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.
49698 Commits
32 Branches
813MB
Tree: a00c4b4db0
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a00c4b4db0'
${ noResults }
FFmpeg/libavcodec/mpegaudiodec.c

2083 lines
68KB
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 "libavutil/avassert.h"

  28. #include "libavutil/channel_layout.h"

  29. #include "libavutil/float_dsp.h"

  30. #include "libavutil/libm.h"

  31. #include "avcodec.h"

  32. #include "get_bits.h"

  33. #include "internal.h"

  34. #include "mathops.h"

  35. #include "mpegaudiodsp.h"

  36. #include "dsputil.h"

  37. 

  38. /*

  39.  * TODO:

  40.  *  - test lsf / mpeg25 extensively.

  41.  */

  42. 

  43. #include "mpegaudio.h"

  44. #include "mpegaudiodecheader.h"

  45. 

  46. #define BACKSTEP_SIZE 512

  47. #define EXTRABYTES 24

  48. #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES

  49. 

  50. /* layer 3 "granule" */

  51. typedef struct GranuleDef {

  52.     uint8_t scfsi;

  53.     int part2_3_length;

  54.     int big_values;

  55.     int global_gain;

  56.     int scalefac_compress;

  57.     uint8_t block_type;

  58.     uint8_t switch_point;

  59.     int table_select[3];

  60.     int subblock_gain[3];

  61.     uint8_t scalefac_scale;

  62.     uint8_t count1table_select;

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

  64.     int preflag;

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

  66.     uint8_t scale_factors[40];

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

  68. } GranuleDef;

  69. 

  70. typedef struct MPADecodeContext {

  71.     MPA_DECODE_HEADER

  72.     uint8_t last_buf[LAST_BUF_SIZE];

  73.     int last_buf_size;

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

  75.     uint32_t free_format_next_header;

  76.     GetBitContext gb;

  77.     GetBitContext in_gb;

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

  79.     int synth_buf_offset[MPA_MAX_CHANNELS];

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

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

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

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

  84.     int dither_state;

  85.     int err_recognition;

  86.     AVCodecContext* avctx;

  87.     MPADSPContext mpadsp;

  88.     AVFloatDSPContext fdsp;

  89.     AVFrame frame;

  90. } MPADecodeContext;

  91. 

  92. #if CONFIG_FLOAT

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

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

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

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

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

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

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

  100. #   define OUT_FMT   AV_SAMPLE_FMT_FLT

  101. #   define OUT_FMT_P AV_SAMPLE_FMT_FLTP

  102. #else

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

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

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

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

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

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

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

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

  111. #   define OUT_FMT   AV_SAMPLE_FMT_S16

  112. #   define OUT_FMT_P AV_SAMPLE_FMT_S16P

  113. #endif

  114. 

  115. /****************/

  116. 

  117. #define HEADER_SIZE 4

  118. 

  119. #include "mpegaudiodata.h"

  120. #include "mpegaudiodectab.h"

  121. 

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

  123. static VLC huff_vlc[16];

  124. static VLC_TYPE huff_vlc_tables[

  125.     0 + 128 + 128 + 128 + 130 + 128 + 154 + 166 +

  126.   142 + 204 + 190 + 170 + 542 + 460 + 662 + 414

  127.   ][2];

  128. static const int huff_vlc_tables_sizes[16] = {

  129.     0,  128,  128,  128,  130,  128,  154,  166,

  130.   142,  204,  190,  170,  542,  460,  662,  414

  131. };

  132. static VLC huff_quad_vlc[2];

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

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

  135. /* computed from band_size_long */

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

  137. #include "mpegaudio_tablegen.h"

  138. /* intensity stereo coef table */

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

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

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

  142. 

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

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

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

  146. 

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

  148.     division_tab3, division_tab5, NULL, division_tab9

  149. };

  150. 

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

  152. static uint16_t scale_factor_modshift[64];

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

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

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

  156. 

  157. #define SCALE_GEN(v) \

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

  159. 

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

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

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

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

  164. };

  165. 

  166. /**

  167.  * Convert region offsets to region sizes and truncate

  168.  * size to big_values.

  169.  */

  170. static void ff_region_offset2size(GranuleDef *g)

  171. {

  172.     int i, k, j = 0;

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

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

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

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

  177.         j = k;

  178.     }

  179. }

  180. 

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

  182. {

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

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

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

  186.         else

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

  188.     } else {

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

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

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

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

  193.         else

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

  195.     }

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

  197. }

  198. 

  199. static void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, 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 ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g)

  209. {

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

  211.         if (g->switch_point) {

  212.             if(s->sample_rate_index == 8)

  213.                 av_log_ask_for_sample(s->avctx, "switch point in 8khz\n");

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

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

  216.                 exponents as long blocks */

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

  218.                 g->long_end = 8;

  219.             else

  220.                 g->long_end = 6;

  221. 

  222.             g->short_start = 3;

  223.         } else {

  224.             g->long_end    = 0;

  225.             g->short_start = 0;

  226.         }

  227.     } else {

  228.         g->short_start = 13;

  229.         g->long_end    = 22;

  230.     }

  231. }

  232. 

  233. /* layer 1 unscaling */

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

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

  236. {

  237.     int shift, mod;

  238.     int64_t val;

  239. 

  240.     shift   = scale_factor_modshift[scale_factor];

  241.     mod     = shift & 3;

  242.     shift >>= 2;

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

  244.     shift  += n;

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

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

  247. }

  248. 

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

  250. {

  251.     int shift, mod, val;

  252. 

  253.     shift   = scale_factor_modshift[scale_factor];

  254.     mod     = shift & 3;

  255.     shift >>= 2;

  256. 

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

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

  259.     if (shift > 0)

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

  261.     return val;

  262. }

  263. 

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

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

  266. {

  267.     unsigned int m;

  268.     int e;

  269. 

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

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

  272.     e -= exponent >> 2;

  273. #ifdef DEBUG

  274.     if(e < 1)

  275.         av_log(NULL, AV_LOG_WARNING, "l3_unscale: e is %d\n", e);

  276. #endif

  277.     if (e > 31)

  278.         return 0;

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

  280. 

  281.     return m;

  282. }

  283. 

  284. static av_cold void decode_init_static(void)

  285. {

  286.     int i, j, k;

  287.     int offset;

  288. 

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

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

  291.         int shift, mod;

  292.         /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

  293.         shift = i / 3;

  294.         mod   = i % 3;

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

  296.     }

  297. 

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

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

  300.         int n, norm;

  301.         n = i + 2;

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

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

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

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

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

  307.                 scale_factor_mult[i][0],

  308.                 scale_factor_mult[i][1],

  309.                 scale_factor_mult[i][2]);

  310.     }

  311. 

  312.     RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));

  313. 

  314.     /* huffman decode tables */

  315.     offset = 0;

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

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

  318.         int xsize, x, y;

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

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

  321. 

  322.         xsize = h->xsize;

  323. 

  324.         j = 0;

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

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

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

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

  329.             }

  330.         }

  331. 

  332.         /* XXX: fail test */

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

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

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

  336.                  tmp_bits, 1, 1, tmp_codes, 2, 2,

  337.                  INIT_VLC_USE_NEW_STATIC);

  338.         offset += huff_vlc_tables_sizes[i];

  339.     }

  340.     av_assert0(offset == FF_ARRAY_ELEMS(huff_vlc_tables));

  341. 

  342.     offset = 0;

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

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

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

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

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

  348.                  INIT_VLC_USE_NEW_STATIC);

  349.         offset += huff_quad_vlc_tables_sizes[i];

  350.     }

  351.     av_assert0(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));

  352. 

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

  354.         k = 0;

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

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

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

  358.         }

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

  360.     }

  361. 

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

  363. 

  364.     mpegaudio_tableinit();

  365. 

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

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

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

  369.                 int val1, val2, val3, steps;

  370.                 int val = j;

  371.                 steps   = ff_mpa_quant_steps[i];

  372.                 val1    = val % steps;

  373.                 val    /= steps;

  374.                 val2    = val % steps;

  375.                 val3    = val / steps;

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

  377.             }

  378.         }

  379.     }

  380. 

  381. 

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

  383.         float f;

  384.         INTFLOAT v;

  385.         if (i != 6) {

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

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

  388.         } else {

  389.             v = FIXR(1.0);

  390.         }

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

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

  393.     }

  394.     /* invalid values */

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

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

  397. 

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

  399.         double f;

  400.         int e, k;

  401. 

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

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

  404.             f = exp2(e / 4.0);

  405.             k = i & 1;

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

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

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

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

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

  411.         }

  412.     }

  413. 

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

  415.         float ci, cs, ca;

  416.         ci = ci_table[i];

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

  418.         ca = cs * ci;

  419. #if !CONFIG_FLOAT

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

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

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

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

  424. #else

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

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

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

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

  429. #endif

  430.     }

  431. }

  432. 

  433. static av_cold int decode_init(AVCodecContext * avctx)

  434. {

  435.     static int initialized_tables = 0;

  436.     MPADecodeContext *s = avctx->priv_data;

  437. 

  438.     if (!initialized_tables) {

  439.         decode_init_static();

  440.         initialized_tables = 1;

  441.     }

  442. 

  443.     s->avctx = avctx;

  444. 

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

  446.     ff_mpadsp_init(&s->mpadsp);

  447. 

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

  449.         avctx->codec_id != AV_CODEC_ID_MP3ON4)

  450.         avctx->sample_fmt = OUT_FMT;

  451.     else

  452.         avctx->sample_fmt = OUT_FMT_P;

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

  454. 

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

  456.         s->adu_mode = 1;

  457. 

  458.     avcodec_get_frame_defaults(&s->frame);

  459.     avctx->coded_frame = &s->frame;

  460. 

  461.     return 0;

  462. }

  463. 

  464. #define C3 FIXHR(0.86602540378443864676/2)

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

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

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

  468. 

  469. /* 12 points IMDCT. We compute it "by hand" by factorizing obvious

  470.    cases. */

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

  472. {

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

  474. 

  475.     in0  = in[0*3];

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

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

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

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

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

  481.     in5 += in3;

  482.     in3 += in1;

  483. 

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

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

  486. 

  487.     t1   = in0 - in4;

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

  489. 

  490.     out[ 7] =

  491.     out[10] = t1 + t2;

  492.     out[ 1] =

  493.     out[ 4] = t1 - t2;

  494. 

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

  496.     in4     = in0 + in2;

  497.     in5    += 2*in1;

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

  499.     out[ 8] =

  500.     out[ 9] = in4 + in1;

  501.     out[ 2] =

  502.     out[ 3] = in4 - in1;

  503. 

  504.     in0    -= in2;

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

  506.     out[ 0] =

  507.     out[ 5] = in0 - in5;

  508.     out[ 6] =

  509.     out[11] = in0 + in5;

  510. }

  511. 

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

  513. static int mp_decode_layer1(MPADecodeContext *s)

  514. {

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

  516.     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];

  517.     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

  518. 

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

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

  521.     else

  522.         bound = SBLIMIT;

  523. 

  524.     /* allocation bits */

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

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

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

  528.         }

  529.     }

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

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

  532. 

  533.     /* scale factors */

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

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

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

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

  538.         }

  539.     }

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

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

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

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

  544.         }

  545.     }

  546. 

  547.     /* compute samples */

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

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

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

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

  552.                 if (n) {

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

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

  555.                 } else {

  556.                     v = 0;

  557.                 }

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

  559.             }

  560.         }

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

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

  563.             if (n) {

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

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

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

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

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

  569.             } else {

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

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

  572.             }

  573.         }

  574.     }

  575.     return 12;

  576. }

  577. 

  578. static int mp_decode_layer2(MPADecodeContext *s)

  579. {

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

  581.     const unsigned char *alloc_table;

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

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

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

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

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

  587. 

  588.     /* select decoding table */

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

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

  591.     sblimit     = ff_mpa_sblimit_table[table];

  592.     alloc_table = ff_mpa_alloc_tables[table];

  593. 

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

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

  596.     else

  597.         bound = sblimit;

  598. 

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

  600. 

  601.     /* sanity check */

  602.     if (bound > sblimit)

  603.         bound = sblimit;

  604. 

  605.     /* parse bit allocation */

  606.     j = 0;

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

  608.         bit_alloc_bits = alloc_table[j];

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

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

  611.         j += 1 << bit_alloc_bits;

  612.     }

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

  614.         bit_alloc_bits = alloc_table[j];

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

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

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

  618.         j += 1 << bit_alloc_bits;

  619.     }

  620. 

  621.     /* scale codes */

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

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

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

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

  626.         }

  627.     }

  628. 

  629.     /* scale factors */

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

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

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

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

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

  635.                 default:

  636.                 case 0:

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

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

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

  640.                     break;

  641.                 case 2:

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

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

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

  645.                     break;

  646.                 case 1:

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

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

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

  650.                     break;

  651.                 case 3:

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

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

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

  655.                     break;

  656.                 }

  657.             }

  658.         }

  659.     }

  660. 

  661.     /* samples */

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

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

  664.             j = 0;

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

  666.                 bit_alloc_bits = alloc_table[j];

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

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

  669.                     if (b) {

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

  671.                         qindex = alloc_table[j+b];

  672.                         bits = ff_mpa_quant_bits[qindex];

  673.                         if (bits < 0) {

  674.                             int v2;

  675.                             /* 3 values at the same time */

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

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

  678.                             steps  = ff_mpa_quant_steps[qindex];

  679. 

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

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

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

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

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

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

  686.                         } else {

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

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

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

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

  691.                             }

  692.                         }

  693.                     } else {

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

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

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

  697.                     }

  698.                 }

  699.                 /* next subband in alloc table */

  700.                 j += 1 << bit_alloc_bits;

  701.             }

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

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

  704.                 bit_alloc_bits = alloc_table[j];

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

  706.                 if (b) {

  707.                     int mant, scale0, scale1;

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

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

  710.                     qindex = alloc_table[j+b];

  711.                     bits = ff_mpa_quant_bits[qindex];

  712.                     if (bits < 0) {

  713.                         /* 3 values at the same time */

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

  715.                         steps = ff_mpa_quant_steps[qindex];

  716.                         mant = v % steps;

  717.                         v = v / steps;

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

  719.                             l2_unscale_group(steps, mant, scale0);

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

  721.                             l2_unscale_group(steps, mant, scale1);

  722.                         mant = v % steps;

  723.                         v = v / steps;

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

  725.                             l2_unscale_group(steps, mant, scale0);

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

  727.                             l2_unscale_group(steps, mant, scale1);

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

  729.                             l2_unscale_group(steps, v, scale0);

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

  731.                             l2_unscale_group(steps, v, scale1);

  732.                     } else {

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

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

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

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

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

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

  739.                         }

  740.                     }

  741.                 } else {

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

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

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

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

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

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

  748.                 }

  749.                 /* next subband in alloc table */

  750.                 j += 1 << bit_alloc_bits;

  751.             }

  752.             /* fill remaining samples to zero */

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

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

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

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

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

  758.                 }

  759.             }

  760.         }

  761.     }

  762.     return 3 * 12;

  763. }

  764. 

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

  766.     if (n == 3) {                   \

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

  768.         dst   = sf - 3 * m;         \

  769.         sf    = m;                  \

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

  771.         dst  = sf & 3;              \

  772.         sf >>= 2;                   \

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

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

  775.         dst   = sf - 5 * m;         \

  776.         sf    = m;                  \

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

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

  779.         dst   = sf - 6 * m;         \

  780.         sf    = m;                  \

  781.     } else {                        \

  782.         dst = 0;                    \

  783.     }

  784. 

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

  786.                                            int n3)

  787. {

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

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

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

  791.     slen[0] = sf;

  792. }

  793. 

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

  795.                                          int16_t *exponents)

  796. {

  797.     const uint8_t *bstab, *pretab;

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

  799.     int16_t *exp_ptr;

  800. 

  801.     exp_ptr = exponents;

  802.     gain    = g->global_gain - 210;

  803.     shift   = g->scalefac_scale + 1;

  804. 

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

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

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

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

  809.         len = bstab[i];

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

  811.             *exp_ptr++ = v0;

  812.     }

  813. 

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

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

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

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

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

  819.         k        = g->long_end;

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

  821.             len = bstab[i];

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

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

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

  825.                     *exp_ptr++ = v0;

  826.             }

  827.         }

  828.     }

  829. }

  830. 

  831. /* handle n = 0 too */

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

  833. {

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

  835. }

  836. 

  837. 

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

  839.                           int *end_pos2)

  840. {

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

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

  843.         s->in_gb.buffer = NULL;

  844.         av_assert2((get_bits_count(&s->gb) & 7) == 0);

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

  846.         *end_pos2 =

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

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

  849.     }

  850. }

  851. 

  852. /* Following is a optimized code for

  853.             INTFLOAT v = *src

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

  855.                 v = -v;

  856.             *dst = v;

  857. */

  858. #if CONFIG_FLOAT

  859. #define READ_FLIP_SIGN(dst,src)                     \

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

  861.     AV_WN32A(dst, v);

  862. #else

  863. #define READ_FLIP_SIGN(dst,src)     \

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

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

  866. #endif

  867. 

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

  869.                           int16_t *exponents, int end_pos2)

  870. {

  871.     int s_index;

  872.     int i;

  873.     int last_pos, bits_left;

  874.     VLC *vlc;

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

  876. 

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

  878.     s_index = 0;

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

  880.         int j, k, l, linbits;

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

  882.         if (j == 0)

  883.             continue;

  884.         /* select vlc table */

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

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

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

  888.         vlc     = &huff_vlc[l];

  889. 

  890.         if (!l) {

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

  892.             s_index += 2 * j;

  893.             continue;

  894.         }

  895. 

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

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

  898.             int exponent, x, y;

  899.             int v;

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

  901. 

  902.             if (pos >= end_pos){

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

  904.                 if (pos >= end_pos)

  905.                     break;

  906.             }

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

  908. 

  909.             if (!y) {

  910.                 g->sb_hybrid[s_index  ] =

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

  912.                 s_index += 2;

  913.                 continue;

  914.             }

  915. 

  916.             exponent= exponents[s_index];

  917. 

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

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

  920.             if (y & 16) {

  921.                 x = y >> 5;

  922.                 y = y & 0x0f;

  923.                 if (x < 15) {

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

  925.                 } else {

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

  927.                     v  = l3_unscale(x, exponent);

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

  929.                         v = -v;

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

  931.                 }

  932.                 if (y < 15) {

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

  934.                 } else {

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

  936.                     v  = l3_unscale(y, exponent);

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

  938.                         v = -v;

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

  940.                 }

  941.             } else {

  942.                 x = y >> 5;

  943.                 y = y & 0x0f;

  944.                 x += y;

  945.                 if (x < 15) {

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

  947.                 } else {

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

  949.                     v  = l3_unscale(x, exponent);

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

  951.                         v = -v;

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

  953.                 }

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

  955.             }

  956.             s_index += 2;

  957.         }

  958.     }

  959. 

  960.     /* high frequencies */

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

  962.     last_pos = 0;

  963.     while (s_index <= 572) {

  964.         int pos, code;

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

  966.         if (pos >= end_pos) {

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

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

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

  970.                 s_index -= 4;

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

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

  973.                 if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT))

  974.                     s_index=0;

  975.                 break;

  976.             }

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

  978.             if (pos >= end_pos)

  979.                 break;

  980.         }

  981.         last_pos = pos;

  982. 

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

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

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

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

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

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

  989.         while (code) {

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

  991.             int v;

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

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

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

  995.         }

  996.         s_index += 4;

  997.     }

  998.     /* skip extension bits */

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

  1000.     if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) {

  1001.         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1002.         s_index=0;

  1003.     } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) {

  1004.         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1005.         s_index = 0;

  1006.     }

  1007.     memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));

  1008.     skip_bits_long(&s->gb, bits_left);

  1009. 

  1010.     i = get_bits_count(&s->gb);

  1011.     switch_buffer(s, &i, &end_pos, &end_pos2);

  1012. 

  1013.     return 0;

  1014. }

  1015. 

  1016. /* Reorder short blocks from bitstream order to interleaved order. It

  1017.    would be faster to do it in parsing, but the code would be far more

  1018.    complicated */

  1019. static void reorder_block(MPADecodeContext *s, GranuleDef *g)

  1020. {

  1021.     int i, j, len;

  1022.     INTFLOAT *ptr, *dst, *ptr1;

  1023.     INTFLOAT tmp[576];

  1024. 

  1025.     if (g->block_type != 2)

  1026.         return;

  1027. 

  1028.     if (g->switch_point) {

  1029.         if (s->sample_rate_index != 8)

  1030.             ptr = g->sb_hybrid + 36;

  1031.         else

  1032.             ptr = g->sb_hybrid + 72;

  1033.     } else {

  1034.         ptr = g->sb_hybrid;

  1035.     }

  1036. 

  1037.     for (i = g->short_start; i < 13; i++) {

  1038.         len  = band_size_short[s->sample_rate_index][i];

  1039.         ptr1 = ptr;

  1040.         dst  = tmp;

  1041.         for (j = len; j > 0; j--) {

  1042.             *dst++ = ptr[0*len];

  1043.             *dst++ = ptr[1*len];

  1044.             *dst++ = ptr[2*len];

  1045.             ptr++;

  1046.         }

  1047.         ptr += 2 * len;

  1048.         memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));

  1049.     }

  1050. }

  1051. 

  1052. #define ISQRT2 FIXR(0.70710678118654752440)

  1053. 

  1054. static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)

  1055. {

  1056.     int i, j, k, l;

  1057.     int sf_max, sf, len, non_zero_found;

  1058.     INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;

  1059.     int non_zero_found_short[3];

  1060. 

  1061.     /* intensity stereo */

  1062.     if (s->mode_ext & MODE_EXT_I_STEREO) {

  1063.         if (!s->lsf) {

  1064.             is_tab = is_table;

  1065.             sf_max = 7;

  1066.         } else {

  1067.             is_tab = is_table_lsf[g1->scalefac_compress & 1];

  1068.             sf_max = 16;

  1069.         }

  1070. 

  1071.         tab0 = g0->sb_hybrid + 576;

  1072.         tab1 = g1->sb_hybrid + 576;

  1073. 

  1074.         non_zero_found_short[0] = 0;

  1075.         non_zero_found_short[1] = 0;

  1076.         non_zero_found_short[2] = 0;

  1077.         k = (13 - g1->short_start) * 3 + g1->long_end - 3;

  1078.         for (i = 12; i >= g1->short_start; i--) {

  1079.             /* for last band, use previous scale factor */

  1080.             if (i != 11)

  1081.                 k -= 3;

  1082.             len = band_size_short[s->sample_rate_index][i];

  1083.             for (l = 2; l >= 0; l--) {

  1084.                 tab0 -= len;

  1085.                 tab1 -= len;

  1086.                 if (!non_zero_found_short[l]) {

  1087.                     /* test if non zero band. if so, stop doing i-stereo */

  1088.                     for (j = 0; j < len; j++) {

  1089.                         if (tab1[j] != 0) {

  1090.                             non_zero_found_short[l] = 1;

  1091.                             goto found1;

  1092.                         }

  1093.                     }

  1094.                     sf = g1->scale_factors[k + l];

  1095.                     if (sf >= sf_max)

  1096.                         goto found1;

  1097. 

  1098.                     v1 = is_tab[0][sf];

  1099.                     v2 = is_tab[1][sf];

  1100.                     for (j = 0; j < len; j++) {

  1101.                         tmp0    = tab0[j];

  1102.                         tab0[j] = MULLx(tmp0, v1, FRAC_BITS);

  1103.                         tab1[j] = MULLx(tmp0, v2, FRAC_BITS);

  1104.                     }

  1105.                 } else {

  1106. found1:

  1107.                     if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1108.                         /* lower part of the spectrum : do ms stereo

  1109.                            if enabled */

  1110.                         for (j = 0; j < len; j++) {

  1111.                             tmp0    = tab0[j];

  1112.                             tmp1    = tab1[j];

  1113.                             tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1114.                             tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1115.                         }

  1116.                     }

  1117.                 }

  1118.             }

  1119.         }

  1120. 

  1121.         non_zero_found = non_zero_found_short[0] |

  1122.                          non_zero_found_short[1] |

  1123.                          non_zero_found_short[2];

  1124. 

  1125.         for (i = g1->long_end - 1;i >= 0;i--) {

  1126.             len   = band_size_long[s->sample_rate_index][i];

  1127.             tab0 -= len;

  1128.             tab1 -= len;

  1129.             /* test if non zero band. if so, stop doing i-stereo */

  1130.             if (!non_zero_found) {

  1131.                 for (j = 0; j < len; j++) {

  1132.                     if (tab1[j] != 0) {

  1133.                         non_zero_found = 1;

  1134.                         goto found2;

  1135.                     }

  1136.                 }

  1137.                 /* for last band, use previous scale factor */

  1138.                 k  = (i == 21) ? 20 : i;

  1139.                 sf = g1->scale_factors[k];

  1140.                 if (sf >= sf_max)

  1141.                     goto found2;

  1142.                 v1 = is_tab[0][sf];

  1143.                 v2 = is_tab[1][sf];

  1144.                 for (j = 0; j < len; j++) {

  1145.                     tmp0    = tab0[j];

  1146.                     tab0[j] = MULLx(tmp0, v1, FRAC_BITS);

  1147.                     tab1[j] = MULLx(tmp0, v2, FRAC_BITS);

  1148.                 }

  1149.             } else {

  1150. found2:

  1151.                 if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1152.                     /* lower part of the spectrum : do ms stereo

  1153.                        if enabled */

  1154.                     for (j = 0; j < len; j++) {

  1155.                         tmp0    = tab0[j];

  1156.                         tmp1    = tab1[j];

  1157.                         tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1158.                         tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1159.                     }

  1160.                 }

  1161.             }

  1162.         }

  1163.     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1164.         /* ms stereo ONLY */

  1165.         /* NOTE: the 1/sqrt(2) normalization factor is included in the

  1166.            global gain */

  1167. #if CONFIG_FLOAT

  1168.        s->fdsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);

  1169. #else

  1170.         tab0 = g0->sb_hybrid;

  1171.         tab1 = g1->sb_hybrid;

  1172.         for (i = 0; i < 576; i++) {

  1173.             tmp0    = tab0[i];

  1174.             tmp1    = tab1[i];

  1175.             tab0[i] = tmp0 + tmp1;

  1176.             tab1[i] = tmp0 - tmp1;

  1177.         }

  1178. #endif

  1179.     }

  1180. }

  1181. 

  1182. #if CONFIG_FLOAT

  1183. #if HAVE_MIPSFPU

  1184. #   include "mips/compute_antialias_float.h"

  1185. #endif /* HAVE_MIPSFPU */

  1186. #else

  1187. #if HAVE_MIPSDSPR1

  1188. #   include "mips/compute_antialias_fixed.h"

  1189. #endif /* HAVE_MIPSDSPR1 */

  1190. #endif /* CONFIG_FLOAT */

  1191. 

  1192. #ifndef compute_antialias

  1193. #if CONFIG_FLOAT

  1194. #define AA(j) do {                                                      \

  1195.         float tmp0 = ptr[-1-j];                                         \

  1196.         float tmp1 = ptr[   j];                                         \

  1197.         ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1];    \

  1198.         ptr[   j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0];    \

  1199.     } while (0)

  1200. #else

  1201. #define AA(j) do {                                              \

  1202.         int tmp0 = ptr[-1-j];                                   \

  1203.         int tmp1 = ptr[   j];                                   \

  1204.         int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]);          \

  1205.         ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2]));   \

  1206.         ptr[   j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3]));   \

  1207.     } while (0)

  1208. #endif

  1209. 

  1210. static void compute_antialias(MPADecodeContext *s, GranuleDef *g)

  1211. {

  1212.     INTFLOAT *ptr;

  1213.     int n, i;

  1214. 

  1215.     /* we antialias only "long" bands */

  1216.     if (g->block_type == 2) {

  1217.         if (!g->switch_point)

  1218.             return;

  1219.         /* XXX: check this for 8000Hz case */

  1220.         n = 1;

  1221.     } else {

  1222.         n = SBLIMIT - 1;

  1223.     }

  1224. 

  1225.     ptr = g->sb_hybrid + 18;

  1226.     for (i = n; i > 0; i--) {

  1227.         AA(0);

  1228.         AA(1);

  1229.         AA(2);

  1230.         AA(3);

  1231.         AA(4);

  1232.         AA(5);

  1233.         AA(6);

  1234.         AA(7);

  1235. 

  1236.         ptr += 18;

  1237.     }

  1238. }

  1239. #endif /* compute_antialias */

  1240. 

  1241. static void compute_imdct(MPADecodeContext *s, GranuleDef *g,

  1242.                           INTFLOAT *sb_samples, INTFLOAT *mdct_buf)

  1243. {

  1244.     INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;

  1245.     INTFLOAT out2[12];

  1246.     int i, j, mdct_long_end, sblimit;

  1247. 

  1248.     /* find last non zero block */

  1249.     ptr  = g->sb_hybrid + 576;

  1250.     ptr1 = g->sb_hybrid + 2 * 18;

  1251.     while (ptr >= ptr1) {

  1252.         int32_t *p;

  1253.         ptr -= 6;

  1254.         p    = (int32_t*)ptr;

  1255.         if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])

  1256.             break;

  1257.     }

  1258.     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

  1259. 

  1260.     if (g->block_type == 2) {

  1261.         /* XXX: check for 8000 Hz */

  1262.         if (g->switch_point)

  1263.             mdct_long_end = 2;

  1264.         else

  1265.             mdct_long_end = 0;

  1266.     } else {

  1267.         mdct_long_end = sblimit;

  1268.     }

  1269. 

  1270.     s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,

  1271.                                      mdct_long_end, g->switch_point,

  1272.                                      g->block_type);

  1273. 

  1274.     buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);

  1275.     ptr = g->sb_hybrid + 18 * mdct_long_end;

  1276. 

  1277.     for (j = mdct_long_end; j < sblimit; j++) {

  1278.         /* select frequency inversion */

  1279.         win     = RENAME(ff_mdct_win)[2 + (4  & -(j & 1))];

  1280.         out_ptr = sb_samples + j;

  1281. 

  1282.         for (i = 0; i < 6; i++) {

  1283.             *out_ptr = buf[4*i];

  1284.             out_ptr += SBLIMIT;

  1285.         }

  1286.         imdct12(out2, ptr + 0);

  1287.         for (i = 0; i < 6; i++) {

  1288.             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*1)];

  1289.             buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);

  1290.             out_ptr += SBLIMIT;

  1291.         }

  1292.         imdct12(out2, ptr + 1);

  1293.         for (i = 0; i < 6; i++) {

  1294.             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*2)];

  1295.             buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);

  1296.             out_ptr += SBLIMIT;

  1297.         }

  1298.         imdct12(out2, ptr + 2);

  1299.         for (i = 0; i < 6; i++) {

  1300.             buf[4*(i + 6*0)] = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*0)];

  1301.             buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);

  1302.             buf[4*(i + 6*2)] = 0;

  1303.         }

  1304.         ptr += 18;

  1305.         buf += (j&3) != 3 ? 1 : (4*18-3);

  1306.     }

  1307.     /* zero bands */

  1308.     for (j = sblimit; j < SBLIMIT; j++) {

  1309.         /* overlap */

  1310.         out_ptr = sb_samples + j;

  1311.         for (i = 0; i < 18; i++) {

  1312.             *out_ptr = buf[4*i];

  1313.             buf[4*i]   = 0;

  1314.             out_ptr += SBLIMIT;

  1315.         }

  1316.         buf += (j&3) != 3 ? 1 : (4*18-3);

  1317.     }

  1318. }

  1319. 

  1320. /* main layer3 decoding function */

  1321. static int mp_decode_layer3(MPADecodeContext *s)

  1322. {

  1323.     int nb_granules, main_data_begin;

  1324.     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;

  1325.     GranuleDef *g;

  1326.     int16_t exponents[576]; //FIXME try INTFLOAT

  1327. 

  1328.     /* read side info */

  1329.     if (s->lsf) {

  1330.         main_data_begin = get_bits(&s->gb, 8);

  1331.         skip_bits(&s->gb, s->nb_channels);

  1332.         nb_granules = 1;

  1333.     } else {

  1334.         main_data_begin = get_bits(&s->gb, 9);

  1335.         if (s->nb_channels == 2)

  1336.             skip_bits(&s->gb, 3);

  1337.         else

  1338.             skip_bits(&s->gb, 5);

  1339.         nb_granules = 2;

  1340.         for (ch = 0; ch < s->nb_channels; ch++) {

  1341.             s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */

  1342.             s->granules[ch][1].scfsi = get_bits(&s->gb, 4);

  1343.         }

  1344.     }

  1345. 

  1346.     for (gr = 0; gr < nb_granules; gr++) {

  1347.         for (ch = 0; ch < s->nb_channels; ch++) {

  1348.             av_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);

  1349.             g = &s->granules[ch][gr];

  1350.             g->part2_3_length = get_bits(&s->gb, 12);

  1351.             g->big_values     = get_bits(&s->gb,  9);

  1352.             if (g->big_values > 288) {

  1353.                 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");

  1354.                 return AVERROR_INVALIDDATA;

  1355.             }

  1356. 

  1357.             g->global_gain = get_bits(&s->gb, 8);

  1358.             /* if MS stereo only is selected, we precompute the

  1359.                1/sqrt(2) renormalization factor */

  1360.             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==

  1361.                 MODE_EXT_MS_STEREO)

  1362.                 g->global_gain -= 2;

  1363.             if (s->lsf)

  1364.                 g->scalefac_compress = get_bits(&s->gb, 9);

  1365.             else

  1366.                 g->scalefac_compress = get_bits(&s->gb, 4);

  1367.             blocksplit_flag = get_bits1(&s->gb);

  1368.             if (blocksplit_flag) {

  1369.                 g->block_type = get_bits(&s->gb, 2);

  1370.                 if (g->block_type == 0) {

  1371.                     av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");

  1372.                     return AVERROR_INVALIDDATA;

  1373.                 }

  1374.                 g->switch_point = get_bits1(&s->gb);

  1375.                 for (i = 0; i < 2; i++)

  1376.                     g->table_select[i] = get_bits(&s->gb, 5);

  1377.                 for (i = 0; i < 3; i++)

  1378.                     g->subblock_gain[i] = get_bits(&s->gb, 3);

  1379.                 ff_init_short_region(s, g);

  1380.             } else {

  1381.                 int region_address1, region_address2;

  1382.                 g->block_type = 0;

  1383.                 g->switch_point = 0;

  1384.                 for (i = 0; i < 3; i++)

  1385.                     g->table_select[i] = get_bits(&s->gb, 5);

  1386.                 /* compute huffman coded region sizes */

  1387.                 region_address1 = get_bits(&s->gb, 4);

  1388.                 region_address2 = get_bits(&s->gb, 3);

  1389.                 av_dlog(s->avctx, "region1=%d region2=%d\n",

  1390.                         region_address1, region_address2);

  1391.                 ff_init_long_region(s, g, region_address1, region_address2);

  1392.             }

  1393.             ff_region_offset2size(g);

  1394.             ff_compute_band_indexes(s, g);

  1395. 

  1396.             g->preflag = 0;

  1397.             if (!s->lsf)

  1398.                 g->preflag = get_bits1(&s->gb);

  1399.             g->scalefac_scale     = get_bits1(&s->gb);

  1400.             g->count1table_select = get_bits1(&s->gb);

  1401.             av_dlog(s->avctx, "block_type=%d switch_point=%d\n",

  1402.                     g->block_type, g->switch_point);

  1403.         }

  1404.     }

  1405. 

  1406.     if (!s->adu_mode) {

  1407.         int skip;

  1408.         const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);

  1409.         int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0, EXTRABYTES);

  1410.         av_assert1((get_bits_count(&s->gb) & 7) == 0);

  1411.         /* now we get bits from the main_data_begin offset */

  1412.         av_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",

  1413.                 main_data_begin, s->last_buf_size);

  1414. 

  1415.         memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);

  1416.         s->in_gb = s->gb;

  1417.         init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);

  1418. #if !UNCHECKED_BITSTREAM_READER

  1419.         s->gb.size_in_bits_plus8 += FFMAX(extrasize, LAST_BUF_SIZE - s->last_buf_size) * 8;

  1420. #endif

  1421.         s->last_buf_size <<= 3;

  1422.         for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {

  1423.             for (ch = 0; ch < s->nb_channels; ch++) {

  1424.                 g = &s->granules[ch][gr];

  1425.                 s->last_buf_size += g->part2_3_length;

  1426.                 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));

  1427.                 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);

  1428.             }

  1429.         }

  1430.         skip = s->last_buf_size - 8 * main_data_begin;

  1431.         if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {

  1432.             skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);

  1433.             s->gb           = s->in_gb;

  1434.             s->in_gb.buffer = NULL;

  1435.         } else {

  1436.             skip_bits_long(&s->gb, skip);

  1437.         }

  1438.     } else {

  1439.         gr = 0;

  1440.     }

  1441. 

  1442.     for (; gr < nb_granules; gr++) {

  1443.         for (ch = 0; ch < s->nb_channels; ch++) {

  1444.             g = &s->granules[ch][gr];

  1445.             bits_pos = get_bits_count(&s->gb);

  1446. 

  1447.             if (!s->lsf) {

  1448.                 uint8_t *sc;

  1449.                 int slen, slen1, slen2;

  1450. 

  1451.                 /* MPEG1 scale factors */

  1452.                 slen1 = slen_table[0][g->scalefac_compress];

  1453.                 slen2 = slen_table[1][g->scalefac_compress];

  1454.                 av_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);

  1455.                 if (g->block_type == 2) {

  1456.                     n = g->switch_point ? 17 : 18;

  1457.                     j = 0;

  1458.                     if (slen1) {

  1459.                         for (i = 0; i < n; i++)

  1460.                             g->scale_factors[j++] = get_bits(&s->gb, slen1);

  1461.                     } else {

  1462.                         for (i = 0; i < n; i++)

  1463.                             g->scale_factors[j++] = 0;

  1464.                     }

  1465.                     if (slen2) {

  1466.                         for (i = 0; i < 18; i++)

  1467.                             g->scale_factors[j++] = get_bits(&s->gb, slen2);

  1468.                         for (i = 0; i < 3; i++)

  1469.                             g->scale_factors[j++] = 0;

  1470.                     } else {

  1471.                         for (i = 0; i < 21; i++)

  1472.                             g->scale_factors[j++] = 0;

  1473.                     }

  1474.                 } else {

  1475.                     sc = s->granules[ch][0].scale_factors;

  1476.                     j = 0;

  1477.                     for (k = 0; k < 4; k++) {

  1478.                         n = k == 0 ? 6 : 5;

  1479.                         if ((g->scfsi & (0x8 >> k)) == 0) {

  1480.                             slen = (k < 2) ? slen1 : slen2;

  1481.                             if (slen) {

  1482.                                 for (i = 0; i < n; i++)

  1483.                                     g->scale_factors[j++] = get_bits(&s->gb, slen);

  1484.                             } else {

  1485.                                 for (i = 0; i < n; i++)

  1486.                                     g->scale_factors[j++] = 0;

  1487.                             }

  1488.                         } else {

  1489.                             /* simply copy from last granule */

  1490.                             for (i = 0; i < n; i++) {

  1491.                                 g->scale_factors[j] = sc[j];

  1492.                                 j++;

  1493.                             }

  1494.                         }

  1495.                     }

  1496.                     g->scale_factors[j++] = 0;

  1497.                 }

  1498.             } else {

  1499.                 int tindex, tindex2, slen[4], sl, sf;

  1500. 

  1501.                 /* LSF scale factors */

  1502.                 if (g->block_type == 2)

  1503.                     tindex = g->switch_point ? 2 : 1;

  1504.                 else

  1505.                     tindex = 0;

  1506. 

  1507.                 sf = g->scalefac_compress;

  1508.                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {

  1509.                     /* intensity stereo case */

  1510.                     sf >>= 1;

  1511.                     if (sf < 180) {

  1512.                         lsf_sf_expand(slen, sf, 6, 6, 0);

  1513.                         tindex2 = 3;

  1514.                     } else if (sf < 244) {

  1515.                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);

  1516.                         tindex2 = 4;

  1517.                     } else {

  1518.                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);

  1519.                         tindex2 = 5;

  1520.                     }

  1521.                 } else {

  1522.                     /* normal case */

  1523.                     if (sf < 400) {

  1524.                         lsf_sf_expand(slen, sf, 5, 4, 4);

  1525.                         tindex2 = 0;

  1526.                     } else if (sf < 500) {

  1527.                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);

  1528.                         tindex2 = 1;

  1529.                     } else {

  1530.                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);

  1531.                         tindex2 = 2;

  1532.                         g->preflag = 1;

  1533.                     }

  1534.                 }

  1535. 

  1536.                 j = 0;

  1537.                 for (k = 0; k < 4; k++) {

  1538.                     n  = lsf_nsf_table[tindex2][tindex][k];

  1539.                     sl = slen[k];

  1540.                     if (sl) {

  1541.                         for (i = 0; i < n; i++)

  1542.                             g->scale_factors[j++] = get_bits(&s->gb, sl);

  1543.                     } else {

  1544.                         for (i = 0; i < n; i++)

  1545.                             g->scale_factors[j++] = 0;

  1546.                     }

  1547.                 }

  1548.                 /* XXX: should compute exact size */

  1549.                 for (; j < 40; j++)

  1550.                     g->scale_factors[j] = 0;

  1551.             }

  1552. 

  1553.             exponents_from_scale_factors(s, g, exponents);

  1554. 

  1555.             /* read Huffman coded residue */

  1556.             huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);

  1557.         } /* ch */

  1558. 

  1559.         if (s->mode == MPA_JSTEREO)

  1560.             compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);

  1561. 

  1562.         for (ch = 0; ch < s->nb_channels; ch++) {

  1563.             g = &s->granules[ch][gr];

  1564. 

  1565.             reorder_block(s, g);

  1566.             compute_antialias(s, g);

  1567.             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);

  1568.         }

  1569.     } /* gr */

  1570.     if (get_bits_count(&s->gb) < 0)

  1571.         skip_bits_long(&s->gb, -get_bits_count(&s->gb));

  1572.     return nb_granules * 18;

  1573. }

  1574. 

  1575. static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,

  1576.                            const uint8_t *buf, int buf_size)

  1577. {

  1578.     int i, nb_frames, ch, ret;

  1579.     OUT_INT *samples_ptr;

  1580. 

  1581.     init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);

  1582. 

  1583.     /* skip error protection field */

  1584.     if (s->error_protection)

  1585.         skip_bits(&s->gb, 16);

  1586. 

  1587.     switch(s->layer) {

  1588.     case 1:

  1589.         s->avctx->frame_size = 384;

  1590.         nb_frames = mp_decode_layer1(s);

  1591.         break;

  1592.     case 2:

  1593.         s->avctx->frame_size = 1152;

  1594.         nb_frames = mp_decode_layer2(s);

  1595.         break;

  1596.     case 3:

  1597.         s->avctx->frame_size = s->lsf ? 576 : 1152;

  1598.     default:

  1599.         nb_frames = mp_decode_layer3(s);

  1600. 

  1601.         s->last_buf_size=0;

  1602.         if (s->in_gb.buffer) {

  1603.             align_get_bits(&s->gb);

  1604.             i = get_bits_left(&s->gb)>>3;

  1605.             if (i >= 0 && i <= BACKSTEP_SIZE) {

  1606.                 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);

  1607.                 s->last_buf_size=i;

  1608.             } else

  1609.                 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);

  1610.             s->gb           = s->in_gb;

  1611.             s->in_gb.buffer = NULL;

  1612.         }

  1613. 

  1614.         align_get_bits(&s->gb);

  1615.         av_assert1((get_bits_count(&s->gb) & 7) == 0);

  1616.         i = get_bits_left(&s->gb) >> 3;

  1617. 

  1618.         if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {

  1619.             if (i < 0)

  1620.                 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);

  1621.             i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);

  1622.         }

  1623.         av_assert1(i <= buf_size - HEADER_SIZE && i >= 0);

  1624.         memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);

  1625.         s->last_buf_size += i;

  1626.     }

  1627. 

  1628.     if(nb_frames < 0)

  1629.         return nb_frames;

  1630. 

  1631.     /* get output buffer */

  1632.     if (!samples) {

  1633.         s->frame.nb_samples = s->avctx->frame_size;

  1634.         if ((ret = ff_get_buffer(s->avctx, &s->frame)) < 0) {

  1635.             av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");

  1636.             return ret;

  1637.         }

  1638.         samples = (OUT_INT **)s->frame.extended_data;

  1639.     }

  1640. 

  1641.     /* apply the synthesis filter */

  1642.     for (ch = 0; ch < s->nb_channels; ch++) {

  1643.         int sample_stride;

  1644.         if (s->avctx->sample_fmt == OUT_FMT_P) {

  1645.             samples_ptr   = samples[ch];

  1646.             sample_stride = 1;

  1647.         } else {

  1648.             samples_ptr   = samples[0] + ch;

  1649.             sample_stride = s->nb_channels;

  1650.         }

  1651.         for (i = 0; i < nb_frames; i++) {

  1652.             RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],

  1653.                                         &(s->synth_buf_offset[ch]),

  1654.                                         RENAME(ff_mpa_synth_window),

  1655.                                         &s->dither_state, samples_ptr,

  1656.                                         sample_stride, s->sb_samples[ch][i]);

  1657.             samples_ptr += 32 * sample_stride;

  1658.         }

  1659.     }

  1660. 

  1661.     return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;

  1662. }

  1663. 

  1664. static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,

  1665.                         AVPacket *avpkt)

  1666. {

  1667.     const uint8_t *buf  = avpkt->data;

  1668.     int buf_size        = avpkt->size;

  1669.     MPADecodeContext *s = avctx->priv_data;

  1670.     uint32_t header;

  1671.     int ret;

  1672. 

  1673.     while(buf_size && !*buf){

  1674.         buf++;

  1675.         buf_size--;

  1676.     }

  1677. 

  1678.     if (buf_size < HEADER_SIZE)

  1679.         return AVERROR_INVALIDDATA;

  1680. 

  1681.     header = AV_RB32(buf);

  1682.     if (header>>8 == AV_RB32("TAG")>>8) {

  1683.         av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n");

  1684.         return buf_size;

  1685.     }

  1686.     if (ff_mpa_check_header(header) < 0) {

  1687.         av_log(avctx, AV_LOG_ERROR, "Header missing\n");

  1688.         return AVERROR_INVALIDDATA;

  1689.     }

  1690. 

  1691.     if (avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {

  1692.         /* free format: prepare to compute frame size */

  1693.         s->frame_size = -1;

  1694.         return AVERROR_INVALIDDATA;

  1695.     }

  1696.     /* update codec info */

  1697.     avctx->channels       = s->nb_channels;

  1698.     avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;

  1699.     if (!avctx->bit_rate)

  1700.         avctx->bit_rate = s->bit_rate;

  1701. 

  1702.     if (s->frame_size <= 0 || s->frame_size > buf_size) {

  1703.         av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");

  1704.         return AVERROR_INVALIDDATA;

  1705.     } else if (s->frame_size < buf_size) {

  1706.         av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");

  1707.         buf_size= s->frame_size;

  1708.     }

  1709. 

  1710.     ret = mp_decode_frame(s, NULL, buf, buf_size);

  1711.     if (ret >= 0) {

  1712.         *got_frame_ptr   = 1;

  1713.         *(AVFrame *)data = s->frame;

  1714.         avctx->sample_rate = s->sample_rate;

  1715.         //FIXME maybe move the other codec info stuff from above here too

  1716.     } else {

  1717.         av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");

  1718.         /* Only return an error if the bad frame makes up the whole packet or

  1719.          * the error is related to buffer management.

  1720.          * If there is more data in the packet, just consume the bad frame

  1721.          * instead of returning an error, which would discard the whole

  1722.          * packet. */

  1723.         *got_frame_ptr = 0;

  1724.         if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)

  1725.             return ret;

  1726.     }

  1727.     s->frame_size = 0;

  1728.     return buf_size;

  1729. }

  1730. 

  1731. static void mp_flush(MPADecodeContext *ctx)

  1732. {

  1733.     memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));

  1734.     ctx->last_buf_size = 0;

  1735. }

  1736. 

  1737. static void flush(AVCodecContext *avctx)

  1738. {

  1739.     mp_flush(avctx->priv_data);

  1740. }

  1741. 

  1742. #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER

  1743. static int decode_frame_adu(AVCodecContext *avctx, void *data,

  1744.                             int *got_frame_ptr, AVPacket *avpkt)

  1745. {

  1746.     const uint8_t *buf  = avpkt->data;

  1747.     int buf_size        = avpkt->size;

  1748.     MPADecodeContext *s = avctx->priv_data;

  1749.     uint32_t header;

  1750.     int len, ret;

  1751.     int av_unused out_size;

  1752. 

  1753.     len = buf_size;

  1754. 

  1755.     // Discard too short frames

  1756.     if (buf_size < HEADER_SIZE) {

  1757.         av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");

  1758.         return AVERROR_INVALIDDATA;

  1759.     }

  1760. 

  1761. 

  1762.     if (len > MPA_MAX_CODED_FRAME_SIZE)

  1763.         len = MPA_MAX_CODED_FRAME_SIZE;

  1764. 

  1765.     // Get header and restore sync word

  1766.     header = AV_RB32(buf) | 0xffe00000;

  1767. 

  1768.     if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame

  1769.         av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");

  1770.         return AVERROR_INVALIDDATA;

  1771.     }

  1772. 

  1773.     avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);

  1774.     /* update codec info */

  1775.     avctx->sample_rate = s->sample_rate;

  1776.     avctx->channels    = s->nb_channels;

  1777.     if (!avctx->bit_rate)

  1778.         avctx->bit_rate = s->bit_rate;

  1779. 

  1780.     s->frame_size = len;

  1781. 

  1782.     ret = mp_decode_frame(s, NULL, buf, buf_size);

  1783.     if (ret < 0) {

  1784.         av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");

  1785.         return ret;

  1786.     }

  1787. 

  1788.     *got_frame_ptr   = 1;

  1789.     *(AVFrame *)data = s->frame;

  1790. 

  1791.     return buf_size;

  1792. }

  1793. #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */

  1794. 

  1795. #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER

  1796. 

  1797. /**

  1798.  * Context for MP3On4 decoder

  1799.  */

  1800. typedef struct MP3On4DecodeContext {

  1801.     AVFrame *frame;

  1802.     int frames;                     ///< number of mp3 frames per block (number of mp3 decoder instances)

  1803.     int syncword;                   ///< syncword patch

  1804.     const uint8_t *coff;            ///< channel offsets in output buffer

  1805.     MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance

  1806. } MP3On4DecodeContext;

  1807. 

  1808. #include "mpeg4audio.h"

  1809. 

  1810. /* Next 3 arrays are indexed by channel config number (passed via codecdata) */

  1811. 

  1812. /* number of mp3 decoder instances */

  1813. static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };

  1814. 

  1815. /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */

  1816. static const uint8_t chan_offset[8][5] = {

  1817.     { 0             },

  1818.     { 0             },  // C

  1819.     { 0             },  // FLR

  1820.     { 2, 0          },  // C FLR

  1821.     { 2, 0, 3       },  // C FLR BS

  1822.     { 2, 0, 3       },  // C FLR BLRS

  1823.     { 2, 0, 4, 3    },  // C FLR BLRS LFE

  1824.     { 2, 0, 6, 4, 3 },  // C FLR BLRS BLR LFE

  1825. };

  1826. 

  1827. /* mp3on4 channel layouts */

  1828. static const int16_t chan_layout[8] = {

  1829.     0,

  1830.     AV_CH_LAYOUT_MONO,

  1831.     AV_CH_LAYOUT_STEREO,

  1832.     AV_CH_LAYOUT_SURROUND,

  1833.     AV_CH_LAYOUT_4POINT0,

  1834.     AV_CH_LAYOUT_5POINT0,

  1835.     AV_CH_LAYOUT_5POINT1,

  1836.     AV_CH_LAYOUT_7POINT1

  1837. };

  1838. 

  1839. static av_cold int decode_close_mp3on4(AVCodecContext * avctx)

  1840. {

  1841.     MP3On4DecodeContext *s = avctx->priv_data;

  1842.     int i;

  1843. 

  1844.     for (i = 0; i < s->frames; i++)

  1845.         av_free(s->mp3decctx[i]);

  1846. 

  1847.     return 0;

  1848. }

  1849. 

  1850. 

  1851. static int decode_init_mp3on4(AVCodecContext * avctx)

  1852. {

  1853.     MP3On4DecodeContext *s = avctx->priv_data;

  1854.     MPEG4AudioConfig cfg;

  1855.     int i;

  1856. 

  1857.     if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {

  1858.         av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");

  1859.         return AVERROR_INVALIDDATA;

  1860.     }

  1861. 

  1862.     avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,

  1863.                                  avctx->extradata_size * 8, 1);

  1864.     if (!cfg.chan_config || cfg.chan_config > 7) {

  1865.         av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");

  1866.         return AVERROR_INVALIDDATA;

  1867.     }

  1868.     s->frames             = mp3Frames[cfg.chan_config];

  1869.     s->coff               = chan_offset[cfg.chan_config];

  1870.     avctx->channels       = ff_mpeg4audio_channels[cfg.chan_config];

  1871.     avctx->channel_layout = chan_layout[cfg.chan_config];

  1872. 

  1873.     if (cfg.sample_rate < 16000)

  1874.         s->syncword = 0xffe00000;

  1875.     else

  1876.         s->syncword = 0xfff00000;

  1877. 

  1878.     /* Init the first mp3 decoder in standard way, so that all tables get builded

  1879.      * We replace avctx->priv_data with the context of the first decoder so that

  1880.      * decode_init() does not have to be changed.

  1881.      * Other decoders will be initialized here copying data from the first context

  1882.      */

  1883.     // Allocate zeroed memory for the first decoder context

  1884.     s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));

  1885.     if (!s->mp3decctx[0])

  1886.         goto alloc_fail;

  1887.     // Put decoder context in place to make init_decode() happy

  1888.     avctx->priv_data = s->mp3decctx[0];

  1889.     decode_init(avctx);

  1890.     s->frame = avctx->coded_frame;

  1891.     // Restore mp3on4 context pointer

  1892.     avctx->priv_data = s;

  1893.     s->mp3decctx[0]->adu_mode = 1; // Set adu mode

  1894. 

  1895.     /* Create a separate codec/context for each frame (first is already ok).

  1896.      * Each frame is 1 or 2 channels - up to 5 frames allowed

  1897.      */

  1898.     for (i = 1; i < s->frames; i++) {

  1899.         s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));

  1900.         if (!s->mp3decctx[i])

  1901.             goto alloc_fail;

  1902.         s->mp3decctx[i]->adu_mode = 1;

  1903.         s->mp3decctx[i]->avctx = avctx;

  1904.         s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;

  1905.     }

  1906. 

  1907.     return 0;

  1908. alloc_fail:

  1909.     decode_close_mp3on4(avctx);

  1910.     return AVERROR(ENOMEM);

  1911. }

  1912. 

  1913. 

  1914. static void flush_mp3on4(AVCodecContext *avctx)

  1915. {

  1916.     int i;

  1917.     MP3On4DecodeContext *s = avctx->priv_data;

  1918. 

  1919.     for (i = 0; i < s->frames; i++)

  1920.         mp_flush(s->mp3decctx[i]);

  1921. }

  1922. 

  1923. 

  1924. static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,

  1925.                                int *got_frame_ptr, AVPacket *avpkt)

  1926. {

  1927.     const uint8_t *buf     = avpkt->data;

  1928.     int buf_size           = avpkt->size;

  1929.     MP3On4DecodeContext *s = avctx->priv_data;

  1930.     MPADecodeContext *m;

  1931.     int fsize, len = buf_size, out_size = 0;

  1932.     uint32_t header;

  1933.     OUT_INT **out_samples;

  1934.     OUT_INT *outptr[2];

  1935.     int fr, ch, ret;

  1936. 

  1937.     /* get output buffer */

  1938.     s->frame->nb_samples = MPA_FRAME_SIZE;

  1939.     if ((ret = ff_get_buffer(avctx, s->frame)) < 0) {

  1940.         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");

  1941.         return ret;

  1942.     }

  1943.     out_samples = (OUT_INT **)s->frame->extended_data;

  1944. 

  1945.     // Discard too short frames

  1946.     if (buf_size < HEADER_SIZE)

  1947.         return AVERROR_INVALIDDATA;

  1948. 

  1949.     avctx->bit_rate = 0;

  1950. 

  1951.     ch = 0;

  1952.     for (fr = 0; fr < s->frames; fr++) {

  1953.         fsize = AV_RB16(buf) >> 4;

  1954.         fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);

  1955.         m     = s->mp3decctx[fr];

  1956.         av_assert1(m);

  1957. 

  1958.         if (fsize < HEADER_SIZE) {

  1959.             av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");

  1960.             return AVERROR_INVALIDDATA;

  1961.         }

  1962.         header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header

  1963. 

  1964.         if (ff_mpa_check_header(header) < 0) // Bad header, discard block

  1965.             break;

  1966. 

  1967.         avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);

  1968. 

  1969.         if (ch + m->nb_channels > avctx->channels || s->coff[fr] + m->nb_channels > avctx->channels) {

  1970.             av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "

  1971.                                         "channel count\n");

  1972.             return AVERROR_INVALIDDATA;

  1973.         }

  1974.         ch += m->nb_channels;

  1975. 

  1976.         outptr[0] = out_samples[s->coff[fr]];

  1977.         if (m->nb_channels > 1)

  1978.             outptr[1] = out_samples[s->coff[fr] + 1];

  1979. 

  1980.         if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0)

  1981.             return ret;

  1982. 

  1983.         out_size += ret;

  1984.         buf      += fsize;

  1985.         len      -= fsize;

  1986. 

  1987.         avctx->bit_rate += m->bit_rate;

  1988.     }

  1989. 

  1990.     /* update codec info */

  1991.     avctx->sample_rate = s->mp3decctx[0]->sample_rate;

  1992. 

  1993.     s->frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));

  1994.     *got_frame_ptr   = 1;

  1995.     *(AVFrame *)data = *s->frame;

  1996. 

  1997.     return buf_size;

  1998. }

  1999. #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */

  2000. 

  2001. #if !CONFIG_FLOAT

  2002. #if CONFIG_MP1_DECODER

  2003. AVCodec ff_mp1_decoder = {

  2004.     .name           = "mp1",

  2005.     .type           = AVMEDIA_TYPE_AUDIO,

  2006.     .id             = AV_CODEC_ID_MP1,

  2007.     .priv_data_size = sizeof(MPADecodeContext),

  2008.     .init           = decode_init,

  2009.     .decode         = decode_frame,

  2010.     .capabilities   = CODEC_CAP_DR1,

  2011.     .flush          = flush,

  2012.     .long_name      = NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),

  2013.     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,

  2014.                                                       AV_SAMPLE_FMT_S16,

  2015.                                                       AV_SAMPLE_FMT_NONE },

  2016. };

  2017. #endif

  2018. #if CONFIG_MP2_DECODER

  2019. AVCodec ff_mp2_decoder = {

  2020.     .name           = "mp2",

  2021.     .type           = AVMEDIA_TYPE_AUDIO,

  2022.     .id             = AV_CODEC_ID_MP2,

  2023.     .priv_data_size = sizeof(MPADecodeContext),

  2024.     .init           = decode_init,

  2025.     .decode         = decode_frame,

  2026.     .capabilities   = CODEC_CAP_DR1,

  2027.     .flush          = flush,

  2028.     .long_name      = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),

  2029.     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,

  2030.                                                       AV_SAMPLE_FMT_S16,

  2031.                                                       AV_SAMPLE_FMT_NONE },

  2032. };

  2033. #endif

  2034. #if CONFIG_MP3_DECODER

  2035. AVCodec ff_mp3_decoder = {

  2036.     .name           = "mp3",

  2037.     .type           = AVMEDIA_TYPE_AUDIO,

  2038.     .id             = AV_CODEC_ID_MP3,

  2039.     .priv_data_size = sizeof(MPADecodeContext),

  2040.     .init           = decode_init,

  2041.     .decode         = decode_frame,

  2042.     .capabilities   = CODEC_CAP_DR1,

  2043.     .flush          = flush,

  2044.     .long_name      = NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),

  2045.     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,

  2046.                                                       AV_SAMPLE_FMT_S16,

  2047.                                                       AV_SAMPLE_FMT_NONE },

  2048. };

  2049. #endif

  2050. #if CONFIG_MP3ADU_DECODER

  2051. AVCodec ff_mp3adu_decoder = {

  2052.     .name           = "mp3adu",

  2053.     .type           = AVMEDIA_TYPE_AUDIO,

  2054.     .id             = AV_CODEC_ID_MP3ADU,

  2055.     .priv_data_size = sizeof(MPADecodeContext),

  2056.     .init           = decode_init,

  2057.     .decode         = decode_frame_adu,

  2058.     .capabilities   = CODEC_CAP_DR1,

  2059.     .flush          = flush,

  2060.     .long_name      = NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),

  2061.     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,

  2062.                                                       AV_SAMPLE_FMT_S16,

  2063.                                                       AV_SAMPLE_FMT_NONE },

  2064. };

  2065. #endif

  2066. #if CONFIG_MP3ON4_DECODER

  2067. AVCodec ff_mp3on4_decoder = {

  2068.     .name           = "mp3on4",

  2069.     .type           = AVMEDIA_TYPE_AUDIO,

  2070.     .id             = AV_CODEC_ID_MP3ON4,

  2071.     .priv_data_size = sizeof(MP3On4DecodeContext),

  2072.     .init           = decode_init_mp3on4,

  2073.     .close          = decode_close_mp3on4,

  2074.     .decode         = decode_frame_mp3on4,

  2075.     .capabilities   = CODEC_CAP_DR1,

  2076.     .flush          = flush_mp3on4,

  2077.     .long_name      = NULL_IF_CONFIG_SMALL("MP3onMP4"),

  2078.     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,

  2079.                                                       AV_SAMPLE_FMT_NONE },

  2080. };

  2081. #endif

  2082. #endif