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