• Main Page
  • Related Pages
  • Modules
  • Data Structures
  • Files
  • Examples
  • File List
  • Globals

libavcodec/mpegaudioenc.c

Go to the documentation of this file.
00001 /*
00002  * The simplest mpeg audio layer 2 encoder
00003  * Copyright (c) 2000, 2001 Fabrice Bellard
00004  *
00005  * This file is part of Libav.
00006  *
00007  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
00019  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00020  */
00021 
00027 #include "avcodec.h"
00028 #include "internal.h"
00029 #include "put_bits.h"
00030 
00031 #define FRAC_BITS   15   /* fractional bits for sb_samples and dct */
00032 #define WFRAC_BITS  14   /* fractional bits for window */
00033 
00034 #include "mpegaudio.h"
00035 
00036 /* currently, cannot change these constants (need to modify
00037    quantization stage) */
00038 #define MUL(a,b) (((int64_t)(a) * (int64_t)(b)) >> FRAC_BITS)
00039 
00040 #define SAMPLES_BUF_SIZE 4096
00041 
00042 typedef struct MpegAudioContext {
00043     PutBitContext pb;
00044     int nb_channels;
00045     int lsf;           /* 1 if mpeg2 low bitrate selected */
00046     int bitrate_index; /* bit rate */
00047     int freq_index;
00048     int frame_size; /* frame size, in bits, without padding */
00049     /* padding computation */
00050     int frame_frac, frame_frac_incr, do_padding;
00051     short samples_buf[MPA_MAX_CHANNELS][SAMPLES_BUF_SIZE]; /* buffer for filter */
00052     int samples_offset[MPA_MAX_CHANNELS];       /* offset in samples_buf */
00053     int sb_samples[MPA_MAX_CHANNELS][3][12][SBLIMIT];
00054     unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3]; /* scale factors */
00055     /* code to group 3 scale factors */
00056     unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
00057     int sblimit; /* number of used subbands */
00058     const unsigned char *alloc_table;
00059 } MpegAudioContext;
00060 
00061 /* define it to use floats in quantization (I don't like floats !) */
00062 #define USE_FLOATS
00063 
00064 #include "mpegaudiodata.h"
00065 #include "mpegaudiotab.h"
00066 
00067 static av_cold int MPA_encode_init(AVCodecContext *avctx)
00068 {
00069     MpegAudioContext *s = avctx->priv_data;
00070     int freq = avctx->sample_rate;
00071     int bitrate = avctx->bit_rate;
00072     int channels = avctx->channels;
00073     int i, v, table;
00074     float a;
00075 
00076     if (channels <= 0 || channels > 2){
00077         av_log(avctx, AV_LOG_ERROR, "encoding %d channel(s) is not allowed in mp2\n", channels);
00078         return -1;
00079     }
00080     bitrate = bitrate / 1000;
00081     s->nb_channels = channels;
00082     avctx->frame_size = MPA_FRAME_SIZE;
00083 
00084     /* encoding freq */
00085     s->lsf = 0;
00086     for(i=0;i<3;i++) {
00087         if (avpriv_mpa_freq_tab[i] == freq)
00088             break;
00089         if ((avpriv_mpa_freq_tab[i] / 2) == freq) {
00090             s->lsf = 1;
00091             break;
00092         }
00093     }
00094     if (i == 3){
00095         av_log(avctx, AV_LOG_ERROR, "Sampling rate %d is not allowed in mp2\n", freq);
00096         return -1;
00097     }
00098     s->freq_index = i;
00099 
00100     /* encoding bitrate & frequency */
00101     for(i=0;i<15;i++) {
00102         if (avpriv_mpa_bitrate_tab[s->lsf][1][i] == bitrate)
00103             break;
00104     }
00105     if (i == 15){
00106         av_log(avctx, AV_LOG_ERROR, "bitrate %d is not allowed in mp2\n", bitrate);
00107         return -1;
00108     }
00109     s->bitrate_index = i;
00110 
00111     /* compute total header size & pad bit */
00112 
00113     a = (float)(bitrate * 1000 * MPA_FRAME_SIZE) / (freq * 8.0);
00114     s->frame_size = ((int)a) * 8;
00115 
00116     /* frame fractional size to compute padding */
00117     s->frame_frac = 0;
00118     s->frame_frac_incr = (int)((a - floor(a)) * 65536.0);
00119 
00120     /* select the right allocation table */
00121     table = ff_mpa_l2_select_table(bitrate, s->nb_channels, freq, s->lsf);
00122 
00123     /* number of used subbands */
00124     s->sblimit = ff_mpa_sblimit_table[table];
00125     s->alloc_table = ff_mpa_alloc_tables[table];
00126 
00127     av_dlog(avctx, "%d kb/s, %d Hz, frame_size=%d bits, table=%d, padincr=%x\n",
00128             bitrate, freq, s->frame_size, table, s->frame_frac_incr);
00129 
00130     for(i=0;i<s->nb_channels;i++)
00131         s->samples_offset[i] = 0;
00132 
00133     for(i=0;i<257;i++) {
00134         int v;
00135         v = ff_mpa_enwindow[i];
00136 #if WFRAC_BITS != 16
00137         v = (v + (1 << (16 - WFRAC_BITS - 1))) >> (16 - WFRAC_BITS);
00138 #endif
00139         filter_bank[i] = v;
00140         if ((i & 63) != 0)
00141             v = -v;
00142         if (i != 0)
00143             filter_bank[512 - i] = v;
00144     }
00145 
00146     for(i=0;i<64;i++) {
00147         v = (int)(pow(2.0, (3 - i) / 3.0) * (1 << 20));
00148         if (v <= 0)
00149             v = 1;
00150         scale_factor_table[i] = v;
00151 #ifdef USE_FLOATS
00152         scale_factor_inv_table[i] = pow(2.0, -(3 - i) / 3.0) / (float)(1 << 20);
00153 #else
00154 #define P 15
00155         scale_factor_shift[i] = 21 - P - (i / 3);
00156         scale_factor_mult[i] = (1 << P) * pow(2.0, (i % 3) / 3.0);
00157 #endif
00158     }
00159     for(i=0;i<128;i++) {
00160         v = i - 64;
00161         if (v <= -3)
00162             v = 0;
00163         else if (v < 0)
00164             v = 1;
00165         else if (v == 0)
00166             v = 2;
00167         else if (v < 3)
00168             v = 3;
00169         else
00170             v = 4;
00171         scale_diff_table[i] = v;
00172     }
00173 
00174     for(i=0;i<17;i++) {
00175         v = ff_mpa_quant_bits[i];
00176         if (v < 0)
00177             v = -v;
00178         else
00179             v = v * 3;
00180         total_quant_bits[i] = 12 * v;
00181     }
00182 
00183     avctx->coded_frame= avcodec_alloc_frame();
00184     avctx->coded_frame->key_frame= 1;
00185 
00186     return 0;
00187 }
00188 
00189 /* 32 point floating point IDCT without 1/sqrt(2) coef zero scaling */
00190 static void idct32(int *out, int *tab)
00191 {
00192     int i, j;
00193     int *t, *t1, xr;
00194     const int *xp = costab32;
00195 
00196     for(j=31;j>=3;j-=2) tab[j] += tab[j - 2];
00197 
00198     t = tab + 30;
00199     t1 = tab + 2;
00200     do {
00201         t[0] += t[-4];
00202         t[1] += t[1 - 4];
00203         t -= 4;
00204     } while (t != t1);
00205 
00206     t = tab + 28;
00207     t1 = tab + 4;
00208     do {
00209         t[0] += t[-8];
00210         t[1] += t[1-8];
00211         t[2] += t[2-8];
00212         t[3] += t[3-8];
00213         t -= 8;
00214     } while (t != t1);
00215 
00216     t = tab;
00217     t1 = tab + 32;
00218     do {
00219         t[ 3] = -t[ 3];
00220         t[ 6] = -t[ 6];
00221 
00222         t[11] = -t[11];
00223         t[12] = -t[12];
00224         t[13] = -t[13];
00225         t[15] = -t[15];
00226         t += 16;
00227     } while (t != t1);
00228 
00229 
00230     t = tab;
00231     t1 = tab + 8;
00232     do {
00233         int x1, x2, x3, x4;
00234 
00235         x3 = MUL(t[16], FIX(SQRT2*0.5));
00236         x4 = t[0] - x3;
00237         x3 = t[0] + x3;
00238 
00239         x2 = MUL(-(t[24] + t[8]), FIX(SQRT2*0.5));
00240         x1 = MUL((t[8] - x2), xp[0]);
00241         x2 = MUL((t[8] + x2), xp[1]);
00242 
00243         t[ 0] = x3 + x1;
00244         t[ 8] = x4 - x2;
00245         t[16] = x4 + x2;
00246         t[24] = x3 - x1;
00247         t++;
00248     } while (t != t1);
00249 
00250     xp += 2;
00251     t = tab;
00252     t1 = tab + 4;
00253     do {
00254         xr = MUL(t[28],xp[0]);
00255         t[28] = (t[0] - xr);
00256         t[0] = (t[0] + xr);
00257 
00258         xr = MUL(t[4],xp[1]);
00259         t[ 4] = (t[24] - xr);
00260         t[24] = (t[24] + xr);
00261 
00262         xr = MUL(t[20],xp[2]);
00263         t[20] = (t[8] - xr);
00264         t[ 8] = (t[8] + xr);
00265 
00266         xr = MUL(t[12],xp[3]);
00267         t[12] = (t[16] - xr);
00268         t[16] = (t[16] + xr);
00269         t++;
00270     } while (t != t1);
00271     xp += 4;
00272 
00273     for (i = 0; i < 4; i++) {
00274         xr = MUL(tab[30-i*4],xp[0]);
00275         tab[30-i*4] = (tab[i*4] - xr);
00276         tab[   i*4] = (tab[i*4] + xr);
00277 
00278         xr = MUL(tab[ 2+i*4],xp[1]);
00279         tab[ 2+i*4] = (tab[28-i*4] - xr);
00280         tab[28-i*4] = (tab[28-i*4] + xr);
00281 
00282         xr = MUL(tab[31-i*4],xp[0]);
00283         tab[31-i*4] = (tab[1+i*4] - xr);
00284         tab[ 1+i*4] = (tab[1+i*4] + xr);
00285 
00286         xr = MUL(tab[ 3+i*4],xp[1]);
00287         tab[ 3+i*4] = (tab[29-i*4] - xr);
00288         tab[29-i*4] = (tab[29-i*4] + xr);
00289 
00290         xp += 2;
00291     }
00292 
00293     t = tab + 30;
00294     t1 = tab + 1;
00295     do {
00296         xr = MUL(t1[0], *xp);
00297         t1[0] = (t[0] - xr);
00298         t[0] = (t[0] + xr);
00299         t -= 2;
00300         t1 += 2;
00301         xp++;
00302     } while (t >= tab);
00303 
00304     for(i=0;i<32;i++) {
00305         out[i] = tab[bitinv32[i]];
00306     }
00307 }
00308 
00309 #define WSHIFT (WFRAC_BITS + 15 - FRAC_BITS)
00310 
00311 static void filter(MpegAudioContext *s, int ch, const short *samples, int incr)
00312 {
00313     short *p, *q;
00314     int sum, offset, i, j;
00315     int tmp[64];
00316     int tmp1[32];
00317     int *out;
00318 
00319     offset = s->samples_offset[ch];
00320     out = &s->sb_samples[ch][0][0][0];
00321     for(j=0;j<36;j++) {
00322         /* 32 samples at once */
00323         for(i=0;i<32;i++) {
00324             s->samples_buf[ch][offset + (31 - i)] = samples[0];
00325             samples += incr;
00326         }
00327 
00328         /* filter */
00329         p = s->samples_buf[ch] + offset;
00330         q = filter_bank;
00331         /* maxsum = 23169 */
00332         for(i=0;i<64;i++) {
00333             sum = p[0*64] * q[0*64];
00334             sum += p[1*64] * q[1*64];
00335             sum += p[2*64] * q[2*64];
00336             sum += p[3*64] * q[3*64];
00337             sum += p[4*64] * q[4*64];
00338             sum += p[5*64] * q[5*64];
00339             sum += p[6*64] * q[6*64];
00340             sum += p[7*64] * q[7*64];
00341             tmp[i] = sum;
00342             p++;
00343             q++;
00344         }
00345         tmp1[0] = tmp[16] >> WSHIFT;
00346         for( i=1; i<=16; i++ ) tmp1[i] = (tmp[i+16]+tmp[16-i]) >> WSHIFT;
00347         for( i=17; i<=31; i++ ) tmp1[i] = (tmp[i+16]-tmp[80-i]) >> WSHIFT;
00348 
00349         idct32(out, tmp1);
00350 
00351         /* advance of 32 samples */
00352         offset -= 32;
00353         out += 32;
00354         /* handle the wrap around */
00355         if (offset < 0) {
00356             memmove(s->samples_buf[ch] + SAMPLES_BUF_SIZE - (512 - 32),
00357                     s->samples_buf[ch], (512 - 32) * 2);
00358             offset = SAMPLES_BUF_SIZE - 512;
00359         }
00360     }
00361     s->samples_offset[ch] = offset;
00362 }
00363 
00364 static void compute_scale_factors(unsigned char scale_code[SBLIMIT],
00365                                   unsigned char scale_factors[SBLIMIT][3],
00366                                   int sb_samples[3][12][SBLIMIT],
00367                                   int sblimit)
00368 {
00369     int *p, vmax, v, n, i, j, k, code;
00370     int index, d1, d2;
00371     unsigned char *sf = &scale_factors[0][0];
00372 
00373     for(j=0;j<sblimit;j++) {
00374         for(i=0;i<3;i++) {
00375             /* find the max absolute value */
00376             p = &sb_samples[i][0][j];
00377             vmax = abs(*p);
00378             for(k=1;k<12;k++) {
00379                 p += SBLIMIT;
00380                 v = abs(*p);
00381                 if (v > vmax)
00382                     vmax = v;
00383             }
00384             /* compute the scale factor index using log 2 computations */
00385             if (vmax > 1) {
00386                 n = av_log2(vmax);
00387                 /* n is the position of the MSB of vmax. now
00388                    use at most 2 compares to find the index */
00389                 index = (21 - n) * 3 - 3;
00390                 if (index >= 0) {
00391                     while (vmax <= scale_factor_table[index+1])
00392                         index++;
00393                 } else {
00394                     index = 0; /* very unlikely case of overflow */
00395                 }
00396             } else {
00397                 index = 62; /* value 63 is not allowed */
00398             }
00399 
00400             av_dlog(NULL, "%2d:%d in=%x %x %d\n",
00401                     j, i, vmax, scale_factor_table[index], index);
00402             /* store the scale factor */
00403             assert(index >=0 && index <= 63);
00404             sf[i] = index;
00405         }
00406 
00407         /* compute the transmission factor : look if the scale factors
00408            are close enough to each other */
00409         d1 = scale_diff_table[sf[0] - sf[1] + 64];
00410         d2 = scale_diff_table[sf[1] - sf[2] + 64];
00411 
00412         /* handle the 25 cases */
00413         switch(d1 * 5 + d2) {
00414         case 0*5+0:
00415         case 0*5+4:
00416         case 3*5+4:
00417         case 4*5+0:
00418         case 4*5+4:
00419             code = 0;
00420             break;
00421         case 0*5+1:
00422         case 0*5+2:
00423         case 4*5+1:
00424         case 4*5+2:
00425             code = 3;
00426             sf[2] = sf[1];
00427             break;
00428         case 0*5+3:
00429         case 4*5+3:
00430             code = 3;
00431             sf[1] = sf[2];
00432             break;
00433         case 1*5+0:
00434         case 1*5+4:
00435         case 2*5+4:
00436             code = 1;
00437             sf[1] = sf[0];
00438             break;
00439         case 1*5+1:
00440         case 1*5+2:
00441         case 2*5+0:
00442         case 2*5+1:
00443         case 2*5+2:
00444             code = 2;
00445             sf[1] = sf[2] = sf[0];
00446             break;
00447         case 2*5+3:
00448         case 3*5+3:
00449             code = 2;
00450             sf[0] = sf[1] = sf[2];
00451             break;
00452         case 3*5+0:
00453         case 3*5+1:
00454         case 3*5+2:
00455             code = 2;
00456             sf[0] = sf[2] = sf[1];
00457             break;
00458         case 1*5+3:
00459             code = 2;
00460             if (sf[0] > sf[2])
00461               sf[0] = sf[2];
00462             sf[1] = sf[2] = sf[0];
00463             break;
00464         default:
00465             assert(0); //cannot happen
00466             code = 0;           /* kill warning */
00467         }
00468 
00469         av_dlog(NULL, "%d: %2d %2d %2d %d %d -> %d\n", j,
00470                 sf[0], sf[1], sf[2], d1, d2, code);
00471         scale_code[j] = code;
00472         sf += 3;
00473     }
00474 }
00475 
00476 /* The most important function : psycho acoustic module. In this
00477    encoder there is basically none, so this is the worst you can do,
00478    but also this is the simpler. */
00479 static void psycho_acoustic_model(MpegAudioContext *s, short smr[SBLIMIT])
00480 {
00481     int i;
00482 
00483     for(i=0;i<s->sblimit;i++) {
00484         smr[i] = (int)(fixed_smr[i] * 10);
00485     }
00486 }
00487 
00488 
00489 #define SB_NOTALLOCATED  0
00490 #define SB_ALLOCATED     1
00491 #define SB_NOMORE        2
00492 
00493 /* Try to maximize the smr while using a number of bits inferior to
00494    the frame size. I tried to make the code simpler, faster and
00495    smaller than other encoders :-) */
00496 static void compute_bit_allocation(MpegAudioContext *s,
00497                                    short smr1[MPA_MAX_CHANNELS][SBLIMIT],
00498                                    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
00499                                    int *padding)
00500 {
00501     int i, ch, b, max_smr, max_ch, max_sb, current_frame_size, max_frame_size;
00502     int incr;
00503     short smr[MPA_MAX_CHANNELS][SBLIMIT];
00504     unsigned char subband_status[MPA_MAX_CHANNELS][SBLIMIT];
00505     const unsigned char *alloc;
00506 
00507     memcpy(smr, smr1, s->nb_channels * sizeof(short) * SBLIMIT);
00508     memset(subband_status, SB_NOTALLOCATED, s->nb_channels * SBLIMIT);
00509     memset(bit_alloc, 0, s->nb_channels * SBLIMIT);
00510 
00511     /* compute frame size and padding */
00512     max_frame_size = s->frame_size;
00513     s->frame_frac += s->frame_frac_incr;
00514     if (s->frame_frac >= 65536) {
00515         s->frame_frac -= 65536;
00516         s->do_padding = 1;
00517         max_frame_size += 8;
00518     } else {
00519         s->do_padding = 0;
00520     }
00521 
00522     /* compute the header + bit alloc size */
00523     current_frame_size = 32;
00524     alloc = s->alloc_table;
00525     for(i=0;i<s->sblimit;i++) {
00526         incr = alloc[0];
00527         current_frame_size += incr * s->nb_channels;
00528         alloc += 1 << incr;
00529     }
00530     for(;;) {
00531         /* look for the subband with the largest signal to mask ratio */
00532         max_sb = -1;
00533         max_ch = -1;
00534         max_smr = INT_MIN;
00535         for(ch=0;ch<s->nb_channels;ch++) {
00536             for(i=0;i<s->sblimit;i++) {
00537                 if (smr[ch][i] > max_smr && subband_status[ch][i] != SB_NOMORE) {
00538                     max_smr = smr[ch][i];
00539                     max_sb = i;
00540                     max_ch = ch;
00541                 }
00542             }
00543         }
00544         if (max_sb < 0)
00545             break;
00546         av_dlog(NULL, "current=%d max=%d max_sb=%d max_ch=%d alloc=%d\n",
00547                 current_frame_size, max_frame_size, max_sb, max_ch,
00548                 bit_alloc[max_ch][max_sb]);
00549 
00550         /* find alloc table entry (XXX: not optimal, should use
00551            pointer table) */
00552         alloc = s->alloc_table;
00553         for(i=0;i<max_sb;i++) {
00554             alloc += 1 << alloc[0];
00555         }
00556 
00557         if (subband_status[max_ch][max_sb] == SB_NOTALLOCATED) {
00558             /* nothing was coded for this band: add the necessary bits */
00559             incr = 2 + nb_scale_factors[s->scale_code[max_ch][max_sb]] * 6;
00560             incr += total_quant_bits[alloc[1]];
00561         } else {
00562             /* increments bit allocation */
00563             b = bit_alloc[max_ch][max_sb];
00564             incr = total_quant_bits[alloc[b + 1]] -
00565                 total_quant_bits[alloc[b]];
00566         }
00567 
00568         if (current_frame_size + incr <= max_frame_size) {
00569             /* can increase size */
00570             b = ++bit_alloc[max_ch][max_sb];
00571             current_frame_size += incr;
00572             /* decrease smr by the resolution we added */
00573             smr[max_ch][max_sb] = smr1[max_ch][max_sb] - quant_snr[alloc[b]];
00574             /* max allocation size reached ? */
00575             if (b == ((1 << alloc[0]) - 1))
00576                 subband_status[max_ch][max_sb] = SB_NOMORE;
00577             else
00578                 subband_status[max_ch][max_sb] = SB_ALLOCATED;
00579         } else {
00580             /* cannot increase the size of this subband */
00581             subband_status[max_ch][max_sb] = SB_NOMORE;
00582         }
00583     }
00584     *padding = max_frame_size - current_frame_size;
00585     assert(*padding >= 0);
00586 }
00587 
00588 /*
00589  * Output the mpeg audio layer 2 frame. Note how the code is small
00590  * compared to other encoders :-)
00591  */
00592 static void encode_frame(MpegAudioContext *s,
00593                          unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT],
00594                          int padding)
00595 {
00596     int i, j, k, l, bit_alloc_bits, b, ch;
00597     unsigned char *sf;
00598     int q[3];
00599     PutBitContext *p = &s->pb;
00600 
00601     /* header */
00602 
00603     put_bits(p, 12, 0xfff);
00604     put_bits(p, 1, 1 - s->lsf); /* 1 = mpeg1 ID, 0 = mpeg2 lsf ID */
00605     put_bits(p, 2, 4-2);  /* layer 2 */
00606     put_bits(p, 1, 1); /* no error protection */
00607     put_bits(p, 4, s->bitrate_index);
00608     put_bits(p, 2, s->freq_index);
00609     put_bits(p, 1, s->do_padding); /* use padding */
00610     put_bits(p, 1, 0);             /* private_bit */
00611     put_bits(p, 2, s->nb_channels == 2 ? MPA_STEREO : MPA_MONO);
00612     put_bits(p, 2, 0); /* mode_ext */
00613     put_bits(p, 1, 0); /* no copyright */
00614     put_bits(p, 1, 1); /* original */
00615     put_bits(p, 2, 0); /* no emphasis */
00616 
00617     /* bit allocation */
00618     j = 0;
00619     for(i=0;i<s->sblimit;i++) {
00620         bit_alloc_bits = s->alloc_table[j];
00621         for(ch=0;ch<s->nb_channels;ch++) {
00622             put_bits(p, bit_alloc_bits, bit_alloc[ch][i]);
00623         }
00624         j += 1 << bit_alloc_bits;
00625     }
00626 
00627     /* scale codes */
00628     for(i=0;i<s->sblimit;i++) {
00629         for(ch=0;ch<s->nb_channels;ch++) {
00630             if (bit_alloc[ch][i])
00631                 put_bits(p, 2, s->scale_code[ch][i]);
00632         }
00633     }
00634 
00635     /* scale factors */
00636     for(i=0;i<s->sblimit;i++) {
00637         for(ch=0;ch<s->nb_channels;ch++) {
00638             if (bit_alloc[ch][i]) {
00639                 sf = &s->scale_factors[ch][i][0];
00640                 switch(s->scale_code[ch][i]) {
00641                 case 0:
00642                     put_bits(p, 6, sf[0]);
00643                     put_bits(p, 6, sf[1]);
00644                     put_bits(p, 6, sf[2]);
00645                     break;
00646                 case 3:
00647                 case 1:
00648                     put_bits(p, 6, sf[0]);
00649                     put_bits(p, 6, sf[2]);
00650                     break;
00651                 case 2:
00652                     put_bits(p, 6, sf[0]);
00653                     break;
00654                 }
00655             }
00656         }
00657     }
00658 
00659     /* quantization & write sub band samples */
00660 
00661     for(k=0;k<3;k++) {
00662         for(l=0;l<12;l+=3) {
00663             j = 0;
00664             for(i=0;i<s->sblimit;i++) {
00665                 bit_alloc_bits = s->alloc_table[j];
00666                 for(ch=0;ch<s->nb_channels;ch++) {
00667                     b = bit_alloc[ch][i];
00668                     if (b) {
00669                         int qindex, steps, m, sample, bits;
00670                         /* we encode 3 sub band samples of the same sub band at a time */
00671                         qindex = s->alloc_table[j+b];
00672                         steps = ff_mpa_quant_steps[qindex];
00673                         for(m=0;m<3;m++) {
00674                             sample = s->sb_samples[ch][k][l + m][i];
00675                             /* divide by scale factor */
00676 #ifdef USE_FLOATS
00677                             {
00678                                 float a;
00679                                 a = (float)sample * scale_factor_inv_table[s->scale_factors[ch][i][k]];
00680                                 q[m] = (int)((a + 1.0) * steps * 0.5);
00681                             }
00682 #else
00683                             {
00684                                 int q1, e, shift, mult;
00685                                 e = s->scale_factors[ch][i][k];
00686                                 shift = scale_factor_shift[e];
00687                                 mult = scale_factor_mult[e];
00688 
00689                                 /* normalize to P bits */
00690                                 if (shift < 0)
00691                                     q1 = sample << (-shift);
00692                                 else
00693                                     q1 = sample >> shift;
00694                                 q1 = (q1 * mult) >> P;
00695                                 q[m] = ((q1 + (1 << P)) * steps) >> (P + 1);
00696                             }
00697 #endif
00698                             if (q[m] >= steps)
00699                                 q[m] = steps - 1;
00700                             assert(q[m] >= 0 && q[m] < steps);
00701                         }
00702                         bits = ff_mpa_quant_bits[qindex];
00703                         if (bits < 0) {
00704                             /* group the 3 values to save bits */
00705                             put_bits(p, -bits,
00706                                      q[0] + steps * (q[1] + steps * q[2]));
00707                         } else {
00708                             put_bits(p, bits, q[0]);
00709                             put_bits(p, bits, q[1]);
00710                             put_bits(p, bits, q[2]);
00711                         }
00712                     }
00713                 }
00714                 /* next subband in alloc table */
00715                 j += 1 << bit_alloc_bits;
00716             }
00717         }
00718     }
00719 
00720     /* padding */
00721     for(i=0;i<padding;i++)
00722         put_bits(p, 1, 0);
00723 
00724     /* flush */
00725     flush_put_bits(p);
00726 }
00727 
00728 static int MPA_encode_frame(AVCodecContext *avctx,
00729                             unsigned char *frame, int buf_size, void *data)
00730 {
00731     MpegAudioContext *s = avctx->priv_data;
00732     const short *samples = data;
00733     short smr[MPA_MAX_CHANNELS][SBLIMIT];
00734     unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
00735     int padding, i;
00736 
00737     for(i=0;i<s->nb_channels;i++) {
00738         filter(s, i, samples + i, s->nb_channels);
00739     }
00740 
00741     for(i=0;i<s->nb_channels;i++) {
00742         compute_scale_factors(s->scale_code[i], s->scale_factors[i],
00743                               s->sb_samples[i], s->sblimit);
00744     }
00745     for(i=0;i<s->nb_channels;i++) {
00746         psycho_acoustic_model(s, smr[i]);
00747     }
00748     compute_bit_allocation(s, smr, bit_alloc, &padding);
00749 
00750     init_put_bits(&s->pb, frame, MPA_MAX_CODED_FRAME_SIZE);
00751 
00752     encode_frame(s, bit_alloc, padding);
00753 
00754     return put_bits_ptr(&s->pb) - s->pb.buf;
00755 }
00756 
00757 static av_cold int MPA_encode_close(AVCodecContext *avctx)
00758 {
00759     av_freep(&avctx->coded_frame);
00760     return 0;
00761 }
00762 
00763 static const AVCodecDefault mp2_defaults[] = {
00764     { "b",    "128k" },
00765     { NULL },
00766 };
00767 
00768 AVCodec ff_mp2_encoder = {
00769     .name           = "mp2",
00770     .type           = AVMEDIA_TYPE_AUDIO,
00771     .id             = CODEC_ID_MP2,
00772     .priv_data_size = sizeof(MpegAudioContext),
00773     .init           = MPA_encode_init,
00774     .encode         = MPA_encode_frame,
00775     .close          = MPA_encode_close,
00776     .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
00777     .supported_samplerates= (const int[]){44100, 48000,  32000, 22050, 24000, 16000, 0},
00778     .long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
00779     .defaults       = mp2_defaults,
00780 };
Generated on Thu Jul 11 2013 15:38:20 for Libav by doxygen 1.7.1