1 /*
2 * MPEG-4 Parametric Stereo decoding functions
3 * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 *
21 * Note: Rounding-to-nearest used unless otherwise stated
22 *
23 */
24
25 #include <stdint.h>
26 #include "libavutil/common.h"
27 #include "libavutil/mathematics.h"
28 #include "libavutil/mem_internal.h"
29 #include "avcodec.h"
30 #include "aacps.h"
31 #if USE_FIXED
32 #include "aacps_fixed_tablegen.h"
33 #else
34 #include "libavutil/internal.h"
35 #include "aacps_tablegen.h"
36 #endif /* USE_FIXED */
37
38 static const INTFLOAT g1_Q2[] = {
39 Q31(0.0f), Q31(0.01899487526049f), Q31(0.0f), Q31(-0.07293139167538f),
40 Q31(0.0f), Q31(0.30596630545168f), Q31(0.5f)
41 };
42
ipdopd_reset(int8_t *ipd_hist, int8_t *opd_hist)43 static void ipdopd_reset(int8_t *ipd_hist, int8_t *opd_hist)
44 {
45 int i;
46 for (i = 0; i < PS_MAX_NR_IPDOPD; i++) {
47 opd_hist[i] = 0;
48 ipd_hist[i] = 0;
49 }
50 }
51
52 /** Split one subband into 2 subsubbands with a symmetric real filter.
53 * The filter must have its non-center even coefficients equal to zero. */
hybrid2_re(INTFLOAT (in)[2], INTFLOAT (*out)[32][2], const INTFLOAT filter[7], int len, int reverse)54 static void hybrid2_re(INTFLOAT (*in)[2], INTFLOAT (*out)[32][2],
55 const INTFLOAT filter[7], int len, int reverse)
56 {
57 int i, j;
58 for (i = 0; i < len; i++, in++) {
59 INT64FLOAT re_in = AAC_MUL31(filter[6], in[6][0]); //real inphase
60 INT64FLOAT re_op = 0.0f; //real out of phase
61 INT64FLOAT im_in = AAC_MUL31(filter[6], in[6][1]); //imag inphase
62 INT64FLOAT im_op = 0.0f; //imag out of phase
63 for (j = 0; j < 6; j += 2) {
64 re_op += (INT64FLOAT)filter[j+1] * (in[j+1][0] + in[12-j-1][0]);
65 im_op += (INT64FLOAT)filter[j+1] * (in[j+1][1] + in[12-j-1][1]);
66 }
67
68 #if USE_FIXED
69 re_op = (re_op + 0x40000000) >> 31;
70 im_op = (im_op + 0x40000000) >> 31;
71 #endif /* USE_FIXED */
72
73 out[ reverse][i][0] = (INTFLOAT)(re_in + re_op);
74 out[ reverse][i][1] = (INTFLOAT)(im_in + im_op);
75 out[!reverse][i][0] = (INTFLOAT)(re_in - re_op);
76 out[!reverse][i][1] = (INTFLOAT)(im_in - im_op);
77 }
78 }
79
80 /** Split one subband into 6 subsubbands with a complex filter */
hybrid6_cx(PSDSPContext *dsp, INTFLOAT (*in)[2], INTFLOAT (*out)[32][2], TABLE_CONST INTFLOAT (*filter)[8][2], int len)81 static void hybrid6_cx(PSDSPContext *dsp, INTFLOAT (*in)[2], INTFLOAT (*out)[32][2],
82 TABLE_CONST INTFLOAT (*filter)[8][2], int len)
83 {
84 int i;
85 int N = 8;
86 LOCAL_ALIGNED_16(INTFLOAT, temp, [8], [2]);
87
88 for (i = 0; i < len; i++, in++) {
89 dsp->hybrid_analysis(temp, in, (const INTFLOAT (*)[8][2]) filter, 1, N);
90 out[0][i][0] = temp[6][0];
91 out[0][i][1] = temp[6][1];
92 out[1][i][0] = temp[7][0];
93 out[1][i][1] = temp[7][1];
94 out[2][i][0] = temp[0][0];
95 out[2][i][1] = temp[0][1];
96 out[3][i][0] = temp[1][0];
97 out[3][i][1] = temp[1][1];
98 out[4][i][0] = temp[2][0] + temp[5][0];
99 out[4][i][1] = temp[2][1] + temp[5][1];
100 out[5][i][0] = temp[3][0] + temp[4][0];
101 out[5][i][1] = temp[3][1] + temp[4][1];
102 }
103 }
104
hybrid4_8_12_cx(PSDSPContext *dsp, INTFLOAT (*in)[2], INTFLOAT (*out)[32][2], TABLE_CONST INTFLOAT (*filter)[8][2], int N, int len)105 static void hybrid4_8_12_cx(PSDSPContext *dsp,
106 INTFLOAT (*in)[2], INTFLOAT (*out)[32][2],
107 TABLE_CONST INTFLOAT (*filter)[8][2], int N, int len)
108 {
109 int i;
110
111 for (i = 0; i < len; i++, in++) {
112 dsp->hybrid_analysis(out[0] + i, in, (const INTFLOAT (*)[8][2]) filter, 32, N);
113 }
114 }
115
hybrid_analysis(PSDSPContext *dsp, INTFLOAT out[91][32][2], INTFLOAT in[5][44][2], INTFLOAT L[2][38][64], int is34, int len)116 static void hybrid_analysis(PSDSPContext *dsp, INTFLOAT out[91][32][2],
117 INTFLOAT in[5][44][2], INTFLOAT L[2][38][64],
118 int is34, int len)
119 {
120 int i, j;
121 for (i = 0; i < 5; i++) {
122 for (j = 0; j < 38; j++) {
123 in[i][j+6][0] = L[0][j][i];
124 in[i][j+6][1] = L[1][j][i];
125 }
126 }
127 if (is34) {
128 hybrid4_8_12_cx(dsp, in[0], out, f34_0_12, 12, len);
129 hybrid4_8_12_cx(dsp, in[1], out+12, f34_1_8, 8, len);
130 hybrid4_8_12_cx(dsp, in[2], out+20, f34_2_4, 4, len);
131 hybrid4_8_12_cx(dsp, in[3], out+24, f34_2_4, 4, len);
132 hybrid4_8_12_cx(dsp, in[4], out+28, f34_2_4, 4, len);
133 dsp->hybrid_analysis_ileave(out + 27, L, 5, len);
134 } else {
135 hybrid6_cx(dsp, in[0], out, f20_0_8, len);
136 hybrid2_re(in[1], out+6, g1_Q2, len, 1);
137 hybrid2_re(in[2], out+8, g1_Q2, len, 0);
138 dsp->hybrid_analysis_ileave(out + 7, L, 3, len);
139 }
140 //update in_buf
141 for (i = 0; i < 5; i++) {
142 memcpy(in[i], in[i]+32, 6 * sizeof(in[i][0]));
143 }
144 }
145
hybrid_synthesis(PSDSPContext *dsp, INTFLOAT out[2][38][64], INTFLOAT in[91][32][2], int is34, int len)146 static void hybrid_synthesis(PSDSPContext *dsp, INTFLOAT out[2][38][64],
147 INTFLOAT in[91][32][2], int is34, int len)
148 {
149 int i, n;
150 if (is34) {
151 for (n = 0; n < len; n++) {
152 memset(out[0][n], 0, 5*sizeof(out[0][n][0]));
153 memset(out[1][n], 0, 5*sizeof(out[1][n][0]));
154 for (i = 0; i < 12; i++) {
155 out[0][n][0] += (UINTFLOAT)in[ i][n][0];
156 out[1][n][0] += (UINTFLOAT)in[ i][n][1];
157 }
158 for (i = 0; i < 8; i++) {
159 out[0][n][1] += (UINTFLOAT)in[12+i][n][0];
160 out[1][n][1] += (UINTFLOAT)in[12+i][n][1];
161 }
162 for (i = 0; i < 4; i++) {
163 out[0][n][2] += (UINTFLOAT)in[20+i][n][0];
164 out[1][n][2] += (UINTFLOAT)in[20+i][n][1];
165 out[0][n][3] += (UINTFLOAT)in[24+i][n][0];
166 out[1][n][3] += (UINTFLOAT)in[24+i][n][1];
167 out[0][n][4] += (UINTFLOAT)in[28+i][n][0];
168 out[1][n][4] += (UINTFLOAT)in[28+i][n][1];
169 }
170 }
171 dsp->hybrid_synthesis_deint(out, in + 27, 5, len);
172 } else {
173 for (n = 0; n < len; n++) {
174 out[0][n][0] = (UINTFLOAT)in[0][n][0] + in[1][n][0] + in[2][n][0] +
175 (UINTFLOAT)in[3][n][0] + in[4][n][0] + in[5][n][0];
176 out[1][n][0] = (UINTFLOAT)in[0][n][1] + in[1][n][1] + in[2][n][1] +
177 (UINTFLOAT)in[3][n][1] + in[4][n][1] + in[5][n][1];
178 out[0][n][1] = (UINTFLOAT)in[6][n][0] + in[7][n][0];
179 out[1][n][1] = (UINTFLOAT)in[6][n][1] + in[7][n][1];
180 out[0][n][2] = (UINTFLOAT)in[8][n][0] + in[9][n][0];
181 out[1][n][2] = (UINTFLOAT)in[8][n][1] + in[9][n][1];
182 }
183 dsp->hybrid_synthesis_deint(out, in + 7, 3, len);
184 }
185 }
186
187 /// All-pass filter decay slope
188 #define DECAY_SLOPE Q30(0.05f)
189 /// Number of frequency bands that can be addressed by the parameter index, b(k)
190 static const int NR_PAR_BANDS[] = { 20, 34 };
191 static const int NR_IPDOPD_BANDS[] = { 11, 17 };
192 /// Number of frequency bands that can be addressed by the sub subband index, k
193 static const int NR_BANDS[] = { 71, 91 };
194 /// Start frequency band for the all-pass filter decay slope
195 static const int DECAY_CUTOFF[] = { 10, 32 };
196 /// Number of all-pass filer bands
197 static const int NR_ALLPASS_BANDS[] = { 30, 50 };
198 /// First stereo band using the short one sample delay
199 static const int SHORT_DELAY_BAND[] = { 42, 62 };
200
201 /** Table 8.46 */
map_idx_10_to_20(int8_t *par_mapped, const int8_t *par, int full)202 static void map_idx_10_to_20(int8_t *par_mapped, const int8_t *par, int full)
203 {
204 int b;
205 if (full)
206 b = 9;
207 else {
208 b = 4;
209 par_mapped[10] = 0;
210 }
211 for (; b >= 0; b--) {
212 par_mapped[2*b+1] = par_mapped[2*b] = par[b];
213 }
214 }
215
map_idx_34_to_20(int8_t *par_mapped, const int8_t *par, int full)216 static void map_idx_34_to_20(int8_t *par_mapped, const int8_t *par, int full)
217 {
218 par_mapped[ 0] = (2*par[ 0] + par[ 1]) / 3;
219 par_mapped[ 1] = ( par[ 1] + 2*par[ 2]) / 3;
220 par_mapped[ 2] = (2*par[ 3] + par[ 4]) / 3;
221 par_mapped[ 3] = ( par[ 4] + 2*par[ 5]) / 3;
222 par_mapped[ 4] = ( par[ 6] + par[ 7]) / 2;
223 par_mapped[ 5] = ( par[ 8] + par[ 9]) / 2;
224 par_mapped[ 6] = par[10];
225 par_mapped[ 7] = par[11];
226 par_mapped[ 8] = ( par[12] + par[13]) / 2;
227 par_mapped[ 9] = ( par[14] + par[15]) / 2;
228 par_mapped[10] = par[16];
229 if (full) {
230 par_mapped[11] = par[17];
231 par_mapped[12] = par[18];
232 par_mapped[13] = par[19];
233 par_mapped[14] = ( par[20] + par[21]) / 2;
234 par_mapped[15] = ( par[22] + par[23]) / 2;
235 par_mapped[16] = ( par[24] + par[25]) / 2;
236 par_mapped[17] = ( par[26] + par[27]) / 2;
237 par_mapped[18] = ( par[28] + par[29] + par[30] + par[31]) / 4;
238 par_mapped[19] = ( par[32] + par[33]) / 2;
239 }
240 }
241
map_val_34_to_20(INTFLOAT par[PS_MAX_NR_IIDICC])242 static void map_val_34_to_20(INTFLOAT par[PS_MAX_NR_IIDICC])
243 {
244 #if USE_FIXED
245 par[ 0] = (int)(((int64_t)(par[ 0] + (unsigned)(par[ 1]>>1)) * 1431655765 + \
246 0x40000000) >> 31);
247 par[ 1] = (int)(((int64_t)((par[ 1]>>1) + (unsigned)par[ 2]) * 1431655765 + \
248 0x40000000) >> 31);
249 par[ 2] = (int)(((int64_t)(par[ 3] + (unsigned)(par[ 4]>>1)) * 1431655765 + \
250 0x40000000) >> 31);
251 par[ 3] = (int)(((int64_t)((par[ 4]>>1) + (unsigned)par[ 5]) * 1431655765 + \
252 0x40000000) >> 31);
253 #else
254 par[ 0] = (2*par[ 0] + par[ 1]) * 0.33333333f;
255 par[ 1] = ( par[ 1] + 2*par[ 2]) * 0.33333333f;
256 par[ 2] = (2*par[ 3] + par[ 4]) * 0.33333333f;
257 par[ 3] = ( par[ 4] + 2*par[ 5]) * 0.33333333f;
258 #endif /* USE_FIXED */
259 par[ 4] = AAC_HALF_SUM(par[ 6], par[ 7]);
260 par[ 5] = AAC_HALF_SUM(par[ 8], par[ 9]);
261 par[ 6] = par[10];
262 par[ 7] = par[11];
263 par[ 8] = AAC_HALF_SUM(par[12], par[13]);
264 par[ 9] = AAC_HALF_SUM(par[14], par[15]);
265 par[10] = par[16];
266 par[11] = par[17];
267 par[12] = par[18];
268 par[13] = par[19];
269 par[14] = AAC_HALF_SUM(par[20], par[21]);
270 par[15] = AAC_HALF_SUM(par[22], par[23]);
271 par[16] = AAC_HALF_SUM(par[24], par[25]);
272 par[17] = AAC_HALF_SUM(par[26], par[27]);
273 #if USE_FIXED
274 par[18] = (((par[28]+2)>>2) + ((par[29]+2)>>2) + ((par[30]+2)>>2) + ((par[31]+2)>>2));
275 #else
276 par[18] = ( par[28] + par[29] + par[30] + par[31]) * 0.25f;
277 #endif /* USE_FIXED */
278 par[19] = AAC_HALF_SUM(par[32], par[33]);
279 }
280
map_idx_10_to_34(int8_t *par_mapped, const int8_t *par, int full)281 static void map_idx_10_to_34(int8_t *par_mapped, const int8_t *par, int full)
282 {
283 if (full) {
284 par_mapped[33] = par[9];
285 par_mapped[32] = par[9];
286 par_mapped[31] = par[9];
287 par_mapped[30] = par[9];
288 par_mapped[29] = par[9];
289 par_mapped[28] = par[9];
290 par_mapped[27] = par[8];
291 par_mapped[26] = par[8];
292 par_mapped[25] = par[8];
293 par_mapped[24] = par[8];
294 par_mapped[23] = par[7];
295 par_mapped[22] = par[7];
296 par_mapped[21] = par[7];
297 par_mapped[20] = par[7];
298 par_mapped[19] = par[6];
299 par_mapped[18] = par[6];
300 par_mapped[17] = par[5];
301 par_mapped[16] = par[5];
302 } else {
303 par_mapped[16] = 0;
304 }
305 par_mapped[15] = par[4];
306 par_mapped[14] = par[4];
307 par_mapped[13] = par[4];
308 par_mapped[12] = par[4];
309 par_mapped[11] = par[3];
310 par_mapped[10] = par[3];
311 par_mapped[ 9] = par[2];
312 par_mapped[ 8] = par[2];
313 par_mapped[ 7] = par[2];
314 par_mapped[ 6] = par[2];
315 par_mapped[ 5] = par[1];
316 par_mapped[ 4] = par[1];
317 par_mapped[ 3] = par[1];
318 par_mapped[ 2] = par[0];
319 par_mapped[ 1] = par[0];
320 par_mapped[ 0] = par[0];
321 }
322
map_idx_20_to_34(int8_t *par_mapped, const int8_t *par, int full)323 static void map_idx_20_to_34(int8_t *par_mapped, const int8_t *par, int full)
324 {
325 if (full) {
326 par_mapped[33] = par[19];
327 par_mapped[32] = par[19];
328 par_mapped[31] = par[18];
329 par_mapped[30] = par[18];
330 par_mapped[29] = par[18];
331 par_mapped[28] = par[18];
332 par_mapped[27] = par[17];
333 par_mapped[26] = par[17];
334 par_mapped[25] = par[16];
335 par_mapped[24] = par[16];
336 par_mapped[23] = par[15];
337 par_mapped[22] = par[15];
338 par_mapped[21] = par[14];
339 par_mapped[20] = par[14];
340 par_mapped[19] = par[13];
341 par_mapped[18] = par[12];
342 par_mapped[17] = par[11];
343 }
344 par_mapped[16] = par[10];
345 par_mapped[15] = par[ 9];
346 par_mapped[14] = par[ 9];
347 par_mapped[13] = par[ 8];
348 par_mapped[12] = par[ 8];
349 par_mapped[11] = par[ 7];
350 par_mapped[10] = par[ 6];
351 par_mapped[ 9] = par[ 5];
352 par_mapped[ 8] = par[ 5];
353 par_mapped[ 7] = par[ 4];
354 par_mapped[ 6] = par[ 4];
355 par_mapped[ 5] = par[ 3];
356 par_mapped[ 4] = (par[ 2] + par[ 3]) / 2;
357 par_mapped[ 3] = par[ 2];
358 par_mapped[ 2] = par[ 1];
359 par_mapped[ 1] = (par[ 0] + par[ 1]) / 2;
360 par_mapped[ 0] = par[ 0];
361 }
362
map_val_20_to_34(INTFLOAT par[PS_MAX_NR_IIDICC])363 static void map_val_20_to_34(INTFLOAT par[PS_MAX_NR_IIDICC])
364 {
365 par[33] = par[19];
366 par[32] = par[19];
367 par[31] = par[18];
368 par[30] = par[18];
369 par[29] = par[18];
370 par[28] = par[18];
371 par[27] = par[17];
372 par[26] = par[17];
373 par[25] = par[16];
374 par[24] = par[16];
375 par[23] = par[15];
376 par[22] = par[15];
377 par[21] = par[14];
378 par[20] = par[14];
379 par[19] = par[13];
380 par[18] = par[12];
381 par[17] = par[11];
382 par[16] = par[10];
383 par[15] = par[ 9];
384 par[14] = par[ 9];
385 par[13] = par[ 8];
386 par[12] = par[ 8];
387 par[11] = par[ 7];
388 par[10] = par[ 6];
389 par[ 9] = par[ 5];
390 par[ 8] = par[ 5];
391 par[ 7] = par[ 4];
392 par[ 6] = par[ 4];
393 par[ 5] = par[ 3];
394 par[ 4] = AAC_HALF_SUM(par[ 2], par[ 3]);
395 par[ 3] = par[ 2];
396 par[ 2] = par[ 1];
397 par[ 1] = AAC_HALF_SUM(par[ 0], par[ 1]);
398 }
399
decorrelation(PSContext *ps, INTFLOAT (*out)[32][2], const INTFLOAT (*s)[32][2], int is34)400 static void decorrelation(PSContext *ps, INTFLOAT (*out)[32][2], const INTFLOAT (*s)[32][2], int is34)
401 {
402 LOCAL_ALIGNED_16(INTFLOAT, power, [34], [PS_QMF_TIME_SLOTS]);
403 LOCAL_ALIGNED_16(INTFLOAT, transient_gain, [34], [PS_QMF_TIME_SLOTS]);
404 INTFLOAT *peak_decay_nrg = ps->peak_decay_nrg;
405 INTFLOAT *power_smooth = ps->power_smooth;
406 INTFLOAT *peak_decay_diff_smooth = ps->peak_decay_diff_smooth;
407 INTFLOAT (*delay)[PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2] = ps->delay;
408 INTFLOAT (*ap_delay)[PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2] = ps->ap_delay;
409 #if !USE_FIXED
410 const float transient_impact = 1.5f;
411 const float a_smooth = 0.25f; ///< Smoothing coefficient
412 #endif /* USE_FIXED */
413 const int8_t *const k_to_i = is34 ? ff_k_to_i_34 : ff_k_to_i_20;
414 int i, k, m, n;
415 int n0 = 0, nL = 32;
416 const INTFLOAT peak_decay_factor = Q31(0.76592833836465f);
417
418 memset(power, 0, 34 * sizeof(*power));
419
420 if (is34 != ps->common.is34bands_old) {
421 memset(ps->peak_decay_nrg, 0, sizeof(ps->peak_decay_nrg));
422 memset(ps->power_smooth, 0, sizeof(ps->power_smooth));
423 memset(ps->peak_decay_diff_smooth, 0, sizeof(ps->peak_decay_diff_smooth));
424 memset(ps->delay, 0, sizeof(ps->delay));
425 memset(ps->ap_delay, 0, sizeof(ps->ap_delay));
426 }
427
428 for (k = 0; k < NR_BANDS[is34]; k++) {
429 int i = k_to_i[k];
430 ps->dsp.add_squares(power[i], s[k], nL - n0);
431 }
432
433 //Transient detection
434 #if USE_FIXED
435 for (i = 0; i < NR_PAR_BANDS[is34]; i++) {
436 for (n = n0; n < nL; n++) {
437 int decayed_peak;
438 decayed_peak = (int)(((int64_t)peak_decay_factor * \
439 peak_decay_nrg[i] + 0x40000000) >> 31);
440 peak_decay_nrg[i] = FFMAX(decayed_peak, power[i][n]);
441 power_smooth[i] += (power[i][n] + 2LL - power_smooth[i]) >> 2;
442 peak_decay_diff_smooth[i] += (peak_decay_nrg[i] + 2LL - power[i][n] - \
443 peak_decay_diff_smooth[i]) >> 2;
444
445 if (peak_decay_diff_smooth[i]) {
446 transient_gain[i][n] = FFMIN(power_smooth[i]*43691LL / peak_decay_diff_smooth[i], 1<<16);
447 } else
448 transient_gain[i][n] = 1 << 16;
449 }
450 }
451 #else
452 for (i = 0; i < NR_PAR_BANDS[is34]; i++) {
453 for (n = n0; n < nL; n++) {
454 float decayed_peak = peak_decay_factor * peak_decay_nrg[i];
455 float denom;
456 peak_decay_nrg[i] = FFMAX(decayed_peak, power[i][n]);
457 power_smooth[i] += a_smooth * (power[i][n] - power_smooth[i]);
458 peak_decay_diff_smooth[i] += a_smooth * (peak_decay_nrg[i] - power[i][n] - peak_decay_diff_smooth[i]);
459 denom = transient_impact * peak_decay_diff_smooth[i];
460 transient_gain[i][n] = (denom > power_smooth[i]) ?
461 power_smooth[i] / denom : 1.0f;
462 }
463 }
464
465 #endif /* USE_FIXED */
466 //Decorrelation and transient reduction
467 // PS_AP_LINKS - 1
468 // -----
469 // | | Q_fract_allpass[k][m]*z^-link_delay[m] - a[m]*g_decay_slope[k]
470 //H[k][z] = z^-2 * phi_fract[k] * | | ----------------------------------------------------------------
471 // | | 1 - a[m]*g_decay_slope[k]*Q_fract_allpass[k][m]*z^-link_delay[m]
472 // m = 0
473 //d[k][z] (out) = transient_gain_mapped[k][z] * H[k][z] * s[k][z]
474 for (k = 0; k < NR_ALLPASS_BANDS[is34]; k++) {
475 int b = k_to_i[k];
476 #if USE_FIXED
477 int g_decay_slope;
478
479 if (k - DECAY_CUTOFF[is34] <= 0) {
480 g_decay_slope = 1 << 30;
481 }
482 else if (k - DECAY_CUTOFF[is34] >= 20) {
483 g_decay_slope = 0;
484 }
485 else {
486 g_decay_slope = (1 << 30) - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]);
487 }
488 #else
489 float g_decay_slope = 1.f - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]);
490 g_decay_slope = av_clipf(g_decay_slope, 0.f, 1.f);
491 #endif /* USE_FIXED */
492 memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
493 memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
494 for (m = 0; m < PS_AP_LINKS; m++) {
495 memcpy(ap_delay[k][m], ap_delay[k][m]+numQMFSlots, 5*sizeof(ap_delay[k][m][0]));
496 }
497 ps->dsp.decorrelate(out[k], delay[k] + PS_MAX_DELAY - 2, ap_delay[k],
498 phi_fract[is34][k],
499 (const INTFLOAT (*)[2]) Q_fract_allpass[is34][k],
500 transient_gain[b], g_decay_slope, nL - n0);
501 }
502 for (; k < SHORT_DELAY_BAND[is34]; k++) {
503 int i = k_to_i[k];
504 memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
505 memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
506 //H = delay 14
507 ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 14,
508 transient_gain[i], nL - n0);
509 }
510 for (; k < NR_BANDS[is34]; k++) {
511 int i = k_to_i[k];
512 memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
513 memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
514 //H = delay 1
515 ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 1,
516 transient_gain[i], nL - n0);
517 }
518 }
519
remap34(int8_t (*p_par_mapped)[PS_MAX_NR_IIDICC], int8_t (*par)[PS_MAX_NR_IIDICC], int num_par, int num_env, int full)520 static void remap34(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC],
521 int8_t (*par)[PS_MAX_NR_IIDICC],
522 int num_par, int num_env, int full)
523 {
524 int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped;
525 int e;
526 if (num_par == 20 || num_par == 11) {
527 for (e = 0; e < num_env; e++) {
528 map_idx_20_to_34(par_mapped[e], par[e], full);
529 }
530 } else if (num_par == 10 || num_par == 5) {
531 for (e = 0; e < num_env; e++) {
532 map_idx_10_to_34(par_mapped[e], par[e], full);
533 }
534 } else {
535 *p_par_mapped = par;
536 }
537 }
538
remap20(int8_t (*p_par_mapped)[PS_MAX_NR_IIDICC], int8_t (*par)[PS_MAX_NR_IIDICC], int num_par, int num_env, int full)539 static void remap20(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC],
540 int8_t (*par)[PS_MAX_NR_IIDICC],
541 int num_par, int num_env, int full)
542 {
543 int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped;
544 int e;
545 if (num_par == 34 || num_par == 17) {
546 for (e = 0; e < num_env; e++) {
547 map_idx_34_to_20(par_mapped[e], par[e], full);
548 }
549 } else if (num_par == 10 || num_par == 5) {
550 for (e = 0; e < num_env; e++) {
551 map_idx_10_to_20(par_mapped[e], par[e], full);
552 }
553 } else {
554 *p_par_mapped = par;
555 }
556 }
557
stereo_processing(PSContext *ps, INTFLOAT (*l)[32][2], INTFLOAT (*r)[32][2], int is34)558 static void stereo_processing(PSContext *ps, INTFLOAT (*l)[32][2], INTFLOAT (*r)[32][2], int is34)
559 {
560 int e, b, k;
561
562 PSCommonContext *const ps2 = &ps->common;
563 INTFLOAT (*H11)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H11;
564 INTFLOAT (*H12)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H12;
565 INTFLOAT (*H21)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H21;
566 INTFLOAT (*H22)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H22;
567 int8_t *opd_hist = ps->opd_hist;
568 int8_t *ipd_hist = ps->ipd_hist;
569 int8_t iid_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
570 int8_t icc_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
571 int8_t ipd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
572 int8_t opd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
573 int8_t (*iid_mapped)[PS_MAX_NR_IIDICC] = iid_mapped_buf;
574 int8_t (*icc_mapped)[PS_MAX_NR_IIDICC] = icc_mapped_buf;
575 int8_t (*ipd_mapped)[PS_MAX_NR_IIDICC] = ipd_mapped_buf;
576 int8_t (*opd_mapped)[PS_MAX_NR_IIDICC] = opd_mapped_buf;
577 const int8_t *const k_to_i = is34 ? ff_k_to_i_34 : ff_k_to_i_20;
578 TABLE_CONST INTFLOAT (*H_LUT)[8][4] = (PS_BASELINE || ps2->icc_mode < 3) ? HA : HB;
579
580 //Remapping
581 if (ps2->num_env_old) {
582 memcpy(H11[0][0], H11[0][ps2->num_env_old], sizeof(H11[0][0]));
583 memcpy(H11[1][0], H11[1][ps2->num_env_old], sizeof(H11[1][0]));
584 memcpy(H12[0][0], H12[0][ps2->num_env_old], sizeof(H12[0][0]));
585 memcpy(H12[1][0], H12[1][ps2->num_env_old], sizeof(H12[1][0]));
586 memcpy(H21[0][0], H21[0][ps2->num_env_old], sizeof(H21[0][0]));
587 memcpy(H21[1][0], H21[1][ps2->num_env_old], sizeof(H21[1][0]));
588 memcpy(H22[0][0], H22[0][ps2->num_env_old], sizeof(H22[0][0]));
589 memcpy(H22[1][0], H22[1][ps2->num_env_old], sizeof(H22[1][0]));
590 }
591
592 if (is34) {
593 remap34(&iid_mapped, ps2->iid_par, ps2->nr_iid_par, ps2->num_env, 1);
594 remap34(&icc_mapped, ps2->icc_par, ps2->nr_icc_par, ps2->num_env, 1);
595 if (ps2->enable_ipdopd) {
596 remap34(&ipd_mapped, ps2->ipd_par, ps2->nr_ipdopd_par, ps2->num_env, 0);
597 remap34(&opd_mapped, ps2->opd_par, ps2->nr_ipdopd_par, ps2->num_env, 0);
598 }
599 if (!ps2->is34bands_old) {
600 map_val_20_to_34(H11[0][0]);
601 map_val_20_to_34(H11[1][0]);
602 map_val_20_to_34(H12[0][0]);
603 map_val_20_to_34(H12[1][0]);
604 map_val_20_to_34(H21[0][0]);
605 map_val_20_to_34(H21[1][0]);
606 map_val_20_to_34(H22[0][0]);
607 map_val_20_to_34(H22[1][0]);
608 ipdopd_reset(ipd_hist, opd_hist);
609 }
610 } else {
611 remap20(&iid_mapped, ps2->iid_par, ps2->nr_iid_par, ps2->num_env, 1);
612 remap20(&icc_mapped, ps2->icc_par, ps2->nr_icc_par, ps2->num_env, 1);
613 if (ps2->enable_ipdopd) {
614 remap20(&ipd_mapped, ps2->ipd_par, ps2->nr_ipdopd_par, ps2->num_env, 0);
615 remap20(&opd_mapped, ps2->opd_par, ps2->nr_ipdopd_par, ps2->num_env, 0);
616 }
617 if (ps2->is34bands_old) {
618 map_val_34_to_20(H11[0][0]);
619 map_val_34_to_20(H11[1][0]);
620 map_val_34_to_20(H12[0][0]);
621 map_val_34_to_20(H12[1][0]);
622 map_val_34_to_20(H21[0][0]);
623 map_val_34_to_20(H21[1][0]);
624 map_val_34_to_20(H22[0][0]);
625 map_val_34_to_20(H22[1][0]);
626 ipdopd_reset(ipd_hist, opd_hist);
627 }
628 }
629
630 //Mixing
631 for (e = 0; e < ps2->num_env; e++) {
632 for (b = 0; b < NR_PAR_BANDS[is34]; b++) {
633 INTFLOAT h11, h12, h21, h22;
634 h11 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps2->iid_quant][icc_mapped[e][b]][0];
635 h12 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps2->iid_quant][icc_mapped[e][b]][1];
636 h21 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps2->iid_quant][icc_mapped[e][b]][2];
637 h22 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps2->iid_quant][icc_mapped[e][b]][3];
638
639 if (!PS_BASELINE && ps2->enable_ipdopd && b < NR_IPDOPD_BANDS[is34]) {
640 //The spec say says to only run this smoother when enable_ipdopd
641 //is set but the reference decoder appears to run it constantly
642 INTFLOAT h11i, h12i, h21i, h22i;
643 INTFLOAT ipd_adj_re, ipd_adj_im;
644 int opd_idx = opd_hist[b] * 8 + opd_mapped[e][b];
645 int ipd_idx = ipd_hist[b] * 8 + ipd_mapped[e][b];
646 INTFLOAT opd_re = pd_re_smooth[opd_idx];
647 INTFLOAT opd_im = pd_im_smooth[opd_idx];
648 INTFLOAT ipd_re = pd_re_smooth[ipd_idx];
649 INTFLOAT ipd_im = pd_im_smooth[ipd_idx];
650 opd_hist[b] = opd_idx & 0x3F;
651 ipd_hist[b] = ipd_idx & 0x3F;
652
653 ipd_adj_re = AAC_MADD30(opd_re, ipd_re, opd_im, ipd_im);
654 ipd_adj_im = AAC_MSUB30(opd_im, ipd_re, opd_re, ipd_im);
655 h11i = AAC_MUL30(h11, opd_im);
656 h11 = AAC_MUL30(h11, opd_re);
657 h12i = AAC_MUL30(h12, ipd_adj_im);
658 h12 = AAC_MUL30(h12, ipd_adj_re);
659 h21i = AAC_MUL30(h21, opd_im);
660 h21 = AAC_MUL30(h21, opd_re);
661 h22i = AAC_MUL30(h22, ipd_adj_im);
662 h22 = AAC_MUL30(h22, ipd_adj_re);
663 H11[1][e+1][b] = h11i;
664 H12[1][e+1][b] = h12i;
665 H21[1][e+1][b] = h21i;
666 H22[1][e+1][b] = h22i;
667 }
668 H11[0][e+1][b] = h11;
669 H12[0][e+1][b] = h12;
670 H21[0][e+1][b] = h21;
671 H22[0][e+1][b] = h22;
672 }
673 for (k = 0; k < NR_BANDS[is34]; k++) {
674 LOCAL_ALIGNED_16(INTFLOAT, h, [2], [4]);
675 LOCAL_ALIGNED_16(INTFLOAT, h_step, [2], [4]);
676 int start = ps2->border_position[e];
677 int stop = ps2->border_position[e+1];
678 INTFLOAT width = Q30(1.f) / ((stop - start) ? (stop - start) : 1);
679 #if USE_FIXED
680 width = FFMIN(2U*width, INT_MAX);
681 #endif
682 b = k_to_i[k];
683 h[0][0] = H11[0][e][b];
684 h[0][1] = H12[0][e][b];
685 h[0][2] = H21[0][e][b];
686 h[0][3] = H22[0][e][b];
687 if (!PS_BASELINE && ps2->enable_ipdopd) {
688 //Is this necessary? ps_04_new seems unchanged
689 if ((is34 && k <= 13 && k >= 9) || (!is34 && k <= 1)) {
690 h[1][0] = -H11[1][e][b];
691 h[1][1] = -H12[1][e][b];
692 h[1][2] = -H21[1][e][b];
693 h[1][3] = -H22[1][e][b];
694 } else {
695 h[1][0] = H11[1][e][b];
696 h[1][1] = H12[1][e][b];
697 h[1][2] = H21[1][e][b];
698 h[1][3] = H22[1][e][b];
699 }
700 }
701 //Interpolation
702 h_step[0][0] = AAC_MSUB31_V3(H11[0][e+1][b], h[0][0], width);
703 h_step[0][1] = AAC_MSUB31_V3(H12[0][e+1][b], h[0][1], width);
704 h_step[0][2] = AAC_MSUB31_V3(H21[0][e+1][b], h[0][2], width);
705 h_step[0][3] = AAC_MSUB31_V3(H22[0][e+1][b], h[0][3], width);
706 if (!PS_BASELINE && ps2->enable_ipdopd) {
707 h_step[1][0] = AAC_MSUB31_V3(H11[1][e+1][b], h[1][0], width);
708 h_step[1][1] = AAC_MSUB31_V3(H12[1][e+1][b], h[1][1], width);
709 h_step[1][2] = AAC_MSUB31_V3(H21[1][e+1][b], h[1][2], width);
710 h_step[1][3] = AAC_MSUB31_V3(H22[1][e+1][b], h[1][3], width);
711 }
712 if (stop - start)
713 ps->dsp.stereo_interpolate[!PS_BASELINE && ps2->enable_ipdopd](
714 l[k] + 1 + start, r[k] + 1 + start,
715 h, h_step, stop - start);
716 }
717 }
718 }
719
ff_ps_apply(AVCodecContext *avctx, PSContext *ps, INTFLOAT L[2][38][64], INTFLOAT R[2][38][64], int top)720 int AAC_RENAME(ff_ps_apply)(AVCodecContext *avctx, PSContext *ps, INTFLOAT L[2][38][64], INTFLOAT R[2][38][64], int top)
721 {
722 INTFLOAT (*Lbuf)[32][2] = ps->Lbuf;
723 INTFLOAT (*Rbuf)[32][2] = ps->Rbuf;
724 const int len = 32;
725 int is34 = ps->common.is34bands;
726
727 top += NR_BANDS[is34] - 64;
728 memset(ps->delay+top, 0, (NR_BANDS[is34] - top)*sizeof(ps->delay[0]));
729 if (top < NR_ALLPASS_BANDS[is34])
730 memset(ps->ap_delay + top, 0, (NR_ALLPASS_BANDS[is34] - top)*sizeof(ps->ap_delay[0]));
731
732 hybrid_analysis(&ps->dsp, Lbuf, ps->in_buf, L, is34, len);
733 decorrelation(ps, Rbuf, (const INTFLOAT (*)[32][2]) Lbuf, is34);
734 stereo_processing(ps, Lbuf, Rbuf, is34);
735 hybrid_synthesis(&ps->dsp, L, Lbuf, is34, len);
736 hybrid_synthesis(&ps->dsp, R, Rbuf, is34, len);
737
738 return 0;
739 }
740
ff_ps_init(void)741 av_cold void AAC_RENAME(ff_ps_init)(void) {
742 ps_tableinit();
743 ff_ps_init_common();
744 }
745
ff_ps_ctx_init(PSContext *ps)746 av_cold void AAC_RENAME(ff_ps_ctx_init)(PSContext *ps)
747 {
748 AAC_RENAME(ff_psdsp_init)(&ps->dsp);
749 }
750