| 1 | /* |
| 2 | * Simple free lossless/lossy audio codec |
| 3 | * Copyright (c) 2004 Alex Beregszaszi |
| 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 | #include "avcodec.h" |
| 22 | #include "get_bits.h" |
| 23 | #include "golomb.h" |
| 24 | #include "internal.h" |
| 25 | #include "rangecoder.h" |
| 26 | |
| 27 | |
| 28 | /** |
| 29 | * @file |
| 30 | * Simple free lossless/lossy audio codec |
| 31 | * Based on Paul Francis Harrison's Bonk (http://www.logarithmic.net/pfh/bonk) |
| 32 | * Written and designed by Alex Beregszaszi |
| 33 | * |
| 34 | * TODO: |
| 35 | * - CABAC put/get_symbol |
| 36 | * - independent quantizer for channels |
| 37 | * - >2 channels support |
| 38 | * - more decorrelation types |
| 39 | * - more tap_quant tests |
| 40 | * - selectable intlist writers/readers (bonk-style, golomb, cabac) |
| 41 | */ |
| 42 | |
| 43 | #define MAX_CHANNELS 2 |
| 44 | |
| 45 | #define MID_SIDE 0 |
| 46 | #define LEFT_SIDE 1 |
| 47 | #define RIGHT_SIDE 2 |
| 48 | |
| 49 | typedef struct SonicContext { |
| 50 | int version; |
| 51 | int minor_version; |
| 52 | int lossless, decorrelation; |
| 53 | |
| 54 | int num_taps, downsampling; |
| 55 | double quantization; |
| 56 | |
| 57 | int channels, samplerate, block_align, frame_size; |
| 58 | |
| 59 | int *tap_quant; |
| 60 | int *int_samples; |
| 61 | int *coded_samples[MAX_CHANNELS]; |
| 62 | |
| 63 | // for encoding |
| 64 | int *tail; |
| 65 | int tail_size; |
| 66 | int *window; |
| 67 | int window_size; |
| 68 | |
| 69 | // for decoding |
| 70 | int *predictor_k; |
| 71 | int *predictor_state[MAX_CHANNELS]; |
| 72 | } SonicContext; |
| 73 | |
| 74 | #define LATTICE_SHIFT 10 |
| 75 | #define SAMPLE_SHIFT 4 |
| 76 | #define LATTICE_FACTOR (1 << LATTICE_SHIFT) |
| 77 | #define SAMPLE_FACTOR (1 << SAMPLE_SHIFT) |
| 78 | |
| 79 | #define BASE_QUANT 0.6 |
| 80 | #define RATE_VARIATION 3.0 |
| 81 | |
| 82 | static inline int shift(int a,int b) |
| 83 | { |
| 84 | return (a+(1<<(b-1))) >> b; |
| 85 | } |
| 86 | |
| 87 | static inline int shift_down(int a,int b) |
| 88 | { |
| 89 | return (a>>b)+(a<0); |
| 90 | } |
| 91 | |
| 92 | static av_always_inline av_flatten void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed, uint64_t rc_stat[256][2], uint64_t rc_stat2[32][2]){ |
| 93 | int i; |
| 94 | |
| 95 | #define put_rac(C,S,B) \ |
| 96 | do{\ |
| 97 | if(rc_stat){\ |
| 98 | rc_stat[*(S)][B]++;\ |
| 99 | rc_stat2[(S)-state][B]++;\ |
| 100 | }\ |
| 101 | put_rac(C,S,B);\ |
| 102 | }while(0) |
| 103 | |
| 104 | if(v){ |
| 105 | const int a= FFABS(v); |
| 106 | const int e= av_log2(a); |
| 107 | put_rac(c, state+0, 0); |
| 108 | if(e<=9){ |
| 109 | for(i=0; i<e; i++){ |
| 110 | put_rac(c, state+1+i, 1); //1..10 |
| 111 | } |
| 112 | put_rac(c, state+1+i, 0); |
| 113 | |
| 114 | for(i=e-1; i>=0; i--){ |
| 115 | put_rac(c, state+22+i, (a>>i)&1); //22..31 |
| 116 | } |
| 117 | |
| 118 | if(is_signed) |
| 119 | put_rac(c, state+11 + e, v < 0); //11..21 |
| 120 | }else{ |
| 121 | for(i=0; i<e; i++){ |
| 122 | put_rac(c, state+1+FFMIN(i,9), 1); //1..10 |
| 123 | } |
| 124 | put_rac(c, state+1+9, 0); |
| 125 | |
| 126 | for(i=e-1; i>=0; i--){ |
| 127 | put_rac(c, state+22+FFMIN(i,9), (a>>i)&1); //22..31 |
| 128 | } |
| 129 | |
| 130 | if(is_signed) |
| 131 | put_rac(c, state+11 + 10, v < 0); //11..21 |
| 132 | } |
| 133 | }else{ |
| 134 | put_rac(c, state+0, 1); |
| 135 | } |
| 136 | #undef put_rac |
| 137 | } |
| 138 | |
| 139 | static inline av_flatten int get_symbol(RangeCoder *c, uint8_t *state, int is_signed){ |
| 140 | if(get_rac(c, state+0)) |
| 141 | return 0; |
| 142 | else{ |
| 143 | int i, e, a; |
| 144 | e= 0; |
| 145 | while(get_rac(c, state+1 + FFMIN(e,9))){ //1..10 |
| 146 | e++; |
| 147 | } |
| 148 | |
| 149 | a= 1; |
| 150 | for(i=e-1; i>=0; i--){ |
| 151 | a += a + get_rac(c, state+22 + FFMIN(i,9)); //22..31 |
| 152 | } |
| 153 | |
| 154 | e= -(is_signed && get_rac(c, state+11 + FFMIN(e, 10))); //11..21 |
| 155 | return (a^e)-e; |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | #if 1 |
| 160 | static inline int intlist_write(RangeCoder *c, uint8_t *state, int *buf, int entries, int base_2_part) |
| 161 | { |
| 162 | int i; |
| 163 | |
| 164 | for (i = 0; i < entries; i++) |
| 165 | put_symbol(c, state, buf[i], 1, NULL, NULL); |
| 166 | |
| 167 | return 1; |
| 168 | } |
| 169 | |
| 170 | static inline int intlist_read(RangeCoder *c, uint8_t *state, int *buf, int entries, int base_2_part) |
| 171 | { |
| 172 | int i; |
| 173 | |
| 174 | for (i = 0; i < entries; i++) |
| 175 | buf[i] = get_symbol(c, state, 1); |
| 176 | |
| 177 | return 1; |
| 178 | } |
| 179 | #elif 1 |
| 180 | static inline int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part) |
| 181 | { |
| 182 | int i; |
| 183 | |
| 184 | for (i = 0; i < entries; i++) |
| 185 | set_se_golomb(pb, buf[i]); |
| 186 | |
| 187 | return 1; |
| 188 | } |
| 189 | |
| 190 | static inline int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part) |
| 191 | { |
| 192 | int i; |
| 193 | |
| 194 | for (i = 0; i < entries; i++) |
| 195 | buf[i] = get_se_golomb(gb); |
| 196 | |
| 197 | return 1; |
| 198 | } |
| 199 | |
| 200 | #else |
| 201 | |
| 202 | #define ADAPT_LEVEL 8 |
| 203 | |
| 204 | static int bits_to_store(uint64_t x) |
| 205 | { |
| 206 | int res = 0; |
| 207 | |
| 208 | while(x) |
| 209 | { |
| 210 | res++; |
| 211 | x >>= 1; |
| 212 | } |
| 213 | return res; |
| 214 | } |
| 215 | |
| 216 | static void write_uint_max(PutBitContext *pb, unsigned int value, unsigned int max) |
| 217 | { |
| 218 | int i, bits; |
| 219 | |
| 220 | if (!max) |
| 221 | return; |
| 222 | |
| 223 | bits = bits_to_store(max); |
| 224 | |
| 225 | for (i = 0; i < bits-1; i++) |
| 226 | put_bits(pb, 1, value & (1 << i)); |
| 227 | |
| 228 | if ( (value | (1 << (bits-1))) <= max) |
| 229 | put_bits(pb, 1, value & (1 << (bits-1))); |
| 230 | } |
| 231 | |
| 232 | static unsigned int read_uint_max(GetBitContext *gb, int max) |
| 233 | { |
| 234 | int i, bits, value = 0; |
| 235 | |
| 236 | if (!max) |
| 237 | return 0; |
| 238 | |
| 239 | bits = bits_to_store(max); |
| 240 | |
| 241 | for (i = 0; i < bits-1; i++) |
| 242 | if (get_bits1(gb)) |
| 243 | value += 1 << i; |
| 244 | |
| 245 | if ( (value | (1<<(bits-1))) <= max) |
| 246 | if (get_bits1(gb)) |
| 247 | value += 1 << (bits-1); |
| 248 | |
| 249 | return value; |
| 250 | } |
| 251 | |
| 252 | static int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part) |
| 253 | { |
| 254 | int i, j, x = 0, low_bits = 0, max = 0; |
| 255 | int step = 256, pos = 0, dominant = 0, any = 0; |
| 256 | int *copy, *bits; |
| 257 | |
| 258 | copy = av_calloc(entries, sizeof(*copy)); |
| 259 | if (!copy) |
| 260 | return AVERROR(ENOMEM); |
| 261 | |
| 262 | if (base_2_part) |
| 263 | { |
| 264 | int energy = 0; |
| 265 | |
| 266 | for (i = 0; i < entries; i++) |
| 267 | energy += abs(buf[i]); |
| 268 | |
| 269 | low_bits = bits_to_store(energy / (entries * 2)); |
| 270 | if (low_bits > 15) |
| 271 | low_bits = 15; |
| 272 | |
| 273 | put_bits(pb, 4, low_bits); |
| 274 | } |
| 275 | |
| 276 | for (i = 0; i < entries; i++) |
| 277 | { |
| 278 | put_bits(pb, low_bits, abs(buf[i])); |
| 279 | copy[i] = abs(buf[i]) >> low_bits; |
| 280 | if (copy[i] > max) |
| 281 | max = abs(copy[i]); |
| 282 | } |
| 283 | |
| 284 | bits = av_calloc(entries*max, sizeof(*bits)); |
| 285 | if (!bits) |
| 286 | { |
| 287 | av_free(copy); |
| 288 | return AVERROR(ENOMEM); |
| 289 | } |
| 290 | |
| 291 | for (i = 0; i <= max; i++) |
| 292 | { |
| 293 | for (j = 0; j < entries; j++) |
| 294 | if (copy[j] >= i) |
| 295 | bits[x++] = copy[j] > i; |
| 296 | } |
| 297 | |
| 298 | // store bitstream |
| 299 | while (pos < x) |
| 300 | { |
| 301 | int steplet = step >> 8; |
| 302 | |
| 303 | if (pos + steplet > x) |
| 304 | steplet = x - pos; |
| 305 | |
| 306 | for (i = 0; i < steplet; i++) |
| 307 | if (bits[i+pos] != dominant) |
| 308 | any = 1; |
| 309 | |
| 310 | put_bits(pb, 1, any); |
| 311 | |
| 312 | if (!any) |
| 313 | { |
| 314 | pos += steplet; |
| 315 | step += step / ADAPT_LEVEL; |
| 316 | } |
| 317 | else |
| 318 | { |
| 319 | int interloper = 0; |
| 320 | |
| 321 | while (((pos + interloper) < x) && (bits[pos + interloper] == dominant)) |
| 322 | interloper++; |
| 323 | |
| 324 | // note change |
| 325 | write_uint_max(pb, interloper, (step >> 8) - 1); |
| 326 | |
| 327 | pos += interloper + 1; |
| 328 | step -= step / ADAPT_LEVEL; |
| 329 | } |
| 330 | |
| 331 | if (step < 256) |
| 332 | { |
| 333 | step = 65536 / step; |
| 334 | dominant = !dominant; |
| 335 | } |
| 336 | } |
| 337 | |
| 338 | // store signs |
| 339 | for (i = 0; i < entries; i++) |
| 340 | if (buf[i]) |
| 341 | put_bits(pb, 1, buf[i] < 0); |
| 342 | |
| 343 | av_free(bits); |
| 344 | av_free(copy); |
| 345 | |
| 346 | return 0; |
| 347 | } |
| 348 | |
| 349 | static int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part) |
| 350 | { |
| 351 | int i, low_bits = 0, x = 0; |
| 352 | int n_zeros = 0, step = 256, dominant = 0; |
| 353 | int pos = 0, level = 0; |
| 354 | int *bits = av_calloc(entries, sizeof(*bits)); |
| 355 | |
| 356 | if (!bits) |
| 357 | return AVERROR(ENOMEM); |
| 358 | |
| 359 | if (base_2_part) |
| 360 | { |
| 361 | low_bits = get_bits(gb, 4); |
| 362 | |
| 363 | if (low_bits) |
| 364 | for (i = 0; i < entries; i++) |
| 365 | buf[i] = get_bits(gb, low_bits); |
| 366 | } |
| 367 | |
| 368 | // av_log(NULL, AV_LOG_INFO, "entries: %d, low bits: %d\n", entries, low_bits); |
| 369 | |
| 370 | while (n_zeros < entries) |
| 371 | { |
| 372 | int steplet = step >> 8; |
| 373 | |
| 374 | if (!get_bits1(gb)) |
| 375 | { |
| 376 | for (i = 0; i < steplet; i++) |
| 377 | bits[x++] = dominant; |
| 378 | |
| 379 | if (!dominant) |
| 380 | n_zeros += steplet; |
| 381 | |
| 382 | step += step / ADAPT_LEVEL; |
| 383 | } |
| 384 | else |
| 385 | { |
| 386 | int actual_run = read_uint_max(gb, steplet-1); |
| 387 | |
| 388 | // av_log(NULL, AV_LOG_INFO, "actual run: %d\n", actual_run); |
| 389 | |
| 390 | for (i = 0; i < actual_run; i++) |
| 391 | bits[x++] = dominant; |
| 392 | |
| 393 | bits[x++] = !dominant; |
| 394 | |
| 395 | if (!dominant) |
| 396 | n_zeros += actual_run; |
| 397 | else |
| 398 | n_zeros++; |
| 399 | |
| 400 | step -= step / ADAPT_LEVEL; |
| 401 | } |
| 402 | |
| 403 | if (step < 256) |
| 404 | { |
| 405 | step = 65536 / step; |
| 406 | dominant = !dominant; |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | // reconstruct unsigned values |
| 411 | n_zeros = 0; |
| 412 | for (i = 0; n_zeros < entries; i++) |
| 413 | { |
| 414 | while(1) |
| 415 | { |
| 416 | if (pos >= entries) |
| 417 | { |
| 418 | pos = 0; |
| 419 | level += 1 << low_bits; |
| 420 | } |
| 421 | |
| 422 | if (buf[pos] >= level) |
| 423 | break; |
| 424 | |
| 425 | pos++; |
| 426 | } |
| 427 | |
| 428 | if (bits[i]) |
| 429 | buf[pos] += 1 << low_bits; |
| 430 | else |
| 431 | n_zeros++; |
| 432 | |
| 433 | pos++; |
| 434 | } |
| 435 | av_free(bits); |
| 436 | |
| 437 | // read signs |
| 438 | for (i = 0; i < entries; i++) |
| 439 | if (buf[i] && get_bits1(gb)) |
| 440 | buf[i] = -buf[i]; |
| 441 | |
| 442 | // av_log(NULL, AV_LOG_INFO, "zeros: %d pos: %d\n", n_zeros, pos); |
| 443 | |
| 444 | return 0; |
| 445 | } |
| 446 | #endif |
| 447 | |
| 448 | static void predictor_init_state(int *k, int *state, int order) |
| 449 | { |
| 450 | int i; |
| 451 | |
| 452 | for (i = order-2; i >= 0; i--) |
| 453 | { |
| 454 | int j, p, x = state[i]; |
| 455 | |
| 456 | for (j = 0, p = i+1; p < order; j++,p++) |
| 457 | { |
| 458 | int tmp = x + shift_down(k[j] * state[p], LATTICE_SHIFT); |
| 459 | state[p] += shift_down(k[j]*x, LATTICE_SHIFT); |
| 460 | x = tmp; |
| 461 | } |
| 462 | } |
| 463 | } |
| 464 | |
| 465 | static int predictor_calc_error(int *k, int *state, int order, int error) |
| 466 | { |
| 467 | int i, x = error - shift_down(k[order-1] * state[order-1], LATTICE_SHIFT); |
| 468 | |
| 469 | #if 1 |
| 470 | int *k_ptr = &(k[order-2]), |
| 471 | *state_ptr = &(state[order-2]); |
| 472 | for (i = order-2; i >= 0; i--, k_ptr--, state_ptr--) |
| 473 | { |
| 474 | int k_value = *k_ptr, state_value = *state_ptr; |
| 475 | x -= shift_down(k_value * state_value, LATTICE_SHIFT); |
| 476 | state_ptr[1] = state_value + shift_down(k_value * x, LATTICE_SHIFT); |
| 477 | } |
| 478 | #else |
| 479 | for (i = order-2; i >= 0; i--) |
| 480 | { |
| 481 | x -= shift_down(k[i] * state[i], LATTICE_SHIFT); |
| 482 | state[i+1] = state[i] + shift_down(k[i] * x, LATTICE_SHIFT); |
| 483 | } |
| 484 | #endif |
| 485 | |
| 486 | // don't drift too far, to avoid overflows |
| 487 | if (x > (SAMPLE_FACTOR<<16)) x = (SAMPLE_FACTOR<<16); |
| 488 | if (x < -(SAMPLE_FACTOR<<16)) x = -(SAMPLE_FACTOR<<16); |
| 489 | |
| 490 | state[0] = x; |
| 491 | |
| 492 | return x; |
| 493 | } |
| 494 | |
| 495 | #if CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER |
| 496 | // Heavily modified Levinson-Durbin algorithm which |
| 497 | // copes better with quantization, and calculates the |
| 498 | // actual whitened result as it goes. |
| 499 | |
| 500 | static void modified_levinson_durbin(int *window, int window_entries, |
| 501 | int *out, int out_entries, int channels, int *tap_quant) |
| 502 | { |
| 503 | int i; |
| 504 | int *state = av_calloc(window_entries, sizeof(*state)); |
| 505 | |
| 506 | memcpy(state, window, 4* window_entries); |
| 507 | |
| 508 | for (i = 0; i < out_entries; i++) |
| 509 | { |
| 510 | int step = (i+1)*channels, k, j; |
| 511 | double xx = 0.0, xy = 0.0; |
| 512 | #if 1 |
| 513 | int *x_ptr = &(window[step]); |
| 514 | int *state_ptr = &(state[0]); |
| 515 | j = window_entries - step; |
| 516 | for (;j>0;j--,x_ptr++,state_ptr++) |
| 517 | { |
| 518 | double x_value = *x_ptr; |
| 519 | double state_value = *state_ptr; |
| 520 | xx += state_value*state_value; |
| 521 | xy += x_value*state_value; |
| 522 | } |
| 523 | #else |
| 524 | for (j = 0; j <= (window_entries - step); j++); |
| 525 | { |
| 526 | double stepval = window[step+j]; |
| 527 | double stateval = window[j]; |
| 528 | // xx += (double)window[j]*(double)window[j]; |
| 529 | // xy += (double)window[step+j]*(double)window[j]; |
| 530 | xx += stateval*stateval; |
| 531 | xy += stepval*stateval; |
| 532 | } |
| 533 | #endif |
| 534 | if (xx == 0.0) |
| 535 | k = 0; |
| 536 | else |
| 537 | k = (int)(floor(-xy/xx * (double)LATTICE_FACTOR / (double)(tap_quant[i]) + 0.5)); |
| 538 | |
| 539 | if (k > (LATTICE_FACTOR/tap_quant[i])) |
| 540 | k = LATTICE_FACTOR/tap_quant[i]; |
| 541 | if (-k > (LATTICE_FACTOR/tap_quant[i])) |
| 542 | k = -(LATTICE_FACTOR/tap_quant[i]); |
| 543 | |
| 544 | out[i] = k; |
| 545 | k *= tap_quant[i]; |
| 546 | |
| 547 | #if 1 |
| 548 | x_ptr = &(window[step]); |
| 549 | state_ptr = &(state[0]); |
| 550 | j = window_entries - step; |
| 551 | for (;j>0;j--,x_ptr++,state_ptr++) |
| 552 | { |
| 553 | int x_value = *x_ptr; |
| 554 | int state_value = *state_ptr; |
| 555 | *x_ptr = x_value + shift_down(k*state_value,LATTICE_SHIFT); |
| 556 | *state_ptr = state_value + shift_down(k*x_value, LATTICE_SHIFT); |
| 557 | } |
| 558 | #else |
| 559 | for (j=0; j <= (window_entries - step); j++) |
| 560 | { |
| 561 | int stepval = window[step+j]; |
| 562 | int stateval=state[j]; |
| 563 | window[step+j] += shift_down(k * stateval, LATTICE_SHIFT); |
| 564 | state[j] += shift_down(k * stepval, LATTICE_SHIFT); |
| 565 | } |
| 566 | #endif |
| 567 | } |
| 568 | |
| 569 | av_free(state); |
| 570 | } |
| 571 | |
| 572 | static inline int code_samplerate(int samplerate) |
| 573 | { |
| 574 | switch (samplerate) |
| 575 | { |
| 576 | case 44100: return 0; |
| 577 | case 22050: return 1; |
| 578 | case 11025: return 2; |
| 579 | case 96000: return 3; |
| 580 | case 48000: return 4; |
| 581 | case 32000: return 5; |
| 582 | case 24000: return 6; |
| 583 | case 16000: return 7; |
| 584 | case 8000: return 8; |
| 585 | } |
| 586 | return AVERROR(EINVAL); |
| 587 | } |
| 588 | |
| 589 | static av_cold int sonic_encode_init(AVCodecContext *avctx) |
| 590 | { |
| 591 | SonicContext *s = avctx->priv_data; |
| 592 | PutBitContext pb; |
| 593 | int i; |
| 594 | |
| 595 | s->version = 2; |
| 596 | |
| 597 | if (avctx->channels > MAX_CHANNELS) |
| 598 | { |
| 599 | av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n"); |
| 600 | return AVERROR(EINVAL); /* only stereo or mono for now */ |
| 601 | } |
| 602 | |
| 603 | if (avctx->channels == 2) |
| 604 | s->decorrelation = MID_SIDE; |
| 605 | else |
| 606 | s->decorrelation = 3; |
| 607 | |
| 608 | if (avctx->codec->id == AV_CODEC_ID_SONIC_LS) |
| 609 | { |
| 610 | s->lossless = 1; |
| 611 | s->num_taps = 32; |
| 612 | s->downsampling = 1; |
| 613 | s->quantization = 0.0; |
| 614 | } |
| 615 | else |
| 616 | { |
| 617 | s->num_taps = 128; |
| 618 | s->downsampling = 2; |
| 619 | s->quantization = 1.0; |
| 620 | } |
| 621 | |
| 622 | // max tap 2048 |
| 623 | if (s->num_taps < 32 || s->num_taps > 1024 || s->num_taps % 32) { |
| 624 | av_log(avctx, AV_LOG_ERROR, "Invalid number of taps\n"); |
| 625 | return AVERROR_INVALIDDATA; |
| 626 | } |
| 627 | |
| 628 | // generate taps |
| 629 | s->tap_quant = av_calloc(s->num_taps, sizeof(*s->tap_quant)); |
| 630 | for (i = 0; i < s->num_taps; i++) |
| 631 | s->tap_quant[i] = ff_sqrt(i+1); |
| 632 | |
| 633 | s->channels = avctx->channels; |
| 634 | s->samplerate = avctx->sample_rate; |
| 635 | |
| 636 | s->block_align = 2048LL*s->samplerate/(44100*s->downsampling); |
| 637 | s->frame_size = s->channels*s->block_align*s->downsampling; |
| 638 | |
| 639 | s->tail_size = s->num_taps*s->channels; |
| 640 | s->tail = av_calloc(s->tail_size, sizeof(*s->tail)); |
| 641 | if (!s->tail) |
| 642 | return AVERROR(ENOMEM); |
| 643 | |
| 644 | s->predictor_k = av_calloc(s->num_taps, sizeof(*s->predictor_k) ); |
| 645 | if (!s->predictor_k) |
| 646 | return AVERROR(ENOMEM); |
| 647 | |
| 648 | for (i = 0; i < s->channels; i++) |
| 649 | { |
| 650 | s->coded_samples[i] = av_calloc(s->block_align, sizeof(**s->coded_samples)); |
| 651 | if (!s->coded_samples[i]) |
| 652 | return AVERROR(ENOMEM); |
| 653 | } |
| 654 | |
| 655 | s->int_samples = av_calloc(s->frame_size, sizeof(*s->int_samples)); |
| 656 | |
| 657 | s->window_size = ((2*s->tail_size)+s->frame_size); |
| 658 | s->window = av_calloc(s->window_size, sizeof(*s->window)); |
| 659 | if (!s->window) |
| 660 | return AVERROR(ENOMEM); |
| 661 | |
| 662 | avctx->extradata = av_mallocz(16); |
| 663 | if (!avctx->extradata) |
| 664 | return AVERROR(ENOMEM); |
| 665 | init_put_bits(&pb, avctx->extradata, 16*8); |
| 666 | |
| 667 | put_bits(&pb, 2, s->version); // version |
| 668 | if (s->version >= 1) |
| 669 | { |
| 670 | if (s->version >= 2) { |
| 671 | put_bits(&pb, 8, s->version); |
| 672 | put_bits(&pb, 8, s->minor_version); |
| 673 | } |
| 674 | put_bits(&pb, 2, s->channels); |
| 675 | put_bits(&pb, 4, code_samplerate(s->samplerate)); |
| 676 | } |
| 677 | put_bits(&pb, 1, s->lossless); |
| 678 | if (!s->lossless) |
| 679 | put_bits(&pb, 3, SAMPLE_SHIFT); // XXX FIXME: sample precision |
| 680 | put_bits(&pb, 2, s->decorrelation); |
| 681 | put_bits(&pb, 2, s->downsampling); |
| 682 | put_bits(&pb, 5, (s->num_taps >> 5)-1); // 32..1024 |
| 683 | put_bits(&pb, 1, 0); // XXX FIXME: no custom tap quant table |
| 684 | |
| 685 | flush_put_bits(&pb); |
| 686 | avctx->extradata_size = put_bits_count(&pb)/8; |
| 687 | |
| 688 | av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d.%d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n", |
| 689 | s->version, s->minor_version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling); |
| 690 | |
| 691 | avctx->frame_size = s->block_align*s->downsampling; |
| 692 | |
| 693 | return 0; |
| 694 | } |
| 695 | |
| 696 | static av_cold int sonic_encode_close(AVCodecContext *avctx) |
| 697 | { |
| 698 | SonicContext *s = avctx->priv_data; |
| 699 | int i; |
| 700 | |
| 701 | for (i = 0; i < s->channels; i++) |
| 702 | av_freep(&s->coded_samples[i]); |
| 703 | |
| 704 | av_freep(&s->predictor_k); |
| 705 | av_freep(&s->tail); |
| 706 | av_freep(&s->tap_quant); |
| 707 | av_freep(&s->window); |
| 708 | av_freep(&s->int_samples); |
| 709 | |
| 710 | return 0; |
| 711 | } |
| 712 | |
| 713 | static int sonic_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, |
| 714 | const AVFrame *frame, int *got_packet_ptr) |
| 715 | { |
| 716 | SonicContext *s = avctx->priv_data; |
| 717 | RangeCoder c; |
| 718 | int i, j, ch, quant = 0, x = 0; |
| 719 | int ret; |
| 720 | const short *samples = (const int16_t*)frame->data[0]; |
| 721 | uint8_t state[32]; |
| 722 | |
| 723 | if ((ret = ff_alloc_packet2(avctx, avpkt, s->frame_size * 5 + 1000)) < 0) |
| 724 | return ret; |
| 725 | |
| 726 | ff_init_range_encoder(&c, avpkt->data, avpkt->size); |
| 727 | ff_build_rac_states(&c, 0.05*(1LL<<32), 256-8); |
| 728 | memset(state, 128, sizeof(state)); |
| 729 | |
| 730 | // short -> internal |
| 731 | for (i = 0; i < s->frame_size; i++) |
| 732 | s->int_samples[i] = samples[i]; |
| 733 | |
| 734 | if (!s->lossless) |
| 735 | for (i = 0; i < s->frame_size; i++) |
| 736 | s->int_samples[i] = s->int_samples[i] << SAMPLE_SHIFT; |
| 737 | |
| 738 | switch(s->decorrelation) |
| 739 | { |
| 740 | case MID_SIDE: |
| 741 | for (i = 0; i < s->frame_size; i += s->channels) |
| 742 | { |
| 743 | s->int_samples[i] += s->int_samples[i+1]; |
| 744 | s->int_samples[i+1] -= shift(s->int_samples[i], 1); |
| 745 | } |
| 746 | break; |
| 747 | case LEFT_SIDE: |
| 748 | for (i = 0; i < s->frame_size; i += s->channels) |
| 749 | s->int_samples[i+1] -= s->int_samples[i]; |
| 750 | break; |
| 751 | case RIGHT_SIDE: |
| 752 | for (i = 0; i < s->frame_size; i += s->channels) |
| 753 | s->int_samples[i] -= s->int_samples[i+1]; |
| 754 | break; |
| 755 | } |
| 756 | |
| 757 | memset(s->window, 0, 4* s->window_size); |
| 758 | |
| 759 | for (i = 0; i < s->tail_size; i++) |
| 760 | s->window[x++] = s->tail[i]; |
| 761 | |
| 762 | for (i = 0; i < s->frame_size; i++) |
| 763 | s->window[x++] = s->int_samples[i]; |
| 764 | |
| 765 | for (i = 0; i < s->tail_size; i++) |
| 766 | s->window[x++] = 0; |
| 767 | |
| 768 | for (i = 0; i < s->tail_size; i++) |
| 769 | s->tail[i] = s->int_samples[s->frame_size - s->tail_size + i]; |
| 770 | |
| 771 | // generate taps |
| 772 | modified_levinson_durbin(s->window, s->window_size, |
| 773 | s->predictor_k, s->num_taps, s->channels, s->tap_quant); |
| 774 | if ((ret = intlist_write(&c, state, s->predictor_k, s->num_taps, 0)) < 0) |
| 775 | return ret; |
| 776 | |
| 777 | for (ch = 0; ch < s->channels; ch++) |
| 778 | { |
| 779 | x = s->tail_size+ch; |
| 780 | for (i = 0; i < s->block_align; i++) |
| 781 | { |
| 782 | int sum = 0; |
| 783 | for (j = 0; j < s->downsampling; j++, x += s->channels) |
| 784 | sum += s->window[x]; |
| 785 | s->coded_samples[ch][i] = sum; |
| 786 | } |
| 787 | } |
| 788 | |
| 789 | // simple rate control code |
| 790 | if (!s->lossless) |
| 791 | { |
| 792 | double energy1 = 0.0, energy2 = 0.0; |
| 793 | for (ch = 0; ch < s->channels; ch++) |
| 794 | { |
| 795 | for (i = 0; i < s->block_align; i++) |
| 796 | { |
| 797 | double sample = s->coded_samples[ch][i]; |
| 798 | energy2 += sample*sample; |
| 799 | energy1 += fabs(sample); |
| 800 | } |
| 801 | } |
| 802 | |
| 803 | energy2 = sqrt(energy2/(s->channels*s->block_align)); |
| 804 | energy1 = M_SQRT2*energy1/(s->channels*s->block_align); |
| 805 | |
| 806 | // increase bitrate when samples are like a gaussian distribution |
| 807 | // reduce bitrate when samples are like a two-tailed exponential distribution |
| 808 | |
| 809 | if (energy2 > energy1) |
| 810 | energy2 += (energy2-energy1)*RATE_VARIATION; |
| 811 | |
| 812 | quant = (int)(BASE_QUANT*s->quantization*energy2/SAMPLE_FACTOR); |
| 813 | // av_log(avctx, AV_LOG_DEBUG, "quant: %d energy: %f / %f\n", quant, energy1, energy2); |
| 814 | |
| 815 | quant = av_clip(quant, 1, 65534); |
| 816 | |
| 817 | put_symbol(&c, state, quant, 0, NULL, NULL); |
| 818 | |
| 819 | quant *= SAMPLE_FACTOR; |
| 820 | } |
| 821 | |
| 822 | // write out coded samples |
| 823 | for (ch = 0; ch < s->channels; ch++) |
| 824 | { |
| 825 | if (!s->lossless) |
| 826 | for (i = 0; i < s->block_align; i++) |
| 827 | s->coded_samples[ch][i] = ROUNDED_DIV(s->coded_samples[ch][i], quant); |
| 828 | |
| 829 | if ((ret = intlist_write(&c, state, s->coded_samples[ch], s->block_align, 1)) < 0) |
| 830 | return ret; |
| 831 | } |
| 832 | |
| 833 | // av_log(avctx, AV_LOG_DEBUG, "used bytes: %d\n", (put_bits_count(&pb)+7)/8); |
| 834 | |
| 835 | avpkt->size = ff_rac_terminate(&c); |
| 836 | *got_packet_ptr = 1; |
| 837 | return 0; |
| 838 | |
| 839 | } |
| 840 | #endif /* CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER */ |
| 841 | |
| 842 | #if CONFIG_SONIC_DECODER |
| 843 | static const int samplerate_table[] = |
| 844 | { 44100, 22050, 11025, 96000, 48000, 32000, 24000, 16000, 8000 }; |
| 845 | |
| 846 | static av_cold int sonic_decode_init(AVCodecContext *avctx) |
| 847 | { |
| 848 | SonicContext *s = avctx->priv_data; |
| 849 | GetBitContext gb; |
| 850 | int i; |
| 851 | |
| 852 | s->channels = avctx->channels; |
| 853 | s->samplerate = avctx->sample_rate; |
| 854 | |
| 855 | if (!avctx->extradata) |
| 856 | { |
| 857 | av_log(avctx, AV_LOG_ERROR, "No mandatory headers present\n"); |
| 858 | return AVERROR_INVALIDDATA; |
| 859 | } |
| 860 | |
| 861 | init_get_bits8(&gb, avctx->extradata, avctx->extradata_size); |
| 862 | |
| 863 | s->version = get_bits(&gb, 2); |
| 864 | if (s->version >= 2) { |
| 865 | s->version = get_bits(&gb, 8); |
| 866 | s->minor_version = get_bits(&gb, 8); |
| 867 | } |
| 868 | if (s->version != 2) |
| 869 | { |
| 870 | av_log(avctx, AV_LOG_ERROR, "Unsupported Sonic version, please report\n"); |
| 871 | return AVERROR_INVALIDDATA; |
| 872 | } |
| 873 | |
| 874 | if (s->version >= 1) |
| 875 | { |
| 876 | s->channels = get_bits(&gb, 2); |
| 877 | s->samplerate = samplerate_table[get_bits(&gb, 4)]; |
| 878 | av_log(avctx, AV_LOG_INFO, "Sonicv2 chans: %d samprate: %d\n", |
| 879 | s->channels, s->samplerate); |
| 880 | } |
| 881 | |
| 882 | if (s->channels > MAX_CHANNELS) |
| 883 | { |
| 884 | av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n"); |
| 885 | return AVERROR_INVALIDDATA; |
| 886 | } |
| 887 | |
| 888 | s->lossless = get_bits1(&gb); |
| 889 | if (!s->lossless) |
| 890 | skip_bits(&gb, 3); // XXX FIXME |
| 891 | s->decorrelation = get_bits(&gb, 2); |
| 892 | if (s->decorrelation != 3 && s->channels != 2) { |
| 893 | av_log(avctx, AV_LOG_ERROR, "invalid decorrelation %d\n", s->decorrelation); |
| 894 | return AVERROR_INVALIDDATA; |
| 895 | } |
| 896 | |
| 897 | s->downsampling = get_bits(&gb, 2); |
| 898 | if (!s->downsampling) { |
| 899 | av_log(avctx, AV_LOG_ERROR, "invalid downsampling value\n"); |
| 900 | return AVERROR_INVALIDDATA; |
| 901 | } |
| 902 | |
| 903 | s->num_taps = (get_bits(&gb, 5)+1)<<5; |
| 904 | if (get_bits1(&gb)) // XXX FIXME |
| 905 | av_log(avctx, AV_LOG_INFO, "Custom quant table\n"); |
| 906 | |
| 907 | s->block_align = 2048LL*s->samplerate/(44100*s->downsampling); |
| 908 | s->frame_size = s->channels*s->block_align*s->downsampling; |
| 909 | // avctx->frame_size = s->block_align; |
| 910 | |
| 911 | av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d.%d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n", |
| 912 | s->version, s->minor_version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling); |
| 913 | |
| 914 | // generate taps |
| 915 | s->tap_quant = av_calloc(s->num_taps, sizeof(*s->tap_quant)); |
| 916 | for (i = 0; i < s->num_taps; i++) |
| 917 | s->tap_quant[i] = ff_sqrt(i+1); |
| 918 | |
| 919 | s->predictor_k = av_calloc(s->num_taps, sizeof(*s->predictor_k)); |
| 920 | |
| 921 | for (i = 0; i < s->channels; i++) |
| 922 | { |
| 923 | s->predictor_state[i] = av_calloc(s->num_taps, sizeof(**s->predictor_state)); |
| 924 | if (!s->predictor_state[i]) |
| 925 | return AVERROR(ENOMEM); |
| 926 | } |
| 927 | |
| 928 | for (i = 0; i < s->channels; i++) |
| 929 | { |
| 930 | s->coded_samples[i] = av_calloc(s->block_align, sizeof(**s->coded_samples)); |
| 931 | if (!s->coded_samples[i]) |
| 932 | return AVERROR(ENOMEM); |
| 933 | } |
| 934 | s->int_samples = av_calloc(s->frame_size, sizeof(*s->int_samples)); |
| 935 | |
| 936 | avctx->sample_fmt = AV_SAMPLE_FMT_S16; |
| 937 | return 0; |
| 938 | } |
| 939 | |
| 940 | static av_cold int sonic_decode_close(AVCodecContext *avctx) |
| 941 | { |
| 942 | SonicContext *s = avctx->priv_data; |
| 943 | int i; |
| 944 | |
| 945 | av_freep(&s->int_samples); |
| 946 | av_freep(&s->tap_quant); |
| 947 | av_freep(&s->predictor_k); |
| 948 | |
| 949 | for (i = 0; i < s->channels; i++) |
| 950 | { |
| 951 | av_freep(&s->predictor_state[i]); |
| 952 | av_freep(&s->coded_samples[i]); |
| 953 | } |
| 954 | |
| 955 | return 0; |
| 956 | } |
| 957 | |
| 958 | static int sonic_decode_frame(AVCodecContext *avctx, |
| 959 | void *data, int *got_frame_ptr, |
| 960 | AVPacket *avpkt) |
| 961 | { |
| 962 | const uint8_t *buf = avpkt->data; |
| 963 | int buf_size = avpkt->size; |
| 964 | SonicContext *s = avctx->priv_data; |
| 965 | RangeCoder c; |
| 966 | uint8_t state[32]; |
| 967 | int i, quant, ch, j, ret; |
| 968 | int16_t *samples; |
| 969 | AVFrame *frame = data; |
| 970 | |
| 971 | if (buf_size == 0) return 0; |
| 972 | |
| 973 | frame->nb_samples = s->frame_size / avctx->channels; |
| 974 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
| 975 | return ret; |
| 976 | samples = (int16_t *)frame->data[0]; |
| 977 | |
| 978 | // av_log(NULL, AV_LOG_INFO, "buf_size: %d\n", buf_size); |
| 979 | |
| 980 | memset(state, 128, sizeof(state)); |
| 981 | ff_init_range_decoder(&c, buf, buf_size); |
| 982 | ff_build_rac_states(&c, 0.05*(1LL<<32), 256-8); |
| 983 | |
| 984 | intlist_read(&c, state, s->predictor_k, s->num_taps, 0); |
| 985 | |
| 986 | // dequantize |
| 987 | for (i = 0; i < s->num_taps; i++) |
| 988 | s->predictor_k[i] *= s->tap_quant[i]; |
| 989 | |
| 990 | if (s->lossless) |
| 991 | quant = 1; |
| 992 | else |
| 993 | quant = get_symbol(&c, state, 0) * SAMPLE_FACTOR; |
| 994 | |
| 995 | // av_log(NULL, AV_LOG_INFO, "quant: %d\n", quant); |
| 996 | |
| 997 | for (ch = 0; ch < s->channels; ch++) |
| 998 | { |
| 999 | int x = ch; |
| 1000 | |
| 1001 | predictor_init_state(s->predictor_k, s->predictor_state[ch], s->num_taps); |
| 1002 | |
| 1003 | intlist_read(&c, state, s->coded_samples[ch], s->block_align, 1); |
| 1004 | |
| 1005 | for (i = 0; i < s->block_align; i++) |
| 1006 | { |
| 1007 | for (j = 0; j < s->downsampling - 1; j++) |
| 1008 | { |
| 1009 | s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, 0); |
| 1010 | x += s->channels; |
| 1011 | } |
| 1012 | |
| 1013 | s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, s->coded_samples[ch][i] * quant); |
| 1014 | x += s->channels; |
| 1015 | } |
| 1016 | |
| 1017 | for (i = 0; i < s->num_taps; i++) |
| 1018 | s->predictor_state[ch][i] = s->int_samples[s->frame_size - s->channels + ch - i*s->channels]; |
| 1019 | } |
| 1020 | |
| 1021 | switch(s->decorrelation) |
| 1022 | { |
| 1023 | case MID_SIDE: |
| 1024 | for (i = 0; i < s->frame_size; i += s->channels) |
| 1025 | { |
| 1026 | s->int_samples[i+1] += shift(s->int_samples[i], 1); |
| 1027 | s->int_samples[i] -= s->int_samples[i+1]; |
| 1028 | } |
| 1029 | break; |
| 1030 | case LEFT_SIDE: |
| 1031 | for (i = 0; i < s->frame_size; i += s->channels) |
| 1032 | s->int_samples[i+1] += s->int_samples[i]; |
| 1033 | break; |
| 1034 | case RIGHT_SIDE: |
| 1035 | for (i = 0; i < s->frame_size; i += s->channels) |
| 1036 | s->int_samples[i] += s->int_samples[i+1]; |
| 1037 | break; |
| 1038 | } |
| 1039 | |
| 1040 | if (!s->lossless) |
| 1041 | for (i = 0; i < s->frame_size; i++) |
| 1042 | s->int_samples[i] = shift(s->int_samples[i], SAMPLE_SHIFT); |
| 1043 | |
| 1044 | // internal -> short |
| 1045 | for (i = 0; i < s->frame_size; i++) |
| 1046 | samples[i] = av_clip_int16(s->int_samples[i]); |
| 1047 | |
| 1048 | *got_frame_ptr = 1; |
| 1049 | |
| 1050 | return buf_size; |
| 1051 | } |
| 1052 | |
| 1053 | AVCodec ff_sonic_decoder = { |
| 1054 | .name = "sonic", |
| 1055 | .long_name = NULL_IF_CONFIG_SMALL("Sonic"), |
| 1056 | .type = AVMEDIA_TYPE_AUDIO, |
| 1057 | .id = AV_CODEC_ID_SONIC, |
| 1058 | .priv_data_size = sizeof(SonicContext), |
| 1059 | .init = sonic_decode_init, |
| 1060 | .close = sonic_decode_close, |
| 1061 | .decode = sonic_decode_frame, |
| 1062 | .capabilities = CODEC_CAP_DR1 | CODEC_CAP_EXPERIMENTAL, |
| 1063 | }; |
| 1064 | #endif /* CONFIG_SONIC_DECODER */ |
| 1065 | |
| 1066 | #if CONFIG_SONIC_ENCODER |
| 1067 | AVCodec ff_sonic_encoder = { |
| 1068 | .name = "sonic", |
| 1069 | .long_name = NULL_IF_CONFIG_SMALL("Sonic"), |
| 1070 | .type = AVMEDIA_TYPE_AUDIO, |
| 1071 | .id = AV_CODEC_ID_SONIC, |
| 1072 | .priv_data_size = sizeof(SonicContext), |
| 1073 | .init = sonic_encode_init, |
| 1074 | .encode2 = sonic_encode_frame, |
| 1075 | .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE }, |
| 1076 | .capabilities = CODEC_CAP_EXPERIMENTAL, |
| 1077 | .close = sonic_encode_close, |
| 1078 | }; |
| 1079 | #endif |
| 1080 | |
| 1081 | #if CONFIG_SONIC_LS_ENCODER |
| 1082 | AVCodec ff_sonic_ls_encoder = { |
| 1083 | .name = "sonicls", |
| 1084 | .long_name = NULL_IF_CONFIG_SMALL("Sonic lossless"), |
| 1085 | .type = AVMEDIA_TYPE_AUDIO, |
| 1086 | .id = AV_CODEC_ID_SONIC_LS, |
| 1087 | .priv_data_size = sizeof(SonicContext), |
| 1088 | .init = sonic_encode_init, |
| 1089 | .encode2 = sonic_encode_frame, |
| 1090 | .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE }, |
| 1091 | .capabilities = CODEC_CAP_EXPERIMENTAL, |
| 1092 | .close = sonic_encode_close, |
| 1093 | }; |
| 1094 | #endif |