] = {

  951.     FIXHR(0.50190991877167369479/2),

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

  953.     FIXHR(0.55168895948124587824/2),

  954.     FIXHR(0.61038729438072803416/2),

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

  956.     FIXHR(0.87172339781054900991/2),

  957.     FIXHR(1.18310079157624925896/4),

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

  959. //    FIXHR(5.73685662283492756461),

  960. };

  961. 

  962. /* 12 points IMDCT. We compute it "by hand" by factorizing obvious

  963.    cases. */

  964. static void imdct12(INTFLOAT *out, INTFLOAT *in)

  965. {

  966.     INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;

  967. 

  968.     in0= in[0*3];

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

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

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

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

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

  974.     in5 += in3;

  975.     in3 += in1;

  976. 

  977.     in2= MULH3(in2, C3, 2);

  978.     in3= MULH3(in3, C3, 4);

  979. 

  980.     t1 = in0 - in4;

  981.     t2 = MULH3(in1 - in5, icos36h[4], 2);

  982. 

  983.     out[ 7]=

  984.     out[10]= t1 + t2;

  985.     out[ 1]=

  986.     out[ 4]= t1 - t2;

  987. 

  988.     in0 += SHR(in4, 1);

  989.     in4 = in0 + in2;

  990.     in5 += 2*in1;

  991.     in1 = MULH3(in5 + in3, icos36h[1], 1);

  992.     out[ 8]=

  993.     out[ 9]= in4 + in1;

  994.     out[ 2]=

  995.     out[ 3]= in4 - in1;

  996. 

  997.     in0 -= in2;

  998.     in5 = MULH3(in5 - in3, icos36h[7], 2);

  999.     out[ 0]=

  1000.     out[ 5]= in0 - in5;

  1001.     out[ 6]=

  1002.     out[11]= in0 + in5;

  1003. }

  1004. 

  1005. /* cos(pi*i/18) */

  1006. #define C1 FIXHR(0.98480775301220805936/2)

  1007. #define C2 FIXHR(0.93969262078590838405/2)

  1008. #define C3 FIXHR(0.86602540378443864676/2)

  1009. #define C4 FIXHR(0.76604444311897803520/2)

  1010. #define C5 FIXHR(0.64278760968653932632/2)

  1011. #define C6 FIXHR(0.5/2)

  1012. #define C7 FIXHR(0.34202014332566873304/2)

  1013. #define C8 FIXHR(0.17364817766693034885/2)

  1014. 

  1015. 

  1016. /* using Lee like decomposition followed by hand coded 9 points DCT */

  1017. static void imdct36(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in, INTFLOAT *win)

  1018. {

  1019.     int i, j;

  1020.     INTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;

  1021.     INTFLOAT tmp[18], *tmp1, *in1;

  1022. 

  1023.     for(i=17;i>=1;i--)

  1024.         in[i] += in[i-1];

  1025.     for(i=17;i>=3;i-=2)

  1026.         in[i] += in[i-2];

  1027. 

  1028.     for(j=0;j<2;j++) {

  1029.         tmp1 = tmp + j;

  1030.         in1 = in + j;

  1031. 

  1032.         t2 = in1[2*4] + in1[2*8] - in1[2*2];

  1033. 

  1034.         t3 = in1[2*0] + SHR(in1[2*6],1);

  1035.         t1 = in1[2*0] - in1[2*6];

  1036.         tmp1[ 6] = t1 - SHR(t2,1);

  1037.         tmp1[16] = t1 + t2;

  1038. 

  1039.         t0 = MULH3(in1[2*2] + in1[2*4] ,    C2, 2);

  1040.         t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);

  1041.         t2 = MULH3(in1[2*2] + in1[2*8] ,   -C4, 2);

  1042. 

  1043.         tmp1[10] = t3 - t0 - t2;

  1044.         tmp1[ 2] = t3 + t0 + t1;

  1045.         tmp1[14] = t3 + t2 - t1;

  1046. 

  1047.         tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);

  1048.         t2 = MULH3(in1[2*1] + in1[2*5],    C1, 2);

  1049.         t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);

  1050.         t0 = MULH3(in1[2*3], C3, 2);

  1051. 

  1052.         t1 = MULH3(in1[2*1] + in1[2*7],   -C5, 2);

  1053. 

  1054.         tmp1[ 0] = t2 + t3 + t0;

  1055.         tmp1[12] = t2 + t1 - t0;

  1056.         tmp1[ 8] = t3 - t1 - t0;

  1057.     }

  1058. 

  1059.     i = 0;

  1060.     for(j=0;j<4;j++) {

  1061.         t0 = tmp[i];

  1062.         t1 = tmp[i + 2];

  1063.         s0 = t1 + t0;

  1064.         s2 = t1 - t0;

  1065. 

  1066.         t2 = tmp[i + 1];

  1067.         t3 = tmp[i + 3];

  1068.         s1 = MULH3(t3 + t2, icos36h[j], 2);

  1069.         s3 = MULLx(t3 - t2, icos36[8 - j], FRAC_BITS);

  1070. 

  1071.         t0 = s0 + s1;

  1072.         t1 = s0 - s1;

  1073.         out[(9 + j)*SBLIMIT] =  MULH3(t1, win[9 + j], 1) + buf[9 + j];

  1074.         out[(8 - j)*SBLIMIT] =  MULH3(t1, win[8 - j], 1) + buf[8 - j];

  1075.         buf[9 + j] = MULH3(t0, win[18 + 9 + j], 1);

  1076.         buf[8 - j] = MULH3(t0, win[18 + 8 - j], 1);

  1077. 

  1078.         t0 = s2 + s3;

  1079.         t1 = s2 - s3;

  1080.         out[(9 + 8 - j)*SBLIMIT] =  MULH3(t1, win[9 + 8 - j], 1) + buf[9 + 8 - j];

  1081.         out[(        j)*SBLIMIT] =  MULH3(t1, win[        j], 1) + buf[        j];

  1082.         buf[9 + 8 - j] = MULH3(t0, win[18 + 9 + 8 - j], 1);

  1083.         buf[      + j] = MULH3(t0, win[18         + j], 1);

  1084.         i += 4;

  1085.     }

  1086. 

  1087.     s0 = tmp[16];

  1088.     s1 = MULH3(tmp[17], icos36h[4], 2);

  1089.     t0 = s0 + s1;

  1090.     t1 = s0 - s1;

  1091.     out[(9 + 4)*SBLIMIT] =  MULH3(t1, win[9 + 4], 1) + buf[9 + 4];

  1092.     out[(8 - 4)*SBLIMIT] =  MULH3(t1, win[8 - 4], 1) + buf[8 - 4];

  1093.     buf[9 + 4] = MULH3(t0, win[18 + 9 + 4], 1);

  1094.     buf[8 - 4] = MULH3(t0, win[18 + 8 - 4], 1);

  1095. }

  1096. 

  1097. /* return the number of decoded frames */

  1098. static int mp_decode_layer1(MPADecodeContext *s)

  1099. {

  1100.     int bound, i, v, n, ch, j, mant;

  1101.     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];

  1102.     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

  1103. 

  1104.     if (s->mode == MPA_JSTEREO)

  1105.         bound = (s->mode_ext + 1) * 4;

  1106.     else

  1107.         bound = SBLIMIT;

  1108. 

  1109.     /* allocation bits */

  1110.     for(i=0;i<bound;i++) {

  1111.         for(ch=0;ch<s->nb_channels;ch++) {

  1112.             allocation[ch][i] = get_bits(&s->gb, 4);

  1113.         }

  1114.     }

  1115.     for(i=bound;i<SBLIMIT;i++) {

  1116.         allocation[0][i] = get_bits(&s->gb, 4);

  1117.     }

  1118. 

  1119.     /* scale factors */

  1120.     for(i=0;i<bound;i++) {

  1121.         for(ch=0;ch<s->nb_channels;ch++) {

  1122.             if (allocation[ch][i])

  1123.                 scale_factors[ch][i] = get_bits(&s->gb, 6);

  1124.         }

  1125.     }

  1126.     for(i=bound;i<SBLIMIT;i++) {

  1127.         if (allocation[0][i]) {

  1128.             scale_factors[0][i] = get_bits(&s->gb, 6);

  1129.             scale_factors[1][i] = get_bits(&s->gb, 6);

  1130.         }

  1131.     }

  1132. 

  1133.     /* compute samples */

  1134.     for(j=0;j<12;j++) {

  1135.         for(i=0;i<bound;i++) {

  1136.             for(ch=0;ch<s->nb_channels;ch++) {

  1137.                 n = allocation[ch][i];

  1138.                 if (n) {

  1139.                     mant = get_bits(&s->gb, n + 1);

  1140.                     v = l1_unscale(n, mant, scale_factors[ch][i]);

  1141.                 } else {

  1142.                     v = 0;

  1143.                 }

  1144.                 s->sb_samples[ch][j][i] = v;

  1145.             }

  1146.         }

  1147.         for(i=bound;i<SBLIMIT;i++) {

  1148.             n = allocation[0][i];

  1149.             if (n) {

  1150.                 mant = get_bits(&s->gb, n + 1);

  1151.                 v = l1_unscale(n, mant, scale_factors[0][i]);

  1152.                 s->sb_samples[0][j][i] = v;

  1153.                 v = l1_unscale(n, mant, scale_factors[1][i]);

  1154.                 s->sb_samples[1][j][i] = v;

  1155.             } else {

  1156.                 s->sb_samples[0][j][i] = 0;

  1157.                 s->sb_samples[1][j][i] = 0;

  1158.             }

  1159.         }

  1160.     }

  1161.     return 12;

  1162. }

  1163. 

  1164. static int mp_decode_layer2(MPADecodeContext *s)

  1165. {

  1166.     int sblimit; /* number of used subbands */

  1167.     const unsigned char *alloc_table;

  1168.     int table, bit_alloc_bits, i, j, ch, bound, v;

  1169.     unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];

  1170.     unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];

  1171.     unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;

  1172.     int scale, qindex, bits, steps, k, l, m, b;

  1173. 

  1174.     /* select decoding table */

  1175.     table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,

  1176.                             s->sample_rate, s->lsf);

  1177.     sblimit = ff_mpa_sblimit_table[table];

  1178.     alloc_table = ff_mpa_alloc_tables[table];

  1179. 

  1180.     if (s->mode == MPA_JSTEREO)

  1181.         bound = (s->mode_ext + 1) * 4;

  1182.     else

  1183.         bound = sblimit;

  1184. 

  1185.     dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);

  1186. 

  1187.     /* sanity check */

  1188.     if( bound > sblimit ) bound = sblimit;

  1189. 

  1190.     /* parse bit allocation */

  1191.     j = 0;

  1192.     for(i=0;i<bound;i++) {

  1193.         bit_alloc_bits = alloc_table[j];

  1194.         for(ch=0;ch<s->nb_channels;ch++) {

  1195.             bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);

  1196.         }

  1197.         j += 1 << bit_alloc_bits;

  1198.     }

  1199.     for(i=bound;i<sblimit;i++) {

  1200.         bit_alloc_bits = alloc_table[j];

  1201.         v = get_bits(&s->gb, bit_alloc_bits);

  1202.         bit_alloc[0][i] = v;

  1203.         bit_alloc[1][i] = v;

  1204.         j += 1 << bit_alloc_bits;

  1205.     }

  1206. 

  1207.     /* scale codes */

  1208.     for(i=0;i<sblimit;i++) {

  1209.         for(ch=0;ch<s->nb_channels;ch++) {

  1210.             if (bit_alloc[ch][i])

  1211.                 scale_code[ch][i] = get_bits(&s->gb, 2);

  1212.         }

  1213.     }

  1214. 

  1215.     /* scale factors */

  1216.     for(i=0;i<sblimit;i++) {

  1217.         for(ch=0;ch<s->nb_channels;ch++) {

  1218.             if (bit_alloc[ch][i]) {

  1219.                 sf = scale_factors[ch][i];

  1220.                 switch(scale_code[ch][i]) {

  1221.                 default:

  1222.                 case 0:

  1223.                     sf[0] = get_bits(&s->gb, 6);

  1224.                     sf[1] = get_bits(&s->gb, 6);

  1225.                     sf[2] = get_bits(&s->gb, 6);

  1226.                     break;

  1227.                 case 2:

  1228.                     sf[0] = get_bits(&s->gb, 6);

  1229.                     sf[1] = sf[0];

  1230.                     sf[2] = sf[0];

  1231.                     break;

  1232.                 case 1:

  1233.                     sf[0] = get_bits(&s->gb, 6);

  1234.                     sf[2] = get_bits(&s->gb, 6);

  1235.                     sf[1] = sf[0];

  1236.                     break;

  1237.                 case 3:

  1238.                     sf[0] = get_bits(&s->gb, 6);

  1239.                     sf[2] = get_bits(&s->gb, 6);

  1240.                     sf[1] = sf[2];

  1241.                     break;

  1242.                 }

  1243.             }

  1244.         }

  1245.     }

  1246. 

  1247.     /* samples */

  1248.     for(k=0;k<3;k++) {

  1249.         for(l=0;l<12;l+=3) {

  1250.             j = 0;

  1251.             for(i=0;i<bound;i++) {

  1252.                 bit_alloc_bits = alloc_table[j];

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

  1254.                     b = bit_alloc[ch][i];

  1255.                     if (b) {

  1256.                         scale = scale_factors[ch][i][k];

  1257.                         qindex = alloc_table[j+b];

  1258.                         bits = ff_mpa_quant_bits[qindex];

  1259.                         if (bits < 0) {

  1260.                             /* 3 values at the same time */

  1261.                             v = get_bits(&s->gb, -bits);

  1262.                             steps = ff_mpa_quant_steps[qindex];

  1263.                             s->sb_samples[ch][k * 12 + l + 0][i] =

  1264.                                 l2_unscale_group(steps, v % steps, scale);

  1265.                             v = v / steps;

  1266.                             s->sb_samples[ch][k * 12 + l + 1][i] =

  1267.                                 l2_unscale_group(steps, v % steps, scale);

  1268.                             v = v / steps;

  1269.                             s->sb_samples[ch][k * 12 + l + 2][i] =

  1270.                                 l2_unscale_group(steps, v, scale);

  1271.                         } else {

  1272.                             for(m=0;m<3;m++) {

  1273.                                 v = get_bits(&s->gb, bits);

  1274.                                 v = l1_unscale(bits - 1, v, scale);

  1275.                                 s->sb_samples[ch][k * 12 + l + m][i] = v;

  1276.                             }

  1277.                         }

  1278.                     } else {

  1279.                         s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1280.                         s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1281.                         s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1282.                     }

  1283.                 }

  1284.                 /* next subband in alloc table */

  1285.                 j += 1 << bit_alloc_bits;

  1286.             }

  1287.             /* XXX: find a way to avoid this duplication of code */

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

  1289.                 bit_alloc_bits = alloc_table[j];

  1290.                 b = bit_alloc[0][i];

  1291.                 if (b) {

  1292.                     int mant, scale0, scale1;

  1293.                     scale0 = scale_factors[0][i][k];

  1294.                     scale1 = scale_factors[1][i][k];

  1295.                     qindex = alloc_table[j+b];

  1296.                     bits = ff_mpa_quant_bits[qindex];

  1297.                     if (bits < 0) {

  1298.                         /* 3 values at the same time */

  1299.                         v = get_bits(&s->gb, -bits);

  1300.                         steps = ff_mpa_quant_steps[qindex];

  1301.                         mant = v % steps;

  1302.                         v = v / steps;

  1303.                         s->sb_samples[0][k * 12 + l + 0][i] =

  1304.                             l2_unscale_group(steps, mant, scale0);

  1305.                         s->sb_samples[1][k * 12 + l + 0][i] =

  1306.                             l2_unscale_group(steps, mant, scale1);

  1307.                         mant = v % steps;

  1308.                         v = v / steps;

  1309.                         s->sb_samples[0][k * 12 + l + 1][i] =

  1310.                             l2_unscale_group(steps, mant, scale0);

  1311.                         s->sb_samples[1][k * 12 + l + 1][i] =

  1312.                             l2_unscale_group(steps, mant, scale1);

  1313.                         s->sb_samples[0][k * 12 + l + 2][i] =

  1314.                             l2_unscale_group(steps, v, scale0);

  1315.                         s->sb_samples[1][k * 12 + l + 2][i] =

  1316.                             l2_unscale_group(steps, v, scale1);

  1317.                     } else {

  1318.                         for(m=0;m<3;m++) {

  1319.                             mant = get_bits(&s->gb, bits);

  1320.                             s->sb_samples[0][k * 12 + l + m][i] =

  1321.                                 l1_unscale(bits - 1, mant, scale0);

  1322.                             s->sb_samples[1][k * 12 + l + m][i] =

  1323.                                 l1_unscale(bits - 1, mant, scale1);

  1324.                         }

  1325.                     }

  1326.                 } else {

  1327.                     s->sb_samples[0][k * 12 + l + 0][i] = 0;

  1328.                     s->sb_samples[0][k * 12 + l + 1][i] = 0;

  1329.                     s->sb_samples[0][k * 12 + l + 2][i] = 0;

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

  1331.                     s->sb_samples[1][k * 12 + l + 1][i] = 0;

  1332.                     s->sb_samples[1][k * 12 + l + 2][i] = 0;

  1333.                 }

  1334.                 /* next subband in alloc table */

  1335.                 j += 1 << bit_alloc_bits;

  1336.             }

  1337.             /* fill remaining samples to zero */

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

  1339.                 for(ch=0;ch<s->nb_channels;ch++) {

  1340.                     s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1341.                     s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1342.                     s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1343.                 }

  1344.             }

  1345.         }

  1346.     }

  1347.     return 3 * 12;

  1348. }

  1349. 

  1350. #define SPLIT(dst,sf,n)\

  1351.     if(n==3){\

  1352.         int m= (sf*171)>>9;\

  1353.         dst= sf - 3*m;\

  1354.         sf=m;\

  1355.     }else if(n==4){\

  1356.         dst= sf&3;\

  1357.         sf>>=2;\

  1358.     }else if(n==5){\

  1359.         int m= (sf*205)>>10;\

  1360.         dst= sf - 5*m;\

  1361.         sf=m;\

  1362.     }else if(n==6){\

  1363.         int m= (sf*171)>>10;\

  1364.         dst= sf - 6*m;\

  1365.         sf=m;\

  1366.     }else{\

  1367.         dst=0;\

  1368.     }

  1369. 

  1370. static av_always_inline void lsf_sf_expand(int *slen,

  1371.                                  int sf, int n1, int n2, int n3)

  1372. {

  1373.     SPLIT(slen[3], sf, n3)

  1374.     SPLIT(slen[2], sf, n2)

  1375.     SPLIT(slen[1], sf, n1)

  1376.     slen[0] = sf;

  1377. }

  1378. 

  1379. static void exponents_from_scale_factors(MPADecodeContext *s,

  1380.                                          GranuleDef *g,

  1381.                                          int16_t *exponents)

  1382. {

  1383.     const uint8_t *bstab, *pretab;

  1384.     int len, i, j, k, l, v0, shift, gain, gains[3];

  1385.     int16_t *exp_ptr;

  1386. 

  1387.     exp_ptr = exponents;

  1388.     gain = g->global_gain - 210;

  1389.     shift = g->scalefac_scale + 1;

  1390. 

  1391.     bstab = band_size_long[s->sample_rate_index];

  1392.     pretab = mpa_pretab[g->preflag];

  1393.     for(i=0;i<g->long_end;i++) {

  1394.         v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;

  1395.         len = bstab[i];

  1396.         for(j=len;j>0;j--)

  1397.             *exp_ptr++ = v0;

  1398.     }

  1399. 

  1400.     if (g->short_start < 13) {

  1401.         bstab = band_size_short[s->sample_rate_index];

  1402.         gains[0] = gain - (g->subblock_gain[0] << 3);

  1403.         gains[1] = gain - (g->subblock_gain[1] << 3);

  1404.         gains[2] = gain - (g->subblock_gain[2] << 3);

  1405.         k = g->long_end;

  1406.         for(i=g->short_start;i<13;i++) {

  1407.             len = bstab[i];

  1408.             for(l=0;l<3;l++) {

  1409.                 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;

  1410.                 for(j=len;j>0;j--)

  1411.                 *exp_ptr++ = v0;

  1412.             }

  1413.         }

  1414.     }

  1415. }

  1416. 

  1417. /* handle n = 0 too */

  1418. static inline int get_bitsz(GetBitContext *s, int n)

  1419. {

  1420.     if (n == 0)

  1421.         return 0;

  1422.     else

  1423.         return get_bits(s, n);

  1424. }

  1425. 

  1426. 

  1427. static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){

  1428.     if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){

  1429.         s->gb= s->in_gb;

  1430.         s->in_gb.buffer=NULL;

  1431.         assert((get_bits_count(&s->gb) & 7) == 0);

  1432.         skip_bits_long(&s->gb, *pos - *end_pos);

  1433.         *end_pos2=

  1434.         *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;

  1435.         *pos= get_bits_count(&s->gb);

  1436.     }

  1437. }

  1438. 

  1439. /* Following is a optimized code for

  1440.             INTFLOAT v = *src

  1441.             if(get_bits1(&s->gb))

  1442.                 v = -v;

  1443.             *dst = v;

  1444. */

  1445. #if CONFIG_FLOAT

  1446. #define READ_FLIP_SIGN(dst,src)\

  1447.             v = AV_RN32A(src) ^ (get_bits1(&s->gb)<<31);\

  1448.             AV_WN32A(dst, v);

  1449. #else

  1450. #define READ_FLIP_SIGN(dst,src)\

  1451.             v= -get_bits1(&s->gb);\

  1452.             *(dst) = (*(src) ^ v) - v;

  1453. #endif

  1454. 

  1455. static int huffman_decode(MPADecodeContext *s, GranuleDef *g,

  1456.                           int16_t *exponents, int end_pos2)

  1457. {

  1458.     int s_index;

  1459.     int i;

  1460.     int last_pos, bits_left;

  1461.     VLC *vlc;

  1462.     int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);

  1463. 

  1464.     /* low frequencies (called big values) */

  1465.     s_index = 0;

  1466.     for(i=0;i<3;i++) {

  1467.         int j, k, l, linbits;

  1468.         j = g->region_size[i];

  1469.         if (j == 0)

  1470.             continue;

  1471.         /* select vlc table */

  1472.         k = g->table_select[i];

  1473.         l = mpa_huff_data[k][0];

  1474.         linbits = mpa_huff_data[k][1];

  1475.         vlc = &huff_vlc[l];

  1476. 

  1477.         if(!l){

  1478.             memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);

  1479.             s_index += 2*j;

  1480.             continue;

  1481.         }

  1482. 

  1483.         /* read huffcode and compute each couple */

  1484.         for(;j>0;j--) {

  1485.             int exponent, x, y;

  1486.             int v;

  1487.             int pos= get_bits_count(&s->gb);

  1488. 

  1489.             if (pos >= end_pos){

  1490. //                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);

  1491.                 switch_buffer(s, &pos, &end_pos, &end_pos2);

  1492. //                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);

  1493.                 if(pos >= end_pos)

  1494.                     break;

  1495.             }

  1496.             y = get_vlc2(&s->gb, vlc->table, 7, 3);

  1497. 

  1498.             if(!y){

  1499.                 g->sb_hybrid[s_index  ] =

  1500.                 g->sb_hybrid[s_index+1] = 0;

  1501.                 s_index += 2;

  1502.                 continue;

  1503.             }

  1504. 

  1505.             exponent= exponents[s_index];

  1506. 

  1507.             dprintf(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",

  1508.                     i, g->region_size[i] - j, x, y, exponent);

  1509.             if(y&16){

  1510.                 x = y >> 5;

  1511.                 y = y & 0x0f;

  1512.                 if (x < 15){

  1513.                     READ_FLIP_SIGN(g->sb_hybrid+s_index, RENAME(expval_table)[ exponent ]+x)

  1514.                 }else{

  1515.                     x += get_bitsz(&s->gb, linbits);

  1516.                     v = l3_unscale(x, exponent);

  1517.                     if (get_bits1(&s->gb))

  1518.                         v = -v;

  1519.                     g->sb_hybrid[s_index] = v;

  1520.                 }

  1521.                 if (y < 15){

  1522.                     READ_FLIP_SIGN(g->sb_hybrid+s_index+1, RENAME(expval_table)[ exponent ]+y)

  1523.                 }else{

  1524.                     y += get_bitsz(&s->gb, linbits);

  1525.                     v = l3_unscale(y, exponent);

  1526.                     if (get_bits1(&s->gb))

  1527.                         v = -v;

  1528.                     g->sb_hybrid[s_index+1] = v;

  1529.                 }

  1530.             }else{

  1531.                 x = y >> 5;

  1532.                 y = y & 0x0f;

  1533.                 x += y;

  1534.                 if (x < 15){

  1535.                     READ_FLIP_SIGN(g->sb_hybrid+s_index+!!y, RENAME(expval_table)[ exponent ]+x)

  1536.                 }else{

  1537.                     x += get_bitsz(&s->gb, linbits);

  1538.                     v = l3_unscale(x, exponent);

  1539.                     if (get_bits1(&s->gb))

  1540.                         v = -v;

  1541.                     g->sb_hybrid[s_index+!!y] = v;

  1542.                 }

  1543.                 g->sb_hybrid[s_index+ !y] = 0;

  1544.             }

  1545.             s_index+=2;

  1546.         }

  1547.     }

  1548. 

  1549.     /* high frequencies */

  1550.     vlc = &huff_quad_vlc[g->count1table_select];

  1551.     last_pos=0;

  1552.     while (s_index <= 572) {

  1553.         int pos, code;

  1554.         pos = get_bits_count(&s->gb);

  1555.         if (pos >= end_pos) {

  1556.             if (pos > end_pos2 && last_pos){

  1557.                 /* some encoders generate an incorrect size for this

  1558.                    part. We must go back into the data */

  1559.                 s_index -= 4;

  1560.                 skip_bits_long(&s->gb, last_pos - pos);

  1561.                 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);

  1562.                 if(s->error_recognition >= FF_ER_COMPLIANT)

  1563.                     s_index=0;

  1564.                 break;

  1565.             }

  1566. //                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);

  1567.             switch_buffer(s, &pos, &end_pos, &end_pos2);

  1568. //                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);

  1569.             if(pos >= end_pos)

  1570.                 break;

  1571.         }

  1572.         last_pos= pos;

  1573. 

  1574.         code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);

  1575.         dprintf(s->avctx, "t=%d code=%d\n", g->count1table_select, code);

  1576.         g->sb_hybrid[s_index+0]=

  1577.         g->sb_hybrid[s_index+1]=

  1578.         g->sb_hybrid[s_index+2]=

  1579.         g->sb_hybrid[s_index+3]= 0;

  1580.         while(code){

  1581.             static const int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};

  1582.             int v;

  1583.             int pos= s_index+idxtab[code];

  1584.             code ^= 8>>idxtab[code];

  1585.             READ_FLIP_SIGN(g->sb_hybrid+pos, RENAME(exp_table)+exponents[pos])

  1586.         }

  1587.         s_index+=4;

  1588.     }

  1589.     /* skip extension bits */

  1590.     bits_left = end_pos2 - get_bits_count(&s->gb);

  1591. //av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);

  1592.     if (bits_left < 0 && s->error_recognition >= FF_ER_COMPLIANT) {

  1593.         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1594.         s_index=0;

  1595.     }else if(bits_left > 0 && s->error_recognition >= FF_ER_AGGRESSIVE){

  1596.         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1597.         s_index=0;

  1598.     }

  1599.     memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));

  1600.     skip_bits_long(&s->gb, bits_left);

  1601. 

  1602.     i= get_bits_count(&s->gb);

  1603.     switch_buffer(s, &i, &end_pos, &end_pos2);

  1604. 

  1605.     return 0;

  1606. }

  1607. 

  1608. /* Reorder short blocks from bitstream order to interleaved order. It

  1609.    would be faster to do it in parsing, but the code would be far more

  1610.    complicated */

  1611. static void reorder_block(MPADecodeContext *s, GranuleDef *g)

  1612. {

  1613.     int i, j, len;

  1614.     INTFLOAT *ptr, *dst, *ptr1;

  1615.     INTFLOAT tmp[576];

  1616. 

  1617.     if (g->block_type != 2)

  1618.         return;

  1619. 

  1620.     if (g->switch_point) {

  1621.         if (s->sample_rate_index != 8) {

  1622.             ptr = g->sb_hybrid + 36;

  1623.         } else {

  1624.             ptr = g->sb_hybrid + 48;

  1625.         }

  1626.     } else {

  1627.         ptr = g->sb_hybrid;

  1628.     }

  1629. 

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

  1631.         len = band_size_short[s->sample_rate_index][i];

  1632.         ptr1 = ptr;

  1633.         dst = tmp;

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

  1635.             *dst++ = ptr[0*len];

  1636.             *dst++ = ptr[1*len];

  1637.             *dst++ = ptr[2*len];

  1638.             ptr++;

  1639.         }

  1640.         ptr+=2*len;

  1641.         memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));

  1642.     }

  1643. }

  1644. 

  1645. #define ISQRT2 FIXR(0.70710678118654752440)

  1646. 

  1647. static void compute_stereo(MPADecodeContext *s,

  1648.                            GranuleDef *g0, GranuleDef *g1)

  1649. {

  1650.     int i, j, k, l;

  1651.     int sf_max, sf, len, non_zero_found;

  1652.     INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;

  1653.     int non_zero_found_short[3];

  1654. 

  1655.     /* intensity stereo */

  1656.     if (s->mode_ext & MODE_EXT_I_STEREO) {

  1657.         if (!s->lsf) {

  1658.             is_tab = is_table;

  1659.             sf_max = 7;

  1660.         } else {

  1661.             is_tab = is_table_lsf[g1->scalefac_compress & 1];

  1662.             sf_max = 16;

  1663.         }

  1664. 

  1665.         tab0 = g0->sb_hybrid + 576;

  1666.         tab1 = g1->sb_hybrid + 576;

  1667. 

  1668.         non_zero_found_short[0] = 0;

  1669.         non_zero_found_short[1] = 0;

  1670.         non_zero_found_short[2] = 0;

  1671.         k = (13 - g1->short_start) * 3 + g1->long_end - 3;

  1672.         for(i = 12;i >= g1->short_start;i--) {

  1673.             /* for last band, use previous scale factor */

  1674.             if (i != 11)

  1675.                 k -= 3;

  1676.             len = band_size_short[s->sample_rate_index][i];

  1677.             for(l=2;l>=0;l--) {

  1678.                 tab0 -= len;

  1679.                 tab1 -= len;

  1680.                 if (!non_zero_found_short[l]) {

  1681.                     /* test if non zero band. if so, stop doing i-stereo */

  1682.                     for(j=0;j<len;j++) {

  1683.                         if (tab1[j] != 0) {

  1684.                             non_zero_found_short[l] = 1;

  1685.                             goto found1;

  1686.                         }

  1687.                     }

  1688.                     sf = g1->scale_factors[k + l];

  1689.                     if (sf >= sf_max)

  1690.                         goto found1;

  1691. 

  1692.                     v1 = is_tab[0][sf];

  1693.                     v2 = is_tab[1][sf];

  1694.                     for(j=0;j<len;j++) {

  1695.                         tmp0 = tab0[j];

  1696.                         tab0[j] = MULLx(tmp0, v1, FRAC_BITS);

  1697.                         tab1[j] = MULLx(tmp0, v2, FRAC_BITS);

  1698.                     }

  1699.                 } else {

  1700.                 found1:

  1701.                     if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1702.                         /* lower part of the spectrum : do ms stereo

  1703.                            if enabled */

  1704.                         for(j=0;j<len;j++) {

  1705.                             tmp0 = tab0[j];

  1706.                             tmp1 = tab1[j];

  1707.                             tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1708.                             tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1709.                         }

  1710.                     }

  1711.                 }

  1712.             }

  1713.         }

  1714. 

  1715.         non_zero_found = non_zero_found_short[0] |

  1716.             non_zero_found_short[1] |

  1717.             non_zero_found_short[2];

  1718. 

  1719.         for(i = g1->long_end - 1;i >= 0;i--) {

  1720.             len = band_size_long[s->sample_rate_index][i];

  1721.             tab0 -= len;

  1722.             tab1 -= len;

  1723.             /* test if non zero band. if so, stop doing i-stereo */

  1724.             if (!non_zero_found) {

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

  1726.                     if (tab1[j] != 0) {

  1727.                         non_zero_found = 1;

  1728.                         goto found2;

  1729.                     }

  1730.                 }

  1731.                 /* for last band, use previous scale factor */

  1732.                 k = (i == 21) ? 20 : i;

  1733.                 sf = g1->scale_factors[k];

  1734.                 if (sf >= sf_max)

  1735.                     goto found2;

  1736.                 v1 = is_tab[0][sf];

  1737.                 v2 = is_tab[1][sf];

  1738.                 for(j=0;j<len;j++) {

  1739.                     tmp0 = tab0[j];

  1740.                     tab0[j] = MULLx(tmp0, v1, FRAC_BITS);

  1741.                     tab1[j] = MULLx(tmp0, v2, FRAC_BITS);

  1742.                 }

  1743.             } else {

  1744.             found2:

  1745.                 if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1746.                     /* lower part of the spectrum : do ms stereo

  1747.                        if enabled */

  1748.                     for(j=0;j<len;j++) {

  1749.                         tmp0 = tab0[j];

  1750.                         tmp1 = tab1[j];

  1751.                         tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1752.                         tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1753.                     }

  1754.                 }

  1755.             }

  1756.         }

  1757.     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1758.         /* ms stereo ONLY */

  1759.         /* NOTE: the 1/sqrt(2) normalization factor is included in the

  1760.            global gain */

  1761.         tab0 = g0->sb_hybrid;

  1762.         tab1 = g1->sb_hybrid;

  1763.         for(i=0;i<576;i++) {

  1764.             tmp0 = tab0[i];

  1765.             tmp1 = tab1[i];

  1766.             tab0[i] = tmp0 + tmp1;

  1767.             tab1[i] = tmp0 - tmp1;

  1768.         }

  1769.     }

  1770. }

  1771. 

  1772. static void compute_antialias_integer(MPADecodeContext *s,

  1773.                               GranuleDef *g)

  1774. {

  1775.     int32_t *ptr, *csa;

  1776.     int n, i;

  1777. 

  1778.     /* we antialias only "long" bands */

  1779.     if (g->block_type == 2) {

  1780.         if (!g->switch_point)

  1781.             return;

  1782.         /* XXX: check this for 8000Hz case */

  1783.         n = 1;

  1784.     } else {

  1785.         n = SBLIMIT - 1;

  1786.     }

  1787. 

  1788.     ptr = g->sb_hybrid + 18;

  1789.     for(i = n;i > 0;i--) {

  1790.         int tmp0, tmp1, tmp2;

  1791.         csa = &csa_table[0][0];

  1792. #define INT_AA(j) \

  1793.             tmp0 = ptr[-1-j];\

  1794.             tmp1 = ptr[   j];\

  1795.             tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\

  1796.             ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\

  1797.             ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));

  1798. 

  1799.         INT_AA(0)

  1800.         INT_AA(1)

  1801.         INT_AA(2)

  1802.         INT_AA(3)

  1803.         INT_AA(4)

  1804.         INT_AA(5)

  1805.         INT_AA(6)

  1806.         INT_AA(7)

  1807. 

  1808.         ptr += 18;

  1809.     }

  1810. }

  1811. 

  1812. static void compute_antialias_float(MPADecodeContext *s,

  1813.                               GranuleDef *g)

  1814. {

  1815.     float *ptr;

  1816.     int n, i;

  1817. 

  1818.     /* we antialias only "long" bands */

  1819.     if (g->block_type == 2) {

  1820.         if (!g->switch_point)

  1821.             return;

  1822.         /* XXX: check this for 8000Hz case */

  1823.         n = 1;

  1824.     } else {

  1825.         n = SBLIMIT - 1;

  1826.     }

  1827. 

  1828.     ptr = g->sb_hybrid + 18;

  1829.     for(i = n;i > 0;i--) {

  1830.         float tmp0, tmp1;

  1831.         float *csa = &csa_table_float[0][0];

  1832. #define FLOAT_AA(j)\

  1833.         tmp0= ptr[-1-j];\

  1834.         tmp1= ptr[   j];\

  1835.         ptr[-1-j] = tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j];\

  1836.         ptr[   j] = tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j];

  1837. 

  1838.         FLOAT_AA(0)

  1839.         FLOAT_AA(1)

  1840.         FLOAT_AA(2)

  1841.         FLOAT_AA(3)

  1842.         FLOAT_AA(4)

  1843.         FLOAT_AA(5)

  1844.         FLOAT_AA(6)

  1845.         FLOAT_AA(7)

  1846. 

  1847.         ptr += 18;

  1848.     }

  1849. }

  1850. 

  1851. static void compute_imdct(MPADecodeContext *s,

  1852.                           GranuleDef *g,

  1853.                           INTFLOAT *sb_samples,

  1854.                           INTFLOAT *mdct_buf)

  1855. {

  1856.     INTFLOAT *win, *win1, *out_ptr, *ptr, *buf, *ptr1;

  1857.     INTFLOAT out2[12];

  1858.     int i, j, mdct_long_end, sblimit;

  1859. 

  1860.     /* find last non zero block */

  1861.     ptr = g->sb_hybrid + 576;

  1862.     ptr1 = g->sb_hybrid + 2 * 18;

  1863.     while (ptr >= ptr1) {

  1864.         int32_t *p;

  1865.         ptr -= 6;

  1866.         p= (int32_t*)ptr;

  1867.         if(p[0] | p[1] | p[2] | p[3] | p[4] | p[5])

  1868.             break;

  1869.     }

  1870.     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

  1871. 

  1872.     if (g->block_type == 2) {

  1873.         /* XXX: check for 8000 Hz */

  1874.         if (g->switch_point)

  1875.             mdct_long_end = 2;

  1876.         else

  1877.             mdct_long_end = 0;

  1878.     } else {

  1879.         mdct_long_end = sblimit;

  1880.     }

  1881. 

  1882.     buf = mdct_buf;

  1883.     ptr = g->sb_hybrid;

  1884.     for(j=0;j<mdct_long_end;j++) {

  1885.         /* apply window & overlap with previous buffer */

  1886.         out_ptr = sb_samples + j;

  1887.         /* select window */

  1888.         if (g->switch_point && j < 2)

  1889.             win1 = mdct_win[0];

  1890.         else

  1891.             win1 = mdct_win[g->block_type];

  1892.         /* select frequency inversion */

  1893.         win = win1 + ((4 * 36) & -(j & 1));

  1894.         imdct36(out_ptr, buf, ptr, win);

  1895.         out_ptr += 18*SBLIMIT;

  1896.         ptr += 18;

  1897.         buf += 18;

  1898.     }

  1899.     for(j=mdct_long_end;j<sblimit;j++) {

  1900.         /* select frequency inversion */

  1901.         win = mdct_win[2] + ((4 * 36) & -(j & 1));

  1902.         out_ptr = sb_samples + j;

  1903. 

  1904.         for(i=0; i<6; i++){

  1905.             *out_ptr = buf[i];

  1906.             out_ptr += SBLIMIT;

  1907.         }

  1908.         imdct12(out2, ptr + 0);

  1909.         for(i=0;i<6;i++) {

  1910.             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*1];

  1911.             buf[i + 6*2] = MULH3(out2[i + 6], win[i + 6], 1);

  1912.             out_ptr += SBLIMIT;

  1913.         }

  1914.         imdct12(out2, ptr + 1);

  1915.         for(i=0;i<6;i++) {

  1916.             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*2];

  1917.             buf[i + 6*0] = MULH3(out2[i + 6], win[i + 6], 1);

  1918.             out_ptr += SBLIMIT;

  1919.         }

  1920.         imdct12(out2, ptr + 2);

  1921.         for(i=0;i<6;i++) {

  1922.             buf[i + 6*0] = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*0];

  1923.             buf[i + 6*1] = MULH3(out2[i + 6], win[i + 6], 1);

  1924.             buf[i + 6*2] = 0;

  1925.         }

  1926.         ptr += 18;

  1927.         buf += 18;

  1928.     }

  1929.     /* zero bands */

  1930.     for(j=sblimit;j<SBLIMIT;j++) {

  1931.         /* overlap */

  1932.         out_ptr = sb_samples + j;

  1933.         for(i=0;i<18;i++) {

  1934.             *out_ptr = buf[i];

  1935.             buf[i] = 0;

  1936.             out_ptr += SBLIMIT;

  1937.         }

  1938.         buf += 18;

  1939.     }

  1940. }

  1941. 

  1942. /* main layer3 decoding function */

  1943. static int mp_decode_layer3(MPADecodeContext *s)

  1944. {

  1945.     int nb_granules, main_data_begin, private_bits;

  1946.     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;

  1947.     GranuleDef *g;

  1948.     int16_t exponents[576]; //FIXME try INTFLOAT

  1949. 

  1950.     /* read side info */

  1951.     if (s->lsf) {

  1952.         main_data_begin = get_bits(&s->gb, 8);

  1953.         private_bits = get_bits(&s->gb, s->nb_channels);

  1954.         nb_granules = 1;

  1955.     } else {

  1956.         main_data_begin = get_bits(&s->gb, 9);

  1957.         if (s->nb_channels == 2)

  1958.             private_bits = get_bits(&s->gb, 3);

  1959.         else

  1960.             private_bits = get_bits(&s->gb, 5);

  1961.         nb_granules = 2;

  1962.         for(ch=0;ch<s->nb_channels;ch++) {

  1963.             s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */

  1964.             s->granules[ch][1].scfsi = get_bits(&s->gb, 4);

  1965.         }

  1966.     }

  1967. 

  1968.     for(gr=0;gr<nb_granules;gr++) {

  1969.         for(ch=0;ch<s->nb_channels;ch++) {

  1970.             dprintf(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);

  1971.             g = &s->granules[ch][gr];

  1972.             g->part2_3_length = get_bits(&s->gb, 12);

  1973.             g->big_values = get_bits(&s->gb, 9);

  1974.             if(g->big_values > 288){

  1975.                 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");

  1976.                 return -1;

  1977.             }

  1978. 

  1979.             g->global_gain = get_bits(&s->gb, 8);

  1980.             /* if MS stereo only is selected, we precompute the

  1981.                1/sqrt(2) renormalization factor */

  1982.             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==

  1983.                 MODE_EXT_MS_STEREO)

  1984.                 g->global_gain -= 2;

  1985.             if (s->lsf)

  1986.                 g->scalefac_compress = get_bits(&s->gb, 9);

  1987.             else

  1988.                 g->scalefac_compress = get_bits(&s->gb, 4);

  1989.             blocksplit_flag = get_bits1(&s->gb);

  1990.             if (blocksplit_flag) {

  1991.                 g->block_type = get_bits(&s->gb, 2);

  1992.                 if (g->block_type == 0){

  1993.                     av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");

  1994.                     return -1;

  1995.                 }

  1996.                 g->switch_point = get_bits1(&s->gb);

  1997.                 for(i=0;i<2;i++)

  1998.                     g->table_select[i] = get_bits(&s->gb, 5);

  1999.                 for(i=0;i<3;i++)

  2000.                     g->subblock_gain[i] = get_bits(&s->gb, 3);

  2001.                 ff_init_short_region(s, g);

  2002.             } else {

  2003.                 int region_address1, region_address2;

  2004.                 g->block_type = 0;

  2005.                 g->switch_point = 0;

  2006.                 for(i=0;i<3;i++)

  2007.                     g->table_select[i] = get_bits(&s->gb, 5);

  2008.                 /* compute huffman coded region sizes */

  2009.                 region_address1 = get_bits(&s->gb, 4);

  2010.                 region_address2 = get_bits(&s->gb, 3);

  2011.                 dprintf(s->avctx, "region1=%d region2=%d\n",

  2012.                         region_address1, region_address2);

  2013.                 ff_init_long_region(s, g, region_address1, region_address2);

  2014.             }

  2015.             ff_region_offset2size(g);

  2016.             ff_compute_band_indexes(s, g);

  2017. 

  2018.             g->preflag = 0;

  2019.             if (!s->lsf)

  2020.                 g->preflag = get_bits1(&s->gb);

  2021.             g->scalefac_scale = get_bits1(&s->gb);

  2022.             g->count1table_select = get_bits1(&s->gb);

  2023.             dprintf(s->avctx, "block_type=%d switch_point=%d\n",

  2024.                     g->block_type, g->switch_point);

  2025.         }

  2026.     }

  2027. 

  2028.   if (!s->adu_mode) {

  2029.     const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);

  2030.     assert((get_bits_count(&s->gb) & 7) == 0);

  2031.     /* now we get bits from the main_data_begin offset */

  2032.     dprintf(s->avctx, "seekback: %d\n", main_data_begin);

  2033. //av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);

  2034. 

  2035.     memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);

  2036.     s->in_gb= s->gb;

  2037.         init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);

  2038.         skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));

  2039.   }

  2040. 

  2041.     for(gr=0;gr<nb_granules;gr++) {

  2042.         for(ch=0;ch<s->nb_channels;ch++) {

  2043.             g = &s->granules[ch][gr];

  2044.             if(get_bits_count(&s->gb)<0){

  2045.                 av_log(s->avctx, AV_LOG_DEBUG, "mdb:%d, lastbuf:%d skipping granule %d\n",

  2046.                                             main_data_begin, s->last_buf_size, gr);

  2047.                 skip_bits_long(&s->gb, g->part2_3_length);

  2048.                 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));

  2049.                 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){

  2050.                     skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);

  2051.                     s->gb= s->in_gb;

  2052.                     s->in_gb.buffer=NULL;

  2053.                 }

  2054.                 continue;

  2055.             }

  2056. 

  2057.             bits_pos = get_bits_count(&s->gb);

  2058. 

  2059.             if (!s->lsf) {

  2060.                 uint8_t *sc;

  2061.                 int slen, slen1, slen2;

  2062. 

  2063.                 /* MPEG1 scale factors */

  2064.                 slen1 = slen_table[0][g->scalefac_compress];

  2065.                 slen2 = slen_table[1][g->scalefac_compress];

  2066.                 dprintf(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);

  2067.                 if (g->block_type == 2) {

  2068.                     n = g->switch_point ? 17 : 18;

  2069.                     j = 0;

  2070.                     if(slen1){

  2071.                         for(i=0;i<n;i++)

  2072.                             g->scale_factors[j++] = get_bits(&s->gb, slen1);

  2073.                     }else{

  2074.                         for(i=0;i<n;i++)

  2075.                             g->scale_factors[j++] = 0;

  2076.                     }

  2077.                     if(slen2){

  2078.                         for(i=0;i<18;i++)

  2079.                             g->scale_factors[j++] = get_bits(&s->gb, slen2);

  2080.                         for(i=0;i<3;i++)

  2081.                             g->scale_factors[j++] = 0;

  2082.                     }else{

  2083.                         for(i=0;i<21;i++)

  2084.                             g->scale_factors[j++] = 0;

  2085.                     }

  2086.                 } else {

  2087.                     sc = s->granules[ch][0].scale_factors;

  2088.                     j = 0;

  2089.                     for(k=0;k<4;k++) {

  2090.                         n = (k == 0 ? 6 : 5);

  2091.                         if ((g->scfsi & (0x8 >> k)) == 0) {

  2092.                             slen = (k < 2) ? slen1 : slen2;

  2093.                             if(slen){

  2094.                                 for(i=0;i<n;i++)

  2095.                                     g->scale_factors[j++] = get_bits(&s->gb, slen);

  2096.                             }else{

  2097.                                 for(i=0;i<n;i++)

  2098.                                     g->scale_factors[j++] = 0;

  2099.                             }

  2100.                         } else {

  2101.                             /* simply copy from last granule */

  2102.                             for(i=0;i<n;i++) {

  2103.                                 g->scale_factors[j] = sc[j];

  2104.                                 j++;

  2105.                             }

  2106.                         }

  2107.                     }

  2108.                     g->scale_factors[j++] = 0;

  2109.                 }

  2110.             } else {

  2111.                 int tindex, tindex2, slen[4], sl, sf;

  2112. 

  2113.                 /* LSF scale factors */

  2114.                 if (g->block_type == 2) {

  2115.                     tindex = g->switch_point ? 2 : 1;

  2116.                 } else {

  2117.                     tindex = 0;

  2118.                 }

  2119.                 sf = g->scalefac_compress;

  2120.                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {

  2121.                     /* intensity stereo case */

  2122.                     sf >>= 1;

  2123.                     if (sf < 180) {

  2124.                         lsf_sf_expand(slen, sf, 6, 6, 0);

  2125.                         tindex2 = 3;

  2126.                     } else if (sf < 244) {

  2127.                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);

  2128.                         tindex2 = 4;

  2129.                     } else {

  2130.                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);

  2131.                         tindex2 = 5;

  2132.                     }

  2133.                 } else {

  2134.                     /* normal case */

  2135.                     if (sf < 400) {

  2136.                         lsf_sf_expand(slen, sf, 5, 4, 4);

  2137.                         tindex2 = 0;

  2138.                     } else if (sf < 500) {

  2139.                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);

  2140.                         tindex2 = 1;

  2141.                     } else {

  2142.                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);

  2143.                         tindex2 = 2;

  2144.                         g->preflag = 1;

  2145.                     }

  2146.                 }

  2147. 

  2148.                 j = 0;

  2149.                 for(k=0;k<4;k++) {

  2150.                     n = lsf_nsf_table[tindex2][tindex][k];

  2151.                     sl = slen[k];

  2152.                     if(sl){

  2153.                         for(i=0;i<n;i++)

  2154.                             g->scale_factors[j++] = get_bits(&s->gb, sl);

  2155.                     }else{

  2156.                         for(i=0;i<n;i++)

  2157.                             g->scale_factors[j++] = 0;

  2158.                     }

  2159.                 }

  2160.                 /* XXX: should compute exact size */

  2161.                 for(;j<40;j++)

  2162.                     g->scale_factors[j] = 0;

  2163.             }

  2164. 

  2165.             exponents_from_scale_factors(s, g, exponents);

  2166. 

  2167.             /* read Huffman coded residue */

  2168.             huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);

  2169.         } /* ch */

  2170. 

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

  2172.             compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);

  2173. 

  2174.         for(ch=0;ch<s->nb_channels;ch++) {

  2175.             g = &s->granules[ch][gr];

  2176. 

  2177.             reorder_block(s, g);

  2178.             compute_antialias(s, g);

  2179.             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);

  2180.         }

  2181.     } /* gr */

  2182.     if(get_bits_count(&s->gb)<0)

  2183.         skip_bits_long(&s->gb, -get_bits_count(&s->gb));

  2184.     return nb_granules * 18;

  2185. }

  2186. 

  2187. static int mp_decode_frame(MPADecodeContext *s,

  2188.                            OUT_INT *samples, const uint8_t *buf, int buf_size)

  2189. {

  2190.     int i, nb_frames, ch;

  2191.     OUT_INT *samples_ptr;

  2192. 

  2193.     init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);

  2194. 

  2195.     /* skip error protection field */

  2196.     if (s->error_protection)

  2197.         skip_bits(&s->gb, 16);

  2198. 

  2199.     dprintf(s->avctx, "frame %d:\n", s->frame_count);

  2200.     switch(s->layer) {

  2201.     case 1:

  2202.         s->avctx->frame_size = 384;

  2203.         nb_frames = mp_decode_layer1(s);

  2204.         break;

  2205.     case 2:

  2206.         s->avctx->frame_size = 1152;

  2207.         nb_frames = mp_decode_layer2(s);

  2208.         break;

  2209.     case 3:

  2210.         s->avctx->frame_size = s->lsf ? 576 : 1152;

  2211.     default:

  2212.         nb_frames = mp_decode_layer3(s);

  2213. 

  2214.         s->last_buf_size=0;

  2215.         if(s->in_gb.buffer){

  2216.             align_get_bits(&s->gb);

  2217.             i= get_bits_left(&s->gb)>>3;

  2218.             if(i >= 0 && i <= BACKSTEP_SIZE){

  2219.                 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);

  2220.                 s->last_buf_size=i;

  2221.             }else

  2222.                 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);

  2223.             s->gb= s->in_gb;

  2224.             s->in_gb.buffer= NULL;

  2225.         }

  2226. 

  2227.         align_get_bits(&s->gb);

  2228.         assert((get_bits_count(&s->gb) & 7) == 0);

  2229.         i= get_bits_left(&s->gb)>>3;

  2230. 

  2231.         if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){

  2232.             if(i<0)

  2233.                 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);

  2234.             i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);

  2235.         }

  2236.         assert(i <= buf_size - HEADER_SIZE && i>= 0);

  2237.         memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);

  2238.         s->last_buf_size += i;

  2239. 

  2240.         break;

  2241.     }

  2242. 

  2243.     /* apply the synthesis filter */

  2244.     for(ch=0;ch<s->nb_channels;ch++) {

  2245.         samples_ptr = samples + ch;

  2246.         for(i=0;i<nb_frames;i++) {

  2247.             RENAME(ff_mpa_synth_filter)(

  2248. #if CONFIG_FLOAT

  2249.                          s,

  2250. #endif

  2251.                          s->synth_buf[ch], &(s->synth_buf_offset[ch]),

  2252.                          RENAME(ff_mpa_synth_window), &s->dither_state,

  2253.                          samples_ptr, s->nb_channels,

  2254.                          s->sb_samples[ch][i]);

  2255.             samples_ptr += 32 * s->nb_channels;

  2256.         }

  2257.     }

  2258. 

  2259.     return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;

  2260. }

  2261. 

  2262. static int decode_frame(AVCodecContext * avctx,

  2263.                         void *data, int *data_size,

  2264.                         AVPacket *avpkt)

  2265. {

  2266.     const uint8_t *buf = avpkt->data;

  2267.     int buf_size = avpkt->size;

  2268.     MPADecodeContext *s = avctx->priv_data;

  2269.     uint32_t header;

  2270.     int out_size;

  2271.     OUT_INT *out_samples = data;

  2272. 

  2273.     if(buf_size < HEADER_SIZE)

  2274.         return -1;

  2275. 

  2276.     header = AV_RB32(buf);

  2277.     if(ff_mpa_check_header(header) < 0){

  2278.         av_log(avctx, AV_LOG_ERROR, "Header missing\n");

  2279.         return -1;

  2280.     }

  2281. 

  2282.     if (ff_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {

  2283.         /* free format: prepare to compute frame size */

  2284.         s->frame_size = -1;

  2285.         return -1;

  2286.     }

  2287.     /* update codec info */

  2288.     avctx->channels = s->nb_channels;

  2289.     avctx->bit_rate = s->bit_rate;

  2290.     avctx->sub_id = s->layer;

  2291. 

  2292.     if(*data_size < 1152*avctx->channels*sizeof(OUT_INT))

  2293.         return -1;

  2294.     *data_size = 0;

  2295. 

  2296.     if(s->frame_size<=0 || s->frame_size > buf_size){

  2297.         av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");

  2298.         return -1;

  2299.     }else if(s->frame_size < buf_size){

  2300.         av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");

  2301.         buf_size= s->frame_size;

  2302.     }

  2303. 

  2304.     out_size = mp_decode_frame(s, out_samples, buf, buf_size);

  2305.     if(out_size>=0){

  2306.         *data_size = out_size;

  2307.         avctx->sample_rate = s->sample_rate;

  2308.         //FIXME maybe move the other codec info stuff from above here too

  2309.     }else

  2310.         av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed

  2311.     s->frame_size = 0;

  2312.     return buf_size;

  2313. }

  2314. 

  2315. static void flush(AVCodecContext *avctx){

  2316.     MPADecodeContext *s = avctx->priv_data;

  2317.     memset(s->synth_buf, 0, sizeof(s->synth_buf));

  2318.     s->last_buf_size= 0;

  2319. }

  2320. 

  2321. #if CONFIG_MP3ADU_DECODER

  2322. static int decode_frame_adu(AVCodecContext * avctx,

  2323.                         void *data, int *data_size,

  2324.                         AVPacket *avpkt)

  2325. {

  2326.     const uint8_t *buf = avpkt->data;

  2327.     int buf_size = avpkt->size;

  2328.     MPADecodeContext *s = avctx->priv_data;

  2329.     uint32_t header;

  2330.     int len, out_size;

  2331.     OUT_INT *out_samples = data;

  2332. 

  2333.     len = buf_size;

  2334. 

  2335.     // Discard too short frames

  2336.     if (buf_size < HEADER_SIZE) {

  2337.         *data_size = 0;

  2338.         return buf_size;

  2339.     }

  2340. 

  2341. 

  2342.     if (len > MPA_MAX_CODED_FRAME_SIZE)

  2343.         len = MPA_MAX_CODED_FRAME_SIZE;

  2344. 

  2345.     // Get header and restore sync word

  2346.     header = AV_RB32(buf) | 0xffe00000;

  2347. 

  2348.     if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame

  2349.         *data_size = 0;

  2350.         return buf_size;

  2351.     }

  2352. 

  2353.     ff_mpegaudio_decode_header((MPADecodeHeader *)s, header);

  2354.     /* update codec info */

  2355.     avctx->sample_rate = s->sample_rate;

  2356.     avctx->channels = s->nb_channels;

  2357.     avctx->bit_rate = s->bit_rate;

  2358.     avctx->sub_id = s->layer;

  2359. 

  2360.     s->frame_size = len;

  2361. 

  2362.     if (avctx->parse_only) {

  2363.         out_size = buf_size;

  2364.     } else {

  2365.         out_size = mp_decode_frame(s, out_samples, buf, buf_size);

  2366.     }

  2367. 

  2368.     *data_size = out_size;

  2369.     return buf_size;

  2370. }

  2371. #endif /* CONFIG_MP3ADU_DECODER */

  2372. 

  2373. #if CONFIG_MP3ON4_DECODER

  2374. 

  2375. /**

  2376.  * Context for MP3On4 decoder

  2377.  */

  2378. typedef struct MP3On4DecodeContext {

  2379.     int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)

  2380.     int syncword; ///< syncword patch

  2381.     const uint8_t *coff; ///< channels offsets in output buffer

  2382.     MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance

  2383. } MP3On4DecodeContext;

  2384. 

  2385. #include "mpeg4audio.h"

  2386. 

  2387. /* Next 3 arrays are indexed by channel config number (passed via codecdata) */

  2388. static const uint8_t mp3Frames[8] = {0,1,1,2,3,3,4,5};   /* number of mp3 decoder instances */

  2389. /* offsets into output buffer, assume output order is FL FR BL BR C LFE */

  2390. static const uint8_t chan_offset[8][5] = {

  2391.     {0},

  2392.     {0},            // C

  2393.     {0},            // FLR

  2394.     {2,0},          // C FLR

  2395.     {2,0,3},        // C FLR BS

  2396.     {4,0,2},        // C FLR BLRS

  2397.     {4,0,2,5},      // C FLR BLRS LFE

  2398.     {4,0,2,6,5},    // C FLR BLRS BLR LFE

  2399. };

  2400. 

  2401. 

  2402. static int decode_init_mp3on4(AVCodecContext * avctx)

  2403. {

  2404.     MP3On4DecodeContext *s = avctx->priv_data;

  2405.     MPEG4AudioConfig cfg;

  2406.     int i;

  2407. 

  2408.     if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {

  2409.         av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");

  2410.         return -1;

  2411.     }

  2412. 

  2413.     ff_mpeg4audio_get_config(&cfg, avctx->extradata, avctx->extradata_size);

  2414.     if (!cfg.chan_config || cfg.chan_config > 7) {

  2415.         av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");

  2416.         return -1;

  2417.     }

  2418.     s->frames = mp3Frames[cfg.chan_config];

  2419.     s->coff = chan_offset[cfg.chan_config];

  2420.     avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];

  2421. 

  2422.     if (cfg.sample_rate < 16000)

  2423.         s->syncword = 0xffe00000;

  2424.     else

  2425.         s->syncword = 0xfff00000;

  2426. 

  2427.     /* Init the first mp3 decoder in standard way, so that all tables get builded

  2428.      * We replace avctx->priv_data with the context of the first decoder so that

  2429.      * decode_init() does not have to be changed.

  2430.      * Other decoders will be initialized here copying data from the first context

  2431.      */

  2432.     // Allocate zeroed memory for the first decoder context

  2433.     s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));

  2434.     // Put decoder context in place to make init_decode() happy

  2435.     avctx->priv_data = s->mp3decctx[0];

  2436.     decode_init(avctx);

  2437.     // Restore mp3on4 context pointer

  2438.     avctx->priv_data = s;

  2439.     s->mp3decctx[0]->adu_mode = 1; // Set adu mode

  2440. 

  2441.     /* Create a separate codec/context for each frame (first is already ok).

  2442.      * Each frame is 1 or 2 channels - up to 5 frames allowed

  2443.      */

  2444.     for (i = 1; i < s->frames; i++) {

  2445.         s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));

  2446.         s->mp3decctx[i]->adu_mode = 1;

  2447.         s->mp3decctx[i]->avctx = avctx;

  2448.     }

  2449. 

  2450.     return 0;

  2451. }

  2452. 

  2453. 

  2454. static av_cold int decode_close_mp3on4(AVCodecContext * avctx)

  2455. {

  2456.     MP3On4DecodeContext *s = avctx->priv_data;

  2457.     int i;

  2458. 

  2459.     for (i = 0; i < s->frames; i++)

  2460.         if (s->mp3decctx[i])

  2461.             av_free(s->mp3decctx[i]);

  2462. 

  2463.     return 0;

  2464. }

  2465. 

  2466. 

  2467. static int decode_frame_mp3on4(AVCodecContext * avctx,

  2468.                         void *data, int *data_size,

  2469.                         AVPacket *avpkt)

  2470. {

  2471.     const uint8_t *buf = avpkt->data;

  2472.     int buf_size = avpkt->size;

  2473.     MP3On4DecodeContext *s = avctx->priv_data;

  2474.     MPADecodeContext *m;

  2475.     int fsize, len = buf_size, out_size = 0;

  2476.     uint32_t header;

  2477.     OUT_INT *out_samples = data;

  2478.     OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];

  2479.     OUT_INT *outptr, *bp;

  2480.     int fr, j, n;

  2481. 

  2482.     if(*data_size < MPA_FRAME_SIZE * MPA_MAX_CHANNELS * s->frames * sizeof(OUT_INT))

  2483.         return -1;

  2484. 

  2485.     *data_size = 0;

  2486.     // Discard too short frames

  2487.     if (buf_size < HEADER_SIZE)

  2488.         return -1;

  2489. 

  2490.     // If only one decoder interleave is not needed

  2491.     outptr = s->frames == 1 ? out_samples : decoded_buf;

  2492. 

  2493.     avctx->bit_rate = 0;

  2494. 

  2495.     for (fr = 0; fr < s->frames; fr++) {

  2496.         fsize = AV_RB16(buf) >> 4;

  2497.         fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);

  2498.         m = s->mp3decctx[fr];

  2499.         assert (m != NULL);

  2500. 

  2501.         header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header

  2502. 

  2503.         if (ff_mpa_check_header(header) < 0) // Bad header, discard block

  2504.             break;

  2505. 

  2506.         ff_mpegaudio_decode_header((MPADecodeHeader *)m, header);

  2507.         out_size += mp_decode_frame(m, outptr, buf, fsize);

  2508.         buf += fsize;

  2509.         len -= fsize;

  2510. 

  2511.         if(s->frames > 1) {

  2512.             n = m->avctx->frame_size*m->nb_channels;

  2513.             /* interleave output data */

  2514.             bp = out_samples + s->coff[fr];

  2515.             if(m->nb_channels == 1) {

  2516.                 for(j = 0; j < n; j++) {

  2517.                     *bp = decoded_buf[j];

  2518.                     bp += avctx->channels;

  2519.                 }

  2520.             } else {

  2521.                 for(j = 0; j < n; j++) {

  2522.                     bp[0] = decoded_buf[j++];

  2523.                     bp[1] = decoded_buf[j];

  2524.                     bp += avctx->channels;

  2525.                 }

  2526.             }

  2527.         }

  2528.         avctx->bit_rate += m->bit_rate;

  2529.     }

  2530. 

  2531.     /* update codec info */

  2532.     avctx->sample_rate = s->mp3decctx[0]->sample_rate;

  2533. 

  2534.     *data_size = out_size;

  2535.     return buf_size;

  2536. }

  2537. #endif /* CONFIG_MP3ON4_DECODER */

  2538. 

  2539. #if !CONFIG_FLOAT

  2540. #if CONFIG_MP1_DECODER

  2541. AVCodec mp1_decoder =

  2542. {

  2543.     "mp1",

  2544.     AVMEDIA_TYPE_AUDIO,

  2545.     CODEC_ID_MP1,

  2546.     sizeof(MPADecodeContext),

  2547.     decode_init,

  2548.     NULL,

  2549.     NULL,

  2550.     decode_frame,

  2551.     CODEC_CAP_PARSE_ONLY,

  2552.     .flush= flush,

  2553.     .long_name= NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),

  2554. };

  2555. #endif

  2556. #if CONFIG_MP2_DECODER

  2557. AVCodec mp2_decoder =

  2558. {

  2559.     "mp2",

  2560.     AVMEDIA_TYPE_AUDIO,

  2561.     CODEC_ID_MP2,

  2562.     sizeof(MPADecodeContext),

  2563.     decode_init,

  2564.     NULL,

  2565.     NULL,

  2566.     decode_frame,

  2567.     CODEC_CAP_PARSE_ONLY,

  2568.     .flush= flush,

  2569.     .long_name= NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),

  2570. };

  2571. #endif

  2572. #if CONFIG_MP3_DECODER

  2573. AVCodec mp3_decoder =

  2574. {

  2575.     "mp3",

  2576.     AVMEDIA_TYPE_AUDIO,

  2577.     CODEC_ID_MP3,

  2578.     sizeof(MPADecodeContext),

  2579.     decode_init,

  2580.     NULL,

  2581.     NULL,

  2582.     decode_frame,

  2583.     CODEC_CAP_PARSE_ONLY,

  2584.     .flush= flush,

  2585.     .long_name= NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),

  2586. };

  2587. #endif

  2588. #if CONFIG_MP3ADU_DECODER

  2589. AVCodec mp3adu_decoder =

  2590. {

  2591.     "mp3adu",

  2592.     AVMEDIA_TYPE_AUDIO,

  2593.     CODEC_ID_MP3ADU,

  2594.     sizeof(MPADecodeContext),

  2595.     decode_init,

  2596.     NULL,

  2597.     NULL,

  2598.     decode_frame_adu,

  2599.     CODEC_CAP_PARSE_ONLY,

  2600.     .flush= flush,

  2601.     .long_name= NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),

  2602. };

  2603. #endif

  2604. #if CONFIG_MP3ON4_DECODER

  2605. AVCodec mp3on4_decoder =

  2606. {

  2607.     "mp3on4",

  2608.     AVMEDIA_TYPE_AUDIO,

  2609.     CODEC_ID_MP3ON4,

  2610.     sizeof(MP3On4DecodeContext),

  2611.     decode_init_mp3on4,

  2612.     NULL,

  2613.     decode_close_mp3on4,

  2614.     decode_frame_mp3on4,

  2615.     .flush= flush,

  2616.     .long_name= NULL_IF_CONFIG_SMALL("MP3onMP4"),

  2617. };

  2618. #endif

  2619. #endif