summaryrefslogtreecommitdiffstats
path: root/lib/rbcodec/codecs/libopus/celt/vq.c
blob: b047b22774aaa17d47aaba4afcf0e93c5662d220 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
/* Copyright (c) 2007-2008 CSIRO
   Copyright (c) 2007-2009 Xiph.Org Foundation
   Written by Jean-Marc Valin */
/*
   Redistribution and use in source and binary forms, with or without
   modification, are permitted provided that the following conditions
   are met:

   - Redistributions of source code must retain the above copyright
   notice, this list of conditions and the following disclaimer.

   - Redistributions in binary form must reproduce the above copyright
   notice, this list of conditions and the following disclaimer in the
   documentation and/or other materials provided with the distribution.

   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
   PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/

#ifdef HAVE_CONFIG_H
#include "config.h"
#endif

#include "mathops.h"
#include "cwrs.h"
#include "vq.h"
#include "arch.h"
#include "os_support.h"
#include "bands.h"
#include "rate.h"
#include "pitch.h"

#if defined(MIPSr1_ASM)
#include "mips/vq_mipsr1.h"
#endif

#ifndef OVERRIDE_vq_exp_rotation1
static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s)
{
   int i;
   opus_val16 ms;
   celt_norm *Xptr;
   Xptr = X;
   ms = NEG16(s);
   for (i=0;i<len-stride;i++)
   {
      celt_norm x1, x2;
      x1 = Xptr[0];
      x2 = Xptr[stride];
      Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2),  s, x1), 15));
      *Xptr++      = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15));
   }
   Xptr = &X[len-2*stride-1];
   for (i=len-2*stride-1;i>=0;i--)
   {
      celt_norm x1, x2;
      x1 = Xptr[0];
      x2 = Xptr[stride];
      Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2),  s, x1), 15));
      *Xptr--      = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15));
   }
}
#endif /* OVERRIDE_vq_exp_rotation1 */

static void exp_rotation(celt_norm *X, int len, int dir, int stride, int K, int spread)
{
   static const int SPREAD_FACTOR[3]={15,10,5};
   int i;
   opus_val16 c, s;
   opus_val16 gain, theta;
   int stride2=0;
   int factor;

   if (2*K>=len || spread==SPREAD_NONE)
      return;
   factor = SPREAD_FACTOR[spread-1];

   gain = celt_div((opus_val32)MULT16_16(Q15_ONE,len),(opus_val32)(len+factor*K));
   theta = HALF16(MULT16_16_Q15(gain,gain));

   c = celt_cos_norm(EXTEND32(theta));
   s = celt_cos_norm(EXTEND32(SUB16(Q15ONE,theta))); /*  sin(theta) */

   if (len>=8*stride)
   {
      stride2 = 1;
      /* This is just a simple (equivalent) way of computing sqrt(len/stride) with rounding.
         It's basically incrementing long as (stride2+0.5)^2 < len/stride. */
      while ((stride2*stride2+stride2)*stride + (stride>>2) < len)
         stride2++;
   }
   /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for
      extract_collapse_mask().*/
   len = celt_udiv(len, stride);
   for (i=0;i<stride;i++)
   {
      if (dir < 0)
      {
         if (stride2)
            exp_rotation1(X+i*len, len, stride2, s, c);
         exp_rotation1(X+i*len, len, 1, c, s);
      } else {
         exp_rotation1(X+i*len, len, 1, c, -s);
         if (stride2)
            exp_rotation1(X+i*len, len, stride2, s, -c);
      }
   }
}

/** Takes the pitch vector and the decoded residual vector, computes the gain
    that will give ||p+g*y||=1 and mixes the residual with the pitch. */
static void normalise_residual(int * OPUS_RESTRICT iy, celt_norm * OPUS_RESTRICT X,
      int N, opus_val32 Ryy, opus_val16 gain)
{
   int i;
#ifdef FIXED_POINT
   int k;
#endif
   opus_val32 t;
   opus_val16 g;

#ifdef FIXED_POINT
   k = celt_ilog2(Ryy)>>1;
#endif
   t = VSHR32(Ryy, 2*(k-7));
   g = MULT16_16_P15(celt_rsqrt_norm(t),gain);

   i=0;
   do
      X[i] = EXTRACT16(PSHR32(MULT16_16(g, iy[i]), k+1));
   while (++i < N);
}

static unsigned extract_collapse_mask(int *iy, int N, int B)
{
   unsigned collapse_mask;
   int N0;
   int i;
   if (B<=1)
      return 1;
   /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for
      exp_rotation().*/
   N0 = celt_udiv(N, B);
   collapse_mask = 0;
   i=0; do {
      int j;
      unsigned tmp=0;
      j=0; do {
         tmp |= iy[i*N0+j];
      } while (++j<N0);
      collapse_mask |= (tmp!=0)<<i;
   } while (++i<B);
   return collapse_mask;
}

unsigned alg_quant(celt_norm *X, int N, int K, int spread, int B, ec_enc *enc
#ifdef RESYNTH
   , opus_val16 gain
#endif
   )
{
   VARDECL(celt_norm, y);
   VARDECL(int, iy);
   VARDECL(opus_val16, signx);
   int i, j;
   opus_val16 s;
   int pulsesLeft;
   opus_val32 sum;
   opus_val32 xy;
   opus_val16 yy;
   unsigned collapse_mask;
   SAVE_STACK;

   celt_assert2(K>0, "alg_quant() needs at least one pulse");
   celt_assert2(N>1, "alg_quant() needs at least two dimensions");

   ALLOC(y, N, celt_norm);
   ALLOC(iy, N, int);
   ALLOC(signx, N, opus_val16);

   exp_rotation(X, N, 1, B, K, spread);

   /* Get rid of the sign */
   sum = 0;
   j=0; do {
      if (X[j]>0)
         signx[j]=1;
      else {
         signx[j]=-1;
         X[j]=-X[j];
      }
      iy[j] = 0;
      y[j] = 0;
   } while (++j<N);

   xy = yy = 0;

   pulsesLeft = K;

   /* Do a pre-search by projecting on the pyramid */
   if (K > (N>>1))
   {
      opus_val16 rcp;
      j=0; do {
         sum += X[j];
      }  while (++j<N);

      /* If X is too small, just replace it with a pulse at 0 */
#ifdef FIXED_POINT
      if (sum <= K)
#else
      /* Prevents infinities and NaNs from causing too many pulses
         to be allocated. 64 is an approximation of infinity here. */
      if (!(sum > EPSILON && sum < 64))
#endif
      {
         X[0] = QCONST16(1.f,14);
         j=1; do
            X[j]=0;
         while (++j<N);
         sum = QCONST16(1.f,14);
      }
      rcp = EXTRACT16(MULT16_32_Q16(K-1, celt_rcp(sum)));
      j=0; do {
#ifdef FIXED_POINT
         /* It's really important to round *towards zero* here */
         iy[j] = MULT16_16_Q15(X[j],rcp);
#else
         iy[j] = (int)floor(rcp*X[j]);
#endif
         y[j] = (celt_norm)iy[j];
         yy = MAC16_16(yy, y[j],y[j]);
         xy = MAC16_16(xy, X[j],y[j]);
         y[j] *= 2;
         pulsesLeft -= iy[j];
      }  while (++j<N);
   }
   celt_assert2(pulsesLeft>=1, "Allocated too many pulses in the quick pass");

   /* This should never happen, but just in case it does (e.g. on silence)
      we fill the first bin with pulses. */
#ifdef FIXED_POINT_DEBUG
   celt_assert2(pulsesLeft<=N+3, "Not enough pulses in the quick pass");
#endif
   if (pulsesLeft > N+3)
   {
      opus_val16 tmp = (opus_val16)pulsesLeft;
      yy = MAC16_16(yy, tmp, tmp);
      yy = MAC16_16(yy, tmp, y[0]);
      iy[0] += pulsesLeft;
      pulsesLeft=0;
   }

   s = 1;
   for (i=0;i<pulsesLeft;i++)
   {
      int best_id;
      opus_val32 best_num = -VERY_LARGE16;
      opus_val16 best_den = 0;
#ifdef FIXED_POINT
      int rshift;
#endif
#ifdef FIXED_POINT
      rshift = 1+celt_ilog2(K-pulsesLeft+i+1);
#endif
      best_id = 0;
      /* The squared magnitude term gets added anyway, so we might as well
         add it outside the loop */
      yy = ADD32(yy, 1);
      j=0;
      do {
         opus_val16 Rxy, Ryy;
         /* Temporary sums of the new pulse(s) */
         Rxy = EXTRACT16(SHR32(ADD32(xy, EXTEND32(X[j])),rshift));
         /* We're multiplying y[j] by two so we don't have to do it here */
         Ryy = ADD16(yy, y[j]);

         /* Approximate score: we maximise Rxy/sqrt(Ryy) (we're guaranteed that
            Rxy is positive because the sign is pre-computed) */
         Rxy = MULT16_16_Q15(Rxy,Rxy);
         /* The idea is to check for num/den >= best_num/best_den, but that way
            we can do it without any division */
         /* OPT: Make sure to use conditional moves here */
         if (MULT16_16(best_den, Rxy) > MULT16_16(Ryy, best_num))
         {
            best_den = Ryy;
            best_num = Rxy;
            best_id = j;
         }
      } while (++j<N);

      /* Updating the sums of the new pulse(s) */
      xy = ADD32(xy, EXTEND32(X[best_id]));
      /* We're multiplying y[j] by two so we don't have to do it here */
      yy = ADD16(yy, y[best_id]);

      /* Only now that we've made the final choice, update y/iy */
      /* Multiplying y[j] by 2 so we don't have to do it everywhere else */
      y[best_id] += 2*s;
      iy[best_id]++;
   }

   /* Put the original sign back */
   j=0;
   do {
      X[j] = MULT16_16(signx[j],X[j]);
      if (signx[j] < 0)
         iy[j] = -iy[j];
   } while (++j<N);
   encode_pulses(iy, N, K, enc);

#ifdef RESYNTH
   normalise_residual(iy, X, N, yy, gain);
   exp_rotation(X, N, -1, B, K, spread);
#endif

   collapse_mask = extract_collapse_mask(iy, N, B);
   RESTORE_STACK;
   return collapse_mask;
}

/** Decode pulse vector and combine the result with the pitch vector to produce
    the final normalised signal in the current band. */
unsigned alg_unquant(celt_norm *X, int N, int K, int spread, int B,
      ec_dec *dec, opus_val16 gain)
{
   opus_val32 Ryy;
   unsigned collapse_mask;
   VARDECL(int, iy);
   SAVE_STACK;

   celt_assert2(K>0, "alg_unquant() needs at least one pulse");
   celt_assert2(N>1, "alg_unquant() needs at least two dimensions");
   ALLOC(iy, N, int);
   Ryy = decode_pulses(iy, N, K, dec);
   normalise_residual(iy, X, N, Ryy, gain);
   exp_rotation(X, N, -1, B, K, spread);
   collapse_mask = extract_collapse_mask(iy, N, B);
   RESTORE_STACK;
   return collapse_mask;
}

#ifndef OVERRIDE_renormalise_vector
void renormalise_vector(celt_norm *X, int N, opus_val16 gain)
{
   int i;
#ifdef FIXED_POINT
   int k;
#endif
   opus_val32 E;
   opus_val16 g;
   opus_val32 t;
   celt_norm *xptr;
   E = EPSILON + celt_inner_prod(X, X, N);
#ifdef FIXED_POINT
   k = celt_ilog2(E)>>1;
#endif
   t = VSHR32(E, 2*(k-7));
   g = MULT16_16_P15(celt_rsqrt_norm(t),gain);

   xptr = X;
   for (i=0;i<N;i++)
   {
      *xptr = EXTRACT16(PSHR32(MULT16_16(g, *xptr), k+1));
      xptr++;
   }
   /*return celt_sqrt(E);*/
}
#endif /* OVERRIDE_renormalise_vector */

int stereo_itheta(const celt_norm *X, const celt_norm *Y, int stereo, int N)
{
   int i;
   int itheta;
   opus_val16 mid, side;
   opus_val32 Emid, Eside;

   Emid = Eside = EPSILON;
   if (stereo)
   {
      for (i=0;i<N;i++)
      {
         celt_norm m, s;
         m = ADD16(SHR16(X[i],1),SHR16(Y[i],1));
         s = SUB16(SHR16(X[i],1),SHR16(Y[i],1));
         Emid = MAC16_16(Emid, m, m);
         Eside = MAC16_16(Eside, s, s);
      }
   } else {
      Emid += celt_inner_prod(X, X, N);
      Eside += celt_inner_prod(Y, Y, N);
   }
   mid = celt_sqrt(Emid);
   side = celt_sqrt(Eside);
#ifdef FIXED_POINT
   /* 0.63662 = 2/pi */
   itheta = MULT16_16_Q15(QCONST16(0.63662f,15),celt_atan2p(side, mid));
#else
   itheta = (int)floor(.5f+16384*0.63662f*atan2(side,mid));
#endif

   return itheta;
}