FFmpeg: libavcodec/mpegaudiodec.c Source File

00001 /*
00002  * MPEG Audio decoder
00003  * Copyright (c) 2001, 2002 Fabrice Bellard
00004  *
00005  * This file is part of FFmpeg.
00006  *
00007  * FFmpeg is free software; you can redistribute it and/or
00008  * modify it under the terms of the GNU Lesser General Public
00009  * License as published by the Free Software Foundation; either
00010  * version 2.1 of the License, or (at your option) any later version.
00011  *
00012  * FFmpeg is distributed in the hope that it will be useful,
00013  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00014  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00015  * Lesser General Public License for more details.
00016  *
00017  * You should have received a copy of the GNU Lesser General Public
00018  * License along with FFmpeg; if not, write to the Free Software
00019  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00020  */
00021 
00027 #include "libavutil/audioconvert.h"
00028 #include "avcodec.h"
00029 #include "get_bits.h"
00030 #include "mathops.h"
00031 #include "mpegaudiodsp.h"
00032 
00033 /*
00034  * TODO:
00035  *  - test lsf / mpeg25 extensively.
00036  */
00037 
00038 #include "mpegaudio.h"
00039 #include "mpegaudiodecheader.h"
00040 
00041 #define BACKSTEP_SIZE 512
00042 #define EXTRABYTES 24
00043 
00044 /* layer 3 "granule" */
00045 typedef struct GranuleDef {
00046     uint8_t scfsi;
00047     int part2_3_length;
00048     int big_values;
00049     int global_gain;
00050     int scalefac_compress;
00051     uint8_t block_type;
00052     uint8_t switch_point;
00053     int table_select[3];
00054     int subblock_gain[3];
00055     uint8_t scalefac_scale;
00056     uint8_t count1table_select;
00057     int region_size[3]; /* number of huffman codes in each region */
00058     int preflag;
00059     int short_start, long_end; /* long/short band indexes */
00060     uint8_t scale_factors[40];
00061     INTFLOAT sb_hybrid[SBLIMIT * 18]; /* 576 samples */
00062 } GranuleDef;
00063 
00064 typedef struct MPADecodeContext {
00065     MPA_DECODE_HEADER
00066     uint8_t last_buf[2*BACKSTEP_SIZE + EXTRABYTES];
00067     int last_buf_size;
00068     /* next header (used in free format parsing) */
00069     uint32_t free_format_next_header;
00070     GetBitContext gb;
00071     GetBitContext in_gb;
00072     DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
00073     int synth_buf_offset[MPA_MAX_CHANNELS];
00074     DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
00075     INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
00076     GranuleDef granules[2][2]; /* Used in Layer 3 */
00077 #ifdef DEBUG
00078     int frame_count;
00079 #endif
00080     int adu_mode; 
00081     int dither_state;
00082     int error_recognition;
00083     AVCodecContext* avctx;
00084     MPADSPContext mpadsp;
00085 } MPADecodeContext;
00086 
00087 #if CONFIG_FLOAT
00088 #   define SHR(a,b)       ((a)*(1.0f/(1<<(b))))
00089 #   define FIXR_OLD(a)    ((int)((a) * FRAC_ONE + 0.5))
00090 #   define FIXR(x)        ((float)(x))
00091 #   define FIXHR(x)       ((float)(x))
00092 #   define MULH3(x, y, s) ((s)*(y)*(x))
00093 #   define MULLx(x, y, s) ((y)*(x))
00094 #   define RENAME(a) a ## _float
00095 #   define OUT_FMT AV_SAMPLE_FMT_FLT
00096 #else
00097 #   define SHR(a,b)       ((a)>>(b))
00098 /* WARNING: only correct for posititive numbers */
00099 #   define FIXR_OLD(a)    ((int)((a) * FRAC_ONE + 0.5))
00100 #   define FIXR(a)        ((int)((a) * FRAC_ONE + 0.5))
00101 #   define FIXHR(a)       ((int)((a) * (1LL<<32) + 0.5))
00102 #   define MULH3(x, y, s) MULH((s)*(x), y)
00103 #   define MULLx(x, y, s) MULL(x,y,s)
00104 #   define RENAME(a)      a ## _fixed
00105 #   define OUT_FMT AV_SAMPLE_FMT_S16
00106 #endif
00107 
00108 /****************/
00109 
00110 #define HEADER_SIZE 4
00111 
00112 #include "mpegaudiodata.h"
00113 #include "mpegaudiodectab.h"
00114 
00115 /* vlc structure for decoding layer 3 huffman tables */
00116 static VLC huff_vlc[16];
00117 static VLC_TYPE huff_vlc_tables[
00118   0+128+128+128+130+128+154+166+
00119   142+204+190+170+542+460+662+414
00120   ][2];
00121 static const int huff_vlc_tables_sizes[16] = {
00122   0, 128, 128, 128, 130, 128, 154, 166,
00123   142, 204, 190, 170, 542, 460, 662, 414
00124 };
00125 static VLC huff_quad_vlc[2];
00126 static VLC_TYPE huff_quad_vlc_tables[128+16][2];
00127 static const int huff_quad_vlc_tables_sizes[2] = {
00128   128, 16
00129 };
00130 /* computed from band_size_long */
00131 static uint16_t band_index_long[9][23];
00132 #include "mpegaudio_tablegen.h"
00133 /* intensity stereo coef table */
00134 static INTFLOAT is_table[2][16];
00135 static INTFLOAT is_table_lsf[2][2][16];
00136 static INTFLOAT csa_table[8][4];
00137 static INTFLOAT mdct_win[8][36];
00138 
00139 static int16_t division_tab3[1<<6 ];
00140 static int16_t division_tab5[1<<8 ];
00141 static int16_t division_tab9[1<<11];
00142 
00143 static int16_t * const division_tabs[4] = {
00144     division_tab3, division_tab5, NULL, division_tab9
00145 };
00146 
00147 /* lower 2 bits: modulo 3, higher bits: shift */
00148 static uint16_t scale_factor_modshift[64];
00149 /* [i][j]:  2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
00150 static int32_t scale_factor_mult[15][3];
00151 /* mult table for layer 2 group quantization */
00152 
00153 #define SCALE_GEN(v) \
00154 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
00155 
00156 static const int32_t scale_factor_mult2[3][3] = {
00157     SCALE_GEN(4.0 / 3.0), /* 3 steps */
00158     SCALE_GEN(4.0 / 5.0), /* 5 steps */
00159     SCALE_GEN(4.0 / 9.0), /* 9 steps */
00160 };
00161 
00166 static void ff_region_offset2size(GranuleDef *g){
00167     int i, k, j=0;
00168     g->region_size[2] = (576 / 2);
00169     for(i=0;i<3;i++) {
00170         k = FFMIN(g->region_size[i], g->big_values);
00171         g->region_size[i] = k - j;
00172         j = k;
00173     }
00174 }
00175 
00176 static void ff_init_short_region(MPADecodeContext *s, GranuleDef *g){
00177     if (g->block_type == 2)
00178         g->region_size[0] = (36 / 2);
00179     else {
00180         if (s->sample_rate_index <= 2)
00181             g->region_size[0] = (36 / 2);
00182         else if (s->sample_rate_index != 8)
00183             g->region_size[0] = (54 / 2);
00184         else
00185             g->region_size[0] = (108 / 2);
00186     }
00187     g->region_size[1] = (576 / 2);
00188 }
00189 
00190 static void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2){
00191     int l;
00192     g->region_size[0] =
00193         band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
00194     /* should not overflow */
00195     l = FFMIN(ra1 + ra2 + 2, 22);
00196     g->region_size[1] =
00197         band_index_long[s->sample_rate_index][l] >> 1;
00198 }
00199 
00200 static void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g){
00201     if (g->block_type == 2) {
00202         if (g->switch_point) {
00203             /* if switched mode, we handle the 36 first samples as
00204                 long blocks.  For 8000Hz, we handle the 48 first
00205                 exponents as long blocks (XXX: check this!) */
00206             if (s->sample_rate_index <= 2)
00207                 g->long_end = 8;
00208             else if (s->sample_rate_index != 8)
00209                 g->long_end = 6;
00210             else
00211                 g->long_end = 4; /* 8000 Hz */
00212 
00213             g->short_start = 2 + (s->sample_rate_index != 8);
00214         } else {
00215             g->long_end = 0;
00216             g->short_start = 0;
00217         }
00218     } else {
00219         g->short_start = 13;
00220         g->long_end = 22;
00221     }
00222 }
00223 
00224 /* layer 1 unscaling */
00225 /* n = number of bits of the mantissa minus 1 */
00226 static inline int l1_unscale(int n, int mant, int scale_factor)
00227 {
00228     int shift, mod;
00229     int64_t val;
00230 
00231     shift = scale_factor_modshift[scale_factor];
00232     mod = shift & 3;
00233     shift >>= 2;
00234     val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
00235     shift += n;
00236     /* NOTE: at this point, 1 <= shift >= 21 + 15 */
00237     return (int)((val + (1LL << (shift - 1))) >> shift);
00238 }
00239 
00240 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
00241 {
00242     int shift, mod, val;
00243 
00244     shift = scale_factor_modshift[scale_factor];
00245     mod = shift & 3;
00246     shift >>= 2;
00247 
00248     val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
00249     /* NOTE: at this point, 0 <= shift <= 21 */
00250     if (shift > 0)
00251         val = (val + (1 << (shift - 1))) >> shift;
00252     return val;
00253 }
00254 
00255 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
00256 static inline int l3_unscale(int value, int exponent)
00257 {
00258     unsigned int m;
00259     int e;
00260 
00261     e = table_4_3_exp  [4*value + (exponent&3)];
00262     m = table_4_3_value[4*value + (exponent&3)];
00263     e -= (exponent >> 2);
00264     assert(e>=1);
00265     if (e > 31)
00266         return 0;
00267     m = (m + (1 << (e-1))) >> e;
00268 
00269     return m;
00270 }
00271 
00272 static av_cold int decode_init(AVCodecContext * avctx)
00273 {
00274     MPADecodeContext *s = avctx->priv_data;
00275     static int init=0;
00276     int i, j, k;
00277 
00278     s->avctx = avctx;
00279 
00280     ff_mpadsp_init(&s->mpadsp);
00281 
00282     avctx->sample_fmt= OUT_FMT;
00283     s->error_recognition= avctx->error_recognition;
00284 
00285     if (!init && !avctx->parse_only) {
00286         int offset;
00287 
00288         /* scale factors table for layer 1/2 */
00289         for(i=0;i<64;i++) {
00290             int shift, mod;
00291             /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
00292             shift = (i / 3);
00293             mod = i % 3;
00294             scale_factor_modshift[i] = mod | (shift << 2);
00295         }
00296 
00297         /* scale factor multiply for layer 1 */
00298         for(i=0;i<15;i++) {
00299             int n, norm;
00300             n = i + 2;
00301             norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
00302             scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0          * 2.0), FRAC_BITS);
00303             scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
00304             scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
00305             av_dlog(avctx, "%d: norm=%x s=%x %x %x\n",
00306                     i, norm,
00307                     scale_factor_mult[i][0],
00308                     scale_factor_mult[i][1],
00309                     scale_factor_mult[i][2]);
00310         }
00311 
00312         RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
00313 
00314         /* huffman decode tables */
00315         offset = 0;
00316         for(i=1;i<16;i++) {
00317             const HuffTable *h = &mpa_huff_tables[i];
00318             int xsize, x, y;
00319             uint8_t  tmp_bits [512];
00320             uint16_t tmp_codes[512];
00321 
00322             memset(tmp_bits , 0, sizeof(tmp_bits ));
00323             memset(tmp_codes, 0, sizeof(tmp_codes));
00324 
00325             xsize = h->xsize;
00326 
00327             j = 0;
00328             for(x=0;x<xsize;x++) {
00329                 for(y=0;y<xsize;y++){
00330                     tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j  ];
00331                     tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
00332                 }
00333             }
00334 
00335             /* XXX: fail test */
00336             huff_vlc[i].table = huff_vlc_tables+offset;
00337             huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
00338             init_vlc(&huff_vlc[i], 7, 512,
00339                      tmp_bits, 1, 1, tmp_codes, 2, 2,
00340                      INIT_VLC_USE_NEW_STATIC);
00341             offset += huff_vlc_tables_sizes[i];
00342         }
00343         assert(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
00344 
00345         offset = 0;
00346         for(i=0;i<2;i++) {
00347             huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
00348             huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
00349             init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
00350                      mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
00351                      INIT_VLC_USE_NEW_STATIC);
00352             offset += huff_quad_vlc_tables_sizes[i];
00353         }
00354         assert(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
00355 
00356         for(i=0;i<9;i++) {
00357             k = 0;
00358             for(j=0;j<22;j++) {
00359                 band_index_long[i][j] = k;
00360                 k += band_size_long[i][j];
00361             }
00362             band_index_long[i][22] = k;
00363         }
00364 
00365         /* compute n ^ (4/3) and store it in mantissa/exp format */
00366 
00367         mpegaudio_tableinit();
00368 
00369         for (i = 0; i < 4; i++)
00370             if (ff_mpa_quant_bits[i] < 0)
00371                 for (j = 0; j < (1<<(-ff_mpa_quant_bits[i]+1)); j++) {
00372                     int val1, val2, val3, steps;
00373                     int val = j;
00374                     steps  = ff_mpa_quant_steps[i];
00375                     val1 = val % steps;
00376                     val /= steps;
00377                     val2 = val % steps;
00378                     val3 = val / steps;
00379                     division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
00380                 }
00381 
00382 
00383         for(i=0;i<7;i++) {
00384             float f;
00385             INTFLOAT v;
00386             if (i != 6) {
00387                 f = tan((double)i * M_PI / 12.0);
00388                 v = FIXR(f / (1.0 + f));
00389             } else {
00390                 v = FIXR(1.0);
00391             }
00392             is_table[0][i] = v;
00393             is_table[1][6 - i] = v;
00394         }
00395         /* invalid values */
00396         for(i=7;i<16;i++)
00397             is_table[0][i] = is_table[1][i] = 0.0;
00398 
00399         for(i=0;i<16;i++) {
00400             double f;
00401             int e, k;
00402 
00403             for(j=0;j<2;j++) {
00404                 e = -(j + 1) * ((i + 1) >> 1);
00405                 f = pow(2.0, e / 4.0);
00406                 k = i & 1;
00407                 is_table_lsf[j][k ^ 1][i] = FIXR(f);
00408                 is_table_lsf[j][k][i] = FIXR(1.0);
00409                 av_dlog(avctx, "is_table_lsf %d %d: %f %f\n",
00410                         i, j, (float) is_table_lsf[j][0][i],
00411                         (float) is_table_lsf[j][1][i]);
00412             }
00413         }
00414 
00415         for(i=0;i<8;i++) {
00416             float ci, cs, ca;
00417             ci = ci_table[i];
00418             cs = 1.0 / sqrt(1.0 + ci * ci);
00419             ca = cs * ci;
00420 #if !CONFIG_FLOAT
00421             csa_table[i][0] = FIXHR(cs/4);
00422             csa_table[i][1] = FIXHR(ca/4);
00423             csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
00424             csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
00425 #else
00426             csa_table[i][0] = cs;
00427             csa_table[i][1] = ca;
00428             csa_table[i][2] = ca + cs;
00429             csa_table[i][3] = ca - cs;
00430 #endif
00431         }
00432 
00433         /* compute mdct windows */
00434         for(i=0;i<36;i++) {
00435             for(j=0; j<4; j++){
00436                 double d;
00437 
00438                 if(j==2 && i%3 != 1)
00439                     continue;
00440 
00441                 d= sin(M_PI * (i + 0.5) / 36.0);
00442                 if(j==1){
00443                     if     (i>=30) d= 0;
00444                     else if(i>=24) d= sin(M_PI * (i - 18 + 0.5) / 12.0);
00445                     else if(i>=18) d= 1;
00446                 }else if(j==3){
00447                     if     (i<  6) d= 0;
00448                     else if(i< 12) d= sin(M_PI * (i -  6 + 0.5) / 12.0);
00449                     else if(i< 18) d= 1;
00450                 }
00451                 //merge last stage of imdct into the window coefficients
00452                 d*= 0.5 / cos(M_PI*(2*i + 19)/72);
00453 
00454                 if(j==2)
00455                     mdct_win[j][i/3] = FIXHR((d / (1<<5)));
00456                 else
00457                     mdct_win[j][i  ] = FIXHR((d / (1<<5)));
00458             }
00459         }
00460 
00461         /* NOTE: we do frequency inversion adter the MDCT by changing
00462            the sign of the right window coefs */
00463         for(j=0;j<4;j++) {
00464             for(i=0;i<36;i+=2) {
00465                 mdct_win[j + 4][i] = mdct_win[j][i];
00466                 mdct_win[j + 4][i + 1] = -mdct_win[j][i + 1];
00467             }
00468         }
00469 
00470         init = 1;
00471     }
00472 
00473     if (avctx->codec_id == CODEC_ID_MP3ADU)
00474         s->adu_mode = 1;
00475     return 0;
00476 }
00477 
00478 #define C3 FIXHR(0.86602540378443864676/2)
00479 
00480 /* 0.5 / cos(pi*(2*i+1)/36) */
00481 static const INTFLOAT icos36[9] = {
00482     FIXR(0.50190991877167369479),
00483     FIXR(0.51763809020504152469), //0
00484     FIXR(0.55168895948124587824),
00485     FIXR(0.61038729438072803416),
00486     FIXR(0.70710678118654752439), //1
00487     FIXR(0.87172339781054900991),
00488     FIXR(1.18310079157624925896),
00489     FIXR(1.93185165257813657349), //2
00490     FIXR(5.73685662283492756461),
00491 };
00492 
00493 /* 0.5 / cos(pi*(2*i+1)/36) */
00494 static const INTFLOAT icos36h[9] = {
00495     FIXHR(0.50190991877167369479/2),
00496     FIXHR(0.51763809020504152469/2), //0
00497     FIXHR(0.55168895948124587824/2),
00498     FIXHR(0.61038729438072803416/2),
00499     FIXHR(0.70710678118654752439/2), //1
00500     FIXHR(0.87172339781054900991/2),
00501     FIXHR(1.18310079157624925896/4),
00502     FIXHR(1.93185165257813657349/4), //2
00503 //    FIXHR(5.73685662283492756461),
00504 };
00505 
00506 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
00507    cases. */
00508 static void imdct12(INTFLOAT *out, INTFLOAT *in)
00509 {
00510     INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
00511 
00512     in0= in[0*3];
00513     in1= in[1*3] + in[0*3];
00514     in2= in[2*3] + in[1*3];
00515     in3= in[3*3] + in[2*3];
00516     in4= in[4*3] + in[3*3];
00517     in5= in[5*3] + in[4*3];
00518     in5 += in3;
00519     in3 += in1;
00520 
00521     in2= MULH3(in2, C3, 2);
00522     in3= MULH3(in3, C3, 4);
00523 
00524     t1 = in0 - in4;
00525     t2 = MULH3(in1 - in5, icos36h[4], 2);
00526 
00527     out[ 7]=
00528     out[10]= t1 + t2;
00529     out[ 1]=
00530     out[ 4]= t1 - t2;
00531 
00532     in0 += SHR(in4, 1);
00533     in4 = in0 + in2;
00534     in5 += 2*in1;
00535     in1 = MULH3(in5 + in3, icos36h[1], 1);
00536     out[ 8]=
00537     out[ 9]= in4 + in1;
00538     out[ 2]=
00539     out[ 3]= in4 - in1;
00540 
00541     in0 -= in2;
00542     in5 = MULH3(in5 - in3, icos36h[7], 2);
00543     out[ 0]=
00544     out[ 5]= in0 - in5;
00545     out[ 6]=
00546     out[11]= in0 + in5;
00547 }
00548 
00549 /* cos(pi*i/18) */
00550 #define C1 FIXHR(0.98480775301220805936/2)
00551 #define C2 FIXHR(0.93969262078590838405/2)
00552 #define C3 FIXHR(0.86602540378443864676/2)
00553 #define C4 FIXHR(0.76604444311897803520/2)
00554 #define C5 FIXHR(0.64278760968653932632/2)
00555 #define C6 FIXHR(0.5/2)
00556 #define C7 FIXHR(0.34202014332566873304/2)
00557 #define C8 FIXHR(0.17364817766693034885/2)
00558 
00559 
00560 /* using Lee like decomposition followed by hand coded 9 points DCT */
00561 static void imdct36(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in, INTFLOAT *win)
00562 {
00563     int i, j;
00564     INTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;
00565     INTFLOAT tmp[18], *tmp1, *in1;
00566 
00567     for(i=17;i>=1;i--)
00568         in[i] += in[i-1];
00569     for(i=17;i>=3;i-=2)
00570         in[i] += in[i-2];
00571 
00572     for(j=0;j<2;j++) {
00573         tmp1 = tmp + j;
00574         in1 = in + j;
00575 
00576         t2 = in1[2*4] + in1[2*8] - in1[2*2];
00577 
00578         t3 = in1[2*0] + SHR(in1[2*6],1);
00579         t1 = in1[2*0] - in1[2*6];
00580         tmp1[ 6] = t1 - SHR(t2,1);
00581         tmp1[16] = t1 + t2;
00582 
00583         t0 = MULH3(in1[2*2] + in1[2*4] ,    C2, 2);
00584         t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);
00585         t2 = MULH3(in1[2*2] + in1[2*8] ,   -C4, 2);
00586 
00587         tmp1[10] = t3 - t0 - t2;
00588         tmp1[ 2] = t3 + t0 + t1;
00589         tmp1[14] = t3 + t2 - t1;
00590 
00591         tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);
00592         t2 = MULH3(in1[2*1] + in1[2*5],    C1, 2);
00593         t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);
00594         t0 = MULH3(in1[2*3], C3, 2);
00595 
00596         t1 = MULH3(in1[2*1] + in1[2*7],   -C5, 2);
00597 
00598         tmp1[ 0] = t2 + t3 + t0;
00599         tmp1[12] = t2 + t1 - t0;
00600         tmp1[ 8] = t3 - t1 - t0;
00601     }
00602 
00603     i = 0;
00604     for(j=0;j<4;j++) {
00605         t0 = tmp[i];
00606         t1 = tmp[i + 2];
00607         s0 = t1 + t0;
00608         s2 = t1 - t0;
00609 
00610         t2 = tmp[i + 1];
00611         t3 = tmp[i + 3];
00612         s1 = MULH3(t3 + t2, icos36h[j], 2);
00613         s3 = MULLx(t3 - t2, icos36[8 - j], FRAC_BITS);
00614 
00615         t0 = s0 + s1;
00616         t1 = s0 - s1;
00617         out[(9 + j)*SBLIMIT] =  MULH3(t1, win[9 + j], 1) + buf[9 + j];
00618         out[(8 - j)*SBLIMIT] =  MULH3(t1, win[8 - j], 1) + buf[8 - j];
00619         buf[9 + j] = MULH3(t0, win[18 + 9 + j], 1);
00620         buf[8 - j] = MULH3(t0, win[18 + 8 - j], 1);
00621 
00622         t0 = s2 + s3;
00623         t1 = s2 - s3;
00624         out[(9 + 8 - j)*SBLIMIT] =  MULH3(t1, win[9 + 8 - j], 1) + buf[9 + 8 - j];
00625         out[(        j)*SBLIMIT] =  MULH3(t1, win[        j], 1) + buf[        j];
00626         buf[9 + 8 - j] = MULH3(t0, win[18 + 9 + 8 - j], 1);
00627         buf[      + j] = MULH3(t0, win[18         + j], 1);
00628         i += 4;
00629     }
00630 
00631     s0 = tmp[16];
00632     s1 = MULH3(tmp[17], icos36h[4], 2);
00633     t0 = s0 + s1;
00634     t1 = s0 - s1;
00635     out[(9 + 4)*SBLIMIT] =  MULH3(t1, win[9 + 4], 1) + buf[9 + 4];
00636     out[(8 - 4)*SBLIMIT] =  MULH3(t1, win[8 - 4], 1) + buf[8 - 4];
00637     buf[9 + 4] = MULH3(t0, win[18 + 9 + 4], 1);
00638     buf[8 - 4] = MULH3(t0, win[18 + 8 - 4], 1);
00639 }
00640 
00641 /* return the number of decoded frames */
00642 static int mp_decode_layer1(MPADecodeContext *s)
00643 {
00644     int bound, i, v, n, ch, j, mant;
00645     uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
00646     uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
00647 
00648     if (s->mode == MPA_JSTEREO)
00649         bound = (s->mode_ext + 1) * 4;
00650     else
00651         bound = SBLIMIT;
00652 
00653     /* allocation bits */
00654     for(i=0;i<bound;i++) {
00655         for(ch=0;ch<s->nb_channels;ch++) {
00656             allocation[ch][i] = get_bits(&s->gb, 4);
00657         }
00658     }
00659     for(i=bound;i<SBLIMIT;i++) {
00660         allocation[0][i] = get_bits(&s->gb, 4);
00661     }
00662 
00663     /* scale factors */
00664     for(i=0;i<bound;i++) {
00665         for(ch=0;ch<s->nb_channels;ch++) {
00666             if (allocation[ch][i])
00667                 scale_factors[ch][i] = get_bits(&s->gb, 6);
00668         }
00669     }
00670     for(i=bound;i<SBLIMIT;i++) {
00671         if (allocation[0][i]) {
00672             scale_factors[0][i] = get_bits(&s->gb, 6);
00673             scale_factors[1][i] = get_bits(&s->gb, 6);
00674         }
00675     }
00676 
00677     /* compute samples */
00678     for(j=0;j<12;j++) {
00679         for(i=0;i<bound;i++) {
00680             for(ch=0;ch<s->nb_channels;ch++) {
00681                 n = allocation[ch][i];
00682                 if (n) {
00683                     mant = get_bits(&s->gb, n + 1);
00684                     v = l1_unscale(n, mant, scale_factors[ch][i]);
00685                 } else {
00686                     v = 0;
00687                 }
00688                 s->sb_samples[ch][j][i] = v;
00689             }
00690         }
00691         for(i=bound;i<SBLIMIT;i++) {
00692             n = allocation[0][i];
00693             if (n) {
00694                 mant = get_bits(&s->gb, n + 1);
00695                 v = l1_unscale(n, mant, scale_factors[0][i]);
00696                 s->sb_samples[0][j][i] = v;
00697                 v = l1_unscale(n, mant, scale_factors[1][i]);
00698                 s->sb_samples[1][j][i] = v;
00699             } else {
00700                 s->sb_samples[0][j][i] = 0;
00701                 s->sb_samples[1][j][i] = 0;
00702             }
00703         }
00704     }
00705     return 12;
00706 }
00707 
00708 static int mp_decode_layer2(MPADecodeContext *s)
00709 {
00710     int sblimit; /* number of used subbands */
00711     const unsigned char *alloc_table;
00712     int table, bit_alloc_bits, i, j, ch, bound, v;
00713     unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
00714     unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
00715     unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
00716     int scale, qindex, bits, steps, k, l, m, b;
00717 
00718     /* select decoding table */
00719     table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
00720                             s->sample_rate, s->lsf);
00721     sblimit = ff_mpa_sblimit_table[table];
00722     alloc_table = ff_mpa_alloc_tables[table];
00723 
00724     if (s->mode == MPA_JSTEREO)
00725         bound = (s->mode_ext + 1) * 4;
00726     else
00727         bound = sblimit;
00728 
00729     av_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
00730 
00731     /* sanity check */
00732     if( bound > sblimit ) bound = sblimit;
00733 
00734     /* parse bit allocation */
00735     j = 0;
00736     for(i=0;i<bound;i++) {
00737         bit_alloc_bits = alloc_table[j];
00738         for(ch=0;ch<s->nb_channels;ch++) {
00739             bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
00740         }
00741         j += 1 << bit_alloc_bits;
00742     }
00743     for(i=bound;i<sblimit;i++) {
00744         bit_alloc_bits = alloc_table[j];
00745         v = get_bits(&s->gb, bit_alloc_bits);
00746         bit_alloc[0][i] = v;
00747         bit_alloc[1][i] = v;
00748         j += 1 << bit_alloc_bits;
00749     }
00750 
00751     /* scale codes */
00752     for(i=0;i<sblimit;i++) {
00753         for(ch=0;ch<s->nb_channels;ch++) {
00754             if (bit_alloc[ch][i])
00755                 scale_code[ch][i] = get_bits(&s->gb, 2);
00756         }
00757     }
00758 
00759     /* scale factors */
00760     for(i=0;i<sblimit;i++) {
00761         for(ch=0;ch<s->nb_channels;ch++) {
00762             if (bit_alloc[ch][i]) {
00763                 sf = scale_factors[ch][i];
00764                 switch(scale_code[ch][i]) {
00765                 default:
00766                 case 0:
00767                     sf[0] = get_bits(&s->gb, 6);
00768                     sf[1] = get_bits(&s->gb, 6);
00769                     sf[2] = get_bits(&s->gb, 6);
00770                     break;
00771                 case 2:
00772                     sf[0] = get_bits(&s->gb, 6);
00773                     sf[1] = sf[0];
00774                     sf[2] = sf[0];
00775                     break;
00776                 case 1:
00777                     sf[0] = get_bits(&s->gb, 6);
00778                     sf[2] = get_bits(&s->gb, 6);
00779                     sf[1] = sf[0];
00780                     break;
00781                 case 3:
00782                     sf[0] = get_bits(&s->gb, 6);
00783                     sf[2] = get_bits(&s->gb, 6);
00784                     sf[1] = sf[2];
00785                     break;
00786                 }
00787             }
00788         }
00789     }
00790 
00791     /* samples */
00792     for(k=0;k<3;k++) {
00793         for(l=0;l<12;l+=3) {
00794             j = 0;
00795             for(i=0;i<bound;i++) {
00796                 bit_alloc_bits = alloc_table[j];
00797                 for(ch=0;ch<s->nb_channels;ch++) {
00798                     b = bit_alloc[ch][i];
00799                     if (b) {
00800                         scale = scale_factors[ch][i][k];
00801                         qindex = alloc_table[j+b];
00802                         bits = ff_mpa_quant_bits[qindex];
00803                         if (bits < 0) {
00804                             int v2;
00805                             /* 3 values at the same time */
00806                             v = get_bits(&s->gb, -bits);
00807                             v2 = division_tabs[qindex][v];
00808                             steps  = ff_mpa_quant_steps[qindex];
00809 
00810                             s->sb_samples[ch][k * 12 + l + 0][i] =
00811                                 l2_unscale_group(steps, v2        & 15, scale);
00812                             s->sb_samples[ch][k * 12 + l + 1][i] =
00813                                 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
00814                             s->sb_samples[ch][k * 12 + l + 2][i] =
00815                                 l2_unscale_group(steps,  v2 >> 8      , scale);
00816                         } else {
00817                             for(m=0;m<3;m++) {
00818                                 v = get_bits(&s->gb, bits);
00819                                 v = l1_unscale(bits - 1, v, scale);
00820                                 s->sb_samples[ch][k * 12 + l + m][i] = v;
00821                             }
00822                         }
00823                     } else {
00824                         s->sb_samples[ch][k * 12 + l + 0][i] = 0;
00825                         s->sb_samples[ch][k * 12 + l + 1][i] = 0;
00826                         s->sb_samples[ch][k * 12 + l + 2][i] = 0;
00827                     }
00828                 }
00829                 /* next subband in alloc table */
00830                 j += 1 << bit_alloc_bits;
00831             }
00832             /* XXX: find a way to avoid this duplication of code */
00833             for(i=bound;i<sblimit;i++) {
00834                 bit_alloc_bits = alloc_table[j];
00835                 b = bit_alloc[0][i];
00836                 if (b) {
00837                     int mant, scale0, scale1;
00838                     scale0 = scale_factors[0][i][k];
00839                     scale1 = scale_factors[1][i][k];
00840                     qindex = alloc_table[j+b];
00841                     bits = ff_mpa_quant_bits[qindex];
00842                     if (bits < 0) {
00843                         /* 3 values at the same time */
00844                         v = get_bits(&s->gb, -bits);
00845                         steps = ff_mpa_quant_steps[qindex];
00846                         mant = v % steps;
00847                         v = v / steps;
00848                         s->sb_samples[0][k * 12 + l + 0][i] =
00849                             l2_unscale_group(steps, mant, scale0);
00850                         s->sb_samples[1][k * 12 + l + 0][i] =
00851                             l2_unscale_group(steps, mant, scale1);
00852                         mant = v % steps;
00853                         v = v / steps;
00854                         s->sb_samples[0][k * 12 + l + 1][i] =
00855                             l2_unscale_group(steps, mant, scale0);
00856                         s->sb_samples[1][k * 12 + l + 1][i] =
00857                             l2_unscale_group(steps, mant, scale1);
00858                         s->sb_samples[0][k * 12 + l + 2][i] =
00859                             l2_unscale_group(steps, v, scale0);
00860                         s->sb_samples[1][k * 12 + l + 2][i] =
00861                             l2_unscale_group(steps, v, scale1);
00862                     } else {
00863                         for(m=0;m<3;m++) {
00864                             mant = get_bits(&s->gb, bits);
00865                             s->sb_samples[0][k * 12 + l + m][i] =
00866                                 l1_unscale(bits - 1, mant, scale0);
00867                             s->sb_samples[1][k * 12 + l + m][i] =
00868                                 l1_unscale(bits - 1, mant, scale1);
00869                         }
00870                     }
00871                 } else {
00872                     s->sb_samples[0][k * 12 + l + 0][i] = 0;
00873                     s->sb_samples[0][k * 12 + l + 1][i] = 0;
00874                     s->sb_samples[0][k * 12 + l + 2][i] = 0;
00875                     s->sb_samples[1][k * 12 + l + 0][i] = 0;
00876                     s->sb_samples[1][k * 12 + l + 1][i] = 0;
00877                     s->sb_samples[1][k * 12 + l + 2][i] = 0;
00878                 }
00879                 /* next subband in alloc table */
00880                 j += 1 << bit_alloc_bits;
00881             }
00882             /* fill remaining samples to zero */
00883             for(i=sblimit;i<SBLIMIT;i++) {
00884                 for(ch=0;ch<s->nb_channels;ch++) {
00885                     s->sb_samples[ch][k * 12 + l + 0][i] = 0;
00886                     s->sb_samples[ch][k * 12 + l + 1][i] = 0;
00887                     s->sb_samples[ch][k * 12 + l + 2][i] = 0;
00888                 }
00889             }
00890         }
00891     }
00892     return 3 * 12;
00893 }
00894 
00895 #define SPLIT(dst,sf,n)\
00896     if(n==3){\
00897         int m= (sf*171)>>9;\
00898         dst= sf - 3*m;\
00899         sf=m;\
00900     }else if(n==4){\
00901         dst= sf&3;\
00902         sf>>=2;\
00903     }else if(n==5){\
00904         int m= (sf*205)>>10;\
00905         dst= sf - 5*m;\
00906         sf=m;\
00907     }else if(n==6){\
00908         int m= (sf*171)>>10;\
00909         dst= sf - 6*m;\
00910         sf=m;\
00911     }else{\
00912         dst=0;\
00913     }
00914 
00915 static av_always_inline void lsf_sf_expand(int *slen,
00916                                  int sf, int n1, int n2, int n3)
00917 {
00918     SPLIT(slen[3], sf, n3)
00919     SPLIT(slen[2], sf, n2)
00920     SPLIT(slen[1], sf, n1)
00921     slen[0] = sf;
00922 }
00923 
00924 static void exponents_from_scale_factors(MPADecodeContext *s,
00925                                          GranuleDef *g,
00926                                          int16_t *exponents)
00927 {
00928     const uint8_t *bstab, *pretab;
00929     int len, i, j, k, l, v0, shift, gain, gains[3];
00930     int16_t *exp_ptr;
00931 
00932     exp_ptr = exponents;
00933     gain = g->global_gain - 210;
00934     shift = g->scalefac_scale + 1;
00935 
00936     bstab = band_size_long[s->sample_rate_index];
00937     pretab = mpa_pretab[g->preflag];
00938     for(i=0;i<g->long_end;i++) {
00939         v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
00940         len = bstab[i];
00941         for(j=len;j>0;j--)
00942             *exp_ptr++ = v0;
00943     }
00944 
00945     if (g->short_start < 13) {
00946         bstab = band_size_short[s->sample_rate_index];
00947         gains[0] = gain - (g->subblock_gain[0] << 3);
00948         gains[1] = gain - (g->subblock_gain[1] << 3);
00949         gains[2] = gain - (g->subblock_gain[2] << 3);
00950         k = g->long_end;
00951         for(i=g->short_start;i<13;i++) {
00952             len = bstab[i];
00953             for(l=0;l<3;l++) {
00954                 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
00955                 for(j=len;j>0;j--)
00956                 *exp_ptr++ = v0;
00957             }
00958         }
00959     }
00960 }
00961 
00962 /* handle n = 0 too */
00963 static inline int get_bitsz(GetBitContext *s, int n)
00964 {
00965     if (n == 0)
00966         return 0;
00967     else
00968         return get_bits(s, n);
00969 }
00970 
00971 
00972 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, int *end_pos2){
00973     if(s->in_gb.buffer && *pos >= s->gb.size_in_bits){
00974         s->gb= s->in_gb;
00975         s->in_gb.buffer=NULL;
00976         assert((get_bits_count(&s->gb) & 7) == 0);
00977         skip_bits_long(&s->gb, *pos - *end_pos);
00978         *end_pos2=
00979         *end_pos= *end_pos2 + get_bits_count(&s->gb) - *pos;
00980         *pos= get_bits_count(&s->gb);
00981     }
00982 }
00983 
00984 /* Following is a optimized code for
00985             INTFLOAT v = *src
00986             if(get_bits1(&s->gb))
00987                 v = -v;
00988             *dst = v;
00989 */
00990 #if CONFIG_FLOAT
00991 #define READ_FLIP_SIGN(dst,src)\
00992             v = AV_RN32A(src) ^ (get_bits1(&s->gb)<<31);\
00993             AV_WN32A(dst, v);
00994 #else
00995 #define READ_FLIP_SIGN(dst,src)\
00996             v= -get_bits1(&s->gb);\
00997             *(dst) = (*(src) ^ v) - v;
00998 #endif
00999 
01000 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
01001                           int16_t *exponents, int end_pos2)
01002 {
01003     int s_index;
01004     int i;
01005     int last_pos, bits_left;
01006     VLC *vlc;
01007     int end_pos= FFMIN(end_pos2, s->gb.size_in_bits);
01008 
01009     /* low frequencies (called big values) */
01010     s_index = 0;
01011     for(i=0;i<3;i++) {
01012         int j, k, l, linbits;
01013         j = g->region_size[i];
01014         if (j == 0)
01015             continue;
01016         /* select vlc table */
01017         k = g->table_select[i];
01018         l = mpa_huff_data[k][0];
01019         linbits = mpa_huff_data[k][1];
01020         vlc = &huff_vlc[l];
01021 
01022         if(!l){
01023             memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*2*j);
01024             s_index += 2*j;
01025             continue;
01026         }
01027 
01028         /* read huffcode and compute each couple */
01029         for(;j>0;j--) {
01030             int exponent, x, y;
01031             int v;
01032             int pos= get_bits_count(&s->gb);
01033 
01034             if (pos >= end_pos){
01035 //                av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
01036                 switch_buffer(s, &pos, &end_pos, &end_pos2);
01037 //                av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);
01038                 if(pos >= end_pos)
01039                     break;
01040             }
01041             y = get_vlc2(&s->gb, vlc->table, 7, 3);
01042 
01043             if(!y){
01044                 g->sb_hybrid[s_index  ] =
01045                 g->sb_hybrid[s_index+1] = 0;
01046                 s_index += 2;
01047                 continue;
01048             }
01049 
01050             exponent= exponents[s_index];
01051 
01052             av_dlog(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
01053                     i, g->region_size[i] - j, x, y, exponent);
01054             if(y&16){
01055                 x = y >> 5;
01056                 y = y & 0x0f;
01057                 if (x < 15){
01058                     READ_FLIP_SIGN(g->sb_hybrid+s_index, RENAME(expval_table)[ exponent ]+x)
01059                 }else{
01060                     x += get_bitsz(&s->gb, linbits);
01061                     v = l3_unscale(x, exponent);
01062                     if (get_bits1(&s->gb))
01063                         v = -v;
01064                     g->sb_hybrid[s_index] = v;
01065                 }
01066                 if (y < 15){
01067                     READ_FLIP_SIGN(g->sb_hybrid+s_index+1, RENAME(expval_table)[ exponent ]+y)
01068                 }else{
01069                     y += get_bitsz(&s->gb, linbits);
01070                     v = l3_unscale(y, exponent);
01071                     if (get_bits1(&s->gb))
01072                         v = -v;
01073                     g->sb_hybrid[s_index+1] = v;
01074                 }
01075             }else{
01076                 x = y >> 5;
01077                 y = y & 0x0f;
01078                 x += y;
01079                 if (x < 15){
01080                     READ_FLIP_SIGN(g->sb_hybrid+s_index+!!y, RENAME(expval_table)[ exponent ]+x)
01081                 }else{
01082                     x += get_bitsz(&s->gb, linbits);
01083                     v = l3_unscale(x, exponent);
01084                     if (get_bits1(&s->gb))
01085                         v = -v;
01086                     g->sb_hybrid[s_index+!!y] = v;
01087                 }
01088                 g->sb_hybrid[s_index+ !y] = 0;
01089             }
01090             s_index+=2;
01091         }
01092     }
01093 
01094     /* high frequencies */
01095     vlc = &huff_quad_vlc[g->count1table_select];
01096     last_pos=0;
01097     while (s_index <= 572) {
01098         int pos, code;
01099         pos = get_bits_count(&s->gb);
01100         if (pos >= end_pos) {
01101             if (pos > end_pos2 && last_pos){
01102                 /* some encoders generate an incorrect size for this
01103                    part. We must go back into the data */
01104                 s_index -= 4;
01105                 skip_bits_long(&s->gb, last_pos - pos);
01106                 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
01107                 if(s->error_recognition >= FF_ER_COMPLIANT)
01108                     s_index=0;
01109                 break;
01110             }
01111 //                av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
01112             switch_buffer(s, &pos, &end_pos, &end_pos2);
01113 //                av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);
01114             if(pos >= end_pos)
01115                 break;
01116         }
01117         last_pos= pos;
01118 
01119         code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
01120         av_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
01121         g->sb_hybrid[s_index+0]=
01122         g->sb_hybrid[s_index+1]=
01123         g->sb_hybrid[s_index+2]=
01124         g->sb_hybrid[s_index+3]= 0;
01125         while(code){
01126             static const int idxtab[16]={3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0};
01127             int v;
01128             int pos= s_index+idxtab[code];
01129             code ^= 8>>idxtab[code];
01130             READ_FLIP_SIGN(g->sb_hybrid+pos, RENAME(exp_table)+exponents[pos])
01131         }
01132         s_index+=4;
01133     }
01134     /* skip extension bits */
01135     bits_left = end_pos2 - get_bits_count(&s->gb);
01136 //av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);
01137     if (bits_left < 0 && s->error_recognition >= FF_ER_COMPLIANT) {
01138         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
01139         s_index=0;
01140     }else if(bits_left > 0 && s->error_recognition >= FF_ER_AGGRESSIVE){
01141         av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
01142         s_index=0;
01143     }
01144     memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid)*(576 - s_index));
01145     skip_bits_long(&s->gb, bits_left);
01146 
01147     i= get_bits_count(&s->gb);
01148     switch_buffer(s, &i, &end_pos, &end_pos2);
01149 
01150     return 0;
01151 }
01152 
01153 /* Reorder short blocks from bitstream order to interleaved order. It
01154    would be faster to do it in parsing, but the code would be far more
01155    complicated */
01156 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
01157 {
01158     int i, j, len;
01159     INTFLOAT *ptr, *dst, *ptr1;
01160     INTFLOAT tmp[576];
01161 
01162     if (g->block_type != 2)
01163         return;
01164 
01165     if (g->switch_point) {
01166         if (s->sample_rate_index != 8) {
01167             ptr = g->sb_hybrid + 36;
01168         } else {
01169             ptr = g->sb_hybrid + 48;
01170         }
01171     } else {
01172         ptr = g->sb_hybrid;
01173     }
01174 
01175     for(i=g->short_start;i<13;i++) {
01176         len = band_size_short[s->sample_rate_index][i];
01177         ptr1 = ptr;
01178         dst = tmp;
01179         for(j=len;j>0;j--) {
01180             *dst++ = ptr[0*len];
01181             *dst++ = ptr[1*len];
01182             *dst++ = ptr[2*len];
01183             ptr++;
01184         }
01185         ptr+=2*len;
01186         memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
01187     }
01188 }
01189 
01190 #define ISQRT2 FIXR(0.70710678118654752440)
01191 
01192 static void compute_stereo(MPADecodeContext *s,
01193                            GranuleDef *g0, GranuleDef *g1)
01194 {
01195     int i, j, k, l;
01196     int sf_max, sf, len, non_zero_found;
01197     INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
01198     int non_zero_found_short[3];
01199 
01200     /* intensity stereo */
01201     if (s->mode_ext & MODE_EXT_I_STEREO) {
01202         if (!s->lsf) {
01203             is_tab = is_table;
01204             sf_max = 7;
01205         } else {
01206             is_tab = is_table_lsf[g1->scalefac_compress & 1];
01207             sf_max = 16;
01208         }
01209 
01210         tab0 = g0->sb_hybrid + 576;
01211         tab1 = g1->sb_hybrid + 576;
01212 
01213         non_zero_found_short[0] = 0;
01214         non_zero_found_short[1] = 0;
01215         non_zero_found_short[2] = 0;
01216         k = (13 - g1->short_start) * 3 + g1->long_end - 3;
01217         for(i = 12;i >= g1->short_start;i--) {
01218             /* for last band, use previous scale factor */
01219             if (i != 11)
01220                 k -= 3;
01221             len = band_size_short[s->sample_rate_index][i];
01222             for(l=2;l>=0;l--) {
01223                 tab0 -= len;
01224                 tab1 -= len;
01225                 if (!non_zero_found_short[l]) {
01226                     /* test if non zero band. if so, stop doing i-stereo */
01227                     for(j=0;j<len;j++) {
01228                         if (tab1[j] != 0) {
01229                             non_zero_found_short[l] = 1;
01230                             goto found1;
01231                         }
01232                     }
01233                     sf = g1->scale_factors[k + l];
01234                     if (sf >= sf_max)
01235                         goto found1;
01236 
01237                     v1 = is_tab[0][sf];
01238                     v2 = is_tab[1][sf];
01239                     for(j=0;j<len;j++) {
01240                         tmp0 = tab0[j];
01241                         tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
01242                         tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
01243                     }
01244                 } else {
01245                 found1:
01246                     if (s->mode_ext & MODE_EXT_MS_STEREO) {
01247                         /* lower part of the spectrum : do ms stereo
01248                            if enabled */
01249                         for(j=0;j<len;j++) {
01250                             tmp0 = tab0[j];
01251                             tmp1 = tab1[j];
01252                             tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
01253                             tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
01254                         }
01255                     }
01256                 }
01257             }
01258         }
01259 
01260         non_zero_found = non_zero_found_short[0] |
01261             non_zero_found_short[1] |
01262             non_zero_found_short[2];
01263 
01264         for(i = g1->long_end - 1;i >= 0;i--) {
01265             len = band_size_long[s->sample_rate_index][i];
01266             tab0 -= len;
01267             tab1 -= len;
01268             /* test if non zero band. if so, stop doing i-stereo */
01269             if (!non_zero_found) {
01270                 for(j=0;j<len;j++) {
01271                     if (tab1[j] != 0) {
01272                         non_zero_found = 1;
01273                         goto found2;
01274                     }
01275                 }
01276                 /* for last band, use previous scale factor */
01277                 k = (i == 21) ? 20 : i;
01278                 sf = g1->scale_factors[k];
01279                 if (sf >= sf_max)
01280                     goto found2;
01281                 v1 = is_tab[0][sf];
01282                 v2 = is_tab[1][sf];
01283                 for(j=0;j<len;j++) {
01284                     tmp0 = tab0[j];
01285                     tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
01286                     tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
01287                 }
01288             } else {
01289             found2:
01290                 if (s->mode_ext & MODE_EXT_MS_STEREO) {
01291                     /* lower part of the spectrum : do ms stereo
01292                        if enabled */
01293                     for(j=0;j<len;j++) {
01294                         tmp0 = tab0[j];
01295                         tmp1 = tab1[j];
01296                         tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
01297                         tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
01298                     }
01299                 }
01300             }
01301         }
01302     } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
01303         /* ms stereo ONLY */
01304         /* NOTE: the 1/sqrt(2) normalization factor is included in the
01305            global gain */
01306         tab0 = g0->sb_hybrid;
01307         tab1 = g1->sb_hybrid;
01308         for(i=0;i<576;i++) {
01309             tmp0 = tab0[i];
01310             tmp1 = tab1[i];
01311             tab0[i] = tmp0 + tmp1;
01312             tab1[i] = tmp0 - tmp1;
01313         }
01314     }
01315 }
01316 
01317 #if CONFIG_FLOAT
01318 #define AA(j) do {                                                      \
01319         float tmp0 = ptr[-1-j];                                         \
01320         float tmp1 = ptr[   j];                                         \
01321         ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1];    \
01322         ptr[   j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0];    \
01323     } while (0)
01324 #else
01325 #define AA(j) do {                                              \
01326         int tmp0 = ptr[-1-j];                                   \
01327         int tmp1 = ptr[   j];                                   \
01328         int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]);          \
01329         ptr[-1-j] = 4*(tmp2 - MULH(tmp1, csa_table[j][2]));     \
01330         ptr[   j] = 4*(tmp2 + MULH(tmp0, csa_table[j][3]));     \
01331     } while (0)
01332 #endif
01333 
01334 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
01335 {
01336     INTFLOAT *ptr;
01337     int n, i;
01338 
01339     /* we antialias only "long" bands */
01340     if (g->block_type == 2) {
01341         if (!g->switch_point)
01342             return;
01343         /* XXX: check this for 8000Hz case */
01344         n = 1;
01345     } else {
01346         n = SBLIMIT - 1;
01347     }
01348 
01349     ptr = g->sb_hybrid + 18;
01350     for(i = n;i > 0;i--) {
01351         AA(0);
01352         AA(1);
01353         AA(2);
01354         AA(3);
01355         AA(4);
01356         AA(5);
01357         AA(6);
01358         AA(7);
01359 
01360         ptr += 18;
01361     }
01362 }
01363 
01364 static void compute_imdct(MPADecodeContext *s,
01365                           GranuleDef *g,
01366                           INTFLOAT *sb_samples,
01367                           INTFLOAT *mdct_buf)
01368 {
01369     INTFLOAT *win, *win1, *out_ptr, *ptr, *buf, *ptr1;
01370     INTFLOAT out2[12];
01371     int i, j, mdct_long_end, sblimit;
01372 
01373     /* find last non zero block */
01374     ptr = g->sb_hybrid + 576;
01375     ptr1 = g->sb_hybrid + 2 * 18;
01376     while (ptr >= ptr1) {
01377         int32_t *p;
01378         ptr -= 6;
01379         p= (int32_t*)ptr;
01380         if(p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
01381             break;
01382     }
01383     sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
01384 
01385     if (g->block_type == 2) {
01386         /* XXX: check for 8000 Hz */
01387         if (g->switch_point)
01388             mdct_long_end = 2;
01389         else
01390             mdct_long_end = 0;
01391     } else {
01392         mdct_long_end = sblimit;
01393     }
01394 
01395     buf = mdct_buf;
01396     ptr = g->sb_hybrid;
01397     for(j=0;j<mdct_long_end;j++) {
01398         /* apply window & overlap with previous buffer */
01399         out_ptr = sb_samples + j;
01400         /* select window */
01401         if (g->switch_point && j < 2)
01402             win1 = mdct_win[0];
01403         else
01404             win1 = mdct_win[g->block_type];
01405         /* select frequency inversion */
01406         win = win1 + ((4 * 36) & -(j & 1));
01407         imdct36(out_ptr, buf, ptr, win);
01408         out_ptr += 18*SBLIMIT;
01409         ptr += 18;
01410         buf += 18;
01411     }
01412     for(j=mdct_long_end;j<sblimit;j++) {
01413         /* select frequency inversion */
01414         win = mdct_win[2] + ((4 * 36) & -(j & 1));
01415         out_ptr = sb_samples + j;
01416 
01417         for(i=0; i<6; i++){
01418             *out_ptr = buf[i];
01419             out_ptr += SBLIMIT;
01420         }
01421         imdct12(out2, ptr + 0);
01422         for(i=0;i<6;i++) {
01423             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*1];
01424             buf[i + 6*2] = MULH3(out2[i + 6], win[i + 6], 1);
01425             out_ptr += SBLIMIT;
01426         }
01427         imdct12(out2, ptr + 1);
01428         for(i=0;i<6;i++) {
01429             *out_ptr     = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*2];
01430             buf[i + 6*0] = MULH3(out2[i + 6], win[i + 6], 1);
01431             out_ptr += SBLIMIT;
01432         }
01433         imdct12(out2, ptr + 2);
01434         for(i=0;i<6;i++) {
01435             buf[i + 6*0] = MULH3(out2[i    ], win[i    ], 1) + buf[i + 6*0];
01436             buf[i + 6*1] = MULH3(out2[i + 6], win[i + 6], 1);
01437             buf[i + 6*2] = 0;
01438         }
01439         ptr += 18;
01440         buf += 18;
01441     }
01442     /* zero bands */
01443     for(j=sblimit;j<SBLIMIT;j++) {
01444         /* overlap */
01445         out_ptr = sb_samples + j;
01446         for(i=0;i<18;i++) {
01447             *out_ptr = buf[i];
01448             buf[i] = 0;
01449             out_ptr += SBLIMIT;
01450         }
01451         buf += 18;
01452     }
01453 }
01454 
01455 /* main layer3 decoding function */
01456 static int mp_decode_layer3(MPADecodeContext *s)
01457 {
01458     int nb_granules, main_data_begin;
01459     int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
01460     GranuleDef *g;
01461     int16_t exponents[576]; //FIXME try INTFLOAT
01462 
01463     /* read side info */
01464     if (s->lsf) {
01465         main_data_begin = get_bits(&s->gb, 8);
01466         skip_bits(&s->gb, s->nb_channels);
01467         nb_granules = 1;
01468     } else {
01469         main_data_begin = get_bits(&s->gb, 9);
01470         if (s->nb_channels == 2)
01471             skip_bits(&s->gb, 3);
01472         else
01473             skip_bits(&s->gb, 5);
01474         nb_granules = 2;
01475         for(ch=0;ch<s->nb_channels;ch++) {
01476             s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
01477             s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
01478         }
01479     }
01480 
01481     for(gr=0;gr<nb_granules;gr++) {
01482         for(ch=0;ch<s->nb_channels;ch++) {
01483             av_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
01484             g = &s->granules[ch][gr];
01485             g->part2_3_length = get_bits(&s->gb, 12);
01486             g->big_values = get_bits(&s->gb, 9);
01487             if(g->big_values > 288){
01488                 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
01489                 return -1;
01490             }
01491 
01492             g->global_gain = get_bits(&s->gb, 8);
01493             /* if MS stereo only is selected, we precompute the
01494                1/sqrt(2) renormalization factor */
01495             if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
01496                 MODE_EXT_MS_STEREO)
01497                 g->global_gain -= 2;
01498             if (s->lsf)
01499                 g->scalefac_compress = get_bits(&s->gb, 9);
01500             else
01501                 g->scalefac_compress = get_bits(&s->gb, 4);
01502             blocksplit_flag = get_bits1(&s->gb);
01503             if (blocksplit_flag) {
01504                 g->block_type = get_bits(&s->gb, 2);
01505                 if (g->block_type == 0){
01506                     av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
01507                     return -1;
01508                 }
01509                 g->switch_point = get_bits1(&s->gb);
01510                 for(i=0;i<2;i++)
01511                     g->table_select[i] = get_bits(&s->gb, 5);
01512                 for(i=0;i<3;i++)
01513                     g->subblock_gain[i] = get_bits(&s->gb, 3);
01514                 ff_init_short_region(s, g);
01515             } else {
01516                 int region_address1, region_address2;
01517                 g->block_type = 0;
01518                 g->switch_point = 0;
01519                 for(i=0;i<3;i++)
01520                     g->table_select[i] = get_bits(&s->gb, 5);
01521                 /* compute huffman coded region sizes */
01522                 region_address1 = get_bits(&s->gb, 4);
01523                 region_address2 = get_bits(&s->gb, 3);
01524                 av_dlog(s->avctx, "region1=%d region2=%d\n",
01525                         region_address1, region_address2);
01526                 ff_init_long_region(s, g, region_address1, region_address2);
01527             }
01528             ff_region_offset2size(g);
01529             ff_compute_band_indexes(s, g);
01530 
01531             g->preflag = 0;
01532             if (!s->lsf)
01533                 g->preflag = get_bits1(&s->gb);
01534             g->scalefac_scale = get_bits1(&s->gb);
01535             g->count1table_select = get_bits1(&s->gb);
01536             av_dlog(s->avctx, "block_type=%d switch_point=%d\n",
01537                     g->block_type, g->switch_point);
01538         }
01539     }
01540 
01541   if (!s->adu_mode) {
01542     const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
01543     assert((get_bits_count(&s->gb) & 7) == 0);
01544     /* now we get bits from the main_data_begin offset */
01545     av_dlog(s->avctx, "seekback: %d\n", main_data_begin);
01546 //av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
01547 
01548     memcpy(s->last_buf + s->last_buf_size, ptr, EXTRABYTES);
01549     s->in_gb= s->gb;
01550         init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
01551         skip_bits_long(&s->gb, 8*(s->last_buf_size - main_data_begin));
01552   }
01553 
01554     for(gr=0;gr<nb_granules;gr++) {
01555         for(ch=0;ch<s->nb_channels;ch++) {
01556             g = &s->granules[ch][gr];
01557             if(get_bits_count(&s->gb)<0){
01558                 av_log(s->avctx, AV_LOG_DEBUG, "mdb:%d, lastbuf:%d skipping granule %d\n",
01559                                             main_data_begin, s->last_buf_size, gr);
01560                 skip_bits_long(&s->gb, g->part2_3_length);
01561                 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
01562                 if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->in_gb.buffer){
01563                     skip_bits_long(&s->in_gb, get_bits_count(&s->gb) - s->gb.size_in_bits);
01564                     s->gb= s->in_gb;
01565                     s->in_gb.buffer=NULL;
01566                 }
01567                 continue;
01568             }
01569 
01570             bits_pos = get_bits_count(&s->gb);
01571 
01572             if (!s->lsf) {
01573                 uint8_t *sc;
01574                 int slen, slen1, slen2;
01575 
01576                 /* MPEG1 scale factors */
01577                 slen1 = slen_table[0][g->scalefac_compress];
01578                 slen2 = slen_table[1][g->scalefac_compress];
01579                 av_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
01580                 if (g->block_type == 2) {
01581                     n = g->switch_point ? 17 : 18;
01582                     j = 0;
01583                     if(slen1){
01584                         for(i=0;i<n;i++)
01585                             g->scale_factors[j++] = get_bits(&s->gb, slen1);
01586                     }else{
01587                         for(i=0;i<n;i++)
01588                             g->scale_factors[j++] = 0;
01589                     }
01590                     if(slen2){
01591                         for(i=0;i<18;i++)
01592                             g->scale_factors[j++] = get_bits(&s->gb, slen2);
01593                         for(i=0;i<3;i++)
01594                             g->scale_factors[j++] = 0;
01595                     }else{
01596                         for(i=0;i<21;i++)
01597                             g->scale_factors[j++] = 0;
01598                     }
01599                 } else {
01600                     sc = s->granules[ch][0].scale_factors;
01601                     j = 0;
01602                     for(k=0;k<4;k++) {
01603                         n = (k == 0 ? 6 : 5);
01604                         if ((g->scfsi & (0x8 >> k)) == 0) {
01605                             slen = (k < 2) ? slen1 : slen2;
01606                             if(slen){
01607                                 for(i=0;i<n;i++)
01608                                     g->scale_factors[j++] = get_bits(&s->gb, slen);
01609                             }else{
01610                                 for(i=0;i<n;i++)
01611                                     g->scale_factors[j++] = 0;
01612                             }
01613                         } else {
01614                             /* simply copy from last granule */
01615                             for(i=0;i<n;i++) {
01616                                 g->scale_factors[j] = sc[j];
01617                                 j++;
01618                             }
01619                         }
01620                     }
01621                     g->scale_factors[j++] = 0;
01622                 }
01623             } else {
01624                 int tindex, tindex2, slen[4], sl, sf;
01625 
01626                 /* LSF scale factors */
01627                 if (g->block_type == 2) {
01628                     tindex = g->switch_point ? 2 : 1;
01629                 } else {
01630                     tindex = 0;
01631                 }
01632                 sf = g->scalefac_compress;
01633                 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
01634                     /* intensity stereo case */
01635                     sf >>= 1;
01636                     if (sf < 180) {
01637                         lsf_sf_expand(slen, sf, 6, 6, 0);
01638                         tindex2 = 3;
01639                     } else if (sf < 244) {
01640                         lsf_sf_expand(slen, sf - 180, 4, 4, 0);
01641                         tindex2 = 4;
01642                     } else {
01643                         lsf_sf_expand(slen, sf - 244, 3, 0, 0);
01644                         tindex2 = 5;
01645                     }
01646                 } else {
01647                     /* normal case */
01648                     if (sf < 400) {
01649                         lsf_sf_expand(slen, sf, 5, 4, 4);
01650                         tindex2 = 0;
01651                     } else if (sf < 500) {
01652                         lsf_sf_expand(slen, sf - 400, 5, 4, 0);
01653                         tindex2 = 1;
01654                     } else {
01655                         lsf_sf_expand(slen, sf - 500, 3, 0, 0);
01656                         tindex2 = 2;
01657                         g->preflag = 1;
01658                     }
01659                 }
01660 
01661                 j = 0;
01662                 for(k=0;k<4;k++) {
01663                     n = lsf_nsf_table[tindex2][tindex][k];
01664                     sl = slen[k];
01665                     if(sl){
01666                         for(i=0;i<n;i++)
01667                             g->scale_factors[j++] = get_bits(&s->gb, sl);
01668                     }else{
01669                         for(i=0;i<n;i++)
01670                             g->scale_factors[j++] = 0;
01671                     }
01672                 }
01673                 /* XXX: should compute exact size */
01674                 for(;j<40;j++)
01675                     g->scale_factors[j] = 0;
01676             }
01677 
01678             exponents_from_scale_factors(s, g, exponents);
01679 
01680             /* read Huffman coded residue */
01681             huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
01682         } /* ch */
01683 
01684         if (s->nb_channels == 2)
01685             compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
01686 
01687         for(ch=0;ch<s->nb_channels;ch++) {
01688             g = &s->granules[ch][gr];
01689 
01690             reorder_block(s, g);
01691             compute_antialias(s, g);
01692             compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
01693         }
01694     } /* gr */
01695     if(get_bits_count(&s->gb)<0)
01696         skip_bits_long(&s->gb, -get_bits_count(&s->gb));
01697     return nb_granules * 18;
01698 }
01699 
01700 static int mp_decode_frame(MPADecodeContext *s,
01701                            OUT_INT *samples, const uint8_t *buf, int buf_size)
01702 {
01703     int i, nb_frames, ch;
01704     OUT_INT *samples_ptr;
01705 
01706     init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE)*8);
01707 
01708     /* skip error protection field */
01709     if (s->error_protection)
01710         skip_bits(&s->gb, 16);
01711 
01712     switch(s->layer) {
01713     case 1:
01714         s->avctx->frame_size = 384;
01715         nb_frames = mp_decode_layer1(s);
01716         break;
01717     case 2:
01718         s->avctx->frame_size = 1152;
01719         nb_frames = mp_decode_layer2(s);
01720         break;
01721     case 3:
01722         s->avctx->frame_size = s->lsf ? 576 : 1152;
01723     default:
01724         nb_frames = mp_decode_layer3(s);
01725 
01726         s->last_buf_size=0;
01727         if(s->in_gb.buffer){
01728             align_get_bits(&s->gb);
01729             i= get_bits_left(&s->gb)>>3;
01730             if(i >= 0 && i <= BACKSTEP_SIZE){
01731                 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
01732                 s->last_buf_size=i;
01733             }else
01734                 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
01735             s->gb= s->in_gb;
01736             s->in_gb.buffer= NULL;
01737         }
01738 
01739         align_get_bits(&s->gb);
01740         assert((get_bits_count(&s->gb) & 7) == 0);
01741         i= get_bits_left(&s->gb)>>3;
01742 
01743         if(i<0 || i > BACKSTEP_SIZE || nb_frames<0){
01744             if(i<0)
01745                 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
01746             i= FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
01747         }
01748         assert(i <= buf_size - HEADER_SIZE && i>= 0);
01749         memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
01750         s->last_buf_size += i;
01751 
01752         break;
01753     }
01754 
01755     /* apply the synthesis filter */
01756     for(ch=0;ch<s->nb_channels;ch++) {
01757         samples_ptr = samples + ch;
01758         for(i=0;i<nb_frames;i++) {
01759             RENAME(ff_mpa_synth_filter)(
01760                          &s->mpadsp,
01761                          s->synth_buf[ch], &(s->synth_buf_offset[ch]),
01762                          RENAME(ff_mpa_synth_window), &s->dither_state,
01763                          samples_ptr, s->nb_channels,
01764                          s->sb_samples[ch][i]);
01765             samples_ptr += 32 * s->nb_channels;
01766         }
01767     }
01768 
01769     return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
01770 }
01771 
01772 static int decode_frame(AVCodecContext * avctx,
01773                         void *data, int *data_size,
01774                         AVPacket *avpkt)
01775 {
01776     const uint8_t *buf = avpkt->data;
01777     int buf_size = avpkt->size;
01778     MPADecodeContext *s = avctx->priv_data;
01779     uint32_t header;
01780     int out_size;
01781     OUT_INT *out_samples = data;
01782 
01783     if(buf_size < HEADER_SIZE)
01784         return -1;
01785 
01786     header = AV_RB32(buf);
01787     if(ff_mpa_check_header(header) < 0){
01788         av_log(avctx, AV_LOG_ERROR, "Header missing\n");
01789         return -1;
01790     }
01791 
01792     if (ff_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
01793         /* free format: prepare to compute frame size */
01794         s->frame_size = -1;
01795         return -1;
01796     }
01797     /* update codec info */
01798     avctx->channels = s->nb_channels;
01799     avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
01800     if (!avctx->bit_rate)
01801         avctx->bit_rate = s->bit_rate;
01802     avctx->sub_id = s->layer;
01803 
01804     if (*data_size < avctx->frame_size * avctx->channels * sizeof(OUT_INT))
01805         return AVERROR(EINVAL);
01806     *data_size = 0;
01807 
01808     if(s->frame_size<=0 || s->frame_size > buf_size){
01809         av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
01810         return -1;
01811     }else if(s->frame_size < buf_size){
01812         av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");
01813         buf_size= s->frame_size;
01814     }
01815 
01816     out_size = mp_decode_frame(s, out_samples, buf, buf_size);
01817     if(out_size>=0){
01818         *data_size = out_size;
01819         avctx->sample_rate = s->sample_rate;
01820         //FIXME maybe move the other codec info stuff from above here too
01821     }else
01822         av_log(avctx, AV_LOG_DEBUG, "Error while decoding MPEG audio frame.\n"); //FIXME return -1 / but also return the number of bytes consumed
01823     s->frame_size = 0;
01824     return buf_size;
01825 }
01826 
01827 static void flush(AVCodecContext *avctx){
01828     MPADecodeContext *s = avctx->priv_data;
01829     memset(s->synth_buf, 0, sizeof(s->synth_buf));
01830     s->last_buf_size= 0;
01831 }
01832 
01833 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
01834 static int decode_frame_adu(AVCodecContext * avctx,
01835                         void *data, int *data_size,
01836                         AVPacket *avpkt)
01837 {
01838     const uint8_t *buf = avpkt->data;
01839     int buf_size = avpkt->size;
01840     MPADecodeContext *s = avctx->priv_data;
01841     uint32_t header;
01842     int len, out_size;
01843     OUT_INT *out_samples = data;
01844 
01845     len = buf_size;
01846 
01847     // Discard too short frames
01848     if (buf_size < HEADER_SIZE) {
01849         *data_size = 0;
01850         return buf_size;
01851     }
01852 
01853 
01854     if (len > MPA_MAX_CODED_FRAME_SIZE)
01855         len = MPA_MAX_CODED_FRAME_SIZE;
01856 
01857     // Get header and restore sync word
01858     header = AV_RB32(buf) | 0xffe00000;
01859 
01860     if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
01861         *data_size = 0;
01862         return buf_size;
01863     }
01864 
01865     ff_mpegaudio_decode_header((MPADecodeHeader *)s, header);
01866     /* update codec info */
01867     avctx->sample_rate = s->sample_rate;
01868     avctx->channels = s->nb_channels;
01869     if (!avctx->bit_rate)
01870         avctx->bit_rate = s->bit_rate;
01871     avctx->sub_id = s->layer;
01872 
01873     if (*data_size < avctx->frame_size * avctx->channels * sizeof(OUT_INT))
01874         return AVERROR(EINVAL);
01875 
01876     s->frame_size = len;
01877 
01878     if (avctx->parse_only) {
01879         out_size = buf_size;
01880     } else {
01881         out_size = mp_decode_frame(s, out_samples, buf, buf_size);
01882     }
01883 
01884     *data_size = out_size;
01885     return buf_size;
01886 }
01887 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
01888 
01889 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
01890 
01894 typedef struct MP3On4DecodeContext {
01895     int frames;   
01896     int syncword; 
01897     const uint8_t *coff; 
01898     MPADecodeContext *mp3decctx[5]; 
01899 } MP3On4DecodeContext;
01900 
01901 #include "mpeg4audio.h"
01902 
01903 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
01904 static const uint8_t mp3Frames[8] = {0,1,1,2,3,3,4,5};   /* number of mp3 decoder instances */
01905 /* offsets into output buffer, assume output order is FL FR BL BR C LFE */
01906 static const uint8_t chan_offset[8][5] = {
01907     {0},
01908     {0},            // C
01909     {0},            // FLR
01910     {2,0},          // C FLR
01911     {2,0,3},        // C FLR BS
01912     {4,0,2},        // C FLR BLRS
01913     {4,0,2,5},      // C FLR BLRS LFE
01914     {4,0,2,6,5},    // C FLR BLRS BLR LFE
01915 };
01916 
01917 
01918 static int decode_init_mp3on4(AVCodecContext * avctx)
01919 {
01920     MP3On4DecodeContext *s = avctx->priv_data;
01921     MPEG4AudioConfig cfg;
01922     int i;
01923 
01924     if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
01925         av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
01926         return -1;
01927     }
01928 
01929     ff_mpeg4audio_get_config(&cfg, avctx->extradata, avctx->extradata_size);
01930     if (!cfg.chan_config || cfg.chan_config > 7) {
01931         av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
01932         return -1;
01933     }
01934     s->frames = mp3Frames[cfg.chan_config];
01935     s->coff = chan_offset[cfg.chan_config];
01936     avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
01937 
01938     if (cfg.sample_rate < 16000)
01939         s->syncword = 0xffe00000;
01940     else
01941         s->syncword = 0xfff00000;
01942 
01943     /* Init the first mp3 decoder in standard way, so that all tables get builded
01944      * We replace avctx->priv_data with the context of the first decoder so that
01945      * decode_init() does not have to be changed.
01946      * Other decoders will be initialized here copying data from the first context
01947      */
01948     // Allocate zeroed memory for the first decoder context
01949     s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
01950     // Put decoder context in place to make init_decode() happy
01951     avctx->priv_data = s->mp3decctx[0];
01952     decode_init(avctx);
01953     // Restore mp3on4 context pointer
01954     avctx->priv_data = s;
01955     s->mp3decctx[0]->adu_mode = 1; // Set adu mode
01956 
01957     /* Create a separate codec/context for each frame (first is already ok).
01958      * Each frame is 1 or 2 channels - up to 5 frames allowed
01959      */
01960     for (i = 1; i < s->frames; i++) {
01961         s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
01962         s->mp3decctx[i]->adu_mode = 1;
01963         s->mp3decctx[i]->avctx = avctx;
01964     }
01965 
01966     return 0;
01967 }
01968 
01969 
01970 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
01971 {
01972     MP3On4DecodeContext *s = avctx->priv_data;
01973     int i;
01974 
01975     for (i = 0; i < s->frames; i++)
01976         av_free(s->mp3decctx[i]);
01977 
01978     return 0;
01979 }
01980 
01981 
01982 static int decode_frame_mp3on4(AVCodecContext * avctx,
01983                         void *data, int *data_size,
01984                         AVPacket *avpkt)
01985 {
01986     const uint8_t *buf = avpkt->data;
01987     int buf_size = avpkt->size;
01988     MP3On4DecodeContext *s = avctx->priv_data;
01989     MPADecodeContext *m;
01990     int fsize, len = buf_size, out_size = 0;
01991     uint32_t header;
01992     OUT_INT *out_samples = data;
01993     OUT_INT decoded_buf[MPA_FRAME_SIZE * MPA_MAX_CHANNELS];
01994     OUT_INT *outptr, *bp;
01995     int fr, j, n;
01996 
01997     if(*data_size < MPA_FRAME_SIZE * MPA_MAX_CHANNELS * s->frames * sizeof(OUT_INT))
01998         return -1;
01999 
02000     *data_size = 0;
02001     // Discard too short frames
02002     if (buf_size < HEADER_SIZE)
02003         return -1;
02004 
02005     // If only one decoder interleave is not needed
02006     outptr = s->frames == 1 ? out_samples : decoded_buf;
02007 
02008     avctx->bit_rate = 0;
02009 
02010     for (fr = 0; fr < s->frames; fr++) {
02011         fsize = AV_RB16(buf) >> 4;
02012         fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
02013         m = s->mp3decctx[fr];
02014         assert (m != NULL);
02015 
02016         header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
02017 
02018         if (ff_mpa_check_header(header) < 0) // Bad header, discard block
02019             break;
02020 
02021         ff_mpegaudio_decode_header((MPADecodeHeader *)m, header);
02022         out_size += mp_decode_frame(m, outptr, buf, fsize);
02023         buf += fsize;
02024         len -= fsize;
02025 
02026         if(s->frames > 1) {
02027             n = m->avctx->frame_size*m->nb_channels;
02028             /* interleave output data */
02029             bp = out_samples + s->coff[fr];
02030             if(m->nb_channels == 1) {
02031                 for(j = 0; j < n; j++) {
02032                     *bp = decoded_buf[j];
02033                     bp += avctx->channels;
02034                 }
02035             } else {
02036                 for(j = 0; j < n; j++) {
02037                     bp[0] = decoded_buf[j++];
02038                     bp[1] = decoded_buf[j];
02039                     bp += avctx->channels;
02040                 }
02041             }
02042         }
02043         avctx->bit_rate += m->bit_rate;
02044     }
02045 
02046     /* update codec info */
02047     avctx->sample_rate = s->mp3decctx[0]->sample_rate;
02048 
02049     *data_size = out_size;
02050     return buf_size;
02051 }
02052 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */
02053 
02054 #if !CONFIG_FLOAT
02055 #if CONFIG_MP1_DECODER
02056 AVCodec ff_mp1_decoder =
02057 {
02058     "mp1",
02059     AVMEDIA_TYPE_AUDIO,
02060     CODEC_ID_MP1,
02061     sizeof(MPADecodeContext),
02062     decode_init,
02063     NULL,
02064     NULL,
02065     decode_frame,
02066     CODEC_CAP_PARSE_ONLY,
02067     .flush= flush,
02068     .long_name= NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),
02069 };
02070 #endif
02071 #if CONFIG_MP2_DECODER
02072 AVCodec ff_mp2_decoder =
02073 {
02074     "mp2",
02075     AVMEDIA_TYPE_AUDIO,
02076     CODEC_ID_MP2,
02077     sizeof(MPADecodeContext),
02078     decode_init,
02079     NULL,
02080     NULL,
02081     decode_frame,
02082     CODEC_CAP_PARSE_ONLY,
02083     .flush= flush,
02084     .long_name= NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
02085 };
02086 #endif
02087 #if CONFIG_MP3_DECODER
02088 AVCodec ff_mp3_decoder =
02089 {
02090     "mp3",
02091     AVMEDIA_TYPE_AUDIO,
02092     CODEC_ID_MP3,
02093     sizeof(MPADecodeContext),
02094     decode_init,
02095     NULL,
02096     NULL,
02097     decode_frame,
02098     CODEC_CAP_PARSE_ONLY,
02099     .flush= flush,
02100     .long_name= NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),
02101 };
02102 #endif
02103 #if CONFIG_MP3ADU_DECODER
02104 AVCodec ff_mp3adu_decoder =
02105 {
02106     "mp3adu",
02107     AVMEDIA_TYPE_AUDIO,
02108     CODEC_ID_MP3ADU,
02109     sizeof(MPADecodeContext),
02110     decode_init,
02111     NULL,
02112     NULL,
02113     decode_frame_adu,
02114     CODEC_CAP_PARSE_ONLY,
02115     .flush= flush,
02116     .long_name= NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),
02117 };
02118 #endif
02119 #if CONFIG_MP3ON4_DECODER
02120 AVCodec ff_mp3on4_decoder =
02121 {
02122     "mp3on4",
02123     AVMEDIA_TYPE_AUDIO,
02124     CODEC_ID_MP3ON4,
02125     sizeof(MP3On4DecodeContext),
02126     decode_init_mp3on4,
02127     NULL,
02128     decode_close_mp3on4,
02129     decode_frame_mp3on4,
02130     .flush= flush,
02131     .long_name= NULL_IF_CONFIG_SMALL("MP3onMP4"),
02132 };
02133 #endif
02134 #endif