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.
24009 Commits
32 Branches
2.4GB
Tree: 495af353db
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '495af353db'
${ noResults }
FFmpeg/libavcodec/mpegaudiodec.c

2562 lines
77KB
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. /* WARNING: only correct for posititive numbers */

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

  44. 

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

  46. 

  47. /****************/

  48. 

  49. #define HEADER_SIZE 4

  50. 

  51. #include "mpegaudiodata.h"

  52. #include "mpegaudiodectab.h"

  53. 

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

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

  56. 

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

  58. static VLC huff_vlc[16];

  59. static VLC_TYPE huff_vlc_tables[

  60.   0+128+128+128+130+128+154+166+

  61.   142+204+190+170+542+460+662+414

  62.   ][2];

  63. static const int huff_vlc_tables_sizes[16] = {

  64.   0, 128, 128, 128, 130, 128, 154, 166,

  65.   142, 204, 190, 170, 542, 460, 662, 414

  66. };

  67. static VLC huff_quad_vlc[2];

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

  69. static const int huff_quad_vlc_tables_sizes[2] = {

  70.   128, 16

  71. };

  72. /* computed from band_size_long */

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

  74. #include "mpegaudio_tablegen.h"

  75. /* intensity stereo coef table */

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

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

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

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

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

  81. 

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

  83. static uint16_t scale_factor_modshift[64];

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

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

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

  87. 

  88. #define SCALE_GEN(v) \

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

  90. 

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

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

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

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

  95. };

  96. 

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

  98. 

  99. /**

  100.  * Convert region offsets to region sizes and truncate

  101.  * size to big_values.

  102.  */

  103. static void ff_region_offset2size(GranuleDef *g){

  104.     int i, k, j=0;

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

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

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

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

  109.         j = k;

  110.     }

  111. }

  112. 

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

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

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

  116.     else {

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

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

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

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

  121.         else

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

  123.     }

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

  125. }

  126. 

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

  128.     int l;

  129.     g->region_size[0] =

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

  131.     /* should not overflow */

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

  133.     g->region_size[1] =

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

  135. }

  136. 

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

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

  139.         if (g->switch_point) {

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

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

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

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

  144.                 g->long_end = 8;

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

  146.                 g->long_end = 6;

  147.             else

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

  149. 

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

  151.         } else {

  152.             g->long_end = 0;

  153.             g->short_start = 0;

  154.         }

  155.     } else {

  156.         g->short_start = 13;

  157.         g->long_end = 22;

  158.     }

  159. }

  160. 

  161. /* layer 1 unscaling */

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

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

  164. {

  165.     int shift, mod;

  166.     int64_t val;

  167. 

  168.     shift = scale_factor_modshift[scale_factor];

  169.     mod = shift & 3;

  170.     shift >>= 2;

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

  172.     shift += n;

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

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

  175. }

  176. 

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

  178. {

  179.     int shift, mod, val;

  180. 

  181.     shift = scale_factor_modshift[scale_factor];

  182.     mod = shift & 3;

  183.     shift >>= 2;

  184. 

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

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

  187.     if (shift > 0)

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

  189.     return val;

  190. }

  191. 

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

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

  194. {

  195.     unsigned int m;

  196.     int e;

  197. 

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

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

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

  201.     assert(e>=1);

  202.     if (e > 31)

  203.         return 0;

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

  205. 

  206.     return m;

  207. }

  208. 

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

  210. #define DEV_ORDER 13

  211. 

  212. #define POW_FRAC_BITS 24

  213. #define POW_FRAC_ONE    (1 << POW_FRAC_BITS)

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

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

  216. 

  217. static int dev_4_3_coefs[DEV_ORDER];

  218. 

  219. #if 0 /* unused */

  220. static int pow_mult3[3] = {

  221.     POW_FIX(1.0),

  222.     POW_FIX(1.25992104989487316476),

  223.     POW_FIX(1.58740105196819947474),

  224. };

  225. #endif

  226. 

  227. static av_cold void int_pow_init(void)

  228. {

  229.     int i, a;

  230. 

  231.     a = POW_FIX(1.0);

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

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

  234.         dev_4_3_coefs[i] = a;

  235.     }

  236. }

  237. 

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

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

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

  241. {

  242.     int e, er, eq, j;

  243.     int a, a1;

  244. 

  245.     /* renormalize */

  246.     a = i;

  247.     e = POW_FRAC_BITS;

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

  249.         a = a << 1;

  250.         e--;

  251.     }

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

  253.     a1 = 0;

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

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

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

  257.     /* exponent compute (exact) */

  258.     e = e * 4;

  259.     er = e % 3;

  260.     eq = e / 3;

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

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

  263.         a = a >> 1;

  264.         eq++;

  265.     }

  266.     /* convert to float */

  267.     while (a < POW_FRAC_ONE) {

  268.         a = a << 1;

  269.         eq--;

  270.     }

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

  272. #if POW_FRAC_BITS > FRAC_BITS

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

  274.     /* correct overflow */

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

  276.         a = a >> 1;

  277.         eq++;

  278.     }

  279. #endif

  280.     *exp_ptr = eq;

  281.     return a;

  282. }

  283. #endif

  284. 

  285. static av_cold int decode_init(AVCodecContext * avctx)

  286. {

  287.     MPADecodeContext *s = avctx->priv_data;

  288.     static int init=0;

  289.     int i, j, k;

  290. 

  291.     s->avctx = avctx;

  292. 

  293.     avctx->sample_fmt= OUT_FMT;

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

  295. 

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

  297.         s->compute_antialias= compute_antialias_integer;

  298.     else

  299.         s->compute_antialias= compute_antialias_float;

  300. 

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

  302.         int offset;

  303. 

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

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

  306.             int shift, mod;

  307.             /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

  308.             shift = (i / 3);

  309.             mod = i % 3;

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

  311.         }

  312. 

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

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

  315.             int n, norm;

  316.             n = i + 2;

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

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

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

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

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

  322.                     i, norm,

  323.                     scale_factor_mult[i][0],

  324.                     scale_factor_mult[i][1],

  325.                     scale_factor_mult[i][2]);

  326.         }

  327. 

  328.         ff_mpa_synth_init(ff_mpa_synth_window);

  329. 

  330.         /* huffman decode tables */

  331.         offset = 0;

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

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

  334.             int xsize, x, y;

  335.             uint8_t  tmp_bits [512];

  336.             uint16_t tmp_codes[512];

  337. 

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

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

  340. 

  341.             xsize = h->xsize;

  342. 

  343.             j = 0;

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

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

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

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

  348.                 }

  349.             }

  350. 

  351.             /* XXX: fail test */

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

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

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

  355.                      tmp_bits, 1, 1, tmp_codes, 2, 2,

  356.                      INIT_VLC_USE_NEW_STATIC);

  357.             offset += huff_vlc_tables_sizes[i];

  358.         }

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

  360. 

  361.         offset = 0;

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

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

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

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

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

  367.                      INIT_VLC_USE_NEW_STATIC);

  368.             offset += huff_quad_vlc_tables_sizes[i];

  369.         }

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

  371. 

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

  373.             k = 0;

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

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

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

  377.             }

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

  379.         }

  380. 

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

  382. 

  383.         int_pow_init();

  384.         mpegaudio_tableinit();

  385. 

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

  387.             float f;

  388.             int v;

  389.             if (i != 6) {

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

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

  392.             } else {

  393.                 v = FIXR(1.0);

  394.             }

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

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

  397.         }

  398.         /* invalid values */

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

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

  401. 

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

  403.             double f;

  404.             int e, k;

  405. 

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

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

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

  409.                 k = i & 1;

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

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

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

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

  414.             }

  415.         }

  416. 

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

  418.             float ci, cs, ca;

  419.             ci = ci_table[i];

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

  421.             ca = cs * ci;

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

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

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

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

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

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

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

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

  430.         }

  431. 

  432.         /* compute mdct windows */

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

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

  435.                 double d;

  436. 

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

  438.                     continue;

  439. 

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

  441.                 if(j==1){

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

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

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

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

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

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

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

  449.                 }

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

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

  452. 

  453.                 if(j==2)

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

  455.                 else

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

  457.             }

  458.         }

  459. 

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

  461.            the sign of the right window coefs */

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

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

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

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

  466.             }

  467.         }

  468. 

  469.         init = 1;

  470.     }

  471. 

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

  473.         s->adu_mode = 1;

  474.     return 0;

  475. }

  476. 

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

  478. 

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

  480. 

  481. #define COS0_0  FIXHR(0.50060299823519630134/2)

  482. #define COS0_1  FIXHR(0.50547095989754365998/2)

  483. #define COS0_2  FIXHR(0.51544730992262454697/2)

  484. #define COS0_3  FIXHR(0.53104259108978417447/2)

  485. #define COS0_4  FIXHR(0.55310389603444452782/2)

  486. #define COS0_5  FIXHR(0.58293496820613387367/2)

  487. #define COS0_6  FIXHR(0.62250412303566481615/2)

  488. #define COS0_7  FIXHR(0.67480834145500574602/2)

  489. #define COS0_8  FIXHR(0.74453627100229844977/2)

  490. #define COS0_9  FIXHR(0.83934964541552703873/2)

  491. #define COS0_10 FIXHR(0.97256823786196069369/2)

  492. #define COS0_11 FIXHR(1.16943993343288495515/4)

  493. #define COS0_12 FIXHR(1.48416461631416627724/4)

  494. #define COS0_13 FIXHR(2.05778100995341155085/8)

  495. #define COS0_14 FIXHR(3.40760841846871878570/8)

  496. #define COS0_15 FIXHR(10.19000812354805681150/32)

  497. 

  498. #define COS1_0 FIXHR(0.50241928618815570551/2)

  499. #define COS1_1 FIXHR(0.52249861493968888062/2)

  500. #define COS1_2 FIXHR(0.56694403481635770368/2)

  501. #define COS1_3 FIXHR(0.64682178335999012954/2)

  502. #define COS1_4 FIXHR(0.78815462345125022473/2)

  503. #define COS1_5 FIXHR(1.06067768599034747134/4)

  504. #define COS1_6 FIXHR(1.72244709823833392782/4)

  505. #define COS1_7 FIXHR(5.10114861868916385802/16)

  506. 

  507. #define COS2_0 FIXHR(0.50979557910415916894/2)

  508. #define COS2_1 FIXHR(0.60134488693504528054/2)

  509. #define COS2_2 FIXHR(0.89997622313641570463/2)

  510. #define COS2_3 FIXHR(2.56291544774150617881/8)

  511. 

  512. #define COS3_0 FIXHR(0.54119610014619698439/2)

  513. #define COS3_1 FIXHR(1.30656296487637652785/4)

  514. 

  515. #define COS4_0 FIXHR(0.70710678118654752439/2)

  516. 

  517. /* butterfly operator */

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

  519. {\

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

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

  522.     tab[a] = tmp0;\

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

  524. }

  525. 

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

  527. {\

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

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

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

  531. }

  532. 

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

  534. {\

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

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

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

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

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

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

  541. }

  542. 

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

  544. 

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

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

  547. {

  548.     int tmp0, tmp1;

  549. 

  550.     /* pass 1 */

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

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

  553.     /* pass 2 */

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

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

  556.     /* pass 1 */

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

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

  559.     /* pass 2 */

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

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

  562.     /* pass 3 */

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

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

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

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

  567.     /* pass 1 */

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

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

  570.     /* pass 2 */

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

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

  573.     /* pass 1 */

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

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

  576.     /* pass 2 */

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

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

  579.     /* pass 3 */

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

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

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

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

  584.     /* pass 4 */

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

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

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

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

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

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

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

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

  593. 

  594. 

  595. 

  596.     /* pass 1 */

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

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

  599.     /* pass 2 */

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

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

  602.     /* pass 1 */

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

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

  605.     /* pass 2 */

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

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

  608.     /* pass 3 */

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

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

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

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

  613. 

  614.     /* pass 1 */

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

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

  617.     /* pass 2 */

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

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

  620.     /* pass 1 */

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

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

  623.     /* pass 2 */

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

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

  626.     /* pass 3 */

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

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

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

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

  631.     /* pass 4 */

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

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

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

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

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

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

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

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

  640. 

  641.     /* pass 5 */

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

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

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

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

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

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

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

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

  650. 

  651.     /* pass 6 */

  652. 

  653.     ADD( 8, 12);

  654.     ADD(12, 10);

  655.     ADD(10, 14);

  656.     ADD(14,  9);

  657.     ADD( 9, 13);

  658.     ADD(13, 11);

  659.     ADD(11, 15);

  660. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  677. 

  678.     ADD(24, 28);

  679.     ADD(28, 26);

  680.     ADD(26, 30);

  681.     ADD(30, 25);

  682.     ADD(25, 29);

  683.     ADD(29, 27);

  684.     ADD(27, 31);

  685. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  702. }

  703. 

  704. #if FRAC_BITS <= 15

  705. 

  706. static inline int round_sample(int *sum)

  707. {

  708.     int sum1;

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

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

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

  712. }

  713. 

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

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

  716. 

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

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

  719. 

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

  721. 

  722. #else

  723. 

  724. static inline int round_sample(int64_t *sum)

  725. {

  726.     int sum1;

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

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

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

  730. }

  731. 

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

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

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

  735. #endif

  736. 

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

  738. {                                         \

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

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

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

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

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

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

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

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

  747. }

  748. 

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

  750. {                                               \

  751.     int tmp;\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  776. }

  777. 

  778. void av_cold ff_mpa_synth_init(MPA_INT *window)

  779. {

  780.     int i;

  781. 

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

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

  784.         int v;

  785.         v = ff_mpa_enwindow[i];

  786. #if WFRAC_BITS < 16

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

  788. #endif

  789.         window[i] = v;

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

  791.             v = -v;

  792.         if (i != 0)

  793.             window[512 - i] = v;

  794.     }

  795. }

  796. 

  797. /* 32 sub band synthesis filter. Input: 32 sub band samples, Output:

  798.    32 samples. */

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

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

  801.                          MPA_INT *window, int *dither_state,

  802.                          OUT_INT *samples, int incr,

  803.                          int32_t sb_samples[SBLIMIT])

  804. {

  805.     register MPA_INT *synth_buf;

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

  807.     int j, offset;

  808.     OUT_INT *samples2;

  809. #if FRAC_BITS <= 15

  810.     int32_t tmp[32];

  811.     int sum, sum2;

  812. #else

  813.     int64_t sum, sum2;

  814. #endif

  815. 

  816.     offset = *synth_buf_offset;

  817.     synth_buf = synth_buf_ptr + offset;

  818. 

  819. #if FRAC_BITS <= 15

  820.     dct32(tmp, sb_samples);

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

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

  823.            sound */

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

  825.     }

  826. #else

  827.     dct32(synth_buf, sb_samples);

  828. #endif

  829. 

  830.     /* copy to avoid wrap */

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

  832. 

  833.     samples2 = samples + 31 * incr;

  834.     w = window;

  835.     w2 = window + 31;

  836. 

  837.     sum = *dither_state;

  838.     p = synth_buf + 16;

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

  840.     p = synth_buf + 48;

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

  842.     *samples = round_sample(&sum);

  843.     samples += incr;

  844.     w++;

  845. 

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

  847.        access per two sample */

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

  849.         sum2 = 0;

  850.         p = synth_buf + 16 + j;

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

  852.         p = synth_buf + 48 - j;

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

  854. 

  855.         *samples = round_sample(&sum);

  856.         samples += incr;

  857.         sum += sum2;

  858.         *samples2 = round_sample(&sum);

  859.         samples2 -= incr;

  860.         w++;

  861.         w2--;

  862.     }

  863. 

  864.     p = synth_buf + 32;

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

  866.     *samples = round_sample(&sum);

  867.     *dither_state= sum;

  868. 

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

  870.     *synth_buf_offset = offset;

  871. }

  872. 

  873. #define C3 FIXHR(0.86602540378443864676/2)

  874. 

  875. /* 0.5 / cos(pi*(2*i+1)/36) */

  876. static const int icos36[9] = {

  877.     FIXR(0.50190991877167369479),

  878.     FIXR(0.51763809020504152469), //0

  879.     FIXR(0.55168895948124587824),

  880.     FIXR(0.61038729438072803416),

  881.     FIXR(0.70710678118654752439), //1

  882.     FIXR(0.87172339781054900991),

  883.     FIXR(1.18310079157624925896),

  884.     FIXR(1.93185165257813657349), //2

  885.     FIXR(5.73685662283492756461),

  886. };

  887. 

  888. /* 0.5 / cos(pi*(2*i+1)/36) */

  889. static const int icos36h[9] = {

  890.     FIXHR(0.50190991877167369479/2),

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

  892.     FIXHR(0.55168895948124587824/2),

  893.     FIXHR(0.61038729438072803416/2),

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

  895.     FIXHR(0.87172339781054900991/2),

  896.     FIXHR(1.18310079157624925896/4),

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

  898. //    FIXHR(5.73685662283492756461),

  899. };

  900. 

  901. /* 12 points IMDCT. We compute it "by hand" by factorizing obvious

  902.    cases. */

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

  904. {

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

  906. 

  907.     in0= in[0*3];

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

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

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

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

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

  913.     in5 += in3;

  914.     in3 += in1;

  915. 

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

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

  918. 

  919.     t1 = in0 - in4;

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

  921. 

  922.     out[ 7]=

  923.     out[10]= t1 + t2;

  924.     out[ 1]=

  925.     out[ 4]= t1 - t2;

  926. 

  927.     in0 += in4>>1;

  928.     in4 = in0 + in2;

  929.     in5 += 2*in1;

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

  931.     out[ 8]=

  932.     out[ 9]= in4 + in1;

  933.     out[ 2]=

  934.     out[ 3]= in4 - in1;

  935. 

  936.     in0 -= in2;

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

  938.     out[ 0]=

  939.     out[ 5]= in0 - in5;

  940.     out[ 6]=

  941.     out[11]= in0 + in5;

  942. }

  943. 

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

  945. #define C1 FIXHR(0.98480775301220805936/2)

  946. #define C2 FIXHR(0.93969262078590838405/2)

  947. #define C3 FIXHR(0.86602540378443864676/2)

  948. #define C4 FIXHR(0.76604444311897803520/2)

  949. #define C5 FIXHR(0.64278760968653932632/2)

  950. #define C6 FIXHR(0.5/2)

  951. #define C7 FIXHR(0.34202014332566873304/2)

  952. #define C8 FIXHR(0.17364817766693034885/2)

  953. 

  954. 

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

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

  957. {

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

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

  960. 

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

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

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

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

  965. 

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

  967.         tmp1 = tmp + j;

  968.         in1 = in + j;

  969. #if 0

  970. //more accurate but slower

  971.         int64_t t0, t1, t2, t3;

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

  973. 

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

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

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

  977.         tmp1[16] = t1 + t2;

  978. 

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

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

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

  982. 

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

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

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

  986. 

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

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

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

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

  991. 

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

  993. 

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

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

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

  997. #else

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

  999. 

  1000.         t3 = in1[2*0] + (in1[2*6]>>1);

  1001.         t1 = in1[2*0] - in1[2*6];

  1002.         tmp1[ 6] = t1 - (t2>>1);

  1003.         tmp1[16] = t1 + t2;

  1004. 

  1005.         t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);

  1006.         t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);

  1007.         t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);

  1008. 

  1009.         tmp1[10] = t3 - t0 - t2;

  1010.         tmp1[ 2] = t3 + t0 + t1;

  1011.         tmp1[14] = t3 + t2 - t1;

  1012. 

  1013.         tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);

  1014.         t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);

  1015.         t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);

  1016.         t0 = MULH(2*in1[2*3], C3);

  1017. 

  1018.         t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);

  1019. 

  1020.         tmp1[ 0] = t2 + t3 + t0;

  1021.         tmp1[12] = t2 + t1 - t0;

  1022.         tmp1[ 8] = t3 - t1 - t0;

  1023. #endif

  1024.     }

  1025. 

  1026.     i = 0;

  1027.     for(j=0;j<4;j++) {

  1028.         t0 = tmp[i];

  1029.         t1 = tmp[i + 2];

  1030.         s0 = t1 + t0;

  1031.         s2 = t1 - t0;

  1032. 

  1033.         t2 = tmp[i + 1];

  1034.         t3 = tmp[i + 3];

  1035.         s1 = MULH(2*(t3 + t2), icos36h[j]);

  1036.         s3 = MULL(t3 - t2, icos36[8 - j], FRAC_BITS);

  1037. 

  1038.         t0 = s0 + s1;

  1039.         t1 = s0 - s1;

  1040.         out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];

  1041.         out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];

  1042.         buf[9 + j] = MULH(t0, win[18 + 9 + j]);

  1043.         buf[8 - j] = MULH(t0, win[18 + 8 - j]);

  1044. 

  1045.         t0 = s2 + s3;

  1046.         t1 = s2 - s3;

  1047.         out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];

  1048.         out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];

  1049.         buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);

  1050.         buf[      + j] = MULH(t0, win[18         + j]);

  1051.         i += 4;

  1052.     }

  1053. 

  1054.     s0 = tmp[16];

  1055.     s1 = MULH(2*tmp[17], icos36h[4]);

  1056.     t0 = s0 + s1;

  1057.     t1 = s0 - s1;

  1058.     out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];

  1059.     out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];

  1060.     buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);

  1061.     buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);

  1062. }

  1063. 

  1064. /* return the number of decoded frames */

  1065. static int mp_decode_layer1(MPADecodeContext *s)

  1066. {

  1067.     int bound, i, v, n, ch, j, mant;

  1068.     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];

  1069.     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

  1070. 

  1071.     if (s->mode == MPA_JSTEREO)

  1072.         bound = (s->mode_ext + 1) * 4;

  1073.     else

  1074.         bound = SBLIMIT;

  1075. 

  1076.     /* allocation bits */

  1077.     for(i=0;i<bound;i++) {

  1078.         for(ch=0;ch<s->nb_channels;ch++) {

  1079.             allocation[ch][i] = get_bits(&s->gb, 4);

  1080.         }

  1081.     }

  1082.     for(i=bound;i<SBLIMIT;i++) {

  1083.         allocation[0][i] = get_bits(&s->gb, 4);

  1084.     }

  1085. 

  1086.     /* scale factors */

  1087.     for(i=0;i<bound;i++) {

  1088.         for(ch=0;ch<s->nb_channels;ch++) {

  1089.             if (allocation[ch][i])

  1090.                 scale_factors[ch][i] = get_bits(&s->gb, 6);

  1091.         }

  1092.     }

  1093.     for(i=bound;i<SBLIMIT;i++) {

  1094.         if (allocation[0][i]) {

  1095.             scale_factors[0][i] = get_bits(&s->gb, 6);

  1096.             scale_factors[1][i] = get_bits(&s->gb, 6);

  1097.         }

  1098.     }

  1099. 

  1100.     /* compute samples */

  1101.     for(j=0;j<12;j++) {

  1102.         for(i=0;i<bound;i++) {

  1103.             for(ch=0;ch<s->nb_channels;ch++) {

  1104.                 n = allocation[ch][i];

  1105.                 if (n) {

  1106.                     mant = get_bits(&s->gb, n + 1);

  1107.                     v = l1_unscale(n, mant, scale_factors[ch][i]);

  1108.                 } else {

  1109.                     v = 0;

  1110.                 }

  1111.                 s->sb_samples[ch][j][i] = v;

  1112.             }

  1113.         }

  1114.         for(i=bound;i<SBLIMIT;i++) {

  1115.             n = allocation[0][i];

  1116.             if (n) {

  1117.                 mant = get_bits(&s->gb, n + 1);

  1118.                 v = l1_unscale(n, mant, scale_factors[0][i]);

  1119.                 s->sb_samples[0][j][i] = v;

  1120.                 v = l1_unscale(n, mant, scale_factors[1][i]);

  1121.                 s->sb_samples[1][j][i] = v;

  1122.             } else {

  1123.                 s->sb_samples[0][j][i] = 0;

  1124.                 s->sb_samples[1][j][i] = 0;

  1125.             }

  1126.         }

  1127.     }

  1128.     return 12;

  1129. }

  1130. 

  1131. static int mp_decode_layer2(MPADecodeContext *s)

  1132. {

  1133.     int sblimit; /* number of used subbands */

  1134.     const unsigned char *alloc_table;

  1135.     int table, bit_alloc_bits, i, j, ch, bound, v;

  1136.     unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];

  1137.     unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];

  1138.     unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;

  1139.     int scale, qindex, bits, steps, k, l, m, b;

  1140. 

  1141.     /* select decoding table */

  1142.     table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,

  1143.                             s->sample_rate, s->lsf);

  1144.     sblimit = ff_mpa_sblimit_table[table];

  1145.     alloc_table = ff_mpa_alloc_tables[table];

  1146. 

  1147.     if (s->mode == MPA_JSTEREO)

  1148.         bound = (s->mode_ext + 1) * 4;

  1149.     else

  1150.         bound = sblimit;

  1151. 

  1152.     dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);

  1153. 

  1154.     /* sanity check */

  1155.     if( bound > sblimit ) bound = sblimit;

  1156. 

  1157.     /* parse bit allocation */

  1158.     j = 0;

  1159.     for(i=0;i<bound;i++) {

  1160.         bit_alloc_bits = alloc_table[j];

  1161.         for(ch=0;ch<s->nb_channels;ch++) {

  1162.             bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);

  1163.         }

  1164.         j += 1 << bit_alloc_bits;

  1165.     }

  1166.     for(i=bound;i<sblimit;i++) {

  1167.         bit_alloc_bits = alloc_table[j];

  1168.         v = get_bits(&s->gb, bit_alloc_bits);

  1169.         bit_alloc[0][i] = v;

  1170.         bit_alloc[1][i] = v;

  1171.         j += 1 << bit_alloc_bits;

  1172.     }

  1173. 

  1174.     /* scale codes */

  1175.     for(i=0;i<sblimit;i++) {

  1176.         for(ch=0;ch<s->nb_channels;ch++) {

  1177.             if (bit_alloc[ch][i])

  1178.                 scale_code[ch][i] = get_bits(&s->gb, 2);

  1179.         }

  1180.     }

  1181. 

  1182.     /* scale factors */

  1183.     for(i=0;i<sblimit;i++) {

  1184.         for(ch=0;ch<s->nb_channels;ch++) {

  1185.             if (bit_alloc[ch][i]) {

  1186.                 sf = scale_factors[ch][i];

  1187.                 switch(scale_code[ch][i]) {

  1188.                 default:

  1189.                 case 0:

  1190.                     sf[0] = get_bits(&s->gb, 6);

  1191.                     sf[1] = get_bits(&s->gb, 6);

  1192.                     sf[2] = get_bits(&s->gb, 6);

  1193.                     break;

  1194.                 case 2:

  1195.                     sf[0] = get_bits(&s->gb, 6);

  1196.                     sf[1] = sf[0];

  1197.                     sf[2] = sf[0];

  1198.                     break;

  1199.                 case 1:

  1200.                     sf[0] = get_bits(&s->gb, 6);

  1201.                     sf[2] = get_bits(&s->gb, 6);

  1202.                     sf[1] = sf[0];

  1203.                     break;

  1204.                 case 3:

  1205.                     sf[0] = get_bits(&s->gb, 6);

  1206.                     sf[2] = get_bits(&s->gb, 6);

  1207.                     sf[1] = sf[2];

  1208.                     break;

  1209.                 }

  1210.             }

  1211.         }

  1212.     }

  1213. 

  1214.     /* samples */

  1215.     for(k=0;k<3;k++) {

  1216.         for(l=0;l<12;l+=3) {

  1217.             j = 0;

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

  1219.                 bit_alloc_bits = alloc_table[j];

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

  1221.                     b = bit_alloc[ch][i];

  1222.                     if (b) {

  1223.                         scale = scale_factors[ch][i][k];

  1224.                         qindex = alloc_table[j+b];

  1225.                         bits = ff_mpa_quant_bits[qindex];

  1226.                         if (bits < 0) {

  1227.                             /* 3 values at the same time */

  1228.                             v = get_bits(&s->gb, -bits);

  1229.                             steps = ff_mpa_quant_steps[qindex];

  1230.                             s->sb_samples[ch][k * 12 + l + 0][i] =

  1231.                                 l2_unscale_group(steps, v % steps, scale);

  1232.                             v = v / steps;

  1233.                             s->sb_samples[ch][k * 12 + l + 1][i] =

  1234.                                 l2_unscale_group(steps, v % steps, scale);

  1235.                             v = v / steps;

  1236.                             s->sb_samples[ch][k * 12 + l + 2][i] =

  1237.                                 l2_unscale_group(steps, v, scale);

  1238.                         } else {

  1239.                             for(m=0;m<3;m++) {

  1240.                                 v = get_bits(&s->gb, bits);

  1241.                                 v = l1_unscale(bits - 1, v, scale);

  1242.                                 s->sb_samples[ch][k * 12 + l + m][i] = v;

  1243.                             }

  1244.                         }

  1245.                     } else {

  1246.                         s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1247.                         s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1248.                         s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1249.                     }

  1250.                 }

  1251.                 /* next subband in alloc table */

  1252.                 j += 1 << bit_alloc_bits;

  1253.             }

  1254.             /* XXX: find a way to avoid this duplication of code */

  1255.             for(i=bound;i<sblimit;i++) {

  1256.                 bit_alloc_bits = alloc_table[j];

  1257.                 b = bit_alloc[0][i];

  1258.                 if (b) {

  1259.                     int mant, scale0, scale1;

  1260.                     scale0 = scale_factors[0][i][k];

  1261.                     scale1 = scale_factors[1][i][k];

  1262.                     qindex = alloc_table[j+b];

  1263.                     bits = ff_mpa_quant_bits[qindex];

  1264.                     if (bits < 0) {

  1265.                         /* 3 values at the same time */

  1266.                         v = get_bits(&s->gb, -bits);

  1267.                         steps = ff_mpa_quant_steps[qindex];

  1268.                         mant = v % steps;

  1269.                         v = v / steps;

  1270.                         s->sb_samples[0][k * 12 + l + 0][i] =

  1271.                             l2_unscale_group(steps, mant, scale0);

  1272.                         s->sb_samples[1][k * 12 + l + 0][i] =

  1273.                             l2_unscale_group(steps, mant, scale1);

  1274.                         mant = v % steps;

  1275.                         v = v / steps;

  1276.                         s->sb_samples[0][k * 12 + l + 1][i] =

  1277.                             l2_unscale_group(steps, mant, scale0);

  1278.                         s->sb_samples[1][k * 12 + l + 1][i] =

  1279.                             l2_unscale_group(steps, mant, scale1);

  1280.                         s->sb_samples[0][k * 12 + l + 2][i] =

  1281.                             l2_unscale_group(steps, v, scale0);

  1282.                         s->sb_samples[1][k * 12 + l + 2][i] =

  1283.                             l2_unscale_group(steps, v, scale1);

  1284.                     } else {

  1285.                         for(m=0;m<3;m++) {

  1286.                             mant = get_bits(&s->gb, bits);

  1287.                             s->sb_samples[0][k * 12 + l + m][i] =

  1288.                                 l1_unscale(bits - 1, mant, scale0);

  1289.                             s->sb_samples[1][k * 12 + l + m][i] =

  1290.                                 l1_unscale(bits - 1, mant, scale1);

  1291.                         }

  1292.                     }

  1293.                 } else {

  1294.                     s->sb_samples[0][k * 12 + l + 0][i] = 0;

  1295.                     s->sb_samples[0][k * 12 + l + 1][i] = 0;

  1296.                     s->sb_samples[0][k * 12 + l + 2][i] = 0;

  1297.                     s->sb_samples[1][k * 12 + l + 0][i] = 0;

  1298.                     s->sb_samples[1][k * 12 + l + 1][i] = 0;

  1299.                     s->sb_samples[1][k * 12 + l + 2][i] = 0;

  1300.                 }

  1301.                 /* next subband in alloc table */

  1302.                 j += 1 << bit_alloc_bits;

  1303.             }

  1304.             /* fill remaining samples to zero */

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

  1306.                 for(ch=0;ch<s->nb_channels;ch++) {

  1307.                     s->sb_samples[ch][k * 12 + l + 0][i] = 0;

  1308.                     s->sb_samples[ch][k * 12 + l + 1][i] = 0;

  1309.                     s->sb_samples[ch][k * 12 + l + 2][i] = 0;

  1310.                 }

  1311.             }

  1312.         }

  1313.     }

  1314.     return 3 * 12;

  1315. }

  1316. 

  1317. static inline void lsf_sf_expand(int *slen,

  1318.                                  int sf, int n1, int n2, int n3)

  1319. {

  1320.     if (n3) {

  1321.         slen[3] = sf % n3;

  1322.         sf /= n3;

  1323.     } else {

  1324.         slen[3] = 0;

  1325.     }

  1326.     if (n2) {

  1327.         slen[2] = sf % n2;

  1328.         sf /= n2;

  1329.     } else {

  1330.         slen[2] = 0;

  1331.     }

  1332.     slen[1] = sf % n1;

  1333.     sf /= n1;

  1334.     slen[0] = sf;

  1335. }

  1336. 

  1337. static void exponents_from_scale_factors(MPADecodeContext *s,

  1338.                                          GranuleDef *g,

  1339.                                          int16_t *exponents)

  1340. {

  1341.     const uint8_t *bstab, *pretab;

  1342.     int len, i, j, k, l, v0, shift, gain, gains[3];

  1343.     int16_t *exp_ptr;

  1344. 

  1345.     exp_ptr = exponents;

  1346.     gain = g->global_gain - 210;

  1347.     shift = g->scalefac_scale + 1;

  1348. 

  1349.     bstab = band_size_long[s->sample_rate_index];

  1350.     pretab = mpa_pretab[g->preflag];

  1351.     for(i=0;i<g->long_end;i++) {

  1352.         v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;

  1353.         len = bstab[i];

  1354.         for(j=len;j>0;j--)

  1355.             *exp_ptr++ = v0;

  1356.     }

  1357. 

  1358.     if (g->short_start < 13) {

  1359.         bstab = band_size_short[s->sample_rate_index];

  1360.         gains[0] = gain - (g->subblock_gain[0] << 3);

  1361.         gains[1] = gain - (g->subblock_gain[1] << 3);

  1362.         gains[2] = gain - (g->subblock_gain[2] << 3);

  1363.         k = g->long_end;

  1364.         for(i=g->short_start;i<13;i++) {

  1365.             len = bstab[i];

  1366.             for(l=0;l<3;l++) {

  1367.                 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;

  1368.                 for(j=len;j>0;j--)

  1369.                 *exp_ptr++ = v0;

  1370.             }

  1371.         }

  1372.     }

  1373. }

  1374. 

  1375. /* handle n = 0 too */

  1376. static inline int get_bitsz(GetBitContext *s, int n)

  1377. {

  1378.     if (n == 0)

  1379.         return 0;

  1380.     else

  1381.         return get_bits(s, n);

  1382. }

  1383. 

  1384. 

  1385. static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){

  1386.     if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){

  1387.         s->gb= s->in_gb;

  1388.         s->in_gb.buffer=NULL;

  1389.         assert((get_bits_count(&s->gb) & 7) == 0);

  1390.         skip_bits_long(&s->gb, *pos - *end_pos);

  1391.         *end_pos2=

  1392.         *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;

  1393.         *pos= get_bits_count(&s->gb);

  1394.     }

  1395. }

  1396. 

  1397. static int huffman_decode(MPADecodeContext *s, GranuleDef *g,

  1398.                           int16_t *exponents, int end_pos2)

  1399. {

  1400.     int s_index;

  1401.     int i;

  1402.     int last_pos, bits_left;

  1403.     VLC *vlc;

  1404.     int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);

  1405. 

  1406.     /* low frequencies (called big values) */

  1407.     s_index = 0;

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

  1409.         int j, k, l, linbits;

  1410.         j = g->region_size[i];

  1411.         if (j == 0)

  1412.             continue;

  1413.         /* select vlc table */

  1414.         k = g->table_select[i];

  1415.         l = mpa_huff_data[k][0];

  1416.         linbits = mpa_huff_data[k][1];

  1417.         vlc = &huff_vlc[l];

  1418. 

  1419.         if(!l){

  1420.             memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);

  1421.             s_index += 2*j;

  1422.             continue;

  1423.         }

  1424. 

  1425.         /* read huffcode and compute each couple */

  1426.         for(;j>0;j--) {

  1427.             int exponent, x, y, v;

  1428.             int pos= get_bits_count(&s->gb);

  1429. 

  1430.             if (pos >= end_pos){

  1431. //                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);

  1432.                 switch_buffer(s, &pos, &end_pos, &end_pos2);

  1433. //                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);

  1434.                 if(pos >= end_pos)

  1435.                     break;

  1436.             }

  1437.             y = get_vlc2(&s->gb, vlc->table, 7, 3);

  1438. 

  1439.             if(!y){

  1440.                 g->sb_hybrid[s_index  ] =

  1441.                 g->sb_hybrid[s_index+1] = 0;

  1442.                 s_index += 2;

  1443.                 continue;

  1444.             }

  1445. 

  1446.             exponent= exponents[s_index];

  1447. 

  1448.             dprintf(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",

  1449.                     i, g->region_size[i] - j, x, y, exponent);

  1450.             if(y&16){

  1451.                 x = y >> 5;

  1452.                 y = y & 0x0f;

  1453.                 if (x < 15){

  1454.                     v = expval_table[ exponent ][ x ];

  1455. //                      v = expval_table[ (exponent&3) ][ x ] >> FFMIN(0 - (exponent>>2), 31);

  1456.                 }else{

  1457.                     x += get_bitsz(&s->gb, linbits);

  1458.                     v = l3_unscale(x, exponent);

  1459.                 }

  1460.                 if (get_bits1(&s->gb))

  1461.                     v = -v;

  1462.                 g->sb_hybrid[s_index] = v;

  1463.                 if (y < 15){

  1464.                     v = expval_table[ exponent ][ y ];

  1465.                 }else{

  1466.                     y += get_bitsz(&s->gb, linbits);

  1467.                     v = l3_unscale(y, exponent);

  1468.                 }

  1469.                 if (get_bits1(&s->gb))

  1470.                     v = -v;

  1471.                 g->sb_hybrid[s_index+1] = v;

  1472.             }else{

  1473.                 x = y >> 5;

  1474.                 y = y & 0x0f;

  1475.                 x += y;

  1476.                 if (x < 15){

  1477.                     v = expval_table[ exponent ][ x ];

  1478.                 }else{

  1479.                     x += get_bitsz(&s->gb, linbits);

  1480.                     v = l3_unscale(x, exponent);

  1481.                 }

  1482.                 if (get_bits1(&s->gb))

  1483.                     v = -v;

  1484.                 g->sb_hybrid[s_index+!!y] = v;

  1485.                 g->sb_hybrid[s_index+ !y] = 0;

  1486.             }

  1487.             s_index+=2;

  1488.         }

  1489.     }

  1490. 

  1491.     /* high frequencies */

  1492.     vlc = &huff_quad_vlc[g->count1table_select];

  1493.     last_pos=0;

  1494.     while (s_index <= 572) {

  1495.         int pos, code;

  1496.         pos = get_bits_count(&s->gb);

  1497.         if (pos >= end_pos) {

  1498.             if (pos > end_pos2 && last_pos){

  1499.                 /* some encoders generate an incorrect size for this

  1500.                    part. We must go back into the data */

  1501.                 s_index -= 4;

  1502.                 skip_bits_long(&s->gb, last_pos - pos);

  1503.                 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);

  1504.                 if(s->error_recognition >= FF_ER_COMPLIANT)

  1505.                     s_index=0;

  1506.                 break;

  1507.             }

  1508. //                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);

  1509.             switch_buffer(s, &pos, &end_pos, &end_pos2);

  1510. //                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);

  1511.             if(pos >= end_pos)

  1512.                 break;

  1513.         }

  1514.         last_pos= pos;

  1515. 

  1516.         code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);

  1517.         dprintf(s->avctx, "t=%d code=%d\n", g->count1table_select, code);

  1518.         g->sb_hybrid[s_index+0]=

  1519.         g->sb_hybrid[s_index+1]=

  1520.         g->sb_hybrid[s_index+2]=

  1521.         g->sb_hybrid[s_index+3]= 0;

  1522.         while(code){

  1523.             static const int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};

  1524.             int v;

  1525.             int pos= s_index+idxtab[code];

  1526.             code ^= 8>>idxtab[code];

  1527.             v = exp_table[ exponents[pos] ];

  1528. //            v = exp_table[ (exponents[pos]&3) ] >> FFMIN(0 - (exponents[pos]>>2), 31);

  1529.             if(get_bits1(&s->gb))

  1530.                 v = -v;

  1531.             g->sb_hybrid[pos] = v;

  1532.         }

  1533.         s_index+=4;

  1534.     }

  1535.     /* skip extension bits */

  1536.     bits_left = end_pos2 - get_bits_count(&s->gb);

  1537. //av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);

  1538.     if (bits_left < 0 && s->error_recognition >= FF_ER_COMPLIANT) {

  1539.         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1540.         s_index=0;

  1541.     }else if(bits_left > 0 && s->error_recognition >= FF_ER_AGGRESSIVE){

  1542.         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1543.         s_index=0;

  1544.     }

  1545.     memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));

  1546.     skip_bits_long(&s->gb, bits_left);

  1547. 

  1548.     i= get_bits_count(&s->gb);

  1549.     switch_buffer(s, &i, &end_pos, &end_pos2);

  1550. 

  1551.     return 0;

  1552. }

  1553. 

  1554. /* Reorder short blocks from bitstream order to interleaved order. It

  1555.    would be faster to do it in parsing, but the code would be far more

  1556.    complicated */

  1557. static void reorder_block(MPADecodeContext *s, GranuleDef *g)

  1558. {

  1559.     int i, j, len;

  1560.     int32_t *ptr, *dst, *ptr1;

  1561.     int32_t tmp[576];

  1562. 

  1563.     if (g->block_type != 2)

  1564.         return;

  1565. 

  1566.     if (g->switch_point) {

  1567.         if (s->sample_rate_index != 8) {

  1568.             ptr = g->sb_hybrid + 36;

  1569.         } else {

  1570.             ptr = g->sb_hybrid + 48;

  1571.         }

  1572.     } else {

  1573.         ptr = g->sb_hybrid;

  1574.     }

  1575. 

  1576.     for(i=g->short_start;i<13;i++) {

  1577.         len = band_size_short[s->sample_rate_index][i];

  1578.         ptr1 = ptr;

  1579.         dst = tmp;

  1580.         for(j=len;j>0;j--) {

  1581.             *dst++ = ptr[0*len];

  1582.             *dst++ = ptr[1*len];

  1583.             *dst++ = ptr[2*len];

  1584.             ptr++;

  1585.         }

  1586.         ptr+=2*len;

  1587.         memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));

  1588.     }

  1589. }

  1590. 

  1591. #define ISQRT2 FIXR(0.70710678118654752440)

  1592. 

  1593. static void compute_stereo(MPADecodeContext *s,

  1594.                            GranuleDef *g0, GranuleDef *g1)

  1595. {

  1596.     int i, j, k, l;

  1597.     int32_t v1, v2;

  1598.     int sf_max, tmp0, tmp1, sf, len, non_zero_found;

  1599.     int32_t (*is_tab)[16];

  1600.     int32_t *tab0, *tab1;

  1601.     int non_zero_found_short[3];

  1602. 

  1603.     /* intensity stereo */

  1604.     if (s->mode_ext & MODE_EXT_I_STEREO) {

  1605.         if (!s->lsf) {

  1606.             is_tab = is_table;

  1607.             sf_max = 7;

  1608.         } else {

  1609.             is_tab = is_table_lsf[g1->scalefac_compress & 1];

  1610.             sf_max = 16;

  1611.         }

  1612. 

  1613.         tab0 = g0->sb_hybrid + 576;

  1614.         tab1 = g1->sb_hybrid + 576;

  1615. 

  1616.         non_zero_found_short[0] = 0;

  1617.         non_zero_found_short[1] = 0;

  1618.         non_zero_found_short[2] = 0;

  1619.         k = (13 - g1->short_start) * 3 + g1->long_end - 3;

  1620.         for(i = 12;i >= g1->short_start;i--) {

  1621.             /* for last band, use previous scale factor */

  1622.             if (i != 11)

  1623.                 k -= 3;

  1624.             len = band_size_short[s->sample_rate_index][i];

  1625.             for(l=2;l>=0;l--) {

  1626.                 tab0 -= len;

  1627.                 tab1 -= len;

  1628.                 if (!non_zero_found_short[l]) {

  1629.                     /* test if non zero band. if so, stop doing i-stereo */

  1630.                     for(j=0;j<len;j++) {

  1631.                         if (tab1[j] != 0) {

  1632.                             non_zero_found_short[l] = 1;

  1633.                             goto found1;

  1634.                         }

  1635.                     }

  1636.                     sf = g1->scale_factors[k + l];

  1637.                     if (sf >= sf_max)

  1638.                         goto found1;

  1639. 

  1640.                     v1 = is_tab[0][sf];

  1641.                     v2 = is_tab[1][sf];

  1642.                     for(j=0;j<len;j++) {

  1643.                         tmp0 = tab0[j];

  1644.                         tab0[j] = MULL(tmp0, v1, FRAC_BITS);

  1645.                         tab1[j] = MULL(tmp0, v2, FRAC_BITS);

  1646.                     }

  1647.                 } else {

  1648.                 found1:

  1649.                     if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1650.                         /* lower part of the spectrum : do ms stereo

  1651.                            if enabled */

  1652.                         for(j=0;j<len;j++) {

  1653.                             tmp0 = tab0[j];

  1654.                             tmp1 = tab1[j];

  1655.                             tab0[j] = MULL(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1656.                             tab1[j] = MULL(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1657.                         }

  1658.                     }

  1659.                 }

  1660.             }

  1661.         }

  1662. 

  1663.         non_zero_found = non_zero_found_short[0] |

  1664.             non_zero_found_short[1] |

  1665.             non_zero_found_short[2];

  1666. 

  1667.         for(i = g1->long_end - 1;i >= 0;i--) {

  1668.             len = band_size_long[s->sample_rate_index][i];

  1669.             tab0 -= len;

  1670.             tab1 -= len;

  1671.             /* test if non zero band. if so, stop doing i-stereo */

  1672.             if (!non_zero_found) {

  1673.                 for(j=0;j<len;j++) {

  1674.                     if (tab1[j] != 0) {

  1675.                         non_zero_found = 1;

  1676.                         goto found2;

  1677.                     }

  1678.                 }

  1679.                 /* for last band, use previous scale factor */

  1680.                 k = (i == 21) ? 20 : i;

  1681.                 sf = g1->scale_factors[k];

  1682.                 if (sf >= sf_max)

  1683.                     goto found2;

  1684.                 v1 = is_tab[0][sf];

  1685.                 v2 = is_tab[1][sf];

  1686.                 for(j=0;j<len;j++) {

  1687.                     tmp0 = tab0[j];

  1688.                     tab0[j] = MULL(tmp0, v1, FRAC_BITS);

  1689.                     tab1[j] = MULL(tmp0, v2, FRAC_BITS);

  1690.                 }

  1691.             } else {

  1692.             found2:

  1693.                 if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1694.                     /* lower part of the spectrum : do ms stereo

  1695.                        if enabled */

  1696.                     for(j=0;j<len;j++) {

  1697.                         tmp0 = tab0[j];

  1698.                         tmp1 = tab1[j];

  1699.                         tab0[j] = MULL(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1700.                         tab1[j] = MULL(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1701.                     }

  1702.                 }

  1703.             }

  1704.         }

  1705.     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1706.         /* ms stereo ONLY */

  1707.         /* NOTE: the 1/sqrt(2) normalization factor is included in the

  1708.            global gain */

  1709.         tab0 = g0->sb_hybrid;

  1710.         tab1 = g1->sb_hybrid;

  1711.         for(i=0;i<576;i++) {

  1712.             tmp0 = tab0[i];

  1713.             tmp1 = tab1[i];

  1714.             tab0[i] = tmp0 + tmp1;

  1715.             tab1[i] = tmp0 - tmp1;

  1716.         }

  1717.     }

  1718. }

  1719. 

  1720. static void compute_antialias_integer(MPADecodeContext *s,

  1721.                               GranuleDef *g)

  1722. {

  1723.     int32_t *ptr, *csa;

  1724.     int n, i;

  1725. 

  1726.     /* we antialias only "long" bands */

  1727.     if (g->block_type == 2) {

  1728.         if (!g->switch_point)

  1729.             return;

  1730.         /* XXX: check this for 8000Hz case */

  1731.         n = 1;

  1732.     } else {

  1733.         n = SBLIMIT - 1;

  1734.     }

  1735. 

  1736.     ptr = g->sb_hybrid + 18;

  1737.     for(i = n;i > 0;i--) {

  1738.         int tmp0, tmp1, tmp2;

  1739.         csa = &csa_table[0][0];

  1740. #define INT_AA(j) \

  1741.             tmp0 = ptr[-1-j];\

  1742.             tmp1 = ptr[   j];\

  1743.             tmp2= MULH(tmp0 + tmp1, csa[0+4*j]);\

  1744.             ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa[2+4*j]));\

  1745.             ptr[   j] = 4*(tmp2 + MULH(tmp0, csa[3+4*j]));

  1746. 

  1747.         INT_AA(0)

  1748.         INT_AA(1)

  1749.         INT_AA(2)

  1750.         INT_AA(3)

  1751.         INT_AA(4)

  1752.         INT_AA(5)

  1753.         INT_AA(6)

  1754.         INT_AA(7)

  1755. 

  1756.         ptr += 18;

  1757.     }

  1758. }

  1759. 

  1760. static void compute_antialias_float(MPADecodeContext *s,

  1761.                               GranuleDef *g)

  1762. {

  1763.     int32_t *ptr;

  1764.     int n, i;

  1765. 

  1766.     /* we antialias only "long" bands */

  1767.     if (g->block_type == 2) {

  1768.         if (!g->switch_point)

  1769.             return;

  1770.         /* XXX: check this for 8000Hz case */

  1771.         n = 1;

  1772.     } else {

  1773.         n = SBLIMIT - 1;

  1774.     }

  1775. 

  1776.     ptr = g->sb_hybrid + 18;

  1777.     for(i = n;i > 0;i--) {

  1778.         float tmp0, tmp1;

  1779.         float *csa = &csa_table_float[0][0];

  1780. #define FLOAT_AA(j)\

  1781.         tmp0= ptr[-1-j];\

  1782.         tmp1= ptr[   j];\

  1783.         ptr[-1-j] = lrintf(tmp0 * csa[0+4*j] - tmp1 * csa[1+4*j]);\

  1784.         ptr[   j] = lrintf(tmp0 * csa[1+4*j] + tmp1 * csa[0+4*j]);

  1785. 

  1786.         FLOAT_AA(0)

  1787.         FLOAT_AA(1)

  1788.         FLOAT_AA(2)

  1789.         FLOAT_AA(3)

  1790.         FLOAT_AA(4)

  1791.         FLOAT_AA(5)

  1792.         FLOAT_AA(6)

  1793.         FLOAT_AA(7)

  1794. 

  1795.         ptr += 18;

  1796.     }

  1797. }

  1798. 

  1799. static void compute_imdct(MPADecodeContext *s,

  1800.                           GranuleDef *g,

  1801.                           int32_t *sb_samples,

  1802.                           int32_t *mdct_buf)

  1803. {

  1804.     int32_t *ptr, *win, *win1, *buf, *out_ptr, *ptr1;

  1805.     int32_t out2[12];

  1806.     int i, j, mdct_long_end, v, sblimit;

  1807. 

  1808.     /* find last non zero block */

  1809.     ptr = g->sb_hybrid + 576;

  1810.     ptr1 = g->sb_hybrid + 2 * 18;

  1811.     while (ptr >= ptr1) {

  1812.         ptr -= 6;

  1813.         v = ptr[0] | ptr[1] | ptr[2] | ptr[3] | ptr[4] | ptr[5];

  1814.         if (v != 0)

  1815.             break;

  1816.     }

  1817.     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

  1818. 

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

  1820.         /* XXX: check for 8000 Hz */

  1821.         if (g->switch_point)

  1822.             mdct_long_end = 2;

  1823.         else

  1824.             mdct_long_end = 0;

  1825.     } else {

  1826.         mdct_long_end = sblimit;

  1827.     }

  1828. 

  1829.     buf = mdct_buf;

  1830.     ptr = g->sb_hybrid;

  1831.     for(j=0;j<mdct_long_end;j++) {

  1832.         /* apply window & overlap with previous buffer */

  1833.         out_ptr = sb_samples + j;

  1834.         /* select window */

  1835.         if (g->switch_point && j < 2)

  1836.             win1 = mdct_win[0];

  1837.         else

  1838.             win1 = mdct_win[g->block_type];

  1839.         /* select frequency inversion */

  1840.         win = win1 + ((4 * 36) & -(j & 1));

  1841.         imdct36(out_ptr, buf, ptr, win);

  1842.         out_ptr += 18*SBLIMIT;

  1843.         ptr += 18;

  1844.         buf += 18;

  1845.     }

  1846.     for(j=mdct_long_end;j<sblimit;j++) {

  1847.         /* select frequency inversion */

  1848.         win = mdct_win[2] + ((4 * 36) & -(j & 1));

  1849.         out_ptr = sb_samples + j;

  1850. 

  1851.         for(i=0; i<6; i++){

  1852.             *out_ptr = buf[i];

  1853.             out_ptr += SBLIMIT;

  1854.         }

  1855.         imdct12(out2, ptr + 0);

  1856.         for(i=0;i<6;i++) {

  1857.             *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*1];

  1858.             buf[i + 6*2] = MULH(out2[i + 6], win[i + 6]);

  1859.             out_ptr += SBLIMIT;

  1860.         }

  1861.         imdct12(out2, ptr + 1);

  1862.         for(i=0;i<6;i++) {

  1863.             *out_ptr = MULH(out2[i], win[i]) + buf[i + 6*2];

  1864.             buf[i + 6*0] = MULH(out2[i + 6], win[i + 6]);

  1865.             out_ptr += SBLIMIT;

  1866.         }

  1867.         imdct12(out2, ptr + 2);

  1868.         for(i=0;i<6;i++) {

  1869.             buf[i + 6*0] = MULH(out2[i], win[i]) + buf[i + 6*0];

  1870.             buf[i + 6*1] = MULH(out2[i + 6], win[i + 6]);

  1871.             buf[i + 6*2] = 0;

  1872.         }

  1873.         ptr += 18;

  1874.         buf += 18;

  1875.     }

  1876.     /* zero bands */

  1877.     for(j=sblimit;j<SBLIMIT;j++) {

  1878.         /* overlap */

  1879.         out_ptr = sb_samples + j;

  1880.         for(i=0;i<18;i++) {

  1881.             *out_ptr = buf[i];

  1882.             buf[i] = 0;

  1883.             out_ptr += SBLIMIT;

  1884.         }

  1885.         buf += 18;

  1886.     }

  1887. }

  1888. 

  1889. /* main layer3 decoding function */

  1890. static int mp_decode_layer3(MPADecodeContext *s)

  1891. {

  1892.     int nb_granules, main_data_begin, private_bits;

  1893.     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;

  1894.     GranuleDef *g;

  1895.     int16_t exponents[576];

  1896. 

  1897.     /* read side info */

  1898.     if (s->lsf) {

  1899.         main_data_begin = get_bits(&s->gb, 8);

  1900.         private_bits = get_bits(&s->gb, s->nb_channels);

  1901.         nb_granules = 1;

  1902.     } else {

  1903.         main_data_begin = get_bits(&s->gb, 9);

  1904.         if (s->nb_channels == 2)

  1905.             private_bits = get_bits(&s->gb, 3);

  1906.         else

  1907.             private_bits = get_bits(&s->gb, 5);

  1908.         nb_granules = 2;

  1909.         for(ch=0;ch<s->nb_channels;ch++) {

  1910.             s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */

  1911.             s->granules[ch][1].scfsi = get_bits(&s->gb, 4);

  1912.         }

  1913.     }

  1914. 

  1915.     for(gr=0;gr<nb_granules;gr++) {

  1916.         for(ch=0;ch<s->nb_channels;ch++) {

  1917.             dprintf(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);

  1918.             g = &s->granules[ch][gr];

  1919.             g->part2_3_length = get_bits(&s->gb, 12);

  1920.             g->big_values = get_bits(&s->gb, 9);

  1921.             if(g->big_values > 288){

  1922.                 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");

  1923.                 return -1;

  1924.             }

  1925. 

  1926.             g->global_gain = get_bits(&s->gb, 8);

  1927.             /* if MS stereo only is selected, we precompute the

  1928.                1/sqrt(2) renormalization factor */

  1929.             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==

  1930.                 MODE_EXT_MS_STEREO)

  1931.                 g->global_gain -= 2;

  1932.             if (s->lsf)

  1933.                 g->scalefac_compress = get_bits(&s->gb, 9);

  1934.             else

  1935.                 g->scalefac_compress = get_bits(&s->gb, 4);

  1936.             blocksplit_flag = get_bits1(&s->gb);

  1937.             if (blocksplit_flag) {

  1938.                 g->block_type = get_bits(&s->gb, 2);

  1939.                 if (g->block_type == 0){

  1940.                     av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");

  1941.                     return -1;

  1942.                 }

  1943.                 g->switch_point = get_bits1(&s->gb);

  1944.                 for(i=0;i<2;i++)

  1945.                     g->table_select[i] = get_bits(&s->gb, 5);

  1946.                 for(i=0;i<3;i++)

  1947.                     g->subblock_gain[i] = get_bits(&s->gb, 3);

  1948.                 ff_init_short_region(s, g);

  1949.             } else {

  1950.                 int region_address1, region_address2;

  1951.                 g->block_type = 0;

  1952.                 g->switch_point = 0;

  1953.                 for(i=0;i<3;i++)

  1954.                     g->table_select[i] = get_bits(&s->gb, 5);

  1955.                 /* compute huffman coded region sizes */

  1956.                 region_address1 = get_bits(&s->gb, 4);

  1957.                 region_address2 = get_bits(&s->gb, 3);

  1958.                 dprintf(s->avctx, "region1=%d region2=%d\n",

  1959.                         region_address1, region_address2);

  1960.                 ff_init_long_region(s, g, region_address1, region_address2);

  1961.             }

  1962.             ff_region_offset2size(g);

  1963.             ff_compute_band_indexes(s, g);

  1964. 

  1965.             g->preflag = 0;

  1966.             if (!s->lsf)

  1967.                 g->preflag = get_bits1(&s->gb);

  1968.             g->scalefac_scale = get_bits1(&s->gb);

  1969.             g->count1table_select = get_bits1(&s->gb);

  1970.             dprintf(s->avctx, "block_type=%d switch_point=%d\n",

  1971.                     g->block_type, g->switch_point);

  1972.         }

  1973.     }

  1974. 

  1975.   if (!s->adu_mode) {

  1976.     const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);

  1977.     assert((get_bits_count(&s->gb) & 7) == 0);

  1978.     /* now we get bits from the main_data_begin offset */

  1979.     dprintf(s->avctx, "seekback: %d\n", main_data_begin);

  1980. //av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);

  1981. 

  1982.     memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);

  1983.     s->in_gb= s->gb;

  1984.         init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);

  1985.         skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));

  1986.   }

  1987. 

  1988.     for(gr=0;gr<nb_granules;gr++) {

  1989.         for(ch=0;ch<s->nb_channels;ch++) {

  1990.             g = &s->granules[ch][gr];

  1991.             if(get_bits_count(&s->gb)<0){

  1992.                 av_log(s->avctx, AV_LOG_DEBUG, "mdb:%d, lastbuf:%d skipping granule %d\n",

  1993.                                             main_data_begin, s->last_buf_size, gr);

  1994.                 skip_bits_long(&s->gb, g->part2_3_length);

  1995.                 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));

  1996.                 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){

  1997.                     skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);

  1998.                     s->gb= s->in_gb;

  1999.                     s->in_gb.buffer=NULL;

  2000.                 }

  2001.                 continue;

  2002.             }

  2003. 

  2004.             bits_pos = get_bits_count(&s->gb);

  2005. 

  2006.             if (!s->lsf) {

  2007.                 uint8_t *sc;

  2008.                 int slen, slen1, slen2;

  2009. 

  2010.                 /* MPEG1 scale factors */

  2011.                 slen1 = slen_table[0][g->scalefac_compress];

  2012.                 slen2 = slen_table[1][g->scalefac_compress];

  2013.                 dprintf(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);

  2014.                 if (g->block_type == 2) {

  2015.                     n = g->switch_point ? 17 : 18;

  2016.                     j = 0;

  2017.                     if(slen1){

  2018.                         for(i=0;i<n;i++)

  2019.                             g->scale_factors[j++] = get_bits(&s->gb, slen1);

  2020.                     }else{

  2021.                         for(i=0;i<n;i++)

  2022.                             g->scale_factors[j++] = 0;

  2023.                     }

  2024.                     if(slen2){

  2025.                         for(i=0;i<18;i++)

  2026.                             g->scale_factors[j++] = get_bits(&s->gb, slen2);

  2027.                         for(i=0;i<3;i++)

  2028.                             g->scale_factors[j++] = 0;

  2029.                     }else{

  2030.                         for(i=0;i<21;i++)

  2031.                             g->scale_factors[j++] = 0;

  2032.                     }

  2033.                 } else {

  2034.                     sc = s->granules[ch][0].scale_factors;

  2035.                     j = 0;

  2036.                     for(k=0;k<4;k++) {

  2037.                         n = (k == 0 ? 6 : 5);

  2038.                         if ((g->scfsi & (0x8 >> k)) == 0) {

  2039.                             slen = (k < 2) ? slen1 : slen2;

  2040.                             if(slen){

  2041.                                 for(i=0;i<n;i++)

  2042.                                     g->scale_factors[j++] = get_bits(&s->gb, slen);

  2043.                             }else{

  2044.                                 for(i=0;i<n;i++)

  2045.                                     g->scale_factors[j++] = 0;

  2046.                             }

  2047.                         } else {

  2048.                             /* simply copy from last granule */

  2049.                             for(i=0;i<n;i++) {

  2050.                                 g->scale_factors[j] = sc[j];

  2051.                                 j++;

  2052.                             }

  2053.                         }

  2054.                     }

  2055.                     g->scale_factors[j++] = 0;

  2056.                 }

  2057.             } else {

  2058.                 int tindex, tindex2, slen[4], sl, sf;

  2059. 

  2060.                 /* LSF scale factors */

  2061.                 if (g->block_type == 2) {

  2062.                     tindex = g->switch_point ? 2 : 1;

  2063.                 } else {

  2064.                     tindex = 0;

  2065.                 }

  2066.                 sf = g->scalefac_compress;

  2067.                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {

  2068.                     /* intensity stereo case */

  2069.                     sf >>= 1;

  2070.                     if (sf < 180) {

  2071.                         lsf_sf_expand(slen, sf, 6, 6, 0);

  2072.                         tindex2 = 3;

  2073.                     } else if (sf < 244) {

  2074.                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);

  2075.                         tindex2 = 4;

  2076.                     } else {

  2077.                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);

  2078.                         tindex2 = 5;

  2079.                     }

  2080.                 } else {

  2081.                     /* normal case */

  2082.                     if (sf < 400) {

  2083.                         lsf_sf_expand(slen, sf, 5, 4, 4);

  2084.                         tindex2 = 0;

  2085.                     } else if (sf < 500) {

  2086.                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);

  2087.                         tindex2 = 1;

  2088.                     } else {

  2089.                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);

  2090.                         tindex2 = 2;

  2091.                         g->preflag = 1;

  2092.                     }

  2093.                 }

  2094. 

  2095.                 j = 0;

  2096.                 for(k=0;k<4;k++) {

  2097.                     n = lsf_nsf_table[tindex2][tindex][k];

  2098.                     sl = slen[k];

  2099.                     if(sl){

  2100.                         for(i=0;i<n;i++)

  2101.                             g->scale_factors[j++] = get_bits(&s->gb, sl);

  2102.                     }else{

  2103.                         for(i=0;i<n;i++)

  2104.                             g->scale_factors[j++] = 0;

  2105.                     }

  2106.                 }

  2107.                 /* XXX: should compute exact size */

  2108.                 for(;j<40;j++)

  2109.                     g->scale_factors[j] = 0;

  2110.             }

  2111. 

  2112.             exponents_from_scale_factors(s, g, exponents);

  2113. 

  2114.             /* read Huffman coded residue */

  2115.             huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);

  2116.         } /* ch */

  2117. 

  2118.         if (s->nb_channels == 2)

  2119.             compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);

  2120. 

  2121.         for(ch=0;ch<s->nb_channels;ch++) {

  2122.             g = &s->granules[ch][gr];

  2123. 

  2124.             reorder_block(s, g);

  2125.             s->compute_antialias(s, g);

  2126.             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);

  2127.         }

  2128.     } /* gr */

  2129.     if(get_bits_count(&s->gb)<0)

  2130.         skip_bits_long(&s->gb, -get_bits_count(&s->gb));

  2131.     return nb_granules * 18;

  2132. }

  2133. 

  2134. static int mp_decode_frame(MPADecodeContext *s,

  2135.                            OUT_INT *samples, const uint8_t *buf, int buf_size)

  2136. {

  2137.     int i, nb_frames, ch;

  2138.     OUT_INT *samples_ptr;

  2139. 

  2140.     init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);

  2141. 

  2142.     /* skip error protection field */

  2143.     if (s->error_protection)

  2144.         skip_bits(&s->gb, 16);

  2145. 

  2146.     dprintf(s->avctx, "frame %d:\n", s->frame_count);

  2147.     switch(s->layer) {

  2148.     case 1:

  2149.         s->avctx->frame_size = 384;

  2150.         nb_frames = mp_decode_layer1(s);

  2151.         break;

  2152.     case 2:

  2153.         s->avctx->frame_size = 1152;

  2154.         nb_frames = mp_decode_layer2(s);

  2155.         break;

  2156.     case 3:

  2157.         s->avctx->frame_size = s->lsf ? 576 : 1152;

  2158.     default:

  2159.         nb_frames = mp_decode_layer3(s);

  2160. 

  2161.         s->last_buf_size=0;

  2162.         if(s->in_gb.buffer){

  2163.             align_get_bits(&s->gb);

  2164.             i= get_bits_left(&s->gb)>>3;

  2165.             if(i >= 0 && i <= BACKSTEP_SIZE){

  2166.                 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);

  2167.                 s->last_buf_size=i;

  2168.             }else

  2169.                 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);

  2170.             s->gb= s->in_gb;

  2171.             s->in_gb.buffer= NULL;

  2172.         }

  2173. 

  2174.         align_get_bits(&s->gb);

  2175.         assert((get_bits_count(&s->gb) & 7) == 0);

  2176.         i= get_bits_left(&s->gb)>>3;

  2177. 

  2178.         if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){

  2179.             if(i<0)

  2180.                 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);

  2181.             i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);

  2182.         }

  2183.         assert(i <= buf_size - HEADER_SIZE && i>= 0);

  2184.         memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);

  2185.         s->last_buf_size += i;

  2186. 

  2187.         break;

  2188.     }

  2189. 

  2190.     /* apply the synthesis filter */

  2191.     for(ch=0;ch<s->nb_channels;ch++) {

  2192.         samples_ptr = samples + ch;

  2193.         for(i=0;i<nb_frames;i++) {

  2194.             ff_mpa_synth_filter(s->synth_buf[ch], &(s->synth_buf_offset[ch]),

  2195.                          ff_mpa_synth_window, &s->dither_state,

  2196.                          samples_ptr, s->nb_channels,

  2197.                          s->sb_samples[ch][i]);

  2198.             samples_ptr += 32 * s->nb_channels;

  2199.         }

  2200.     }

  2201. 

  2202.     return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;

  2203. }

  2204. 

  2205. static int decode_frame(AVCodecContext * avctx,

  2206.                         void *data, int *data_size,

  2207.                         AVPacket *avpkt)

  2208. {

  2209.     const uint8_t *buf = avpkt->data;

  2210.     int buf_size = avpkt->size;

  2211.     MPADecodeContext *s = avctx->priv_data;

  2212.     uint32_t header;

  2213.     int out_size;

  2214.     OUT_INT *out_samples = data;

  2215. 

  2216.     if(buf_size < HEADER_SIZE)

  2217.         return -1;

  2218. 

  2219.     header = AV_RB32(buf);

  2220.     if(ff_mpa_check_header(header) < 0){

  2221.         av_log(avctx, AV_LOG_ERROR, "Header missing\n");

  2222.         return -1;

  2223.     }

  2224. 

  2225.     if (ff_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {

  2226.         /* free format: prepare to compute frame size */

  2227.         s->frame_size = -1;

  2228.         return -1;

  2229.     }

  2230.     /* update codec info */

  2231.     avctx->channels = s->nb_channels;

  2232.     avctx->bit_rate = s->bit_rate;

  2233.     avctx->sub_id = s->layer;

  2234. 

  2235.     if(*data_size < 1152*avctx->channels*sizeof(OUT_INT))

  2236.         return -1;

  2237.     *data_size = 0;

  2238. 

  2239.     if(s->frame_size<=0 || s->frame_size > buf_size){

  2240.         av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");

  2241.         return -1;

  2242.     }else if(s->frame_size < buf_size){

  2243.         av_log(avctx, AV_LOG_ERROR, "incorrect frame size\n");

  2244.         buf_size= s->frame_size;

  2245.     }

  2246. 

  2247.     out_size = mp_decode_frame(s, out_samples, buf, buf_size);

  2248.     if(out_size>=0){

  2249.         *data_size = out_size;

  2250.         avctx->sample_rate = s->sample_rate;

  2251.         //FIXME maybe move the other codec info stuff from above here too

  2252.     }else

  2253.         av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed

  2254.     s->frame_size = 0;

  2255.     return buf_size;

  2256. }

  2257. 

  2258. static void flush(AVCodecContext *avctx){

  2259.     MPADecodeContext *s = avctx->priv_data;

  2260.     memset(s->synth_buf, 0, sizeof(s->synth_buf));

  2261.     s->last_buf_size= 0;

  2262. }

  2263. 

  2264. #if CONFIG_MP3ADU_DECODER

  2265. static int decode_frame_adu(AVCodecContext * avctx,

  2266.                         void *data, int *data_size,

  2267.                         AVPacket *avpkt)

  2268. {

  2269.     const uint8_t *buf = avpkt->data;

  2270.     int buf_size = avpkt->size;

  2271.     MPADecodeContext *s = avctx->priv_data;

  2272.     uint32_t header;

  2273.     int len, out_size;

  2274.     OUT_INT *out_samples = data;

  2275. 

  2276.     len = buf_size;

  2277. 

  2278.     // Discard too short frames

  2279.     if (buf_size < HEADER_SIZE) {

  2280.         *data_size = 0;

  2281.         return buf_size;

  2282.     }

  2283. 

  2284. 

  2285.     if (len > MPA_MAX_CODED_FRAME_SIZE)

  2286.         len = MPA_MAX_CODED_FRAME_SIZE;

  2287. 

  2288.     // Get header and restore sync word

  2289.     header = AV_RB32(buf) | 0xffe00000;

  2290. 

  2291.     if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame

  2292.         *data_size = 0;

  2293.         return buf_size;

  2294.     }

  2295. 

  2296.     ff_mpegaudio_decode_header((MPADecodeHeader *)s, header);

  2297.     /* update codec info */

  2298.     avctx->sample_rate = s->sample_rate;

  2299.     avctx->channels = s->nb_channels;

  2300.     avctx->bit_rate = s->bit_rate;

  2301.     avctx->sub_id = s->layer;

  2302. 

  2303.     s->frame_size = len;

  2304. 

  2305.     if (avctx->parse_only) {

  2306.         out_size = buf_size;

  2307.     } else {

  2308.         out_size = mp_decode_frame(s, out_samples, buf, buf_size);

  2309.     }

  2310. 

  2311.     *data_size = out_size;

  2312.     return buf_size;

  2313. }

  2314. #endif /* CONFIG_MP3ADU_DECODER */

  2315. 

  2316. #if CONFIG_MP3ON4_DECODER

  2317. 

  2318. /**

  2319.  * Context for MP3On4 decoder

  2320.  */

  2321. typedef struct MP3On4DecodeContext {

  2322.     int frames;   ///< number of mp3 frames per block (number of mp3 decoder instances)

  2323.     int syncword; ///< syncword patch

  2324.     const uint8_t *coff; ///< channels offsets in output buffer

  2325.     MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance

  2326. } MP3On4DecodeContext;

  2327. 

  2328. #include "mpeg4audio.h"

  2329. 

  2330. /* Next 3 arrays are indexed by channel config number (passed via codecdata) */

  2331. static const uint8_t mp3Frames[8] = {0,1,1,2,3,3,4,5};   /* number of mp3 decoder instances */

  2332. /* offsets into output buffer, assume output order is FL FR BL BR C LFE */

  2333. static const uint8_t chan_offset[8][5] = {

  2334.     {0},

  2335.     {0},            // C

  2336.     {0},            // FLR

  2337.     {2,0},          // C FLR

  2338.     {2,0,3},        // C FLR BS

  2339.     {4,0,2},        // C FLR BLRS

  2340.     {4,0,2,5},      // C FLR BLRS LFE

  2341.     {4,0,2,6,5},    // C FLR BLRS BLR LFE

  2342. };

  2343. 

  2344. 

  2345. static int decode_init_mp3on4(AVCodecContext * avctx)

  2346. {

  2347.     MP3On4DecodeContext *s = avctx->priv_data;

  2348.     MPEG4AudioConfig cfg;

  2349.     int i;

  2350. 

  2351.     if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {

  2352.         av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");

  2353.         return -1;

  2354.     }

  2355. 

  2356.     ff_mpeg4audio_get_config(&cfg, avctx->extradata, avctx->extradata_size);

  2357.     if (!cfg.chan_config || cfg.chan_config > 7) {

  2358.         av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");

  2359.         return -1;

  2360.     }

  2361.     s->frames = mp3Frames[cfg.chan_config];

  2362.     s->coff = chan_offset[cfg.chan_config];

  2363.     avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];

  2364. 

  2365.     if (cfg.sample_rate < 16000)

  2366.         s->syncword = 0xffe00000;

  2367.     else

  2368.         s->syncword = 0xfff00000;

  2369. 

  2370.     /* Init the first mp3 decoder in standard way, so that all tables get builded

  2371.      * We replace avctx->priv_data with the context of the first decoder so that

  2372.      * decode_init() does not have to be changed.

  2373.      * Other decoders will be initialized here copying data from the first context

  2374.      */

  2375.     // Allocate zeroed memory for the first decoder context

  2376.     s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));

  2377.     // Put decoder context in place to make init_decode() happy

  2378.     avctx->priv_data = s->mp3decctx[0];

  2379.     decode_init(avctx);

  2380.     // Restore mp3on4 context pointer

  2381.     avctx->priv_data = s;

  2382.     s->mp3decctx[0]->adu_mode = 1; // Set adu mode

  2383. 

  2384.     /* Create a separate codec/context for each frame (first is already ok).

  2385.      * Each frame is 1 or 2 channels - up to 5 frames allowed

  2386.      */

  2387.     for (i = 1; i < s->frames; i++) {

  2388.         s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));

  2389.         s->mp3decctx[i]->compute_antialias = s->mp3decctx[0]->compute_antialias;

  2390.         s->mp3decctx[i]->adu_mode = 1;

  2391.         s->mp3decctx[i]->avctx = avctx;

  2392.     }

  2393. 

  2394.     return 0;

  2395. }

  2396. 

  2397. 

  2398. static av_cold int decode_close_mp3on4(AVCodecContext * avctx)

  2399. {

  2400.     MP3On4DecodeContext *s = avctx->priv_data;

  2401.     int i;

  2402. 

  2403.     for (i = 0; i < s->frames; i++)

  2404.         if (s->mp3decctx[i])

  2405.             av_free(s->mp3decctx[i]);

  2406. 

  2407.     return 0;

  2408. }

  2409. 

  2410. 

  2411. static int decode_frame_mp3on4(AVCodecContext * avctx,

  2412.                         void *data, int *data_size,

  2413.                         AVPacket *avpkt)

  2414. {

  2415.     const uint8_t *buf = avpkt->data;

  2416.     int buf_size = avpkt->size;

  2417.     MP3On4DecodeContext *s = avctx->priv_data;

  2418.     MPADecodeContext *m;

  2419.     int fsize, len = buf_size, out_size = 0;

  2420.     uint32_t header;

  2421.     OUT_INT *out_samples = data;

  2422.     OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];

  2423.     OUT_INT *outptr, *bp;

  2424.     int fr, j, n;

  2425. 

  2426.     if(*data_size < MPA_FRAME_SIZE * MPA_MAX_CHANNELS * s->frames * sizeof(OUT_INT))

  2427.         return -1;

  2428. 

  2429.     *data_size = 0;

  2430.     // Discard too short frames

  2431.     if (buf_size < HEADER_SIZE)

  2432.         return -1;

  2433. 

  2434.     // If only one decoder interleave is not needed

  2435.     outptr = s->frames == 1 ? out_samples : decoded_buf;

  2436. 

  2437.     avctx->bit_rate = 0;

  2438. 

  2439.     for (fr = 0; fr < s->frames; fr++) {

  2440.         fsize = AV_RB16(buf) >> 4;

  2441.         fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);

  2442.         m = s->mp3decctx[fr];

  2443.         assert (m != NULL);

  2444. 

  2445.         header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header

  2446. 

  2447.         if (ff_mpa_check_header(header) < 0) // Bad header, discard block

  2448.             break;

  2449. 

  2450.         ff_mpegaudio_decode_header((MPADecodeHeader *)m, header);

  2451.         out_size += mp_decode_frame(m, outptr, buf, fsize);

  2452.         buf += fsize;

  2453.         len -= fsize;

  2454. 

  2455.         if(s->frames > 1) {

  2456.             n = m->avctx->frame_size*m->nb_channels;

  2457.             /* interleave output data */

  2458.             bp = out_samples + s->coff[fr];

  2459.             if(m->nb_channels == 1) {

  2460.                 for(j = 0; j < n; j++) {

  2461.                     *bp = decoded_buf[j];

  2462.                     bp += avctx->channels;

  2463.                 }

  2464.             } else {

  2465.                 for(j = 0; j < n; j++) {

  2466.                     bp[0] = decoded_buf[j++];

  2467.                     bp[1] = decoded_buf[j];

  2468.                     bp += avctx->channels;

  2469.                 }

  2470.             }

  2471.         }

  2472.         avctx->bit_rate += m->bit_rate;

  2473.     }

  2474. 

  2475.     /* update codec info */

  2476.     avctx->sample_rate = s->mp3decctx[0]->sample_rate;

  2477. 

  2478.     *data_size = out_size;

  2479.     return buf_size;

  2480. }

  2481. #endif /* CONFIG_MP3ON4_DECODER */

  2482. 

  2483. #if CONFIG_MP1_DECODER

  2484. AVCodec mp1_decoder =

  2485. {

  2486.     "mp1",

  2487.     AVMEDIA_TYPE_AUDIO,

  2488.     CODEC_ID_MP1,

  2489.     sizeof(MPADecodeContext),

  2490.     decode_init,

  2491.     NULL,

  2492.     NULL,

  2493.     decode_frame,

  2494.     CODEC_CAP_PARSE_ONLY,

  2495.     .flush= flush,

  2496.     .long_name= NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),

  2497. };

  2498. #endif

  2499. #if CONFIG_MP2_DECODER

  2500. AVCodec mp2_decoder =

  2501. {

  2502.     "mp2",

  2503.     AVMEDIA_TYPE_AUDIO,

  2504.     CODEC_ID_MP2,

  2505.     sizeof(MPADecodeContext),

  2506.     decode_init,

  2507.     NULL,

  2508.     NULL,

  2509.     decode_frame,

  2510.     CODEC_CAP_PARSE_ONLY,

  2511.     .flush= flush,

  2512.     .long_name= NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),

  2513. };

  2514. #endif

  2515. #if CONFIG_MP3_DECODER

  2516. AVCodec mp3_decoder =

  2517. {

  2518.     "mp3",

  2519.     AVMEDIA_TYPE_AUDIO,

  2520.     CODEC_ID_MP3,

  2521.     sizeof(MPADecodeContext),

  2522.     decode_init,

  2523.     NULL,

  2524.     NULL,

  2525.     decode_frame,

  2526.     CODEC_CAP_PARSE_ONLY,

  2527.     .flush= flush,

  2528.     .long_name= NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),

  2529. };

  2530. #endif

  2531. #if CONFIG_MP3ADU_DECODER

  2532. AVCodec mp3adu_decoder =

  2533. {

  2534.     "mp3adu",

  2535.     AVMEDIA_TYPE_AUDIO,

  2536.     CODEC_ID_MP3ADU,

  2537.     sizeof(MPADecodeContext),

  2538.     decode_init,

  2539.     NULL,

  2540.     NULL,

  2541.     decode_frame_adu,

  2542.     CODEC_CAP_PARSE_ONLY,

  2543.     .flush= flush,

  2544.     .long_name= NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),

  2545. };

  2546. #endif

  2547. #if CONFIG_MP3ON4_DECODER

  2548. AVCodec mp3on4_decoder =

  2549. {

  2550.     "mp3on4",

  2551.     AVMEDIA_TYPE_AUDIO,

  2552.     CODEC_ID_MP3ON4,

  2553.     sizeof(MP3On4DecodeContext),

  2554.     decode_init_mp3on4,

  2555.     NULL,

  2556.     decode_close_mp3on4,

  2557.     decode_frame_mp3on4,

  2558.     .flush= flush,

  2559.     .long_name= NULL_IF_CONFIG_SMALL("MP3onMP4"),

  2560. };

  2561. #endif