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.
100005 Commits
32 Branches
303MB
Tree: 16bb8247b4
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '16bb8247b4'
${ noResults }
FFmpeg/libavcodec/mpegaudiodec_template.c

2032 lines
66KB
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/attributes.h"

  28. #include "libavutil/avassert.h"

  29. #include "libavutil/channel_layout.h"

  30. #include "libavutil/crc.h"

  31. #include "libavutil/float_dsp.h"

  32. #include "libavutil/libm.h"

  33. #include "avcodec.h"

  34. #include "get_bits.h"

  35. #include "internal.h"

  36. #include "mathops.h"

  37. #include "mpegaudiodsp.h"

  38. 

  39. /*

  40.  * TODO:

  41.  *  - test lsf / mpeg25 extensively.

  42.  */

  43. 

  44. #include "mpegaudio.h"

  45. #include "mpegaudiodecheader.h"

  46. 

  47. #define BACKSTEP_SIZE 512

  48. #define EXTRABYTES 24

  49. #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES

  50. 

  51. /* layer 3 "granule" */

  52. typedef struct GranuleDef {

  53.     uint8_t scfsi;

  54.     int part2_3_length;

  55.     int big_values;

  56.     int global_gain;

  57.     int scalefac_compress;

  58.     uint8_t block_type;

  59.     uint8_t switch_point;

  60.     int table_select[3];

  61.     int subblock_gain[3];

  62.     uint8_t scalefac_scale;

  63.     uint8_t count1table_select;

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

  65.     int preflag;

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

  67.     uint8_t scale_factors[40];

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

  69. } GranuleDef;

  70. 

  71. typedef struct MPADecodeContext {

  72.     MPA_DECODE_HEADER

  73.     uint8_t last_buf[LAST_BUF_SIZE];

  74.     int last_buf_size;

  75.     int extrasize;

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

  77.     uint32_t free_format_next_header;

  78.     GetBitContext gb;

  79.     GetBitContext in_gb;

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

  81.     int synth_buf_offset[MPA_MAX_CHANNELS];

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

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

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

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

  86.     int dither_state;

  87.     int err_recognition;

  88.     AVCodecContext* avctx;

  89.     MPADSPContext mpadsp;

  90.     void (*butterflies_float)(float *av_restrict v1, float *av_restrict v2, int len);

  91.     AVFrame *frame;

  92.     uint32_t crc;

  93. } MPADecodeContext;

  94. 

  95. #define HEADER_SIZE 4

  96. 

  97. #include "mpegaudiodata.h"

  98. #include "mpegaudiodectab.h"

  99. 

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

  101. static VLC huff_vlc[16];

  102. static VLC_TYPE huff_vlc_tables[

  103.     0 + 128 + 128 + 128 + 130 + 128 + 154 + 166 +

  104.   142 + 204 + 190 + 170 + 542 + 460 + 662 + 414

  105.   ][2];

  106. static const int huff_vlc_tables_sizes[16] = {

  107.     0,  128,  128,  128,  130,  128,  154,  166,

  108.   142,  204,  190,  170,  542,  460,  662,  414

  109. };

  110. static VLC huff_quad_vlc[2];

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

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

  113. /* computed from band_size_long */

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

  115. #include "mpegaudio_tablegen.h"

  116. /* intensity stereo coef table */

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

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

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

  120. 

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

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

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

  124. 

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

  126.     division_tab3, division_tab5, NULL, division_tab9

  127. };

  128. 

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

  130. static uint16_t scale_factor_modshift[64];

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

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

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

  134. 

  135. #define SCALE_GEN(v) \

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

  137. 

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

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

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

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

  142. };

  143. 

  144. /**

  145.  * Convert region offsets to region sizes and truncate

  146.  * size to big_values.

  147.  */

  148. static void region_offset2size(GranuleDef *g)

  149. {

  150.     int i, k, j = 0;

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

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

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

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

  155.         j = k;

  156.     }

  157. }

  158. 

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

  160. {

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

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

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

  164.         else

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

  166.     } else {

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

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

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

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

  171.         else

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

  173.     }

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

  175. }

  176. 

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

  178.                              int ra1, int ra2)

  179. {

  180.     int l;

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

  182.     /* should not overflow */

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

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

  185. }

  186. 

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

  188. {

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

  190.         if (g->switch_point) {

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

  192.                 avpriv_request_sample(s->avctx, "switch point in 8khz");

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

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

  195.                 exponents as long blocks */

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

  197.                 g->long_end = 8;

  198.             else

  199.                 g->long_end = 6;

  200. 

  201.             g->short_start = 3;

  202.         } else {

  203.             g->long_end    = 0;

  204.             g->short_start = 0;

  205.         }

  206.     } else {

  207.         g->short_start = 13;

  208.         g->long_end    = 22;

  209.     }

  210. }

  211. 

  212. /* layer 1 unscaling */

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

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

  215. {

  216.     int shift, mod;

  217.     int64_t val;

  218. 

  219.     shift   = scale_factor_modshift[scale_factor];

  220.     mod     = shift & 3;

  221.     shift >>= 2;

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

  223.     shift  += n;

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

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

  226. }

  227. 

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

  229. {

  230.     int shift, mod, val;

  231. 

  232.     shift   = scale_factor_modshift[scale_factor];

  233.     mod     = shift & 3;

  234.     shift >>= 2;

  235. 

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

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

  238.     if (shift > 0)

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

  240.     return val;

  241. }

  242. 

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

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

  245. {

  246.     unsigned int m;

  247.     int e;

  248. 

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

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

  251.     e -= exponent >> 2;

  252. #ifdef DEBUG

  253.     if(e < 1)

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

  255. #endif

  256.     if (e > (SUINT)31)

  257.         return 0;

  258.     m = (m + ((1U << e) >> 1)) >> e;

  259. 

  260.     return m;

  261. }

  262. 

  263. static av_cold void decode_init_static(void)

  264. {

  265.     int i, j, k;

  266.     int offset;

  267. 

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

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

  270.         int shift, mod;

  271.         /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */

  272.         shift = i / 3;

  273.         mod   = i % 3;

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

  275.     }

  276. 

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

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

  279.         int n, norm;

  280.         n = i + 2;

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

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

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

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

  285.         ff_dlog(NULL, "%d: norm=%x s=%"PRIx32" %"PRIx32" %"PRIx32"\n", i,

  286.                 (unsigned)norm,

  287.                 scale_factor_mult[i][0],

  288.                 scale_factor_mult[i][1],

  289.                 scale_factor_mult[i][2]);

  290.     }

  291. 

  292.     /* huffman decode tables */

  293.     offset = 0;

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

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

  296.         int xsize, x, y;

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

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

  299. 

  300.         xsize = h->xsize;

  301. 

  302.         j = 0;

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

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

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

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

  307.             }

  308.         }

  309. 

  310.         /* XXX: fail test */

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

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

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

  314.                  tmp_bits, 1, 1, tmp_codes, 2, 2,

  315.                  INIT_VLC_USE_NEW_STATIC);

  316.         offset += huff_vlc_tables_sizes[i];

  317.     }

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

  319. 

  320.     offset = 0;

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

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

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

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

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

  326.                  INIT_VLC_USE_NEW_STATIC);

  327.         offset += huff_quad_vlc_tables_sizes[i];

  328.     }

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

  330. 

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

  332.         k = 0;

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

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

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

  336.         }

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

  338.     }

  339. 

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

  341. 

  342.     mpegaudio_tableinit();

  343. 

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

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

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

  347.                 int val1, val2, val3, steps;

  348.                 int val = j;

  349.                 steps   = ff_mpa_quant_steps[i];

  350.                 val1    = val % steps;

  351.                 val    /= steps;

  352.                 val2    = val % steps;

  353.                 val3    = val / steps;

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

  355.             }

  356.         }

  357.     }

  358. 

  359. 

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

  361.         float f;

  362.         INTFLOAT v;

  363.         if (i != 6) {

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

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

  366.         } else {

  367.             v = FIXR(1.0);

  368.         }

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

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

  371.     }

  372.     /* invalid values */

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

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

  375. 

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

  377.         double f;

  378.         int e, k;

  379. 

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

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

  382.             f = exp2(e / 4.0);

  383.             k = i & 1;

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

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

  386.             ff_dlog(NULL, "is_table_lsf %d %d: %f %f\n",

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

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

  389.         }

  390.     }

  391. 

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

  393.         double ci, cs, ca;

  394.         ci = ci_table[i];

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

  396.         ca = cs * ci;

  397. #if !USE_FLOATS

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

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

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

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

  402. #else

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

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

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

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

  407. #endif

  408.     }

  409.     RENAME(ff_mpa_synth_init)();

  410. }

  411. 

  412. static av_cold int decode_init(AVCodecContext * avctx)

  413. {

  414.     static int initialized_tables = 0;

  415.     MPADecodeContext *s = avctx->priv_data;

  416. 

  417.     if (!initialized_tables) {

  418.         decode_init_static();

  419.         initialized_tables = 1;

  420.     }

  421. 

  422.     s->avctx = avctx;

  423. 

  424. #if USE_FLOATS

  425.     {

  426.         AVFloatDSPContext *fdsp;

  427.         fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);

  428.         if (!fdsp)

  429.             return AVERROR(ENOMEM);

  430.         s->butterflies_float = fdsp->butterflies_float;

  431.         av_free(fdsp);

  432.     }

  433. #endif

  434. 

  435.     ff_mpadsp_init(&s->mpadsp);

  436. 

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

  438.         avctx->codec_id != AV_CODEC_ID_MP3ON4)

  439.         avctx->sample_fmt = OUT_FMT;

  440.     else

  441.         avctx->sample_fmt = OUT_FMT_P;

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

  443. 

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

  445.         s->adu_mode = 1;

  446. 

  447.     return 0;

  448. }

  449. 

  450. #define C3 FIXHR(0.86602540378443864676/2)

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

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

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

  454. 

  455. /* 12 points IMDCT. We compute it "by hand" by factorizing obvious

  456.    cases. */

  457. static void imdct12(INTFLOAT *out, SUINTFLOAT *in)

  458. {

  459.     SUINTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;

  460. 

  461.     in0  = in[0*3];

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

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

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

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

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

  467.     in5 += in3;

  468.     in3 += in1;

  469. 

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

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

  472. 

  473.     t1   = in0 - in4;

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

  475. 

  476.     out[ 7] =

  477.     out[10] = t1 + t2;

  478.     out[ 1] =

  479.     out[ 4] = t1 - t2;

  480. 

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

  482.     in4     = in0 + in2;

  483.     in5    += 2*in1;

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

  485.     out[ 8] =

  486.     out[ 9] = in4 + in1;

  487.     out[ 2] =

  488.     out[ 3] = in4 - in1;

  489. 

  490.     in0    -= in2;

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

  492.     out[ 0] =

  493.     out[ 5] = in0 - in5;

  494.     out[ 6] =

  495.     out[11] = in0 + in5;

  496. }

  497. 

  498. static int handle_crc(MPADecodeContext *s, int sec_len)

  499. {

  500.     if (s->error_protection && (s->err_recognition & AV_EF_CRCCHECK)) {

  501.         const uint8_t *buf = s->gb.buffer - HEADER_SIZE;

  502.         int sec_byte_len  = sec_len >> 3;

  503.         int sec_rem_bits  = sec_len & 7;

  504.         const AVCRC *crc_tab = av_crc_get_table(AV_CRC_16_ANSI);

  505.         uint8_t tmp_buf[4];

  506.         uint32_t crc_val = av_crc(crc_tab, UINT16_MAX, &buf[2], 2);

  507.         crc_val = av_crc(crc_tab, crc_val, &buf[6], sec_byte_len);

  508. 

  509.         AV_WB32(tmp_buf,

  510.                 ((buf[6 + sec_byte_len] & (0xFF00 >> sec_rem_bits)) << 24) +

  511.                 ((s->crc << 16) >> sec_rem_bits));

  512. 

  513.         crc_val = av_crc(crc_tab, crc_val, tmp_buf, 3);

  514. 

  515.         if (crc_val) {

  516.             av_log(s->avctx, AV_LOG_ERROR, "CRC mismatch %X!\n", crc_val);

  517.             if (s->err_recognition & AV_EF_EXPLODE)

  518.                 return AVERROR_INVALIDDATA;

  519.         }

  520.     }

  521.     return 0;

  522. }

  523. 

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

  525. static int mp_decode_layer1(MPADecodeContext *s)

  526. {

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

  528.     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];

  529.     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

  530.     int ret;

  531. 

  532.     ret = handle_crc(s, (s->nb_channels == 1) ? 8*16  : 8*32);

  533.     if (ret < 0)

  534.         return ret;

  535. 

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

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

  538.     else

  539.         bound = SBLIMIT;

  540. 

  541.     /* allocation bits */

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

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

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

  545.         }

  546.     }

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

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

  549. 

  550.     /* scale factors */

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

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

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

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

  555.         }

  556.     }

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

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

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

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

  561.         }

  562.     }

  563. 

  564.     /* compute samples */

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

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

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

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

  569.                 if (n) {

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

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

  572.                 } else {

  573.                     v = 0;

  574.                 }

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

  576.             }

  577.         }

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

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

  580.             if (n) {

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

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

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

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

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

  586.             } else {

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

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

  589.             }

  590.         }

  591.     }

  592.     return 12;

  593. }

  594. 

  595. static int mp_decode_layer2(MPADecodeContext *s)

  596. {

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

  598.     const unsigned char *alloc_table;

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

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

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

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

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

  604.     int ret;

  605. 

  606.     /* select decoding table */

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

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

  609.     sblimit     = ff_mpa_sblimit_table[table];

  610.     alloc_table = ff_mpa_alloc_tables[table];

  611. 

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

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

  614.     else

  615.         bound = sblimit;

  616. 

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

  618. 

  619.     /* sanity check */

  620.     if (bound > sblimit)

  621.         bound = sblimit;

  622. 

  623.     /* parse bit allocation */

  624.     j = 0;

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

  626.         bit_alloc_bits = alloc_table[j];

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

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

  629.         j += 1 << bit_alloc_bits;

  630.     }

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

  632.         bit_alloc_bits = alloc_table[j];

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

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

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

  636.         j += 1 << bit_alloc_bits;

  637.     }

  638. 

  639.     /* scale codes */

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

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

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

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

  644.         }

  645.     }

  646. 

  647.     ret = handle_crc(s, get_bits_count(&s->gb) - 16);

  648.     if (ret < 0)

  649.         return ret;

  650. 

  651.     /* scale factors */

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

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

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

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

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

  657.                 default:

  658.                 case 0:

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

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

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

  662.                     break;

  663.                 case 2:

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

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

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

  667.                     break;

  668.                 case 1:

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

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

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

  672.                     break;

  673.                 case 3:

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

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

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

  677.                     break;

  678.                 }

  679.             }

  680.         }

  681.     }

  682. 

  683.     /* samples */

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

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

  686.             j = 0;

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

  688.                 bit_alloc_bits = alloc_table[j];

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

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

  691.                     if (b) {

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

  693.                         qindex = alloc_table[j+b];

  694.                         bits = ff_mpa_quant_bits[qindex];

  695.                         if (bits < 0) {

  696.                             int v2;

  697.                             /* 3 values at the same time */

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

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

  700.                             steps  = ff_mpa_quant_steps[qindex];

  701. 

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

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

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

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

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

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

  708.                         } else {

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

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

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

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

  713.                             }

  714.                         }

  715.                     } else {

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

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

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

  719.                     }

  720.                 }

  721.                 /* next subband in alloc table */

  722.                 j += 1 << bit_alloc_bits;

  723.             }

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

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

  726.                 bit_alloc_bits = alloc_table[j];

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

  728.                 if (b) {

  729.                     int mant, scale0, scale1;

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

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

  732.                     qindex = alloc_table[j + b];

  733.                     bits = ff_mpa_quant_bits[qindex];

  734.                     if (bits < 0) {

  735.                         /* 3 values at the same time */

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

  737.                         steps = ff_mpa_quant_steps[qindex];

  738.                         mant = v % steps;

  739.                         v = v / steps;

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

  741.                             l2_unscale_group(steps, mant, scale0);

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

  743.                             l2_unscale_group(steps, mant, scale1);

  744.                         mant = v % steps;

  745.                         v = v / steps;

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

  747.                             l2_unscale_group(steps, mant, scale0);

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

  749.                             l2_unscale_group(steps, mant, scale1);

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

  751.                             l2_unscale_group(steps, v, scale0);

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

  753.                             l2_unscale_group(steps, v, scale1);

  754.                     } else {

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

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

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

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

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

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

  761.                         }

  762.                     }

  763.                 } else {

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

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

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

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

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

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

  770.                 }

  771.                 /* next subband in alloc table */

  772.                 j += 1 << bit_alloc_bits;

  773.             }

  774.             /* fill remaining samples to zero */

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

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

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

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

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

  780.                 }

  781.             }

  782.         }

  783.     }

  784.     return 3 * 12;

  785. }

  786. 

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

  788.     if (n == 3) {                   \

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

  790.         dst   = sf - 3 * m;         \

  791.         sf    = m;                  \

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

  793.         dst  = sf & 3;              \

  794.         sf >>= 2;                   \

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

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

  797.         dst   = sf - 5 * m;         \

  798.         sf    = m;                  \

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

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

  801.         dst   = sf - 6 * m;         \

  802.         sf    = m;                  \

  803.     } else {                        \

  804.         dst = 0;                    \

  805.     }

  806. 

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

  808.                                            int n3)

  809. {

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

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

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

  813.     slen[0] = sf;

  814. }

  815. 

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

  817.                                          int16_t *exponents)

  818. {

  819.     const uint8_t *bstab, *pretab;

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

  821.     int16_t *exp_ptr;

  822. 

  823.     exp_ptr = exponents;

  824.     gain    = g->global_gain - 210;

  825.     shift   = g->scalefac_scale + 1;

  826. 

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

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

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

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

  831.         len = bstab[i];

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

  833.             *exp_ptr++ = v0;

  834.     }

  835. 

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

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

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

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

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

  841.         k        = g->long_end;

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

  843.             len = bstab[i];

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

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

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

  847.                     *exp_ptr++ = v0;

  848.             }

  849.         }

  850.     }

  851. }

  852. 

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

  854.                           int *end_pos2)

  855. {

  856.     if (s->in_gb.buffer && *pos >= s->gb.size_in_bits - s->extrasize * 8) {

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

  858.         s->in_gb.buffer = NULL;

  859.         s->extrasize    = 0;

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

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

  862.         *end_pos2 =

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

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

  865.     }

  866. }

  867. 

  868. /* Following is an optimized code for

  869.             INTFLOAT v = *src

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

  871.                 v = -v;

  872.             *dst = v;

  873. */

  874. #if USE_FLOATS

  875. #define READ_FLIP_SIGN(dst,src)                     \

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

  877.     AV_WN32A(dst, v);

  878. #else

  879. #define READ_FLIP_SIGN(dst,src)     \

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

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

  882. #endif

  883. 

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

  885.                           int16_t *exponents, int end_pos2)

  886. {

  887.     int s_index;

  888.     int i;

  889.     int last_pos, bits_left;

  890.     VLC *vlc;

  891.     int end_pos = FFMIN(end_pos2, s->gb.size_in_bits - s->extrasize * 8);

  892. 

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

  894.     s_index = 0;

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

  896.         int j, k, l, linbits;

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

  898.         if (j == 0)

  899.             continue;

  900.         /* select vlc table */

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

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

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

  904.         vlc     = &huff_vlc[l];

  905. 

  906.         if (!l) {

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

  908.             s_index += 2 * j;

  909.             continue;

  910.         }

  911. 

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

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

  914.             int exponent, x, y;

  915.             int v;

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

  917. 

  918.             if (pos >= end_pos){

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

  920.                 if (pos >= end_pos)

  921.                     break;

  922.             }

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

  924. 

  925.             if (!y) {

  926.                 g->sb_hybrid[s_index    ] =

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

  928.                 s_index += 2;

  929.                 continue;

  930.             }

  931. 

  932.             exponent= exponents[s_index];

  933. 

  934.             ff_dlog(s->avctx, "region=%d n=%d y=%d exp=%d\n",

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

  936.             if (y & 16) {

  937.                 x = y >> 5;

  938.                 y = y & 0x0f;

  939.                 if (x < 15) {

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

  941.                 } else {

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

  943.                     v  = l3_unscale(x, exponent);

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

  945.                         v = -v;

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

  947.                 }

  948.                 if (y < 15) {

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

  950.                 } else {

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

  952.                     v  = l3_unscale(y, exponent);

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

  954.                         v = -v;

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

  956.                 }

  957.             } else {

  958.                 x = y >> 5;

  959.                 y = y & 0x0f;

  960.                 x += y;

  961.                 if (x < 15) {

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

  963.                 } else {

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

  965.                     v  = l3_unscale(x, exponent);

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

  967.                         v = -v;

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

  969.                 }

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

  971.             }

  972.             s_index += 2;

  973.         }

  974.     }

  975. 

  976.     /* high frequencies */

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

  978.     last_pos = 0;

  979.     while (s_index <= 572) {

  980.         int pos, code;

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

  982.         if (pos >= end_pos) {

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

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

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

  986.                 s_index -= 4;

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

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

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

  990.                     s_index=0;

  991.                 break;

  992.             }

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

  994.             if (pos >= end_pos)

  995.                 break;

  996.         }

  997.         last_pos = pos;

  998. 

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

  1000.         ff_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);

  1001.         g->sb_hybrid[s_index + 0] =

  1002.         g->sb_hybrid[s_index + 1] =

  1003.         g->sb_hybrid[s_index + 2] =

  1004.         g->sb_hybrid[s_index + 3] = 0;

  1005.         while (code) {

  1006.             static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };

  1007.             int v;

  1008.             int pos = s_index + idxtab[code];

  1009.             code   ^= 8 >> idxtab[code];

  1010.             READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])

  1011.         }

  1012.         s_index += 4;

  1013.     }

  1014.     /* skip extension bits */

  1015.     bits_left = end_pos2 - get_bits_count(&s->gb);

  1016.     if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) {

  1017.         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1018.         s_index=0;

  1019.     } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) {

  1020.         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);

  1021.         s_index = 0;

  1022.     }

  1023.     memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));

  1024.     skip_bits_long(&s->gb, bits_left);

  1025. 

  1026.     i = get_bits_count(&s->gb);

  1027.     switch_buffer(s, &i, &end_pos, &end_pos2);

  1028. 

  1029.     return 0;

  1030. }

  1031. 

  1032. /* Reorder short blocks from bitstream order to interleaved order. It

  1033.    would be faster to do it in parsing, but the code would be far more

  1034.    complicated */

  1035. static void reorder_block(MPADecodeContext *s, GranuleDef *g)

  1036. {

  1037.     int i, j, len;

  1038.     INTFLOAT *ptr, *dst, *ptr1;

  1039.     INTFLOAT tmp[576];

  1040. 

  1041.     if (g->block_type != 2)

  1042.         return;

  1043. 

  1044.     if (g->switch_point) {

  1045.         if (s->sample_rate_index != 8)

  1046.             ptr = g->sb_hybrid + 36;

  1047.         else

  1048.             ptr = g->sb_hybrid + 72;

  1049.     } else {

  1050.         ptr = g->sb_hybrid;

  1051.     }

  1052. 

  1053.     for (i = g->short_start; i < 13; i++) {

  1054.         len  = band_size_short[s->sample_rate_index][i];

  1055.         ptr1 = ptr;

  1056.         dst  = tmp;

  1057.         for (j = len; j > 0; j--) {

  1058.             *dst++ = ptr[0*len];

  1059.             *dst++ = ptr[1*len];

  1060.             *dst++ = ptr[2*len];

  1061.             ptr++;

  1062.         }

  1063.         ptr += 2 * len;

  1064.         memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));

  1065.     }

  1066. }

  1067. 

  1068. #define ISQRT2 FIXR(0.70710678118654752440)

  1069. 

  1070. static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)

  1071. {

  1072.     int i, j, k, l;

  1073.     int sf_max, sf, len, non_zero_found;

  1074.     INTFLOAT (*is_tab)[16], *tab0, *tab1, v1, v2;

  1075.     SUINTFLOAT tmp0, tmp1;

  1076.     int non_zero_found_short[3];

  1077. 

  1078.     /* intensity stereo */

  1079.     if (s->mode_ext & MODE_EXT_I_STEREO) {

  1080.         if (!s->lsf) {

  1081.             is_tab = is_table;

  1082.             sf_max = 7;

  1083.         } else {

  1084.             is_tab = is_table_lsf[g1->scalefac_compress & 1];

  1085.             sf_max = 16;

  1086.         }

  1087. 

  1088.         tab0 = g0->sb_hybrid + 576;

  1089.         tab1 = g1->sb_hybrid + 576;

  1090. 

  1091.         non_zero_found_short[0] = 0;

  1092.         non_zero_found_short[1] = 0;

  1093.         non_zero_found_short[2] = 0;

  1094.         k = (13 - g1->short_start) * 3 + g1->long_end - 3;

  1095.         for (i = 12; i >= g1->short_start; i--) {

  1096.             /* for last band, use previous scale factor */

  1097.             if (i != 11)

  1098.                 k -= 3;

  1099.             len = band_size_short[s->sample_rate_index][i];

  1100.             for (l = 2; l >= 0; l--) {

  1101.                 tab0 -= len;

  1102.                 tab1 -= len;

  1103.                 if (!non_zero_found_short[l]) {

  1104.                     /* test if non zero band. if so, stop doing i-stereo */

  1105.                     for (j = 0; j < len; j++) {

  1106.                         if (tab1[j] != 0) {

  1107.                             non_zero_found_short[l] = 1;

  1108.                             goto found1;

  1109.                         }

  1110.                     }

  1111.                     sf = g1->scale_factors[k + l];

  1112.                     if (sf >= sf_max)

  1113.                         goto found1;

  1114. 

  1115.                     v1 = is_tab[0][sf];

  1116.                     v2 = is_tab[1][sf];

  1117.                     for (j = 0; j < len; j++) {

  1118.                         tmp0    = tab0[j];

  1119.                         tab0[j] = MULLx(tmp0, v1, FRAC_BITS);

  1120.                         tab1[j] = MULLx(tmp0, v2, FRAC_BITS);

  1121.                     }

  1122.                 } else {

  1123. found1:

  1124.                     if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1125.                         /* lower part of the spectrum : do ms stereo

  1126.                            if enabled */

  1127.                         for (j = 0; j < len; j++) {

  1128.                             tmp0    = tab0[j];

  1129.                             tmp1    = tab1[j];

  1130.                             tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1131.                             tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1132.                         }

  1133.                     }

  1134.                 }

  1135.             }

  1136.         }

  1137. 

  1138.         non_zero_found = non_zero_found_short[0] |

  1139.                          non_zero_found_short[1] |

  1140.                          non_zero_found_short[2];

  1141. 

  1142.         for (i = g1->long_end - 1;i >= 0;i--) {

  1143.             len   = band_size_long[s->sample_rate_index][i];

  1144.             tab0 -= len;

  1145.             tab1 -= len;

  1146.             /* test if non zero band. if so, stop doing i-stereo */

  1147.             if (!non_zero_found) {

  1148.                 for (j = 0; j < len; j++) {

  1149.                     if (tab1[j] != 0) {

  1150.                         non_zero_found = 1;

  1151.                         goto found2;

  1152.                     }

  1153.                 }

  1154.                 /* for last band, use previous scale factor */

  1155.                 k  = (i == 21) ? 20 : i;

  1156.                 sf = g1->scale_factors[k];

  1157.                 if (sf >= sf_max)

  1158.                     goto found2;

  1159.                 v1 = is_tab[0][sf];

  1160.                 v2 = is_tab[1][sf];

  1161.                 for (j = 0; j < len; j++) {

  1162.                     tmp0    = tab0[j];

  1163.                     tab0[j] = MULLx(tmp0, v1, FRAC_BITS);

  1164.                     tab1[j] = MULLx(tmp0, v2, FRAC_BITS);

  1165.                 }

  1166.             } else {

  1167. found2:

  1168.                 if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1169.                     /* lower part of the spectrum : do ms stereo

  1170.                        if enabled */

  1171.                     for (j = 0; j < len; j++) {

  1172.                         tmp0    = tab0[j];

  1173.                         tmp1    = tab1[j];

  1174.                         tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);

  1175.                         tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);

  1176.                     }

  1177.                 }

  1178.             }

  1179.         }

  1180.     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {

  1181.         /* ms stereo ONLY */

  1182.         /* NOTE: the 1/sqrt(2) normalization factor is included in the

  1183.            global gain */

  1184. #if USE_FLOATS

  1185.        s->butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);

  1186. #else

  1187.         tab0 = g0->sb_hybrid;

  1188.         tab1 = g1->sb_hybrid;

  1189.         for (i = 0; i < 576; i++) {

  1190.             tmp0    = tab0[i];

  1191.             tmp1    = tab1[i];

  1192.             tab0[i] = tmp0 + tmp1;

  1193.             tab1[i] = tmp0 - tmp1;

  1194.         }

  1195. #endif

  1196.     }

  1197. }

  1198. 

  1199. #if USE_FLOATS

  1200. #if HAVE_MIPSFPU

  1201. #   include "mips/compute_antialias_float.h"

  1202. #endif /* HAVE_MIPSFPU */

  1203. #else

  1204. #if HAVE_MIPSDSP

  1205. #   include "mips/compute_antialias_fixed.h"

  1206. #endif /* HAVE_MIPSDSP */

  1207. #endif /* USE_FLOATS */

  1208. 

  1209. #ifndef compute_antialias

  1210. #if USE_FLOATS

  1211. #define AA(j) do {                                                      \

  1212.         float tmp0 = ptr[-1-j];                                         \

  1213.         float tmp1 = ptr[   j];                                         \

  1214.         ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1];    \

  1215.         ptr[   j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0];    \

  1216.     } while (0)

  1217. #else

  1218. #define AA(j) do {                                              \

  1219.         SUINT tmp0 = ptr[-1-j];                                   \

  1220.         SUINT tmp1 = ptr[   j];                                   \

  1221.         SUINT tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]);          \

  1222.         ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2]));   \

  1223.         ptr[   j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3]));   \

  1224.     } while (0)

  1225. #endif

  1226. 

  1227. static void compute_antialias(MPADecodeContext *s, GranuleDef *g)

  1228. {

  1229.     INTFLOAT *ptr;

  1230.     int n, i;

  1231. 

  1232.     /* we antialias only "long" bands */

  1233.     if (g->block_type == 2) {

  1234.         if (!g->switch_point)

  1235.             return;

  1236.         /* XXX: check this for 8000Hz case */

  1237.         n = 1;

  1238.     } else {

  1239.         n = SBLIMIT - 1;

  1240.     }

  1241. 

  1242.     ptr = g->sb_hybrid + 18;

  1243.     for (i = n; i > 0; i--) {

  1244.         AA(0);

  1245.         AA(1);

  1246.         AA(2);

  1247.         AA(3);

  1248.         AA(4);

  1249.         AA(5);

  1250.         AA(6);

  1251.         AA(7);

  1252. 

  1253.         ptr += 18;

  1254.     }

  1255. }

  1256. #endif /* compute_antialias */

  1257. 

  1258. static void compute_imdct(MPADecodeContext *s, GranuleDef *g,

  1259.                           INTFLOAT *sb_samples, INTFLOAT *mdct_buf)

  1260. {

  1261.     INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;

  1262.     INTFLOAT out2[12];

  1263.     int i, j, mdct_long_end, sblimit;

  1264. 

  1265.     /* find last non zero block */

  1266.     ptr  = g->sb_hybrid + 576;

  1267.     ptr1 = g->sb_hybrid + 2 * 18;

  1268.     while (ptr >= ptr1) {

  1269.         int32_t *p;

  1270.         ptr -= 6;

  1271.         p    = (int32_t*)ptr;

  1272.         if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])

  1273.             break;

  1274.     }

  1275.     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;

  1276. 

  1277.     if (g->block_type == 2) {

  1278.         /* XXX: check for 8000 Hz */

  1279.         if (g->switch_point)

  1280.             mdct_long_end = 2;

  1281.         else

  1282.             mdct_long_end = 0;

  1283.     } else {

  1284.         mdct_long_end = sblimit;

  1285.     }

  1286. 

  1287.     s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,

  1288.                                      mdct_long_end, g->switch_point,

  1289.                                      g->block_type);

  1290. 

  1291.     buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);

  1292.     ptr = g->sb_hybrid + 18 * mdct_long_end;

  1293. 

  1294.     for (j = mdct_long_end; j < sblimit; j++) {

  1295.         /* select frequency inversion */

  1296.         win     = RENAME(ff_mdct_win)[2 + (4  & -(j & 1))];

  1297.         out_ptr = sb_samples + j;

  1298. 

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

  1300.             *out_ptr = buf[4*i];

  1301.             out_ptr += SBLIMIT;

  1302.         }

  1303.         imdct12(out2, ptr + 0);

  1304.         for (i = 0; i < 6; i++) {

  1305.             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*1)];

  1306.             buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);

  1307.             out_ptr += SBLIMIT;

  1308.         }

  1309.         imdct12(out2, ptr + 1);

  1310.         for (i = 0; i < 6; i++) {

  1311.             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*2)];

  1312.             buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);

  1313.             out_ptr += SBLIMIT;

  1314.         }

  1315.         imdct12(out2, ptr + 2);

  1316.         for (i = 0; i < 6; i++) {

  1317.             buf[4*(i + 6*0)] = MULH3(out2[i    ], win[i    ], 1) + buf[4*(i + 6*0)];

  1318.             buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);

  1319.             buf[4*(i + 6*2)] = 0;

  1320.         }

  1321.         ptr += 18;

  1322.         buf += (j&3) != 3 ? 1 : (4*18-3);

  1323.     }

  1324.     /* zero bands */

  1325.     for (j = sblimit; j < SBLIMIT; j++) {

  1326.         /* overlap */

  1327.         out_ptr = sb_samples + j;

  1328.         for (i = 0; i < 18; i++) {

  1329.             *out_ptr = buf[4*i];

  1330.             buf[4*i]   = 0;

  1331.             out_ptr += SBLIMIT;

  1332.         }

  1333.         buf += (j&3) != 3 ? 1 : (4*18-3);

  1334.     }

  1335. }

  1336. 

  1337. /* main layer3 decoding function */

  1338. static int mp_decode_layer3(MPADecodeContext *s)

  1339. {

  1340.     int nb_granules, main_data_begin;

  1341.     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;

  1342.     GranuleDef *g;

  1343.     int16_t exponents[576]; //FIXME try INTFLOAT

  1344.     int ret;

  1345. 

  1346.     /* read side info */

  1347.     if (s->lsf) {

  1348.         ret = handle_crc(s, ((s->nb_channels == 1) ? 8*9  : 8*17));

  1349.         main_data_begin = get_bits(&s->gb, 8);

  1350.         skip_bits(&s->gb, s->nb_channels);

  1351.         nb_granules = 1;

  1352.     } else {

  1353.         ret = handle_crc(s, ((s->nb_channels == 1) ? 8*17 : 8*32));

  1354.         main_data_begin = get_bits(&s->gb, 9);

  1355.         if (s->nb_channels == 2)

  1356.             skip_bits(&s->gb, 3);

  1357.         else

  1358.             skip_bits(&s->gb, 5);

  1359.         nb_granules = 2;

  1360.         for (ch = 0; ch < s->nb_channels; ch++) {

  1361.             s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */

  1362.             s->granules[ch][1].scfsi = get_bits(&s->gb, 4);

  1363.         }

  1364.     }

  1365.     if (ret < 0)

  1366.         return ret;

  1367. 

  1368.     for (gr = 0; gr < nb_granules; gr++) {

  1369.         for (ch = 0; ch < s->nb_channels; ch++) {

  1370.             ff_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);

  1371.             g = &s->granules[ch][gr];

  1372.             g->part2_3_length = get_bits(&s->gb, 12);

  1373.             g->big_values     = get_bits(&s->gb,  9);

  1374.             if (g->big_values > 288) {

  1375.                 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");

  1376.                 return AVERROR_INVALIDDATA;

  1377.             }

  1378. 

  1379.             g->global_gain = get_bits(&s->gb, 8);

  1380.             /* if MS stereo only is selected, we precompute the

  1381.                1/sqrt(2) renormalization factor */

  1382.             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==

  1383.                 MODE_EXT_MS_STEREO)

  1384.                 g->global_gain -= 2;

  1385.             if (s->lsf)

  1386.                 g->scalefac_compress = get_bits(&s->gb, 9);

  1387.             else

  1388.                 g->scalefac_compress = get_bits(&s->gb, 4);

  1389.             blocksplit_flag = get_bits1(&s->gb);

  1390.             if (blocksplit_flag) {

  1391.                 g->block_type = get_bits(&s->gb, 2);

  1392.                 if (g->block_type == 0) {

  1393.                     av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");

  1394.                     return AVERROR_INVALIDDATA;

  1395.                 }

  1396.                 g->switch_point = get_bits1(&s->gb);

  1397.                 for (i = 0; i < 2; i++)

  1398.                     g->table_select[i] = get_bits(&s->gb, 5);

  1399.                 for (i = 0; i < 3; i++)

  1400.                     g->subblock_gain[i] = get_bits(&s->gb, 3);

  1401.                 init_short_region(s, g);

  1402.             } else {

  1403.                 int region_address1, region_address2;

  1404.                 g->block_type = 0;

  1405.                 g->switch_point = 0;

  1406.                 for (i = 0; i < 3; i++)

  1407.                     g->table_select[i] = get_bits(&s->gb, 5);

  1408.                 /* compute huffman coded region sizes */

  1409.                 region_address1 = get_bits(&s->gb, 4);

  1410.                 region_address2 = get_bits(&s->gb, 3);

  1411.                 ff_dlog(s->avctx, "region1=%d region2=%d\n",

  1412.                         region_address1, region_address2);

  1413.                 init_long_region(s, g, region_address1, region_address2);

  1414.             }

  1415.             region_offset2size(g);

  1416.             compute_band_indexes(s, g);

  1417. 

  1418.             g->preflag = 0;

  1419.             if (!s->lsf)

  1420.                 g->preflag = get_bits1(&s->gb);

  1421.             g->scalefac_scale     = get_bits1(&s->gb);

  1422.             g->count1table_select = get_bits1(&s->gb);

  1423.             ff_dlog(s->avctx, "block_type=%d switch_point=%d\n",

  1424.                     g->block_type, g->switch_point);

  1425.         }

  1426.     }

  1427. 

  1428.     if (!s->adu_mode) {

  1429.         int skip;

  1430.         const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb) >> 3);

  1431.         s->extrasize = av_clip((get_bits_left(&s->gb) >> 3) - s->extrasize, 0,

  1432.                                FFMAX(0, LAST_BUF_SIZE - s->last_buf_size));

  1433.         av_assert1((get_bits_count(&s->gb) & 7) == 0);

  1434.         /* now we get bits from the main_data_begin offset */

  1435.         ff_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",

  1436.                 main_data_begin, s->last_buf_size);

  1437. 

  1438.         memcpy(s->last_buf + s->last_buf_size, ptr, s->extrasize);

  1439.         s->in_gb = s->gb;

  1440.         init_get_bits(&s->gb, s->last_buf, (s->last_buf_size + s->extrasize) * 8);

  1441.         s->last_buf_size <<= 3;

  1442.         for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {

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

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

  1445.                 s->last_buf_size += g->part2_3_length;

  1446.                 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));

  1447.                 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);

  1448.             }

  1449.         }

  1450.         skip = s->last_buf_size - 8 * main_data_begin;

  1451.         if (skip >= s->gb.size_in_bits - s->extrasize * 8 && s->in_gb.buffer) {

  1452.             skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits + s->extrasize * 8);

  1453.             s->gb           = s->in_gb;

  1454.             s->in_gb.buffer = NULL;

  1455.             s->extrasize    = 0;

  1456.         } else {

  1457.             skip_bits_long(&s->gb, skip);

  1458.         }

  1459.     } else {

  1460.         gr = 0;

  1461.         s->extrasize = 0;

  1462.     }

  1463. 

  1464.     for (; gr < nb_granules; gr++) {

  1465.         for (ch = 0; ch < s->nb_channels; ch++) {

  1466.             g = &s->granules[ch][gr];

  1467.             bits_pos = get_bits_count(&s->gb);

  1468. 

  1469.             if (!s->lsf) {

  1470.                 uint8_t *sc;

  1471.                 int slen, slen1, slen2;

  1472. 

  1473.                 /* MPEG-1 scale factors */

  1474.                 slen1 = slen_table[0][g->scalefac_compress];

  1475.                 slen2 = slen_table[1][g->scalefac_compress];

  1476.                 ff_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);

  1477.                 if (g->block_type == 2) {

  1478.                     n = g->switch_point ? 17 : 18;

  1479.                     j = 0;

  1480.                     if (slen1) {

  1481.                         for (i = 0; i < n; i++)

  1482.                             g->scale_factors[j++] = get_bits(&s->gb, slen1);

  1483.                     } else {

  1484.                         for (i = 0; i < n; i++)

  1485.                             g->scale_factors[j++] = 0;

  1486.                     }

  1487.                     if (slen2) {

  1488.                         for (i = 0; i < 18; i++)

  1489.                             g->scale_factors[j++] = get_bits(&s->gb, slen2);

  1490.                         for (i = 0; i < 3; i++)

  1491.                             g->scale_factors[j++] = 0;

  1492.                     } else {

  1493.                         for (i = 0; i < 21; i++)

  1494.                             g->scale_factors[j++] = 0;

  1495.                     }

  1496.                 } else {

  1497.                     sc = s->granules[ch][0].scale_factors;

  1498.                     j = 0;

  1499.                     for (k = 0; k < 4; k++) {

  1500.                         n = k == 0 ? 6 : 5;

  1501.                         if ((g->scfsi & (0x8 >> k)) == 0) {

  1502.                             slen = (k < 2) ? slen1 : slen2;

  1503.                             if (slen) {

  1504.                                 for (i = 0; i < n; i++)

  1505.                                     g->scale_factors[j++] = get_bits(&s->gb, slen);

  1506.                             } else {

  1507.                                 for (i = 0; i < n; i++)

  1508.                                     g->scale_factors[j++] = 0;

  1509.                             }

  1510.                         } else {

  1511.                             /* simply copy from last granule */

  1512.                             for (i = 0; i < n; i++) {

  1513.                                 g->scale_factors[j] = sc[j];

  1514.                                 j++;

  1515.                             }

  1516.                         }

  1517.                     }

  1518.                     g->scale_factors[j++] = 0;

  1519.                 }

  1520.             } else {

  1521.                 int tindex, tindex2, slen[4], sl, sf;

  1522. 

  1523.                 /* LSF scale factors */

  1524.                 if (g->block_type == 2)

  1525.                     tindex = g->switch_point ? 2 : 1;

  1526.                 else

  1527.                     tindex = 0;

  1528. 

  1529.                 sf = g->scalefac_compress;

  1530.                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {

  1531.                     /* intensity stereo case */

  1532.                     sf >>= 1;

  1533.                     if (sf < 180) {

  1534.                         lsf_sf_expand(slen, sf, 6, 6, 0);

  1535.                         tindex2 = 3;

  1536.                     } else if (sf < 244) {

  1537.                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);

  1538.                         tindex2 = 4;

  1539.                     } else {

  1540.                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);

  1541.                         tindex2 = 5;

  1542.                     }

  1543.                 } else {

  1544.                     /* normal case */

  1545.                     if (sf < 400) {

  1546.                         lsf_sf_expand(slen, sf, 5, 4, 4);

  1547.                         tindex2 = 0;

  1548.                     } else if (sf < 500) {

  1549.                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);

  1550.                         tindex2 = 1;

  1551.                     } else {

  1552.                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);

  1553.                         tindex2 = 2;

  1554.                         g->preflag = 1;

  1555.                     }

  1556.                 }

  1557. 

  1558.                 j = 0;

  1559.                 for (k = 0; k < 4; k++) {

  1560.                     n  = lsf_nsf_table[tindex2][tindex][k];

  1561.                     sl = slen[k];

  1562.                     if (sl) {

  1563.                         for (i = 0; i < n; i++)

  1564.                             g->scale_factors[j++] = get_bits(&s->gb, sl);

  1565.                     } else {

  1566.                         for (i = 0; i < n; i++)

  1567.                             g->scale_factors[j++] = 0;

  1568.                     }

  1569.                 }

  1570.                 /* XXX: should compute exact size */

  1571.                 for (; j < 40; j++)

  1572.                     g->scale_factors[j] = 0;

  1573.             }

  1574. 

  1575.             exponents_from_scale_factors(s, g, exponents);

  1576. 

  1577.             /* read Huffman coded residue */

  1578.             huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);

  1579.         } /* ch */

  1580. 

  1581.         if (s->mode == MPA_JSTEREO)

  1582.             compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);

  1583. 

  1584.         for (ch = 0; ch < s->nb_channels; ch++) {

  1585.             g = &s->granules[ch][gr];

  1586. 

  1587.             reorder_block(s, g);

  1588.             compute_antialias(s, g);

  1589.             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);

  1590.         }

  1591.     } /* gr */

  1592.     if (get_bits_count(&s->gb) < 0)

  1593.         skip_bits_long(&s->gb, -get_bits_count(&s->gb));

  1594.     return nb_granules * 18;

  1595. }

  1596. 

  1597. static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,

  1598.                            const uint8_t *buf, int buf_size)

  1599. {

  1600.     int i, nb_frames, ch, ret;

  1601.     OUT_INT *samples_ptr;

  1602. 

  1603.     init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);

  1604.     if (s->error_protection)

  1605.         s->crc = get_bits(&s->gb, 16);

  1606. 

  1607.     switch(s->layer) {

  1608.     case 1:

  1609.         s->avctx->frame_size = 384;

  1610.         nb_frames = mp_decode_layer1(s);

  1611.         break;

  1612.     case 2:

  1613.         s->avctx->frame_size = 1152;

  1614.         nb_frames = mp_decode_layer2(s);

  1615.         break;

  1616.     case 3:

  1617.         s->avctx->frame_size = s->lsf ? 576 : 1152;

  1618.     default:

  1619.         nb_frames = mp_decode_layer3(s);

  1620. 

  1621.         s->last_buf_size=0;

  1622.         if (s->in_gb.buffer) {

  1623.             align_get_bits(&s->gb);

  1624.             i = (get_bits_left(&s->gb) >> 3) - s->extrasize;

  1625.             if (i >= 0 && i <= BACKSTEP_SIZE) {

  1626.                 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb) >> 3), i);

  1627.                 s->last_buf_size=i;

  1628.             } else

  1629.                 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);

  1630.             s->gb           = s->in_gb;

  1631.             s->in_gb.buffer = NULL;

  1632.             s->extrasize    = 0;

  1633.         }

  1634. 

  1635.         align_get_bits(&s->gb);

  1636.         av_assert1((get_bits_count(&s->gb) & 7) == 0);

  1637.         i = (get_bits_left(&s->gb) >> 3) - s->extrasize;

  1638.         if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {

  1639.             if (i < 0)

  1640.                 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);

  1641.             i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);

  1642.         }

  1643.         av_assert1(i <= buf_size - HEADER_SIZE && i >= 0);

  1644.         memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);

  1645.         s->last_buf_size += i;

  1646.     }

  1647. 

  1648.     if(nb_frames < 0)

  1649.         return nb_frames;

  1650. 

  1651.     /* get output buffer */

  1652.     if (!samples) {

  1653.         av_assert0(s->frame);

  1654.         s->frame->nb_samples = s->avctx->frame_size;

  1655.         if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0)

  1656.             return ret;

  1657.         samples = (OUT_INT **)s->frame->extended_data;

  1658.     }

  1659. 

  1660.     /* apply the synthesis filter */

  1661.     for (ch = 0; ch < s->nb_channels; ch++) {

  1662.         int sample_stride;

  1663.         if (s->avctx->sample_fmt == OUT_FMT_P) {

  1664.             samples_ptr   = samples[ch];

  1665.             sample_stride = 1;

  1666.         } else {

  1667.             samples_ptr   = samples[0] + ch;

  1668.             sample_stride = s->nb_channels;

  1669.         }

  1670.         for (i = 0; i < nb_frames; i++) {

  1671.             RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],

  1672.                                         &(s->synth_buf_offset[ch]),

  1673.                                         RENAME(ff_mpa_synth_window),

  1674.                                         &s->dither_state, samples_ptr,

  1675.                                         sample_stride, s->sb_samples[ch][i]);

  1676.             samples_ptr += 32 * sample_stride;

  1677.         }

  1678.     }

  1679. 

  1680.     return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;

  1681. }

  1682. 

  1683. static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,

  1684.                         AVPacket *avpkt)

  1685. {

  1686.     const uint8_t *buf  = avpkt->data;

  1687.     int buf_size        = avpkt->size;

  1688.     MPADecodeContext *s = avctx->priv_data;

  1689.     uint32_t header;

  1690.     int ret;

  1691. 

  1692.     int skipped = 0;

  1693.     while(buf_size && !*buf){

  1694.         buf++;

  1695.         buf_size--;

  1696.         skipped++;

  1697.     }

  1698. 

  1699.     if (buf_size < HEADER_SIZE)

  1700.         return AVERROR_INVALIDDATA;

  1701. 

  1702.     header = AV_RB32(buf);

  1703.     if (header >> 8 == AV_RB32("TAG") >> 8) {

  1704.         av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n");

  1705.         return buf_size + skipped;

  1706.     }

  1707.     ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);

  1708.     if (ret < 0) {

  1709.         av_log(avctx, AV_LOG_ERROR, "Header missing\n");

  1710.         return AVERROR_INVALIDDATA;

  1711.     } else if (ret == 1) {

  1712.         /* free format: prepare to compute frame size */

  1713.         s->frame_size = -1;

  1714.         return AVERROR_INVALIDDATA;

  1715.     }

  1716.     /* update codec info */

  1717.     avctx->channels       = s->nb_channels;

  1718.     avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;

  1719.     if (!avctx->bit_rate)

  1720.         avctx->bit_rate = s->bit_rate;

  1721. 

  1722.     if (s->frame_size <= 0) {

  1723.         av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");

  1724.         return AVERROR_INVALIDDATA;

  1725.     } else if (s->frame_size < buf_size) {

  1726.         av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");

  1727.         buf_size= s->frame_size;

  1728.     }

  1729. 

  1730.     s->frame = data;

  1731. 

  1732.     ret = mp_decode_frame(s, NULL, buf, buf_size);

  1733.     if (ret >= 0) {

  1734.         s->frame->nb_samples = avctx->frame_size;

  1735.         *got_frame_ptr       = 1;

  1736.         avctx->sample_rate   = s->sample_rate;

  1737.         //FIXME maybe move the other codec info stuff from above here too

  1738.     } else {

  1739.         av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");

  1740.         /* Only return an error if the bad frame makes up the whole packet or

  1741.          * the error is related to buffer management.

  1742.          * If there is more data in the packet, just consume the bad frame

  1743.          * instead of returning an error, which would discard the whole

  1744.          * packet. */

  1745.         *got_frame_ptr = 0;

  1746.         if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)

  1747.             return ret;

  1748.     }

  1749.     s->frame_size = 0;

  1750.     return buf_size + skipped;

  1751. }

  1752. 

  1753. static void mp_flush(MPADecodeContext *ctx)

  1754. {

  1755.     memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));

  1756.     memset(ctx->mdct_buf, 0, sizeof(ctx->mdct_buf));

  1757.     ctx->last_buf_size = 0;

  1758.     ctx->dither_state = 0;

  1759. }

  1760. 

  1761. static void flush(AVCodecContext *avctx)

  1762. {

  1763.     mp_flush(avctx->priv_data);

  1764. }

  1765. 

  1766. #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER

  1767. static int decode_frame_adu(AVCodecContext *avctx, void *data,

  1768.                             int *got_frame_ptr, AVPacket *avpkt)

  1769. {

  1770.     const uint8_t *buf  = avpkt->data;

  1771.     int buf_size        = avpkt->size;

  1772.     MPADecodeContext *s = avctx->priv_data;

  1773.     uint32_t header;

  1774.     int len, ret;

  1775.     int av_unused out_size;

  1776. 

  1777.     len = buf_size;

  1778. 

  1779.     // Discard too short frames

  1780.     if (buf_size < HEADER_SIZE) {

  1781.         av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");

  1782.         return AVERROR_INVALIDDATA;

  1783.     }

  1784. 

  1785. 

  1786.     if (len > MPA_MAX_CODED_FRAME_SIZE)

  1787.         len = MPA_MAX_CODED_FRAME_SIZE;

  1788. 

  1789.     // Get header and restore sync word

  1790.     header = AV_RB32(buf) | 0xffe00000;

  1791. 

  1792.     ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);

  1793.     if (ret < 0) {

  1794.         av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");

  1795.         return ret;

  1796.     }

  1797.     /* update codec info */

  1798.     avctx->sample_rate = s->sample_rate;

  1799.     avctx->channels    = s->nb_channels;

  1800.     avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;

  1801.     if (!avctx->bit_rate)

  1802.         avctx->bit_rate = s->bit_rate;

  1803. 

  1804.     s->frame_size = len;

  1805. 

  1806.     s->frame = data;

  1807. 

  1808.     ret = mp_decode_frame(s, NULL, buf, buf_size);

  1809.     if (ret < 0) {

  1810.         av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");

  1811.         return ret;

  1812.     }

  1813. 

  1814.     *got_frame_ptr = 1;

  1815. 

  1816.     return buf_size;

  1817. }

  1818. #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */

  1819. 

  1820. #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER

  1821. 

  1822. /**

  1823.  * Context for MP3On4 decoder

  1824.  */

  1825. typedef struct MP3On4DecodeContext {

  1826.     int frames;                     ///< number of mp3 frames per block (number of mp3 decoder instances)

  1827.     int syncword;                   ///< syncword patch

  1828.     const uint8_t *coff;            ///< channel offsets in output buffer

  1829.     MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance

  1830. } MP3On4DecodeContext;

  1831. 

  1832. #include "mpeg4audio.h"

  1833. 

  1834. /* Next 3 arrays are indexed by channel config number (passed via codecdata) */

  1835. 

  1836. /* number of mp3 decoder instances */

  1837. static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };

  1838. 

  1839. /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */

  1840. static const uint8_t chan_offset[8][5] = {

  1841.     { 0             },

  1842.     { 0             },  // C

  1843.     { 0             },  // FLR

  1844.     { 2, 0          },  // C FLR

  1845.     { 2, 0, 3       },  // C FLR BS

  1846.     { 2, 0, 3       },  // C FLR BLRS

  1847.     { 2, 0, 4, 3    },  // C FLR BLRS LFE

  1848.     { 2, 0, 6, 4, 3 },  // C FLR BLRS BLR LFE

  1849. };

  1850. 

  1851. /* mp3on4 channel layouts */

  1852. static const int16_t chan_layout[8] = {

  1853.     0,

  1854.     AV_CH_LAYOUT_MONO,

  1855.     AV_CH_LAYOUT_STEREO,

  1856.     AV_CH_LAYOUT_SURROUND,

  1857.     AV_CH_LAYOUT_4POINT0,

  1858.     AV_CH_LAYOUT_5POINT0,

  1859.     AV_CH_LAYOUT_5POINT1,

  1860.     AV_CH_LAYOUT_7POINT1

  1861. };

  1862. 

  1863. static av_cold int decode_close_mp3on4(AVCodecContext * avctx)

  1864. {

  1865.     MP3On4DecodeContext *s = avctx->priv_data;

  1866.     int i;

  1867. 

  1868.     for (i = 0; i < s->frames; i++)

  1869.         av_freep(&s->mp3decctx[i]);

  1870. 

  1871.     return 0;

  1872. }

  1873. 

  1874. 

  1875. static av_cold int decode_init_mp3on4(AVCodecContext * avctx)

  1876. {

  1877.     MP3On4DecodeContext *s = avctx->priv_data;

  1878.     MPEG4AudioConfig cfg;

  1879.     int i, ret;

  1880. 

  1881.     if ((avctx->extradata_size < 2) || !avctx->extradata) {

  1882.         av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");

  1883.         return AVERROR_INVALIDDATA;

  1884.     }

  1885. 

  1886.     avpriv_mpeg4audio_get_config2(&cfg, avctx->extradata,

  1887.                                   avctx->extradata_size, 1, avctx);

  1888.     if (!cfg.chan_config || cfg.chan_config > 7) {

  1889.         av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");

  1890.         return AVERROR_INVALIDDATA;

  1891.     }

  1892.     s->frames             = mp3Frames[cfg.chan_config];

  1893.     s->coff               = chan_offset[cfg.chan_config];

  1894.     avctx->channels       = ff_mpeg4audio_channels[cfg.chan_config];

  1895.     avctx->channel_layout = chan_layout[cfg.chan_config];

  1896. 

  1897.     if (cfg.sample_rate < 16000)

  1898.         s->syncword = 0xffe00000;

  1899.     else

  1900.         s->syncword = 0xfff00000;

  1901. 

  1902.     /* Init the first mp3 decoder in standard way, so that all tables get builded

  1903.      * We replace avctx->priv_data with the context of the first decoder so that

  1904.      * decode_init() does not have to be changed.

  1905.      * Other decoders will be initialized here copying data from the first context

  1906.      */

  1907.     // Allocate zeroed memory for the first decoder context

  1908.     s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));

  1909.     if (!s->mp3decctx[0])

  1910.         return AVERROR(ENOMEM);

  1911.     // Put decoder context in place to make init_decode() happy

  1912.     avctx->priv_data = s->mp3decctx[0];

  1913.     ret = decode_init(avctx);

  1914.     // Restore mp3on4 context pointer

  1915.     avctx->priv_data = s;

  1916.     if (ret < 0)

  1917.         return ret;

  1918.     s->mp3decctx[0]->adu_mode = 1; // Set adu mode

  1919. 

  1920.     /* Create a separate codec/context for each frame (first is already ok).

  1921.      * Each frame is 1 or 2 channels - up to 5 frames allowed

  1922.      */

  1923.     for (i = 1; i < s->frames; i++) {

  1924.         s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));

  1925.         if (!s->mp3decctx[i])

  1926.             return AVERROR(ENOMEM);

  1927.         s->mp3decctx[i]->adu_mode = 1;

  1928.         s->mp3decctx[i]->avctx = avctx;

  1929.         s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;

  1930.         s->mp3decctx[i]->butterflies_float = s->mp3decctx[0]->butterflies_float;

  1931.     }

  1932. 

  1933.     return 0;

  1934. }

  1935. 

  1936. 

  1937. static void flush_mp3on4(AVCodecContext *avctx)

  1938. {

  1939.     int i;

  1940.     MP3On4DecodeContext *s = avctx->priv_data;

  1941. 

  1942.     for (i = 0; i < s->frames; i++)

  1943.         mp_flush(s->mp3decctx[i]);

  1944. }

  1945. 

  1946. 

  1947. static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,

  1948.                                int *got_frame_ptr, AVPacket *avpkt)

  1949. {

  1950.     AVFrame *frame         = data;

  1951.     const uint8_t *buf     = avpkt->data;

  1952.     int buf_size           = avpkt->size;

  1953.     MP3On4DecodeContext *s = avctx->priv_data;

  1954.     MPADecodeContext *m;

  1955.     int fsize, len = buf_size, out_size = 0;

  1956.     uint32_t header;

  1957.     OUT_INT **out_samples;

  1958.     OUT_INT *outptr[2];

  1959.     int fr, ch, ret;

  1960. 

  1961.     /* get output buffer */

  1962.     frame->nb_samples = MPA_FRAME_SIZE;

  1963.     if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)

  1964.         return ret;

  1965.     out_samples = (OUT_INT **)frame->extended_data;

  1966. 

  1967.     // Discard too short frames

  1968.     if (buf_size < HEADER_SIZE)

  1969.         return AVERROR_INVALIDDATA;

  1970. 

  1971.     avctx->bit_rate = 0;

  1972. 

  1973.     ch = 0;

  1974.     for (fr = 0; fr < s->frames; fr++) {

  1975.         fsize = AV_RB16(buf) >> 4;

  1976.         fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);

  1977.         m     = s->mp3decctx[fr];

  1978.         av_assert1(m);

  1979. 

  1980.         if (fsize < HEADER_SIZE) {

  1981.             av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");

  1982.             return AVERROR_INVALIDDATA;

  1983.         }

  1984.         header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header

  1985. 

  1986.         ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);

  1987.         if (ret < 0) {

  1988.             av_log(avctx, AV_LOG_ERROR, "Bad header, discard block\n");

  1989.             return AVERROR_INVALIDDATA;

  1990.         }

  1991. 

  1992.         if (ch + m->nb_channels > avctx->channels ||

  1993.             s->coff[fr] + m->nb_channels > avctx->channels) {

  1994.             av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "

  1995.                                         "channel count\n");

  1996.             return AVERROR_INVALIDDATA;

  1997.         }

  1998.         ch += m->nb_channels;

  1999. 

  2000.         outptr[0] = out_samples[s->coff[fr]];

  2001.         if (m->nb_channels > 1)

  2002.             outptr[1] = out_samples[s->coff[fr] + 1];

  2003. 

  2004.         if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0) {

  2005.             av_log(avctx, AV_LOG_ERROR, "failed to decode channel %d\n", ch);

  2006.             memset(outptr[0], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));

  2007.             if (m->nb_channels > 1)

  2008.                 memset(outptr[1], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));

  2009.             ret = m->nb_channels * MPA_FRAME_SIZE*sizeof(OUT_INT);

  2010.         }

  2011. 

  2012.         out_size += ret;

  2013.         buf      += fsize;

  2014.         len      -= fsize;

  2015. 

  2016.         avctx->bit_rate += m->bit_rate;

  2017.     }

  2018.     if (ch != avctx->channels) {

  2019.         av_log(avctx, AV_LOG_ERROR, "failed to decode all channels\n");

  2020.         return AVERROR_INVALIDDATA;

  2021.     }

  2022. 

  2023.     /* update codec info */

  2024.     avctx->sample_rate = s->mp3decctx[0]->sample_rate;

  2025. 

  2026.     frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));

  2027.     *got_frame_ptr    = 1;

  2028. 

  2029.     return buf_size;

  2030. }

  2031. #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */