FFmpeg
opus_pvq.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2007-2008 CSIRO
3  * Copyright (c) 2007-2009 Xiph.Org Foundation
4  * Copyright (c) 2008-2009 Gregory Maxwell
5  * Copyright (c) 2012 Andrew D'Addesio
6  * Copyright (c) 2013-2014 Mozilla Corporation
7  * Copyright (c) 2017 Rostislav Pehlivanov <atomnuker@gmail.com>
8  *
9  * This file is part of FFmpeg.
10  *
11  * FFmpeg is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU Lesser General Public
13  * License as published by the Free Software Foundation; either
14  * version 2.1 of the License, or (at your option) any later version.
15  *
16  * FFmpeg is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19  * Lesser General Public License for more details.
20  *
21  * You should have received a copy of the GNU Lesser General Public
22  * License along with FFmpeg; if not, write to the Free Software
23  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24  */
25 
26 #include "config_components.h"
27 
28 #include "opustab.h"
29 #include "opus_pvq.h"
30 
31 #define CELT_PVQ_U(n, k) (ff_celt_pvq_u_row[FFMIN(n, k)][FFMAX(n, k)])
32 #define CELT_PVQ_V(n, k) (CELT_PVQ_U(n, k) + CELT_PVQ_U(n, (k) + 1))
33 
34 static inline int16_t celt_cos(int16_t x)
35 {
36  x = (MUL16(x, x) + 4096) >> 13;
37  x = (32767-x) + ROUND_MUL16(x, (-7651 + ROUND_MUL16(x, (8277 + ROUND_MUL16(-626, x)))));
38  return x + 1;
39 }
40 
41 static inline int celt_log2tan(int isin, int icos)
42 {
43  int lc, ls;
44  lc = opus_ilog(icos);
45  ls = opus_ilog(isin);
46  icos <<= 15 - lc;
47  isin <<= 15 - ls;
48  return (ls << 11) - (lc << 11) +
49  ROUND_MUL16(isin, ROUND_MUL16(isin, -2597) + 7932) -
50  ROUND_MUL16(icos, ROUND_MUL16(icos, -2597) + 7932);
51 }
52 
53 static inline int celt_bits2pulses(const uint8_t *cache, int bits)
54 {
55  // TODO: Find the size of cache and make it into an array in the parameters list
56  int i, low = 0, high;
57 
58  high = cache[0];
59  bits--;
60 
61  for (i = 0; i < 6; i++) {
62  int center = (low + high + 1) >> 1;
63  if (cache[center] >= bits)
64  high = center;
65  else
66  low = center;
67  }
68 
69  return (bits - (low == 0 ? -1 : cache[low]) <= cache[high] - bits) ? low : high;
70 }
71 
72 static inline int celt_pulses2bits(const uint8_t *cache, int pulses)
73 {
74  // TODO: Find the size of cache and make it into an array in the parameters list
75  return (pulses == 0) ? 0 : cache[pulses] + 1;
76 }
77 
78 static inline void celt_normalize_residual(const int * av_restrict iy, float * av_restrict X,
79  int N, float g)
80 {
81  int i;
82  for (i = 0; i < N; i++)
83  X[i] = g * iy[i];
84 }
85 
86 static void celt_exp_rotation_impl(float *X, uint32_t len, uint32_t stride,
87  float c, float s)
88 {
89  float *Xptr;
90  int i;
91 
92  Xptr = X;
93  for (i = 0; i < len - stride; i++) {
94  float x1 = Xptr[0];
95  float x2 = Xptr[stride];
96  Xptr[stride] = c * x2 + s * x1;
97  *Xptr++ = c * x1 - s * x2;
98  }
99 
100  Xptr = &X[len - 2 * stride - 1];
101  for (i = len - 2 * stride - 1; i >= 0; i--) {
102  float x1 = Xptr[0];
103  float x2 = Xptr[stride];
104  Xptr[stride] = c * x2 + s * x1;
105  *Xptr-- = c * x1 - s * x2;
106  }
107 }
108 
109 static inline void celt_exp_rotation(float *X, uint32_t len,
110  uint32_t stride, uint32_t K,
111  enum CeltSpread spread, const int encode)
112 {
113  uint32_t stride2 = 0;
114  float c, s;
115  float gain, theta;
116  int i;
117 
118  if (2*K >= len || spread == CELT_SPREAD_NONE)
119  return;
120 
121  gain = (float)len / (len + (20 - 5*spread) * K);
122  theta = M_PI * gain * gain / 4;
123 
124  c = cosf(theta);
125  s = sinf(theta);
126 
127  if (len >= stride << 3) {
128  stride2 = 1;
129  /* This is just a simple (equivalent) way of computing sqrt(len/stride) with rounding.
130  It's basically incrementing long as (stride2+0.5)^2 < len/stride. */
131  while ((stride2 * stride2 + stride2) * stride + (stride >> 2) < len)
132  stride2++;
133  }
134 
135  len /= stride;
136  for (i = 0; i < stride; i++) {
137  if (encode) {
138  celt_exp_rotation_impl(X + i * len, len, 1, c, -s);
139  if (stride2)
140  celt_exp_rotation_impl(X + i * len, len, stride2, s, -c);
141  } else {
142  if (stride2)
143  celt_exp_rotation_impl(X + i * len, len, stride2, s, c);
144  celt_exp_rotation_impl(X + i * len, len, 1, c, s);
145  }
146  }
147 }
148 
149 static inline uint32_t celt_extract_collapse_mask(const int *iy, uint32_t N, uint32_t B)
150 {
151  int i, j, N0 = N / B;
152  uint32_t collapse_mask = 0;
153 
154  if (B <= 1)
155  return 1;
156 
157  for (i = 0; i < B; i++)
158  for (j = 0; j < N0; j++)
159  collapse_mask |= (!!iy[i*N0+j]) << i;
160  return collapse_mask;
161 }
162 
163 static inline void celt_stereo_merge(float *X, float *Y, float mid, int N)
164 {
165  int i;
166  float xp = 0, side = 0;
167  float E[2];
168  float mid2;
169  float gain[2];
170 
171  /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */
172  for (i = 0; i < N; i++) {
173  xp += X[i] * Y[i];
174  side += Y[i] * Y[i];
175  }
176 
177  /* Compensating for the mid normalization */
178  xp *= mid;
179  mid2 = mid;
180  E[0] = mid2 * mid2 + side - 2 * xp;
181  E[1] = mid2 * mid2 + side + 2 * xp;
182  if (E[0] < 6e-4f || E[1] < 6e-4f) {
183  for (i = 0; i < N; i++)
184  Y[i] = X[i];
185  return;
186  }
187 
188  gain[0] = 1.0f / sqrtf(E[0]);
189  gain[1] = 1.0f / sqrtf(E[1]);
190 
191  for (i = 0; i < N; i++) {
192  float value[2];
193  /* Apply mid scaling (side is already scaled) */
194  value[0] = mid * X[i];
195  value[1] = Y[i];
196  X[i] = gain[0] * (value[0] - value[1]);
197  Y[i] = gain[1] * (value[0] + value[1]);
198  }
199 }
200 
201 static void celt_interleave_hadamard(float *tmp, float *X, int N0,
202  int stride, int hadamard)
203 {
204  int i, j, N = N0*stride;
205  const uint8_t *order = &ff_celt_hadamard_order[hadamard ? stride - 2 : 30];
206 
207  for (i = 0; i < stride; i++)
208  for (j = 0; j < N0; j++)
209  tmp[j*stride+i] = X[order[i]*N0+j];
210 
211  memcpy(X, tmp, N*sizeof(float));
212 }
213 
214 static void celt_deinterleave_hadamard(float *tmp, float *X, int N0,
215  int stride, int hadamard)
216 {
217  int i, j, N = N0*stride;
218  const uint8_t *order = &ff_celt_hadamard_order[hadamard ? stride - 2 : 30];
219 
220  for (i = 0; i < stride; i++)
221  for (j = 0; j < N0; j++)
222  tmp[order[i]*N0+j] = X[j*stride+i];
223 
224  memcpy(X, tmp, N*sizeof(float));
225 }
226 
227 static void celt_haar1(float *X, int N0, int stride)
228 {
229  int i, j;
230  N0 >>= 1;
231  for (i = 0; i < stride; i++) {
232  for (j = 0; j < N0; j++) {
233  float x0 = X[stride * (2 * j + 0) + i];
234  float x1 = X[stride * (2 * j + 1) + i];
235  X[stride * (2 * j + 0) + i] = (x0 + x1) * M_SQRT1_2;
236  X[stride * (2 * j + 1) + i] = (x0 - x1) * M_SQRT1_2;
237  }
238  }
239 }
240 
241 static inline int celt_compute_qn(int N, int b, int offset, int pulse_cap,
242  int stereo)
243 {
244  int qn, qb;
245  int N2 = 2 * N - 1;
246  if (stereo && N == 2)
247  N2--;
248 
249  /* The upper limit ensures that in a stereo split with itheta==16384, we'll
250  * always have enough bits left over to code at least one pulse in the
251  * side; otherwise it would collapse, since it doesn't get folded. */
252  qb = FFMIN3(b - pulse_cap - (4 << 3), (b + N2 * offset) / N2, 8 << 3);
253  qn = (qb < (1 << 3 >> 1)) ? 1 : ((ff_celt_qn_exp2[qb & 0x7] >> (14 - (qb >> 3))) + 1) >> 1 << 1;
254  return qn;
255 }
256 
257 /* Convert the quantized vector to an index */
258 static inline uint32_t celt_icwrsi(uint32_t N, uint32_t K, const int *y)
259 {
260  int i, idx = 0, sum = 0;
261  for (i = N - 1; i >= 0; i--) {
262  const uint32_t i_s = CELT_PVQ_U(N - i, sum + FFABS(y[i]) + 1);
263  idx += CELT_PVQ_U(N - i, sum) + (y[i] < 0)*i_s;
264  sum += FFABS(y[i]);
265  }
266  return idx;
267 }
268 
269 // this code was adapted from libopus
270 static inline uint64_t celt_cwrsi(uint32_t N, uint32_t K, uint32_t i, int *y)
271 {
272  uint64_t norm = 0;
273  uint32_t q, p;
274  int s, val;
275  int k0;
276 
277  while (N > 2) {
278  /*Lots of pulses case:*/
279  if (K >= N) {
280  const uint32_t *row = ff_celt_pvq_u_row[N];
281 
282  /* Are the pulses in this dimension negative? */
283  p = row[K + 1];
284  s = -(i >= p);
285  i -= p & s;
286 
287  /*Count how many pulses were placed in this dimension.*/
288  k0 = K;
289  q = row[N];
290  if (q > i) {
291  K = N;
292  do {
293  p = ff_celt_pvq_u_row[--K][N];
294  } while (p > i);
295  } else
296  for (p = row[K]; p > i; p = row[K])
297  K--;
298 
299  i -= p;
300  val = (k0 - K + s) ^ s;
301  norm += val * val;
302  *y++ = val;
303  } else { /*Lots of dimensions case:*/
304  /*Are there any pulses in this dimension at all?*/
305  p = ff_celt_pvq_u_row[K ][N];
306  q = ff_celt_pvq_u_row[K + 1][N];
307 
308  if (p <= i && i < q) {
309  i -= p;
310  *y++ = 0;
311  } else {
312  /*Are the pulses in this dimension negative?*/
313  s = -(i >= q);
314  i -= q & s;
315 
316  /*Count how many pulses were placed in this dimension.*/
317  k0 = K;
318  do p = ff_celt_pvq_u_row[--K][N];
319  while (p > i);
320 
321  i -= p;
322  val = (k0 - K + s) ^ s;
323  norm += val * val;
324  *y++ = val;
325  }
326  }
327  N--;
328  }
329 
330  /* N == 2 */
331  p = 2 * K + 1;
332  s = -(i >= p);
333  i -= p & s;
334  k0 = K;
335  K = (i + 1) / 2;
336 
337  if (K)
338  i -= 2 * K - 1;
339 
340  val = (k0 - K + s) ^ s;
341  norm += val * val;
342  *y++ = val;
343 
344  /* N==1 */
345  s = -i;
346  val = (K + s) ^ s;
347  norm += val * val;
348  *y = val;
349 
350  return norm;
351 }
352 
353 static inline void celt_encode_pulses(OpusRangeCoder *rc, int *y, uint32_t N, uint32_t K)
354 {
355  ff_opus_rc_enc_uint(rc, celt_icwrsi(N, K, y), CELT_PVQ_V(N, K));
356 }
357 
358 static inline float celt_decode_pulses(OpusRangeCoder *rc, int *y, uint32_t N, uint32_t K)
359 {
360  const uint32_t idx = ff_opus_rc_dec_uint(rc, CELT_PVQ_V(N, K));
361  return celt_cwrsi(N, K, idx, y);
362 }
363 
364 /*
365  * Faster than libopus's search, operates entirely in the signed domain.
366  * Slightly worse/better depending on N, K and the input vector.
367  */
368 static float ppp_pvq_search_c(float *X, int *y, int K, int N)
369 {
370  int i, y_norm = 0;
371  float res = 0.0f, xy_norm = 0.0f;
372 
373  for (i = 0; i < N; i++)
374  res += FFABS(X[i]);
375 
376  res = K/(res + FLT_EPSILON);
377 
378  for (i = 0; i < N; i++) {
379  y[i] = lrintf(res*X[i]);
380  y_norm += y[i]*y[i];
381  xy_norm += y[i]*X[i];
382  K -= FFABS(y[i]);
383  }
384 
385  while (K) {
386  int max_idx = 0, phase = FFSIGN(K);
387  float max_num = 0.0f;
388  float max_den = 1.0f;
389  y_norm += 1.0f;
390 
391  for (i = 0; i < N; i++) {
392  /* If the sum has been overshot and the best place has 0 pulses allocated
393  * to it, attempting to decrease it further will actually increase the
394  * sum. Prevent this by disregarding any 0 positions when decrementing. */
395  const int ca = 1 ^ ((y[i] == 0) & (phase < 0));
396  const int y_new = y_norm + 2*phase*FFABS(y[i]);
397  float xy_new = xy_norm + 1*phase*FFABS(X[i]);
398  xy_new = xy_new * xy_new;
399  if (ca && (max_den*xy_new) > (y_new*max_num)) {
400  max_den = y_new;
401  max_num = xy_new;
402  max_idx = i;
403  }
404  }
405 
406  K -= phase;
407 
408  phase *= FFSIGN(X[max_idx]);
409  xy_norm += 1*phase*X[max_idx];
410  y_norm += 2*phase*y[max_idx];
411  y[max_idx] += phase;
412  }
413 
414  return (float)y_norm;
415 }
416 
417 static uint32_t celt_alg_quant(OpusRangeCoder *rc, float *X, uint32_t N, uint32_t K,
418  enum CeltSpread spread, uint32_t blocks, float gain,
419  CeltPVQ *pvq)
420 {
421  int *y = pvq->qcoeff;
422 
423  celt_exp_rotation(X, N, blocks, K, spread, 1);
424  gain /= sqrtf(pvq->pvq_search(X, y, K, N));
425  celt_encode_pulses(rc, y, N, K);
426  celt_normalize_residual(y, X, N, gain);
427  celt_exp_rotation(X, N, blocks, K, spread, 0);
428  return celt_extract_collapse_mask(y, N, blocks);
429 }
430 
431 /** Decode pulse vector and combine the result with the pitch vector to produce
432  the final normalised signal in the current band. */
433 static uint32_t celt_alg_unquant(OpusRangeCoder *rc, float *X, uint32_t N, uint32_t K,
434  enum CeltSpread spread, uint32_t blocks, float gain,
435  CeltPVQ *pvq)
436 {
437  int *y = pvq->qcoeff;
438 
439  gain /= sqrtf(celt_decode_pulses(rc, y, N, K));
440  celt_normalize_residual(y, X, N, gain);
441  celt_exp_rotation(X, N, blocks, K, spread, 0);
442  return celt_extract_collapse_mask(y, N, blocks);
443 }
444 
445 static int celt_calc_theta(const float *X, const float *Y, int coupling, int N)
446 {
447  int i;
448  float e[2] = { 0.0f, 0.0f };
449  if (coupling) { /* Coupling case */
450  for (i = 0; i < N; i++) {
451  e[0] += (X[i] + Y[i])*(X[i] + Y[i]);
452  e[1] += (X[i] - Y[i])*(X[i] - Y[i]);
453  }
454  } else {
455  for (i = 0; i < N; i++) {
456  e[0] += X[i]*X[i];
457  e[1] += Y[i]*Y[i];
458  }
459  }
460  return lrintf(32768.0f*atan2f(sqrtf(e[1]), sqrtf(e[0]))/M_PI);
461 }
462 
463 static void celt_stereo_is_decouple(float *X, float *Y, float e_l, float e_r, int N)
464 {
465  int i;
466  const float energy_n = 1.0f/(sqrtf(e_l*e_l + e_r*e_r) + FLT_EPSILON);
467  e_l *= energy_n;
468  e_r *= energy_n;
469  for (i = 0; i < N; i++)
470  X[i] = e_l*X[i] + e_r*Y[i];
471 }
472 
473 static void celt_stereo_ms_decouple(float *X, float *Y, int N)
474 {
475  int i;
476  for (i = 0; i < N; i++) {
477  const float Xret = X[i];
478  X[i] = (X[i] + Y[i])*M_SQRT1_2;
479  Y[i] = (Y[i] - Xret)*M_SQRT1_2;
480  }
481 }
482 
484  OpusRangeCoder *rc,
485  const int band, float *X,
486  float *Y, int N, int b,
487  uint32_t blocks, float *lowband,
488  int duration, float *lowband_out,
489  int level, float gain,
490  float *lowband_scratch,
491  int fill, int quant)
492 {
493  int i;
494  const uint8_t *cache;
495  int stereo = !!Y, split = stereo;
496  int imid = 0, iside = 0;
497  uint32_t N0 = N;
498  int N_B = N / blocks;
499  int N_B0 = N_B;
500  int B0 = blocks;
501  int time_divide = 0;
502  int recombine = 0;
503  int inv = 0;
504  float mid = 0, side = 0;
505  int longblocks = (B0 == 1);
506  uint32_t cm = 0;
507 
508  if (N == 1) {
509  float *x = X;
510  for (i = 0; i <= stereo; i++) {
511  int sign = 0;
512  if (f->remaining2 >= 1 << 3) {
513  if (quant) {
514  sign = x[0] < 0;
515  ff_opus_rc_put_raw(rc, sign, 1);
516  } else {
517  sign = ff_opus_rc_get_raw(rc, 1);
518  }
519  f->remaining2 -= 1 << 3;
520  }
521  x[0] = 1.0f - 2.0f*sign;
522  x = Y;
523  }
524  if (lowband_out)
525  lowband_out[0] = X[0];
526  return 1;
527  }
528 
529  if (!stereo && level == 0) {
530  int tf_change = f->tf_change[band];
531  int k;
532  if (tf_change > 0)
533  recombine = tf_change;
534  /* Band recombining to increase frequency resolution */
535 
536  if (lowband &&
537  (recombine || ((N_B & 1) == 0 && tf_change < 0) || B0 > 1)) {
538  for (i = 0; i < N; i++)
539  lowband_scratch[i] = lowband[i];
540  lowband = lowband_scratch;
541  }
542 
543  for (k = 0; k < recombine; k++) {
544  if (quant || lowband)
545  celt_haar1(quant ? X : lowband, N >> k, 1 << k);
546  fill = ff_celt_bit_interleave[fill & 0xF] | ff_celt_bit_interleave[fill >> 4] << 2;
547  }
548  blocks >>= recombine;
549  N_B <<= recombine;
550 
551  /* Increasing the time resolution */
552  while ((N_B & 1) == 0 && tf_change < 0) {
553  if (quant || lowband)
554  celt_haar1(quant ? X : lowband, N_B, blocks);
555  fill |= fill << blocks;
556  blocks <<= 1;
557  N_B >>= 1;
558  time_divide++;
559  tf_change++;
560  }
561  B0 = blocks;
562  N_B0 = N_B;
563 
564  /* Reorganize the samples in time order instead of frequency order */
565  if (B0 > 1 && (quant || lowband))
567  N_B >> recombine, B0 << recombine,
568  longblocks);
569  }
570 
571  /* If we need 1.5 more bit than we can produce, split the band in two. */
572  cache = ff_celt_cache_bits +
574  if (!stereo && duration >= 0 && b > cache[cache[0]] + 12 && N > 2) {
575  N >>= 1;
576  Y = X + N;
577  split = 1;
578  duration -= 1;
579  if (blocks == 1)
580  fill = (fill & 1) | (fill << 1);
581  blocks = (blocks + 1) >> 1;
582  }
583 
584  if (split) {
585  int qn;
586  int itheta = quant ? celt_calc_theta(X, Y, stereo, N) : 0;
587  int mbits, sbits, delta;
588  int qalloc;
589  int pulse_cap;
590  int offset;
591  int orig_fill;
592  int tell;
593 
594  /* Decide on the resolution to give to the split parameter theta */
595  pulse_cap = ff_celt_log_freq_range[band] + duration * 8;
596  offset = (pulse_cap >> 1) - (stereo && N == 2 ? CELT_QTHETA_OFFSET_TWOPHASE :
598  qn = (stereo && band >= f->intensity_stereo) ? 1 :
599  celt_compute_qn(N, b, offset, pulse_cap, stereo);
600  tell = opus_rc_tell_frac(rc);
601  if (qn != 1) {
602  if (quant)
603  itheta = (itheta*qn + 8192) >> 14;
604  /* Entropy coding of the angle. We use a uniform pdf for the
605  * time split, a step for stereo, and a triangular one for the rest. */
606  if (quant) {
607  if (stereo && N > 2)
608  ff_opus_rc_enc_uint_step(rc, itheta, qn / 2);
609  else if (stereo || B0 > 1)
610  ff_opus_rc_enc_uint(rc, itheta, qn + 1);
611  else
612  ff_opus_rc_enc_uint_tri(rc, itheta, qn);
613  itheta = itheta * 16384 / qn;
614  if (stereo) {
615  if (itheta == 0)
616  celt_stereo_is_decouple(X, Y, f->block[0].lin_energy[band],
617  f->block[1].lin_energy[band], N);
618  else
620  }
621  } else {
622  if (stereo && N > 2)
623  itheta = ff_opus_rc_dec_uint_step(rc, qn / 2);
624  else if (stereo || B0 > 1)
625  itheta = ff_opus_rc_dec_uint(rc, qn+1);
626  else
627  itheta = ff_opus_rc_dec_uint_tri(rc, qn);
628  itheta = itheta * 16384 / qn;
629  }
630  } else if (stereo) {
631  if (quant) {
632  inv = f->apply_phase_inv ? itheta > 8192 : 0;
633  if (inv) {
634  for (i = 0; i < N; i++)
635  Y[i] *= -1;
636  }
637  celt_stereo_is_decouple(X, Y, f->block[0].lin_energy[band],
638  f->block[1].lin_energy[band], N);
639 
640  if (b > 2 << 3 && f->remaining2 > 2 << 3) {
641  ff_opus_rc_enc_log(rc, inv, 2);
642  } else {
643  inv = 0;
644  }
645  } else {
646  inv = (b > 2 << 3 && f->remaining2 > 2 << 3) ? ff_opus_rc_dec_log(rc, 2) : 0;
647  inv = f->apply_phase_inv ? inv : 0;
648  }
649  itheta = 0;
650  }
651  qalloc = opus_rc_tell_frac(rc) - tell;
652  b -= qalloc;
653 
654  orig_fill = fill;
655  if (itheta == 0) {
656  imid = 32767;
657  iside = 0;
658  fill = av_mod_uintp2(fill, blocks);
659  delta = -16384;
660  } else if (itheta == 16384) {
661  imid = 0;
662  iside = 32767;
663  fill &= ((1 << blocks) - 1) << blocks;
664  delta = 16384;
665  } else {
666  imid = celt_cos(itheta);
667  iside = celt_cos(16384-itheta);
668  /* This is the mid vs side allocation that minimizes squared error
669  in that band. */
670  delta = ROUND_MUL16((N - 1) << 7, celt_log2tan(iside, imid));
671  }
672 
673  mid = imid / 32768.0f;
674  side = iside / 32768.0f;
675 
676  /* This is a special case for N=2 that only works for stereo and takes
677  advantage of the fact that mid and side are orthogonal to encode
678  the side with just one bit. */
679  if (N == 2 && stereo) {
680  int c;
681  int sign = 0;
682  float tmp;
683  float *x2, *y2;
684  mbits = b;
685  /* Only need one bit for the side */
686  sbits = (itheta != 0 && itheta != 16384) ? 1 << 3 : 0;
687  mbits -= sbits;
688  c = (itheta > 8192);
689  f->remaining2 -= qalloc+sbits;
690 
691  x2 = c ? Y : X;
692  y2 = c ? X : Y;
693  if (sbits) {
694  if (quant) {
695  sign = x2[0]*y2[1] - x2[1]*y2[0] < 0;
696  ff_opus_rc_put_raw(rc, sign, 1);
697  } else {
698  sign = ff_opus_rc_get_raw(rc, 1);
699  }
700  }
701  sign = 1 - 2 * sign;
702  /* We use orig_fill here because we want to fold the side, but if
703  itheta==16384, we'll have cleared the low bits of fill. */
704  cm = pvq->quant_band(pvq, f, rc, band, x2, NULL, N, mbits, blocks, lowband, duration,
705  lowband_out, level, gain, lowband_scratch, orig_fill);
706  /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse),
707  and there's no need to worry about mixing with the other channel. */
708  y2[0] = -sign * x2[1];
709  y2[1] = sign * x2[0];
710  X[0] *= mid;
711  X[1] *= mid;
712  Y[0] *= side;
713  Y[1] *= side;
714  tmp = X[0];
715  X[0] = tmp - Y[0];
716  Y[0] = tmp + Y[0];
717  tmp = X[1];
718  X[1] = tmp - Y[1];
719  Y[1] = tmp + Y[1];
720  } else {
721  /* "Normal" split code */
722  float *next_lowband2 = NULL;
723  float *next_lowband_out1 = NULL;
724  int next_level = 0;
725  int rebalance;
726  uint32_t cmt;
727 
728  /* Give more bits to low-energy MDCTs than they would
729  * otherwise deserve */
730  if (B0 > 1 && !stereo && (itheta & 0x3fff)) {
731  if (itheta > 8192)
732  /* Rough approximation for pre-echo masking */
733  delta -= delta >> (4 - duration);
734  else
735  /* Corresponds to a forward-masking slope of
736  * 1.5 dB per 10 ms */
737  delta = FFMIN(0, delta + (N << 3 >> (5 - duration)));
738  }
739  mbits = av_clip((b - delta) / 2, 0, b);
740  sbits = b - mbits;
741  f->remaining2 -= qalloc;
742 
743  if (lowband && !stereo)
744  next_lowband2 = lowband + N; /* >32-bit split case */
745 
746  /* Only stereo needs to pass on lowband_out.
747  * Otherwise, it's handled at the end */
748  if (stereo)
749  next_lowband_out1 = lowband_out;
750  else
751  next_level = level + 1;
752 
753  rebalance = f->remaining2;
754  if (mbits >= sbits) {
755  /* In stereo mode, we do not apply a scaling to the mid
756  * because we need the normalized mid for folding later */
757  cm = pvq->quant_band(pvq, f, rc, band, X, NULL, N, mbits, blocks,
758  lowband, duration, next_lowband_out1, next_level,
759  stereo ? 1.0f : (gain * mid), lowband_scratch, fill);
760  rebalance = mbits - (rebalance - f->remaining2);
761  if (rebalance > 3 << 3 && itheta != 0)
762  sbits += rebalance - (3 << 3);
763 
764  /* For a stereo split, the high bits of fill are always zero,
765  * so no folding will be done to the side. */
766  cmt = pvq->quant_band(pvq, f, rc, band, Y, NULL, N, sbits, blocks,
767  next_lowband2, duration, NULL, next_level,
768  gain * side, NULL, fill >> blocks);
769  cm |= cmt << ((B0 >> 1) & (stereo - 1));
770  } else {
771  /* For a stereo split, the high bits of fill are always zero,
772  * so no folding will be done to the side. */
773  cm = pvq->quant_band(pvq, f, rc, band, Y, NULL, N, sbits, blocks,
774  next_lowband2, duration, NULL, next_level,
775  gain * side, NULL, fill >> blocks);
776  cm <<= ((B0 >> 1) & (stereo - 1));
777  rebalance = sbits - (rebalance - f->remaining2);
778  if (rebalance > 3 << 3 && itheta != 16384)
779  mbits += rebalance - (3 << 3);
780 
781  /* In stereo mode, we do not apply a scaling to the mid because
782  * we need the normalized mid for folding later */
783  cm |= pvq->quant_band(pvq, f, rc, band, X, NULL, N, mbits, blocks,
784  lowband, duration, next_lowband_out1, next_level,
785  stereo ? 1.0f : (gain * mid), lowband_scratch, fill);
786  }
787  }
788  } else {
789  /* This is the basic no-split case */
790  uint32_t q = celt_bits2pulses(cache, b);
791  uint32_t curr_bits = celt_pulses2bits(cache, q);
792  f->remaining2 -= curr_bits;
793 
794  /* Ensures we can never bust the budget */
795  while (f->remaining2 < 0 && q > 0) {
796  f->remaining2 += curr_bits;
797  curr_bits = celt_pulses2bits(cache, --q);
798  f->remaining2 -= curr_bits;
799  }
800 
801  if (q != 0) {
802  /* Finally do the actual (de)quantization */
803  if (quant) {
804  cm = celt_alg_quant(rc, X, N, (q < 8) ? q : (8 + (q & 7)) << ((q >> 3) - 1),
805  f->spread, blocks, gain, pvq);
806  } else {
807  cm = celt_alg_unquant(rc, X, N, (q < 8) ? q : (8 + (q & 7)) << ((q >> 3) - 1),
808  f->spread, blocks, gain, pvq);
809  }
810  } else {
811  /* If there's no pulse, fill the band anyway */
812  uint32_t cm_mask = (1 << blocks) - 1;
813  fill &= cm_mask;
814  if (fill) {
815  if (!lowband) {
816  /* Noise */
817  for (i = 0; i < N; i++)
818  X[i] = (((int32_t)celt_rng(f)) >> 20);
819  cm = cm_mask;
820  } else {
821  /* Folded spectrum */
822  for (i = 0; i < N; i++) {
823  /* About 48 dB below the "normal" folding level */
824  X[i] = lowband[i] + (((celt_rng(f)) & 0x8000) ? 1.0f / 256 : -1.0f / 256);
825  }
826  cm = fill;
827  }
828  celt_renormalize_vector(X, N, gain);
829  } else {
830  memset(X, 0, N*sizeof(float));
831  }
832  }
833  }
834 
835  /* This code is used by the decoder and by the resynthesis-enabled encoder */
836  if (stereo) {
837  if (N > 2)
838  celt_stereo_merge(X, Y, mid, N);
839  if (inv) {
840  for (i = 0; i < N; i++)
841  Y[i] *= -1;
842  }
843  } else if (level == 0) {
844  int k;
845 
846  /* Undo the sample reorganization going from time order to frequency order */
847  if (B0 > 1)
848  celt_interleave_hadamard(pvq->hadamard_tmp, X, N_B >> recombine,
849  B0 << recombine, longblocks);
850 
851  /* Undo time-freq changes that we did earlier */
852  N_B = N_B0;
853  blocks = B0;
854  for (k = 0; k < time_divide; k++) {
855  blocks >>= 1;
856  N_B <<= 1;
857  cm |= cm >> blocks;
858  celt_haar1(X, N_B, blocks);
859  }
860 
861  for (k = 0; k < recombine; k++) {
863  celt_haar1(X, N0>>k, 1<<k);
864  }
865  blocks <<= recombine;
866 
867  /* Scale output for later folding */
868  if (lowband_out) {
869  float n = sqrtf(N0);
870  for (i = 0; i < N0; i++)
871  lowband_out[i] = n * X[i];
872  }
873  cm = av_mod_uintp2(cm, blocks);
874  }
875 
876  return cm;
877 }
878 
879 static QUANT_FN(pvq_decode_band)
880 {
881 #if CONFIG_OPUS_DECODER
882  return quant_band_template(pvq, f, rc, band, X, Y, N, b, blocks, lowband, duration,
883  lowband_out, level, gain, lowband_scratch, fill, 0);
884 #else
885  return 0;
886 #endif
887 }
888 
889 static QUANT_FN(pvq_encode_band)
890 {
891 #if CONFIG_OPUS_ENCODER
892  return quant_band_template(pvq, f, rc, band, X, Y, N, b, blocks, lowband, duration,
893  lowband_out, level, gain, lowband_scratch, fill, 1);
894 #else
895  return 0;
896 #endif
897 }
898 
900 {
901  CeltPVQ *s = av_malloc(sizeof(CeltPVQ));
902  if (!s)
903  return AVERROR(ENOMEM);
904 
905  s->pvq_search = ppp_pvq_search_c;
906  s->quant_band = encode ? pvq_encode_band : pvq_decode_band;
907 
908 #if CONFIG_OPUS_ENCODER && ARCH_X86
910 #endif
911 
912  *pvq = s;
913 
914  return 0;
915 }
916 
918 {
919  av_freep(pvq);
920 }
ff_celt_cache_bits
const uint8_t ff_celt_cache_bits[392]
Definition: opustab.c:890
ff_celt_bit_deinterleave
const uint8_t ff_celt_bit_deinterleave[]
Definition: opustab.c:938
celt_normalize_residual
static void celt_normalize_residual(const int *av_restrict iy, float *av_restrict X, int N, float g)
Definition: opus_pvq.c:78
level
uint8_t level
Definition: svq3.c:206
av_clip
#define av_clip
Definition: common.h:95
AVERROR
Filter the word “frame” indicates either a video frame or a group of audio as stored in an AVFrame structure Format for each input and each output the list of supported formats For video that means pixel format For audio that means channel sample they are references to shared objects When the negotiation mechanism computes the intersection of the formats supported at each end of a all references to both lists are replaced with a reference to the intersection And when a single format is eventually chosen for a link amongst the remaining all references to the list are updated That means that if a filter requires that its input and output have the same format amongst a supported all it has to do is use a reference to the same list of formats query_formats can leave some formats unset and return AVERROR(EAGAIN) to cause the negotiation mechanism toagain later. That can be used by filters with complex requirements to use the format negotiated on one link to set the formats supported on another. Frame references ownership and permissions
ff_celt_qn_exp2
const uint16_t ff_celt_qn_exp2[]
Definition: opustab.c:951
celt_alg_unquant
static uint32_t celt_alg_unquant(OpusRangeCoder *rc, float *X, uint32_t N, uint32_t K, enum CeltSpread spread, uint32_t blocks, float gain, CeltPVQ *pvq)
Decode pulse vector and combine the result with the pitch vector to produce the final normalised sign...
Definition: opus_pvq.c:433
ppp_pvq_search_c
static float ppp_pvq_search_c(float *X, int *y, int K, int N)
Definition: opus_pvq.c:368
opus_ilog
#define opus_ilog(i)
Definition: opus_rc.h:31
ff_opus_rc_get_raw
uint32_t ff_opus_rc_get_raw(OpusRangeCoder *rc, uint32_t count)
CELT: read 1-25 raw bits at the end of the frame, backwards byte-wise.
Definition: opus_rc.c:140
quant_band_template
static av_always_inline uint32_t quant_band_template(CeltPVQ *pvq, CeltFrame *f, OpusRangeCoder *rc, const int band, float *X, float *Y, int N, int b, uint32_t blocks, float *lowband, int duration, float *lowband_out, int level, float gain, float *lowband_scratch, int fill, int quant)
Definition: opus_pvq.c:483
celt_cwrsi
static uint64_t celt_cwrsi(uint32_t N, uint32_t K, uint32_t i, int *y)
Definition: opus_pvq.c:270
ff_opus_rc_enc_uint
void ff_opus_rc_enc_uint(OpusRangeCoder *rc, uint32_t val, uint32_t size)
CELT: write a uniformly distributed integer.
Definition: opus_rc.c:204
atan2f
#define atan2f(y, x)
Definition: libm.h:45
av_mod_uintp2
#define av_mod_uintp2
Definition: common.h:122
tmp
static uint8_t tmp[11]
Definition: aes_ctr.c:28
celt_haar1
static void celt_haar1(float *X, int N0, int stride)
Definition: opus_pvq.c:227
B0
#define B0
Definition: faandct.c:40
CELT_SPREAD_NONE
@ CELT_SPREAD_NONE
Definition: opus_celt.h:51
b
#define b
Definition: input.c:34
ff_celt_log_freq_range
const uint8_t ff_celt_log_freq_range[]
Definition: opustab.c:776
celt_stereo_is_decouple
static void celt_stereo_is_decouple(float *X, float *Y, float e_l, float e_r, int N)
Definition: opus_pvq.c:463
celt_exp_rotation_impl
static void celt_exp_rotation_impl(float *X, uint32_t len, uint32_t stride, float c, float s)
Definition: opus_pvq.c:86
ff_celt_hadamard_order
const uint8_t ff_celt_hadamard_order[]
Definition: opustab.c:943
celt_calc_theta
static int celt_calc_theta(const float *X, const float *Y, int coupling, int N)
Definition: opus_pvq.c:445
celt_decode_pulses
static float celt_decode_pulses(OpusRangeCoder *rc, int *y, uint32_t N, uint32_t K)
Definition: opus_pvq.c:358
N2
#define N2
Definition: vf_pp7.c:70
QUANT_FN
static QUANT_FN(pvq_decode_band)
Definition: opus_pvq.c:879
ff_celt_pvq_init
int av_cold ff_celt_pvq_init(CeltPVQ **pvq, int encode)
Definition: opus_pvq.c:899
ff_celt_pvq_u_row
const uint32_t *const ff_celt_pvq_u_row[15]
Definition: opustab.c:1157
av_malloc
#define av_malloc(s)
Definition: tableprint_vlc.h:30
celt_exp_rotation
static void celt_exp_rotation(float *X, uint32_t len, uint32_t stride, uint32_t K, enum CeltSpread spread, const int encode)
Definition: opus_pvq.c:109
CeltPVQ
Definition: opus_pvq.h:37
celt_log2tan
static int celt_log2tan(int isin, int icos)
Definition: opus_pvq.c:41
cosf
#define cosf(x)
Definition: libm.h:78
FFSIGN
#define FFSIGN(a)
Definition: common.h:65
celt_pulses2bits
static int celt_pulses2bits(const uint8_t *cache, int pulses)
Definition: opus_pvq.c:72
val
static double val(void *priv, double ch)
Definition: aeval.c:77
celt_bits2pulses
static int celt_bits2pulses(const uint8_t *cache, int bits)
Definition: opus_pvq.c:53
quant
static int quant(float coef, const float Q, const float rounding)
Quantize one coefficient.
Definition: aacenc_utils.h:59
celt_renormalize_vector
static av_always_inline void celt_renormalize_vector(float *X, int N, float gain)
Definition: opus_celt.h:150
MUL16
#define MUL16(ra, rb)
Definition: mathops.h:88
av_cold
#define av_cold
Definition: attributes.h:90
CeltPVQ::quant_band
QUANT_FN * quant_band
Definition: opus_pvq.h:42
duration
int64_t duration
Definition: movenc.c:64
float
float
Definition: af_crystalizer.c:122
CELT_PVQ_U
#define CELT_PVQ_U(n, k)
Definition: opus_pvq.c:31
ff_opus_rc_dec_uint
uint32_t ff_opus_rc_dec_uint(OpusRangeCoder *rc, uint32_t size)
CELT: read a uniform distribution.
Definition: opus_rc.c:182
s
#define s(width, name)
Definition: cbs_vp9.c:256
celt_alg_quant
static uint32_t celt_alg_quant(OpusRangeCoder *rc, float *X, uint32_t N, uint32_t K, enum CeltSpread spread, uint32_t blocks, float gain, CeltPVQ *pvq)
Definition: opus_pvq.c:417
g
const char * g
Definition: vf_curves.c:117
bits
uint8_t bits
Definition: vp3data.h:141
CELT_QTHETA_OFFSET_TWOPHASE
#define CELT_QTHETA_OFFSET_TWOPHASE
Definition: opus_celt.h:44
CeltPVQ::hadamard_tmp
float hadamard_tmp[256]
Definition: opus_pvq.h:39
ff_opus_rc_enc_uint_tri
void ff_opus_rc_enc_uint_tri(OpusRangeCoder *rc, uint32_t k, int qn)
Definition: opus_rc.c:258
celt_icwrsi
static uint32_t celt_icwrsi(uint32_t N, uint32_t K, const int *y)
Definition: opus_pvq.c:258
E
#define E
Definition: avdct.c:32
FFABS
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
Definition: common.h:64
CeltPVQ::qcoeff
int qcoeff[256]
Definition: opus_pvq.h:38
ff_celt_cache_index
const int16_t ff_celt_cache_index[105]
Definition: opustab.c:920
NULL
#define NULL
Definition: coverity.c:32
ff_celt_pvq_uninit
void av_cold ff_celt_pvq_uninit(CeltPVQ **pvq)
Definition: opus_pvq.c:917
CELT_MAX_BANDS
#define CELT_MAX_BANDS
Definition: opus.h:46
CELT_QTHETA_OFFSET
#define CELT_QTHETA_OFFSET
Definition: opus_celt.h:43
sqrtf
static __device__ float sqrtf(float a)
Definition: cuda_runtime.h:184
sinf
#define sinf(x)
Definition: libm.h:419
ff_opus_rc_dec_uint_tri
uint32_t ff_opus_rc_dec_uint_tri(OpusRangeCoder *rc, int qn)
Definition: opus_rc.c:234
celt_deinterleave_hadamard
static void celt_deinterleave_hadamard(float *tmp, float *X, int N0, int stride, int hadamard)
Definition: opus_pvq.c:214
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
opustab.h
f
f
Definition: af_crystalizer.c:122
opus_pvq.h
N0
#define N0
Definition: vf_pp7.c:68
celt_cos
static int16_t celt_cos(int16_t x)
Definition: opus_pvq.c:34
OpusRangeCoder
Definition: opus_rc.h:40
split
static char * split(char *message, char delim)
Definition: af_channelmap.c:81
celt_rng
static av_always_inline uint32_t celt_rng(CeltFrame *f)
Definition: opus_celt.h:144
encode
static void encode(AVCodecContext *ctx, AVFrame *frame, AVPacket *pkt, FILE *output)
Definition: encode_audio.c:94
offset
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf offset
Definition: writing_filters.txt:86
N
#define N
Definition: af_mcompand.c:53
ROUND_MUL16
#define ROUND_MUL16(a, b)
Definition: opus.h:52
M_PI
#define M_PI
Definition: mathematics.h:52
Y
#define Y
Definition: boxblur.h:37
ff_opus_rc_enc_uint_step
void ff_opus_rc_enc_uint_step(OpusRangeCoder *rc, uint32_t val, int k0)
Definition: opus_rc.c:226
X
@ X
Definition: vf_addroi.c:26
lrintf
#define lrintf(x)
Definition: libm_mips.h:72
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:269
FFMIN3
#define FFMIN3(a, b, c)
Definition: macros.h:50
ff_opus_rc_put_raw
void ff_opus_rc_put_raw(OpusRangeCoder *rc, uint32_t val, uint32_t count)
CELT: write 0 - 31 bits to the rawbits buffer.
Definition: opus_rc.c:161
delta
float delta
Definition: vorbis_enc_data.h:430
av_always_inline
#define av_always_inline
Definition: attributes.h:49
value
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf default value
Definition: writing_filters.txt:86
celt_encode_pulses
static void celt_encode_pulses(OpusRangeCoder *rc, int *y, uint32_t N, uint32_t K)
Definition: opus_pvq.c:353
FFMIN
#define FFMIN(a, b)
Definition: macros.h:49
len
int len
Definition: vorbis_enc_data.h:426
opus_rc_tell_frac
static av_always_inline uint32_t opus_rc_tell_frac(const OpusRangeCoder *rc)
Definition: opus_rc.h:66
stride
#define stride
Definition: h264pred_template.c:537
ff_opus_rc_dec_uint_step
uint32_t ff_opus_rc_dec_uint_step(OpusRangeCoder *rc, int k0)
Definition: opus_rc.c:211
celt_stereo_merge
static void celt_stereo_merge(float *X, float *Y, float mid, int N)
Definition: opus_pvq.c:163
M_SQRT1_2
#define M_SQRT1_2
Definition: mathematics.h:58
B
#define B
Definition: huffyuvdsp.h:32
CeltSpread
CeltSpread
Definition: opus_celt.h:50
cm
#define cm
Definition: dvbsubdec.c:39
ff_opus_rc_enc_log
void ff_opus_rc_enc_log(OpusRangeCoder *rc, int val, uint32_t bits)
Definition: opus_rc.c:131
celt_compute_qn
static int celt_compute_qn(int N, int b, int offset, int pulse_cap, int stereo)
Definition: opus_pvq.c:241
ff_celt_pvq_init_x86
void ff_celt_pvq_init_x86(struct CeltPVQ *s)
Definition: celt_pvq_init.c:32
celt_interleave_hadamard
static void celt_interleave_hadamard(float *tmp, float *X, int N0, int stride, int hadamard)
Definition: opus_pvq.c:201
av_freep
#define av_freep(p)
Definition: tableprint_vlc.h:34
int32_t
int32_t
Definition: audioconvert.c:56
ff_celt_bit_interleave
const uint8_t ff_celt_bit_interleave[]
Definition: opustab.c:934
ff_opus_rc_dec_log
uint32_t ff_opus_rc_dec_log(OpusRangeCoder *rc, uint32_t bits)
Definition: opus_rc.c:114
CeltPVQ::pvq_search
float(* pvq_search)(float *X, int *y, int K, int N)
Definition: opus_pvq.h:41
pulses
static const int8_t pulses[4]
Number of non-zero pulses in the MP-MLQ excitation.
Definition: g723_1.h:260
celt_extract_collapse_mask
static uint32_t celt_extract_collapse_mask(const int *iy, uint32_t N, uint32_t B)
Definition: opus_pvq.c:149
celt_stereo_ms_decouple
static void celt_stereo_ms_decouple(float *X, float *Y, int N)
Definition: opus_pvq.c:473
CeltFrame
Definition: opus_celt.h:93
CELT_PVQ_V
#define CELT_PVQ_V(n, k)
Definition: opus_pvq.c:32