• 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     int lossless, decorrelation;
00048 
00049     int num_taps, downsampling;
00050     double quantization;
00051 
00052     int channels, samplerate, block_align, frame_size;
00053 
00054     int *tap_quant;
00055     int *int_samples;
00056     int *coded_samples[MAX_CHANNELS];
00057 
00058     // for encoding
00059     int *tail;
00060     int tail_size;
00061     int *window;
00062     int window_size;
00063 
00064     // for decoding
00065     int *predictor_k;
00066     int *predictor_state[MAX_CHANNELS];
00067 } SonicContext;
00068 
00069 #define LATTICE_SHIFT   10
00070 #define SAMPLE_SHIFT    4
00071 #define LATTICE_FACTOR  (1 << LATTICE_SHIFT)
00072 #define SAMPLE_FACTOR   (1 << SAMPLE_SHIFT)
00073 
00074 #define BASE_QUANT      0.6
00075 #define RATE_VARIATION  3.0
00076 
00077 static inline int divide(int a, int b)
00078 {
00079     if (a < 0)
00080         return -( (-a + b/2)/b );
00081     else
00082         return (a + b/2)/b;
00083 }
00084 
00085 static inline int shift(int a,int b)
00086 {
00087     return (a+(1<<(b-1))) >> b;
00088 }
00089 
00090 static inline int shift_down(int a,int b)
00091 {
00092     return (a>>b)+((a<0)?1:0);
00093 }
00094 
00095 #if 1
00096 static inline int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
00097 {
00098     int i;
00099 
00100     for (i = 0; i < entries; i++)
00101         set_se_golomb(pb, buf[i]);
00102 
00103     return 1;
00104 }
00105 
00106 static inline int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
00107 {
00108     int i;
00109 
00110     for (i = 0; i < entries; i++)
00111         buf[i] = get_se_golomb(gb);
00112 
00113     return 1;
00114 }
00115 
00116 #else
00117 
00118 #define ADAPT_LEVEL 8
00119 
00120 static int bits_to_store(uint64_t x)
00121 {
00122     int res = 0;
00123 
00124     while(x)
00125     {
00126         res++;
00127         x >>= 1;
00128     }
00129     return res;
00130 }
00131 
00132 static void write_uint_max(PutBitContext *pb, unsigned int value, unsigned int max)
00133 {
00134     int i, bits;
00135 
00136     if (!max)
00137         return;
00138 
00139     bits = bits_to_store(max);
00140 
00141     for (i = 0; i < bits-1; i++)
00142         put_bits(pb, 1, value & (1 << i));
00143 
00144     if ( (value | (1 << (bits-1))) <= max)
00145         put_bits(pb, 1, value & (1 << (bits-1)));
00146 }
00147 
00148 static unsigned int read_uint_max(GetBitContext *gb, int max)
00149 {
00150     int i, bits, value = 0;
00151 
00152     if (!max)
00153         return 0;
00154 
00155     bits = bits_to_store(max);
00156 
00157     for (i = 0; i < bits-1; i++)
00158         if (get_bits1(gb))
00159             value += 1 << i;
00160 
00161     if ( (value | (1<<(bits-1))) <= max)
00162         if (get_bits1(gb))
00163             value += 1 << (bits-1);
00164 
00165     return value;
00166 }
00167 
00168 static int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
00169 {
00170     int i, j, x = 0, low_bits = 0, max = 0;
00171     int step = 256, pos = 0, dominant = 0, any = 0;
00172     int *copy, *bits;
00173 
00174     copy = av_mallocz(4* entries);
00175     if (!copy)
00176         return -1;
00177 
00178     if (base_2_part)
00179     {
00180         int energy = 0;
00181 
00182         for (i = 0; i < entries; i++)
00183             energy += abs(buf[i]);
00184 
00185         low_bits = bits_to_store(energy / (entries * 2));
00186         if (low_bits > 15)
00187             low_bits = 15;
00188 
00189         put_bits(pb, 4, low_bits);
00190     }
00191 
00192     for (i = 0; i < entries; i++)
00193     {
00194         put_bits(pb, low_bits, abs(buf[i]));
00195         copy[i] = abs(buf[i]) >> low_bits;
00196         if (copy[i] > max)
00197             max = abs(copy[i]);
00198     }
00199 
00200     bits = av_mallocz(4* entries*max);
00201     if (!bits)
00202     {
00203 //        av_free(copy);
00204         return -1;
00205     }
00206 
00207     for (i = 0; i <= max; i++)
00208     {
00209         for (j = 0; j < entries; j++)
00210             if (copy[j] >= i)
00211                 bits[x++] = copy[j] > i;
00212     }
00213 
00214     // store bitstream
00215     while (pos < x)
00216     {
00217         int steplet = step >> 8;
00218 
00219         if (pos + steplet > x)
00220             steplet = x - pos;
00221 
00222         for (i = 0; i < steplet; i++)
00223             if (bits[i+pos] != dominant)
00224                 any = 1;
00225 
00226         put_bits(pb, 1, any);
00227 
00228         if (!any)
00229         {
00230             pos += steplet;
00231             step += step / ADAPT_LEVEL;
00232         }
00233         else
00234         {
00235             int interloper = 0;
00236 
00237             while (((pos + interloper) < x) && (bits[pos + interloper] == dominant))
00238                 interloper++;
00239 
00240             // note change
00241             write_uint_max(pb, interloper, (step >> 8) - 1);
00242 
00243             pos += interloper + 1;
00244             step -= step / ADAPT_LEVEL;
00245         }
00246 
00247         if (step < 256)
00248         {
00249             step = 65536 / step;
00250             dominant = !dominant;
00251         }
00252     }
00253 
00254     // store signs
00255     for (i = 0; i < entries; i++)
00256         if (buf[i])
00257             put_bits(pb, 1, buf[i] < 0);
00258 
00259 //    av_free(bits);
00260 //    av_free(copy);
00261 
00262     return 0;
00263 }
00264 
00265 static int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
00266 {
00267     int i, low_bits = 0, x = 0;
00268     int n_zeros = 0, step = 256, dominant = 0;
00269     int pos = 0, level = 0;
00270     int *bits = av_mallocz(4* entries);
00271 
00272     if (!bits)
00273         return -1;
00274 
00275     if (base_2_part)
00276     {
00277         low_bits = get_bits(gb, 4);
00278 
00279         if (low_bits)
00280             for (i = 0; i < entries; i++)
00281                 buf[i] = get_bits(gb, low_bits);
00282     }
00283 
00284 //    av_log(NULL, AV_LOG_INFO, "entries: %d, low bits: %d\n", entries, low_bits);
00285 
00286     while (n_zeros < entries)
00287     {
00288         int steplet = step >> 8;
00289 
00290         if (!get_bits1(gb))
00291         {
00292             for (i = 0; i < steplet; i++)
00293                 bits[x++] = dominant;
00294 
00295             if (!dominant)
00296                 n_zeros += steplet;
00297 
00298             step += step / ADAPT_LEVEL;
00299         }
00300         else
00301         {
00302             int actual_run = read_uint_max(gb, steplet-1);
00303 
00304 //            av_log(NULL, AV_LOG_INFO, "actual run: %d\n", actual_run);
00305 
00306             for (i = 0; i < actual_run; i++)
00307                 bits[x++] = dominant;
00308 
00309             bits[x++] = !dominant;
00310 
00311             if (!dominant)
00312                 n_zeros += actual_run;
00313             else
00314                 n_zeros++;
00315 
00316             step -= step / ADAPT_LEVEL;
00317         }
00318 
00319         if (step < 256)
00320         {
00321             step = 65536 / step;
00322             dominant = !dominant;
00323         }
00324     }
00325 
00326     // reconstruct unsigned values
00327     n_zeros = 0;
00328     for (i = 0; n_zeros < entries; i++)
00329     {
00330         while(1)
00331         {
00332             if (pos >= entries)
00333             {
00334                 pos = 0;
00335                 level += 1 << low_bits;
00336             }
00337 
00338             if (buf[pos] >= level)
00339                 break;
00340 
00341             pos++;
00342         }
00343 
00344         if (bits[i])
00345             buf[pos] += 1 << low_bits;
00346         else
00347             n_zeros++;
00348 
00349         pos++;
00350     }
00351 //    av_free(bits);
00352 
00353     // read signs
00354     for (i = 0; i < entries; i++)
00355         if (buf[i] && get_bits1(gb))
00356             buf[i] = -buf[i];
00357 
00358 //    av_log(NULL, AV_LOG_INFO, "zeros: %d pos: %d\n", n_zeros, pos);
00359 
00360     return 0;
00361 }
00362 #endif
00363 
00364 static void predictor_init_state(int *k, int *state, int order)
00365 {
00366     int i;
00367 
00368     for (i = order-2; i >= 0; i--)
00369     {
00370         int j, p, x = state[i];
00371 
00372         for (j = 0, p = i+1; p < order; j++,p++)
00373             {
00374             int tmp = x + shift_down(k[j] * state[p], LATTICE_SHIFT);
00375             state[p] += shift_down(k[j]*x, LATTICE_SHIFT);
00376             x = tmp;
00377         }
00378     }
00379 }
00380 
00381 static int predictor_calc_error(int *k, int *state, int order, int error)
00382 {
00383     int i, x = error - shift_down(k[order-1] * state[order-1], LATTICE_SHIFT);
00384 
00385 #if 1
00386     int *k_ptr = &(k[order-2]),
00387         *state_ptr = &(state[order-2]);
00388     for (i = order-2; i >= 0; i--, k_ptr--, state_ptr--)
00389     {
00390         int k_value = *k_ptr, state_value = *state_ptr;
00391         x -= shift_down(k_value * state_value, LATTICE_SHIFT);
00392         state_ptr[1] = state_value + shift_down(k_value * x, LATTICE_SHIFT);
00393     }
00394 #else
00395     for (i = order-2; i >= 0; i--)
00396     {
00397         x -= shift_down(k[i] * state[i], LATTICE_SHIFT);
00398         state[i+1] = state[i] + shift_down(k[i] * x, LATTICE_SHIFT);
00399     }
00400 #endif
00401 
00402     // don't drift too far, to avoid overflows
00403     if (x >  (SAMPLE_FACTOR<<16)) x =  (SAMPLE_FACTOR<<16);
00404     if (x < -(SAMPLE_FACTOR<<16)) x = -(SAMPLE_FACTOR<<16);
00405 
00406     state[0] = x;
00407 
00408     return x;
00409 }
00410 
00411 #if CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER
00412 // Heavily modified Levinson-Durbin algorithm which
00413 // copes better with quantization, and calculates the
00414 // actual whitened result as it goes.
00415 
00416 static void modified_levinson_durbin(int *window, int window_entries,
00417         int *out, int out_entries, int channels, int *tap_quant)
00418 {
00419     int i;
00420     int *state = av_mallocz(4* window_entries);
00421 
00422     memcpy(state, window, 4* window_entries);
00423 
00424     for (i = 0; i < out_entries; i++)
00425     {
00426         int step = (i+1)*channels, k, j;
00427         double xx = 0.0, xy = 0.0;
00428 #if 1
00429         int *x_ptr = &(window[step]), *state_ptr = &(state[0]);
00430         j = window_entries - step;
00431         for (;j>=0;j--,x_ptr++,state_ptr++)
00432         {
00433             double x_value = *x_ptr, state_value = *state_ptr;
00434             xx += state_value*state_value;
00435             xy += x_value*state_value;
00436         }
00437 #else
00438         for (j = 0; j <= (window_entries - step); j++);
00439         {
00440             double stepval = window[step+j], stateval = window[j];
00441 //            xx += (double)window[j]*(double)window[j];
00442 //            xy += (double)window[step+j]*(double)window[j];
00443             xx += stateval*stateval;
00444             xy += stepval*stateval;
00445         }
00446 #endif
00447         if (xx == 0.0)
00448             k = 0;
00449         else
00450             k = (int)(floor(-xy/xx * (double)LATTICE_FACTOR / (double)(tap_quant[i]) + 0.5));
00451 
00452         if (k > (LATTICE_FACTOR/tap_quant[i]))
00453             k = LATTICE_FACTOR/tap_quant[i];
00454         if (-k > (LATTICE_FACTOR/tap_quant[i]))
00455             k = -(LATTICE_FACTOR/tap_quant[i]);
00456 
00457         out[i] = k;
00458         k *= tap_quant[i];
00459 
00460 #if 1
00461         x_ptr = &(window[step]);
00462         state_ptr = &(state[0]);
00463         j = window_entries - step;
00464         for (;j>=0;j--,x_ptr++,state_ptr++)
00465         {
00466             int x_value = *x_ptr, state_value = *state_ptr;
00467             *x_ptr = x_value + shift_down(k*state_value,LATTICE_SHIFT);
00468             *state_ptr = state_value + shift_down(k*x_value, LATTICE_SHIFT);
00469         }
00470 #else
00471         for (j=0; j <= (window_entries - step); j++)
00472         {
00473             int stepval = window[step+j], stateval=state[j];
00474             window[step+j] += shift_down(k * stateval, LATTICE_SHIFT);
00475             state[j] += shift_down(k * stepval, LATTICE_SHIFT);
00476         }
00477 #endif
00478     }
00479 
00480     av_free(state);
00481 }
00482 
00483 static inline int code_samplerate(int samplerate)
00484 {
00485     switch (samplerate)
00486     {
00487         case 44100: return 0;
00488         case 22050: return 1;
00489         case 11025: return 2;
00490         case 96000: return 3;
00491         case 48000: return 4;
00492         case 32000: return 5;
00493         case 24000: return 6;
00494         case 16000: return 7;
00495         case 8000: return 8;
00496     }
00497     return -1;
00498 }
00499 
00500 static av_cold int sonic_encode_init(AVCodecContext *avctx)
00501 {
00502     SonicContext *s = avctx->priv_data;
00503     PutBitContext pb;
00504     int i, version = 0;
00505 
00506     if (avctx->channels > MAX_CHANNELS)
00507     {
00508         av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n");
00509         return -1; /* only stereo or mono for now */
00510     }
00511 
00512     if (avctx->channels == 2)
00513         s->decorrelation = MID_SIDE;
00514 
00515     if (avctx->codec->id == CODEC_ID_SONIC_LS)
00516     {
00517         s->lossless = 1;
00518         s->num_taps = 32;
00519         s->downsampling = 1;
00520         s->quantization = 0.0;
00521     }
00522     else
00523     {
00524         s->num_taps = 128;
00525         s->downsampling = 2;
00526         s->quantization = 1.0;
00527     }
00528 
00529     // max tap 2048
00530     if ((s->num_taps < 32) || (s->num_taps > 1024) ||
00531         ((s->num_taps>>5)<<5 != s->num_taps))
00532     {
00533         av_log(avctx, AV_LOG_ERROR, "Invalid number of taps\n");
00534         return -1;
00535     }
00536 
00537     // generate taps
00538     s->tap_quant = av_mallocz(4* s->num_taps);
00539     for (i = 0; i < s->num_taps; i++)
00540         s->tap_quant[i] = (int)(sqrt(i+1));
00541 
00542     s->channels = avctx->channels;
00543     s->samplerate = avctx->sample_rate;
00544 
00545     s->block_align = (int)(2048.0*s->samplerate/44100)/s->downsampling;
00546     s->frame_size = s->channels*s->block_align*s->downsampling;
00547 
00548     s->tail = av_mallocz(4* s->num_taps*s->channels);
00549     if (!s->tail)
00550         return -1;
00551     s->tail_size = s->num_taps*s->channels;
00552 
00553     s->predictor_k = av_mallocz(4 * s->num_taps);
00554     if (!s->predictor_k)
00555         return -1;
00556 
00557     for (i = 0; i < s->channels; i++)
00558     {
00559         s->coded_samples[i] = av_mallocz(4* s->block_align);
00560         if (!s->coded_samples[i])
00561             return -1;
00562     }
00563 
00564     s->int_samples = av_mallocz(4* s->frame_size);
00565 
00566     s->window_size = ((2*s->tail_size)+s->frame_size);
00567     s->window = av_mallocz(4* s->window_size);
00568     if (!s->window)
00569         return -1;
00570 
00571     avctx->extradata = av_mallocz(16);
00572     if (!avctx->extradata)
00573         return -1;
00574     init_put_bits(&pb, avctx->extradata, 16*8);
00575 
00576     put_bits(&pb, 2, version); // version
00577     if (version == 1)
00578     {
00579         put_bits(&pb, 2, s->channels);
00580         put_bits(&pb, 4, code_samplerate(s->samplerate));
00581     }
00582     put_bits(&pb, 1, s->lossless);
00583     if (!s->lossless)
00584         put_bits(&pb, 3, SAMPLE_SHIFT); // XXX FIXME: sample precision
00585     put_bits(&pb, 2, s->decorrelation);
00586     put_bits(&pb, 2, s->downsampling);
00587     put_bits(&pb, 5, (s->num_taps >> 5)-1); // 32..1024
00588     put_bits(&pb, 1, 0); // XXX FIXME: no custom tap quant table
00589 
00590     flush_put_bits(&pb);
00591     avctx->extradata_size = put_bits_count(&pb)/8;
00592 
00593     av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
00594         version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling);
00595 
00596     avctx->coded_frame = avcodec_alloc_frame();
00597     if (!avctx->coded_frame)
00598         return AVERROR(ENOMEM);
00599     avctx->coded_frame->key_frame = 1;
00600     avctx->frame_size = s->block_align*s->downsampling;
00601 
00602     return 0;
00603 }
00604 
00605 static av_cold int sonic_encode_close(AVCodecContext *avctx)
00606 {
00607     SonicContext *s = avctx->priv_data;
00608     int i;
00609 
00610     av_freep(&avctx->coded_frame);
00611 
00612     for (i = 0; i < s->channels; i++)
00613         av_free(s->coded_samples[i]);
00614 
00615     av_free(s->predictor_k);
00616     av_free(s->tail);
00617     av_free(s->tap_quant);
00618     av_free(s->window);
00619     av_free(s->int_samples);
00620 
00621     return 0;
00622 }
00623 
00624 static int sonic_encode_frame(AVCodecContext *avctx,
00625                             uint8_t *buf, int buf_size, void *data)
00626 {
00627     SonicContext *s = avctx->priv_data;
00628     PutBitContext pb;
00629     int i, j, ch, quant = 0, x = 0;
00630     short *samples = data;
00631 
00632     init_put_bits(&pb, buf, buf_size*8);
00633 
00634     // short -> internal
00635     for (i = 0; i < s->frame_size; i++)
00636         s->int_samples[i] = samples[i];
00637 
00638     if (!s->lossless)
00639         for (i = 0; i < s->frame_size; i++)
00640             s->int_samples[i] = s->int_samples[i] << SAMPLE_SHIFT;
00641 
00642     switch(s->decorrelation)
00643     {
00644         case MID_SIDE:
00645             for (i = 0; i < s->frame_size; i += s->channels)
00646             {
00647                 s->int_samples[i] += s->int_samples[i+1];
00648                 s->int_samples[i+1] -= shift(s->int_samples[i], 1);
00649             }
00650             break;
00651         case LEFT_SIDE:
00652             for (i = 0; i < s->frame_size; i += s->channels)
00653                 s->int_samples[i+1] -= s->int_samples[i];
00654             break;
00655         case RIGHT_SIDE:
00656             for (i = 0; i < s->frame_size; i += s->channels)
00657                 s->int_samples[i] -= s->int_samples[i+1];
00658             break;
00659     }
00660 
00661     memset(s->window, 0, 4* s->window_size);
00662 
00663     for (i = 0; i < s->tail_size; i++)
00664         s->window[x++] = s->tail[i];
00665 
00666     for (i = 0; i < s->frame_size; i++)
00667         s->window[x++] = s->int_samples[i];
00668 
00669     for (i = 0; i < s->tail_size; i++)
00670         s->window[x++] = 0;
00671 
00672     for (i = 0; i < s->tail_size; i++)
00673         s->tail[i] = s->int_samples[s->frame_size - s->tail_size + i];
00674 
00675     // generate taps
00676     modified_levinson_durbin(s->window, s->window_size,
00677                 s->predictor_k, s->num_taps, s->channels, s->tap_quant);
00678     if (intlist_write(&pb, s->predictor_k, s->num_taps, 0) < 0)
00679         return -1;
00680 
00681     for (ch = 0; ch < s->channels; ch++)
00682     {
00683         x = s->tail_size+ch;
00684         for (i = 0; i < s->block_align; i++)
00685         {
00686             int sum = 0;
00687             for (j = 0; j < s->downsampling; j++, x += s->channels)
00688                 sum += s->window[x];
00689             s->coded_samples[ch][i] = sum;
00690         }
00691     }
00692 
00693     // simple rate control code
00694     if (!s->lossless)
00695     {
00696         double energy1 = 0.0, energy2 = 0.0;
00697         for (ch = 0; ch < s->channels; ch++)
00698         {
00699             for (i = 0; i < s->block_align; i++)
00700             {
00701                 double sample = s->coded_samples[ch][i];
00702                 energy2 += sample*sample;
00703                 energy1 += fabs(sample);
00704             }
00705         }
00706 
00707         energy2 = sqrt(energy2/(s->channels*s->block_align));
00708         energy1 = sqrt(2.0)*energy1/(s->channels*s->block_align);
00709 
00710         // increase bitrate when samples are like a gaussian distribution
00711         // reduce bitrate when samples are like a two-tailed exponential distribution
00712 
00713         if (energy2 > energy1)
00714             energy2 += (energy2-energy1)*RATE_VARIATION;
00715 
00716         quant = (int)(BASE_QUANT*s->quantization*energy2/SAMPLE_FACTOR);
00717 //        av_log(avctx, AV_LOG_DEBUG, "quant: %d energy: %f / %f\n", quant, energy1, energy2);
00718 
00719         if (quant < 1)
00720             quant = 1;
00721         if (quant > 65535)
00722             quant = 65535;
00723 
00724         set_ue_golomb(&pb, quant);
00725 
00726         quant *= SAMPLE_FACTOR;
00727     }
00728 
00729     // write out coded samples
00730     for (ch = 0; ch < s->channels; ch++)
00731     {
00732         if (!s->lossless)
00733             for (i = 0; i < s->block_align; i++)
00734                 s->coded_samples[ch][i] = divide(s->coded_samples[ch][i], quant);
00735 
00736         if (intlist_write(&pb, s->coded_samples[ch], s->block_align, 1) < 0)
00737             return -1;
00738     }
00739 
00740 //    av_log(avctx, AV_LOG_DEBUG, "used bytes: %d\n", (put_bits_count(&pb)+7)/8);
00741 
00742     flush_put_bits(&pb);
00743     return (put_bits_count(&pb)+7)/8;
00744 }
00745 #endif /* CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER */
00746 
00747 #if CONFIG_SONIC_DECODER
00748 static const int samplerate_table[] =
00749     { 44100, 22050, 11025, 96000, 48000, 32000, 24000, 16000, 8000 };
00750 
00751 static av_cold int sonic_decode_init(AVCodecContext *avctx)
00752 {
00753     SonicContext *s = avctx->priv_data;
00754     GetBitContext gb;
00755     int i, version;
00756 
00757     s->channels = avctx->channels;
00758     s->samplerate = avctx->sample_rate;
00759 
00760     if (!avctx->extradata)
00761     {
00762         av_log(avctx, AV_LOG_ERROR, "No mandatory headers present\n");
00763         return -1;
00764     }
00765 
00766     init_get_bits(&gb, avctx->extradata, avctx->extradata_size);
00767 
00768     version = get_bits(&gb, 2);
00769     if (version > 1)
00770     {
00771         av_log(avctx, AV_LOG_ERROR, "Unsupported Sonic version, please report\n");
00772         return -1;
00773     }
00774 
00775     if (version == 1)
00776     {
00777         s->channels = get_bits(&gb, 2);
00778         s->samplerate = samplerate_table[get_bits(&gb, 4)];
00779         av_log(avctx, AV_LOG_INFO, "Sonicv2 chans: %d samprate: %d\n",
00780             s->channels, s->samplerate);
00781     }
00782 
00783     if (s->channels > MAX_CHANNELS)
00784     {
00785         av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n");
00786         return -1;
00787     }
00788 
00789     s->lossless = get_bits1(&gb);
00790     if (!s->lossless)
00791         skip_bits(&gb, 3); // XXX FIXME
00792     s->decorrelation = get_bits(&gb, 2);
00793 
00794     s->downsampling = get_bits(&gb, 2);
00795     s->num_taps = (get_bits(&gb, 5)+1)<<5;
00796     if (get_bits1(&gb)) // XXX FIXME
00797         av_log(avctx, AV_LOG_INFO, "Custom quant table\n");
00798 
00799     s->block_align = (int)(2048.0*(s->samplerate/44100))/s->downsampling;
00800     s->frame_size = s->channels*s->block_align*s->downsampling;
00801 //    avctx->frame_size = s->block_align;
00802 
00803     av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
00804         version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling);
00805 
00806     // generate taps
00807     s->tap_quant = av_mallocz(4* s->num_taps);
00808     for (i = 0; i < s->num_taps; i++)
00809         s->tap_quant[i] = (int)(sqrt(i+1));
00810 
00811     s->predictor_k = av_mallocz(4* s->num_taps);
00812 
00813     for (i = 0; i < s->channels; i++)
00814     {
00815         s->predictor_state[i] = av_mallocz(4* s->num_taps);
00816         if (!s->predictor_state[i])
00817             return -1;
00818     }
00819 
00820     for (i = 0; i < s->channels; i++)
00821     {
00822         s->coded_samples[i] = av_mallocz(4* s->block_align);
00823         if (!s->coded_samples[i])
00824             return -1;
00825     }
00826     s->int_samples = av_mallocz(4* s->frame_size);
00827 
00828     avctx->sample_fmt = SAMPLE_FMT_S16;
00829     return 0;
00830 }
00831 
00832 static av_cold int sonic_decode_close(AVCodecContext *avctx)
00833 {
00834     SonicContext *s = avctx->priv_data;
00835     int i;
00836 
00837     av_free(s->int_samples);
00838     av_free(s->tap_quant);
00839     av_free(s->predictor_k);
00840 
00841     for (i = 0; i < s->channels; i++)
00842     {
00843         av_free(s->predictor_state[i]);
00844         av_free(s->coded_samples[i]);
00845     }
00846 
00847     return 0;
00848 }
00849 
00850 static int sonic_decode_frame(AVCodecContext *avctx,
00851                             void *data, int *data_size,
00852                             AVPacket *avpkt)
00853 {
00854     const uint8_t *buf = avpkt->data;
00855     int buf_size = avpkt->size;
00856     SonicContext *s = avctx->priv_data;
00857     GetBitContext gb;
00858     int i, quant, ch, j;
00859     short *samples = data;
00860 
00861     if (buf_size == 0) return 0;
00862 
00863 //    av_log(NULL, AV_LOG_INFO, "buf_size: %d\n", buf_size);
00864 
00865     init_get_bits(&gb, buf, buf_size*8);
00866 
00867     intlist_read(&gb, s->predictor_k, s->num_taps, 0);
00868 
00869     // dequantize
00870     for (i = 0; i < s->num_taps; i++)
00871         s->predictor_k[i] *= s->tap_quant[i];
00872 
00873     if (s->lossless)
00874         quant = 1;
00875     else
00876         quant = get_ue_golomb(&gb) * SAMPLE_FACTOR;
00877 
00878 //    av_log(NULL, AV_LOG_INFO, "quant: %d\n", quant);
00879 
00880     for (ch = 0; ch < s->channels; ch++)
00881     {
00882         int x = ch;
00883 
00884         predictor_init_state(s->predictor_k, s->predictor_state[ch], s->num_taps);
00885 
00886         intlist_read(&gb, s->coded_samples[ch], s->block_align, 1);
00887 
00888         for (i = 0; i < s->block_align; i++)
00889         {
00890             for (j = 0; j < s->downsampling - 1; j++)
00891             {
00892                 s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, 0);
00893                 x += s->channels;
00894             }
00895 
00896             s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, s->coded_samples[ch][i] * quant);
00897             x += s->channels;
00898         }
00899 
00900         for (i = 0; i < s->num_taps; i++)
00901             s->predictor_state[ch][i] = s->int_samples[s->frame_size - s->channels + ch - i*s->channels];
00902     }
00903 
00904     switch(s->decorrelation)
00905     {
00906         case MID_SIDE:
00907             for (i = 0; i < s->frame_size; i += s->channels)
00908             {
00909                 s->int_samples[i+1] += shift(s->int_samples[i], 1);
00910                 s->int_samples[i] -= s->int_samples[i+1];
00911             }
00912             break;
00913         case LEFT_SIDE:
00914             for (i = 0; i < s->frame_size; i += s->channels)
00915                 s->int_samples[i+1] += s->int_samples[i];
00916             break;
00917         case RIGHT_SIDE:
00918             for (i = 0; i < s->frame_size; i += s->channels)
00919                 s->int_samples[i] += s->int_samples[i+1];
00920             break;
00921     }
00922 
00923     if (!s->lossless)
00924         for (i = 0; i < s->frame_size; i++)
00925             s->int_samples[i] = shift(s->int_samples[i], SAMPLE_SHIFT);
00926 
00927     // internal -> short
00928     for (i = 0; i < s->frame_size; i++)
00929         samples[i] = av_clip_int16(s->int_samples[i]);
00930 
00931     align_get_bits(&gb);
00932 
00933     *data_size = s->frame_size * 2;
00934 
00935     return (get_bits_count(&gb)+7)/8;
00936 }
00937 
00938 AVCodec sonic_decoder = {
00939     "sonic",
00940     AVMEDIA_TYPE_AUDIO,
00941     CODEC_ID_SONIC,
00942     sizeof(SonicContext),
00943     sonic_decode_init,
00944     NULL,
00945     sonic_decode_close,
00946     sonic_decode_frame,
00947     .long_name = NULL_IF_CONFIG_SMALL("Sonic"),
00948 };
00949 #endif /* CONFIG_SONIC_DECODER */
00950 
00951 #if CONFIG_SONIC_ENCODER
00952 AVCodec sonic_encoder = {
00953     "sonic",
00954     AVMEDIA_TYPE_AUDIO,
00955     CODEC_ID_SONIC,
00956     sizeof(SonicContext),
00957     sonic_encode_init,
00958     sonic_encode_frame,
00959     sonic_encode_close,
00960     NULL,
00961     .long_name = NULL_IF_CONFIG_SMALL("Sonic"),
00962 };
00963 #endif
00964 
00965 #if CONFIG_SONIC_LS_ENCODER
00966 AVCodec sonic_ls_encoder = {
00967     "sonicls",
00968     AVMEDIA_TYPE_AUDIO,
00969     CODEC_ID_SONIC_LS,
00970     sizeof(SonicContext),
00971     sonic_encode_init,
00972     sonic_encode_frame,
00973     sonic_encode_close,
00974     NULL,
00975     .long_name = NULL_IF_CONFIG_SMALL("Sonic lossless"),
00976 };
00977 #endif

Generated on Fri Sep 16 2011 17:17:43 for FFmpeg by  doxygen 1.7.1