1 /* Copyright (c) 2007-2008 CSIRO 2 Copyright (c) 2007-2009 Xiph.Org Foundation 3 Written by Jean-Marc Valin */ 4 /* 5 Redistribution and use in source and binary forms, with or without 6 modification, are permitted provided that the following conditions 7 are met: 8 9 - Redistributions of source code must retain the above copyright 10 notice, this list of conditions and the following disclaimer. 11 12 - Redistributions in binary form must reproduce the above copyright 13 notice, this list of conditions and the following disclaimer in the 14 documentation and/or other materials provided with the distribution. 15 16 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 17 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 18 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 19 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER 20 OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 21 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 22 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 23 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 24 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 25 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 26 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 #ifdef HAVE_CONFIG_H 30 #include "config.h" 31 #endif 32 33 #include "quant_bands.h" 34 #include "laplace.h" 35 #include <math.h> 36 #include "os_support.h" 37 #include "arch.h" 38 #include "mathops.h" 39 #include "stack_alloc.h" 40 #include "rate.h" 41 42 #ifdef FIXED_POINT 43 /* Mean energy in each band quantized in Q4 */ 44 const signed char eMeans[25] = { 45 103,100, 92, 85, 81, 46 77, 72, 70, 78, 75, 47 73, 71, 78, 74, 69, 48 72, 70, 74, 76, 71, 49 60, 60, 60, 60, 60 50 }; 51 #else 52 /* Mean energy in each band quantized in Q4 and converted back to float */ 53 const opus_val16 eMeans[25] = { 54 6.437500f, 6.250000f, 5.750000f, 5.312500f, 5.062500f, 55 4.812500f, 4.500000f, 4.375000f, 4.875000f, 4.687500f, 56 4.562500f, 4.437500f, 4.875000f, 4.625000f, 4.312500f, 57 4.500000f, 4.375000f, 4.625000f, 4.750000f, 4.437500f, 58 3.750000f, 3.750000f, 3.750000f, 3.750000f, 3.750000f 59 }; 60 #endif 61 /* prediction coefficients: 0.9, 0.8, 0.65, 0.5 */ 62 #ifdef FIXED_POINT 63 static const opus_val16 pred_coef[4] = {29440, 26112, 21248, 16384}; 64 static const opus_val16 beta_coef[4] = {30147, 22282, 12124, 6554}; 65 static const opus_val16 beta_intra = 4915; 66 #else 67 static const opus_val16 pred_coef[4] = {29440/32768., 26112/32768., 21248/32768., 16384/32768.}; 68 static const opus_val16 beta_coef[4] = {30147/32768., 22282/32768., 12124/32768., 6554/32768.}; 69 static const opus_val16 beta_intra = 4915/32768.; 70 #endif 71 72 /*Parameters of the Laplace-like probability models used for the coarse energy. 73 There is one pair of parameters for each frame size, prediction type 74 (inter/intra), and band number. 75 The first number of each pair is the probability of 0, and the second is the 76 decay rate, both in Q8 precision.*/ 77 static const unsigned char e_prob_model[4][2][42] = { 78 /*120 sample frames.*/ 79 { 80 /*Inter*/ 81 { 82 72, 127, 65, 129, 66, 128, 65, 128, 64, 128, 62, 128, 64, 128, 83 64, 128, 92, 78, 92, 79, 92, 78, 90, 79, 116, 41, 115, 40, 84 114, 40, 132, 26, 132, 26, 145, 17, 161, 12, 176, 10, 177, 11 85 }, 86 /*Intra*/ 87 { 88 24, 179, 48, 138, 54, 135, 54, 132, 53, 134, 56, 133, 55, 132, 89 55, 132, 61, 114, 70, 96, 74, 88, 75, 88, 87, 74, 89, 66, 90 91, 67, 100, 59, 108, 50, 120, 40, 122, 37, 97, 43, 78, 50 91 } 92 }, 93 /*240 sample frames.*/ 94 { 95 /*Inter*/ 96 { 97 83, 78, 84, 81, 88, 75, 86, 74, 87, 71, 90, 73, 93, 74, 98 93, 74, 109, 40, 114, 36, 117, 34, 117, 34, 143, 17, 145, 18, 99 146, 19, 162, 12, 165, 10, 178, 7, 189, 6, 190, 8, 177, 9 100 }, 101 /*Intra*/ 102 { 103 23, 178, 54, 115, 63, 102, 66, 98, 69, 99, 74, 89, 71, 91, 104 73, 91, 78, 89, 86, 80, 92, 66, 93, 64, 102, 59, 103, 60, 105 104, 60, 117, 52, 123, 44, 138, 35, 133, 31, 97, 38, 77, 45 106 } 107 }, 108 /*480 sample frames.*/ 109 { 110 /*Inter*/ 111 { 112 61, 90, 93, 60, 105, 42, 107, 41, 110, 45, 116, 38, 113, 38, 113 112, 38, 124, 26, 132, 27, 136, 19, 140, 20, 155, 14, 159, 16, 114 158, 18, 170, 13, 177, 10, 187, 8, 192, 6, 175, 9, 159, 10 115 }, 116 /*Intra*/ 117 { 118 21, 178, 59, 110, 71, 86, 75, 85, 84, 83, 91, 66, 88, 73, 119 87, 72, 92, 75, 98, 72, 105, 58, 107, 54, 115, 52, 114, 55, 120 112, 56, 129, 51, 132, 40, 150, 33, 140, 29, 98, 35, 77, 42 121 } 122 }, 123 /*960 sample frames.*/ 124 { 125 /*Inter*/ 126 { 127 42, 121, 96, 66, 108, 43, 111, 40, 117, 44, 123, 32, 120, 36, 128 119, 33, 127, 33, 134, 34, 139, 21, 147, 23, 152, 20, 158, 25, 129 154, 26, 166, 21, 173, 16, 184, 13, 184, 10, 150, 13, 139, 15 130 }, 131 /*Intra*/ 132 { 133 22, 178, 63, 114, 74, 82, 84, 83, 92, 82, 103, 62, 96, 72, 134 96, 67, 101, 73, 107, 72, 113, 55, 118, 52, 125, 52, 118, 52, 135 117, 55, 135, 49, 137, 39, 157, 32, 145, 29, 97, 33, 77, 40 136 } 137 } 138 }; 139 140 static const unsigned char small_energy_icdf[3]={2,1,0}; 141 142 static opus_val32 loss_distortion(const opus_val16 *eBands, opus_val16 *oldEBands, int start, int end, int len, int C) 143 { 144 int c, i; 145 opus_val32 dist = 0; 146 c=0; do { 147 for (i=start;i<end;i++) 148 { 149 opus_val16 d = SUB16(SHR16(eBands[i+c*len], 3), SHR16(oldEBands[i+c*len], 3)); 150 dist = MAC16_16(dist, d,d); 151 } 152 } while (++c<C); 153 return MIN32(200,SHR32(dist,2*DB_SHIFT-6)); 154 } 155 156 static int quant_coarse_energy_impl(const CELTMode *m, int start, int end, 157 const opus_val16 *eBands, opus_val16 *oldEBands, 158 opus_int32 budget, opus_int32 tell, 159 const unsigned char *prob_model, opus_val16 *error, ec_enc *enc, 160 int C, int LM, int intra, opus_val16 max_decay, int lfe) 161 { 162 int i, c; 163 int badness = 0; 164 opus_val32 prev[2] = {0,0}; 165 opus_val16 coef; 166 opus_val16 beta; 167 168 if (tell+3 <= budget) 169 ec_enc_bit_logp(enc, intra, 3); 170 if (intra) 171 { 172 coef = 0; 173 beta = beta_intra; 174 } else { 175 beta = beta_coef[LM]; 176 coef = pred_coef[LM]; 177 } 178 179 /* Encode at a fixed coarse resolution */ 180 for (i=start;i<end;i++) 181 { 182 c=0; 183 do { 184 int bits_left; 185 int qi, qi0; 186 opus_val32 q; 187 opus_val16 x; 188 opus_val32 f, tmp; 189 opus_val16 oldE; 190 opus_val16 decay_bound; 191 x = eBands[i+c*m->nbEBands]; 192 oldE = MAX16(-QCONST16(9.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]); 193 #ifdef FIXED_POINT 194 f = SHL32(EXTEND32(x),7) - PSHR32(MULT16_16(coef,oldE), 8) - prev[c]; 195 /* Rounding to nearest integer here is really important! */ 196 qi = (f+QCONST32(.5f,DB_SHIFT+7))>>(DB_SHIFT+7); 197 decay_bound = EXTRACT16(MAX32(-QCONST16(28.f,DB_SHIFT), 198 SUB32((opus_val32)oldEBands[i+c*m->nbEBands],max_decay))); 199 #else 200 f = x-coef*oldE-prev[c]; 201 /* Rounding to nearest integer here is really important! */ 202 qi = (int)floor(.5f+f); 203 decay_bound = MAX16(-QCONST16(28.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]) - max_decay; 204 #endif 205 /* Prevent the energy from going down too quickly (e.g. for bands 206 that have just one bin) */ 207 if (qi < 0 && x < decay_bound) 208 { 209 qi += (int)SHR16(SUB16(decay_bound,x), DB_SHIFT); 210 if (qi > 0) 211 qi = 0; 212 } 213 qi0 = qi; 214 /* If we don't have enough bits to encode all the energy, just assume 215 something safe. */ 216 tell = ec_tell(enc); 217 bits_left = budget-tell-3*C*(end-i); 218 if (i!=start && bits_left < 30) 219 { 220 if (bits_left < 24) 221 qi = IMIN(1, qi); 222 if (bits_left < 16) 223 qi = IMAX(-1, qi); 224 } 225 if (lfe && i>=2) 226 qi = IMIN(qi, 0); 227 if (budget-tell >= 15) 228 { 229 int pi; 230 pi = 2*IMIN(i,20); 231 ec_laplace_encode(enc, &qi, 232 prob_model[pi]<<7, prob_model[pi+1]<<6); 233 } 234 else if(budget-tell >= 2) 235 { 236 qi = IMAX(-1, IMIN(qi, 1)); 237 ec_enc_icdf(enc, 2*qi^-(qi<0), small_energy_icdf, 2); 238 } 239 else if(budget-tell >= 1) 240 { 241 qi = IMIN(0, qi); 242 ec_enc_bit_logp(enc, -qi, 1); 243 } 244 else 245 qi = -1; 246 error[i+c*m->nbEBands] = PSHR32(f,7) - SHL16(qi,DB_SHIFT); 247 badness += abs(qi0-qi); 248 q = (opus_val32)SHL32(EXTEND32(qi),DB_SHIFT); 249 250 tmp = PSHR32(MULT16_16(coef,oldE),8) + prev[c] + SHL32(q,7); 251 #ifdef FIXED_POINT 252 tmp = MAX32(-QCONST32(28.f, DB_SHIFT+7), tmp); 253 #endif 254 oldEBands[i+c*m->nbEBands] = PSHR32(tmp, 7); 255 prev[c] = prev[c] + SHL32(q,7) - MULT16_16(beta,PSHR32(q,8)); 256 } while (++c < C); 257 } 258 return lfe ? 0 : badness; 259 } 260 261 void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, 262 const opus_val16 *eBands, opus_val16 *oldEBands, opus_uint32 budget, 263 opus_val16 *error, ec_enc *enc, int C, int LM, int nbAvailableBytes, 264 int force_intra, opus_val32 *delayedIntra, int two_pass, int loss_rate, int lfe) 265 { 266 int intra; 267 opus_val16 max_decay; 268 VARDECL(opus_val16, oldEBands_intra); 269 VARDECL(opus_val16, error_intra); 270 ec_enc enc_start_state; 271 opus_uint32 tell; 272 int badness1=0; 273 opus_int32 intra_bias; 274 opus_val32 new_distortion; 275 SAVE_STACK; 276 277 intra = force_intra || (!two_pass && *delayedIntra>2*C*(end-start) && nbAvailableBytes > (end-start)*C); 278 intra_bias = (opus_int32)((budget**delayedIntra*loss_rate)/(C*512)); 279 new_distortion = loss_distortion(eBands, oldEBands, start, effEnd, m->nbEBands, C); 280 281 tell = ec_tell(enc); 282 if (tell+3 > budget) 283 two_pass = intra = 0; 284 285 max_decay = QCONST16(16.f,DB_SHIFT); 286 if (end-start>10) 287 { 288 #ifdef FIXED_POINT 289 max_decay = MIN32(max_decay, SHL32(EXTEND32(nbAvailableBytes),DB_SHIFT-3)); 290 #else 291 max_decay = MIN32(max_decay, .125f*nbAvailableBytes); 292 #endif 293 } 294 if (lfe) 295 max_decay=3; 296 enc_start_state = *enc; 297 298 ALLOC(oldEBands_intra, C*m->nbEBands, opus_val16); 299 ALLOC(error_intra, C*m->nbEBands, opus_val16); 300 OPUS_COPY(oldEBands_intra, oldEBands, C*m->nbEBands); 301 302 if (two_pass || intra) 303 { 304 badness1 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands_intra, budget, 305 tell, e_prob_model[LM][1], error_intra, enc, C, LM, 1, max_decay, lfe); 306 } 307 308 if (!intra) 309 { 310 unsigned char *intra_buf; 311 ec_enc enc_intra_state; 312 opus_int32 tell_intra; 313 opus_uint32 nstart_bytes; 314 opus_uint32 nintra_bytes; 315 int badness2; 316 VARDECL(unsigned char, intra_bits); 317 318 tell_intra = ec_tell_frac(enc); 319 320 enc_intra_state = *enc; 321 322 nstart_bytes = ec_range_bytes(&enc_start_state); 323 nintra_bytes = ec_range_bytes(&enc_intra_state); 324 intra_buf = ec_get_buffer(&enc_intra_state) + nstart_bytes; 325 ALLOC(intra_bits, nintra_bytes-nstart_bytes, unsigned char); 326 /* Copy bits from intra bit-stream */ 327 OPUS_COPY(intra_bits, intra_buf, nintra_bytes - nstart_bytes); 328 329 *enc = enc_start_state; 330 331 badness2 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands, budget, 332 tell, e_prob_model[LM][intra], error, enc, C, LM, 0, max_decay, lfe); 333 334 if (two_pass && (badness1 < badness2 || (badness1 == badness2 && ((opus_int32)ec_tell_frac(enc))+intra_bias > tell_intra))) 335 { 336 *enc = enc_intra_state; 337 /* Copy intra bits to bit-stream */ 338 OPUS_COPY(intra_buf, intra_bits, nintra_bytes - nstart_bytes); 339 OPUS_COPY(oldEBands, oldEBands_intra, C*m->nbEBands); 340 OPUS_COPY(error, error_intra, C*m->nbEBands); 341 intra = 1; 342 } 343 } else { 344 OPUS_COPY(oldEBands, oldEBands_intra, C*m->nbEBands); 345 OPUS_COPY(error, error_intra, C*m->nbEBands); 346 } 347 348 if (intra) 349 *delayedIntra = new_distortion; 350 else 351 *delayedIntra = ADD32(MULT16_32_Q15(MULT16_16_Q15(pred_coef[LM], pred_coef[LM]),*delayedIntra), 352 new_distortion); 353 354 RESTORE_STACK; 355 } 356 357 void quant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, ec_enc *enc, int C) 358 { 359 int i, c; 360 361 /* Encode finer resolution */ 362 for (i=start;i<end;i++) 363 { 364 opus_int16 frac = 1<<fine_quant[i]; 365 if (fine_quant[i] <= 0) 366 continue; 367 c=0; 368 do { 369 int q2; 370 opus_val16 offset; 371 #ifdef FIXED_POINT 372 /* Has to be without rounding */ 373 q2 = (error[i+c*m->nbEBands]+QCONST16(.5f,DB_SHIFT))>>(DB_SHIFT-fine_quant[i]); 374 #else 375 q2 = (int)floor((error[i+c*m->nbEBands]+.5f)*frac); 376 #endif 377 if (q2 > frac-1) 378 q2 = frac-1; 379 if (q2<0) 380 q2 = 0; 381 ec_enc_bits(enc, q2, fine_quant[i]); 382 #ifdef FIXED_POINT 383 offset = SUB16(SHR32(SHL32(EXTEND32(q2),DB_SHIFT)+QCONST16(.5f,DB_SHIFT),fine_quant[i]),QCONST16(.5f,DB_SHIFT)); 384 #else 385 offset = (q2+.5f)*(1<<(14-fine_quant[i]))*(1.f/16384) - .5f; 386 #endif 387 oldEBands[i+c*m->nbEBands] += offset; 388 error[i+c*m->nbEBands] -= offset; 389 /*printf ("%f ", error[i] - offset);*/ 390 } while (++c < C); 391 } 392 } 393 394 void quant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, int *fine_priority, int bits_left, ec_enc *enc, int C) 395 { 396 int i, prio, c; 397 398 /* Use up the remaining bits */ 399 for (prio=0;prio<2;prio++) 400 { 401 for (i=start;i<end && bits_left>=C ;i++) 402 { 403 if (fine_quant[i] >= MAX_FINE_BITS || fine_priority[i]!=prio) 404 continue; 405 c=0; 406 do { 407 int q2; 408 opus_val16 offset; 409 q2 = error[i+c*m->nbEBands]<0 ? 0 : 1; 410 ec_enc_bits(enc, q2, 1); 411 #ifdef FIXED_POINT 412 offset = SHR16(SHL16(q2,DB_SHIFT)-QCONST16(.5f,DB_SHIFT),fine_quant[i]+1); 413 #else 414 offset = (q2-.5f)*(1<<(14-fine_quant[i]-1))*(1.f/16384); 415 #endif 416 oldEBands[i+c*m->nbEBands] += offset; 417 bits_left--; 418 } while (++c < C); 419 } 420 } 421 } 422 423 void unquant_coarse_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int intra, ec_dec *dec, int C, int LM) 424 { 425 const unsigned char *prob_model = e_prob_model[LM][intra]; 426 int i, c; 427 opus_val32 prev[2] = {0, 0}; 428 opus_val16 coef; 429 opus_val16 beta; 430 opus_int32 budget; 431 opus_int32 tell; 432 433 if (intra) 434 { 435 coef = 0; 436 beta = beta_intra; 437 } else { 438 beta = beta_coef[LM]; 439 coef = pred_coef[LM]; 440 } 441 442 budget = dec->storage*8; 443 444 /* Decode at a fixed coarse resolution */ 445 for (i=start;i<end;i++) 446 { 447 c=0; 448 do { 449 int qi; 450 opus_val32 q; 451 opus_val32 tmp; 452 /* It would be better to express this invariant as a 453 test on C at function entry, but that isn't enough 454 to make the static analyzer happy. */ 455 celt_assert(c<2); 456 tell = ec_tell(dec); 457 if(budget-tell>=15) 458 { 459 int pi; 460 pi = 2*IMIN(i,20); 461 qi = ec_laplace_decode(dec, 462 prob_model[pi]<<7, prob_model[pi+1]<<6); 463 } 464 else if(budget-tell>=2) 465 { 466 qi = ec_dec_icdf(dec, small_energy_icdf, 2); 467 qi = (qi>>1)^-(qi&1); 468 } 469 else if(budget-tell>=1) 470 { 471 qi = -ec_dec_bit_logp(dec, 1); 472 } 473 else 474 qi = -1; 475 q = (opus_val32)SHL32(EXTEND32(qi),DB_SHIFT); 476 477 oldEBands[i+c*m->nbEBands] = MAX16(-QCONST16(9.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]); 478 tmp = PSHR32(MULT16_16(coef,oldEBands[i+c*m->nbEBands]),8) + prev[c] + SHL32(q,7); 479 #ifdef FIXED_POINT 480 tmp = MAX32(-QCONST32(28.f, DB_SHIFT+7), tmp); 481 #endif 482 oldEBands[i+c*m->nbEBands] = PSHR32(tmp, 7); 483 prev[c] = prev[c] + SHL32(q,7) - MULT16_16(beta,PSHR32(q,8)); 484 } while (++c < C); 485 } 486 } 487 488 void unquant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, ec_dec *dec, int C) 489 { 490 int i, c; 491 /* Decode finer resolution */ 492 for (i=start;i<end;i++) 493 { 494 if (fine_quant[i] <= 0) 495 continue; 496 c=0; 497 do { 498 int q2; 499 opus_val16 offset; 500 q2 = ec_dec_bits(dec, fine_quant[i]); 501 #ifdef FIXED_POINT 502 offset = SUB16(SHR32(SHL32(EXTEND32(q2),DB_SHIFT)+QCONST16(.5f,DB_SHIFT),fine_quant[i]),QCONST16(.5f,DB_SHIFT)); 503 #else 504 offset = (q2+.5f)*(1<<(14-fine_quant[i]))*(1.f/16384) - .5f; 505 #endif 506 oldEBands[i+c*m->nbEBands] += offset; 507 } while (++c < C); 508 } 509 } 510 511 void unquant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, int *fine_priority, int bits_left, ec_dec *dec, int C) 512 { 513 int i, prio, c; 514 515 /* Use up the remaining bits */ 516 for (prio=0;prio<2;prio++) 517 { 518 for (i=start;i<end && bits_left>=C ;i++) 519 { 520 if (fine_quant[i] >= MAX_FINE_BITS || fine_priority[i]!=prio) 521 continue; 522 c=0; 523 do { 524 int q2; 525 opus_val16 offset; 526 q2 = ec_dec_bits(dec, 1); 527 #ifdef FIXED_POINT 528 offset = SHR16(SHL16(q2,DB_SHIFT)-QCONST16(.5f,DB_SHIFT),fine_quant[i]+1); 529 #else 530 offset = (q2-.5f)*(1<<(14-fine_quant[i]-1))*(1.f/16384); 531 #endif 532 oldEBands[i+c*m->nbEBands] += offset; 533 bits_left--; 534 } while (++c < C); 535 } 536 } 537 } 538 539 void amp2Log2(const CELTMode *m, int effEnd, int end, 540 celt_ener *bandE, opus_val16 *bandLogE, int C) 541 { 542 int c, i; 543 c=0; 544 do { 545 for (i=0;i<effEnd;i++) 546 bandLogE[i+c*m->nbEBands] = 547 celt_log2(SHL32(bandE[i+c*m->nbEBands],2)) 548 - SHL16((opus_val16)eMeans[i],6); 549 for (i=effEnd;i<end;i++) 550 bandLogE[c*m->nbEBands+i] = -QCONST16(14.f,DB_SHIFT); 551 } while (++c < C); 552 } 553