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.
38308 Commits
32 Branches
283MB
Tree: ce1e8045e0
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ce1e8045e0'
${ noResults }
FFmpeg/libavcodec/mpegaudiodec.c

2056 lines
67KB
Raw Blame History 

  1. /*

  2.  * MPEG Audio decoder

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

  4.  *

  5.  * This file is part of Libav.

  6.  *

  7.  * Libav 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.  * Libav 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 Libav; 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 "libavutil/attributes.h"

  28. #include "libavutil/avassert.h"

  29. #include "libavutil/channel_layout.h"

  30. #include "libavutil/float_dsp.h"

  31. #include "avcodec.h"

  32. #include "get_bits.h"

  33. #include "internal.h"

  34. #include "mathops.h"

  35. #include "mpegaudiodsp.h"

  36. 

  37. /*

  38.  * TODO:

  39.  *  - test lsf / mpeg25 extensively.

  40.  */

  41. 

  42. #include "mpegaudio.h"

  43. #include "mpegaudiodecheader.h"

  44. 

  45. #define BACKSTEP_SIZE 512

  46. #define EXTRABYTES 24

  47. #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES

  48. 

  49. /* layer 3 "granule" */

  50. typedef struct GranuleDef {

  51.     uint8_t scfsi;

  52.     int part2_3_length;

  53.     int big_values;

  54.     int global_gain;

  55.     int scalefac_compress;

  56.     uint8_t block_type;

  57.     uint8_t switch_point;

  58.     int table_select[3];

  59.     int subblock_gain[3];

  60.     uint8_t scalefac_scale;

  61.     uint8_t count1table_select;

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

  63.     int preflag;

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

  65.     uint8_t scale_factors[40];

  66.     DECLARE_ALIGNED(16, INTFLOAT, sb_hybrid)[SBLIMIT * 18]; /* 576 samples */

  67. } GranuleDef;

  68. 

  69. typedef struct MPADecodeContext {

  70.     MPA_DECODE_HEADER

  71.     uint8_t last_buf[LAST_BUF_SIZE];

  72.     int last_buf_size;

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

  74.     uint32_t free_format_next_header;

  75.     GetBitContext gb;

  76.     GetBitContext in_gb;

  77.     DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];

  78.     int synth_buf_offset[MPA_MAX_CHANNELS];

  79.     DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];

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

  81.     GranuleDef granules[2][2]; /* Used in Layer 3 */

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

  83.     int dither_state;

  84.     int err_recognition;

  85.     AVCodecContext* avctx;

  86.     MPADSPContext mpadsp;

  87.     AVFloatDSPContext fdsp;

  88.     AVFrame *frame;

  89. } MPADecodeContext;

  90. 

  91. #if CONFIG_FLOAT

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

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

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

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

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

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

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

  99. #   define OUT_FMT   AV_SAMPLE_FMT_FLT

  100. #   define OUT_FMT_P AV_SAMPLE_FMT_FLTP

  101. #else

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

  103. /* WARNING: only correct for positive numbers */

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

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

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

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

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

  109. #   define RENAME(a)      a ## _fixed

  110. #   define OUT_FMT   AV_SAMPLE_FMT_S16

  111. #   define OUT_FMT_P AV_SAMPLE_FMT_S16P

  112. #endif

  113. 

  114. /****************/

  115. 

  116. #define HEADER_SIZE 4

  117. 

  118. #include "mpegaudiodata.h"

  119. #include "mpegaudiodectab.h"

  120. 

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

  122. static VLC huff_vlc[16];

  123. static VLC_TYPE huff_vlc_tables[

  124.     0 + 128 + 128 + 128 + 130 + 128 + 154 + 166 +

  125.   142 + 204 + 190 + 170 + 542 + 460 + 662 + 414

  126.   ][2];

  127. static const int huff_vlc_tables_sizes[16] = {

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

  129.   142,  204,  190,  170,  542,  460,  662,  414

  130. };

  131. static VLC huff_quad_vlc[2];

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

  133. static const int huff_quad_vlc_tables_sizes[2] = { 128, 16 };

  134. /* computed from band_size_long */

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

  136. #include "mpegaudio_tablegen.h"

  137. /* intensity stereo coef table */

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

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

  140. static INTFLOAT csa_table[8][4];

  141. 

  142. static int16_t division_tab3[1<<6 ];

  143. static int16_t division_tab5[1<<8 ];

  144. static int16_t division_tab9[1<<11];

  145. 

  146. static int16_t * const division_tabs[4] = {

  147.     division_tab3, division_tab5, NULL, division_tab9

  148. };

  149. 

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

  151. static uint16_t scale_factor_modshift[64];

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

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

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

  155. 

  156. #define SCALE_GEN(v) \

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

  158. 

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

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

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

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

  163. };

  164. 

  165. /**

  166.  * Convert region offsets to region sizes and truncate

  167.  * size to big_values.

  168.  */

  169. static void region_offset2size(GranuleDef *g)

  170. {

  171.     int i, k, j = 0;

  172.     g->region_size[2] = 576 / 2;

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

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

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

  176.         j = k;

  177.     }

  178. }

  179. 

  180. static void init_short_region(MPADecodeContext *s, GranuleDef *g)

  181. {

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

  183.         if (s->sample_rate_index != 8)

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

  185.         else

  186.             g->region_size[0] = (72 / 2);

  187.     } else {

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

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

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

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

  192.         else

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

  194.     }

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

  196. }

  197. 

  198. static void init_long_region(MPADecodeContext *s, GranuleDef *g,

  199.                              int ra1, int ra2)

  200. {

  201.     int l;

  202.     g->region_size[0] = band_index_long[s->sample_rate_index][ra1 + 1] >> 1;

  203.     /* should not overflow */

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

  205.     g->region_size[1] = band_index_long[s->sample_rate_index][      l] >> 1;

  206. }

  207. 

  208. static void compute_band_indexes(MPADecodeContext *s, GranuleDef *g)

  209. {

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

  211.         if (g->switch_point) {

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

  213.                 long blocks.  For 8000Hz, we handle the 72 first

  214.                 exponents as long blocks */

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

  216.                 g->long_end = 8;

  217.             else

  218.                 g->long_end = 6;

  219. 

  220.             g->short_start = 3;

  221.         } else {

  222.             g->long_end    = 0;

  223.             g->short_start = 0;

  224.         }

  225.     } else {

  226.         g->short_start = 13;

  227.         g->long_end    = 22;

  228.     }

  229. }

  230. 

  231. /* layer 1 unscaling */

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

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

  234. {

  235.     int shift, mod;

  236.     int64_t val;

  237. 

  238.     shift   = scale_factor_modshift[scale_factor];

  239.     mod     = shift & 3;

  240.     shift >>= 2;

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

  242.     shift  += n;

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

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

  245. }

  246. 

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

  248. {

  249.     int shift, mod, val;

  250. 

  251.     shift   = scale_factor_modshift[scale_factor];

  252.     mod     = shift & 3;

  253.     shift >>= 2;

  254. 

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

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

  257.     if (shift > 0)

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

  259.     return val;

  260. }

  261. 

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

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

  264. {

  265.     unsigned int m;

  266.     int e;

  267. 

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

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

  270.     e -= exponent >> 2;

  271.     assert(e >= 1);

  272.     if (e > 31)

  273.         return 0;

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

  275. 

  276.     return m;

  277. }

  278. 

  279. static av_cold void decode_init_static(void)

  280. {

  281.     int i, j, k;

  282.     int offset;

  283. 

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

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

  286.         int shift, mod;

  287.         /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

  288.         shift = i / 3;

  289.         mod   = i % 3;

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

  291.     }

  292. 

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

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

  295.         int n, norm;

  296.         n = i + 2;

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

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

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

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

  301.         av_dlog(NULL, "%d: norm=%x s=%x %x %x\n", i, norm,

  302.                 scale_factor_mult[i][0],

  303.                 scale_factor_mult[i][1],

  304.                 scale_factor_mult[i][2]);

  305.     }

  306. 

  307.     RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));

  308. 

  309.     /* huffman decode tables */

  310.     offset = 0;

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

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

  313.         int xsize, x, y;

  314.         uint8_t  tmp_bits [512] = { 0 };

  315.         uint16_t tmp_codes[512] = { 0 };

  316. 

  317.         xsize = h->xsize;

  318. 

  319.         j = 0;

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

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

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

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

  324.             }

  325.         }

  326. 

  327.         /* XXX: fail test */

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

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

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

  331.                  tmp_bits, 1, 1, tmp_codes, 2, 2,

  332.                  INIT_VLC_USE_NEW_STATIC);

  333.         offset += huff_vlc_tables_sizes[i];

  334.     }

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

  336. 

  337.     offset = 0;

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

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

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

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

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

  343.                  INIT_VLC_USE_NEW_STATIC);

  344.         offset += huff_quad_vlc_tables_sizes[i];

  345.     }

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

  347. 

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

  349.         k = 0;

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

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

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

  353.         }

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

  355.     }

  356. 

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

  358. 

  359.     mpegaudio_tableinit();

  360. 

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

  362.         if (ff_mpa_quant_bits[i] < 0) {

  363.             for (j = 0; j < (1 << (-ff_mpa_quant_bits[i]+1)); j++) {

  364.                 int val1, val2, val3, steps;

  365.                 int val = j;

  366.                 steps   = ff_mpa_quant_steps[i];

  367.                 val1    = val % steps;

  368.                 val    /= steps;

  369.                 val2    = val % steps;

  370.                 val3    = val / steps;

  371.                 division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);

  372.             }

  373.         }

  374.     }

  375. 

  376. 

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

  378.         float f;

  379.         INTFLOAT v;

  380.         if (i != 6) {

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

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

  383.         } else {

  384.             v = FIXR(1.0);

  385.         }

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

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

  388.     }

  389.     /* invalid values */

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

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

  392. 

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

  394.         double f;

  395.         int e, k;

  396. 

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

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

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

  400.             k = i & 1;

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

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

  403.             av_dlog(NULL, "is_table_lsf %d %d: %f %f\n",

  404.                     i, j, (float) is_table_lsf[j][0][i],

  405.                     (float) is_table_lsf[j][1][i]);

  406.         }

  407.     }

  408. 

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

  410.         float ci, cs, ca;

  411.         ci = ci_table[i];

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

  413.         ca = cs * ci;

  414. #if !CONFIG_FLOAT

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

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

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

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

  419. #else

  420.         csa_table[i][0] = cs;

  421.         csa_table[i][1] = ca;

  422.         csa_table[i][2] = ca + cs;

  423.         csa_table[i][3] = ca - cs;

  424. #endif

  425.     }

  426. }

  427. 

  428. static av_cold int decode_init(AVCodecContext * avctx)

  429. {

  430.     static int initialized_tables = 0;

  431.     MPADecodeContext *s = avctx->priv_data;

  432. 

  433.     if (!initialized_tables) {

  434.         decode_init_static();

  435.         initialized_tables = 1;

  436.     }

  437. 

  438.     s->avctx = avctx;

  439. 

  440.     avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);

  441.     ff_mpadsp_init(&s->mpadsp);

  442. 

  443.     if (avctx->request_sample_fmt == OUT_FMT &&

  444.         avctx->codec_id != AV_CODEC_ID_MP3ON4)

  445.         avctx->sample_fmt = OUT_FMT;

  446.     else

  447.         avctx->sample_fmt = OUT_FMT_P;

  448.     s->err_recognition = avctx->err_recognition;

  449. 

  450.     if (avctx->codec_id == AV_CODEC_ID_MP3ADU)

  451.         s->adu_mode = 1;

  452. 

  453.     return 0;

  454. }

  455. 

  456. #define C3 FIXHR(0.86602540378443864676/2)

  457. #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)

  458. #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)

  459. #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)

  460. 

  461. /* 12 points IMDCT. We compute it "by hand" by factorizing obvious

  462.    cases. */

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

  464. {

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

  466. 

  467.     in0  = in[0*3];

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

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

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

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

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

  473.     in5 += in3;

  474.     in3 += in1;

  475. 

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

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

  478. 

  479.     t1   = in0 - in4;

  480.     t2   = MULH3(in1 - in5, C4, 2);

  481. 

  482.     out[ 7] =

  483.     out[10] = t1 + t2;

  484.     out[ 1] =

  485.     out[ 4] = t1 - t2;

  486. 

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

  488.     in4     = in0 + in2;

  489.     in5    += 2*in1;

  490.     in1     = MULH3(in5 + in3, C5, 1);

  491.     out[ 8] =

  492.     out[ 9] = in4 + in1;

  493.     out[ 2] =

  494.     out[ 3] = in4 - in1;

  495. 

  496.     in0    -= in2;

  497.     in5     = MULH3(in5 - in3, C6, 2);

  498.     out[ 0] =

  499.     out[ 5] = in0 - in5;

  500.     out[ 6] =

  501.     out[11] = in0 + in5;

  502. }

  503. 

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

  505. static int mp_decode_layer1(MPADecodeContext *s)

  506. {

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

  508.     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];

  509.     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

  510. 

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

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

  513.     else

  514.         bound = SBLIMIT;

  515. 

  516.     /* allocation bits */

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

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

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

  520.         }

  521.     }

  522.     for (i = bound; i < SBLIMIT; i++)

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

  524. 

  525.     /* scale factors */

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

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

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

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

  530.         }

  531.     }

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

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

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

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

  536.         }

  537.     }

  538. 

  539.     /* compute samples */

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

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

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

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

  544.                 if (n) {

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

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

  547.                 } else {

  548.                     v = 0;

  549.                 }

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

  551.             }

  552.         }

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

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

  555.             if (n) {

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

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

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

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

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

  561.             } else {

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

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

  564.             }

  565.         }

  566.     }

  567.     return 12;

  568. }

  569. 

  570. static int mp_decode_layer2(MPADecodeContext *s)

  571. {

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

  573.     const unsigned char *alloc_table;

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

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

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

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

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

  579. 

  580.     /* select decoding table */

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

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

  583.     sblimit     = ff_mpa_sblimit_table[table];

  584.     alloc_table = ff_mpa_alloc_tables[table];

  585. 

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

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

  588.     else

  589.         bound = sblimit;

  590. 

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

  592. 

  593.     /* sanity check */

  594.     if (bound > sblimit)

  595.         bound = sblimit;

  596. 

  597.     /* parse bit allocation */

  598.     j = 0;

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

  600.         bit_alloc_bits = alloc_table[j];

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

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

  603.         j += 1 << bit_alloc_bits;

  604.     }

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

  606.         bit_alloc_bits = alloc_table[j];

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

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

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

  610.         j += 1 << bit_alloc_bits;

  611.     }

  612. 

  613.     /* scale codes */

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

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

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

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

  618.         }

  619.     }

  620. 

  621.     /* scale factors */

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

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

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

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

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

  627.                 default:

  628.                 case 0:

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

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

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

  632.                     break;

  633.                 case 2:

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

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

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

  637.                     break;

  638.                 case 1:

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

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

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

  642.                     break;

  643.                 case 3:

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

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

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

  647.                     break;

  648.                 }

  649.             }

  650.         }

  651.     }

  652. 

  653.     /* samples */

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

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

  656.             j = 0;

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

  658.                 bit_alloc_bits = alloc_table[j];

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

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

  661.                     if (b) {

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

  663.                         qindex = alloc_table[j+b];

  664.                         bits = ff_mpa_quant_bits[qindex];

  665.                         if (bits < 0) {

  666.                             int v2;

  667.                             /* 3 values at the same time */

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

  669.                             v2 = division_tabs[qindex][v];

  670.                             steps  = ff_mpa_quant_steps[qindex];

  671. 

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

  673.                                 l2_unscale_group(steps,  v2       & 15, scale);

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

  675.                                 l2_unscale_group(steps, (v2 >> 4) & 15, scale);

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

  677.                                 l2_unscale_group(steps,  v2 >> 8      , scale);

  678.                         } else {

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

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

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

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

  683.                             }

  684.                         }

  685.                     } else {

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

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

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

  689.                     }

  690.                 }

  691.                 /* next subband in alloc table */

  692.                 j += 1 << bit_alloc_bits;

  693.             }

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

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

  696.                 bit_alloc_bits = alloc_table[j];

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

  698.                 if (b) {

  699.                     int mant, scale0, scale1;

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

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

  702.                     qindex = alloc_table[j+b];

  703.                     bits = ff_mpa_quant_bits[qindex];

  704.                     if (bits < 0) {

  705.                         /* 3 values at the same time */

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

  707.                         steps = ff_mpa_quant_steps[qindex];

  708.                         mant = v % steps;

  709.                         v = v / steps;

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

  711.                             l2_unscale_group(steps, mant, scale0);

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

  713.                             l2_unscale_group(steps, mant, scale1);

  714.                         mant = v % steps;

  715.                         v = v / steps;

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

  717.                             l2_unscale_group(steps, mant, scale0);

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

  719.                             l2_unscale_group(steps, mant, scale1);

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

  721.                             l2_unscale_group(steps, v, scale0);

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

  723.                             l2_unscale_group(steps, v, scale1);

  724.                     } else {

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

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

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

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

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

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

  731.                         }

  732.                     }

  733.                 } else {

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

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

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

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

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

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

  740.                 }

  741.                 /* next subband in alloc table */

  742.                 j += 1 << bit_alloc_bits;

  743.             }

  744.             /* fill remaining samples to zero */

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

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

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

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

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

  750.                 }

  751.             }

  752.         }

  753.     }

  754.     return 3 * 12;

  755. }

  756. 

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

  758.     if (n == 3) {                   \

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

  760.         dst   = sf - 3 * m;         \

  761.         sf    = m;                  \

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

  763.         dst  = sf & 3;              \

  764.         sf >>= 2;                   \

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

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

  767.         dst   = sf - 5 * m;         \

  768.         sf    = m;                  \

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

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

  771.         dst   = sf - 6 * m;         \

  772.         sf    = m;                  \

  773.     } else {                        \

  774.         dst = 0;                    \

  775.     }

  776. 

  777. static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,

  778.                                            int n3)

  779. {

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

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

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

  783.     slen[0] = sf;

  784. }

  785. 

  786. static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,

  787.                                          int16_t *exponents)

  788. {

  789.     const uint8_t *bstab, *pretab;

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

  791.     int16_t *exp_ptr;

  792. 

  793.     exp_ptr = exponents;

  794.     gain    = g->global_gain - 210;

  795.     shift   = g->scalefac_scale + 1;

  796. 

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

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

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

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

  801.         len = bstab[i];

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

  803.             *exp_ptr++ = v0;

  804.     }

  805. 

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

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

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

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

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

  811.         k        = g->long_end;

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

  813.             len = bstab[i];

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

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

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

  817.                     *exp_ptr++ = v0;

  818.             }

  819.         }

  820.     }

  821. }

  822. 

  823. /* handle n = 0 too */

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

  825. {

  826.     return n ? get_bits(s, n) : 0;

  827. }

  828. 

  829. 

  830. static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,

  831.                           int *end_pos2)

  832. {

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

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

  835.         s->in_gb.buffer = NULL;

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

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

  838.         *end_pos2 =

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

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

  841.     }

  842. }

  843. 

  844. /* Following is a optimized code for

  845.             INTFLOAT v = *src

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

  847.                 v = -v;

  848.             *dst = v;

  849. */

  850. #if CONFIG_FLOAT

  851. #define READ_FLIP_SIGN(dst,src)                     \

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

  853.     AV_WN32A(dst, v);

  854. #else

  855. #define READ_FLIP_SIGN(dst,src)     \

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

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

  858. #endif

  859. 

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

  861.                           int16_t *exponents, int end_pos2)

  862. {

  863.     int s_index;

  864.     int i;

  865.     int last_pos, bits_left;

  866.     VLC *vlc;

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

  868. 

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

  870.     s_index = 0;

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

  872.         int j, k, l, linbits;

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

  874.         if (j == 0)

  875.             continue;

  876.         /* select vlc table */

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

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

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

  880.         vlc     = &huff_vlc[l];

  881. 

  882.         if (!l) {

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

  884.             s_index += 2 * j;

  885.             continue;

  886.         }

  887. 

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

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

  890.             int exponent, x, y;

  891.             int v;

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

  893. 

  894.             if (pos >= end_pos){

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

  896.                 if (pos >= end_pos)

  897.                     break;

  898.             }

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

  900. 

  901.             if (!y) {

  902.                 g->sb_hybrid[s_index  ] =

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

  904.                 s_index += 2;

  905.                 continue;

  906.             }

  907. 

  908.             exponent= exponents[s_index];

  909. 

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

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

  912.             if (y & 16) {

  913.                 x = y >> 5;

  914.                 y = y & 0x0f;

  915.                 if (x < 15) {

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

  917.                 } else {

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

  919.                     v  = l3_unscale(x, exponent);

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

  921.                         v = -v;

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

  923.                 }

  924.                 if (y < 15) {

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

  926.                 } else {

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

  928.                     v  = l3_unscale(y, exponent);

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

  930.                         v = -v;

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

  932.                 }

  933.             } else {

  934.                 x = y >> 5;

  935.                 y = y & 0x0f;

  936.                 x += y;

  937.                 if (x < 15) {

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

  939.                 } else {

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

  941.                     v  = l3_unscale(x, exponent);

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

  943.                         v = -v;

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

  945.                 }

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

  947.             }

  948.             s_index += 2;

  949.         }

  950.     }

  951. 

  952.     /* high frequencies */

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

  954.     last_pos = 0;

  955.     while (s_index <= 572) {

  956.         int pos, code;

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

  958.         if (pos >= end_pos) {

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

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

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

  962.                 s_index -= 4;

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

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

  965.                 if(s->err_recognition & AV_EF_BITSTREAM)

  966.                     s_index=0;

  967.                 break;

  968.             }

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

  970.             if (pos >= end_pos)

  971.                 break;

  972.         }

  973.         last_pos = pos;

  974. 

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

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

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

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

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

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

  981.         while (code) {

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

  983.             int v;

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

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

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

  987.         }

  988.         s_index += 4;

  989.     }

  990.     /* skip extension bits */

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

  992.     if (bits_left < 0 && (s->err_recognition & AV_EF_BUFFER)) {

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

  994.         s_index=0;

  995.     } else if (bits_left > 0 && (s->err_recognition & AV_EF_BUFFER)) {

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

  997.         s_index = 0;

  998.     }

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

  1000.     skip_bits_long(&s->gb, bits_left);

  1001. 

  1002.     i = get_bits_count(&s->gb);

  1003.     switch_buffer(s, &i, &end_pos, &end_pos2);

  1004. 

  1005.     return 0;

  1006. }

  1007. 

  1008. /* Reorder short blocks from bitstream order to interleaved order. It

  1009.    would be faster to do it in parsing, but the code would be far more

  1010.    complicated */

  1011. static void reorder_block(MPADecodeContext *s, GranuleDef *g)

  1012. {

  1013.     int i, j, len;

  1014.     INTFLOAT *ptr, *dst, *ptr1;

  1015.     INTFLOAT tmp[576];

  1016. 

  1017.     if (g->block_type != 2)

  1018.         return;

  1019. 

  1020.     if (g->switch_point) {

  1021.         if (s->sample_rate_index != 8)

  1022.             ptr = g->sb_hybrid + 36;

  1023.         else

  1024.             ptr = g->sb_hybrid + 72;

  1025.     } else {

  1026.         ptr = g->sb_hybrid;

  1027.     }

  1028. 

  1029.     for (i = g->short_start; i < 13; i++) {

  1030.         len  = band_size_short[s->sample_rate_index][i];

  1031.         ptr1 = ptr;

  1032.         dst  = tmp;

  1033.         for (j = len; j > 0; j--) {

  1034.             *dst++ = ptr[0*len];

  1035.             *dst++ = ptr[1*len];

  1036.             *dst++ = ptr[2*len];

  1037.             ptr++;

  1038.         }

  1039.         ptr += 2 * len;

  1040.         memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));

  1041.     }

  1042. }

  1043. 

  1044. #define ISQRT2 FIXR(0.70710678118654752440)

  1045. 

  1046. static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)

  1047. {

  1048.     int i, j, k, l;

  1049.     int sf_max, sf, len, non_zero_found;

  1050.     INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;

  1051.     int non_zero_found_short[3];

  1052. 

  1053.     /* intensity stereo */

  1054.     if (s->mode_ext & MODE_EXT_I_STEREO) {

  1055.         if (!s->lsf) {

  1056.             is_tab = is_table;

  1057.             sf_max = 7;

  1058.         } else {

  1059.             is_tab = is_table_lsf[g1->scalefac_compress & 1];

  1060.             sf_max = 16;

  1061.         }

  1062. 

  1063.         tab0 = g0->sb_hybrid + 576;

  1064.         tab1 = g1->sb_hybrid + 576;

  1065. 

  1066.         non_zero_found_short[0] = 0;

  1067.         non_zero_found_short[1] = 0;

  1068.         non_zero_found_short[2] = 0;

  1069.         k = (13 - g1->short_start) * 3 + g1->long_end - 3;

  1070.         for (i = 12; i >= g1->short_start; i--) {

  1071.             /* for last band, use previous scale factor */

  1072.             if (i != 11)

  1073.                 k -= 3;

  1074.             len = band_size_short[s->sample_rate_index][i];

  1075.             for (l = 2; l >= 0; l--) {

  1076.                 tab0 -= len;

  1077.                 tab1 -= len;

  1078.                 if (!non_zero_found_short[l]) {

  1079.                     /* test if non zero band. if so, stop doing i-stereo */

  1080.                     for (j = 0; j < len; j++) {

  1081.                         if (tab1[j] != 0) {

  1082.                             non_zero_found_short[l] = 1;

  1083.                             goto found1;

  1084.                         }

  1085.                     }

  1086.                     sf = g1->scale_factors[k + l];

  1087.                     if (sf >= sf_max)

  1088.                         goto found1;

  1089. 

  1090.                     v1 = is_tab[0][sf];

  1091.                     v2 = is_tab[1][sf];

  1092.                     for (j = 0; j < len; j++) {

  1093.                         tmp0    = tab0[j];

  1094.                         tab0[j] = MULLx(tmp0, v1, FRAC_BITS);

  1095.                         tab1[j] = MULLx(tmp0, v2, FRAC_BITS);

  1096.                     }

  1097.                 } else {

  1098. found1:

  1099.                     if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1100.                         /* lower part of the spectrum : do ms stereo

  1101.                            if enabled */

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

  1103.                             tmp0    = tab0[j];

  1104.                             tmp1    = tab1[j];

  1105.                             tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1106.                             tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1107.                         }

  1108.                     }

  1109.                 }

  1110.             }

  1111.         }

  1112. 

  1113.         non_zero_found = non_zero_found_short[0] |

  1114.                          non_zero_found_short[1] |

  1115.                          non_zero_found_short[2];

  1116. 

  1117.         for (i = g1->long_end - 1;i >= 0;i--) {

  1118.             len   = band_size_long[s->sample_rate_index][i];

  1119.             tab0 -= len;

  1120.             tab1 -= len;

  1121.             /* test if non zero band. if so, stop doing i-stereo */

  1122.             if (!non_zero_found) {

  1123.                 for (j = 0; j < len; j++) {

  1124.                     if (tab1[j] != 0) {

  1125.                         non_zero_found = 1;

  1126.                         goto found2;

  1127.                     }

  1128.                 }

  1129.                 /* for last band, use previous scale factor */

  1130.                 k  = (i == 21) ? 20 : i;

  1131.                 sf = g1->scale_factors[k];

  1132.                 if (sf >= sf_max)

  1133.                     goto found2;

  1134.                 v1 = is_tab[0][sf];

  1135.                 v2 = is_tab[1][sf];

  1136.                 for (j = 0; j < len; j++) {

  1137.                     tmp0    = tab0[j];

  1138.                     tab0[j] = MULLx(tmp0, v1, FRAC_BITS);

  1139.                     tab1[j] = MULLx(tmp0, v2, FRAC_BITS);

  1140.                 }

  1141.             } else {

  1142. found2:

  1143.                 if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1144.                     /* lower part of the spectrum : do ms stereo

  1145.                        if enabled */

  1146.                     for (j = 0; j < len; j++) {

  1147.                         tmp0    = tab0[j];

  1148.                         tmp1    = tab1[j];

  1149.                         tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1150.                         tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1151.                     }

  1152.                 }

  1153.             }

  1154.         }

  1155.     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1156.         /* ms stereo ONLY */

  1157.         /* NOTE: the 1/sqrt(2) normalization factor is included in the

  1158.            global gain */

  1159. #if CONFIG_FLOAT

  1160.        s->fdsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);

  1161. #else

  1162.         tab0 = g0->sb_hybrid;

  1163.         tab1 = g1->sb_hybrid;

  1164.         for (i = 0; i < 576; i++) {

  1165.             tmp0    = tab0[i];

  1166.             tmp1    = tab1[i];

  1167.             tab0[i] = tmp0 + tmp1;

  1168.             tab1[i] = tmp0 - tmp1;

  1169.         }

  1170. #endif

  1171.     }

  1172. }

  1173. 

  1174. #if CONFIG_FLOAT

  1175. #define AA(j) do {                                                      \

  1176.         float tmp0 = ptr[-1-j];                                         \

  1177.         float tmp1 = ptr[   j];                                         \

  1178.         ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1];    \

  1179.         ptr[   j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0];    \

  1180.     } while (0)

  1181. #else

  1182. #define AA(j) do {                                              \

  1183.         int tmp0 = ptr[-1-j];                                   \

  1184.         int tmp1 = ptr[   j];                                   \

  1185.         int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]);          \

  1186.         ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2]));   \

  1187.         ptr[   j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3]));   \

  1188.     } while (0)

  1189. #endif

  1190. 

  1191. static void compute_antialias(MPADecodeContext *s, GranuleDef *g)

  1192. {

  1193.     INTFLOAT *ptr;

  1194.     int n, i;

  1195. 

  1196.     /* we antialias only "long" bands */

  1197.     if (g->block_type == 2) {

  1198.         if (!g->switch_point)

  1199.             return;

  1200.         /* XXX: check this for 8000Hz case */

  1201.         n = 1;

  1202.     } else {

  1203.         n = SBLIMIT - 1;

  1204.     }

  1205. 

  1206.     ptr = g->sb_hybrid + 18;

  1207.     for (i = n; i > 0; i--) {

  1208.         AA(0);

  1209.         AA(1);

  1210.         AA(2);

  1211.         AA(3);

  1212.         AA(4);

  1213.         AA(5);

  1214.         AA(6);

  1215.         AA(7);

  1216. 

  1217.         ptr += 18;

  1218.     }

  1219. }

  1220. 

  1221. static void compute_imdct(MPADecodeContext *s, GranuleDef *g,

  1222.                           INTFLOAT *sb_samples, INTFLOAT *mdct_buf)

  1223. {

  1224.     INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;

  1225.     INTFLOAT out2[12];

  1226.     int i, j, mdct_long_end, sblimit;

  1227. 

  1228.     /* find last non zero block */

  1229.     ptr  = g->sb_hybrid + 576;

  1230.     ptr1 = g->sb_hybrid + 2 * 18;

  1231.     while (ptr >= ptr1) {

  1232.         int32_t *p;

  1233.         ptr -= 6;

  1234.         p    = (int32_t*)ptr;

  1235.         if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])

  1236.             break;

  1237.     }

  1238.     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

  1239. 

  1240.     if (g->block_type == 2) {

  1241.         /* XXX: check for 8000 Hz */

  1242.         if (g->switch_point)

  1243.             mdct_long_end = 2;

  1244.         else

  1245.             mdct_long_end = 0;

  1246.     } else {

  1247.         mdct_long_end = sblimit;

  1248.     }

  1249. 

  1250.     s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,

  1251.                                      mdct_long_end, g->switch_point,

  1252.                                      g->block_type);

  1253. 

  1254.     buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);

  1255.     ptr = g->sb_hybrid + 18 * mdct_long_end;

  1256. 

  1257.     for (j = mdct_long_end; j < sblimit; j++) {

  1258.         /* select frequency inversion */

  1259.         win     = RENAME(ff_mdct_win)[2 + (4  & -(j & 1))];

  1260.         out_ptr = sb_samples + j;

  1261. 

  1262.         for (i = 0; i < 6; i++) {

  1263.             *out_ptr = buf[4*i];

  1264.             out_ptr += SBLIMIT;

  1265.         }

  1266.         imdct12(out2, ptr + 0);

  1267.         for (i = 0; i < 6; i++) {

  1268.             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*1)];

  1269.             buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);

  1270.             out_ptr += SBLIMIT;

  1271.         }

  1272.         imdct12(out2, ptr + 1);

  1273.         for (i = 0; i < 6; i++) {

  1274.             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*2)];

  1275.             buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);

  1276.             out_ptr += SBLIMIT;

  1277.         }

  1278.         imdct12(out2, ptr + 2);

  1279.         for (i = 0; i < 6; i++) {

  1280.             buf[4*(i + 6*0)] = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*0)];

  1281.             buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);

  1282.             buf[4*(i + 6*2)] = 0;

  1283.         }

  1284.         ptr += 18;

  1285.         buf += (j&3) != 3 ? 1 : (4*18-3);

  1286.     }

  1287.     /* zero bands */

  1288.     for (j = sblimit; j < SBLIMIT; j++) {

  1289.         /* overlap */

  1290.         out_ptr = sb_samples + j;

  1291.         for (i = 0; i < 18; i++) {

  1292.             *out_ptr = buf[4*i];

  1293.             buf[4*i]   = 0;

  1294.             out_ptr += SBLIMIT;

  1295.         }

  1296.         buf += (j&3) != 3 ? 1 : (4*18-3);

  1297.     }

  1298. }

  1299. 

  1300. /* main layer3 decoding function */

  1301. static int mp_decode_layer3(MPADecodeContext *s)

  1302. {

  1303.     int nb_granules, main_data_begin;

  1304.     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;

  1305.     GranuleDef *g;

  1306.     int16_t exponents[576]; //FIXME try INTFLOAT

  1307. 

  1308.     /* read side info */

  1309.     if (s->lsf) {

  1310.         main_data_begin = get_bits(&s->gb, 8);

  1311.         skip_bits(&s->gb, s->nb_channels);

  1312.         nb_granules = 1;

  1313.     } else {

  1314.         main_data_begin = get_bits(&s->gb, 9);

  1315.         if (s->nb_channels == 2)

  1316.             skip_bits(&s->gb, 3);

  1317.         else

  1318.             skip_bits(&s->gb, 5);

  1319.         nb_granules = 2;

  1320.         for (ch = 0; ch < s->nb_channels; ch++) {

  1321.             s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */

  1322.             s->granules[ch][1].scfsi = get_bits(&s->gb, 4);

  1323.         }

  1324.     }

  1325. 

  1326.     for (gr = 0; gr < nb_granules; gr++) {

  1327.         for (ch = 0; ch < s->nb_channels; ch++) {

  1328.             av_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);

  1329.             g = &s->granules[ch][gr];

  1330.             g->part2_3_length = get_bits(&s->gb, 12);

  1331.             g->big_values     = get_bits(&s->gb,  9);

  1332.             if (g->big_values > 288) {

  1333.                 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");

  1334.                 return AVERROR_INVALIDDATA;

  1335.             }

  1336. 

  1337.             g->global_gain = get_bits(&s->gb, 8);

  1338.             /* if MS stereo only is selected, we precompute the

  1339.                1/sqrt(2) renormalization factor */

  1340.             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==

  1341.                 MODE_EXT_MS_STEREO)

  1342.                 g->global_gain -= 2;

  1343.             if (s->lsf)

  1344.                 g->scalefac_compress = get_bits(&s->gb, 9);

  1345.             else

  1346.                 g->scalefac_compress = get_bits(&s->gb, 4);

  1347.             blocksplit_flag = get_bits1(&s->gb);

  1348.             if (blocksplit_flag) {

  1349.                 g->block_type = get_bits(&s->gb, 2);

  1350.                 if (g->block_type == 0) {

  1351.                     av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");

  1352.                     return AVERROR_INVALIDDATA;

  1353.                 }

  1354.                 g->switch_point = get_bits1(&s->gb);

  1355.                 for (i = 0; i < 2; i++)

  1356.                     g->table_select[i] = get_bits(&s->gb, 5);

  1357.                 for (i = 0; i < 3; i++)

  1358.                     g->subblock_gain[i] = get_bits(&s->gb, 3);

  1359.                 init_short_region(s, g);

  1360.             } else {

  1361.                 int region_address1, region_address2;

  1362.                 g->block_type = 0;

  1363.                 g->switch_point = 0;

  1364.                 for (i = 0; i < 3; i++)

  1365.                     g->table_select[i] = get_bits(&s->gb, 5);

  1366.                 /* compute huffman coded region sizes */

  1367.                 region_address1 = get_bits(&s->gb, 4);

  1368.                 region_address2 = get_bits(&s->gb, 3);

  1369.                 av_dlog(s->avctx, "region1=%d region2=%d\n",

  1370.                         region_address1, region_address2);

  1371.                 init_long_region(s, g, region_address1, region_address2);

  1372.             }

  1373.             region_offset2size(g);

  1374.             compute_band_indexes(s, g);

  1375. 

  1376.             g->preflag = 0;

  1377.             if (!s->lsf)

  1378.                 g->preflag = get_bits1(&s->gb);

  1379.             g->scalefac_scale     = get_bits1(&s->gb);

  1380.             g->count1table_select = get_bits1(&s->gb);

  1381.             av_dlog(s->avctx, "block_type=%d switch_point=%d\n",

  1382.                     g->block_type, g->switch_point);

  1383.         }

  1384.     }

  1385. 

  1386.     if (!s->adu_mode) {

  1387.         int skip;

  1388.         const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);

  1389.         int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0,

  1390.                                 FFMAX(0, LAST_BUF_SIZE - s->last_buf_size));

  1391.         assert((get_bits_count(&s->gb) & 7) == 0);

  1392.         /* now we get bits from the main_data_begin offset */

  1393.         av_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",

  1394.                 main_data_begin, s->last_buf_size);

  1395. 

  1396.         memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);

  1397.         s->in_gb = s->gb;

  1398.         init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);

  1399. #if !UNCHECKED_BITSTREAM_READER

  1400.         s->gb.size_in_bits_plus8 += extrasize * 8;

  1401. #endif

  1402.         s->last_buf_size <<= 3;

  1403.         for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {

  1404.             for (ch = 0; ch < s->nb_channels; ch++) {

  1405.                 g = &s->granules[ch][gr];

  1406.                 s->last_buf_size += g->part2_3_length;

  1407.                 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));

  1408.                 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);

  1409.             }

  1410.         }

  1411.         skip = s->last_buf_size - 8 * main_data_begin;

  1412.         if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {

  1413.             skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);

  1414.             s->gb           = s->in_gb;

  1415.             s->in_gb.buffer = NULL;

  1416.         } else {

  1417.             skip_bits_long(&s->gb, skip);

  1418.         }

  1419.     } else {

  1420.         gr = 0;

  1421.     }

  1422. 

  1423.     for (; gr < nb_granules; gr++) {

  1424.         for (ch = 0; ch < s->nb_channels; ch++) {

  1425.             g = &s->granules[ch][gr];

  1426.             bits_pos = get_bits_count(&s->gb);

  1427. 

  1428.             if (!s->lsf) {

  1429.                 uint8_t *sc;

  1430.                 int slen, slen1, slen2;

  1431. 

  1432.                 /* MPEG1 scale factors */

  1433.                 slen1 = slen_table[0][g->scalefac_compress];

  1434.                 slen2 = slen_table[1][g->scalefac_compress];

  1435.                 av_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);

  1436.                 if (g->block_type == 2) {

  1437.                     n = g->switch_point ? 17 : 18;

  1438.                     j = 0;

  1439.                     if (slen1) {

  1440.                         for (i = 0; i < n; i++)

  1441.                             g->scale_factors[j++] = get_bits(&s->gb, slen1);

  1442.                     } else {

  1443.                         for (i = 0; i < n; i++)

  1444.                             g->scale_factors[j++] = 0;

  1445.                     }

  1446.                     if (slen2) {

  1447.                         for (i = 0; i < 18; i++)

  1448.                             g->scale_factors[j++] = get_bits(&s->gb, slen2);

  1449.                         for (i = 0; i < 3; i++)

  1450.                             g->scale_factors[j++] = 0;

  1451.                     } else {

  1452.                         for (i = 0; i < 21; i++)

  1453.                             g->scale_factors[j++] = 0;

  1454.                     }

  1455.                 } else {

  1456.                     sc = s->granules[ch][0].scale_factors;

  1457.                     j = 0;

  1458.                     for (k = 0; k < 4; k++) {

  1459.                         n = k == 0 ? 6 : 5;

  1460.                         if ((g->scfsi & (0x8 >> k)) == 0) {

  1461.                             slen = (k < 2) ? slen1 : slen2;

  1462.                             if (slen) {

  1463.                                 for (i = 0; i < n; i++)

  1464.                                     g->scale_factors[j++] = get_bits(&s->gb, slen);

  1465.                             } else {

  1466.                                 for (i = 0; i < n; i++)

  1467.                                     g->scale_factors[j++] = 0;

  1468.                             }

  1469.                         } else {

  1470.                             /* simply copy from last granule */

  1471.                             for (i = 0; i < n; i++) {

  1472.                                 g->scale_factors[j] = sc[j];

  1473.                                 j++;

  1474.                             }

  1475.                         }

  1476.                     }

  1477.                     g->scale_factors[j++] = 0;

  1478.                 }

  1479.             } else {

  1480.                 int tindex, tindex2, slen[4], sl, sf;

  1481. 

  1482.                 /* LSF scale factors */

  1483.                 if (g->block_type == 2)

  1484.                     tindex = g->switch_point ? 2 : 1;

  1485.                 else

  1486.                     tindex = 0;

  1487. 

  1488.                 sf = g->scalefac_compress;

  1489.                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {

  1490.                     /* intensity stereo case */

  1491.                     sf >>= 1;

  1492.                     if (sf < 180) {

  1493.                         lsf_sf_expand(slen, sf, 6, 6, 0);

  1494.                         tindex2 = 3;

  1495.                     } else if (sf < 244) {

  1496.                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);

  1497.                         tindex2 = 4;

  1498.                     } else {

  1499.                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);

  1500.                         tindex2 = 5;

  1501.                     }

  1502.                 } else {

  1503.                     /* normal case */

  1504.                     if (sf < 400) {

  1505.                         lsf_sf_expand(slen, sf, 5, 4, 4);

  1506.                         tindex2 = 0;

  1507.                     } else if (sf < 500) {

  1508.                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);

  1509.                         tindex2 = 1;

  1510.                     } else {

  1511.                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);

  1512.                         tindex2 = 2;

  1513.                         g->preflag = 1;

  1514.                     }

  1515.                 }

  1516. 

  1517.                 j = 0;

  1518.                 for (k = 0; k < 4; k++) {

  1519.                     n  = lsf_nsf_table[tindex2][tindex][k];

  1520.                     sl = slen[k];

  1521.                     if (sl) {

  1522.                         for (i = 0; i < n; i++)

  1523.                             g->scale_factors[j++] = get_bits(&s->gb, sl);

  1524.                     } else {

  1525.                         for (i = 0; i < n; i++)

  1526.                             g->scale_factors[j++] = 0;

  1527.                     }

  1528.                 }

  1529.                 /* XXX: should compute exact size */

  1530.                 for (; j < 40; j++)

  1531.                     g->scale_factors[j] = 0;

  1532.             }

  1533. 

  1534.             exponents_from_scale_factors(s, g, exponents);

  1535. 

  1536.             /* read Huffman coded residue */

  1537.             huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);

  1538.         } /* ch */

  1539. 

  1540.         if (s->mode == MPA_JSTEREO)

  1541.             compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);

  1542. 

  1543.         for (ch = 0; ch < s->nb_channels; ch++) {

  1544.             g = &s->granules[ch][gr];

  1545. 

  1546.             reorder_block(s, g);

  1547.             compute_antialias(s, g);

  1548.             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);

  1549.         }

  1550.     } /* gr */

  1551.     if (get_bits_count(&s->gb) < 0)

  1552.         skip_bits_long(&s->gb, -get_bits_count(&s->gb));

  1553.     return nb_granules * 18;

  1554. }

  1555. 

  1556. static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,

  1557.                            const uint8_t *buf, int buf_size)

  1558. {

  1559.     int i, nb_frames, ch, ret;

  1560.     OUT_INT *samples_ptr;

  1561. 

  1562.     init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);

  1563. 

  1564.     /* skip error protection field */

  1565.     if (s->error_protection)

  1566.         skip_bits(&s->gb, 16);

  1567. 

  1568.     switch(s->layer) {

  1569.     case 1:

  1570.         s->avctx->frame_size = 384;

  1571.         nb_frames = mp_decode_layer1(s);

  1572.         break;

  1573.     case 2:

  1574.         s->avctx->frame_size = 1152;

  1575.         nb_frames = mp_decode_layer2(s);

  1576.         break;

  1577.     case 3:

  1578.         s->avctx->frame_size = s->lsf ? 576 : 1152;

  1579.     default:

  1580.         nb_frames = mp_decode_layer3(s);

  1581. 

  1582.         if (nb_frames < 0)

  1583.             return nb_frames;

  1584. 

  1585.         s->last_buf_size=0;

  1586.         if (s->in_gb.buffer) {

  1587.             align_get_bits(&s->gb);

  1588.             i = get_bits_left(&s->gb)>>3;

  1589.             if (i >= 0 && i <= BACKSTEP_SIZE) {

  1590.                 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);

  1591.                 s->last_buf_size=i;

  1592.             } else

  1593.                 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);

  1594.             s->gb           = s->in_gb;

  1595.             s->in_gb.buffer = NULL;

  1596.         }

  1597. 

  1598.         align_get_bits(&s->gb);

  1599.         assert((get_bits_count(&s->gb) & 7) == 0);

  1600.         i = get_bits_left(&s->gb) >> 3;

  1601. 

  1602.         if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {

  1603.             if (i < 0)

  1604.                 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);

  1605.             i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);

  1606.         }

  1607.         assert(i <= buf_size - HEADER_SIZE && i >= 0);

  1608.         memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);

  1609.         s->last_buf_size += i;

  1610.     }

  1611. 

  1612.     /* get output buffer */

  1613.     if (!samples) {

  1614.         av_assert0(s->frame != NULL);

  1615.         s->frame->nb_samples = s->avctx->frame_size;

  1616.         if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0) {

  1617.             av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");

  1618.             return ret;

  1619.         }

  1620.         samples = (OUT_INT **)s->frame->extended_data;

  1621.     }

  1622. 

  1623.     /* apply the synthesis filter */

  1624.     for (ch = 0; ch < s->nb_channels; ch++) {

  1625.         int sample_stride;

  1626.         if (s->avctx->sample_fmt == OUT_FMT_P) {

  1627.             samples_ptr   = samples[ch];

  1628.             sample_stride = 1;

  1629.         } else {

  1630.             samples_ptr   = samples[0] + ch;

  1631.             sample_stride = s->nb_channels;

  1632.         }

  1633.         for (i = 0; i < nb_frames; i++) {

  1634.             RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],

  1635.                                         &(s->synth_buf_offset[ch]),

  1636.                                         RENAME(ff_mpa_synth_window),

  1637.                                         &s->dither_state, samples_ptr,

  1638.                                         sample_stride, s->sb_samples[ch][i]);

  1639.             samples_ptr += 32 * sample_stride;

  1640.         }

  1641.     }

  1642. 

  1643.     return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;

  1644. }

  1645. 

  1646. static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,

  1647.                         AVPacket *avpkt)

  1648. {

  1649.     const uint8_t *buf  = avpkt->data;

  1650.     int buf_size        = avpkt->size;

  1651.     MPADecodeContext *s = avctx->priv_data;

  1652.     uint32_t header;

  1653.     int ret;

  1654. 

  1655.     if (buf_size < HEADER_SIZE)

  1656.         return AVERROR_INVALIDDATA;

  1657. 

  1658.     header = AV_RB32(buf);

  1659.     if (ff_mpa_check_header(header) < 0) {

  1660.         av_log(avctx, AV_LOG_ERROR, "Header missing\n");

  1661.         return AVERROR_INVALIDDATA;

  1662.     }

  1663. 

  1664.     if (avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {

  1665.         /* free format: prepare to compute frame size */

  1666.         s->frame_size = -1;

  1667.         return AVERROR_INVALIDDATA;

  1668.     }

  1669.     /* update codec info */

  1670.     avctx->channels       = s->nb_channels;

  1671.     avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;

  1672.     if (!avctx->bit_rate)

  1673.         avctx->bit_rate = s->bit_rate;

  1674. 

  1675.     if (s->frame_size <= 0 || s->frame_size > buf_size) {

  1676.         av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");

  1677.         return AVERROR_INVALIDDATA;

  1678.     } else if (s->frame_size < buf_size) {

  1679.         buf_size= s->frame_size;

  1680.     }

  1681. 

  1682.     s->frame = data;

  1683. 

  1684.     ret = mp_decode_frame(s, NULL, buf, buf_size);

  1685.     if (ret >= 0) {

  1686.         s->frame->nb_samples = avctx->frame_size;

  1687.         *got_frame_ptr       = 1;

  1688.         avctx->sample_rate   = s->sample_rate;

  1689.         //FIXME maybe move the other codec info stuff from above here too

  1690.     } else {

  1691.         av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");

  1692.         /* Only return an error if the bad frame makes up the whole packet or

  1693.          * the error is related to buffer management.

  1694.          * If there is more data in the packet, just consume the bad frame

  1695.          * instead of returning an error, which would discard the whole

  1696.          * packet. */

  1697.         *got_frame_ptr = 0;

  1698.         if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)

  1699.             return ret;

  1700.     }

  1701.     s->frame_size = 0;

  1702.     return buf_size;

  1703. }

  1704. 

  1705. static void mp_flush(MPADecodeContext *ctx)

  1706. {

  1707.     memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));

  1708.     ctx->last_buf_size = 0;

  1709. }

  1710. 

  1711. static void flush(AVCodecContext *avctx)

  1712. {

  1713.     mp_flush(avctx->priv_data);

  1714. }

  1715. 

  1716. #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER

  1717. static int decode_frame_adu(AVCodecContext *avctx, void *data,

  1718.                             int *got_frame_ptr, AVPacket *avpkt)

  1719. {

  1720.     const uint8_t *buf  = avpkt->data;

  1721.     int buf_size        = avpkt->size;

  1722.     MPADecodeContext *s = avctx->priv_data;

  1723.     uint32_t header;

  1724.     int len, ret;

  1725. 

  1726.     len = buf_size;

  1727. 

  1728.     // Discard too short frames

  1729.     if (buf_size < HEADER_SIZE) {

  1730.         av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");

  1731.         return AVERROR_INVALIDDATA;

  1732.     }

  1733. 

  1734. 

  1735.     if (len > MPA_MAX_CODED_FRAME_SIZE)

  1736.         len = MPA_MAX_CODED_FRAME_SIZE;

  1737. 

  1738.     // Get header and restore sync word

  1739.     header = AV_RB32(buf) | 0xffe00000;

  1740. 

  1741.     if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame

  1742.         av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");

  1743.         return AVERROR_INVALIDDATA;

  1744.     }

  1745. 

  1746.     avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);

  1747.     /* update codec info */

  1748.     avctx->sample_rate = s->sample_rate;

  1749.     avctx->channels    = s->nb_channels;

  1750.     if (!avctx->bit_rate)

  1751.         avctx->bit_rate = s->bit_rate;

  1752. 

  1753.     s->frame_size = len;

  1754. 

  1755.     s->frame = data;

  1756. 

  1757.     ret = mp_decode_frame(s, NULL, buf, buf_size);

  1758.     if (ret < 0) {

  1759.         av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");

  1760.         return ret;

  1761.     }

  1762. 

  1763.     *got_frame_ptr = 1;

  1764. 

  1765.     return buf_size;

  1766. }

  1767. #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */

  1768. 

  1769. #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER

  1770. 

  1771. /**

  1772.  * Context for MP3On4 decoder

  1773.  */

  1774. typedef struct MP3On4DecodeContext {

  1775.     int frames;                     ///< number of mp3 frames per block (number of mp3 decoder instances)

  1776.     int syncword;                   ///< syncword patch

  1777.     const uint8_t *coff;            ///< channel offsets in output buffer

  1778.     MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance

  1779. } MP3On4DecodeContext;

  1780. 

  1781. #include "mpeg4audio.h"

  1782. 

  1783. /* Next 3 arrays are indexed by channel config number (passed via codecdata) */

  1784. 

  1785. /* number of mp3 decoder instances */

  1786. static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };

  1787. 

  1788. /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */

  1789. static const uint8_t chan_offset[8][5] = {

  1790.     { 0             },

  1791.     { 0             },  // C

  1792.     { 0             },  // FLR

  1793.     { 2, 0          },  // C FLR

  1794.     { 2, 0, 3       },  // C FLR BS

  1795.     { 2, 0, 3       },  // C FLR BLRS

  1796.     { 2, 0, 4, 3    },  // C FLR BLRS LFE

  1797.     { 2, 0, 6, 4, 3 },  // C FLR BLRS BLR LFE

  1798. };

  1799. 

  1800. /* mp3on4 channel layouts */

  1801. static const int16_t chan_layout[8] = {

  1802.     0,

  1803.     AV_CH_LAYOUT_MONO,

  1804.     AV_CH_LAYOUT_STEREO,

  1805.     AV_CH_LAYOUT_SURROUND,

  1806.     AV_CH_LAYOUT_4POINT0,

  1807.     AV_CH_LAYOUT_5POINT0,

  1808.     AV_CH_LAYOUT_5POINT1,

  1809.     AV_CH_LAYOUT_7POINT1

  1810. };

  1811. 

  1812. static av_cold int decode_close_mp3on4(AVCodecContext * avctx)

  1813. {

  1814.     MP3On4DecodeContext *s = avctx->priv_data;

  1815.     int i;

  1816. 

  1817.     for (i = 0; i < s->frames; i++)

  1818.         av_free(s->mp3decctx[i]);

  1819. 

  1820.     return 0;

  1821. }

  1822. 

  1823. 

  1824. static av_cold int decode_init_mp3on4(AVCodecContext * avctx)

  1825. {

  1826.     MP3On4DecodeContext *s = avctx->priv_data;

  1827.     MPEG4AudioConfig cfg;

  1828.     int i;

  1829. 

  1830.     if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {

  1831.         av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");

  1832.         return AVERROR_INVALIDDATA;

  1833.     }

  1834. 

  1835.     avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,

  1836.                                  avctx->extradata_size * 8, 1);

  1837.     if (!cfg.chan_config || cfg.chan_config > 7) {

  1838.         av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");

  1839.         return AVERROR_INVALIDDATA;

  1840.     }

  1841.     s->frames             = mp3Frames[cfg.chan_config];

  1842.     s->coff               = chan_offset[cfg.chan_config];

  1843.     avctx->channels       = ff_mpeg4audio_channels[cfg.chan_config];

  1844.     avctx->channel_layout = chan_layout[cfg.chan_config];

  1845. 

  1846.     if (cfg.sample_rate < 16000)

  1847.         s->syncword = 0xffe00000;

  1848.     else

  1849.         s->syncword = 0xfff00000;

  1850. 

  1851.     /* Init the first mp3 decoder in standard way, so that all tables get builded

  1852.      * We replace avctx->priv_data with the context of the first decoder so that

  1853.      * decode_init() does not have to be changed.

  1854.      * Other decoders will be initialized here copying data from the first context

  1855.      */

  1856.     // Allocate zeroed memory for the first decoder context

  1857.     s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));

  1858.     if (!s->mp3decctx[0])

  1859.         goto alloc_fail;

  1860.     // Put decoder context in place to make init_decode() happy

  1861.     avctx->priv_data = s->mp3decctx[0];

  1862.     decode_init(avctx);

  1863.     // Restore mp3on4 context pointer

  1864.     avctx->priv_data = s;

  1865.     s->mp3decctx[0]->adu_mode = 1; // Set adu mode

  1866. 

  1867.     /* Create a separate codec/context for each frame (first is already ok).

  1868.      * Each frame is 1 or 2 channels - up to 5 frames allowed

  1869.      */

  1870.     for (i = 1; i < s->frames; i++) {

  1871.         s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));

  1872.         if (!s->mp3decctx[i])

  1873.             goto alloc_fail;

  1874.         s->mp3decctx[i]->adu_mode = 1;

  1875.         s->mp3decctx[i]->avctx = avctx;

  1876.         s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;

  1877.     }

  1878. 

  1879.     return 0;

  1880. alloc_fail:

  1881.     decode_close_mp3on4(avctx);

  1882.     return AVERROR(ENOMEM);

  1883. }

  1884. 

  1885. 

  1886. static void flush_mp3on4(AVCodecContext *avctx)

  1887. {

  1888.     int i;

  1889.     MP3On4DecodeContext *s = avctx->priv_data;

  1890. 

  1891.     for (i = 0; i < s->frames; i++)

  1892.         mp_flush(s->mp3decctx[i]);

  1893. }

  1894. 

  1895. 

  1896. static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,

  1897.                                int *got_frame_ptr, AVPacket *avpkt)

  1898. {

  1899.     AVFrame *frame         = data;

  1900.     const uint8_t *buf     = avpkt->data;

  1901.     int buf_size           = avpkt->size;

  1902.     MP3On4DecodeContext *s = avctx->priv_data;

  1903.     MPADecodeContext *m;

  1904.     int fsize, len = buf_size, out_size = 0;

  1905.     uint32_t header;

  1906.     OUT_INT **out_samples;

  1907.     OUT_INT *outptr[2];

  1908.     int fr, ch, ret;

  1909. 

  1910.     /* get output buffer */

  1911.     frame->nb_samples = MPA_FRAME_SIZE;

  1912.     if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {

  1913.         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");

  1914.         return ret;

  1915.     }

  1916.     out_samples = (OUT_INT **)frame->extended_data;

  1917. 

  1918.     // Discard too short frames

  1919.     if (buf_size < HEADER_SIZE)

  1920.         return AVERROR_INVALIDDATA;

  1921. 

  1922.     avctx->bit_rate = 0;

  1923. 

  1924.     ch = 0;

  1925.     for (fr = 0; fr < s->frames; fr++) {

  1926.         fsize = AV_RB16(buf) >> 4;

  1927.         fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);

  1928.         m     = s->mp3decctx[fr];

  1929.         assert(m != NULL);

  1930. 

  1931.         if (fsize < HEADER_SIZE) {

  1932.             av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");

  1933.             return AVERROR_INVALIDDATA;

  1934.         }

  1935.         header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header

  1936. 

  1937.         if (ff_mpa_check_header(header) < 0) // Bad header, discard block

  1938.             break;

  1939. 

  1940.         avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);

  1941. 

  1942.         if (ch + m->nb_channels > avctx->channels ||

  1943.             s->coff[fr] + m->nb_channels > avctx->channels) {

  1944.             av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "

  1945.                                         "channel count\n");

  1946.             return AVERROR_INVALIDDATA;

  1947.         }

  1948.         ch += m->nb_channels;

  1949. 

  1950.         outptr[0] = out_samples[s->coff[fr]];

  1951.         if (m->nb_channels > 1)

  1952.             outptr[1] = out_samples[s->coff[fr] + 1];

  1953. 

  1954.         if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0)

  1955.             return ret;

  1956. 

  1957.         out_size += ret;

  1958.         buf      += fsize;

  1959.         len      -= fsize;

  1960. 

  1961.         avctx->bit_rate += m->bit_rate;

  1962.     }

  1963. 

  1964.     /* update codec info */

  1965.     avctx->sample_rate = s->mp3decctx[0]->sample_rate;

  1966. 

  1967.     frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));

  1968.     *got_frame_ptr    = 1;

  1969. 

  1970.     return buf_size;

  1971. }

  1972. #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */

  1973. 

  1974. #if !CONFIG_FLOAT

  1975. #if CONFIG_MP1_DECODER

  1976. AVCodec ff_mp1_decoder = {

  1977.     .name           = "mp1",

  1978.     .long_name      = NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),

  1979.     .type           = AVMEDIA_TYPE_AUDIO,

  1980.     .id             = AV_CODEC_ID_MP1,

  1981.     .priv_data_size = sizeof(MPADecodeContext),

  1982.     .init           = decode_init,

  1983.     .decode         = decode_frame,

  1984.     .capabilities   = CODEC_CAP_DR1,

  1985.     .flush          = flush,

  1986.     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,

  1987.                                                       AV_SAMPLE_FMT_S16,

  1988.                                                       AV_SAMPLE_FMT_NONE },

  1989. };

  1990. #endif

  1991. #if CONFIG_MP2_DECODER

  1992. AVCodec ff_mp2_decoder = {

  1993.     .name           = "mp2",

  1994.     .long_name      = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),

  1995.     .type           = AVMEDIA_TYPE_AUDIO,

  1996.     .id             = AV_CODEC_ID_MP2,

  1997.     .priv_data_size = sizeof(MPADecodeContext),

  1998.     .init           = decode_init,

  1999.     .decode         = decode_frame,

  2000.     .capabilities   = CODEC_CAP_DR1,

  2001.     .flush          = flush,

  2002.     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,

  2003.                                                       AV_SAMPLE_FMT_S16,

  2004.                                                       AV_SAMPLE_FMT_NONE },

  2005. };

  2006. #endif

  2007. #if CONFIG_MP3_DECODER

  2008. AVCodec ff_mp3_decoder = {

  2009.     .name           = "mp3",

  2010.     .long_name      = NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),

  2011.     .type           = AVMEDIA_TYPE_AUDIO,

  2012.     .id             = AV_CODEC_ID_MP3,

  2013.     .priv_data_size = sizeof(MPADecodeContext),

  2014.     .init           = decode_init,

  2015.     .decode         = decode_frame,

  2016.     .capabilities   = CODEC_CAP_DR1,

  2017.     .flush          = flush,

  2018.     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,

  2019.                                                       AV_SAMPLE_FMT_S16,

  2020.                                                       AV_SAMPLE_FMT_NONE },

  2021. };

  2022. #endif

  2023. #if CONFIG_MP3ADU_DECODER

  2024. AVCodec ff_mp3adu_decoder = {

  2025.     .name           = "mp3adu",

  2026.     .long_name      = NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),

  2027.     .type           = AVMEDIA_TYPE_AUDIO,

  2028.     .id             = AV_CODEC_ID_MP3ADU,

  2029.     .priv_data_size = sizeof(MPADecodeContext),

  2030.     .init           = decode_init,

  2031.     .decode         = decode_frame_adu,

  2032.     .capabilities   = CODEC_CAP_DR1,

  2033.     .flush          = flush,

  2034.     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,

  2035.                                                       AV_SAMPLE_FMT_S16,

  2036.                                                       AV_SAMPLE_FMT_NONE },

  2037. };

  2038. #endif

  2039. #if CONFIG_MP3ON4_DECODER

  2040. AVCodec ff_mp3on4_decoder = {

  2041.     .name           = "mp3on4",

  2042.     .long_name      = NULL_IF_CONFIG_SMALL("MP3onMP4"),

  2043.     .type           = AVMEDIA_TYPE_AUDIO,

  2044.     .id             = AV_CODEC_ID_MP3ON4,

  2045.     .priv_data_size = sizeof(MP3On4DecodeContext),

  2046.     .init           = decode_init_mp3on4,

  2047.     .close          = decode_close_mp3on4,

  2048.     .decode         = decode_frame_mp3on4,

  2049.     .capabilities   = CODEC_CAP_DR1,

  2050.     .flush          = flush_mp3on4,

  2051.     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,

  2052.                                                       AV_SAMPLE_FMT_NONE },

  2053. };

  2054. #endif

  2055. #endif