libavcodec/alacenc.c

   1 /**
   2  * ALAC audio encoder
   3  * Copyright (c) 2008  Jaikrishnan Menon <realityman@gmx.net>
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #include "avcodec.h"
  23 #include "bitstream.h"
  24 #include "dsputil.h"
  25 #include "lpc.h"
  26 #include "mathops.h"
  27
  28 #define DEFAULT_FRAME_SIZE        4096
  29 #define DEFAULT_SAMPLE_SIZE       16
  30 #define MAX_CHANNELS              8
  31 #define ALAC_EXTRADATA_SIZE       36
  32 #define ALAC_FRAME_HEADER_SIZE    55
  33 #define ALAC_FRAME_FOOTER_SIZE    3
  34
  35 #define ALAC_ESCAPE_CODE          0x1FF
  36 #define ALAC_MAX_LPC_ORDER        30
  37 #define DEFAULT_MAX_PRED_ORDER    6
  38 #define DEFAULT_MIN_PRED_ORDER    4
  39 #define ALAC_MAX_LPC_PRECISION    9
  40 #define ALAC_MAX_LPC_SHIFT        9
  41
  42 #define ALAC_CHMODE_LEFT_RIGHT    0
  43 #define ALAC_CHMODE_LEFT_SIDE     1
  44 #define ALAC_CHMODE_RIGHT_SIDE    2
  45 #define ALAC_CHMODE_MID_SIDE      3
  46
  47 typedef struct RiceContext {
  48     int history_mult;
  49     int initial_history;
  50     int k_modifier;
  51     int rice_modifier;
  52 } RiceContext;
  53
  54 typedef struct LPCContext {
  55     int lpc_order;
  56     int lpc_coeff[ALAC_MAX_LPC_ORDER+1];
  57     int lpc_quant;
  58 } LPCContext;
  59
  60 typedef struct AlacEncodeContext {
  61     int compression_level;
  62     int min_prediction_order;
  63     int max_prediction_order;
  64     int max_coded_frame_size;
  65     int write_sample_size;
  66     int32_t sample_buf[MAX_CHANNELS][DEFAULT_FRAME_SIZE];
  67     int32_t predictor_buf[DEFAULT_FRAME_SIZE];
  68     int interlacing_shift;
  69     int interlacing_leftweight;
  70     PutBitContext pbctx;
  71     RiceContext rc;
  72     LPCContext lpc[MAX_CHANNELS];
  73     DSPContext dspctx;
  74     AVCodecContext *avctx;
  75 } AlacEncodeContext;
  76
  77
  78 static void init_sample_buffers(AlacEncodeContext *s, int16_t *input_samples)
  79 {
  80     int ch, i;
  81
  82     for(ch=0;ch<s->avctx->channels;ch++) {
  83         int16_t *sptr = input_samples + ch;
  84         for(i=0;i<s->avctx->frame_size;i++) {
  85             s->sample_buf[ch][i] = *sptr;
  86             sptr += s->avctx->channels;
  87         }
  88     }
  89 }
  90
  91 static void encode_scalar(AlacEncodeContext *s, int x, int k, int write_sample_size)
  92 {
  93     int divisor, q, r;
  94
  95     k = FFMIN(k, s->rc.k_modifier);
  96     divisor = (1<<k) - 1;
  97     q = x / divisor;
  98     r = x % divisor;
  99
 100     if(q > 8) {
 101         // write escape code and sample value directly
 102         put_bits(&s->pbctx, 9, ALAC_ESCAPE_CODE);
 103         put_bits(&s->pbctx, write_sample_size, x);
 104     } else {
 105         if(q)
 106             put_bits(&s->pbctx, q, (1<<q) - 1);
 107         put_bits(&s->pbctx, 1, 0);
 108
 109         if(k != 1) {
 110             if(r > 0)
 111                 put_bits(&s->pbctx, k, r+1);
 112             else
 113                 put_bits(&s->pbctx, k-1, 0);
 114         }
 115     }
 116 }
 117
 118 static void write_frame_header(AlacEncodeContext *s, int is_verbatim)
 119 {
 120     put_bits(&s->pbctx, 3,  s->avctx->channels-1);          // No. of channels -1
 121     put_bits(&s->pbctx, 16, 0);                             // Seems to be zero
 122     put_bits(&s->pbctx, 1,  1);                             // Sample count is in the header
 123     put_bits(&s->pbctx, 2,  0);                             // FIXME: Wasted bytes field
 124     put_bits(&s->pbctx, 1,  is_verbatim);                   // Audio block is verbatim
 125     put_bits(&s->pbctx, 32, s->avctx->frame_size);          // No. of samples in the frame
 126 }
 127
 128 static void calc_predictor_params(AlacEncodeContext *s, int ch)
 129 {
 130     int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
 131     int shift[MAX_LPC_ORDER];
 132     int opt_order;
 133
 134     opt_order = ff_lpc_calc_coefs(&s->dspctx, s->sample_buf[ch], s->avctx->frame_size, s->min_prediction_order, s->max_prediction_order,
 135                                    ALAC_MAX_LPC_PRECISION, coefs, shift, 1, ORDER_METHOD_EST, ALAC_MAX_LPC_SHIFT, 1);
 136
 137     s->lpc[ch].lpc_order = opt_order;
 138     s->lpc[ch].lpc_quant = shift[opt_order-1];
 139     memcpy(s->lpc[ch].lpc_coeff, coefs[opt_order-1], opt_order*sizeof(int));
 140 }
 141
 142 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
 143 {
 144     int i, best;
 145     int32_t lt, rt;
 146     uint64_t sum[4];
 147     uint64_t score[4];
 148
 149     /* calculate sum of 2nd order residual for each channel */
 150     sum[0] = sum[1] = sum[2] = sum[3] = 0;
 151     for(i=2; i<n; i++) {
 152         lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
 153         rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
 154         sum[2] += FFABS((lt + rt) >> 1);
 155         sum[3] += FFABS(lt - rt);
 156         sum[0] += FFABS(lt);
 157         sum[1] += FFABS(rt);
 158     }
 159
 160     /* calculate score for each mode */
 161     score[0] = sum[0] + sum[1];
 162     score[1] = sum[0] + sum[3];
 163     score[2] = sum[1] + sum[3];
 164     score[3] = sum[2] + sum[3];
 165
 166     /* return mode with lowest score */
 167     best = 0;
 168     for(i=1; i<4; i++) {
 169         if(score[i] < score[best]) {
 170             best = i;
 171         }
 172     }
 173     return best;
 174 }
 175
 176 static void alac_stereo_decorrelation(AlacEncodeContext *s)
 177 {
 178     int32_t *left = s->sample_buf[0], *right = s->sample_buf[1];
 179     int i, mode, n = s->avctx->frame_size;
 180     int32_t tmp;
 181
 182     mode = estimate_stereo_mode(left, right, n);
 183
 184     switch(mode)
 185     {
 186         case ALAC_CHMODE_LEFT_RIGHT:
 187             s->interlacing_leftweight = 0;
 188             s->interlacing_shift = 0;
 189             break;
 190
 191         case ALAC_CHMODE_LEFT_SIDE:
 192             for(i=0; i<n; i++) {
 193                 right[i] = left[i] - right[i];
 194             }
 195             s->interlacing_leftweight = 1;
 196             s->interlacing_shift = 0;
 197             break;
 198
 199         case ALAC_CHMODE_RIGHT_SIDE:
 200             for(i=0; i<n; i++) {
 201                 tmp = right[i];
 202                 right[i] = left[i] - right[i];
 203                 left[i] = tmp + (right[i] >> 31);
 204             }
 205             s->interlacing_leftweight = 1;
 206             s->interlacing_shift = 31;
 207             break;
 208
 209         default:
 210             for(i=0; i<n; i++) {
 211                 tmp = left[i];
 212                 left[i] = (tmp + right[i]) >> 1;
 213                 right[i] = tmp - right[i];
 214             }
 215             s->interlacing_leftweight = 1;
 216             s->interlacing_shift = 1;
 217             break;
 218     }
 219 }
 220
 221 static void alac_linear_predictor(AlacEncodeContext *s, int ch)
 222 {
 223     int i;
 224     LPCContext lpc = s->lpc[ch];
 225
 226     if(lpc.lpc_order == 31) {
 227         s->predictor_buf[0] = s->sample_buf[ch][0];
 228
 229         for(i=1; i<s->avctx->frame_size; i++)
 230             s->predictor_buf[i] = s->sample_buf[ch][i] - s->sample_buf[ch][i-1];
 231
 232         return;
 233     }
 234
 235     // generalised linear predictor
 236
 237     if(lpc.lpc_order > 0) {
 238         int32_t *samples  = s->sample_buf[ch];
 239         int32_t *residual = s->predictor_buf;
 240
 241         // generate warm-up samples
 242         residual[0] = samples[0];
 243         for(i=1;i<=lpc.lpc_order;i++)
 244             residual[i] = samples[i] - samples[i-1];
 245
 246         // perform lpc on remaining samples
 247         for(i = lpc.lpc_order + 1; i < s->avctx->frame_size; i++) {
 248             int sum = 1 << (lpc.lpc_quant - 1), res_val, j;
 249
 250             for (j = 0; j < lpc.lpc_order; j++) {
 251                 sum += (samples[lpc.lpc_order-j] - samples[0]) *
 252                         lpc.lpc_coeff[j];
 253             }
 254
 255             sum >>= lpc.lpc_quant;
 256             sum += samples[0];
 257             residual[i] = sign_extend(samples[lpc.lpc_order+1] - sum,
 258                                       s->write_sample_size);
 259             res_val = residual[i];
 260
 261             if(res_val) {
 262                 int index = lpc.lpc_order - 1;
 263                 int neg = (res_val < 0);
 264
 265                 while(index >= 0 && (neg ? (res_val < 0):(res_val > 0))) {
 266                     int val = samples[0] - samples[lpc.lpc_order - index];
 267                     int sign = (val ? FFSIGN(val) : 0);
 268
 269                     if(neg)
 270                         sign*=-1;
 271
 272                     lpc.lpc_coeff[index] -= sign;
 273                     val *= sign;
 274                     res_val -= ((val >> lpc.lpc_quant) *
 275                             (lpc.lpc_order - index));
 276                     index--;
 277                 }
 278             }
 279             samples++;
 280         }
 281     }
 282 }
 283
 284 static void alac_entropy_coder(AlacEncodeContext *s)
 285 {
 286     unsigned int history = s->rc.initial_history;
 287     int sign_modifier = 0, i, k;
 288     int32_t *samples = s->predictor_buf;
 289
 290     for(i=0;i < s->avctx->frame_size;) {
 291         int x;
 292
 293         k = av_log2((history >> 9) + 3);
 294
 295         x = -2*(*samples)-1;
 296         x ^= (x>>31);
 297
 298         samples++;
 299         i++;
 300
 301         encode_scalar(s, x - sign_modifier, k, s->write_sample_size);
 302
 303         history += x * s->rc.history_mult
 304                    - ((history * s->rc.history_mult) >> 9);
 305
 306         sign_modifier = 0;
 307         if(x > 0xFFFF)
 308             history = 0xFFFF;
 309
 310         if((history < 128) && (i < s->avctx->frame_size)) {
 311             unsigned int block_size = 0;
 312
 313             k = 7 - av_log2(history) + ((history + 16) >> 6);
 314
 315             while((*samples == 0) && (i < s->avctx->frame_size)) {
 316                 samples++;
 317                 i++;
 318                 block_size++;
 319             }
 320             encode_scalar(s, block_size, k, 16);
 321
 322             sign_modifier = (block_size <= 0xFFFF);
 323
 324             history = 0;
 325         }
 326
 327     }
 328 }
 329
 330 static void write_compressed_frame(AlacEncodeContext *s)
 331 {
 332     int i, j;
 333
 334     /* only simple mid/side decorrelation supported as of now */
 335     if(s->avctx->channels == 2)
 336         alac_stereo_decorrelation(s);
 337     put_bits(&s->pbctx, 8, s->interlacing_shift);
 338     put_bits(&s->pbctx, 8, s->interlacing_leftweight);
 339
 340     for(i=0;i<s->avctx->channels;i++) {
 341
 342         calc_predictor_params(s, i);
 343
 344         put_bits(&s->pbctx, 4, 0);  // prediction type : currently only type 0 has been RE'd
 345         put_bits(&s->pbctx, 4, s->lpc[i].lpc_quant);
 346
 347         put_bits(&s->pbctx, 3, s->rc.rice_modifier);
 348         put_bits(&s->pbctx, 5, s->lpc[i].lpc_order);
 349         // predictor coeff. table
 350         for(j=0;j<s->lpc[i].lpc_order;j++) {
 351             put_sbits(&s->pbctx, 16, s->lpc[i].lpc_coeff[j]);
 352         }
 353     }
 354
 355     // apply lpc and entropy coding to audio samples
 356
 357     for(i=0;i<s->avctx->channels;i++) {
 358         alac_linear_predictor(s, i);
 359         alac_entropy_coder(s);
 360     }
 361 }
 362
 363 static av_cold int alac_encode_init(AVCodecContext *avctx)
 364 {
 365     AlacEncodeContext *s    = avctx->priv_data;
 366     uint8_t *alac_extradata = av_mallocz(ALAC_EXTRADATA_SIZE+1);
 367
 368     avctx->frame_size      = DEFAULT_FRAME_SIZE;
 369     avctx->bits_per_coded_sample = DEFAULT_SAMPLE_SIZE;
 370
 371     if(avctx->sample_fmt != SAMPLE_FMT_S16) {
 372         av_log(avctx, AV_LOG_ERROR, "only pcm_s16 input samples are supported\n");
 373         return -1;
 374     }
 375
 376     // Set default compression level
 377     if(avctx->compression_level == FF_COMPRESSION_DEFAULT)
 378         s->compression_level = 1;
 379     else
 380         s->compression_level = av_clip(avctx->compression_level, 0, 1);
 381
 382     // Initialize default Rice parameters
 383     s->rc.history_mult    = 40;
 384     s->rc.initial_history = 10;
 385     s->rc.k_modifier      = 14;
 386     s->rc.rice_modifier   = 4;
 387
 388     s->max_coded_frame_size = (ALAC_FRAME_HEADER_SIZE + ALAC_FRAME_FOOTER_SIZE +
 389                                avctx->frame_size*avctx->channels*avctx->bits_per_coded_sample)>>3;
 390
 391     s->write_sample_size  = avctx->bits_per_coded_sample + avctx->channels - 1; // FIXME: consider wasted_bytes
 392
 393     AV_WB32(alac_extradata,    ALAC_EXTRADATA_SIZE);
 394     AV_WB32(alac_extradata+4,  MKBETAG('a','l','a','c'));
 395     AV_WB32(alac_extradata+12, avctx->frame_size);
 396     AV_WB8 (alac_extradata+17, avctx->bits_per_coded_sample);
 397     AV_WB8 (alac_extradata+21, avctx->channels);
 398     AV_WB32(alac_extradata+24, s->max_coded_frame_size);
 399     AV_WB32(alac_extradata+28, avctx->sample_rate*avctx->channels*avctx->bits_per_coded_sample); // average bitrate
 400     AV_WB32(alac_extradata+32, avctx->sample_rate);
 401
 402     // Set relevant extradata fields
 403     if(s->compression_level > 0) {
 404         AV_WB8(alac_extradata+18, s->rc.history_mult);
 405         AV_WB8(alac_extradata+19, s->rc.initial_history);
 406         AV_WB8(alac_extradata+20, s->rc.k_modifier);
 407     }
 408
 409     s->min_prediction_order = DEFAULT_MIN_PRED_ORDER;
 410     if(avctx->min_prediction_order >= 0) {
 411         if(avctx->min_prediction_order < MIN_LPC_ORDER ||
 412            avctx->min_prediction_order > ALAC_MAX_LPC_ORDER) {
 413             av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", avctx->min_prediction_order);
 414                 return -1;
 415         }
 416
 417         s->min_prediction_order = avctx->min_prediction_order;
 418     }
 419
 420     s->max_prediction_order = DEFAULT_MAX_PRED_ORDER;
 421     if(avctx->max_prediction_order >= 0) {
 422         if(avctx->max_prediction_order < MIN_LPC_ORDER ||
 423            avctx->max_prediction_order > ALAC_MAX_LPC_ORDER) {
 424             av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", avctx->max_prediction_order);
 425                 return -1;
 426         }
 427
 428         s->max_prediction_order = avctx->max_prediction_order;
 429     }
 430
 431     if(s->max_prediction_order < s->min_prediction_order) {
 432         av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
 433                s->min_prediction_order, s->max_prediction_order);
 434         return -1;
 435     }
 436
 437     avctx->extradata = alac_extradata;
 438     avctx->extradata_size = ALAC_EXTRADATA_SIZE;
 439
 440     avctx->coded_frame = avcodec_alloc_frame();
 441     avctx->coded_frame->key_frame = 1;
 442
 443     s->avctx = avctx;
 444     dsputil_init(&s->dspctx, avctx);
 445
 446     return 0;
 447 }
 448
 449 static int alac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
 450                              int buf_size, void *data)
 451 {
 452     AlacEncodeContext *s = avctx->priv_data;
 453     PutBitContext *pb = &s->pbctx;
 454     int i, out_bytes, verbatim_flag = 0;
 455
 456     if(avctx->frame_size > DEFAULT_FRAME_SIZE) {
 457         av_log(avctx, AV_LOG_ERROR, "input frame size exceeded\n");
 458         return -1;
 459     }
 460
 461     if(buf_size < 2*s->max_coded_frame_size) {
 462         av_log(avctx, AV_LOG_ERROR, "buffer size is too small\n");
 463         return -1;
 464     }
 465
 466 verbatim:
 467     init_put_bits(pb, frame, buf_size);
 468
 469     if((s->compression_level == 0) || verbatim_flag) {
 470         // Verbatim mode
 471         int16_t *samples = data;
 472         write_frame_header(s, 1);
 473         for(i=0; i<avctx->frame_size*avctx->channels; i++) {
 474             put_sbits(pb, 16, *samples++);
 475         }
 476     } else {
 477         init_sample_buffers(s, data);
 478         write_frame_header(s, 0);
 479         write_compressed_frame(s);
 480     }
 481
 482     put_bits(pb, 3, 7);
 483     flush_put_bits(pb);
 484     out_bytes = put_bits_count(pb) >> 3;
 485
 486     if(out_bytes > s->max_coded_frame_size) {
 487         /* frame too large. use verbatim mode */
 488         if(verbatim_flag || (s->compression_level == 0)) {
 489             /* still too large. must be an error. */
 490             av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
 491             return -1;
 492         }
 493         verbatim_flag = 1;
 494         goto verbatim;
 495     }
 496
 497     return out_bytes;
 498 }
 499
 500 static av_cold int alac_encode_close(AVCodecContext *avctx)
 501 {
 502     av_freep(&avctx->extradata);
 503     avctx->extradata_size = 0;
 504     av_freep(&avctx->coded_frame);
 505     return 0;
 506 }
 507
 508 AVCodec alac_encoder = {
 509     "alac",
 510     CODEC_TYPE_AUDIO,
 511     CODEC_ID_ALAC,
 512     sizeof(AlacEncodeContext),
 513     alac_encode_init,
 514     alac_encode_frame,
 515     alac_encode_close,
 516     .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
 517     .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
 518 };