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 "put_bits.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 47typedef struct RiceContext { 48 int history_mult; 49 int initial_history; 50 int k_modifier; 51 int rice_modifier; 52} RiceContext; 53 54typedef struct LPCContext { 55 int lpc_order; 56 int lpc_coeff[ALAC_MAX_LPC_ORDER+1]; 57 int lpc_quant; 58} LPCContext; 59 60typedef 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 78static 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 91static 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 118static 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_bits32(&s->pbctx, s->avctx->frame_size); // No. of samples in the frame 126} 127 128static 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 if (s->compression_level == 1) { 135 s->lpc[ch].lpc_order = 6; 136 s->lpc[ch].lpc_quant = 6; 137 s->lpc[ch].lpc_coeff[0] = 160; 138 s->lpc[ch].lpc_coeff[1] = -190; 139 s->lpc[ch].lpc_coeff[2] = 170; 140 s->lpc[ch].lpc_coeff[3] = -130; 141 s->lpc[ch].lpc_coeff[4] = 80; 142 s->lpc[ch].lpc_coeff[5] = -25; 143 } else { 144 opt_order = ff_lpc_calc_coefs(&s->dspctx, s->sample_buf[ch], 145 s->avctx->frame_size, 146 s->min_prediction_order, 147 s->max_prediction_order, 148 ALAC_MAX_LPC_PRECISION, coefs, shift, 1, 149 ORDER_METHOD_EST, ALAC_MAX_LPC_SHIFT, 1); 150 151 s->lpc[ch].lpc_order = opt_order; 152 s->lpc[ch].lpc_quant = shift[opt_order-1]; 153 memcpy(s->lpc[ch].lpc_coeff, coefs[opt_order-1], opt_order*sizeof(int)); 154 } 155} 156 157static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n) 158{ 159 int i, best; 160 int32_t lt, rt; 161 uint64_t sum[4]; 162 uint64_t score[4]; 163 164 /* calculate sum of 2nd order residual for each channel */ 165 sum[0] = sum[1] = sum[2] = sum[3] = 0; 166 for(i=2; i<n; i++) { 167 lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2]; 168 rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2]; 169 sum[2] += FFABS((lt + rt) >> 1); 170 sum[3] += FFABS(lt - rt); 171 sum[0] += FFABS(lt); 172 sum[1] += FFABS(rt); 173 } 174 175 /* calculate score for each mode */ 176 score[0] = sum[0] + sum[1]; 177 score[1] = sum[0] + sum[3]; 178 score[2] = sum[1] + sum[3]; 179 score[3] = sum[2] + sum[3]; 180 181 /* return mode with lowest score */ 182 best = 0; 183 for(i=1; i<4; i++) { 184 if(score[i] < score[best]) { 185 best = i; 186 } 187 } 188 return best; 189} 190 191static void alac_stereo_decorrelation(AlacEncodeContext *s) 192{ 193 int32_t *left = s->sample_buf[0], *right = s->sample_buf[1]; 194 int i, mode, n = s->avctx->frame_size; 195 int32_t tmp; 196 197 mode = estimate_stereo_mode(left, right, n); 198 199 switch(mode) 200 { 201 case ALAC_CHMODE_LEFT_RIGHT: 202 s->interlacing_leftweight = 0; 203 s->interlacing_shift = 0; 204 break; 205 206 case ALAC_CHMODE_LEFT_SIDE: 207 for(i=0; i<n; i++) { 208 right[i] = left[i] - right[i]; 209 } 210 s->interlacing_leftweight = 1; 211 s->interlacing_shift = 0; 212 break; 213 214 case ALAC_CHMODE_RIGHT_SIDE: 215 for(i=0; i<n; i++) { 216 tmp = right[i]; 217 right[i] = left[i] - right[i]; 218 left[i] = tmp + (right[i] >> 31); 219 } 220 s->interlacing_leftweight = 1; 221 s->interlacing_shift = 31; 222 break; 223 224 default: 225 for(i=0; i<n; i++) { 226 tmp = left[i]; 227 left[i] = (tmp + right[i]) >> 1; 228 right[i] = tmp - right[i]; 229 } 230 s->interlacing_leftweight = 1; 231 s->interlacing_shift = 1; 232 break; 233 } 234} 235 236static void alac_linear_predictor(AlacEncodeContext *s, int ch) 237{ 238 int i; 239 LPCContext lpc = s->lpc[ch]; 240 241 if(lpc.lpc_order == 31) { 242 s->predictor_buf[0] = s->sample_buf[ch][0]; 243 244 for(i=1; i<s->avctx->frame_size; i++) 245 s->predictor_buf[i] = s->sample_buf[ch][i] - s->sample_buf[ch][i-1]; 246 247 return; 248 } 249 250 // generalised linear predictor 251 252 if(lpc.lpc_order > 0) { 253 int32_t *samples = s->sample_buf[ch]; 254 int32_t *residual = s->predictor_buf; 255 256 // generate warm-up samples 257 residual[0] = samples[0]; 258 for(i=1;i<=lpc.lpc_order;i++) 259 residual[i] = samples[i] - samples[i-1]; 260 261 // perform lpc on remaining samples 262 for(i = lpc.lpc_order + 1; i < s->avctx->frame_size; i++) { 263 int sum = 1 << (lpc.lpc_quant - 1), res_val, j; 264 265 for (j = 0; j < lpc.lpc_order; j++) { 266 sum += (samples[lpc.lpc_order-j] - samples[0]) * 267 lpc.lpc_coeff[j]; 268 } 269 270 sum >>= lpc.lpc_quant; 271 sum += samples[0]; 272 residual[i] = sign_extend(samples[lpc.lpc_order+1] - sum, 273 s->write_sample_size); 274 res_val = residual[i]; 275 276 if(res_val) { 277 int index = lpc.lpc_order - 1; 278 int neg = (res_val < 0); 279 280 while(index >= 0 && (neg ? (res_val < 0):(res_val > 0))) { 281 int val = samples[0] - samples[lpc.lpc_order - index]; 282 int sign = (val ? FFSIGN(val) : 0); 283 284 if(neg) 285 sign*=-1; 286 287 lpc.lpc_coeff[index] -= sign; 288 val *= sign; 289 res_val -= ((val >> lpc.lpc_quant) * 290 (lpc.lpc_order - index)); 291 index--; 292 } 293 } 294 samples++; 295 } 296 } 297} 298 299static void alac_entropy_coder(AlacEncodeContext *s) 300{ 301 unsigned int history = s->rc.initial_history; 302 int sign_modifier = 0, i, k; 303 int32_t *samples = s->predictor_buf; 304 305 for(i=0;i < s->avctx->frame_size;) { 306 int x; 307 308 k = av_log2((history >> 9) + 3); 309 310 x = -2*(*samples)-1; 311 x ^= (x>>31); 312 313 samples++; 314 i++; 315 316 encode_scalar(s, x - sign_modifier, k, s->write_sample_size); 317 318 history += x * s->rc.history_mult 319 - ((history * s->rc.history_mult) >> 9); 320 321 sign_modifier = 0; 322 if(x > 0xFFFF) 323 history = 0xFFFF; 324 325 if((history < 128) && (i < s->avctx->frame_size)) { 326 unsigned int block_size = 0; 327 328 k = 7 - av_log2(history) + ((history + 16) >> 6); 329 330 while((*samples == 0) && (i < s->avctx->frame_size)) { 331 samples++; 332 i++; 333 block_size++; 334 } 335 encode_scalar(s, block_size, k, 16); 336 337 sign_modifier = (block_size <= 0xFFFF); 338 339 history = 0; 340 } 341 342 } 343} 344 345static void write_compressed_frame(AlacEncodeContext *s) 346{ 347 int i, j; 348 349 if(s->avctx->channels == 2) 350 alac_stereo_decorrelation(s); 351 put_bits(&s->pbctx, 8, s->interlacing_shift); 352 put_bits(&s->pbctx, 8, s->interlacing_leftweight); 353 354 for(i=0;i<s->avctx->channels;i++) { 355 356 calc_predictor_params(s, i); 357 358 put_bits(&s->pbctx, 4, 0); // prediction type : currently only type 0 has been RE'd 359 put_bits(&s->pbctx, 4, s->lpc[i].lpc_quant); 360 361 put_bits(&s->pbctx, 3, s->rc.rice_modifier); 362 put_bits(&s->pbctx, 5, s->lpc[i].lpc_order); 363 // predictor coeff. table 364 for(j=0;j<s->lpc[i].lpc_order;j++) { 365 put_sbits(&s->pbctx, 16, s->lpc[i].lpc_coeff[j]); 366 } 367 } 368 369 // apply lpc and entropy coding to audio samples 370 371 for(i=0;i<s->avctx->channels;i++) { 372 alac_linear_predictor(s, i); 373 alac_entropy_coder(s); 374 } 375} 376 377static av_cold int alac_encode_init(AVCodecContext *avctx) 378{ 379 AlacEncodeContext *s = avctx->priv_data; 380 uint8_t *alac_extradata = av_mallocz(ALAC_EXTRADATA_SIZE+1); 381 382 avctx->frame_size = DEFAULT_FRAME_SIZE; 383 avctx->bits_per_coded_sample = DEFAULT_SAMPLE_SIZE; 384 385 if(avctx->sample_fmt != SAMPLE_FMT_S16) { 386 av_log(avctx, AV_LOG_ERROR, "only pcm_s16 input samples are supported\n"); 387 return -1; 388 } 389 390 // Set default compression level 391 if(avctx->compression_level == FF_COMPRESSION_DEFAULT) 392 s->compression_level = 2; 393 else 394 s->compression_level = av_clip(avctx->compression_level, 0, 2); 395 396 // Initialize default Rice parameters 397 s->rc.history_mult = 40; 398 s->rc.initial_history = 10; 399 s->rc.k_modifier = 14; 400 s->rc.rice_modifier = 4; 401 402 s->max_coded_frame_size = 8 + (avctx->frame_size*avctx->channels*avctx->bits_per_coded_sample>>3); 403 404 s->write_sample_size = avctx->bits_per_coded_sample + avctx->channels - 1; // FIXME: consider wasted_bytes 405 406 AV_WB32(alac_extradata, ALAC_EXTRADATA_SIZE); 407 AV_WB32(alac_extradata+4, MKBETAG('a','l','a','c')); 408 AV_WB32(alac_extradata+12, avctx->frame_size); 409 AV_WB8 (alac_extradata+17, avctx->bits_per_coded_sample); 410 AV_WB8 (alac_extradata+21, avctx->channels); 411 AV_WB32(alac_extradata+24, s->max_coded_frame_size); 412 AV_WB32(alac_extradata+28, avctx->sample_rate*avctx->channels*avctx->bits_per_coded_sample); // average bitrate 413 AV_WB32(alac_extradata+32, avctx->sample_rate); 414 415 // Set relevant extradata fields 416 if(s->compression_level > 0) { 417 AV_WB8(alac_extradata+18, s->rc.history_mult); 418 AV_WB8(alac_extradata+19, s->rc.initial_history); 419 AV_WB8(alac_extradata+20, s->rc.k_modifier); 420 } 421 422 s->min_prediction_order = DEFAULT_MIN_PRED_ORDER; 423 if(avctx->min_prediction_order >= 0) { 424 if(avctx->min_prediction_order < MIN_LPC_ORDER || 425 avctx->min_prediction_order > ALAC_MAX_LPC_ORDER) { 426 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", avctx->min_prediction_order); 427 return -1; 428 } 429 430 s->min_prediction_order = avctx->min_prediction_order; 431 } 432 433 s->max_prediction_order = DEFAULT_MAX_PRED_ORDER; 434 if(avctx->max_prediction_order >= 0) { 435 if(avctx->max_prediction_order < MIN_LPC_ORDER || 436 avctx->max_prediction_order > ALAC_MAX_LPC_ORDER) { 437 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", avctx->max_prediction_order); 438 return -1; 439 } 440 441 s->max_prediction_order = avctx->max_prediction_order; 442 } 443 444 if(s->max_prediction_order < s->min_prediction_order) { 445 av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n", 446 s->min_prediction_order, s->max_prediction_order); 447 return -1; 448 } 449 450 avctx->extradata = alac_extradata; 451 avctx->extradata_size = ALAC_EXTRADATA_SIZE; 452 453 avctx->coded_frame = avcodec_alloc_frame(); 454 avctx->coded_frame->key_frame = 1; 455 456 s->avctx = avctx; 457 dsputil_init(&s->dspctx, avctx); 458 459 return 0; 460} 461 462static int alac_encode_frame(AVCodecContext *avctx, uint8_t *frame, 463 int buf_size, void *data) 464{ 465 AlacEncodeContext *s = avctx->priv_data; 466 PutBitContext *pb = &s->pbctx; 467 int i, out_bytes, verbatim_flag = 0; 468 469 if(avctx->frame_size > DEFAULT_FRAME_SIZE) { 470 av_log(avctx, AV_LOG_ERROR, "input frame size exceeded\n"); 471 return -1; 472 } 473 474 if(buf_size < 2*s->max_coded_frame_size) { 475 av_log(avctx, AV_LOG_ERROR, "buffer size is too small\n"); 476 return -1; 477 } 478 479verbatim: 480 init_put_bits(pb, frame, buf_size); 481 482 if((s->compression_level == 0) || verbatim_flag) { 483 // Verbatim mode 484 int16_t *samples = data; 485 write_frame_header(s, 1); 486 for(i=0; i<avctx->frame_size*avctx->channels; i++) { 487 put_sbits(pb, 16, *samples++); 488 } 489 } else { 490 init_sample_buffers(s, data); 491 write_frame_header(s, 0); 492 write_compressed_frame(s); 493 } 494 495 put_bits(pb, 3, 7); 496 flush_put_bits(pb); 497 out_bytes = put_bits_count(pb) >> 3; 498 499 if(out_bytes > s->max_coded_frame_size) { 500 /* frame too large. use verbatim mode */ 501 if(verbatim_flag || (s->compression_level == 0)) { 502 /* still too large. must be an error. */ 503 av_log(avctx, AV_LOG_ERROR, "error encoding frame\n"); 504 return -1; 505 } 506 verbatim_flag = 1; 507 goto verbatim; 508 } 509 510 return out_bytes; 511} 512 513static av_cold int alac_encode_close(AVCodecContext *avctx) 514{ 515 av_freep(&avctx->extradata); 516 avctx->extradata_size = 0; 517 av_freep(&avctx->coded_frame); 518 return 0; 519} 520 521AVCodec alac_encoder = { 522 "alac", 523 AVMEDIA_TYPE_AUDIO, 524 CODEC_ID_ALAC, 525 sizeof(AlacEncodeContext), 526 alac_encode_init, 527 alac_encode_frame, 528 alac_encode_close, 529 .capabilities = CODEC_CAP_SMALL_LAST_FRAME, 530 .sample_fmts = (const enum SampleFormat[]){ SAMPLE_FMT_S16, SAMPLE_FMT_NONE}, 531 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"), 532}; 533