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 "opustab.h"
27 #include "opus_pvq.h"
28
29 #define CELT_PVQ_U(n, k) (ff_celt_pvq_u_row[FFMIN(n, k)][FFMAX(n, k)])
30 #define CELT_PVQ_V(n, k) (CELT_PVQ_U(n, k) + CELT_PVQ_U(n, (k) + 1))
31
celt_cos(int16_t x)32 static inline int16_t celt_cos(int16_t x)
33 {
34 x = (MUL16(x, x) + 4096) >> 13;
35 x = (32767-x) + ROUND_MUL16(x, (-7651 + ROUND_MUL16(x, (8277 + ROUND_MUL16(-626, x)))));
36 return x + 1;
37 }
38
celt_log2tan(int isin,int icos)39 static inline int celt_log2tan(int isin, int icos)
40 {
41 int lc, ls;
42 lc = opus_ilog(icos);
43 ls = opus_ilog(isin);
44 icos <<= 15 - lc;
45 isin <<= 15 - ls;
46 return (ls << 11) - (lc << 11) +
47 ROUND_MUL16(isin, ROUND_MUL16(isin, -2597) + 7932) -
48 ROUND_MUL16(icos, ROUND_MUL16(icos, -2597) + 7932);
49 }
50
celt_bits2pulses(const uint8_t * cache,int bits)51 static inline int celt_bits2pulses(const uint8_t *cache, int bits)
52 {
53 // TODO: Find the size of cache and make it into an array in the parameters list
54 int i, low = 0, high;
55
56 high = cache[0];
57 bits--;
58
59 for (i = 0; i < 6; i++) {
60 int center = (low + high + 1) >> 1;
61 if (cache[center] >= bits)
62 high = center;
63 else
64 low = center;
65 }
66
67 return (bits - (low == 0 ? -1 : cache[low]) <= cache[high] - bits) ? low : high;
68 }
69
celt_pulses2bits(const uint8_t * cache,int pulses)70 static inline int celt_pulses2bits(const uint8_t *cache, int pulses)
71 {
72 // TODO: Find the size of cache and make it into an array in the parameters list
73 return (pulses == 0) ? 0 : cache[pulses] + 1;
74 }
75
celt_normalize_residual(const int * av_restrict iy,float * av_restrict X,int N,float g)76 static inline void celt_normalize_residual(const int * av_restrict iy, float * av_restrict X,
77 int N, float g)
78 {
79 int i;
80 for (i = 0; i < N; i++)
81 X[i] = g * iy[i];
82 }
83
celt_exp_rotation_impl(float * X,uint32_t len,uint32_t stride,float c,float s)84 static void celt_exp_rotation_impl(float *X, uint32_t len, uint32_t stride,
85 float c, float s)
86 {
87 float *Xptr;
88 int i;
89
90 Xptr = X;
91 for (i = 0; i < len - stride; i++) {
92 float x1 = Xptr[0];
93 float x2 = Xptr[stride];
94 Xptr[stride] = c * x2 + s * x1;
95 *Xptr++ = c * x1 - s * x2;
96 }
97
98 Xptr = &X[len - 2 * stride - 1];
99 for (i = len - 2 * stride - 1; i >= 0; i--) {
100 float x1 = Xptr[0];
101 float x2 = Xptr[stride];
102 Xptr[stride] = c * x2 + s * x1;
103 *Xptr-- = c * x1 - s * x2;
104 }
105 }
106
celt_exp_rotation(float * X,uint32_t len,uint32_t stride,uint32_t K,enum CeltSpread spread,const int encode)107 static inline void celt_exp_rotation(float *X, uint32_t len,
108 uint32_t stride, uint32_t K,
109 enum CeltSpread spread, const int encode)
110 {
111 uint32_t stride2 = 0;
112 float c, s;
113 float gain, theta;
114 int i;
115
116 if (2*K >= len || spread == CELT_SPREAD_NONE)
117 return;
118
119 gain = (float)len / (len + (20 - 5*spread) * K);
120 theta = M_PI * gain * gain / 4;
121
122 c = cosf(theta);
123 s = sinf(theta);
124
125 if (len >= stride << 3) {
126 stride2 = 1;
127 /* This is just a simple (equivalent) way of computing sqrt(len/stride) with rounding.
128 It's basically incrementing long as (stride2+0.5)^2 < len/stride. */
129 while ((stride2 * stride2 + stride2) * stride + (stride >> 2) < len)
130 stride2++;
131 }
132
133 len /= stride;
134 for (i = 0; i < stride; i++) {
135 if (encode) {
136 celt_exp_rotation_impl(X + i * len, len, 1, c, -s);
137 if (stride2)
138 celt_exp_rotation_impl(X + i * len, len, stride2, s, -c);
139 } else {
140 if (stride2)
141 celt_exp_rotation_impl(X + i * len, len, stride2, s, c);
142 celt_exp_rotation_impl(X + i * len, len, 1, c, s);
143 }
144 }
145 }
146
celt_extract_collapse_mask(const int * iy,uint32_t N,uint32_t B)147 static inline uint32_t celt_extract_collapse_mask(const int *iy, uint32_t N, uint32_t B)
148 {
149 int i, j, N0 = N / B;
150 uint32_t collapse_mask = 0;
151
152 if (B <= 1)
153 return 1;
154
155 for (i = 0; i < B; i++)
156 for (j = 0; j < N0; j++)
157 collapse_mask |= (!!iy[i*N0+j]) << i;
158 return collapse_mask;
159 }
160
celt_stereo_merge(float * X,float * Y,float mid,int N)161 static inline void celt_stereo_merge(float *X, float *Y, float mid, int N)
162 {
163 int i;
164 float xp = 0, side = 0;
165 float E[2];
166 float mid2;
167 float gain[2];
168
169 /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */
170 for (i = 0; i < N; i++) {
171 xp += X[i] * Y[i];
172 side += Y[i] * Y[i];
173 }
174
175 /* Compensating for the mid normalization */
176 xp *= mid;
177 mid2 = mid;
178 E[0] = mid2 * mid2 + side - 2 * xp;
179 E[1] = mid2 * mid2 + side + 2 * xp;
180 if (E[0] < 6e-4f || E[1] < 6e-4f) {
181 for (i = 0; i < N; i++)
182 Y[i] = X[i];
183 return;
184 }
185
186 gain[0] = 1.0f / sqrtf(E[0]);
187 gain[1] = 1.0f / sqrtf(E[1]);
188
189 for (i = 0; i < N; i++) {
190 float value[2];
191 /* Apply mid scaling (side is already scaled) */
192 value[0] = mid * X[i];
193 value[1] = Y[i];
194 X[i] = gain[0] * (value[0] - value[1]);
195 Y[i] = gain[1] * (value[0] + value[1]);
196 }
197 }
198
celt_interleave_hadamard(float * tmp,float * X,int N0,int stride,int hadamard)199 static void celt_interleave_hadamard(float *tmp, float *X, int N0,
200 int stride, int hadamard)
201 {
202 int i, j, N = N0*stride;
203 const uint8_t *order = &ff_celt_hadamard_order[hadamard ? stride - 2 : 30];
204
205 for (i = 0; i < stride; i++)
206 for (j = 0; j < N0; j++)
207 tmp[j*stride+i] = X[order[i]*N0+j];
208
209 memcpy(X, tmp, N*sizeof(float));
210 }
211
celt_deinterleave_hadamard(float * tmp,float * X,int N0,int stride,int hadamard)212 static void celt_deinterleave_hadamard(float *tmp, float *X, int N0,
213 int stride, int hadamard)
214 {
215 int i, j, N = N0*stride;
216 const uint8_t *order = &ff_celt_hadamard_order[hadamard ? stride - 2 : 30];
217
218 for (i = 0; i < stride; i++)
219 for (j = 0; j < N0; j++)
220 tmp[order[i]*N0+j] = X[j*stride+i];
221
222 memcpy(X, tmp, N*sizeof(float));
223 }
224
celt_haar1(float * X,int N0,int stride)225 static void celt_haar1(float *X, int N0, int stride)
226 {
227 int i, j;
228 N0 >>= 1;
229 for (i = 0; i < stride; i++) {
230 for (j = 0; j < N0; j++) {
231 float x0 = X[stride * (2 * j + 0) + i];
232 float x1 = X[stride * (2 * j + 1) + i];
233 X[stride * (2 * j + 0) + i] = (x0 + x1) * M_SQRT1_2;
234 X[stride * (2 * j + 1) + i] = (x0 - x1) * M_SQRT1_2;
235 }
236 }
237 }
238
celt_compute_qn(int N,int b,int offset,int pulse_cap,int stereo)239 static inline int celt_compute_qn(int N, int b, int offset, int pulse_cap,
240 int stereo)
241 {
242 int qn, qb;
243 int N2 = 2 * N - 1;
244 if (stereo && N == 2)
245 N2--;
246
247 /* The upper limit ensures that in a stereo split with itheta==16384, we'll
248 * always have enough bits left over to code at least one pulse in the
249 * side; otherwise it would collapse, since it doesn't get folded. */
250 qb = FFMIN3(b - pulse_cap - (4 << 3), (b + N2 * offset) / N2, 8 << 3);
251 qn = (qb < (1 << 3 >> 1)) ? 1 : ((ff_celt_qn_exp2[qb & 0x7] >> (14 - (qb >> 3))) + 1) >> 1 << 1;
252 return qn;
253 }
254
255 /* Convert the quantized vector to an index */
celt_icwrsi(uint32_t N,uint32_t K,const int * y)256 static inline uint32_t celt_icwrsi(uint32_t N, uint32_t K, const int *y)
257 {
258 int i, idx = 0, sum = 0;
259 for (i = N - 1; i >= 0; i--) {
260 const uint32_t i_s = CELT_PVQ_U(N - i, sum + FFABS(y[i]) + 1);
261 idx += CELT_PVQ_U(N - i, sum) + (y[i] < 0)*i_s;
262 sum += FFABS(y[i]);
263 }
264 return idx;
265 }
266
267 // this code was adapted from libopus
celt_cwrsi(uint32_t N,uint32_t K,uint32_t i,int * y)268 static inline uint64_t celt_cwrsi(uint32_t N, uint32_t K, uint32_t i, int *y)
269 {
270 uint64_t norm = 0;
271 uint32_t q, p;
272 int s, val;
273 int k0;
274
275 while (N > 2) {
276 /*Lots of pulses case:*/
277 if (K >= N) {
278 const uint32_t *row = ff_celt_pvq_u_row[N];
279
280 /* Are the pulses in this dimension negative? */
281 p = row[K + 1];
282 s = -(i >= p);
283 i -= p & s;
284
285 /*Count how many pulses were placed in this dimension.*/
286 k0 = K;
287 q = row[N];
288 if (q > i) {
289 K = N;
290 do {
291 p = ff_celt_pvq_u_row[--K][N];
292 } while (p > i);
293 } else
294 for (p = row[K]; p > i; p = row[K])
295 K--;
296
297 i -= p;
298 val = (k0 - K + s) ^ s;
299 norm += val * val;
300 *y++ = val;
301 } else { /*Lots of dimensions case:*/
302 /*Are there any pulses in this dimension at all?*/
303 p = ff_celt_pvq_u_row[K ][N];
304 q = ff_celt_pvq_u_row[K + 1][N];
305
306 if (p <= i && i < q) {
307 i -= p;
308 *y++ = 0;
309 } else {
310 /*Are the pulses in this dimension negative?*/
311 s = -(i >= q);
312 i -= q & s;
313
314 /*Count how many pulses were placed in this dimension.*/
315 k0 = K;
316 do p = ff_celt_pvq_u_row[--K][N];
317 while (p > i);
318
319 i -= p;
320 val = (k0 - K + s) ^ s;
321 norm += val * val;
322 *y++ = val;
323 }
324 }
325 N--;
326 }
327
328 /* N == 2 */
329 p = 2 * K + 1;
330 s = -(i >= p);
331 i -= p & s;
332 k0 = K;
333 K = (i + 1) / 2;
334
335 if (K)
336 i -= 2 * K - 1;
337
338 val = (k0 - K + s) ^ s;
339 norm += val * val;
340 *y++ = val;
341
342 /* N==1 */
343 s = -i;
344 val = (K + s) ^ s;
345 norm += val * val;
346 *y = val;
347
348 return norm;
349 }
350
celt_encode_pulses(OpusRangeCoder * rc,int * y,uint32_t N,uint32_t K)351 static inline void celt_encode_pulses(OpusRangeCoder *rc, int *y, uint32_t N, uint32_t K)
352 {
353 ff_opus_rc_enc_uint(rc, celt_icwrsi(N, K, y), CELT_PVQ_V(N, K));
354 }
355
celt_decode_pulses(OpusRangeCoder * rc,int * y,uint32_t N,uint32_t K)356 static inline float celt_decode_pulses(OpusRangeCoder *rc, int *y, uint32_t N, uint32_t K)
357 {
358 const uint32_t idx = ff_opus_rc_dec_uint(rc, CELT_PVQ_V(N, K));
359 return celt_cwrsi(N, K, idx, y);
360 }
361
362 /*
363 * Faster than libopus's search, operates entirely in the signed domain.
364 * Slightly worse/better depending on N, K and the input vector.
365 */
ppp_pvq_search_c(float * X,int * y,int K,int N)366 static float ppp_pvq_search_c(float *X, int *y, int K, int N)
367 {
368 int i, y_norm = 0;
369 float res = 0.0f, xy_norm = 0.0f;
370
371 for (i = 0; i < N; i++)
372 res += FFABS(X[i]);
373
374 res = K/(res + FLT_EPSILON);
375
376 for (i = 0; i < N; i++) {
377 y[i] = lrintf(res*X[i]);
378 y_norm += y[i]*y[i];
379 xy_norm += y[i]*X[i];
380 K -= FFABS(y[i]);
381 }
382
383 while (K) {
384 int max_idx = 0, phase = FFSIGN(K);
385 float max_num = 0.0f;
386 float max_den = 1.0f;
387 y_norm += 1.0f;
388
389 for (i = 0; i < N; i++) {
390 /* If the sum has been overshot and the best place has 0 pulses allocated
391 * to it, attempting to decrease it further will actually increase the
392 * sum. Prevent this by disregarding any 0 positions when decrementing. */
393 const int ca = 1 ^ ((y[i] == 0) & (phase < 0));
394 const int y_new = y_norm + 2*phase*FFABS(y[i]);
395 float xy_new = xy_norm + 1*phase*FFABS(X[i]);
396 xy_new = xy_new * xy_new;
397 if (ca && (max_den*xy_new) > (y_new*max_num)) {
398 max_den = y_new;
399 max_num = xy_new;
400 max_idx = i;
401 }
402 }
403
404 K -= phase;
405
406 phase *= FFSIGN(X[max_idx]);
407 xy_norm += 1*phase*X[max_idx];
408 y_norm += 2*phase*y[max_idx];
409 y[max_idx] += phase;
410 }
411
412 return (float)y_norm;
413 }
414
celt_alg_quant(OpusRangeCoder * rc,float * X,uint32_t N,uint32_t K,enum CeltSpread spread,uint32_t blocks,float gain,CeltPVQ * pvq)415 static uint32_t celt_alg_quant(OpusRangeCoder *rc, float *X, uint32_t N, uint32_t K,
416 enum CeltSpread spread, uint32_t blocks, float gain,
417 CeltPVQ *pvq)
418 {
419 int *y = pvq->qcoeff;
420
421 celt_exp_rotation(X, N, blocks, K, spread, 1);
422 gain /= sqrtf(pvq->pvq_search(X, y, K, N));
423 celt_encode_pulses(rc, y, N, K);
424 celt_normalize_residual(y, X, N, gain);
425 celt_exp_rotation(X, N, blocks, K, spread, 0);
426 return celt_extract_collapse_mask(y, N, blocks);
427 }
428
429 /** Decode pulse vector and combine the result with the pitch vector to produce
430 the final normalised signal in the current band. */
celt_alg_unquant(OpusRangeCoder * rc,float * X,uint32_t N,uint32_t K,enum CeltSpread spread,uint32_t blocks,float gain,CeltPVQ * pvq)431 static uint32_t celt_alg_unquant(OpusRangeCoder *rc, float *X, uint32_t N, uint32_t K,
432 enum CeltSpread spread, uint32_t blocks, float gain,
433 CeltPVQ *pvq)
434 {
435 int *y = pvq->qcoeff;
436
437 gain /= sqrtf(celt_decode_pulses(rc, y, N, K));
438 celt_normalize_residual(y, X, N, gain);
439 celt_exp_rotation(X, N, blocks, K, spread, 0);
440 return celt_extract_collapse_mask(y, N, blocks);
441 }
442
celt_calc_theta(const float * X,const float * Y,int coupling,int N)443 static int celt_calc_theta(const float *X, const float *Y, int coupling, int N)
444 {
445 int i;
446 float e[2] = { 0.0f, 0.0f };
447 if (coupling) { /* Coupling case */
448 for (i = 0; i < N; i++) {
449 e[0] += (X[i] + Y[i])*(X[i] + Y[i]);
450 e[1] += (X[i] - Y[i])*(X[i] - Y[i]);
451 }
452 } else {
453 for (i = 0; i < N; i++) {
454 e[0] += X[i]*X[i];
455 e[1] += Y[i]*Y[i];
456 }
457 }
458 return lrintf(32768.0f*atan2f(sqrtf(e[1]), sqrtf(e[0]))/M_PI);
459 }
460
celt_stereo_is_decouple(float * X,float * Y,float e_l,float e_r,int N)461 static void celt_stereo_is_decouple(float *X, float *Y, float e_l, float e_r, int N)
462 {
463 int i;
464 const float energy_n = 1.0f/(sqrtf(e_l*e_l + e_r*e_r) + FLT_EPSILON);
465 e_l *= energy_n;
466 e_r *= energy_n;
467 for (i = 0; i < N; i++)
468 X[i] = e_l*X[i] + e_r*Y[i];
469 }
470
celt_stereo_ms_decouple(float * X,float * Y,int N)471 static void celt_stereo_ms_decouple(float *X, float *Y, int N)
472 {
473 int i;
474 for (i = 0; i < N; i++) {
475 const float Xret = X[i];
476 X[i] = (X[i] + Y[i])*M_SQRT1_2;
477 Y[i] = (Y[i] - Xret)*M_SQRT1_2;
478 }
479 }
480
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)481 static av_always_inline uint32_t quant_band_template(CeltPVQ *pvq, CeltFrame *f,
482 OpusRangeCoder *rc,
483 const int band, float *X,
484 float *Y, int N, int b,
485 uint32_t blocks, float *lowband,
486 int duration, float *lowband_out,
487 int level, float gain,
488 float *lowband_scratch,
489 int fill, int quant)
490 {
491 int i;
492 const uint8_t *cache;
493 int stereo = !!Y, split = stereo;
494 int imid = 0, iside = 0;
495 uint32_t N0 = N;
496 int N_B = N / blocks;
497 int N_B0 = N_B;
498 int B0 = blocks;
499 int time_divide = 0;
500 int recombine = 0;
501 int inv = 0;
502 float mid = 0, side = 0;
503 int longblocks = (B0 == 1);
504 uint32_t cm = 0;
505
506 if (N == 1) {
507 float *x = X;
508 for (i = 0; i <= stereo; i++) {
509 int sign = 0;
510 if (f->remaining2 >= 1 << 3) {
511 if (quant) {
512 sign = x[0] < 0;
513 ff_opus_rc_put_raw(rc, sign, 1);
514 } else {
515 sign = ff_opus_rc_get_raw(rc, 1);
516 }
517 f->remaining2 -= 1 << 3;
518 }
519 x[0] = 1.0f - 2.0f*sign;
520 x = Y;
521 }
522 if (lowband_out)
523 lowband_out[0] = X[0];
524 return 1;
525 }
526
527 if (!stereo && level == 0) {
528 int tf_change = f->tf_change[band];
529 int k;
530 if (tf_change > 0)
531 recombine = tf_change;
532 /* Band recombining to increase frequency resolution */
533
534 if (lowband &&
535 (recombine || ((N_B & 1) == 0 && tf_change < 0) || B0 > 1)) {
536 for (i = 0; i < N; i++)
537 lowband_scratch[i] = lowband[i];
538 lowband = lowband_scratch;
539 }
540
541 for (k = 0; k < recombine; k++) {
542 if (quant || lowband)
543 celt_haar1(quant ? X : lowband, N >> k, 1 << k);
544 fill = ff_celt_bit_interleave[fill & 0xF] | ff_celt_bit_interleave[fill >> 4] << 2;
545 }
546 blocks >>= recombine;
547 N_B <<= recombine;
548
549 /* Increasing the time resolution */
550 while ((N_B & 1) == 0 && tf_change < 0) {
551 if (quant || lowband)
552 celt_haar1(quant ? X : lowband, N_B, blocks);
553 fill |= fill << blocks;
554 blocks <<= 1;
555 N_B >>= 1;
556 time_divide++;
557 tf_change++;
558 }
559 B0 = blocks;
560 N_B0 = N_B;
561
562 /* Reorganize the samples in time order instead of frequency order */
563 if (B0 > 1 && (quant || lowband))
564 celt_deinterleave_hadamard(pvq->hadamard_tmp, quant ? X : lowband,
565 N_B >> recombine, B0 << recombine,
566 longblocks);
567 }
568
569 /* If we need 1.5 more bit than we can produce, split the band in two. */
570 cache = ff_celt_cache_bits +
571 ff_celt_cache_index[(duration + 1) * CELT_MAX_BANDS + band];
572 if (!stereo && duration >= 0 && b > cache[cache[0]] + 12 && N > 2) {
573 N >>= 1;
574 Y = X + N;
575 split = 1;
576 duration -= 1;
577 if (blocks == 1)
578 fill = (fill & 1) | (fill << 1);
579 blocks = (blocks + 1) >> 1;
580 }
581
582 if (split) {
583 int qn;
584 int itheta = quant ? celt_calc_theta(X, Y, stereo, N) : 0;
585 int mbits, sbits, delta;
586 int qalloc;
587 int pulse_cap;
588 int offset;
589 int orig_fill;
590 int tell;
591
592 /* Decide on the resolution to give to the split parameter theta */
593 pulse_cap = ff_celt_log_freq_range[band] + duration * 8;
594 offset = (pulse_cap >> 1) - (stereo && N == 2 ? CELT_QTHETA_OFFSET_TWOPHASE :
595 CELT_QTHETA_OFFSET);
596 qn = (stereo && band >= f->intensity_stereo) ? 1 :
597 celt_compute_qn(N, b, offset, pulse_cap, stereo);
598 tell = opus_rc_tell_frac(rc);
599 if (qn != 1) {
600 if (quant)
601 itheta = (itheta*qn + 8192) >> 14;
602 /* Entropy coding of the angle. We use a uniform pdf for the
603 * time split, a step for stereo, and a triangular one for the rest. */
604 if (quant) {
605 if (stereo && N > 2)
606 ff_opus_rc_enc_uint_step(rc, itheta, qn / 2);
607 else if (stereo || B0 > 1)
608 ff_opus_rc_enc_uint(rc, itheta, qn + 1);
609 else
610 ff_opus_rc_enc_uint_tri(rc, itheta, qn);
611 itheta = itheta * 16384 / qn;
612 if (stereo) {
613 if (itheta == 0)
614 celt_stereo_is_decouple(X, Y, f->block[0].lin_energy[band],
615 f->block[1].lin_energy[band], N);
616 else
617 celt_stereo_ms_decouple(X, Y, N);
618 }
619 } else {
620 if (stereo && N > 2)
621 itheta = ff_opus_rc_dec_uint_step(rc, qn / 2);
622 else if (stereo || B0 > 1)
623 itheta = ff_opus_rc_dec_uint(rc, qn+1);
624 else
625 itheta = ff_opus_rc_dec_uint_tri(rc, qn);
626 itheta = itheta * 16384 / qn;
627 }
628 } else if (stereo) {
629 if (quant) {
630 inv = f->apply_phase_inv ? itheta > 8192 : 0;
631 if (inv) {
632 for (i = 0; i < N; i++)
633 Y[i] *= -1;
634 }
635 celt_stereo_is_decouple(X, Y, f->block[0].lin_energy[band],
636 f->block[1].lin_energy[band], N);
637
638 if (b > 2 << 3 && f->remaining2 > 2 << 3) {
639 ff_opus_rc_enc_log(rc, inv, 2);
640 } else {
641 inv = 0;
642 }
643 } else {
644 inv = (b > 2 << 3 && f->remaining2 > 2 << 3) ? ff_opus_rc_dec_log(rc, 2) : 0;
645 inv = f->apply_phase_inv ? inv : 0;
646 }
647 itheta = 0;
648 }
649 qalloc = opus_rc_tell_frac(rc) - tell;
650 b -= qalloc;
651
652 orig_fill = fill;
653 if (itheta == 0) {
654 imid = 32767;
655 iside = 0;
656 fill = av_mod_uintp2(fill, blocks);
657 delta = -16384;
658 } else if (itheta == 16384) {
659 imid = 0;
660 iside = 32767;
661 fill &= ((1 << blocks) - 1) << blocks;
662 delta = 16384;
663 } else {
664 imid = celt_cos(itheta);
665 iside = celt_cos(16384-itheta);
666 /* This is the mid vs side allocation that minimizes squared error
667 in that band. */
668 delta = ROUND_MUL16((N - 1) << 7, celt_log2tan(iside, imid));
669 }
670
671 mid = imid / 32768.0f;
672 side = iside / 32768.0f;
673
674 /* This is a special case for N=2 that only works for stereo and takes
675 advantage of the fact that mid and side are orthogonal to encode
676 the side with just one bit. */
677 if (N == 2 && stereo) {
678 int c;
679 int sign = 0;
680 float tmp;
681 float *x2, *y2;
682 mbits = b;
683 /* Only need one bit for the side */
684 sbits = (itheta != 0 && itheta != 16384) ? 1 << 3 : 0;
685 mbits -= sbits;
686 c = (itheta > 8192);
687 f->remaining2 -= qalloc+sbits;
688
689 x2 = c ? Y : X;
690 y2 = c ? X : Y;
691 if (sbits) {
692 if (quant) {
693 sign = x2[0]*y2[1] - x2[1]*y2[0] < 0;
694 ff_opus_rc_put_raw(rc, sign, 1);
695 } else {
696 sign = ff_opus_rc_get_raw(rc, 1);
697 }
698 }
699 sign = 1 - 2 * sign;
700 /* We use orig_fill here because we want to fold the side, but if
701 itheta==16384, we'll have cleared the low bits of fill. */
702 cm = pvq->quant_band(pvq, f, rc, band, x2, NULL, N, mbits, blocks, lowband, duration,
703 lowband_out, level, gain, lowband_scratch, orig_fill);
704 /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse),
705 and there's no need to worry about mixing with the other channel. */
706 y2[0] = -sign * x2[1];
707 y2[1] = sign * x2[0];
708 X[0] *= mid;
709 X[1] *= mid;
710 Y[0] *= side;
711 Y[1] *= side;
712 tmp = X[0];
713 X[0] = tmp - Y[0];
714 Y[0] = tmp + Y[0];
715 tmp = X[1];
716 X[1] = tmp - Y[1];
717 Y[1] = tmp + Y[1];
718 } else {
719 /* "Normal" split code */
720 float *next_lowband2 = NULL;
721 float *next_lowband_out1 = NULL;
722 int next_level = 0;
723 int rebalance;
724 uint32_t cmt;
725
726 /* Give more bits to low-energy MDCTs than they would
727 * otherwise deserve */
728 if (B0 > 1 && !stereo && (itheta & 0x3fff)) {
729 if (itheta > 8192)
730 /* Rough approximation for pre-echo masking */
731 delta -= delta >> (4 - duration);
732 else
733 /* Corresponds to a forward-masking slope of
734 * 1.5 dB per 10 ms */
735 delta = FFMIN(0, delta + (N << 3 >> (5 - duration)));
736 }
737 mbits = av_clip((b - delta) / 2, 0, b);
738 sbits = b - mbits;
739 f->remaining2 -= qalloc;
740
741 if (lowband && !stereo)
742 next_lowband2 = lowband + N; /* >32-bit split case */
743
744 /* Only stereo needs to pass on lowband_out.
745 * Otherwise, it's handled at the end */
746 if (stereo)
747 next_lowband_out1 = lowband_out;
748 else
749 next_level = level + 1;
750
751 rebalance = f->remaining2;
752 if (mbits >= sbits) {
753 /* In stereo mode, we do not apply a scaling to the mid
754 * because we need the normalized mid for folding later */
755 cm = pvq->quant_band(pvq, f, rc, band, X, NULL, N, mbits, blocks,
756 lowband, duration, next_lowband_out1, next_level,
757 stereo ? 1.0f : (gain * mid), lowband_scratch, fill);
758 rebalance = mbits - (rebalance - f->remaining2);
759 if (rebalance > 3 << 3 && itheta != 0)
760 sbits += rebalance - (3 << 3);
761
762 /* For a stereo split, the high bits of fill are always zero,
763 * so no folding will be done to the side. */
764 cmt = pvq->quant_band(pvq, f, rc, band, Y, NULL, N, sbits, blocks,
765 next_lowband2, duration, NULL, next_level,
766 gain * side, NULL, fill >> blocks);
767 cm |= cmt << ((B0 >> 1) & (stereo - 1));
768 } else {
769 /* For a stereo split, the high bits of fill are always zero,
770 * so no folding will be done to the side. */
771 cm = pvq->quant_band(pvq, f, rc, band, Y, NULL, N, sbits, blocks,
772 next_lowband2, duration, NULL, next_level,
773 gain * side, NULL, fill >> blocks);
774 cm <<= ((B0 >> 1) & (stereo - 1));
775 rebalance = sbits - (rebalance - f->remaining2);
776 if (rebalance > 3 << 3 && itheta != 16384)
777 mbits += rebalance - (3 << 3);
778
779 /* In stereo mode, we do not apply a scaling to the mid because
780 * we need the normalized mid for folding later */
781 cm |= pvq->quant_band(pvq, f, rc, band, X, NULL, N, mbits, blocks,
782 lowband, duration, next_lowband_out1, next_level,
783 stereo ? 1.0f : (gain * mid), lowband_scratch, fill);
784 }
785 }
786 } else {
787 /* This is the basic no-split case */
788 uint32_t q = celt_bits2pulses(cache, b);
789 uint32_t curr_bits = celt_pulses2bits(cache, q);
790 f->remaining2 -= curr_bits;
791
792 /* Ensures we can never bust the budget */
793 while (f->remaining2 < 0 && q > 0) {
794 f->remaining2 += curr_bits;
795 curr_bits = celt_pulses2bits(cache, --q);
796 f->remaining2 -= curr_bits;
797 }
798
799 if (q != 0) {
800 /* Finally do the actual (de)quantization */
801 if (quant) {
802 cm = celt_alg_quant(rc, X, N, (q < 8) ? q : (8 + (q & 7)) << ((q >> 3) - 1),
803 f->spread, blocks, gain, pvq);
804 } else {
805 cm = celt_alg_unquant(rc, X, N, (q < 8) ? q : (8 + (q & 7)) << ((q >> 3) - 1),
806 f->spread, blocks, gain, pvq);
807 }
808 } else {
809 /* If there's no pulse, fill the band anyway */
810 uint32_t cm_mask = (1 << blocks) - 1;
811 fill &= cm_mask;
812 if (fill) {
813 if (!lowband) {
814 /* Noise */
815 for (i = 0; i < N; i++)
816 X[i] = (((int32_t)celt_rng(f)) >> 20);
817 cm = cm_mask;
818 } else {
819 /* Folded spectrum */
820 for (i = 0; i < N; i++) {
821 /* About 48 dB below the "normal" folding level */
822 X[i] = lowband[i] + (((celt_rng(f)) & 0x8000) ? 1.0f / 256 : -1.0f / 256);
823 }
824 cm = fill;
825 }
826 celt_renormalize_vector(X, N, gain);
827 } else {
828 memset(X, 0, N*sizeof(float));
829 }
830 }
831 }
832
833 /* This code is used by the decoder and by the resynthesis-enabled encoder */
834 if (stereo) {
835 if (N > 2)
836 celt_stereo_merge(X, Y, mid, N);
837 if (inv) {
838 for (i = 0; i < N; i++)
839 Y[i] *= -1;
840 }
841 } else if (level == 0) {
842 int k;
843
844 /* Undo the sample reorganization going from time order to frequency order */
845 if (B0 > 1)
846 celt_interleave_hadamard(pvq->hadamard_tmp, X, N_B >> recombine,
847 B0 << recombine, longblocks);
848
849 /* Undo time-freq changes that we did earlier */
850 N_B = N_B0;
851 blocks = B0;
852 for (k = 0; k < time_divide; k++) {
853 blocks >>= 1;
854 N_B <<= 1;
855 cm |= cm >> blocks;
856 celt_haar1(X, N_B, blocks);
857 }
858
859 for (k = 0; k < recombine; k++) {
860 cm = ff_celt_bit_deinterleave[cm];
861 celt_haar1(X, N0>>k, 1<<k);
862 }
863 blocks <<= recombine;
864
865 /* Scale output for later folding */
866 if (lowband_out) {
867 float n = sqrtf(N0);
868 for (i = 0; i < N0; i++)
869 lowband_out[i] = n * X[i];
870 }
871 cm = av_mod_uintp2(cm, blocks);
872 }
873
874 return cm;
875 }
876
QUANT_FN(pvq_decode_band)877 static QUANT_FN(pvq_decode_band)
878 {
879 #if CONFIG_OPUS_DECODER
880 return quant_band_template(pvq, f, rc, band, X, Y, N, b, blocks, lowband, duration,
881 lowband_out, level, gain, lowband_scratch, fill, 0);
882 #else
883 return 0;
884 #endif
885 }
886
QUANT_FN(pvq_encode_band)887 static QUANT_FN(pvq_encode_band)
888 {
889 #if CONFIG_OPUS_ENCODER
890 return quant_band_template(pvq, f, rc, band, X, Y, N, b, blocks, lowband, duration,
891 lowband_out, level, gain, lowband_scratch, fill, 1);
892 #else
893 return 0;
894 #endif
895 }
896
ff_celt_pvq_init(CeltPVQ ** pvq,int encode)897 int av_cold ff_celt_pvq_init(CeltPVQ **pvq, int encode)
898 {
899 CeltPVQ *s = av_malloc(sizeof(CeltPVQ));
900 if (!s)
901 return AVERROR(ENOMEM);
902
903 s->pvq_search = ppp_pvq_search_c;
904 s->quant_band = encode ? pvq_encode_band : pvq_decode_band;
905
906 if (CONFIG_OPUS_ENCODER && ARCH_X86)
907 ff_celt_pvq_init_x86(s);
908
909 *pvq = s;
910
911 return 0;
912 }
913
ff_celt_pvq_uninit(CeltPVQ ** pvq)914 void av_cold ff_celt_pvq_uninit(CeltPVQ **pvq)
915 {
916 av_freep(pvq);
917 }
918