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

libavcodec/sonic.c

Go to the documentation of this file.
00001 /*
00002  * Simple free lossless/lossy audio codec
00003  * Copyright (c) 2004 Alex Beregszaszi
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 #include "avcodec.h"
00022 #include "get_bits.h"
00023 #include "golomb.h"
00024 
00040 #define MAX_CHANNELS 2
00041 
00042 #define MID_SIDE 0
00043 #define LEFT_SIDE 1
00044 #define RIGHT_SIDE 2
00045 
00046 typedef struct SonicContext {
00047     AVFrame frame;
00048     int lossless, decorrelation;
00049 
00050     int num_taps, downsampling;
00051     double quantization;
00052 
00053     int channels, samplerate, block_align, frame_size;
00054 
00055     int *tap_quant;
00056     int *int_samples;
00057     int *coded_samples[MAX_CHANNELS];
00058 
00059     // for encoding
00060     int *tail;
00061     int tail_size;
00062     int *window;
00063     int window_size;
00064 
00065     // for decoding
00066     int *predictor_k;
00067     int *predictor_state[MAX_CHANNELS];
00068 } SonicContext;
00069 
00070 #define LATTICE_SHIFT   10
00071 #define SAMPLE_SHIFT    4
00072 #define LATTICE_FACTOR  (1 << LATTICE_SHIFT)
00073 #define SAMPLE_FACTOR   (1 << SAMPLE_SHIFT)
00074 
00075 #define BASE_QUANT      0.6
00076 #define RATE_VARIATION  3.0
00077 
00078 static inline int divide(int a, int b)
00079 {
00080     if (a < 0)
00081         return -( (-a + b/2)/b );
00082     else
00083         return (a + b/2)/b;
00084 }
00085 
00086 static inline int shift(int a,int b)
00087 {
00088     return (a+(1<<(b-1))) >> b;
00089 }
00090 
00091 static inline int shift_down(int a,int b)
00092 {
00093     return (a>>b)+((a<0)?1:0);
00094 }
00095 
00096 #if 1
00097 static inline int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
00098 {
00099     int i;
00100 
00101     for (i = 0; i < entries; i++)
00102         set_se_golomb(pb, buf[i]);
00103 
00104     return 1;
00105 }
00106 
00107 static inline int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
00108 {
00109     int i;
00110 
00111     for (i = 0; i < entries; i++)
00112         buf[i] = get_se_golomb(gb);
00113 
00114     return 1;
00115 }
00116 
00117 #else
00118 
00119 #define ADAPT_LEVEL 8
00120 
00121 static int bits_to_store(uint64_t x)
00122 {
00123     int res = 0;
00124 
00125     while(x)
00126     {
00127         res++;
00128         x >>= 1;
00129     }
00130     return res;
00131 }
00132 
00133 static void write_uint_max(PutBitContext *pb, unsigned int value, unsigned int max)
00134 {
00135     int i, bits;
00136 
00137     if (!max)
00138         return;
00139 
00140     bits = bits_to_store(max);
00141 
00142     for (i = 0; i < bits-1; i++)
00143         put_bits(pb, 1, value & (1 << i));
00144 
00145     if ( (value | (1 << (bits-1))) <= max)
00146         put_bits(pb, 1, value & (1 << (bits-1)));
00147 }
00148 
00149 static unsigned int read_uint_max(GetBitContext *gb, int max)
00150 {
00151     int i, bits, value = 0;
00152 
00153     if (!max)
00154         return 0;
00155 
00156     bits = bits_to_store(max);
00157 
00158     for (i = 0; i < bits-1; i++)
00159         if (get_bits1(gb))
00160             value += 1 << i;
00161 
00162     if ( (value | (1<<(bits-1))) <= max)
00163         if (get_bits1(gb))
00164             value += 1 << (bits-1);
00165 
00166     return value;
00167 }
00168 
00169 static int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
00170 {
00171     int i, j, x = 0, low_bits = 0, max = 0;
00172     int step = 256, pos = 0, dominant = 0, any = 0;
00173     int *copy, *bits;
00174 
00175     copy = av_mallocz(4* entries);
00176     if (!copy)
00177         return -1;
00178 
00179     if (base_2_part)
00180     {
00181         int energy = 0;
00182 
00183         for (i = 0; i < entries; i++)
00184             energy += abs(buf[i]);
00185 
00186         low_bits = bits_to_store(energy / (entries * 2));
00187         if (low_bits > 15)
00188             low_bits = 15;
00189 
00190         put_bits(pb, 4, low_bits);
00191     }
00192 
00193     for (i = 0; i < entries; i++)
00194     {
00195         put_bits(pb, low_bits, abs(buf[i]));
00196         copy[i] = abs(buf[i]) >> low_bits;
00197         if (copy[i] > max)
00198             max = abs(copy[i]);
00199     }
00200 
00201     bits = av_mallocz(4* entries*max);
00202     if (!bits)
00203     {
00204 //        av_free(copy);
00205         return -1;
00206     }
00207 
00208     for (i = 0; i <= max; i++)
00209     {
00210         for (j = 0; j < entries; j++)
00211             if (copy[j] >= i)
00212                 bits[x++] = copy[j] > i;
00213     }
00214 
00215     // store bitstream
00216     while (pos < x)
00217     {
00218         int steplet = step >> 8;
00219 
00220         if (pos + steplet > x)
00221             steplet = x - pos;
00222 
00223         for (i = 0; i < steplet; i++)
00224             if (bits[i+pos] != dominant)
00225                 any = 1;
00226 
00227         put_bits(pb, 1, any);
00228 
00229         if (!any)
00230         {
00231             pos += steplet;
00232             step += step / ADAPT_LEVEL;
00233         }
00234         else
00235         {
00236             int interloper = 0;
00237 
00238             while (((pos + interloper) < x) && (bits[pos + interloper] == dominant))
00239                 interloper++;
00240 
00241             // note change
00242             write_uint_max(pb, interloper, (step >> 8) - 1);
00243 
00244             pos += interloper + 1;
00245             step -= step / ADAPT_LEVEL;
00246         }
00247 
00248         if (step < 256)
00249         {
00250             step = 65536 / step;
00251             dominant = !dominant;
00252         }
00253     }
00254 
00255     // store signs
00256     for (i = 0; i < entries; i++)
00257         if (buf[i])
00258             put_bits(pb, 1, buf[i] < 0);
00259 
00260 //    av_free(bits);
00261 //    av_free(copy);
00262 
00263     return 0;
00264 }
00265 
00266 static int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
00267 {
00268     int i, low_bits = 0, x = 0;
00269     int n_zeros = 0, step = 256, dominant = 0;
00270     int pos = 0, level = 0;
00271     int *bits = av_mallocz(4* entries);
00272 
00273     if (!bits)
00274         return -1;
00275 
00276     if (base_2_part)
00277     {
00278         low_bits = get_bits(gb, 4);
00279 
00280         if (low_bits)
00281             for (i = 0; i < entries; i++)
00282                 buf[i] = get_bits(gb, low_bits);
00283     }
00284 
00285 //    av_log(NULL, AV_LOG_INFO, "entries: %d, low bits: %d\n", entries, low_bits);
00286 
00287     while (n_zeros < entries)
00288     {
00289         int steplet = step >> 8;
00290 
00291         if (!get_bits1(gb))
00292         {
00293             for (i = 0; i < steplet; i++)
00294                 bits[x++] = dominant;
00295 
00296             if (!dominant)
00297                 n_zeros += steplet;
00298 
00299             step += step / ADAPT_LEVEL;
00300         }
00301         else
00302         {
00303             int actual_run = read_uint_max(gb, steplet-1);
00304 
00305 //            av_log(NULL, AV_LOG_INFO, "actual run: %d\n", actual_run);
00306 
00307             for (i = 0; i < actual_run; i++)
00308                 bits[x++] = dominant;
00309 
00310             bits[x++] = !dominant;
00311 
00312             if (!dominant)
00313                 n_zeros += actual_run;
00314             else
00315                 n_zeros++;
00316 
00317             step -= step / ADAPT_LEVEL;
00318         }
00319 
00320         if (step < 256)
00321         {
00322             step = 65536 / step;
00323             dominant = !dominant;
00324         }
00325     }
00326 
00327     // reconstruct unsigned values
00328     n_zeros = 0;
00329     for (i = 0; n_zeros < entries; i++)
00330     {
00331         while(1)
00332         {
00333             if (pos >= entries)
00334             {
00335                 pos = 0;
00336                 level += 1 << low_bits;
00337             }
00338 
00339             if (buf[pos] >= level)
00340                 break;
00341 
00342             pos++;
00343         }
00344 
00345         if (bits[i])
00346             buf[pos] += 1 << low_bits;
00347         else
00348             n_zeros++;
00349 
00350         pos++;
00351     }
00352 //    av_free(bits);
00353 
00354     // read signs
00355     for (i = 0; i < entries; i++)
00356         if (buf[i] && get_bits1(gb))
00357             buf[i] = -buf[i];
00358 
00359 //    av_log(NULL, AV_LOG_INFO, "zeros: %d pos: %d\n", n_zeros, pos);
00360 
00361     return 0;
00362 }
00363 #endif
00364 
00365 static void predictor_init_state(int *k, int *state, int order)
00366 {
00367     int i;
00368 
00369     for (i = order-2; i >= 0; i--)
00370     {
00371         int j, p, x = state[i];
00372 
00373         for (j = 0, p = i+1; p < order; j++,p++)
00374             {
00375             int tmp = x + shift_down(k[j] * state[p], LATTICE_SHIFT);
00376             state[p] += shift_down(k[j]*x, LATTICE_SHIFT);
00377             x = tmp;
00378         }
00379     }
00380 }
00381 
00382 static int predictor_calc_error(int *k, int *state, int order, int error)
00383 {
00384     int i, x = error - shift_down(k[order-1] * state[order-1], LATTICE_SHIFT);
00385 
00386 #if 1
00387     int *k_ptr = &(k[order-2]),
00388         *state_ptr = &(state[order-2]);
00389     for (i = order-2; i >= 0; i--, k_ptr--, state_ptr--)
00390     {
00391         int k_value = *k_ptr, state_value = *state_ptr;
00392         x -= shift_down(k_value * state_value, LATTICE_SHIFT);
00393         state_ptr[1] = state_value + shift_down(k_value * x, LATTICE_SHIFT);
00394     }
00395 #else
00396     for (i = order-2; i >= 0; i--)
00397     {
00398         x -= shift_down(k[i] * state[i], LATTICE_SHIFT);
00399         state[i+1] = state[i] + shift_down(k[i] * x, LATTICE_SHIFT);
00400     }
00401 #endif
00402 
00403     // don't drift too far, to avoid overflows
00404     if (x >  (SAMPLE_FACTOR<<16)) x =  (SAMPLE_FACTOR<<16);
00405     if (x < -(SAMPLE_FACTOR<<16)) x = -(SAMPLE_FACTOR<<16);
00406 
00407     state[0] = x;
00408 
00409     return x;
00410 }
00411 
00412 #if CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER
00413 // Heavily modified Levinson-Durbin algorithm which
00414 // copes better with quantization, and calculates the
00415 // actual whitened result as it goes.
00416 
00417 static void modified_levinson_durbin(int *window, int window_entries,
00418         int *out, int out_entries, int channels, int *tap_quant)
00419 {
00420     int i;
00421     int *state = av_mallocz(4* window_entries);
00422 
00423     memcpy(state, window, 4* window_entries);
00424 
00425     for (i = 0; i < out_entries; i++)
00426     {
00427         int step = (i+1)*channels, k, j;
00428         double xx = 0.0, xy = 0.0;
00429 #if 1
00430         int *x_ptr = &(window[step]), *state_ptr = &(state[0]);
00431         j = window_entries - step;
00432         for (;j>=0;j--,x_ptr++,state_ptr++)
00433         {
00434             double x_value = *x_ptr, state_value = *state_ptr;
00435             xx += state_value*state_value;
00436             xy += x_value*state_value;
00437         }
00438 #else
00439         for (j = 0; j <= (window_entries - step); j++);
00440         {
00441             double stepval = window[step+j], stateval = window[j];
00442 //            xx += (double)window[j]*(double)window[j];
00443 //            xy += (double)window[step+j]*(double)window[j];
00444             xx += stateval*stateval;
00445             xy += stepval*stateval;
00446         }
00447 #endif
00448         if (xx == 0.0)
00449             k = 0;
00450         else
00451             k = (int)(floor(-xy/xx * (double)LATTICE_FACTOR / (double)(tap_quant[i]) + 0.5));
00452 
00453         if (k > (LATTICE_FACTOR/tap_quant[i]))
00454             k = LATTICE_FACTOR/tap_quant[i];
00455         if (-k > (LATTICE_FACTOR/tap_quant[i]))
00456             k = -(LATTICE_FACTOR/tap_quant[i]);
00457 
00458         out[i] = k;
00459         k *= tap_quant[i];
00460 
00461 #if 1
00462         x_ptr = &(window[step]);
00463         state_ptr = &(state[0]);
00464         j = window_entries - step;
00465         for (;j>=0;j--,x_ptr++,state_ptr++)
00466         {
00467             int x_value = *x_ptr, state_value = *state_ptr;
00468             *x_ptr = x_value + shift_down(k*state_value,LATTICE_SHIFT);
00469             *state_ptr = state_value + shift_down(k*x_value, LATTICE_SHIFT);
00470         }
00471 #else
00472         for (j=0; j <= (window_entries - step); j++)
00473         {
00474             int stepval = window[step+j], stateval=state[j];
00475             window[step+j] += shift_down(k * stateval, LATTICE_SHIFT);
00476             state[j] += shift_down(k * stepval, LATTICE_SHIFT);
00477         }
00478 #endif
00479     }
00480 
00481     av_free(state);
00482 }
00483 
00484 static inline int code_samplerate(int samplerate)
00485 {
00486     switch (samplerate)
00487     {
00488         case 44100: return 0;
00489         case 22050: return 1;
00490         case 11025: return 2;
00491         case 96000: return 3;
00492         case 48000: return 4;
00493         case 32000: return 5;
00494         case 24000: return 6;
00495         case 16000: return 7;
00496         case 8000: return 8;
00497     }
00498     return -1;
00499 }
00500 
00501 static av_cold int sonic_encode_init(AVCodecContext *avctx)
00502 {
00503     SonicContext *s = avctx->priv_data;
00504     PutBitContext pb;
00505     int i, version = 0;
00506 
00507     if (avctx->channels > MAX_CHANNELS)
00508     {
00509         av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n");
00510         return -1; /* only stereo or mono for now */
00511     }
00512 
00513     if (avctx->channels == 2)
00514         s->decorrelation = MID_SIDE;
00515 
00516     if (avctx->codec->id == CODEC_ID_SONIC_LS)
00517     {
00518         s->lossless = 1;
00519         s->num_taps = 32;
00520         s->downsampling = 1;
00521         s->quantization = 0.0;
00522     }
00523     else
00524     {
00525         s->num_taps = 128;
00526         s->downsampling = 2;
00527         s->quantization = 1.0;
00528     }
00529 
00530     // max tap 2048
00531     if ((s->num_taps < 32) || (s->num_taps > 1024) ||
00532         ((s->num_taps>>5)<<5 != s->num_taps))
00533     {
00534         av_log(avctx, AV_LOG_ERROR, "Invalid number of taps\n");
00535         return -1;
00536     }
00537 
00538     // generate taps
00539     s->tap_quant = av_mallocz(4* s->num_taps);
00540     for (i = 0; i < s->num_taps; i++)
00541         s->tap_quant[i] = (int)(sqrt(i+1));
00542 
00543     s->channels = avctx->channels;
00544     s->samplerate = avctx->sample_rate;
00545 
00546     s->block_align = (int)(2048.0*s->samplerate/44100)/s->downsampling;
00547     s->frame_size = s->channels*s->block_align*s->downsampling;
00548 
00549     s->tail = av_mallocz(4* s->num_taps*s->channels);
00550     if (!s->tail)
00551         return -1;
00552     s->tail_size = s->num_taps*s->channels;
00553 
00554     s->predictor_k = av_mallocz(4 * s->num_taps);
00555     if (!s->predictor_k)
00556         return -1;
00557 
00558     for (i = 0; i < s->channels; i++)
00559     {
00560         s->coded_samples[i] = av_mallocz(4* s->block_align);
00561         if (!s->coded_samples[i])
00562             return -1;
00563     }
00564 
00565     s->int_samples = av_mallocz(4* s->frame_size);
00566 
00567     s->window_size = ((2*s->tail_size)+s->frame_size);
00568     s->window = av_mallocz(4* s->window_size);
00569     if (!s->window)
00570         return -1;
00571 
00572     avctx->extradata = av_mallocz(16);
00573     if (!avctx->extradata)
00574         return -1;
00575     init_put_bits(&pb, avctx->extradata, 16*8);
00576 
00577     put_bits(&pb, 2, version); // version
00578     if (version == 1)
00579     {
00580         put_bits(&pb, 2, s->channels);
00581         put_bits(&pb, 4, code_samplerate(s->samplerate));
00582     }
00583     put_bits(&pb, 1, s->lossless);
00584     if (!s->lossless)
00585         put_bits(&pb, 3, SAMPLE_SHIFT); // XXX FIXME: sample precision
00586     put_bits(&pb, 2, s->decorrelation);
00587     put_bits(&pb, 2, s->downsampling);
00588     put_bits(&pb, 5, (s->num_taps >> 5)-1); // 32..1024
00589     put_bits(&pb, 1, 0); // XXX FIXME: no custom tap quant table
00590 
00591     flush_put_bits(&pb);
00592     avctx->extradata_size = put_bits_count(&pb)/8;
00593 
00594     av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
00595         version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling);
00596 
00597     avctx->coded_frame = avcodec_alloc_frame();
00598     if (!avctx->coded_frame)
00599         return AVERROR(ENOMEM);
00600     avctx->coded_frame->key_frame = 1;
00601     avctx->frame_size = s->block_align*s->downsampling;
00602 
00603     return 0;
00604 }
00605 
00606 static av_cold int sonic_encode_close(AVCodecContext *avctx)
00607 {
00608     SonicContext *s = avctx->priv_data;
00609     int i;
00610 
00611     av_freep(&avctx->coded_frame);
00612 
00613     for (i = 0; i < s->channels; i++)
00614         av_free(s->coded_samples[i]);
00615 
00616     av_free(s->predictor_k);
00617     av_free(s->tail);
00618     av_free(s->tap_quant);
00619     av_free(s->window);
00620     av_free(s->int_samples);
00621 
00622     return 0;
00623 }
00624 
00625 static int sonic_encode_frame(AVCodecContext *avctx,
00626                             uint8_t *buf, int buf_size, void *data)
00627 {
00628     SonicContext *s = avctx->priv_data;
00629     PutBitContext pb;
00630     int i, j, ch, quant = 0, x = 0;
00631     short *samples = data;
00632 
00633     init_put_bits(&pb, buf, buf_size*8);
00634 
00635     // short -> internal
00636     for (i = 0; i < s->frame_size; i++)
00637         s->int_samples[i] = samples[i];
00638 
00639     if (!s->lossless)
00640         for (i = 0; i < s->frame_size; i++)
00641             s->int_samples[i] = s->int_samples[i] << SAMPLE_SHIFT;
00642 
00643     switch(s->decorrelation)
00644     {
00645         case MID_SIDE:
00646             for (i = 0; i < s->frame_size; i += s->channels)
00647             {
00648                 s->int_samples[i] += s->int_samples[i+1];
00649                 s->int_samples[i+1] -= shift(s->int_samples[i], 1);
00650             }
00651             break;
00652         case LEFT_SIDE:
00653             for (i = 0; i < s->frame_size; i += s->channels)
00654                 s->int_samples[i+1] -= s->int_samples[i];
00655             break;
00656         case RIGHT_SIDE:
00657             for (i = 0; i < s->frame_size; i += s->channels)
00658                 s->int_samples[i] -= s->int_samples[i+1];
00659             break;
00660     }
00661 
00662     memset(s->window, 0, 4* s->window_size);
00663 
00664     for (i = 0; i < s->tail_size; i++)
00665         s->window[x++] = s->tail[i];
00666 
00667     for (i = 0; i < s->frame_size; i++)
00668         s->window[x++] = s->int_samples[i];
00669 
00670     for (i = 0; i < s->tail_size; i++)
00671         s->window[x++] = 0;
00672 
00673     for (i = 0; i < s->tail_size; i++)
00674         s->tail[i] = s->int_samples[s->frame_size - s->tail_size + i];
00675 
00676     // generate taps
00677     modified_levinson_durbin(s->window, s->window_size,
00678                 s->predictor_k, s->num_taps, s->channels, s->tap_quant);
00679     if (intlist_write(&pb, s->predictor_k, s->num_taps, 0) < 0)
00680         return -1;
00681 
00682     for (ch = 0; ch < s->channels; ch++)
00683     {
00684         x = s->tail_size+ch;
00685         for (i = 0; i < s->block_align; i++)
00686         {
00687             int sum = 0;
00688             for (j = 0; j < s->downsampling; j++, x += s->channels)
00689                 sum += s->window[x];
00690             s->coded_samples[ch][i] = sum;
00691         }
00692     }
00693 
00694     // simple rate control code
00695     if (!s->lossless)
00696     {
00697         double energy1 = 0.0, energy2 = 0.0;
00698         for (ch = 0; ch < s->channels; ch++)
00699         {
00700             for (i = 0; i < s->block_align; i++)
00701             {
00702                 double sample = s->coded_samples[ch][i];
00703                 energy2 += sample*sample;
00704                 energy1 += fabs(sample);
00705             }
00706         }
00707 
00708         energy2 = sqrt(energy2/(s->channels*s->block_align));
00709         energy1 = sqrt(2.0)*energy1/(s->channels*s->block_align);
00710 
00711         // increase bitrate when samples are like a gaussian distribution
00712         // reduce bitrate when samples are like a two-tailed exponential distribution
00713 
00714         if (energy2 > energy1)
00715             energy2 += (energy2-energy1)*RATE_VARIATION;
00716 
00717         quant = (int)(BASE_QUANT*s->quantization*energy2/SAMPLE_FACTOR);
00718 //        av_log(avctx, AV_LOG_DEBUG, "quant: %d energy: %f / %f\n", quant, energy1, energy2);
00719 
00720         if (quant < 1)
00721             quant = 1;
00722         if (quant > 65535)
00723             quant = 65535;
00724 
00725         set_ue_golomb(&pb, quant);
00726 
00727         quant *= SAMPLE_FACTOR;
00728     }
00729 
00730     // write out coded samples
00731     for (ch = 0; ch < s->channels; ch++)
00732     {
00733         if (!s->lossless)
00734             for (i = 0; i < s->block_align; i++)
00735                 s->coded_samples[ch][i] = divide(s->coded_samples[ch][i], quant);
00736 
00737         if (intlist_write(&pb, s->coded_samples[ch], s->block_align, 1) < 0)
00738             return -1;
00739     }
00740 
00741 //    av_log(avctx, AV_LOG_DEBUG, "used bytes: %d\n", (put_bits_count(&pb)+7)/8);
00742 
00743     flush_put_bits(&pb);
00744     return (put_bits_count(&pb)+7)/8;
00745 }
00746 #endif /* CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER */
00747 
00748 #if CONFIG_SONIC_DECODER
00749 static const int samplerate_table[] =
00750     { 44100, 22050, 11025, 96000, 48000, 32000, 24000, 16000, 8000 };
00751 
00752 static av_cold int sonic_decode_init(AVCodecContext *avctx)
00753 {
00754     SonicContext *s = avctx->priv_data;
00755     GetBitContext gb;
00756     int i, version;
00757 
00758     s->channels = avctx->channels;
00759     s->samplerate = avctx->sample_rate;
00760 
00761     avcodec_get_frame_defaults(&s->frame);
00762     avctx->coded_frame = &s->frame;
00763 
00764     if (!avctx->extradata)
00765     {
00766         av_log(avctx, AV_LOG_ERROR, "No mandatory headers present\n");
00767         return -1;
00768     }
00769 
00770     init_get_bits(&gb, avctx->extradata, avctx->extradata_size);
00771 
00772     version = get_bits(&gb, 2);
00773     if (version > 1)
00774     {
00775         av_log(avctx, AV_LOG_ERROR, "Unsupported Sonic version, please report\n");
00776         return -1;
00777     }
00778 
00779     if (version == 1)
00780     {
00781         s->channels = get_bits(&gb, 2);
00782         s->samplerate = samplerate_table[get_bits(&gb, 4)];
00783         av_log(avctx, AV_LOG_INFO, "Sonicv2 chans: %d samprate: %d\n",
00784             s->channels, s->samplerate);
00785     }
00786 
00787     if (s->channels > MAX_CHANNELS)
00788     {
00789         av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n");
00790         return -1;
00791     }
00792 
00793     s->lossless = get_bits1(&gb);
00794     if (!s->lossless)
00795         skip_bits(&gb, 3); // XXX FIXME
00796     s->decorrelation = get_bits(&gb, 2);
00797 
00798     s->downsampling = get_bits(&gb, 2);
00799     s->num_taps = (get_bits(&gb, 5)+1)<<5;
00800     if (get_bits1(&gb)) // XXX FIXME
00801         av_log(avctx, AV_LOG_INFO, "Custom quant table\n");
00802 
00803     s->block_align = (int)(2048.0*s->samplerate/44100)/s->downsampling;
00804     s->frame_size = s->channels*s->block_align*s->downsampling;
00805 //    avctx->frame_size = s->block_align;
00806 
00807     av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
00808         version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling);
00809 
00810     // generate taps
00811     s->tap_quant = av_mallocz(4* s->num_taps);
00812     for (i = 0; i < s->num_taps; i++)
00813         s->tap_quant[i] = (int)(sqrt(i+1));
00814 
00815     s->predictor_k = av_mallocz(4* s->num_taps);
00816 
00817     for (i = 0; i < s->channels; i++)
00818     {
00819         s->predictor_state[i] = av_mallocz(4* s->num_taps);
00820         if (!s->predictor_state[i])
00821             return -1;
00822     }
00823 
00824     for (i = 0; i < s->channels; i++)
00825     {
00826         s->coded_samples[i] = av_mallocz(4* s->block_align);
00827         if (!s->coded_samples[i])
00828             return -1;
00829     }
00830     s->int_samples = av_mallocz(4* s->frame_size);
00831 
00832     avctx->sample_fmt = AV_SAMPLE_FMT_S16;
00833     return 0;
00834 }
00835 
00836 static av_cold int sonic_decode_close(AVCodecContext *avctx)
00837 {
00838     SonicContext *s = avctx->priv_data;
00839     int i;
00840 
00841     av_free(s->int_samples);
00842     av_free(s->tap_quant);
00843     av_free(s->predictor_k);
00844 
00845     for (i = 0; i < s->channels; i++)
00846     {
00847         av_free(s->predictor_state[i]);
00848         av_free(s->coded_samples[i]);
00849     }
00850 
00851     return 0;
00852 }
00853 
00854 static int sonic_decode_frame(AVCodecContext *avctx,
00855                             void *data, int *got_frame_ptr,
00856                             AVPacket *avpkt)
00857 {
00858     const uint8_t *buf = avpkt->data;
00859     int buf_size = avpkt->size;
00860     SonicContext *s = avctx->priv_data;
00861     GetBitContext gb;
00862     int i, quant, ch, j, ret;
00863     short *samples;
00864 
00865     if (buf_size == 0) return 0;
00866 
00867     s->frame.nb_samples = s->frame_size;
00868     if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
00869         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
00870         return ret;
00871     }
00872     samples = s->frame.data[0];
00873 
00874 //    av_log(NULL, AV_LOG_INFO, "buf_size: %d\n", buf_size);
00875 
00876     init_get_bits(&gb, buf, buf_size*8);
00877 
00878     intlist_read(&gb, s->predictor_k, s->num_taps, 0);
00879 
00880     // dequantize
00881     for (i = 0; i < s->num_taps; i++)
00882         s->predictor_k[i] *= s->tap_quant[i];
00883 
00884     if (s->lossless)
00885         quant = 1;
00886     else
00887         quant = get_ue_golomb(&gb) * SAMPLE_FACTOR;
00888 
00889 //    av_log(NULL, AV_LOG_INFO, "quant: %d\n", quant);
00890 
00891     for (ch = 0; ch < s->channels; ch++)
00892     {
00893         int x = ch;
00894 
00895         predictor_init_state(s->predictor_k, s->predictor_state[ch], s->num_taps);
00896 
00897         intlist_read(&gb, s->coded_samples[ch], s->block_align, 1);
00898 
00899         for (i = 0; i < s->block_align; i++)
00900         {
00901             for (j = 0; j < s->downsampling - 1; j++)
00902             {
00903                 s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, 0);
00904                 x += s->channels;
00905             }
00906 
00907             s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, s->coded_samples[ch][i] * quant);
00908             x += s->channels;
00909         }
00910 
00911         for (i = 0; i < s->num_taps; i++)
00912             s->predictor_state[ch][i] = s->int_samples[s->frame_size - s->channels + ch - i*s->channels];
00913     }
00914 
00915     switch(s->decorrelation)
00916     {
00917         case MID_SIDE:
00918             for (i = 0; i < s->frame_size; i += s->channels)
00919             {
00920                 s->int_samples[i+1] += shift(s->int_samples[i], 1);
00921                 s->int_samples[i] -= s->int_samples[i+1];
00922             }
00923             break;
00924         case LEFT_SIDE:
00925             for (i = 0; i < s->frame_size; i += s->channels)
00926                 s->int_samples[i+1] += s->int_samples[i];
00927             break;
00928         case RIGHT_SIDE:
00929             for (i = 0; i < s->frame_size; i += s->channels)
00930                 s->int_samples[i] += s->int_samples[i+1];
00931             break;
00932     }
00933 
00934     if (!s->lossless)
00935         for (i = 0; i < s->frame_size; i++)
00936             s->int_samples[i] = shift(s->int_samples[i], SAMPLE_SHIFT);
00937 
00938     // internal -> short
00939     for (i = 0; i < s->frame_size; i++)
00940         samples[i] = av_clip_int16(s->int_samples[i]);
00941 
00942     align_get_bits(&gb);
00943 
00944     *got_frame_ptr = 1;
00945     *(AVFrame*)data = s->frame;
00946 
00947     return (get_bits_count(&gb)+7)/8;
00948 }
00949 
00950 AVCodec ff_sonic_decoder = {
00951     .name           = "sonic",
00952     .type           = AVMEDIA_TYPE_AUDIO,
00953     .id             = CODEC_ID_SONIC,
00954     .priv_data_size = sizeof(SonicContext),
00955     .init           = sonic_decode_init,
00956     .close          = sonic_decode_close,
00957     .decode         = sonic_decode_frame,
00958     .capabilities   = CODEC_CAP_DR1,
00959     .long_name = NULL_IF_CONFIG_SMALL("Sonic"),
00960 };
00961 #endif /* CONFIG_SONIC_DECODER */
00962 
00963 #if CONFIG_SONIC_ENCODER
00964 AVCodec ff_sonic_encoder = {
00965     .name           = "sonic",
00966     .type           = AVMEDIA_TYPE_AUDIO,
00967     .id             = CODEC_ID_SONIC,
00968     .priv_data_size = sizeof(SonicContext),
00969     .init           = sonic_encode_init,
00970     .encode         = sonic_encode_frame,
00971     .close          = sonic_encode_close,
00972     .long_name = NULL_IF_CONFIG_SMALL("Sonic"),
00973 };
00974 #endif
00975 
00976 #if CONFIG_SONIC_LS_ENCODER
00977 AVCodec ff_sonic_ls_encoder = {
00978     .name           = "sonicls",
00979     .type           = AVMEDIA_TYPE_AUDIO,
00980     .id             = CODEC_ID_SONIC_LS,
00981     .priv_data_size = sizeof(SonicContext),
00982     .init           = sonic_encode_init,
00983     .encode         = sonic_encode_frame,
00984     .close          = sonic_encode_close,
00985     .long_name = NULL_IF_CONFIG_SMALL("Sonic lossless"),
00986 };
00987 #endif
Generated on Fri Feb 1 2013 14:34:43 for FFmpeg by doxygen 1.7.1