1/** 2 * ALAC audio encoder 3 * Copyright (c) 2008 Jaikrishnan Menon <realityman@gmx.net> 4 * 5 * This file is part of Libav. 6 * 7 * Libav 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 * Libav 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 Libav; 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 AlacLPCContext { 55 int lpc_order; 56 int lpc_coeff[ALAC_MAX_LPC_ORDER+1]; 57 int lpc_quant; 58} AlacLPCContext; 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 AlacLPCContext lpc[MAX_CHANNELS]; 73 LPCContext lpc_ctx; 74 AVCodecContext *avctx; 75} AlacEncodeContext; 76 77 78static void init_sample_buffers(AlacEncodeContext *s, 79 const int16_t *input_samples) 80{ 81 int ch, i; 82 83 for (ch = 0; ch < s->avctx->channels; ch++) { 84 const int16_t *sptr = input_samples + ch; 85 for (i = 0; i < s->avctx->frame_size; i++) { 86 s->sample_buf[ch][i] = *sptr; 87 sptr += s->avctx->channels; 88 } 89 } 90} 91 92static void encode_scalar(AlacEncodeContext *s, int x, 93 int k, int write_sample_size) 94{ 95 int divisor, q, r; 96 97 k = FFMIN(k, s->rc.k_modifier); 98 divisor = (1<<k) - 1; 99 q = x / divisor; 100 r = x % divisor; 101 102 if (q > 8) { 103 // write escape code and sample value directly 104 put_bits(&s->pbctx, 9, ALAC_ESCAPE_CODE); 105 put_bits(&s->pbctx, write_sample_size, x); 106 } else { 107 if (q) 108 put_bits(&s->pbctx, q, (1<<q) - 1); 109 put_bits(&s->pbctx, 1, 0); 110 111 if (k != 1) { 112 if (r > 0) 113 put_bits(&s->pbctx, k, r+1); 114 else 115 put_bits(&s->pbctx, k-1, 0); 116 } 117 } 118} 119 120static void write_frame_header(AlacEncodeContext *s, int is_verbatim) 121{ 122 put_bits(&s->pbctx, 3, s->avctx->channels-1); // No. of channels -1 123 put_bits(&s->pbctx, 16, 0); // Seems to be zero 124 put_bits(&s->pbctx, 1, 1); // Sample count is in the header 125 put_bits(&s->pbctx, 2, 0); // FIXME: Wasted bytes field 126 put_bits(&s->pbctx, 1, is_verbatim); // Audio block is verbatim 127 put_bits32(&s->pbctx, s->avctx->frame_size); // No. of samples in the frame 128} 129 130static void calc_predictor_params(AlacEncodeContext *s, int ch) 131{ 132 int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER]; 133 int shift[MAX_LPC_ORDER]; 134 int opt_order; 135 136 if (s->compression_level == 1) { 137 s->lpc[ch].lpc_order = 6; 138 s->lpc[ch].lpc_quant = 6; 139 s->lpc[ch].lpc_coeff[0] = 160; 140 s->lpc[ch].lpc_coeff[1] = -190; 141 s->lpc[ch].lpc_coeff[2] = 170; 142 s->lpc[ch].lpc_coeff[3] = -130; 143 s->lpc[ch].lpc_coeff[4] = 80; 144 s->lpc[ch].lpc_coeff[5] = -25; 145 } else { 146 opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, s->sample_buf[ch], 147 s->avctx->frame_size, 148 s->min_prediction_order, 149 s->max_prediction_order, 150 ALAC_MAX_LPC_PRECISION, coefs, shift, 151 FF_LPC_TYPE_LEVINSON, 0, 152 ORDER_METHOD_EST, ALAC_MAX_LPC_SHIFT, 1); 153 154 s->lpc[ch].lpc_order = opt_order; 155 s->lpc[ch].lpc_quant = shift[opt_order-1]; 156 memcpy(s->lpc[ch].lpc_coeff, coefs[opt_order-1], opt_order*sizeof(int)); 157 } 158} 159 160static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n) 161{ 162 int i, best; 163 int32_t lt, rt; 164 uint64_t sum[4]; 165 uint64_t score[4]; 166 167 /* calculate sum of 2nd order residual for each channel */ 168 sum[0] = sum[1] = sum[2] = sum[3] = 0; 169 for (i = 2; i < n; i++) { 170 lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2]; 171 rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2]; 172 sum[2] += FFABS((lt + rt) >> 1); 173 sum[3] += FFABS(lt - rt); 174 sum[0] += FFABS(lt); 175 sum[1] += FFABS(rt); 176 } 177 178 /* calculate score for each mode */ 179 score[0] = sum[0] + sum[1]; 180 score[1] = sum[0] + sum[3]; 181 score[2] = sum[1] + sum[3]; 182 score[3] = sum[2] + sum[3]; 183 184 /* return mode with lowest score */ 185 best = 0; 186 for (i = 1; i < 4; i++) { 187 if (score[i] < score[best]) { 188 best = i; 189 } 190 } 191 return best; 192} 193 194static void alac_stereo_decorrelation(AlacEncodeContext *s) 195{ 196 int32_t *left = s->sample_buf[0], *right = s->sample_buf[1]; 197 int i, mode, n = s->avctx->frame_size; 198 int32_t tmp; 199 200 mode = estimate_stereo_mode(left, right, n); 201 202 switch(mode) 203 { 204 case ALAC_CHMODE_LEFT_RIGHT: 205 s->interlacing_leftweight = 0; 206 s->interlacing_shift = 0; 207 break; 208 209 case ALAC_CHMODE_LEFT_SIDE: 210 for (i = 0; i < n; i++) { 211 right[i] = left[i] - right[i]; 212 } 213 s->interlacing_leftweight = 1; 214 s->interlacing_shift = 0; 215 break; 216 217 case ALAC_CHMODE_RIGHT_SIDE: 218 for (i = 0; i < n; i++) { 219 tmp = right[i]; 220 right[i] = left[i] - right[i]; 221 left[i] = tmp + (right[i] >> 31); 222 } 223 s->interlacing_leftweight = 1; 224 s->interlacing_shift = 31; 225 break; 226 227 default: 228 for (i = 0; i < n; i++) { 229 tmp = left[i]; 230 left[i] = (tmp + right[i]) >> 1; 231 right[i] = tmp - right[i]; 232 } 233 s->interlacing_leftweight = 1; 234 s->interlacing_shift = 1; 235 break; 236 } 237} 238 239static void alac_linear_predictor(AlacEncodeContext *s, int ch) 240{ 241 int i; 242 AlacLPCContext lpc = s->lpc[ch]; 243 244 if (lpc.lpc_order == 31) { 245 s->predictor_buf[0] = s->sample_buf[ch][0]; 246 247 for (i = 1; i < s->avctx->frame_size; i++) 248 s->predictor_buf[i] = s->sample_buf[ch][i] - s->sample_buf[ch][i-1]; 249 250 return; 251 } 252 253 // generalised linear predictor 254 255 if (lpc.lpc_order > 0) { 256 int32_t *samples = s->sample_buf[ch]; 257 int32_t *residual = s->predictor_buf; 258 259 // generate warm-up samples 260 residual[0] = samples[0]; 261 for (i = 1; i <= lpc.lpc_order; i++) 262 residual[i] = samples[i] - samples[i-1]; 263 264 // perform lpc on remaining samples 265 for (i = lpc.lpc_order + 1; i < s->avctx->frame_size; i++) { 266 int sum = 1 << (lpc.lpc_quant - 1), res_val, j; 267 268 for (j = 0; j < lpc.lpc_order; j++) { 269 sum += (samples[lpc.lpc_order-j] - samples[0]) * 270 lpc.lpc_coeff[j]; 271 } 272 273 sum >>= lpc.lpc_quant; 274 sum += samples[0]; 275 residual[i] = sign_extend(samples[lpc.lpc_order+1] - sum, 276 s->write_sample_size); 277 res_val = residual[i]; 278 279 if(res_val) { 280 int index = lpc.lpc_order - 1; 281 int neg = (res_val < 0); 282 283 while(index >= 0 && (neg ? (res_val < 0):(res_val > 0))) { 284 int val = samples[0] - samples[lpc.lpc_order - index]; 285 int sign = (val ? FFSIGN(val) : 0); 286 287 if(neg) 288 sign*=-1; 289 290 lpc.lpc_coeff[index] -= sign; 291 val *= sign; 292 res_val -= ((val >> lpc.lpc_quant) * 293 (lpc.lpc_order - index)); 294 index--; 295 } 296 } 297 samples++; 298 } 299 } 300} 301 302static void alac_entropy_coder(AlacEncodeContext *s) 303{ 304 unsigned int history = s->rc.initial_history; 305 int sign_modifier = 0, i, k; 306 int32_t *samples = s->predictor_buf; 307 308 for (i = 0; i < s->avctx->frame_size;) { 309 int x; 310 311 k = av_log2((history >> 9) + 3); 312 313 x = -2*(*samples)-1; 314 x ^= (x>>31); 315 316 samples++; 317 i++; 318 319 encode_scalar(s, x - sign_modifier, k, s->write_sample_size); 320 321 history += x * s->rc.history_mult 322 - ((history * s->rc.history_mult) >> 9); 323 324 sign_modifier = 0; 325 if (x > 0xFFFF) 326 history = 0xFFFF; 327 328 if (history < 128 && i < s->avctx->frame_size) { 329 unsigned int block_size = 0; 330 331 k = 7 - av_log2(history) + ((history + 16) >> 6); 332 333 while (*samples == 0 && i < s->avctx->frame_size) { 334 samples++; 335 i++; 336 block_size++; 337 } 338 encode_scalar(s, block_size, k, 16); 339 340 sign_modifier = (block_size <= 0xFFFF); 341 342 history = 0; 343 } 344 345 } 346} 347 348static void write_compressed_frame(AlacEncodeContext *s) 349{ 350 int i, j; 351 int prediction_type = 0; 352 353 if (s->avctx->channels == 2) 354 alac_stereo_decorrelation(s); 355 put_bits(&s->pbctx, 8, s->interlacing_shift); 356 put_bits(&s->pbctx, 8, s->interlacing_leftweight); 357 358 for (i = 0; i < s->avctx->channels; i++) { 359 360 calc_predictor_params(s, i); 361 362 put_bits(&s->pbctx, 4, prediction_type); 363 put_bits(&s->pbctx, 4, s->lpc[i].lpc_quant); 364 365 put_bits(&s->pbctx, 3, s->rc.rice_modifier); 366 put_bits(&s->pbctx, 5, s->lpc[i].lpc_order); 367 // predictor coeff. table 368 for (j = 0; j < s->lpc[i].lpc_order; j++) { 369 put_sbits(&s->pbctx, 16, s->lpc[i].lpc_coeff[j]); 370 } 371 } 372 373 // apply lpc and entropy coding to audio samples 374 375 for (i = 0; i < s->avctx->channels; i++) { 376 alac_linear_predictor(s, i); 377 378 // TODO: determine when this will actually help. for now it's not used. 379 if (prediction_type == 15) { 380 // 2nd pass 1st order filter 381 for (j = s->avctx->frame_size - 1; j > 0; j--) 382 s->predictor_buf[j] -= s->predictor_buf[j - 1]; 383 } 384 385 alac_entropy_coder(s); 386 } 387} 388 389static av_cold int alac_encode_init(AVCodecContext *avctx) 390{ 391 AlacEncodeContext *s = avctx->priv_data; 392 int ret; 393 uint8_t *alac_extradata = av_mallocz(ALAC_EXTRADATA_SIZE+1); 394 395 avctx->frame_size = DEFAULT_FRAME_SIZE; 396 avctx->bits_per_coded_sample = DEFAULT_SAMPLE_SIZE; 397 398 if (avctx->sample_fmt != AV_SAMPLE_FMT_S16) { 399 av_log(avctx, AV_LOG_ERROR, "only pcm_s16 input samples are supported\n"); 400 return -1; 401 } 402 403 /* TODO: Correctly implement multi-channel ALAC. 404 It is similar to multi-channel AAC, in that it has a series of 405 single-channel (SCE), channel-pair (CPE), and LFE elements. */ 406 if (avctx->channels > 2) { 407 av_log(avctx, AV_LOG_ERROR, "only mono or stereo input is currently supported\n"); 408 return AVERROR_PATCHWELCOME; 409 } 410 411 // Set default compression level 412 if (avctx->compression_level == FF_COMPRESSION_DEFAULT) 413 s->compression_level = 2; 414 else 415 s->compression_level = av_clip(avctx->compression_level, 0, 2); 416 417 // Initialize default Rice parameters 418 s->rc.history_mult = 40; 419 s->rc.initial_history = 10; 420 s->rc.k_modifier = 14; 421 s->rc.rice_modifier = 4; 422 423 s->max_coded_frame_size = 8 + (avctx->frame_size*avctx->channels*avctx->bits_per_coded_sample>>3); 424 425 s->write_sample_size = avctx->bits_per_coded_sample + avctx->channels - 1; // FIXME: consider wasted_bytes 426 427 AV_WB32(alac_extradata, ALAC_EXTRADATA_SIZE); 428 AV_WB32(alac_extradata+4, MKBETAG('a','l','a','c')); 429 AV_WB32(alac_extradata+12, avctx->frame_size); 430 AV_WB8 (alac_extradata+17, avctx->bits_per_coded_sample); 431 AV_WB8 (alac_extradata+21, avctx->channels); 432 AV_WB32(alac_extradata+24, s->max_coded_frame_size); 433 AV_WB32(alac_extradata+28, 434 avctx->sample_rate * avctx->channels * avctx->bits_per_coded_sample); // average bitrate 435 AV_WB32(alac_extradata+32, avctx->sample_rate); 436 437 // Set relevant extradata fields 438 if (s->compression_level > 0) { 439 AV_WB8(alac_extradata+18, s->rc.history_mult); 440 AV_WB8(alac_extradata+19, s->rc.initial_history); 441 AV_WB8(alac_extradata+20, s->rc.k_modifier); 442 } 443 444 s->min_prediction_order = DEFAULT_MIN_PRED_ORDER; 445 if (avctx->min_prediction_order >= 0) { 446 if (avctx->min_prediction_order < MIN_LPC_ORDER || 447 avctx->min_prediction_order > ALAC_MAX_LPC_ORDER) { 448 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", 449 avctx->min_prediction_order); 450 return -1; 451 } 452 453 s->min_prediction_order = avctx->min_prediction_order; 454 } 455 456 s->max_prediction_order = DEFAULT_MAX_PRED_ORDER; 457 if (avctx->max_prediction_order >= 0) { 458 if (avctx->max_prediction_order < MIN_LPC_ORDER || 459 avctx->max_prediction_order > ALAC_MAX_LPC_ORDER) { 460 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", 461 avctx->max_prediction_order); 462 return -1; 463 } 464 465 s->max_prediction_order = avctx->max_prediction_order; 466 } 467 468 if (s->max_prediction_order < s->min_prediction_order) { 469 av_log(avctx, AV_LOG_ERROR, 470 "invalid prediction orders: min=%d max=%d\n", 471 s->min_prediction_order, s->max_prediction_order); 472 return -1; 473 } 474 475 avctx->extradata = alac_extradata; 476 avctx->extradata_size = ALAC_EXTRADATA_SIZE; 477 478 avctx->coded_frame = avcodec_alloc_frame(); 479 avctx->coded_frame->key_frame = 1; 480 481 s->avctx = avctx; 482 ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size, s->max_prediction_order, 483 FF_LPC_TYPE_LEVINSON); 484 485 return ret; 486} 487 488static int alac_encode_frame(AVCodecContext *avctx, uint8_t *frame, 489 int buf_size, void *data) 490{ 491 AlacEncodeContext *s = avctx->priv_data; 492 PutBitContext *pb = &s->pbctx; 493 int i, out_bytes, verbatim_flag = 0; 494 495 if (avctx->frame_size > DEFAULT_FRAME_SIZE) { 496 av_log(avctx, AV_LOG_ERROR, "input frame size exceeded\n"); 497 return -1; 498 } 499 500 if (buf_size < 2 * s->max_coded_frame_size) { 501 av_log(avctx, AV_LOG_ERROR, "buffer size is too small\n"); 502 return -1; 503 } 504 505verbatim: 506 init_put_bits(pb, frame, buf_size); 507 508 if (s->compression_level == 0 || verbatim_flag) { 509 // Verbatim mode 510 const int16_t *samples = data; 511 write_frame_header(s, 1); 512 for (i = 0; i < avctx->frame_size * avctx->channels; i++) { 513 put_sbits(pb, 16, *samples++); 514 } 515 } else { 516 init_sample_buffers(s, data); 517 write_frame_header(s, 0); 518 write_compressed_frame(s); 519 } 520 521 put_bits(pb, 3, 7); 522 flush_put_bits(pb); 523 out_bytes = put_bits_count(pb) >> 3; 524 525 if (out_bytes > s->max_coded_frame_size) { 526 /* frame too large. use verbatim mode */ 527 if (verbatim_flag || s->compression_level == 0) { 528 /* still too large. must be an error. */ 529 av_log(avctx, AV_LOG_ERROR, "error encoding frame\n"); 530 return -1; 531 } 532 verbatim_flag = 1; 533 goto verbatim; 534 } 535 536 return out_bytes; 537} 538 539static av_cold int alac_encode_close(AVCodecContext *avctx) 540{ 541 AlacEncodeContext *s = avctx->priv_data; 542 ff_lpc_end(&s->lpc_ctx); 543 av_freep(&avctx->extradata); 544 avctx->extradata_size = 0; 545 av_freep(&avctx->coded_frame); 546 return 0; 547} 548 549AVCodec ff_alac_encoder = { 550 .name = "alac", 551 .type = AVMEDIA_TYPE_AUDIO, 552 .id = CODEC_ID_ALAC, 553 .priv_data_size = sizeof(AlacEncodeContext), 554 .init = alac_encode_init, 555 .encode = alac_encode_frame, 556 .close = alac_encode_close, 557 .capabilities = CODEC_CAP_SMALL_LAST_FRAME, 558 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16, 559 AV_SAMPLE_FMT_NONE }, 560 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"), 561}; 562