Line | Count | Source (jump to first uncovered line) |
1 | | /* Copyright (c) 2007-2008 CSIRO |
2 | | Copyright (c) 2007-2009 Xiph.Org Foundation |
3 | | Copyright (c) 2008-2009 Gregory Maxwell |
4 | | Written by Jean-Marc Valin and Gregory Maxwell */ |
5 | | /* |
6 | | Redistribution and use in source and binary forms, with or without |
7 | | modification, are permitted provided that the following conditions |
8 | | are met: |
9 | | |
10 | | - Redistributions of source code must retain the above copyright |
11 | | notice, this list of conditions and the following disclaimer. |
12 | | |
13 | | - Redistributions in binary form must reproduce the above copyright |
14 | | notice, this list of conditions and the following disclaimer in the |
15 | | documentation and/or other materials provided with the distribution. |
16 | | |
17 | | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
18 | | ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
19 | | LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
20 | | A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER |
21 | | OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
22 | | EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
23 | | PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR |
24 | | PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF |
25 | | LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING |
26 | | NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS |
27 | | SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
28 | | */ |
29 | | |
30 | | #ifdef HAVE_CONFIG_H |
31 | | #include "config.h" |
32 | | #endif |
33 | | |
34 | | #include <math.h> |
35 | | #include "bands.h" |
36 | | #include "modes.h" |
37 | | #include "vq.h" |
38 | | #include "cwrs.h" |
39 | | #include "stack_alloc.h" |
40 | | #include "os_support.h" |
41 | | #include "mathops.h" |
42 | | #include "rate.h" |
43 | | #include "quant_bands.h" |
44 | | #include "pitch.h" |
45 | | |
46 | | int hysteresis_decision(opus_val16 val, const opus_val16 *thresholds, const opus_val16 *hysteresis, int N, int prev) |
47 | 16.7M | { |
48 | 16.7M | int i; |
49 | 250M | for (i=0;i<N;i++) |
50 | 250M | { |
51 | 250M | if (val < thresholds[i]) |
52 | 16.7M | break; |
53 | 250M | } |
54 | 16.7M | if (i>prev && val < thresholds[prev]+hysteresis[prev]) |
55 | 27.8k | i=prev; |
56 | 16.7M | if (i<prev && val > thresholds[prev-1]-hysteresis[prev-1]) |
57 | 10.2k | i=prev; |
58 | 16.7M | return i; |
59 | 16.7M | } |
60 | | |
61 | | opus_uint32 celt_lcg_rand(opus_uint32 seed) |
62 | 857M | { |
63 | 857M | return 1664525 * seed + 1013904223; |
64 | 857M | } |
65 | | |
66 | | /* This is a cos() approximation designed to be bit-exact on any platform. Bit exactness |
67 | | with this approximation is important because it has an impact on the bit allocation */ |
68 | | opus_int16 bitexact_cos(opus_int16 x) |
69 | 18.0M | { |
70 | 18.0M | opus_int32 tmp; |
71 | 18.0M | opus_int16 x2; |
72 | 18.0M | tmp = (4096+((opus_int32)(x)*(x)))>>13; |
73 | 18.0M | celt_sig_assert(tmp<=32767); |
74 | 18.0M | x2 = tmp; |
75 | 18.0M | x2 = (32767-x2) + FRAC_MUL16(x2, (-7651 + FRAC_MUL16(x2, (8277 + FRAC_MUL16(-626, x2))))); |
76 | 18.0M | celt_sig_assert(x2<=32766); |
77 | 18.0M | return 1+x2; |
78 | 18.0M | } |
79 | | |
80 | | int bitexact_log2tan(int isin,int icos) |
81 | 9.03M | { |
82 | 9.03M | int lc; |
83 | 9.03M | int ls; |
84 | 9.03M | lc=EC_ILOG(icos); |
85 | 9.03M | ls=EC_ILOG(isin); |
86 | 9.03M | icos<<=15-lc; |
87 | 9.03M | isin<<=15-ls; |
88 | 9.03M | return (ls-lc)*(1<<11) |
89 | 9.03M | +FRAC_MUL16(isin, FRAC_MUL16(isin, -2597) + 7932) |
90 | 9.03M | -FRAC_MUL16(icos, FRAC_MUL16(icos, -2597) + 7932); |
91 | 9.03M | } |
92 | | |
93 | | #ifdef FIXED_POINT |
94 | | /* Compute the amplitude (sqrt energy) in each of the bands */ |
95 | | void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch) |
96 | | { |
97 | | int i, c, N; |
98 | | const opus_int16 *eBands = m->eBands; |
99 | | (void)arch; |
100 | | N = m->shortMdctSize<<LM; |
101 | | c=0; do { |
102 | | for (i=0;i<end;i++) |
103 | | { |
104 | | int j; |
105 | | opus_val32 maxval=0; |
106 | | opus_val32 sum = 0; |
107 | | |
108 | | maxval = celt_maxabs32(&X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM); |
109 | | if (maxval > 0) |
110 | | { |
111 | | int shift = IMAX(0, 30 - celt_ilog2(maxval+(maxval>>14)+1) - ((((m->logN[i]+7)>>BITRES)+LM+1)>>1)); |
112 | | j=eBands[i]<<LM; do { |
113 | | opus_val32 x = SHL32(X[j+c*N],shift); |
114 | | sum = ADD32(sum, MULT32_32_Q31(x, x)); |
115 | | } while (++j<eBands[i+1]<<LM); |
116 | | bandE[i+c*m->nbEBands] = MAX32(maxval, PSHR32(celt_sqrt32(SHR32(sum,1)), shift)); |
117 | | } else { |
118 | | bandE[i+c*m->nbEBands] = EPSILON; |
119 | | } |
120 | | } |
121 | | } while (++c<C); |
122 | | } |
123 | | |
124 | | /* Normalise each band such that the energy is one. */ |
125 | | void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M) |
126 | | { |
127 | | int i, c, N; |
128 | | const opus_int16 *eBands = m->eBands; |
129 | | N = M*m->shortMdctSize; |
130 | | c=0; do { |
131 | | i=0; do { |
132 | | int j,shift; |
133 | | opus_val32 E; |
134 | | opus_val32 g; |
135 | | E = bandE[i+c*m->nbEBands]; |
136 | | /* For very low energies, we need this to make sure not to prevent energy rounding from |
137 | | blowing up the normalized signal. */ |
138 | | if (E < 10) E += EPSILON; |
139 | | shift = 30-celt_zlog2(E); |
140 | | E = SHL32(E, shift); |
141 | | g = celt_rcp_norm32(E); |
142 | | j=M*eBands[i]; do { |
143 | | X[j+c*N] = PSHR32(MULT32_32_Q31(g, SHL32(freq[j+c*N], shift)), 30-NORM_SHIFT); |
144 | | } while (++j<M*eBands[i+1]); |
145 | | } while (++i<end); |
146 | | } while (++c<C); |
147 | | } |
148 | | |
149 | | #else /* FIXED_POINT */ |
150 | | /* Compute the amplitude (sqrt energy) in each of the bands */ |
151 | | void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch) |
152 | 55.3M | { |
153 | 55.3M | int i, c, N; |
154 | 55.3M | const opus_int16 *eBands = m->eBands; |
155 | 55.3M | N = m->shortMdctSize<<LM; |
156 | 72.0M | c=0; do { |
157 | 1.21G | for (i=0;i<end;i++) |
158 | 1.14G | { |
159 | 1.14G | opus_val32 sum; |
160 | 1.14G | sum = 1e-27f + celt_inner_prod(&X[c*N+(eBands[i]<<LM)], &X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM, arch); |
161 | 1.14G | bandE[i+c*m->nbEBands] = celt_sqrt(sum); |
162 | | /*printf ("%f ", bandE[i+c*m->nbEBands]);*/ |
163 | 1.14G | } |
164 | 72.0M | } while (++c<C); |
165 | | /*printf ("\n");*/ |
166 | 55.3M | } |
167 | | |
168 | | /* Normalise each band such that the energy is one. */ |
169 | | void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M) |
170 | 55.1M | { |
171 | 55.1M | int i, c, N; |
172 | 55.1M | const opus_int16 *eBands = m->eBands; |
173 | 55.1M | N = M*m->shortMdctSize; |
174 | 71.8M | c=0; do { |
175 | 1.21G | for (i=0;i<end;i++) |
176 | 1.13G | { |
177 | 1.13G | int j; |
178 | 1.13G | opus_val16 g = 1.f/(1e-27f+bandE[i+c*m->nbEBands]); |
179 | 8.83G | for (j=M*eBands[i];j<M*eBands[i+1];j++) |
180 | 7.69G | X[j+c*N] = freq[j+c*N]*g; |
181 | 1.13G | } |
182 | 71.8M | } while (++c<C); |
183 | 55.1M | } |
184 | | |
185 | | #endif /* FIXED_POINT */ |
186 | | |
187 | | /* De-normalise the energy to produce the synthesis from the unit-energy bands */ |
188 | | void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, |
189 | | celt_sig * OPUS_RESTRICT freq, const celt_glog *bandLogE, int start, |
190 | | int end, int M, int downsample, int silence) |
191 | 0 | { |
192 | 0 | int i, N; |
193 | 0 | int bound; |
194 | 0 | celt_sig * OPUS_RESTRICT f; |
195 | 0 | const celt_norm * OPUS_RESTRICT x; |
196 | 0 | const opus_int16 *eBands = m->eBands; |
197 | 0 | N = M*m->shortMdctSize; |
198 | 0 | bound = M*eBands[end]; |
199 | 0 | if (downsample!=1) |
200 | 0 | bound = IMIN(bound, N/downsample); |
201 | 0 | if (silence) |
202 | 0 | { |
203 | 0 | bound = 0; |
204 | 0 | start = end = 0; |
205 | 0 | } |
206 | 0 | f = freq; |
207 | 0 | x = X+M*eBands[start]; |
208 | 0 | if (start != 0) |
209 | 0 | { |
210 | 0 | for (i=0;i<M*eBands[start];i++) |
211 | 0 | *f++ = 0; |
212 | 0 | } else { |
213 | 0 | f += M*eBands[start]; |
214 | 0 | } |
215 | 0 | for (i=start;i<end;i++) |
216 | 0 | { |
217 | 0 | int j, band_end; |
218 | 0 | opus_val32 g; |
219 | 0 | celt_glog lg; |
220 | | #ifdef FIXED_POINT |
221 | | int shift; |
222 | | #endif |
223 | 0 | j=M*eBands[i]; |
224 | 0 | band_end = M*eBands[i+1]; |
225 | 0 | lg = ADD32(bandLogE[i], SHL32((opus_val32)eMeans[i],DB_SHIFT-4)); |
226 | 0 | #ifndef FIXED_POINT |
227 | 0 | g = celt_exp2_db(MIN32(32.f, lg)); |
228 | | #else |
229 | | /* Handle the integer part of the log energy */ |
230 | | shift = 17-(lg>>DB_SHIFT); |
231 | | if (shift>=31) |
232 | | { |
233 | | shift=0; |
234 | | g=0; |
235 | | } else { |
236 | | /* Handle the fractional part. */ |
237 | | g = SHL32(celt_exp2_db_frac((lg&((1<<DB_SHIFT)-1))), 2); |
238 | | } |
239 | | /* Handle extreme gains with negative shift. */ |
240 | | if (shift<0) |
241 | | { |
242 | | /* To avoid overflow, we're |
243 | | capping the gain here, which is equivalent to a cap of 18 on lg. |
244 | | This shouldn't trigger unless the bitstream is already corrupted. */ |
245 | | g = 2147483647; |
246 | | shift = 0; |
247 | | } |
248 | | #endif |
249 | 0 | do { |
250 | 0 | *f++ = PSHR32(MULT32_32_Q31(SHL32(*x, 30-NORM_SHIFT), g), shift); |
251 | 0 | x++; |
252 | 0 | } while (++j<band_end); |
253 | 0 | } |
254 | 0 | celt_assert(start <= end); |
255 | 0 | OPUS_CLEAR(&freq[bound], N-bound); |
256 | 0 | } |
257 | | |
258 | | /* This prevents energy collapse for transients with multiple short MDCTs */ |
259 | | void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_masks, int LM, int C, int size, |
260 | | int start, int end, const celt_glog *logE, const celt_glog *prev1logE, |
261 | | const celt_glog *prev2logE, const int *pulses, opus_uint32 seed, int encode, int arch) |
262 | 0 | { |
263 | 0 | int c, i, j, k; |
264 | 0 | for (i=start;i<end;i++) |
265 | 0 | { |
266 | 0 | int N0; |
267 | 0 | opus_val16 thresh, sqrt_1; |
268 | 0 | int depth; |
269 | | #ifdef FIXED_POINT |
270 | | int shift; |
271 | | opus_val32 thresh32; |
272 | | #endif |
273 | |
|
274 | 0 | N0 = m->eBands[i+1]-m->eBands[i]; |
275 | | /* depth in 1/8 bits */ |
276 | 0 | celt_sig_assert(pulses[i]>=0); |
277 | 0 | depth = celt_udiv(1+pulses[i], (m->eBands[i+1]-m->eBands[i]))>>LM; |
278 | |
|
279 | | #ifdef FIXED_POINT |
280 | | thresh32 = SHR32(celt_exp2(-SHL16(depth, 10-BITRES)),1); |
281 | | thresh = MULT16_32_Q15(QCONST16(0.5f, 15), MIN32(32767,thresh32)); |
282 | | { |
283 | | opus_val32 t; |
284 | | t = N0<<LM; |
285 | | shift = celt_ilog2(t)>>1; |
286 | | t = SHL32(t, (7-shift)<<1); |
287 | | sqrt_1 = celt_rsqrt_norm(t); |
288 | | } |
289 | | #else |
290 | 0 | thresh = .5f*celt_exp2(-.125f*depth); |
291 | 0 | sqrt_1 = celt_rsqrt(N0<<LM); |
292 | 0 | #endif |
293 | |
|
294 | 0 | c=0; do |
295 | 0 | { |
296 | 0 | celt_norm *X; |
297 | 0 | celt_glog prev1; |
298 | 0 | celt_glog prev2; |
299 | 0 | opus_val32 Ediff; |
300 | 0 | celt_norm r; |
301 | 0 | int renormalize=0; |
302 | 0 | prev1 = prev1logE[c*m->nbEBands+i]; |
303 | 0 | prev2 = prev2logE[c*m->nbEBands+i]; |
304 | 0 | if (!encode && C==1) |
305 | 0 | { |
306 | 0 | prev1 = MAXG(prev1,prev1logE[m->nbEBands+i]); |
307 | 0 | prev2 = MAXG(prev2,prev2logE[m->nbEBands+i]); |
308 | 0 | } |
309 | 0 | Ediff = logE[c*m->nbEBands+i]-MING(prev1,prev2); |
310 | 0 | Ediff = MAX32(0, Ediff); |
311 | |
|
312 | | #ifdef FIXED_POINT |
313 | | if (Ediff < GCONST(16.f)) |
314 | | { |
315 | | opus_val32 r32 = SHR32(celt_exp2_db(-Ediff),1); |
316 | | r = 2*MIN16(16383,r32); |
317 | | } else { |
318 | | r = 0; |
319 | | } |
320 | | if (LM==3) |
321 | | r = MULT16_16_Q14(23170, MIN32(23169, r)); |
322 | | r = SHR16(MIN16(thresh, r),1); |
323 | | r = VSHR32(MULT16_16_Q15(sqrt_1, r),shift+14-NORM_SHIFT); |
324 | | #else |
325 | | /* r needs to be multiplied by 2 or 2*sqrt(2) depending on LM because |
326 | | short blocks don't have the same energy as long */ |
327 | 0 | r = 2.f*celt_exp2_db(-Ediff); |
328 | 0 | if (LM==3) |
329 | 0 | r *= 1.41421356f; |
330 | 0 | r = MIN16(thresh, r); |
331 | 0 | r = r*sqrt_1; |
332 | 0 | #endif |
333 | 0 | X = X_+c*size+(m->eBands[i]<<LM); |
334 | 0 | for (k=0;k<1<<LM;k++) |
335 | 0 | { |
336 | | /* Detect collapse */ |
337 | 0 | if (!(collapse_masks[i*C+c]&1<<k)) |
338 | 0 | { |
339 | | /* Fill with noise */ |
340 | 0 | for (j=0;j<N0;j++) |
341 | 0 | { |
342 | 0 | seed = celt_lcg_rand(seed); |
343 | 0 | X[(j<<LM)+k] = (seed&0x8000 ? r : -r); |
344 | 0 | } |
345 | 0 | renormalize = 1; |
346 | 0 | } |
347 | 0 | } |
348 | | /* We just added some energy, so we need to renormalise */ |
349 | 0 | if (renormalize) |
350 | 0 | renormalise_vector(X, N0<<LM, Q31ONE, arch); |
351 | 0 | } while (++c<C); |
352 | 0 | } |
353 | 0 | } |
354 | | |
355 | | /* Compute the weights to use for optimizing normalized distortion across |
356 | | channels. We use the amplitude to weight square distortion, which means |
357 | | that we use the square root of the value we would have been using if we |
358 | | wanted to minimize the MSE in the non-normalized domain. This roughly |
359 | | corresponds to some quick-and-dirty perceptual experiments I ran to |
360 | | measure inter-aural masking (there doesn't seem to be any published data |
361 | | on the topic). */ |
362 | | static void compute_channel_weights(celt_ener Ex, celt_ener Ey, opus_val16 w[2]) |
363 | 2.15M | { |
364 | 2.15M | celt_ener minE; |
365 | | #ifdef FIXED_POINT |
366 | | int shift; |
367 | | #endif |
368 | 2.15M | minE = MIN32(Ex, Ey); |
369 | | /* Adjustment to make the weights a bit more conservative. */ |
370 | 2.15M | Ex = ADD32(Ex, minE/3); |
371 | 2.15M | Ey = ADD32(Ey, minE/3); |
372 | | #ifdef FIXED_POINT |
373 | | shift = celt_ilog2(EPSILON+MAX32(Ex, Ey))-14; |
374 | | #endif |
375 | 2.15M | w[0] = VSHR32(Ex, shift); |
376 | 2.15M | w[1] = VSHR32(Ey, shift); |
377 | 2.15M | } |
378 | | |
379 | | static void intensity_stereo(const CELTMode *m, celt_norm * OPUS_RESTRICT X, const celt_norm * OPUS_RESTRICT Y, const celt_ener *bandE, int bandID, int N) |
380 | 189M | { |
381 | 189M | int i = bandID; |
382 | 189M | int j; |
383 | 189M | opus_val16 a1, a2; |
384 | 189M | opus_val16 left, right; |
385 | 189M | opus_val16 norm; |
386 | | #ifdef FIXED_POINT |
387 | | int shift = celt_zlog2(MAX32(bandE[i], bandE[i+m->nbEBands]))-13; |
388 | | #endif |
389 | 189M | left = VSHR32(bandE[i],shift); |
390 | 189M | right = VSHR32(bandE[i+m->nbEBands],shift); |
391 | 189M | norm = EPSILON + celt_sqrt(EPSILON+MULT16_16(left,left)+MULT16_16(right,right)); |
392 | | #ifdef FIXED_POINT |
393 | | left = MIN32(left, norm-1); |
394 | | right = MIN32(right, norm-1); |
395 | | #endif |
396 | 189M | a1 = DIV32_16(SHL32(EXTEND32(left),15),norm); |
397 | 189M | a2 = DIV32_16(SHL32(EXTEND32(right),15),norm); |
398 | 1.91G | for (j=0;j<N;j++) |
399 | 1.72G | { |
400 | 1.72G | X[j] = ADD32(MULT16_32_Q15(a1, X[j]), MULT16_32_Q15(a2, Y[j])); |
401 | | /* Side is not encoded, no need to calculate */ |
402 | 1.72G | } |
403 | 189M | } |
404 | | |
405 | | static void stereo_split(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT Y, int N) |
406 | 1.77M | { |
407 | 1.77M | int j; |
408 | 22.4M | for (j=0;j<N;j++) |
409 | 20.6M | { |
410 | 20.6M | opus_val32 r, l; |
411 | 20.6M | l = MULT32_32_Q31(QCONST32(.70710678f,31), X[j]); |
412 | 20.6M | r = MULT32_32_Q31(QCONST32(.70710678f,31), Y[j]); |
413 | 20.6M | X[j] = ADD32(l, r); |
414 | 20.6M | Y[j] = SUB32(r, l); |
415 | 20.6M | } |
416 | 1.77M | } |
417 | | |
418 | | static void stereo_merge(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT Y, opus_val32 mid, int N, int arch) |
419 | 64.8M | { |
420 | 64.8M | int j; |
421 | 64.8M | opus_val32 xp=0, side=0; |
422 | 64.8M | opus_val32 El, Er; |
423 | | #ifdef FIXED_POINT |
424 | | int kl, kr; |
425 | | #endif |
426 | 64.8M | opus_val32 t, lgain, rgain; |
427 | | |
428 | | /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */ |
429 | 64.8M | xp = celt_inner_prod_norm_shift(Y, X, N, arch); |
430 | 64.8M | side = celt_inner_prod_norm_shift(Y, Y, N, arch); |
431 | | /* Compensating for the mid normalization */ |
432 | 64.8M | xp = MULT32_32_Q31(mid, xp); |
433 | | /* mid and side are in Q15, not Q14 like X and Y */ |
434 | 64.8M | El = SHR32(MULT32_32_Q31(mid, mid),3) + side - 2*xp; |
435 | 64.8M | Er = SHR32(MULT32_32_Q31(mid, mid),3) + side + 2*xp; |
436 | 64.8M | if (Er < QCONST32(6e-4f, 28) || El < QCONST32(6e-4f, 28)) |
437 | 3.64k | { |
438 | 3.64k | OPUS_COPY(Y, X, N); |
439 | 3.64k | return; |
440 | 3.64k | } |
441 | | |
442 | | #ifdef FIXED_POINT |
443 | | kl = celt_ilog2(El)>>1; |
444 | | kr = celt_ilog2(Er)>>1; |
445 | | #endif |
446 | 64.8M | t = VSHR32(El, (kl<<1)-29); |
447 | 64.8M | lgain = celt_rsqrt_norm32(t); |
448 | 64.8M | t = VSHR32(Er, (kr<<1)-29); |
449 | 64.8M | rgain = celt_rsqrt_norm32(t); |
450 | | |
451 | | #ifdef FIXED_POINT |
452 | | if (kl < 7) |
453 | | kl = 7; |
454 | | if (kr < 7) |
455 | | kr = 7; |
456 | | #endif |
457 | | |
458 | 910M | for (j=0;j<N;j++) |
459 | 845M | { |
460 | 845M | celt_norm r, l; |
461 | | /* Apply mid scaling (side is already scaled) */ |
462 | 845M | l = MULT32_32_Q31(mid, X[j]); |
463 | 845M | r = Y[j]; |
464 | 845M | X[j] = VSHR32(MULT32_32_Q31(lgain, SUB32(l,r)), kl-15); |
465 | 845M | Y[j] = VSHR32(MULT32_32_Q31(rgain, ADD32(l,r)), kr-15); |
466 | 845M | } |
467 | 64.8M | } |
468 | | |
469 | | /* Decide whether we should spread the pulses in the current frame */ |
470 | | int spreading_decision(const CELTMode *m, const celt_norm *X, int *average, |
471 | | int last_decision, int *hf_average, int *tapset_decision, int update_hf, |
472 | | int end, int C, int M, const int *spread_weight) |
473 | 927k | { |
474 | 927k | int i, c, N0; |
475 | 927k | int sum = 0, nbBands=0; |
476 | 927k | const opus_int16 * OPUS_RESTRICT eBands = m->eBands; |
477 | 927k | int decision; |
478 | 927k | int hf_sum=0; |
479 | | |
480 | 927k | celt_assert(end>0); |
481 | | |
482 | 927k | N0 = M*m->shortMdctSize; |
483 | | |
484 | 927k | if (M*(eBands[end]-eBands[end-1]) <= 8) |
485 | 405k | return SPREAD_NONE; |
486 | 661k | c=0; do { |
487 | 12.4M | for (i=0;i<end;i++) |
488 | 11.8M | { |
489 | 11.8M | int j, N, tmp=0; |
490 | 11.8M | int tcount[3] = {0,0,0}; |
491 | 11.8M | const celt_norm * OPUS_RESTRICT x = X+M*eBands[i]+c*N0; |
492 | 11.8M | N = M*(eBands[i+1]-eBands[i]); |
493 | 11.8M | if (N<=8) |
494 | 8.81M | continue; |
495 | | /* Compute rough CDF of |x[j]| */ |
496 | 92.5M | for (j=0;j<N;j++) |
497 | 89.5M | { |
498 | 89.5M | opus_val32 x2N; /* Q13 */ |
499 | | |
500 | 89.5M | x2N = MULT16_16(MULT16_16_Q15(SHR32(x[j], NORM_SHIFT-14), SHR32(x[j], NORM_SHIFT-14)), N); |
501 | 89.5M | if (x2N < QCONST16(0.25f,13)) |
502 | 36.0M | tcount[0]++; |
503 | 89.5M | if (x2N < QCONST16(0.0625f,13)) |
504 | 20.0M | tcount[1]++; |
505 | 89.5M | if (x2N < QCONST16(0.015625f,13)) |
506 | 10.8M | tcount[2]++; |
507 | 89.5M | } |
508 | | |
509 | | /* Only include four last bands (8 kHz and up) */ |
510 | 2.98M | if (i>m->nbEBands-4) |
511 | 622k | hf_sum += celt_udiv(32*(tcount[1]+tcount[0]), N); |
512 | 2.98M | tmp = (2*tcount[2] >= N) + (2*tcount[1] >= N) + (2*tcount[0] >= N); |
513 | 2.98M | sum += tmp*spread_weight[i]; |
514 | 2.98M | nbBands+=spread_weight[i]; |
515 | 2.98M | } |
516 | 661k | } while (++c<C); |
517 | | |
518 | 521k | if (update_hf) |
519 | 24.8k | { |
520 | 24.8k | if (hf_sum) |
521 | 22.2k | hf_sum = celt_udiv(hf_sum, C*(4-m->nbEBands+end)); |
522 | 24.8k | *hf_average = (*hf_average+hf_sum)>>1; |
523 | 24.8k | hf_sum = *hf_average; |
524 | 24.8k | if (*tapset_decision==2) |
525 | 5.78k | hf_sum += 4; |
526 | 19.0k | else if (*tapset_decision==0) |
527 | 17.3k | hf_sum -= 4; |
528 | 24.8k | if (hf_sum > 22) |
529 | 5.79k | *tapset_decision=2; |
530 | 19.0k | else if (hf_sum > 18) |
531 | 1.70k | *tapset_decision=1; |
532 | 17.3k | else |
533 | 17.3k | *tapset_decision=0; |
534 | 24.8k | } |
535 | | /*printf("%d %d %d\n", hf_sum, *hf_average, *tapset_decision);*/ |
536 | 521k | celt_assert(nbBands>0); /* end has to be non-zero */ |
537 | 521k | celt_assert(sum>=0); |
538 | 521k | sum = celt_udiv((opus_int32)sum<<8, nbBands); |
539 | | /* Recursive averaging */ |
540 | 521k | sum = (sum+*average)>>1; |
541 | 521k | *average = sum; |
542 | | /* Hysteresis */ |
543 | 521k | sum = (3*sum + (((3-last_decision)<<7) + 64) + 2)>>2; |
544 | 521k | if (sum < 80) |
545 | 310k | { |
546 | 310k | decision = SPREAD_AGGRESSIVE; |
547 | 310k | } else if (sum < 256) |
548 | 173k | { |
549 | 173k | decision = SPREAD_NORMAL; |
550 | 173k | } else if (sum < 384) |
551 | 22.4k | { |
552 | 22.4k | decision = SPREAD_LIGHT; |
553 | 22.4k | } else { |
554 | 15.0k | decision = SPREAD_NONE; |
555 | 15.0k | } |
556 | | #ifdef FUZZING |
557 | | decision = rand()&0x3; |
558 | | *tapset_decision=rand()%3; |
559 | | #endif |
560 | 521k | return decision; |
561 | 521k | } |
562 | | |
563 | | /* Indexing table for converting from natural Hadamard to ordery Hadamard |
564 | | This is essentially a bit-reversed Gray, on top of which we've added |
565 | | an inversion of the order because we want the DC at the end rather than |
566 | | the beginning. The lines are for N=2, 4, 8, 16 */ |
567 | | static const int ordery_table[] = { |
568 | | 1, 0, |
569 | | 3, 0, 2, 1, |
570 | | 7, 0, 4, 3, 6, 1, 5, 2, |
571 | | 15, 0, 8, 7, 12, 3, 11, 4, 14, 1, 9, 6, 13, 2, 10, 5, |
572 | | }; |
573 | | |
574 | | static void deinterleave_hadamard(celt_norm *X, int N0, int stride, int hadamard) |
575 | 8.11M | { |
576 | 8.11M | int i,j; |
577 | 8.11M | VARDECL(celt_norm, tmp); |
578 | 8.11M | int N; |
579 | 8.11M | SAVE_STACK; |
580 | 8.11M | N = N0*stride; |
581 | 8.11M | ALLOC(tmp, N, celt_norm); |
582 | 8.11M | celt_assert(stride>0); |
583 | 8.11M | if (hadamard) |
584 | 1.66M | { |
585 | 1.66M | const int *ordery = ordery_table+stride-2; |
586 | 8.24M | for (i=0;i<stride;i++) |
587 | 6.57M | { |
588 | 36.7M | for (j=0;j<N0;j++) |
589 | 30.1M | tmp[ordery[i]*N0+j] = X[j*stride+i]; |
590 | 6.57M | } |
591 | 6.44M | } else { |
592 | 54.0M | for (i=0;i<stride;i++) |
593 | 170M | for (j=0;j<N0;j++) |
594 | 122M | tmp[i*N0+j] = X[j*stride+i]; |
595 | 6.44M | } |
596 | 8.11M | OPUS_COPY(X, tmp, N); |
597 | 8.11M | RESTORE_STACK; |
598 | 8.11M | } |
599 | | |
600 | | static void interleave_hadamard(celt_norm *X, int N0, int stride, int hadamard) |
601 | 1.55M | { |
602 | 1.55M | int i,j; |
603 | 1.55M | VARDECL(celt_norm, tmp); |
604 | 1.55M | int N; |
605 | 1.55M | SAVE_STACK; |
606 | 1.55M | N = N0*stride; |
607 | 1.55M | ALLOC(tmp, N, celt_norm); |
608 | 1.55M | if (hadamard) |
609 | 594k | { |
610 | 594k | const int *ordery = ordery_table+stride-2; |
611 | 2.97M | for (i=0;i<stride;i++) |
612 | 12.1M | for (j=0;j<N0;j++) |
613 | 9.76M | tmp[j*stride+i] = X[ordery[i]*N0+j]; |
614 | 956k | } else { |
615 | 9.84M | for (i=0;i<stride;i++) |
616 | 27.6M | for (j=0;j<N0;j++) |
617 | 18.7M | tmp[j*stride+i] = X[i*N0+j]; |
618 | 956k | } |
619 | 1.55M | OPUS_COPY(X, tmp, N); |
620 | 1.55M | RESTORE_STACK; |
621 | 1.55M | } |
622 | | |
623 | | void haar1(celt_norm *X, int N0, int stride) |
624 | 398M | { |
625 | 398M | int i, j; |
626 | 398M | N0 >>= 1; |
627 | 1.27G | for (i=0;i<stride;i++) |
628 | 3.84G | for (j=0;j<N0;j++) |
629 | 2.97G | { |
630 | 2.97G | opus_val32 tmp1, tmp2; |
631 | 2.97G | tmp1 = MULT32_32_Q31(QCONST32(.70710678f,31), X[stride*2*j+i]); |
632 | 2.97G | tmp2 = MULT32_32_Q31(QCONST32(.70710678f,31), X[stride*(2*j+1)+i]); |
633 | 2.97G | X[stride*2*j+i] = ADD32(tmp1, tmp2); |
634 | 2.97G | X[stride*(2*j+1)+i] = SUB32(tmp1, tmp2); |
635 | 2.97G | } |
636 | 398M | } |
637 | | |
638 | | static int compute_qn(int N, int b, int offset, int pulse_cap, int stereo) |
639 | 213M | { |
640 | 213M | static const opus_int16 exp2_table8[8] = |
641 | 213M | {16384, 17866, 19483, 21247, 23170, 25267, 27554, 30048}; |
642 | 213M | int qn, qb; |
643 | 213M | int N2 = 2*N-1; |
644 | 213M | if (stereo && N==2) |
645 | 62.1M | N2--; |
646 | | /* The upper limit ensures that in a stereo split with itheta==16384, we'll |
647 | | always have enough bits left over to code at least one pulse in the |
648 | | side; otherwise it would collapse, since it doesn't get folded. */ |
649 | 213M | qb = celt_sudiv(b+N2*offset, N2); |
650 | 213M | qb = IMIN(b-pulse_cap-(4<<BITRES), qb); |
651 | | |
652 | 213M | qb = IMIN(8<<BITRES, qb); |
653 | | |
654 | 213M | if (qb<(1<<BITRES>>1)) { |
655 | 187M | qn = 1; |
656 | 187M | } else { |
657 | 26.1M | qn = exp2_table8[qb&0x7]>>(14-(qb>>BITRES)); |
658 | 26.1M | qn = (qn+1)>>1<<1; |
659 | 26.1M | } |
660 | 213M | celt_assert(qn <= 256); |
661 | 213M | return qn; |
662 | 213M | } |
663 | | |
664 | | struct band_ctx { |
665 | | int encode; |
666 | | int resynth; |
667 | | const CELTMode *m; |
668 | | int i; |
669 | | int intensity; |
670 | | int spread; |
671 | | int tf_change; |
672 | | ec_ctx *ec; |
673 | | opus_int32 remaining_bits; |
674 | | const celt_ener *bandE; |
675 | | opus_uint32 seed; |
676 | | int arch; |
677 | | int theta_round; |
678 | | int disable_inv; |
679 | | int avoid_split_noise; |
680 | | #ifdef ENABLE_QEXT |
681 | | ec_ctx *ext_ec; |
682 | | int extra_bits; |
683 | | opus_int32 ext_total_bits; |
684 | | int extra_bands; |
685 | | #endif |
686 | | }; |
687 | | |
688 | | struct split_ctx { |
689 | | int inv; |
690 | | int imid; |
691 | | int iside; |
692 | | int delta; |
693 | | int itheta; |
694 | | #ifdef ENABLE_QEXT |
695 | | int itheta_q30; |
696 | | #endif |
697 | | int qalloc; |
698 | | }; |
699 | | |
700 | | static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx, |
701 | | celt_norm *X, celt_norm *Y, int N, int *b, int B, int B0, |
702 | | int LM, |
703 | | int stereo, int *fill ARG_QEXT(int *ext_b)) |
704 | 213M | { |
705 | 213M | int qn; |
706 | 213M | int itheta=0; |
707 | 213M | int itheta_q30=0; |
708 | 213M | int delta; |
709 | 213M | int imid, iside; |
710 | 213M | int qalloc; |
711 | 213M | int pulse_cap; |
712 | 213M | int offset; |
713 | 213M | opus_int32 tell; |
714 | 213M | int inv=0; |
715 | 213M | int encode; |
716 | 213M | const CELTMode *m; |
717 | 213M | int i; |
718 | 213M | int intensity; |
719 | 213M | ec_ctx *ec; |
720 | 213M | const celt_ener *bandE; |
721 | | |
722 | 213M | encode = ctx->encode; |
723 | 213M | m = ctx->m; |
724 | 213M | i = ctx->i; |
725 | 213M | intensity = ctx->intensity; |
726 | 213M | ec = ctx->ec; |
727 | 213M | bandE = ctx->bandE; |
728 | | |
729 | | /* Decide on the resolution to give to the split parameter theta */ |
730 | 213M | pulse_cap = m->logN[i]+LM*(1<<BITRES); |
731 | 213M | offset = (pulse_cap>>1) - (stereo&&N==2 ? QTHETA_OFFSET_TWOPHASE : QTHETA_OFFSET); |
732 | 213M | qn = compute_qn(N, *b, offset, pulse_cap, stereo); |
733 | 213M | if (stereo && i>=intensity) |
734 | 188M | qn = 1; |
735 | 213M | if (encode) |
736 | 213M | { |
737 | | /* theta is the atan() of the ratio between the (normalized) |
738 | | side and mid. With just that parameter, we can re-scale both |
739 | | mid and side because we know that 1) they have unit norm and |
740 | | 2) they are orthogonal. */ |
741 | 213M | itheta_q30 = stereo_itheta(X, Y, stereo, N, ctx->arch); |
742 | 213M | itheta = itheta_q30>>16; |
743 | 213M | } |
744 | 213M | tell = ec_tell_frac(ec); |
745 | 213M | if (qn!=1) |
746 | 25.0M | { |
747 | 25.0M | if (encode) |
748 | 25.0M | { |
749 | 25.0M | if (!stereo || ctx->theta_round == 0) |
750 | 23.1M | { |
751 | 23.1M | itheta = (itheta*(opus_int32)qn+8192)>>14; |
752 | 23.1M | if (!stereo && ctx->avoid_split_noise && itheta > 0 && itheta < qn) |
753 | 280k | { |
754 | | /* Check if the selected value of theta will cause the bit allocation |
755 | | to inject noise on one side. If so, make sure the energy of that side |
756 | | is zero. */ |
757 | 280k | int unquantized = celt_udiv((opus_int32)itheta*16384, qn); |
758 | 280k | imid = bitexact_cos((opus_int16)unquantized); |
759 | 280k | iside = bitexact_cos((opus_int16)(16384-unquantized)); |
760 | 280k | delta = FRAC_MUL16((N-1)<<7,bitexact_log2tan(iside,imid)); |
761 | 280k | if (delta > *b) |
762 | 911 | itheta = qn; |
763 | 279k | else if (delta < -*b) |
764 | 1.36k | itheta = 0; |
765 | 280k | } |
766 | 23.1M | } else { |
767 | 1.93M | int down; |
768 | | /* Bias quantization towards itheta=0 and itheta=16384. */ |
769 | 1.93M | int bias = itheta > 8192 ? 32767/qn : -32767/qn; |
770 | 1.93M | down = IMIN(qn-1, IMAX(0, (itheta*(opus_int32)qn + bias)>>14)); |
771 | 1.93M | if (ctx->theta_round < 0) |
772 | 965k | itheta = down; |
773 | 965k | else |
774 | 965k | itheta = down+1; |
775 | 1.93M | } |
776 | 25.0M | } |
777 | | /* Entropy coding of the angle. We use a uniform pdf for the |
778 | | time split, a step for stereo, and a triangular one for the rest. */ |
779 | 25.0M | if (stereo && N>2) |
780 | 2.26M | { |
781 | 2.26M | int p0 = 3; |
782 | 2.26M | int x = itheta; |
783 | 2.26M | int x0 = qn/2; |
784 | 2.26M | int ft = p0*(x0+1) + x0; |
785 | | /* Use a probability of p0 up to itheta=8192 and then use 1 after */ |
786 | 2.26M | if (encode) |
787 | 2.26M | { |
788 | 2.26M | ec_encode(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); |
789 | 2.26M | } else { |
790 | 0 | int fs; |
791 | 0 | fs=ec_decode(ec,ft); |
792 | 0 | if (fs<(x0+1)*p0) |
793 | 0 | x=fs/p0; |
794 | 0 | else |
795 | 0 | x=x0+1+(fs-(x0+1)*p0); |
796 | 0 | ec_dec_update(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); |
797 | 0 | itheta = x; |
798 | 0 | } |
799 | 22.7M | } else if (B0>1 || stereo) { |
800 | | /* Uniform pdf */ |
801 | 4.45M | if (encode) |
802 | 4.45M | ec_enc_uint(ec, itheta, qn+1); |
803 | 0 | else |
804 | 0 | itheta = ec_dec_uint(ec, qn+1); |
805 | 18.3M | } else { |
806 | 18.3M | int fs=1, ft; |
807 | 18.3M | ft = ((qn>>1)+1)*((qn>>1)+1); |
808 | 18.3M | if (encode) |
809 | 18.3M | { |
810 | 18.3M | int fl; |
811 | | |
812 | 18.3M | fs = itheta <= (qn>>1) ? itheta + 1 : qn + 1 - itheta; |
813 | 18.3M | fl = itheta <= (qn>>1) ? itheta*(itheta + 1)>>1 : |
814 | 18.3M | ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); |
815 | | |
816 | 18.3M | ec_encode(ec, fl, fl+fs, ft); |
817 | 18.3M | } else { |
818 | | /* Triangular pdf */ |
819 | 0 | int fl=0; |
820 | 0 | int fm; |
821 | 0 | fm = ec_decode(ec, ft); |
822 | |
|
823 | 0 | if (fm < ((qn>>1)*((qn>>1) + 1)>>1)) |
824 | 0 | { |
825 | 0 | itheta = (isqrt32(8*(opus_uint32)fm + 1) - 1)>>1; |
826 | 0 | fs = itheta + 1; |
827 | 0 | fl = itheta*(itheta + 1)>>1; |
828 | 0 | } |
829 | 0 | else |
830 | 0 | { |
831 | 0 | itheta = (2*(qn + 1) |
832 | 0 | - isqrt32(8*(opus_uint32)(ft - fm - 1) + 1))>>1; |
833 | 0 | fs = qn + 1 - itheta; |
834 | 0 | fl = ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); |
835 | 0 | } |
836 | |
|
837 | 0 | ec_dec_update(ec, fl, fl+fs, ft); |
838 | 0 | } |
839 | 18.3M | } |
840 | 25.0M | celt_assert(itheta>=0); |
841 | 25.0M | itheta = celt_udiv((opus_int32)itheta*16384, qn); |
842 | | #ifdef ENABLE_QEXT |
843 | | *ext_b = IMIN(*ext_b, ctx->ext_total_bits - (opus_int32)ec_tell_frac(ctx->ext_ec)); |
844 | | if (*ext_b >= 2*N<<BITRES && ctx->ext_total_bits-ec_tell_frac(ctx->ext_ec)-1 > 2<<BITRES) { |
845 | | int extra_bits; |
846 | | int ext_tell = ec_tell_frac(ctx->ext_ec); |
847 | | extra_bits = IMIN(12, IMAX(2, celt_sudiv(*ext_b, (2*N-1)<<BITRES))); |
848 | | if (encode) { |
849 | | itheta_q30 = itheta_q30 - (itheta<<16); |
850 | | itheta_q30 = (itheta_q30*(opus_int64)qn*((1<<extra_bits)-1)+(1<<29))>>30; |
851 | | itheta_q30 += (1<<(extra_bits-1))-1; |
852 | | itheta_q30 = IMAX(0, IMIN((1<<extra_bits)-2, itheta_q30)); |
853 | | ec_enc_uint(ctx->ext_ec, itheta_q30, (1<<extra_bits)-1); |
854 | | } else { |
855 | | itheta_q30 = ec_dec_uint(ctx->ext_ec, (1<<extra_bits)-1); |
856 | | } |
857 | | itheta_q30 -= (1<<(extra_bits-1))-1; |
858 | | itheta_q30 = (itheta<<16) + itheta_q30*(opus_int64)(1<<30)/(qn*((1<<extra_bits)-1)); |
859 | | /* Hard bounds on itheta (can only trigger on corrupted bitstreams). */ |
860 | | itheta_q30 = IMAX(0, IMIN(itheta_q30, 1073741824)); |
861 | | *ext_b -= ec_tell_frac(ctx->ext_ec) - ext_tell; |
862 | | } else { |
863 | | itheta_q30 = (opus_int32)itheta<<16; |
864 | | } |
865 | | #endif |
866 | 25.0M | if (encode && stereo) |
867 | 2.78M | { |
868 | 2.78M | if (itheta==0) |
869 | 1.01M | intensity_stereo(m, X, Y, bandE, i, N); |
870 | 1.77M | else |
871 | 1.77M | stereo_split(X, Y, N); |
872 | 2.78M | } |
873 | | /* NOTE: Renormalising X and Y *may* help fixed-point a bit at very high rate. |
874 | | Let's do that at higher complexity */ |
875 | 188M | } else if (stereo) { |
876 | 188M | if (encode) |
877 | 188M | { |
878 | 188M | inv = itheta > 8192 && !ctx->disable_inv; |
879 | 188M | if (inv) |
880 | 252k | { |
881 | 252k | int j; |
882 | 5.28M | for (j=0;j<N;j++) |
883 | 5.03M | Y[j] = -Y[j]; |
884 | 252k | } |
885 | 188M | intensity_stereo(m, X, Y, bandE, i, N); |
886 | 188M | } |
887 | 188M | if (*b>2<<BITRES && ctx->remaining_bits > 2<<BITRES) |
888 | 2.03M | { |
889 | 2.03M | if (encode) |
890 | 2.03M | ec_enc_bit_logp(ec, inv, 2); |
891 | 0 | else |
892 | 0 | inv = ec_dec_bit_logp(ec, 2); |
893 | 2.03M | } else |
894 | 186M | inv = 0; |
895 | | /* inv flag override to avoid problems with downmixing. */ |
896 | 188M | if (ctx->disable_inv) |
897 | 89.3M | inv = 0; |
898 | 188M | itheta = 0; |
899 | 188M | itheta_q30 = 0; |
900 | 188M | } |
901 | 213M | qalloc = ec_tell_frac(ec) - tell; |
902 | 213M | *b -= qalloc; |
903 | | |
904 | 213M | if (itheta == 0) |
905 | 204M | { |
906 | 204M | imid = 32767; |
907 | 204M | iside = 0; |
908 | 204M | *fill &= (1<<B)-1; |
909 | 204M | delta = -16384; |
910 | 204M | } else if (itheta == 16384) |
911 | 464k | { |
912 | 464k | imid = 0; |
913 | 464k | iside = 32767; |
914 | 464k | *fill &= ((1<<B)-1)<<B; |
915 | 464k | delta = 16384; |
916 | 8.75M | } else { |
917 | 8.75M | imid = bitexact_cos((opus_int16)itheta); |
918 | 8.75M | iside = bitexact_cos((opus_int16)(16384-itheta)); |
919 | | /* This is the mid vs side allocation that minimizes squared error |
920 | | in that band. */ |
921 | 8.75M | delta = FRAC_MUL16((N-1)<<7,bitexact_log2tan(iside,imid)); |
922 | 8.75M | } |
923 | | |
924 | 213M | sctx->inv = inv; |
925 | 213M | sctx->imid = imid; |
926 | 213M | sctx->iside = iside; |
927 | 213M | sctx->delta = delta; |
928 | 213M | sctx->itheta = itheta; |
929 | | #ifdef ENABLE_QEXT |
930 | | sctx->itheta_q30 = itheta_q30; |
931 | | #endif |
932 | 213M | sctx->qalloc = qalloc; |
933 | 213M | } |
934 | | static unsigned quant_band_n1(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, |
935 | | celt_norm *lowband_out) |
936 | 260M | { |
937 | 260M | int c; |
938 | 260M | int stereo; |
939 | 260M | celt_norm *x = X; |
940 | 260M | int encode; |
941 | 260M | ec_ctx *ec; |
942 | | |
943 | 260M | encode = ctx->encode; |
944 | 260M | ec = ctx->ec; |
945 | | |
946 | 260M | stereo = Y != NULL; |
947 | 332M | c=0; do { |
948 | 332M | int sign=0; |
949 | 332M | if (ctx->remaining_bits>=1<<BITRES) |
950 | 13.1M | { |
951 | 13.1M | if (encode) |
952 | 13.1M | { |
953 | 13.1M | sign = x[0]<0; |
954 | 13.1M | ec_enc_bits(ec, sign, 1); |
955 | 13.1M | } else { |
956 | 0 | sign = ec_dec_bits(ec, 1); |
957 | 0 | } |
958 | 13.1M | ctx->remaining_bits -= 1<<BITRES; |
959 | 13.1M | } |
960 | 332M | if (ctx->resynth) |
961 | 52.2M | x[0] = sign ? -NORM_SCALING : NORM_SCALING; |
962 | 332M | x = Y; |
963 | 332M | } while (++c<1+stereo); |
964 | 260M | if (lowband_out) |
965 | 260M | lowband_out[0] = SHR32(X[0],4); |
966 | 260M | return 1; |
967 | 260M | } |
968 | | |
969 | | /* This function is responsible for encoding and decoding a mono partition. |
970 | | It can split the band in two and transmit the energy difference with |
971 | | the two half-bands. It can be called recursively so bands can end up being |
972 | | split in 8 parts. */ |
973 | | static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X, |
974 | | int N, int b, int B, celt_norm *lowband, |
975 | | int LM, |
976 | | opus_val32 gain, int fill |
977 | | ARG_QEXT(int ext_b)) |
978 | 756M | { |
979 | 756M | const unsigned char *cache; |
980 | 756M | int q; |
981 | 756M | int curr_bits; |
982 | 756M | int imid=0, iside=0; |
983 | 756M | int B0=B; |
984 | 756M | opus_val32 mid=0, side=0; |
985 | 756M | unsigned cm=0; |
986 | 756M | celt_norm *Y=NULL; |
987 | 756M | int encode; |
988 | 756M | const CELTMode *m; |
989 | 756M | int i; |
990 | 756M | int spread; |
991 | 756M | ec_ctx *ec; |
992 | | |
993 | 756M | encode = ctx->encode; |
994 | 756M | m = ctx->m; |
995 | 756M | i = ctx->i; |
996 | 756M | spread = ctx->spread; |
997 | 756M | ec = ctx->ec; |
998 | | |
999 | | /* If we need 1.5 more bit than we can produce, split the band in two. */ |
1000 | 756M | cache = m->cache.bits + m->cache.index[(LM+1)*m->nbEBands+i]; |
1001 | 756M | if (LM != -1 && b > cache[cache[0]]+12 && N>2) |
1002 | 22.2M | { |
1003 | 22.2M | int mbits, sbits, delta; |
1004 | 22.2M | int itheta; |
1005 | 22.2M | int qalloc; |
1006 | 22.2M | struct split_ctx sctx; |
1007 | 22.2M | celt_norm *next_lowband2=NULL; |
1008 | 22.2M | opus_int32 rebalance; |
1009 | | |
1010 | 22.2M | N >>= 1; |
1011 | 22.2M | Y = X+N; |
1012 | 22.2M | LM -= 1; |
1013 | 22.2M | if (B==1) |
1014 | 18.3M | fill = (fill&1)|(fill<<1); |
1015 | 22.2M | B = (B+1)>>1; |
1016 | | |
1017 | 22.2M | compute_theta(ctx, &sctx, X, Y, N, &b, B, B0, LM, 0, &fill ARG_QEXT(&ext_b)); |
1018 | 22.2M | imid = sctx.imid; |
1019 | 22.2M | iside = sctx.iside; |
1020 | 22.2M | delta = sctx.delta; |
1021 | 22.2M | itheta = sctx.itheta; |
1022 | 22.2M | qalloc = sctx.qalloc; |
1023 | | #ifdef FIXED_POINT |
1024 | | # ifdef ENABLE_QEXT |
1025 | | (void)imid; |
1026 | | (void)iside; |
1027 | | mid = celt_cos_norm32(sctx.itheta_q30); |
1028 | | side = celt_cos_norm32((1<<30)-sctx.itheta_q30); |
1029 | | # else |
1030 | | mid = SHL32(EXTEND32(imid), 16); |
1031 | | side = SHL32(EXTEND32(iside), 16); |
1032 | | # endif |
1033 | | #else |
1034 | | # ifdef ENABLE_QEXT |
1035 | | (void)imid; |
1036 | | (void)iside; |
1037 | | mid = celt_cos_norm2(sctx.itheta_q30*(1.f/(1<<30))); |
1038 | | side = celt_cos_norm2(1.f-sctx.itheta_q30*(1.f/(1<<30))); |
1039 | | # else |
1040 | 22.2M | mid = (1.f/32768)*imid; |
1041 | 22.2M | side = (1.f/32768)*iside; |
1042 | 22.2M | # endif |
1043 | 22.2M | #endif |
1044 | | |
1045 | | /* Give more bits to low-energy MDCTs than they would otherwise deserve */ |
1046 | 22.2M | if (B0>1 && (itheta&0x3fff)) |
1047 | 3.19M | { |
1048 | 3.19M | if (itheta > 8192) |
1049 | | /* Rough approximation for pre-echo masking */ |
1050 | 1.25M | delta -= delta>>(4-LM); |
1051 | 1.93M | else |
1052 | | /* Corresponds to a forward-masking slope of 1.5 dB per 10 ms */ |
1053 | 1.93M | delta = IMIN(0, delta + (N<<BITRES>>(5-LM))); |
1054 | 3.19M | } |
1055 | 22.2M | mbits = IMAX(0, IMIN(b, (b-delta)/2)); |
1056 | 22.2M | sbits = b-mbits; |
1057 | 22.2M | ctx->remaining_bits -= qalloc; |
1058 | | |
1059 | 22.2M | if (lowband) |
1060 | 837k | next_lowband2 = lowband+N; /* >32-bit split case */ |
1061 | | |
1062 | 22.2M | rebalance = ctx->remaining_bits; |
1063 | 22.2M | if (mbits >= sbits) |
1064 | 18.5M | { |
1065 | 18.5M | cm = quant_partition(ctx, X, N, mbits, B, lowband, LM, |
1066 | 18.5M | MULT32_32_Q31(gain,mid), fill ARG_QEXT(ext_b/2)); |
1067 | 18.5M | rebalance = mbits - (rebalance-ctx->remaining_bits); |
1068 | 18.5M | if (rebalance > 3<<BITRES && itheta!=0) |
1069 | 506k | sbits += rebalance - (3<<BITRES); |
1070 | 18.5M | cm |= quant_partition(ctx, Y, N, sbits, B, next_lowband2, LM, |
1071 | 18.5M | MULT32_32_Q31(gain,side), fill>>B ARG_QEXT(ext_b/2))<<(B0>>1); |
1072 | 18.5M | } else { |
1073 | 3.68M | cm = quant_partition(ctx, Y, N, sbits, B, next_lowband2, LM, |
1074 | 3.68M | MULT32_32_Q31(gain,side), fill>>B ARG_QEXT(ext_b/2))<<(B0>>1); |
1075 | 3.68M | rebalance = sbits - (rebalance-ctx->remaining_bits); |
1076 | 3.68M | if (rebalance > 3<<BITRES && itheta!=16384) |
1077 | 383k | mbits += rebalance - (3<<BITRES); |
1078 | 3.68M | cm |= quant_partition(ctx, X, N, mbits, B, lowband, LM, |
1079 | 3.68M | MULT32_32_Q31(gain,mid), fill ARG_QEXT(ext_b/2)); |
1080 | 3.68M | } |
1081 | 734M | } else { |
1082 | | #ifdef ENABLE_QEXT |
1083 | | int extra_bits; |
1084 | | int ext_remaining_bits; |
1085 | | extra_bits = ext_b/(N-1)>>BITRES; |
1086 | | ext_remaining_bits = ctx->ext_total_bits-(opus_int32)ec_tell_frac(ctx->ext_ec); |
1087 | | if (ext_remaining_bits < ((extra_bits+1)*(N-1)+N)<<BITRES) { |
1088 | | extra_bits = (ext_remaining_bits-(N<<BITRES))/(N-1)>>BITRES; |
1089 | | extra_bits = IMAX(extra_bits-1, 0); |
1090 | | } |
1091 | | extra_bits = IMIN(12, extra_bits); |
1092 | | #endif |
1093 | | /* This is the basic no-split case */ |
1094 | 734M | q = bits2pulses(m, i, LM, b); |
1095 | 734M | curr_bits = pulses2bits(m, i, LM, q); |
1096 | 734M | ctx->remaining_bits -= curr_bits; |
1097 | | |
1098 | | /* Ensures we can never bust the budget */ |
1099 | 735M | while (ctx->remaining_bits < 0 && q > 0) |
1100 | 1.07M | { |
1101 | 1.07M | ctx->remaining_bits += curr_bits; |
1102 | 1.07M | q--; |
1103 | 1.07M | curr_bits = pulses2bits(m, i, LM, q); |
1104 | 1.07M | ctx->remaining_bits -= curr_bits; |
1105 | 1.07M | } |
1106 | | |
1107 | 734M | if (q!=0) |
1108 | 33.1M | { |
1109 | 33.1M | int K = get_pulses(q); |
1110 | | |
1111 | | /* Finally do the actual quantization */ |
1112 | 33.1M | if (encode) |
1113 | 33.1M | { |
1114 | 33.1M | cm = alg_quant(X, N, K, spread, B, ec, gain, ctx->resynth |
1115 | 33.1M | ARG_QEXT(ctx->ext_ec) ARG_QEXT(extra_bits), |
1116 | 33.1M | ctx->arch); |
1117 | 33.1M | } else { |
1118 | 0 | cm = alg_unquant(X, N, K, spread, B, ec, gain |
1119 | 0 | ARG_QEXT(ctx->ext_ec) ARG_QEXT(extra_bits)); |
1120 | 0 | } |
1121 | | #ifdef ENABLE_QEXT |
1122 | | } else if (ext_b > 2*N<<BITRES) |
1123 | | { |
1124 | | extra_bits = ext_b/(N-1)>>BITRES; |
1125 | | ext_remaining_bits = ctx->ext_total_bits-ec_tell_frac(ctx->ext_ec); |
1126 | | if (ext_remaining_bits < ((extra_bits+1)*(N-1)+N)<<BITRES) { |
1127 | | extra_bits = (ext_remaining_bits-(N<<BITRES))/(N-1)>>BITRES; |
1128 | | extra_bits = IMAX(extra_bits-1, 0); |
1129 | | } |
1130 | | extra_bits = IMIN(14, extra_bits); |
1131 | | if (encode) cm = cubic_quant(X, N, extra_bits, B, ctx->ext_ec, gain, ctx->resynth); |
1132 | | else cm = cubic_unquant(X, N, extra_bits, B, ctx->ext_ec, gain); |
1133 | | #endif |
1134 | 700M | } else { |
1135 | | /* If there's no pulse, fill the band anyway */ |
1136 | 700M | int j; |
1137 | 700M | if (ctx->resynth) |
1138 | 150M | { |
1139 | 150M | unsigned cm_mask; |
1140 | | /* B can be as large as 16, so this shift might overflow an int on a |
1141 | | 16-bit platform; use a long to get defined behavior.*/ |
1142 | 150M | cm_mask = (unsigned)(1UL<<B)-1; |
1143 | 150M | fill &= cm_mask; |
1144 | 150M | if (!fill) |
1145 | 64.4M | { |
1146 | 64.4M | OPUS_CLEAR(X, N); |
1147 | 86.1M | } else { |
1148 | 86.1M | if (lowband == NULL) |
1149 | 3.75M | { |
1150 | | /* Noise */ |
1151 | 25.3M | for (j=0;j<N;j++) |
1152 | 21.6M | { |
1153 | 21.6M | ctx->seed = celt_lcg_rand(ctx->seed); |
1154 | 21.6M | X[j] = SHL32((celt_norm)((opus_int32)ctx->seed>>20), NORM_SHIFT-14); |
1155 | 21.6M | } |
1156 | 3.75M | cm = cm_mask; |
1157 | 82.4M | } else { |
1158 | | /* Folded spectrum */ |
1159 | 918M | for (j=0;j<N;j++) |
1160 | 835M | { |
1161 | 835M | opus_val16 tmp; |
1162 | 835M | ctx->seed = celt_lcg_rand(ctx->seed); |
1163 | | /* About 48 dB below the "normal" folding level */ |
1164 | 835M | tmp = QCONST16(1.0f/256, NORM_SHIFT-4); |
1165 | 835M | tmp = (ctx->seed)&0x8000 ? tmp : -tmp; |
1166 | 835M | X[j] = lowband[j]+tmp; |
1167 | 835M | } |
1168 | 82.4M | cm = fill; |
1169 | 82.4M | } |
1170 | 86.1M | renormalise_vector(X, N, gain, ctx->arch); |
1171 | 86.1M | } |
1172 | 150M | } |
1173 | 700M | } |
1174 | 734M | } |
1175 | | |
1176 | 756M | return cm; |
1177 | 756M | } |
1178 | | |
1179 | | #ifdef ENABLE_QEXT |
1180 | | static unsigned cubic_quant_partition(struct band_ctx *ctx, celt_norm *X, int N, int b, int B, ec_ctx *ec, int LM, opus_val32 gain, int resynth, int encode) |
1181 | | { |
1182 | | celt_assert(LM>=0); |
1183 | | ctx->remaining_bits = ctx->ec->storage*8*8 - ec_tell_frac(ctx->ec); |
1184 | | b = IMIN(b, ctx->remaining_bits); |
1185 | | /* As long as we have at least two bits of depth, split all the way to LM=0 (not -1 like PVQ). */ |
1186 | | if (LM==0 || b<=2*N<<BITRES) { |
1187 | | int res, ret; |
1188 | | b = IMIN(b + ((N-1)<<BITRES)/2, ctx->remaining_bits); |
1189 | | /* Resolution left after taking into account coding the cube face. */ |
1190 | | res = (b-(1<<BITRES)-ctx->m->logN[ctx->i]-(LM<<BITRES)-1)/(N-1)>>BITRES; |
1191 | | res = IMIN(14, IMAX(0, res)); |
1192 | | if (encode) ret = cubic_quant(X, N, res, B, ec, gain, resynth); |
1193 | | else ret = cubic_unquant(X, N, res, B, ec, gain); |
1194 | | ctx->remaining_bits = ctx->ec->storage*8*8 - ec_tell_frac(ctx->ec); |
1195 | | return ret; |
1196 | | } else { |
1197 | | celt_norm *Y; |
1198 | | opus_int32 itheta_q30; |
1199 | | opus_val32 g1, g2; |
1200 | | opus_int32 theta_res; |
1201 | | opus_int32 qtheta; |
1202 | | int delta; |
1203 | | int b1, b2; |
1204 | | int cm; |
1205 | | int N0; |
1206 | | N0 = N; |
1207 | | N >>= 1; |
1208 | | Y = X+N; |
1209 | | LM -= 1; |
1210 | | B = (B+1)>>1; |
1211 | | theta_res = IMIN(16, (b>>BITRES)/(N0-1) + 1); |
1212 | | if (encode) { |
1213 | | itheta_q30 = stereo_itheta(X, Y, 0, N, ctx->arch); |
1214 | | qtheta = (itheta_q30+(1<<(29-theta_res)))>>(30-theta_res); |
1215 | | ec_enc_uint(ec, qtheta, (1<<theta_res)+1); |
1216 | | } else { |
1217 | | qtheta = ec_dec_uint(ec, (1<<theta_res)+1); |
1218 | | } |
1219 | | itheta_q30 = qtheta<<(30-theta_res); |
1220 | | b -= theta_res<<BITRES; |
1221 | | delta = (N0-1) * 23 * ((itheta_q30>>16)-8192) >> (17-BITRES); |
1222 | | |
1223 | | #ifdef FIXED_POINT |
1224 | | g1 = celt_cos_norm32(itheta_q30); |
1225 | | g2 = celt_cos_norm32((1<<30)-itheta_q30); |
1226 | | #else |
1227 | | g1 = celt_cos_norm2(itheta_q30*(1.f/(1<<30))); |
1228 | | g2 = celt_cos_norm2(1.f-itheta_q30*(1.f/(1<<30))); |
1229 | | #endif |
1230 | | if (itheta_q30 == 0) { |
1231 | | b1=b; |
1232 | | b2=0; |
1233 | | } else if (itheta_q30==1073741824) { |
1234 | | b1=0; |
1235 | | b2=b; |
1236 | | } else { |
1237 | | b1 = IMIN(b, IMAX(0, (b-delta)/2)); |
1238 | | b2 = b-b1; |
1239 | | } |
1240 | | cm = cubic_quant_partition(ctx, X, N, b1, B, ec, LM, MULT32_32_Q31(gain, g1), resynth, encode); |
1241 | | cm |= cubic_quant_partition(ctx, Y, N, b2, B, ec, LM, MULT32_32_Q31(gain, g2), resynth, encode); |
1242 | | return cm; |
1243 | | } |
1244 | | } |
1245 | | #endif |
1246 | | |
1247 | | /* This function is responsible for encoding and decoding a band for the mono case. */ |
1248 | | static unsigned quant_band(struct band_ctx *ctx, celt_norm *X, |
1249 | | int N, int b, int B, celt_norm *lowband, |
1250 | | int LM, celt_norm *lowband_out, |
1251 | | opus_val32 gain, celt_norm *lowband_scratch, int fill |
1252 | | ARG_QEXT(int ext_b)) |
1253 | 899M | { |
1254 | 899M | int N0=N; |
1255 | 899M | int N_B=N; |
1256 | 899M | int N_B0; |
1257 | 899M | int B0=B; |
1258 | 899M | int time_divide=0; |
1259 | 899M | int recombine=0; |
1260 | 899M | int longBlocks; |
1261 | 899M | unsigned cm=0; |
1262 | 899M | int k; |
1263 | 899M | int encode; |
1264 | 899M | int tf_change; |
1265 | | |
1266 | 899M | encode = ctx->encode; |
1267 | 899M | tf_change = ctx->tf_change; |
1268 | | |
1269 | 899M | longBlocks = B0==1; |
1270 | | |
1271 | 899M | N_B = celt_udiv(N_B, B); |
1272 | | |
1273 | | /* Special case for one sample */ |
1274 | 899M | if (N==1) |
1275 | 187M | { |
1276 | 187M | return quant_band_n1(ctx, X, NULL, lowband_out); |
1277 | 187M | } |
1278 | | |
1279 | 711M | if (tf_change>0) |
1280 | 1.46M | recombine = tf_change; |
1281 | | /* Band recombining to increase frequency resolution */ |
1282 | | |
1283 | 711M | if (lowband_scratch && lowband && (recombine || ((N_B&1) == 0 && tf_change<0) || B0>1)) |
1284 | 829k | { |
1285 | 829k | OPUS_COPY(lowband_scratch, lowband, N); |
1286 | 829k | lowband = lowband_scratch; |
1287 | 829k | } |
1288 | | |
1289 | 714M | for (k=0;k<recombine;k++) |
1290 | 2.80M | { |
1291 | 2.80M | static const unsigned char bit_interleave_table[16]={ |
1292 | 2.80M | 0,1,1,1,2,3,3,3,2,3,3,3,2,3,3,3 |
1293 | 2.80M | }; |
1294 | 2.80M | if (encode) |
1295 | 2.80M | haar1(X, N>>k, 1<<k); |
1296 | 2.80M | if (lowband) |
1297 | 282k | haar1(lowband, N>>k, 1<<k); |
1298 | 2.80M | fill = bit_interleave_table[fill&0xF]|bit_interleave_table[fill>>4]<<2; |
1299 | 2.80M | } |
1300 | 711M | B>>=recombine; |
1301 | 711M | N_B<<=recombine; |
1302 | | |
1303 | | /* Increasing the time resolution */ |
1304 | 716M | while ((N_B&1) == 0 && tf_change<0) |
1305 | 4.74M | { |
1306 | 4.74M | if (encode) |
1307 | 4.74M | haar1(X, N_B, B); |
1308 | 4.74M | if (lowband) |
1309 | 703k | haar1(lowband, N_B, B); |
1310 | 4.74M | fill |= fill<<B; |
1311 | 4.74M | B <<= 1; |
1312 | 4.74M | N_B >>= 1; |
1313 | 4.74M | time_divide++; |
1314 | 4.74M | tf_change++; |
1315 | 4.74M | } |
1316 | 711M | B0=B; |
1317 | 711M | N_B0 = N_B; |
1318 | | |
1319 | | /* Reorganize the samples in time order instead of frequency order */ |
1320 | 711M | if (B0>1) |
1321 | 7.36M | { |
1322 | 7.36M | if (encode) |
1323 | 7.36M | deinterleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); |
1324 | 7.36M | if (lowband) |
1325 | 747k | deinterleave_hadamard(lowband, N_B>>recombine, B0<<recombine, longBlocks); |
1326 | 7.36M | } |
1327 | | |
1328 | | #ifdef ENABLE_QEXT |
1329 | | if (ctx->extra_bands && b > (3*N<<BITRES)+(ctx->m->logN[ctx->i]+8+8*LM)) { |
1330 | | cm = cubic_quant_partition(ctx, X, N, b, B, ctx->ec, LM, gain, ctx->resynth, encode); |
1331 | | } else |
1332 | | #endif |
1333 | 711M | { |
1334 | 711M | cm = quant_partition(ctx, X, N, b, B, lowband, LM, gain, fill ARG_QEXT(ext_b)); |
1335 | 711M | } |
1336 | | |
1337 | | /* This code is used by the decoder and by the resynthesis-enabled encoder */ |
1338 | 711M | if (ctx->resynth) |
1339 | 153M | { |
1340 | | /* Undo the sample reorganization going from time order to frequency order */ |
1341 | 153M | if (B0>1) |
1342 | 1.55M | interleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); |
1343 | | |
1344 | | /* Undo time-freq changes that we did earlier */ |
1345 | 153M | N_B = N_B0; |
1346 | 153M | B = B0; |
1347 | 155M | for (k=0;k<time_divide;k++) |
1348 | 1.45M | { |
1349 | 1.45M | B >>= 1; |
1350 | 1.45M | N_B <<= 1; |
1351 | 1.45M | cm |= cm>>B; |
1352 | 1.45M | haar1(X, N_B, B); |
1353 | 1.45M | } |
1354 | | |
1355 | 154M | for (k=0;k<recombine;k++) |
1356 | 607k | { |
1357 | 607k | static const unsigned char bit_deinterleave_table[16]={ |
1358 | 607k | 0x00,0x03,0x0C,0x0F,0x30,0x33,0x3C,0x3F, |
1359 | 607k | 0xC0,0xC3,0xCC,0xCF,0xF0,0xF3,0xFC,0xFF |
1360 | 607k | }; |
1361 | 607k | cm = bit_deinterleave_table[cm]; |
1362 | 607k | haar1(X, N0>>k, 1<<k); |
1363 | 607k | } |
1364 | 153M | B<<=recombine; |
1365 | | |
1366 | | /* Scale output for later folding */ |
1367 | 153M | if (lowband_out) |
1368 | 82.0M | { |
1369 | 82.0M | int j; |
1370 | 82.0M | opus_val16 n; |
1371 | 82.0M | n = celt_sqrt(SHL32(EXTEND32(N0),22)); |
1372 | 800M | for (j=0;j<N0;j++) |
1373 | 718M | lowband_out[j] = MULT16_32_Q15(n,X[j]); |
1374 | 82.0M | } |
1375 | 153M | cm &= (1<<B)-1; |
1376 | 153M | } |
1377 | 711M | return cm; |
1378 | 899M | } |
1379 | | |
1380 | | #ifdef FIXED_POINT |
1381 | | #define MIN_STEREO_ENERGY 2 |
1382 | | #else |
1383 | 388M | #define MIN_STEREO_ENERGY 1e-10f |
1384 | | #endif |
1385 | | |
1386 | | /* This function is responsible for encoding and decoding a band for the stereo case. */ |
1387 | | static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, |
1388 | | int N, int b, int B, celt_norm *lowband, |
1389 | | int LM, celt_norm *lowband_out, |
1390 | | celt_norm *lowband_scratch, int fill |
1391 | | ARG_QEXT(int ext_b) ARG_QEXT(const int *cap)) |
1392 | 264M | { |
1393 | 264M | int imid=0, iside=0; |
1394 | 264M | int inv = 0; |
1395 | 264M | opus_val32 mid=0, side=0; |
1396 | 264M | unsigned cm=0; |
1397 | 264M | int mbits, sbits, delta; |
1398 | 264M | int itheta; |
1399 | 264M | int qalloc; |
1400 | 264M | struct split_ctx sctx; |
1401 | 264M | int orig_fill; |
1402 | 264M | int encode; |
1403 | 264M | ec_ctx *ec; |
1404 | | |
1405 | 264M | encode = ctx->encode; |
1406 | 264M | ec = ctx->ec; |
1407 | | |
1408 | | /* Special case for one sample */ |
1409 | 264M | if (N==1) |
1410 | 72.5M | { |
1411 | 72.5M | return quant_band_n1(ctx, X, Y, lowband_out); |
1412 | 72.5M | } |
1413 | | |
1414 | 191M | orig_fill = fill; |
1415 | | |
1416 | 191M | if (encode) { |
1417 | 191M | if (ctx->bandE[ctx->i] < MIN_STEREO_ENERGY || ctx->bandE[ctx->m->nbEBands+ctx->i] < MIN_STEREO_ENERGY) { |
1418 | 185M | if (ctx->bandE[ctx->i] > ctx->bandE[ctx->m->nbEBands+ctx->i]) OPUS_COPY(Y, X, N); |
1419 | 185M | else OPUS_COPY(X, Y, N); |
1420 | 185M | } |
1421 | 191M | } |
1422 | 191M | compute_theta(ctx, &sctx, X, Y, N, &b, B, B, LM, 1, &fill ARG_QEXT(&ext_b)); |
1423 | 191M | inv = sctx.inv; |
1424 | 191M | imid = sctx.imid; |
1425 | 191M | iside = sctx.iside; |
1426 | 191M | delta = sctx.delta; |
1427 | 191M | itheta = sctx.itheta; |
1428 | 191M | qalloc = sctx.qalloc; |
1429 | | #ifdef FIXED_POINT |
1430 | | # ifdef ENABLE_QEXT |
1431 | | (void)imid; |
1432 | | (void)iside; |
1433 | | mid = celt_cos_norm32(sctx.itheta_q30); |
1434 | | side = celt_cos_norm32((1<<30)-sctx.itheta_q30); |
1435 | | # else |
1436 | | mid = SHL32(EXTEND32(imid), 16); |
1437 | | side = SHL32(EXTEND32(iside), 16); |
1438 | | # endif |
1439 | | #else |
1440 | | # ifdef ENABLE_QEXT |
1441 | | (void)imid; |
1442 | | (void)iside; |
1443 | | mid = celt_cos_norm2(sctx.itheta_q30*(1.f/(1<<30))); |
1444 | | side = celt_cos_norm2(1.f-sctx.itheta_q30*(1.f/(1<<30))); |
1445 | | # else |
1446 | 191M | mid = (1.f/32768)*imid; |
1447 | 191M | side = (1.f/32768)*iside; |
1448 | 191M | # endif |
1449 | 191M | #endif |
1450 | | |
1451 | | /* This is a special case for N=2 that only works for stereo and takes |
1452 | | advantage of the fact that mid and side are orthogonal to encode |
1453 | | the side with just one bit. */ |
1454 | 191M | if (N==2) |
1455 | 62.1M | { |
1456 | 62.1M | int c; |
1457 | 62.1M | int sign=0; |
1458 | 62.1M | celt_norm *x2, *y2; |
1459 | 62.1M | mbits = b; |
1460 | 62.1M | sbits = 0; |
1461 | | /* Only need one bit for the side. */ |
1462 | 62.1M | if (itheta != 0 && itheta != 16384) |
1463 | 278k | sbits = 1<<BITRES; |
1464 | 62.1M | mbits -= sbits; |
1465 | 62.1M | c = itheta > 8192; |
1466 | 62.1M | ctx->remaining_bits -= qalloc+sbits; |
1467 | | |
1468 | 62.1M | x2 = c ? Y : X; |
1469 | 62.1M | y2 = c ? X : Y; |
1470 | 62.1M | if (sbits) |
1471 | 278k | { |
1472 | 278k | if (encode) |
1473 | 278k | { |
1474 | | /* Here we only need to encode a sign for the side. */ |
1475 | | /* FIXME: Need to increase fixed-point precision? */ |
1476 | 278k | sign = MULT32_32_Q31(x2[0],y2[1]) - MULT32_32_Q31(x2[1],y2[0]) < 0; |
1477 | 278k | ec_enc_bits(ec, sign, 1); |
1478 | 278k | } else { |
1479 | 0 | sign = ec_dec_bits(ec, 1); |
1480 | 0 | } |
1481 | 278k | } |
1482 | 62.1M | sign = 1-2*sign; |
1483 | | /* We use orig_fill here because we want to fold the side, but if |
1484 | | itheta==16384, we'll have cleared the low bits of fill. */ |
1485 | 62.1M | cm = quant_band(ctx, x2, N, mbits, B, lowband, LM, lowband_out, Q31ONE, |
1486 | 62.1M | lowband_scratch, orig_fill ARG_QEXT(ext_b)); |
1487 | | /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse), |
1488 | | and there's no need to worry about mixing with the other channel. */ |
1489 | 62.1M | y2[0] = -sign*x2[1]; |
1490 | 62.1M | y2[1] = sign*x2[0]; |
1491 | 62.1M | if (ctx->resynth) |
1492 | 24.1M | { |
1493 | 24.1M | celt_norm tmp; |
1494 | 24.1M | X[0] = MULT32_32_Q31(mid, X[0]); |
1495 | 24.1M | X[1] = MULT32_32_Q31(mid, X[1]); |
1496 | 24.1M | Y[0] = MULT32_32_Q31(side, Y[0]); |
1497 | 24.1M | Y[1] = MULT32_32_Q31(side, Y[1]); |
1498 | 24.1M | tmp = X[0]; |
1499 | 24.1M | X[0] = SUB32(tmp,Y[0]); |
1500 | 24.1M | Y[0] = ADD32(tmp,Y[0]); |
1501 | 24.1M | tmp = X[1]; |
1502 | 24.1M | X[1] = SUB32(tmp,Y[1]); |
1503 | 24.1M | Y[1] = ADD32(tmp,Y[1]); |
1504 | 24.1M | } |
1505 | 129M | } else { |
1506 | | /* "Normal" split code */ |
1507 | 129M | opus_int32 rebalance; |
1508 | | |
1509 | 129M | mbits = IMAX(0, IMIN(b, (b-delta)/2)); |
1510 | 129M | sbits = b-mbits; |
1511 | 129M | ctx->remaining_bits -= qalloc; |
1512 | | |
1513 | 129M | rebalance = ctx->remaining_bits; |
1514 | 129M | if (mbits >= sbits) |
1515 | 128M | { |
1516 | | #ifdef ENABLE_QEXT |
1517 | | int qext_extra = 0; |
1518 | | /* Reallocate any mid bits that cannot be used to extra mid bits. */ |
1519 | | if (cap != NULL && ext_b != 0) qext_extra = IMAX(0, IMIN(ext_b/2, mbits - cap[ctx->i]/2)); |
1520 | | #endif |
1521 | | /* In stereo mode, we do not apply a scaling to the mid because we need the normalized |
1522 | | mid for folding later. */ |
1523 | 128M | cm = quant_band(ctx, X, N, mbits, B, lowband, LM, lowband_out, Q31ONE, |
1524 | 128M | lowband_scratch, fill ARG_QEXT(ext_b/2+qext_extra)); |
1525 | 128M | rebalance = mbits - (rebalance-ctx->remaining_bits); |
1526 | 128M | if (rebalance > 3<<BITRES && itheta!=0) |
1527 | 46.2k | sbits += rebalance - (3<<BITRES); |
1528 | | #ifdef ENABLE_QEXT |
1529 | | /* Guard against overflowing the EC with the angle if the cubic quant used too many bits for the mid. */ |
1530 | | if (ctx->extra_bands) sbits = IMIN(sbits, ctx->remaining_bits); |
1531 | | #endif |
1532 | | /* For a stereo split, the high bits of fill are always zero, so no |
1533 | | folding will be done to the side. */ |
1534 | 128M | cm |= quant_band(ctx, Y, N, sbits, B, NULL, LM, NULL, side, NULL, fill>>B ARG_QEXT(ext_b/2-qext_extra)); |
1535 | 128M | } else { |
1536 | | #ifdef ENABLE_QEXT |
1537 | | int qext_extra = 0; |
1538 | | /* Reallocate any side bits that cannot be used to extra side bits. */ |
1539 | | if (cap != NULL && ext_b != 0) qext_extra = IMAX(0, IMIN(ext_b/2, sbits - cap[ctx->i]/2)); |
1540 | | #endif |
1541 | | /* For a stereo split, the high bits of fill are always zero, so no |
1542 | | folding will be done to the side. */ |
1543 | 344k | cm = quant_band(ctx, Y, N, sbits, B, NULL, LM, NULL, side, NULL, fill>>B ARG_QEXT(ext_b/2+qext_extra)); |
1544 | 344k | rebalance = sbits - (rebalance-ctx->remaining_bits); |
1545 | 344k | if (rebalance > 3<<BITRES && itheta!=16384) |
1546 | 10.2k | mbits += rebalance - (3<<BITRES); |
1547 | | #ifdef ENABLE_QEXT |
1548 | | /* Guard against overflowing the EC with the angle if the cubic quant used too many bits for the side. */ |
1549 | | if (ctx->extra_bands) mbits = IMIN(mbits, ctx->remaining_bits); |
1550 | | #endif |
1551 | | /* In stereo mode, we do not apply a scaling to the mid because we need the normalized |
1552 | | mid for folding later. */ |
1553 | 344k | cm |= quant_band(ctx, X, N, mbits, B, lowband, LM, lowband_out, Q31ONE, |
1554 | 344k | lowband_scratch, fill ARG_QEXT(ext_b/2-qext_extra)); |
1555 | 344k | } |
1556 | 129M | } |
1557 | | |
1558 | | |
1559 | | /* This code is used by the decoder and by the resynthesis-enabled encoder */ |
1560 | 191M | if (ctx->resynth) |
1561 | 89.0M | { |
1562 | 89.0M | if (N!=2) |
1563 | 64.8M | stereo_merge(X, Y, mid, N, ctx->arch); |
1564 | 89.0M | if (inv) |
1565 | 99.5k | { |
1566 | 99.5k | int j; |
1567 | 1.44M | for (j=0;j<N;j++) |
1568 | 1.34M | Y[j] = -Y[j]; |
1569 | 99.5k | } |
1570 | 89.0M | } |
1571 | 191M | return cm; |
1572 | 264M | } |
1573 | | |
1574 | | #ifndef DISABLE_UPDATE_DRAFT |
1575 | | static void special_hybrid_folding(const CELTMode *m, celt_norm *norm, celt_norm *norm2, int start, int M, int dual_stereo) |
1576 | 55.3M | { |
1577 | 55.3M | int n1, n2; |
1578 | 55.3M | const opus_int16 * OPUS_RESTRICT eBands = m->eBands; |
1579 | 55.3M | n1 = M*(eBands[start+1]-eBands[start]); |
1580 | 55.3M | n2 = M*(eBands[start+2]-eBands[start+1]); |
1581 | | /* Duplicate enough of the first band folding data to be able to fold the second band. |
1582 | | Copies no data for CELT-only mode. */ |
1583 | 55.3M | OPUS_COPY(&norm[n1], &norm[2*n1 - n2], n2-n1); |
1584 | 55.3M | if (dual_stereo) |
1585 | 41.3k | OPUS_COPY(&norm2[n1], &norm2[2*n1 - n2], n2-n1); |
1586 | 55.3M | } |
1587 | | #endif |
1588 | | |
1589 | | void quant_all_bands(int encode, const CELTMode *m, int start, int end, |
1590 | | celt_norm *X_, celt_norm *Y_, unsigned char *collapse_masks, |
1591 | | const celt_ener *bandE, int *pulses, int shortBlocks, int spread, |
1592 | | int dual_stereo, int intensity, int *tf_res, opus_int32 total_bits, |
1593 | | opus_int32 balance, ec_ctx *ec, int LM, int codedBands, |
1594 | | opus_uint32 *seed, int complexity, int arch, int disable_inv |
1595 | | ARG_QEXT(ec_ctx *ext_ec) ARG_QEXT(int *extra_pulses) |
1596 | | ARG_QEXT(opus_int32 ext_total_bits) ARG_QEXT(const int *cap)) |
1597 | 55.1M | { |
1598 | 55.1M | int i; |
1599 | 55.1M | opus_int32 remaining_bits; |
1600 | 55.1M | const opus_int16 * OPUS_RESTRICT eBands = m->eBands; |
1601 | 55.1M | celt_norm * OPUS_RESTRICT norm, * OPUS_RESTRICT norm2; |
1602 | 55.1M | VARDECL(celt_norm, _norm); |
1603 | 55.1M | VARDECL(celt_norm, _lowband_scratch); |
1604 | 55.1M | VARDECL(celt_norm, X_save); |
1605 | 55.1M | VARDECL(celt_norm, Y_save); |
1606 | 55.1M | VARDECL(celt_norm, X_save2); |
1607 | 55.1M | VARDECL(celt_norm, Y_save2); |
1608 | 55.1M | VARDECL(celt_norm, norm_save2); |
1609 | 55.1M | int resynth_alloc; |
1610 | 55.1M | celt_norm *lowband_scratch; |
1611 | 55.1M | int B; |
1612 | 55.1M | int M; |
1613 | 55.1M | int lowband_offset; |
1614 | 55.1M | int update_lowband = 1; |
1615 | 55.1M | int C = Y_ != NULL ? 2 : 1; |
1616 | 55.1M | int norm_offset; |
1617 | 55.1M | int theta_rdo = encode && Y_!=NULL && !dual_stereo && complexity>=8; |
1618 | | #ifdef RESYNTH |
1619 | | int resynth = 1; |
1620 | | #else |
1621 | 55.1M | int resynth = !encode || theta_rdo; |
1622 | 55.1M | #endif |
1623 | 55.1M | struct band_ctx ctx; |
1624 | | #ifdef ENABLE_QEXT |
1625 | | int ext_b; |
1626 | | opus_int32 ext_balance=0; |
1627 | | opus_int32 ext_tell=0; |
1628 | | #endif |
1629 | 55.1M | SAVE_STACK; |
1630 | | |
1631 | 55.1M | M = 1<<LM; |
1632 | 55.1M | B = shortBlocks ? M : 1; |
1633 | 55.1M | norm_offset = M*eBands[start]; |
1634 | | /* No need to allocate norm for the last band because we don't need an |
1635 | | output in that band. */ |
1636 | 55.1M | ALLOC(_norm, C*(M*eBands[m->nbEBands-1]-norm_offset), celt_norm); |
1637 | 55.1M | norm = _norm; |
1638 | 55.1M | norm2 = norm + M*eBands[m->nbEBands-1]-norm_offset; |
1639 | | |
1640 | | /* For decoding, we can use the last band as scratch space because we don't need that |
1641 | | scratch space for the last band and we don't care about the data there until we're |
1642 | | decoding the last band. */ |
1643 | 55.1M | if (encode && resynth) |
1644 | 6.93M | resynth_alloc = M*(eBands[m->nbEBands]-eBands[m->nbEBands-1]); |
1645 | 48.2M | else |
1646 | 48.2M | resynth_alloc = ALLOC_NONE; |
1647 | 55.1M | ALLOC(_lowband_scratch, resynth_alloc, celt_norm); |
1648 | 55.1M | if (encode && resynth) |
1649 | 6.93M | lowband_scratch = _lowband_scratch; |
1650 | 48.2M | else |
1651 | 48.2M | lowband_scratch = X_+M*eBands[m->effEBands-1]; |
1652 | 55.1M | ALLOC(X_save, resynth_alloc, celt_norm); |
1653 | 55.1M | ALLOC(Y_save, resynth_alloc, celt_norm); |
1654 | 55.1M | ALLOC(X_save2, resynth_alloc, celt_norm); |
1655 | 55.1M | ALLOC(Y_save2, resynth_alloc, celt_norm); |
1656 | 55.1M | ALLOC(norm_save2, resynth_alloc, celt_norm); |
1657 | | |
1658 | 55.1M | lowband_offset = 0; |
1659 | 55.1M | ctx.bandE = bandE; |
1660 | 55.1M | ctx.ec = ec; |
1661 | 55.1M | ctx.encode = encode; |
1662 | 55.1M | ctx.intensity = intensity; |
1663 | 55.1M | ctx.m = m; |
1664 | 55.1M | ctx.seed = *seed; |
1665 | 55.1M | ctx.spread = spread; |
1666 | 55.1M | ctx.arch = arch; |
1667 | 55.1M | ctx.disable_inv = disable_inv; |
1668 | 55.1M | ctx.resynth = resynth; |
1669 | 55.1M | ctx.theta_round = 0; |
1670 | | #ifdef ENABLE_QEXT |
1671 | | ctx.ext_ec = ext_ec; |
1672 | | ctx.ext_total_bits = ext_total_bits; |
1673 | | ctx.extra_bands = end == NB_QEXT_BANDS || end == 2; |
1674 | | if (ctx.extra_bands) theta_rdo = 0; |
1675 | | #endif |
1676 | | /* Avoid injecting noise in the first band on transients. */ |
1677 | 55.1M | ctx.avoid_split_noise = B > 1; |
1678 | 895M | for (i=start;i<end;i++) |
1679 | 840M | { |
1680 | 840M | opus_int32 tell; |
1681 | 840M | int b; |
1682 | 840M | int N; |
1683 | 840M | opus_int32 curr_balance; |
1684 | 840M | int effective_lowband=-1; |
1685 | 840M | celt_norm * OPUS_RESTRICT X, * OPUS_RESTRICT Y; |
1686 | 840M | int tf_change=0; |
1687 | 840M | unsigned x_cm; |
1688 | 840M | unsigned y_cm; |
1689 | 840M | int last; |
1690 | | |
1691 | 840M | ctx.i = i; |
1692 | 840M | last = (i==end-1); |
1693 | | |
1694 | 840M | X = X_+M*eBands[i]; |
1695 | 840M | if (Y_!=NULL) |
1696 | 262M | Y = Y_+M*eBands[i]; |
1697 | 577M | else |
1698 | 577M | Y = NULL; |
1699 | 840M | N = M*eBands[i+1]-M*eBands[i]; |
1700 | 840M | celt_assert(N > 0); |
1701 | 840M | tell = ec_tell_frac(ec); |
1702 | | |
1703 | | /* Compute how many bits we want to allocate to this band */ |
1704 | 840M | if (i != start) |
1705 | 784M | balance -= tell; |
1706 | 840M | remaining_bits = total_bits-tell-1; |
1707 | 840M | ctx.remaining_bits = remaining_bits; |
1708 | | #ifdef ENABLE_QEXT |
1709 | | if (i != start) { |
1710 | | ext_balance += extra_pulses[i-1] + ext_tell; |
1711 | | } |
1712 | | ext_tell = ec_tell_frac(ext_ec); |
1713 | | ctx.extra_bits = extra_pulses[i]; |
1714 | | if (i != start) |
1715 | | ext_balance -= ext_tell; |
1716 | | if (i <= codedBands-1) |
1717 | | { |
1718 | | opus_int32 ext_curr_balance = celt_sudiv(ext_balance, IMIN(3, codedBands-i)); |
1719 | | ext_b = IMAX(0, IMIN(16383, IMIN(ext_total_bits-ext_tell,extra_pulses[i]+ext_curr_balance))); |
1720 | | } else { |
1721 | | ext_b = 0; |
1722 | | } |
1723 | | #endif |
1724 | 840M | if (i <= codedBands-1) |
1725 | 84.5M | { |
1726 | 84.5M | curr_balance = celt_sudiv(balance, IMIN(3, codedBands-i)); |
1727 | 84.5M | b = IMAX(0, IMIN(16383, IMIN(remaining_bits+1,pulses[i]+curr_balance))); |
1728 | 755M | } else { |
1729 | 755M | b = 0; |
1730 | 755M | } |
1731 | | |
1732 | 840M | #ifndef DISABLE_UPDATE_DRAFT |
1733 | 840M | if (resynth && (M*eBands[i]-N >= M*eBands[start] || i==start+1) && (update_lowband || lowband_offset==0)) |
1734 | 9.02M | lowband_offset = i; |
1735 | 840M | if (i == start+1) |
1736 | 55.1M | special_hybrid_folding(m, norm, norm2, start, M, dual_stereo); |
1737 | | #else |
1738 | | if (resynth && M*eBands[i]-N >= M*eBands[start] && (update_lowband || lowband_offset==0)) |
1739 | | lowband_offset = i; |
1740 | | #endif |
1741 | | |
1742 | 840M | tf_change = tf_res[i]; |
1743 | 840M | ctx.tf_change = tf_change; |
1744 | 840M | if (i>=m->effEBands) |
1745 | 0 | { |
1746 | 0 | X=norm; |
1747 | 0 | if (Y_!=NULL) |
1748 | 0 | Y = norm; |
1749 | 0 | lowband_scratch = NULL; |
1750 | 0 | } |
1751 | 840M | if (last && !theta_rdo) |
1752 | 48.2M | lowband_scratch = NULL; |
1753 | | |
1754 | | /* Get a conservative estimate of the collapse_mask's for the bands we're |
1755 | | going to be folding from. */ |
1756 | 840M | if (lowband_offset != 0 && (spread!=SPREAD_AGGRESSIVE || B>1 || tf_change<0)) |
1757 | 105M | { |
1758 | 105M | int fold_start; |
1759 | 105M | int fold_end; |
1760 | 105M | int fold_i; |
1761 | | /* This ensures we never repeat spectral content within one band */ |
1762 | 105M | effective_lowband = IMAX(0, M*eBands[lowband_offset]-norm_offset-N); |
1763 | 105M | fold_start = lowband_offset; |
1764 | 105M | while(M*eBands[--fold_start] > effective_lowband+norm_offset); |
1765 | 105M | fold_end = lowband_offset-1; |
1766 | 105M | #ifndef DISABLE_UPDATE_DRAFT |
1767 | 283M | while(++fold_end < i && M*eBands[fold_end] < effective_lowband+norm_offset+N); |
1768 | | #else |
1769 | | while(M*eBands[++fold_end] < effective_lowband+norm_offset+N); |
1770 | | #endif |
1771 | 105M | x_cm = y_cm = 0; |
1772 | 284M | fold_i = fold_start; do { |
1773 | 284M | x_cm |= collapse_masks[fold_i*C+0]; |
1774 | 284M | y_cm |= collapse_masks[fold_i*C+C-1]; |
1775 | 284M | } while (++fold_i<fold_end); |
1776 | 105M | } |
1777 | | /* Otherwise, we'll be using the LCG to fold, so all blocks will (almost |
1778 | | always) be non-zero. */ |
1779 | 734M | else |
1780 | 734M | x_cm = y_cm = (1<<B)-1; |
1781 | | |
1782 | 840M | if (dual_stereo && i==intensity) |
1783 | 26.9k | { |
1784 | 26.9k | int j; |
1785 | | |
1786 | | /* Switch off dual stereo to do intensity. */ |
1787 | 26.9k | dual_stereo = 0; |
1788 | 26.9k | if (resynth) |
1789 | 0 | for (j=0;j<M*eBands[i]-norm_offset;j++) |
1790 | 0 | norm[j] = HALF32(norm[j]+norm2[j]); |
1791 | 26.9k | } |
1792 | 840M | if (dual_stereo) |
1793 | 543k | { |
1794 | 543k | x_cm = quant_band(&ctx, X, N, b/2, B, |
1795 | 543k | effective_lowband != -1 ? norm+effective_lowband : NULL, LM, |
1796 | 543k | last?NULL:norm+M*eBands[i]-norm_offset, Q31ONE, lowband_scratch, x_cm ARG_QEXT(ext_b/2)); |
1797 | 543k | y_cm = quant_band(&ctx, Y, N, b/2, B, |
1798 | 543k | effective_lowband != -1 ? norm2+effective_lowband : NULL, LM, |
1799 | 543k | last?NULL:norm2+M*eBands[i]-norm_offset, Q31ONE, lowband_scratch, y_cm ARG_QEXT(ext_b/2)); |
1800 | 839M | } else { |
1801 | 839M | if (Y!=NULL) |
1802 | 261M | { |
1803 | 261M | if (theta_rdo && i < intensity) |
1804 | 2.15M | { |
1805 | 2.15M | ec_ctx ec_save, ec_save2; |
1806 | 2.15M | struct band_ctx ctx_save, ctx_save2; |
1807 | 2.15M | opus_val32 dist0, dist1; |
1808 | 2.15M | unsigned cm, cm2; |
1809 | 2.15M | int nstart_bytes, nend_bytes, save_bytes; |
1810 | 2.15M | unsigned char *bytes_buf; |
1811 | 2.15M | unsigned char bytes_save[1275]; |
1812 | | #ifdef ENABLE_QEXT |
1813 | | ec_ctx ext_ec_save, ext_ec_save2; |
1814 | | unsigned char *ext_bytes_buf; |
1815 | | int ext_nstart_bytes, ext_nend_bytes, ext_save_bytes; |
1816 | | unsigned char ext_bytes_save[QEXT_PACKET_SIZE_CAP]; |
1817 | | #endif |
1818 | 2.15M | opus_val16 w[2]; |
1819 | 2.15M | compute_channel_weights(bandE[i], bandE[i+m->nbEBands], w); |
1820 | | /* Make a copy. */ |
1821 | 2.15M | cm = x_cm|y_cm; |
1822 | 2.15M | ec_save = *ec; |
1823 | | #ifdef ENABLE_QEXT |
1824 | | ext_ec_save = *ext_ec; |
1825 | | #endif |
1826 | 2.15M | ctx_save = ctx; |
1827 | 2.15M | OPUS_COPY(X_save, X, N); |
1828 | 2.15M | OPUS_COPY(Y_save, Y, N); |
1829 | | /* Encode and round down. */ |
1830 | 2.15M | ctx.theta_round = -1; |
1831 | 2.15M | x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, |
1832 | 2.15M | effective_lowband != -1 ? norm+effective_lowband : NULL, LM, |
1833 | 2.15M | last?NULL:norm+M*eBands[i]-norm_offset, lowband_scratch, cm ARG_QEXT(ext_b) ARG_QEXT(cap)); |
1834 | 2.15M | dist0 = MULT16_32_Q15(w[0], celt_inner_prod_norm_shift(X_save, X, N, arch)) + MULT16_32_Q15(w[1], celt_inner_prod_norm_shift(Y_save, Y, N, arch)); |
1835 | | |
1836 | | /* Save first result. */ |
1837 | 2.15M | cm2 = x_cm; |
1838 | 2.15M | ec_save2 = *ec; |
1839 | | #ifdef ENABLE_QEXT |
1840 | | ext_ec_save2 = *ext_ec; |
1841 | | #endif |
1842 | 2.15M | ctx_save2 = ctx; |
1843 | 2.15M | OPUS_COPY(X_save2, X, N); |
1844 | 2.15M | OPUS_COPY(Y_save2, Y, N); |
1845 | 2.15M | if (!last) |
1846 | 2.12M | OPUS_COPY(norm_save2, norm+M*eBands[i]-norm_offset, N); |
1847 | 2.15M | nstart_bytes = ec_save.offs; |
1848 | 2.15M | nend_bytes = ec_save.storage; |
1849 | 2.15M | bytes_buf = ec_save.buf+nstart_bytes; |
1850 | 2.15M | save_bytes = nend_bytes-nstart_bytes; |
1851 | 2.15M | OPUS_COPY(bytes_save, bytes_buf, save_bytes); |
1852 | | #ifdef ENABLE_QEXT |
1853 | | ext_nstart_bytes = ext_ec_save.offs; |
1854 | | ext_nend_bytes = ext_ec_save.storage; |
1855 | | ext_bytes_buf = ext_ec_save.buf!=NULL ? ext_ec_save.buf+ext_nstart_bytes : NULL; |
1856 | | ext_save_bytes = ext_nend_bytes-ext_nstart_bytes; |
1857 | | if (ext_save_bytes) OPUS_COPY(ext_bytes_save, ext_bytes_buf, ext_save_bytes); |
1858 | | #endif |
1859 | | /* Restore */ |
1860 | 2.15M | *ec = ec_save; |
1861 | | #ifdef ENABLE_QEXT |
1862 | | *ext_ec = ext_ec_save; |
1863 | | #endif |
1864 | 2.15M | ctx = ctx_save; |
1865 | 2.15M | OPUS_COPY(X, X_save, N); |
1866 | 2.15M | OPUS_COPY(Y, Y_save, N); |
1867 | 2.15M | #ifndef DISABLE_UPDATE_DRAFT |
1868 | 2.15M | if (i == start+1) |
1869 | 196k | special_hybrid_folding(m, norm, norm2, start, M, dual_stereo); |
1870 | 2.15M | #endif |
1871 | | /* Encode and round up. */ |
1872 | 2.15M | ctx.theta_round = 1; |
1873 | 2.15M | x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, |
1874 | 2.15M | effective_lowband != -1 ? norm+effective_lowband : NULL, LM, |
1875 | 2.15M | last?NULL:norm+M*eBands[i]-norm_offset, lowband_scratch, cm ARG_QEXT(ext_b) ARG_QEXT(cap)); |
1876 | 2.15M | dist1 = MULT16_32_Q15(w[0], celt_inner_prod_norm_shift(X_save, X, N, arch)) + MULT16_32_Q15(w[1], celt_inner_prod_norm_shift(Y_save, Y, N, arch)); |
1877 | 2.15M | if (dist0 >= dist1) { |
1878 | 1.83M | x_cm = cm2; |
1879 | 1.83M | *ec = ec_save2; |
1880 | | #ifdef ENABLE_QEXT |
1881 | | *ext_ec = ext_ec_save2; |
1882 | | #endif |
1883 | 1.83M | ctx = ctx_save2; |
1884 | 1.83M | OPUS_COPY(X, X_save2, N); |
1885 | 1.83M | OPUS_COPY(Y, Y_save2, N); |
1886 | 1.83M | if (!last) |
1887 | 1.82M | OPUS_COPY(norm+M*eBands[i]-norm_offset, norm_save2, N); |
1888 | 1.83M | OPUS_COPY(bytes_buf, bytes_save, save_bytes); |
1889 | | #ifdef ENABLE_QEXT |
1890 | | if (ext_save_bytes) OPUS_COPY(ext_bytes_buf, ext_bytes_save, ext_save_bytes); |
1891 | | #endif |
1892 | 1.83M | } |
1893 | 259M | } else { |
1894 | 259M | ctx.theta_round = 0; |
1895 | 259M | x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, |
1896 | 259M | effective_lowband != -1 ? norm+effective_lowband : NULL, LM, |
1897 | 259M | last?NULL:norm+M*eBands[i]-norm_offset, lowband_scratch, x_cm|y_cm ARG_QEXT(ext_b) ARG_QEXT(cap)); |
1898 | 259M | } |
1899 | 577M | } else { |
1900 | 577M | x_cm = quant_band(&ctx, X, N, b, B, |
1901 | 577M | effective_lowband != -1 ? norm+effective_lowband : NULL, LM, |
1902 | 577M | last?NULL:norm+M*eBands[i]-norm_offset, Q31ONE, lowband_scratch, x_cm|y_cm ARG_QEXT(ext_b)); |
1903 | 577M | } |
1904 | 839M | y_cm = x_cm; |
1905 | 839M | } |
1906 | 840M | collapse_masks[i*C+0] = (unsigned char)x_cm; |
1907 | 840M | collapse_masks[i*C+C-1] = (unsigned char)y_cm; |
1908 | 840M | balance += pulses[i] + tell; |
1909 | | |
1910 | | /* Update the folding position only as long as we have 1 bit/sample depth. */ |
1911 | 840M | update_lowband = b>(N<<BITRES); |
1912 | | /* We only need to avoid noise on a split for the first band. After that, we |
1913 | | have folding. */ |
1914 | 840M | ctx.avoid_split_noise = 0; |
1915 | 840M | } |
1916 | 55.1M | *seed = ctx.seed; |
1917 | | |
1918 | 55.1M | RESTORE_STACK; |
1919 | 55.1M | } |