2 * MPEG-4 Parametric Stereo decoding functions
3 * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com>
5 * This file is part of FFmpeg.
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.
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.
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
21 * Note: Rounding-to-nearest used unless otherwise stated
26 #include "libavutil/common.h"
27 #include "libavutil/mathematics.h"
32 #include "aacps_fixed_tablegen.h"
34 #include "libavutil/internal.h"
35 #include "aacps_tablegen.h"
36 #endif /* USE_FIXED */
37 #include "aacpsdata.c"
39 #define PS_BASELINE 0 ///< Operate in Baseline PS mode
40 ///< Baseline implies 10 or 20 stereo bands,
41 ///< mixing mode A, and no ipd/opd
43 #define numQMFSlots 32 //numTimeSlots * RATE
45 static const int8_t num_env_tab[2][4] = {
50 static const int8_t nr_iidicc_par_tab[] = {
51 10, 20, 34, 10, 20, 34,
54 static const int8_t nr_iidopd_par_tab[] = {
71 static const int huff_iid[] = {
78 static VLC vlc_ps[10];
80 #define READ_PAR_DATA(PAR, OFFSET, MASK, ERR_CONDITION) \
82 * Read Inter-channel Intensity Difference/Inter-Channel Coherence/ \
83 * Inter-channel Phase Difference/Overall Phase Difference parameters from the \
86 * @param avctx contains the current codec context \
87 * @param gb pointer to the input bitstream \
88 * @param ps pointer to the Parametric Stereo context \
89 * @param PAR pointer to the parameter to be read \
90 * @param e envelope to decode \
91 * @param dt 1: time delta-coded, 0: frequency delta-coded \
93 static int read_ ## PAR ## _data(AVCodecContext *avctx, GetBitContext *gb, PSContext *ps, \
94 int8_t (*PAR)[PS_MAX_NR_IIDICC], int table_idx, int e, int dt) \
96 int b, num = ps->nr_ ## PAR ## _par; \
97 VLC_TYPE (*vlc_table)[2] = vlc_ps[table_idx].table; \
99 int e_prev = e ? e - 1 : ps->num_env_old - 1; \
100 e_prev = FFMAX(e_prev, 0); \
101 for (b = 0; b < num; b++) { \
102 int val = PAR[e_prev][b] + get_vlc2(gb, vlc_table, 9, 3) - OFFSET; \
103 if (MASK) val &= MASK; \
110 for (b = 0; b < num; b++) { \
111 val += get_vlc2(gb, vlc_table, 9, 3) - OFFSET; \
112 if (MASK) val &= MASK; \
120 av_log(avctx, AV_LOG_ERROR, "illegal "#PAR"\n"); \
124 READ_PAR_DATA(iid, huff_offset[table_idx], 0, FFABS(ps->iid_par[e][b]) > 7 + 8 * ps->iid_quant)
125 READ_PAR_DATA(icc, huff_offset[table_idx], 0, ps->icc_par[e][b] > 7U)
126 READ_PAR_DATA(ipdopd, 0, 0x07, 0)
128 static int ps_read_extension_data(GetBitContext *gb, PSContext *ps, int ps_extension_id)
131 int count = get_bits_count(gb);
136 ps->enable_ipdopd = get_bits1(gb);
137 if (ps->enable_ipdopd) {
138 for (e = 0; e < ps->num_env; e++) {
139 int dt = get_bits1(gb);
140 read_ipdopd_data(NULL, gb, ps, ps->ipd_par, dt ? huff_ipd_dt : huff_ipd_df, e, dt);
142 read_ipdopd_data(NULL, gb, ps, ps->opd_par, dt ? huff_opd_dt : huff_opd_df, e, dt);
145 skip_bits1(gb); //reserved_ps
146 return get_bits_count(gb) - count;
149 static void ipdopd_reset(int8_t *ipd_hist, int8_t *opd_hist)
152 for (i = 0; i < PS_MAX_NR_IPDOPD; i++) {
158 int AAC_RENAME(ff_ps_read_data)(AVCodecContext *avctx, GetBitContext *gb_host, PSContext *ps, int bits_left)
161 int bit_count_start = get_bits_count(gb_host);
164 GetBitContext gbc = *gb_host, *gb = &gbc;
166 header = get_bits1(gb);
167 if (header) { //enable_ps_header
168 ps->enable_iid = get_bits1(gb);
169 if (ps->enable_iid) {
170 int iid_mode = get_bits(gb, 3);
172 av_log(avctx, AV_LOG_ERROR, "iid_mode %d is reserved.\n",
176 ps->nr_iid_par = nr_iidicc_par_tab[iid_mode];
177 ps->iid_quant = iid_mode > 2;
178 ps->nr_ipdopd_par = nr_iidopd_par_tab[iid_mode];
180 ps->enable_icc = get_bits1(gb);
181 if (ps->enable_icc) {
182 ps->icc_mode = get_bits(gb, 3);
183 if (ps->icc_mode > 5) {
184 av_log(avctx, AV_LOG_ERROR, "icc_mode %d is reserved.\n",
188 ps->nr_icc_par = nr_iidicc_par_tab[ps->icc_mode];
190 ps->enable_ext = get_bits1(gb);
193 ps->frame_class = get_bits1(gb);
194 ps->num_env_old = ps->num_env;
195 ps->num_env = num_env_tab[ps->frame_class][get_bits(gb, 2)];
197 ps->border_position[0] = -1;
198 if (ps->frame_class) {
199 for (e = 1; e <= ps->num_env; e++)
200 ps->border_position[e] = get_bits(gb, 5);
202 for (e = 1; e <= ps->num_env; e++)
203 ps->border_position[e] = (e * numQMFSlots >> ff_log2_tab[ps->num_env]) - 1;
205 if (ps->enable_iid) {
206 for (e = 0; e < ps->num_env; e++) {
207 int dt = get_bits1(gb);
208 if (read_iid_data(avctx, gb, ps, ps->iid_par, huff_iid[2*dt+ps->iid_quant], e, dt))
212 memset(ps->iid_par, 0, sizeof(ps->iid_par));
215 for (e = 0; e < ps->num_env; e++) {
216 int dt = get_bits1(gb);
217 if (read_icc_data(avctx, gb, ps, ps->icc_par, dt ? huff_icc_dt : huff_icc_df, e, dt))
221 memset(ps->icc_par, 0, sizeof(ps->icc_par));
223 if (ps->enable_ext) {
224 int cnt = get_bits(gb, 4);
226 cnt += get_bits(gb, 8);
230 int ps_extension_id = get_bits(gb, 2);
231 cnt -= 2 + ps_read_extension_data(gb, ps, ps_extension_id);
234 av_log(avctx, AV_LOG_ERROR, "ps extension overflow %d\n", cnt);
240 ps->enable_ipdopd &= !PS_BASELINE;
243 if (!ps->num_env || ps->border_position[ps->num_env] < numQMFSlots - 1) {
244 //Create a fake envelope
245 int source = ps->num_env ? ps->num_env - 1 : ps->num_env_old - 1;
247 if (source >= 0 && source != ps->num_env) {
248 if (ps->enable_iid) {
249 memcpy(ps->iid_par+ps->num_env, ps->iid_par+source, sizeof(ps->iid_par[0]));
251 if (ps->enable_icc) {
252 memcpy(ps->icc_par+ps->num_env, ps->icc_par+source, sizeof(ps->icc_par[0]));
254 if (ps->enable_ipdopd) {
255 memcpy(ps->ipd_par+ps->num_env, ps->ipd_par+source, sizeof(ps->ipd_par[0]));
256 memcpy(ps->opd_par+ps->num_env, ps->opd_par+source, sizeof(ps->opd_par[0]));
260 for (b = 0; b < ps->nr_iid_par; b++) {
261 if (FFABS(ps->iid_par[ps->num_env][b]) > 7 + 8 * ps->iid_quant) {
262 av_log(avctx, AV_LOG_ERROR, "iid_par invalid\n");
268 for (b = 0; b < ps->nr_iid_par; b++) {
269 if (ps->icc_par[ps->num_env][b] > 7U) {
270 av_log(avctx, AV_LOG_ERROR, "icc_par invalid\n");
276 ps->border_position[ps->num_env] = numQMFSlots - 1;
280 ps->is34bands_old = ps->is34bands;
281 if (!PS_BASELINE && (ps->enable_iid || ps->enable_icc))
282 ps->is34bands = (ps->enable_iid && ps->nr_iid_par == 34) ||
283 (ps->enable_icc && ps->nr_icc_par == 34);
286 if (!ps->enable_ipdopd) {
287 memset(ps->ipd_par, 0, sizeof(ps->ipd_par));
288 memset(ps->opd_par, 0, sizeof(ps->opd_par));
294 bits_consumed = get_bits_count(gb) - bit_count_start;
295 if (bits_consumed <= bits_left) {
296 skip_bits_long(gb_host, bits_consumed);
297 return bits_consumed;
299 av_log(avctx, AV_LOG_ERROR, "Expected to read %d PS bits actually read %d.\n", bits_left, bits_consumed);
302 skip_bits_long(gb_host, bits_left);
303 memset(ps->iid_par, 0, sizeof(ps->iid_par));
304 memset(ps->icc_par, 0, sizeof(ps->icc_par));
305 memset(ps->ipd_par, 0, sizeof(ps->ipd_par));
306 memset(ps->opd_par, 0, sizeof(ps->opd_par));
310 /** Split one subband into 2 subsubbands with a symmetric real filter.
311 * The filter must have its non-center even coefficients equal to zero. */
312 static void hybrid2_re(INTFLOAT (*in)[2], INTFLOAT (*out)[32][2], const INTFLOAT filter[8], int len, int reverse)
315 for (i = 0; i < len; i++, in++) {
316 INT64FLOAT re_in = AAC_MUL31(filter[6], in[6][0]); //real inphase
317 INT64FLOAT re_op = 0.0f; //real out of phase
318 INT64FLOAT im_in = AAC_MUL31(filter[6], in[6][1]); //imag inphase
319 INT64FLOAT im_op = 0.0f; //imag out of phase
320 for (j = 0; j < 6; j += 2) {
321 re_op += (INT64FLOAT)filter[j+1] * (in[j+1][0] + in[12-j-1][0]);
322 im_op += (INT64FLOAT)filter[j+1] * (in[j+1][1] + in[12-j-1][1]);
326 re_op = (re_op + 0x40000000) >> 31;
327 im_op = (im_op + 0x40000000) >> 31;
328 #endif /* USE_FIXED */
330 out[ reverse][i][0] = (INTFLOAT)(re_in + re_op);
331 out[ reverse][i][1] = (INTFLOAT)(im_in + im_op);
332 out[!reverse][i][0] = (INTFLOAT)(re_in - re_op);
333 out[!reverse][i][1] = (INTFLOAT)(im_in - im_op);
337 /** Split one subband into 6 subsubbands with a complex filter */
338 static void hybrid6_cx(PSDSPContext *dsp, INTFLOAT (*in)[2], INTFLOAT (*out)[32][2],
339 TABLE_CONST INTFLOAT (*filter)[8][2], int len)
343 LOCAL_ALIGNED_16(INTFLOAT, temp, [8], [2]);
345 for (i = 0; i < len; i++, in++) {
346 dsp->hybrid_analysis(temp, in, (const INTFLOAT (*)[8][2]) filter, 1, N);
347 out[0][i][0] = temp[6][0];
348 out[0][i][1] = temp[6][1];
349 out[1][i][0] = temp[7][0];
350 out[1][i][1] = temp[7][1];
351 out[2][i][0] = temp[0][0];
352 out[2][i][1] = temp[0][1];
353 out[3][i][0] = temp[1][0];
354 out[3][i][1] = temp[1][1];
355 out[4][i][0] = temp[2][0] + temp[5][0];
356 out[4][i][1] = temp[2][1] + temp[5][1];
357 out[5][i][0] = temp[3][0] + temp[4][0];
358 out[5][i][1] = temp[3][1] + temp[4][1];
362 static void hybrid4_8_12_cx(PSDSPContext *dsp,
363 INTFLOAT (*in)[2], INTFLOAT (*out)[32][2],
364 TABLE_CONST INTFLOAT (*filter)[8][2], int N, int len)
368 for (i = 0; i < len; i++, in++) {
369 dsp->hybrid_analysis(out[0] + i, in, (const INTFLOAT (*)[8][2]) filter, 32, N);
373 static void hybrid_analysis(PSDSPContext *dsp, INTFLOAT out[91][32][2],
374 INTFLOAT in[5][44][2], INTFLOAT L[2][38][64],
378 for (i = 0; i < 5; i++) {
379 for (j = 0; j < 38; j++) {
380 in[i][j+6][0] = L[0][j][i];
381 in[i][j+6][1] = L[1][j][i];
385 hybrid4_8_12_cx(dsp, in[0], out, f34_0_12, 12, len);
386 hybrid4_8_12_cx(dsp, in[1], out+12, f34_1_8, 8, len);
387 hybrid4_8_12_cx(dsp, in[2], out+20, f34_2_4, 4, len);
388 hybrid4_8_12_cx(dsp, in[3], out+24, f34_2_4, 4, len);
389 hybrid4_8_12_cx(dsp, in[4], out+28, f34_2_4, 4, len);
390 dsp->hybrid_analysis_ileave(out + 27, L, 5, len);
392 hybrid6_cx(dsp, in[0], out, f20_0_8, len);
393 hybrid2_re(in[1], out+6, g1_Q2, len, 1);
394 hybrid2_re(in[2], out+8, g1_Q2, len, 0);
395 dsp->hybrid_analysis_ileave(out + 7, L, 3, len);
398 for (i = 0; i < 5; i++) {
399 memcpy(in[i], in[i]+32, 6 * sizeof(in[i][0]));
403 static void hybrid_synthesis(PSDSPContext *dsp, INTFLOAT out[2][38][64],
404 INTFLOAT in[91][32][2], int is34, int len)
408 for (n = 0; n < len; n++) {
409 memset(out[0][n], 0, 5*sizeof(out[0][n][0]));
410 memset(out[1][n], 0, 5*sizeof(out[1][n][0]));
411 for (i = 0; i < 12; i++) {
412 out[0][n][0] += in[ i][n][0];
413 out[1][n][0] += in[ i][n][1];
415 for (i = 0; i < 8; i++) {
416 out[0][n][1] += in[12+i][n][0];
417 out[1][n][1] += in[12+i][n][1];
419 for (i = 0; i < 4; i++) {
420 out[0][n][2] += in[20+i][n][0];
421 out[1][n][2] += in[20+i][n][1];
422 out[0][n][3] += in[24+i][n][0];
423 out[1][n][3] += in[24+i][n][1];
424 out[0][n][4] += in[28+i][n][0];
425 out[1][n][4] += in[28+i][n][1];
428 dsp->hybrid_synthesis_deint(out, in + 27, 5, len);
430 for (n = 0; n < len; n++) {
431 out[0][n][0] = in[0][n][0] + in[1][n][0] + in[2][n][0] +
432 in[3][n][0] + in[4][n][0] + in[5][n][0];
433 out[1][n][0] = in[0][n][1] + in[1][n][1] + in[2][n][1] +
434 in[3][n][1] + in[4][n][1] + in[5][n][1];
435 out[0][n][1] = in[6][n][0] + in[7][n][0];
436 out[1][n][1] = in[6][n][1] + in[7][n][1];
437 out[0][n][2] = in[8][n][0] + in[9][n][0];
438 out[1][n][2] = in[8][n][1] + in[9][n][1];
440 dsp->hybrid_synthesis_deint(out, in + 7, 3, len);
444 /// All-pass filter decay slope
445 #define DECAY_SLOPE Q30(0.05f)
446 /// Number of frequency bands that can be addressed by the parameter index, b(k)
447 static const int NR_PAR_BANDS[] = { 20, 34 };
448 static const int NR_IPDOPD_BANDS[] = { 11, 17 };
449 /// Number of frequency bands that can be addressed by the sub subband index, k
450 static const int NR_BANDS[] = { 71, 91 };
451 /// Start frequency band for the all-pass filter decay slope
452 static const int DECAY_CUTOFF[] = { 10, 32 };
453 /// Number of all-pass filer bands
454 static const int NR_ALLPASS_BANDS[] = { 30, 50 };
455 /// First stereo band using the short one sample delay
456 static const int SHORT_DELAY_BAND[] = { 42, 62 };
459 static void map_idx_10_to_20(int8_t *par_mapped, const int8_t *par, int full)
468 for (; b >= 0; b--) {
469 par_mapped[2*b+1] = par_mapped[2*b] = par[b];
473 static void map_idx_34_to_20(int8_t *par_mapped, const int8_t *par, int full)
475 par_mapped[ 0] = (2*par[ 0] + par[ 1]) / 3;
476 par_mapped[ 1] = ( par[ 1] + 2*par[ 2]) / 3;
477 par_mapped[ 2] = (2*par[ 3] + par[ 4]) / 3;
478 par_mapped[ 3] = ( par[ 4] + 2*par[ 5]) / 3;
479 par_mapped[ 4] = ( par[ 6] + par[ 7]) / 2;
480 par_mapped[ 5] = ( par[ 8] + par[ 9]) / 2;
481 par_mapped[ 6] = par[10];
482 par_mapped[ 7] = par[11];
483 par_mapped[ 8] = ( par[12] + par[13]) / 2;
484 par_mapped[ 9] = ( par[14] + par[15]) / 2;
485 par_mapped[10] = par[16];
487 par_mapped[11] = par[17];
488 par_mapped[12] = par[18];
489 par_mapped[13] = par[19];
490 par_mapped[14] = ( par[20] + par[21]) / 2;
491 par_mapped[15] = ( par[22] + par[23]) / 2;
492 par_mapped[16] = ( par[24] + par[25]) / 2;
493 par_mapped[17] = ( par[26] + par[27]) / 2;
494 par_mapped[18] = ( par[28] + par[29] + par[30] + par[31]) / 4;
495 par_mapped[19] = ( par[32] + par[33]) / 2;
499 static void map_val_34_to_20(INTFLOAT par[PS_MAX_NR_IIDICC])
502 par[ 0] = (int)(((int64_t)(par[ 0] + (par[ 1]>>1)) * 1431655765 + \
504 par[ 1] = (int)(((int64_t)((par[ 1]>>1) + par[ 2]) * 1431655765 + \
506 par[ 2] = (int)(((int64_t)(par[ 3] + (par[ 4]>>1)) * 1431655765 + \
508 par[ 3] = (int)(((int64_t)((par[ 4]>>1) + par[ 5]) * 1431655765 + \
511 par[ 0] = (2*par[ 0] + par[ 1]) * 0.33333333f;
512 par[ 1] = ( par[ 1] + 2*par[ 2]) * 0.33333333f;
513 par[ 2] = (2*par[ 3] + par[ 4]) * 0.33333333f;
514 par[ 3] = ( par[ 4] + 2*par[ 5]) * 0.33333333f;
515 #endif /* USE_FIXED */
516 par[ 4] = AAC_HALF_SUM(par[ 6], par[ 7]);
517 par[ 5] = AAC_HALF_SUM(par[ 8], par[ 9]);
520 par[ 8] = AAC_HALF_SUM(par[12], par[13]);
521 par[ 9] = AAC_HALF_SUM(par[14], par[15]);
526 par[14] = AAC_HALF_SUM(par[20], par[21]);
527 par[15] = AAC_HALF_SUM(par[22], par[23]);
528 par[16] = AAC_HALF_SUM(par[24], par[25]);
529 par[17] = AAC_HALF_SUM(par[26], par[27]);
531 par[18] = (((par[28]+2)>>2) + ((par[29]+2)>>2) + ((par[30]+2)>>2) + ((par[31]+2)>>2));
533 par[18] = ( par[28] + par[29] + par[30] + par[31]) * 0.25f;
534 #endif /* USE_FIXED */
535 par[19] = AAC_HALF_SUM(par[32], par[33]);
538 static void map_idx_10_to_34(int8_t *par_mapped, const int8_t *par, int full)
541 par_mapped[33] = par[9];
542 par_mapped[32] = par[9];
543 par_mapped[31] = par[9];
544 par_mapped[30] = par[9];
545 par_mapped[29] = par[9];
546 par_mapped[28] = par[9];
547 par_mapped[27] = par[8];
548 par_mapped[26] = par[8];
549 par_mapped[25] = par[8];
550 par_mapped[24] = par[8];
551 par_mapped[23] = par[7];
552 par_mapped[22] = par[7];
553 par_mapped[21] = par[7];
554 par_mapped[20] = par[7];
555 par_mapped[19] = par[6];
556 par_mapped[18] = par[6];
557 par_mapped[17] = par[5];
558 par_mapped[16] = par[5];
562 par_mapped[15] = par[4];
563 par_mapped[14] = par[4];
564 par_mapped[13] = par[4];
565 par_mapped[12] = par[4];
566 par_mapped[11] = par[3];
567 par_mapped[10] = par[3];
568 par_mapped[ 9] = par[2];
569 par_mapped[ 8] = par[2];
570 par_mapped[ 7] = par[2];
571 par_mapped[ 6] = par[2];
572 par_mapped[ 5] = par[1];
573 par_mapped[ 4] = par[1];
574 par_mapped[ 3] = par[1];
575 par_mapped[ 2] = par[0];
576 par_mapped[ 1] = par[0];
577 par_mapped[ 0] = par[0];
580 static void map_idx_20_to_34(int8_t *par_mapped, const int8_t *par, int full)
583 par_mapped[33] = par[19];
584 par_mapped[32] = par[19];
585 par_mapped[31] = par[18];
586 par_mapped[30] = par[18];
587 par_mapped[29] = par[18];
588 par_mapped[28] = par[18];
589 par_mapped[27] = par[17];
590 par_mapped[26] = par[17];
591 par_mapped[25] = par[16];
592 par_mapped[24] = par[16];
593 par_mapped[23] = par[15];
594 par_mapped[22] = par[15];
595 par_mapped[21] = par[14];
596 par_mapped[20] = par[14];
597 par_mapped[19] = par[13];
598 par_mapped[18] = par[12];
599 par_mapped[17] = par[11];
601 par_mapped[16] = par[10];
602 par_mapped[15] = par[ 9];
603 par_mapped[14] = par[ 9];
604 par_mapped[13] = par[ 8];
605 par_mapped[12] = par[ 8];
606 par_mapped[11] = par[ 7];
607 par_mapped[10] = par[ 6];
608 par_mapped[ 9] = par[ 5];
609 par_mapped[ 8] = par[ 5];
610 par_mapped[ 7] = par[ 4];
611 par_mapped[ 6] = par[ 4];
612 par_mapped[ 5] = par[ 3];
613 par_mapped[ 4] = (par[ 2] + par[ 3]) / 2;
614 par_mapped[ 3] = par[ 2];
615 par_mapped[ 2] = par[ 1];
616 par_mapped[ 1] = (par[ 0] + par[ 1]) / 2;
617 par_mapped[ 0] = par[ 0];
620 static void map_val_20_to_34(INTFLOAT par[PS_MAX_NR_IIDICC])
651 par[ 4] = AAC_HALF_SUM(par[ 2], par[ 3]);
654 par[ 1] = AAC_HALF_SUM(par[ 0], par[ 1]);
657 static void decorrelation(PSContext *ps, INTFLOAT (*out)[32][2], const INTFLOAT (*s)[32][2], int is34)
659 LOCAL_ALIGNED_16(INTFLOAT, power, [34], [PS_QMF_TIME_SLOTS]);
660 LOCAL_ALIGNED_16(INTFLOAT, transient_gain, [34], [PS_QMF_TIME_SLOTS]);
661 INTFLOAT *peak_decay_nrg = ps->peak_decay_nrg;
662 INTFLOAT *power_smooth = ps->power_smooth;
663 INTFLOAT *peak_decay_diff_smooth = ps->peak_decay_diff_smooth;
664 INTFLOAT (*delay)[PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2] = ps->delay;
665 INTFLOAT (*ap_delay)[PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2] = ps->ap_delay;
667 const float transient_impact = 1.5f;
668 const float a_smooth = 0.25f; ///< Smoothing coefficient
669 #endif /* USE_FIXED */
670 const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20;
673 const INTFLOAT peak_decay_factor = Q31(0.76592833836465f);
675 memset(power, 0, 34 * sizeof(*power));
677 if (is34 != ps->is34bands_old) {
678 memset(ps->peak_decay_nrg, 0, sizeof(ps->peak_decay_nrg));
679 memset(ps->power_smooth, 0, sizeof(ps->power_smooth));
680 memset(ps->peak_decay_diff_smooth, 0, sizeof(ps->peak_decay_diff_smooth));
681 memset(ps->delay, 0, sizeof(ps->delay));
682 memset(ps->ap_delay, 0, sizeof(ps->ap_delay));
685 for (k = 0; k < NR_BANDS[is34]; k++) {
687 ps->dsp.add_squares(power[i], s[k], nL - n0);
690 //Transient detection
692 for (i = 0; i < NR_PAR_BANDS[is34]; i++) {
693 for (n = n0; n < nL; n++) {
697 decayed_peak = (int)(((int64_t)peak_decay_factor * \
698 peak_decay_nrg[i] + 0x40000000) >> 31);
699 peak_decay_nrg[i] = FFMAX(decayed_peak, power[i][n]);
700 power_smooth[i] += (power[i][n] - power_smooth[i] + 2) >> 2;
701 peak_decay_diff_smooth[i] += (peak_decay_nrg[i] - power[i][n] - \
702 peak_decay_diff_smooth[i] + 2) >> 2;
703 denom = peak_decay_diff_smooth[i] + (peak_decay_diff_smooth[i] >> 1);
704 if (denom > power_smooth[i]) {
705 int p = power_smooth[i];
706 while (denom < 0x40000000) {
710 transient_gain[i][n] = p / (denom >> 16);
713 transient_gain[i][n] = 1 << 16;
718 for (i = 0; i < NR_PAR_BANDS[is34]; i++) {
719 for (n = n0; n < nL; n++) {
720 float decayed_peak = peak_decay_factor * peak_decay_nrg[i];
722 peak_decay_nrg[i] = FFMAX(decayed_peak, power[i][n]);
723 power_smooth[i] += a_smooth * (power[i][n] - power_smooth[i]);
724 peak_decay_diff_smooth[i] += a_smooth * (peak_decay_nrg[i] - power[i][n] - peak_decay_diff_smooth[i]);
725 denom = transient_impact * peak_decay_diff_smooth[i];
726 transient_gain[i][n] = (denom > power_smooth[i]) ?
727 power_smooth[i] / denom : 1.0f;
731 #endif /* USE_FIXED */
732 //Decorrelation and transient reduction
735 // | | Q_fract_allpass[k][m]*z^-link_delay[m] - a[m]*g_decay_slope[k]
736 //H[k][z] = z^-2 * phi_fract[k] * | | ----------------------------------------------------------------
737 // | | 1 - a[m]*g_decay_slope[k]*Q_fract_allpass[k][m]*z^-link_delay[m]
739 //d[k][z] (out) = transient_gain_mapped[k][z] * H[k][z] * s[k][z]
740 for (k = 0; k < NR_ALLPASS_BANDS[is34]; k++) {
745 if (k - DECAY_CUTOFF[is34] <= 0) {
746 g_decay_slope = 1 << 30;
748 else if (k - DECAY_CUTOFF[is34] >= 20) {
752 g_decay_slope = (1 << 30) - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]);
755 float g_decay_slope = 1.f - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]);
756 g_decay_slope = av_clipf(g_decay_slope, 0.f, 1.f);
757 #endif /* USE_FIXED */
758 memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
759 memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
760 for (m = 0; m < PS_AP_LINKS; m++) {
761 memcpy(ap_delay[k][m], ap_delay[k][m]+numQMFSlots, 5*sizeof(ap_delay[k][m][0]));
763 ps->dsp.decorrelate(out[k], delay[k] + PS_MAX_DELAY - 2, ap_delay[k],
765 (const INTFLOAT (*)[2]) Q_fract_allpass[is34][k],
766 transient_gain[b], g_decay_slope, nL - n0);
768 for (; k < SHORT_DELAY_BAND[is34]; k++) {
770 memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
771 memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
773 ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 14,
774 transient_gain[i], nL - n0);
776 for (; k < NR_BANDS[is34]; k++) {
778 memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
779 memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
781 ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 1,
782 transient_gain[i], nL - n0);
786 static void remap34(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC],
787 int8_t (*par)[PS_MAX_NR_IIDICC],
788 int num_par, int num_env, int full)
790 int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped;
792 if (num_par == 20 || num_par == 11) {
793 for (e = 0; e < num_env; e++) {
794 map_idx_20_to_34(par_mapped[e], par[e], full);
796 } else if (num_par == 10 || num_par == 5) {
797 for (e = 0; e < num_env; e++) {
798 map_idx_10_to_34(par_mapped[e], par[e], full);
805 static void remap20(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC],
806 int8_t (*par)[PS_MAX_NR_IIDICC],
807 int num_par, int num_env, int full)
809 int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped;
811 if (num_par == 34 || num_par == 17) {
812 for (e = 0; e < num_env; e++) {
813 map_idx_34_to_20(par_mapped[e], par[e], full);
815 } else if (num_par == 10 || num_par == 5) {
816 for (e = 0; e < num_env; e++) {
817 map_idx_10_to_20(par_mapped[e], par[e], full);
824 static void stereo_processing(PSContext *ps, INTFLOAT (*l)[32][2], INTFLOAT (*r)[32][2], int is34)
828 INTFLOAT (*H11)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H11;
829 INTFLOAT (*H12)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H12;
830 INTFLOAT (*H21)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H21;
831 INTFLOAT (*H22)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H22;
832 int8_t *opd_hist = ps->opd_hist;
833 int8_t *ipd_hist = ps->ipd_hist;
834 int8_t iid_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
835 int8_t icc_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
836 int8_t ipd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
837 int8_t opd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
838 int8_t (*iid_mapped)[PS_MAX_NR_IIDICC] = iid_mapped_buf;
839 int8_t (*icc_mapped)[PS_MAX_NR_IIDICC] = icc_mapped_buf;
840 int8_t (*ipd_mapped)[PS_MAX_NR_IIDICC] = ipd_mapped_buf;
841 int8_t (*opd_mapped)[PS_MAX_NR_IIDICC] = opd_mapped_buf;
842 const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20;
843 TABLE_CONST INTFLOAT (*H_LUT)[8][4] = (PS_BASELINE || ps->icc_mode < 3) ? HA : HB;
846 if (ps->num_env_old) {
847 memcpy(H11[0][0], H11[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H11[0][0][0]));
848 memcpy(H11[1][0], H11[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H11[1][0][0]));
849 memcpy(H12[0][0], H12[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H12[0][0][0]));
850 memcpy(H12[1][0], H12[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H12[1][0][0]));
851 memcpy(H21[0][0], H21[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H21[0][0][0]));
852 memcpy(H21[1][0], H21[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H21[1][0][0]));
853 memcpy(H22[0][0], H22[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H22[0][0][0]));
854 memcpy(H22[1][0], H22[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H22[1][0][0]));
858 remap34(&iid_mapped, ps->iid_par, ps->nr_iid_par, ps->num_env, 1);
859 remap34(&icc_mapped, ps->icc_par, ps->nr_icc_par, ps->num_env, 1);
860 if (ps->enable_ipdopd) {
861 remap34(&ipd_mapped, ps->ipd_par, ps->nr_ipdopd_par, ps->num_env, 0);
862 remap34(&opd_mapped, ps->opd_par, ps->nr_ipdopd_par, ps->num_env, 0);
864 if (!ps->is34bands_old) {
865 map_val_20_to_34(H11[0][0]);
866 map_val_20_to_34(H11[1][0]);
867 map_val_20_to_34(H12[0][0]);
868 map_val_20_to_34(H12[1][0]);
869 map_val_20_to_34(H21[0][0]);
870 map_val_20_to_34(H21[1][0]);
871 map_val_20_to_34(H22[0][0]);
872 map_val_20_to_34(H22[1][0]);
873 ipdopd_reset(ipd_hist, opd_hist);
876 remap20(&iid_mapped, ps->iid_par, ps->nr_iid_par, ps->num_env, 1);
877 remap20(&icc_mapped, ps->icc_par, ps->nr_icc_par, ps->num_env, 1);
878 if (ps->enable_ipdopd) {
879 remap20(&ipd_mapped, ps->ipd_par, ps->nr_ipdopd_par, ps->num_env, 0);
880 remap20(&opd_mapped, ps->opd_par, ps->nr_ipdopd_par, ps->num_env, 0);
882 if (ps->is34bands_old) {
883 map_val_34_to_20(H11[0][0]);
884 map_val_34_to_20(H11[1][0]);
885 map_val_34_to_20(H12[0][0]);
886 map_val_34_to_20(H12[1][0]);
887 map_val_34_to_20(H21[0][0]);
888 map_val_34_to_20(H21[1][0]);
889 map_val_34_to_20(H22[0][0]);
890 map_val_34_to_20(H22[1][0]);
891 ipdopd_reset(ipd_hist, opd_hist);
896 for (e = 0; e < ps->num_env; e++) {
897 for (b = 0; b < NR_PAR_BANDS[is34]; b++) {
898 INTFLOAT h11, h12, h21, h22;
899 h11 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][0];
900 h12 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][1];
901 h21 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][2];
902 h22 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][3];
904 if (!PS_BASELINE && ps->enable_ipdopd && b < NR_IPDOPD_BANDS[is34]) {
905 //The spec say says to only run this smoother when enable_ipdopd
906 //is set but the reference decoder appears to run it constantly
907 INTFLOAT h11i, h12i, h21i, h22i;
908 INTFLOAT ipd_adj_re, ipd_adj_im;
909 int opd_idx = opd_hist[b] * 8 + opd_mapped[e][b];
910 int ipd_idx = ipd_hist[b] * 8 + ipd_mapped[e][b];
911 INTFLOAT opd_re = pd_re_smooth[opd_idx];
912 INTFLOAT opd_im = pd_im_smooth[opd_idx];
913 INTFLOAT ipd_re = pd_re_smooth[ipd_idx];
914 INTFLOAT ipd_im = pd_im_smooth[ipd_idx];
915 opd_hist[b] = opd_idx & 0x3F;
916 ipd_hist[b] = ipd_idx & 0x3F;
918 ipd_adj_re = AAC_MADD30(opd_re, ipd_re, opd_im, ipd_im);
919 ipd_adj_im = AAC_MSUB30(opd_im, ipd_re, opd_re, ipd_im);
920 h11i = AAC_MUL30(h11, opd_im);
921 h11 = AAC_MUL30(h11, opd_re);
922 h12i = AAC_MUL30(h12, ipd_adj_im);
923 h12 = AAC_MUL30(h12, ipd_adj_re);
924 h21i = AAC_MUL30(h21, opd_im);
925 h21 = AAC_MUL30(h21, opd_re);
926 h22i = AAC_MUL30(h22, ipd_adj_im);
927 h22 = AAC_MUL30(h22, ipd_adj_re);
928 H11[1][e+1][b] = h11i;
929 H12[1][e+1][b] = h12i;
930 H21[1][e+1][b] = h21i;
931 H22[1][e+1][b] = h22i;
933 H11[0][e+1][b] = h11;
934 H12[0][e+1][b] = h12;
935 H21[0][e+1][b] = h21;
936 H22[0][e+1][b] = h22;
938 for (k = 0; k < NR_BANDS[is34]; k++) {
939 LOCAL_ALIGNED_16(INTFLOAT, h, [2], [4]);
940 LOCAL_ALIGNED_16(INTFLOAT, h_step, [2], [4]);
941 int start = ps->border_position[e];
942 int stop = ps->border_position[e+1];
943 INTFLOAT width = Q30(1.f) / ((stop - start) ? (stop - start) : 1);
948 h[0][0] = H11[0][e][b];
949 h[0][1] = H12[0][e][b];
950 h[0][2] = H21[0][e][b];
951 h[0][3] = H22[0][e][b];
952 if (!PS_BASELINE && ps->enable_ipdopd) {
953 //Is this necessary? ps_04_new seems unchanged
954 if ((is34 && k <= 13 && k >= 9) || (!is34 && k <= 1)) {
955 h[1][0] = -H11[1][e][b];
956 h[1][1] = -H12[1][e][b];
957 h[1][2] = -H21[1][e][b];
958 h[1][3] = -H22[1][e][b];
960 h[1][0] = H11[1][e][b];
961 h[1][1] = H12[1][e][b];
962 h[1][2] = H21[1][e][b];
963 h[1][3] = H22[1][e][b];
967 h_step[0][0] = AAC_MSUB31_V3(H11[0][e+1][b], h[0][0], width);
968 h_step[0][1] = AAC_MSUB31_V3(H12[0][e+1][b], h[0][1], width);
969 h_step[0][2] = AAC_MSUB31_V3(H21[0][e+1][b], h[0][2], width);
970 h_step[0][3] = AAC_MSUB31_V3(H22[0][e+1][b], h[0][3], width);
971 if (!PS_BASELINE && ps->enable_ipdopd) {
972 h_step[1][0] = AAC_MSUB31_V3(H11[1][e+1][b], h[1][0], width);
973 h_step[1][1] = AAC_MSUB31_V3(H12[1][e+1][b], h[1][1], width);
974 h_step[1][2] = AAC_MSUB31_V3(H21[1][e+1][b], h[1][2], width);
975 h_step[1][3] = AAC_MSUB31_V3(H22[1][e+1][b], h[1][3], width);
977 ps->dsp.stereo_interpolate[!PS_BASELINE && ps->enable_ipdopd](
978 l[k] + start + 1, r[k] + start + 1,
979 h, h_step, stop - start);
984 int AAC_RENAME(ff_ps_apply)(AVCodecContext *avctx, PSContext *ps, INTFLOAT L[2][38][64], INTFLOAT R[2][38][64], int top)
986 INTFLOAT (*Lbuf)[32][2] = ps->Lbuf;
987 INTFLOAT (*Rbuf)[32][2] = ps->Rbuf;
989 int is34 = ps->is34bands;
991 top += NR_BANDS[is34] - 64;
992 memset(ps->delay+top, 0, (NR_BANDS[is34] - top)*sizeof(ps->delay[0]));
993 if (top < NR_ALLPASS_BANDS[is34])
994 memset(ps->ap_delay + top, 0, (NR_ALLPASS_BANDS[is34] - top)*sizeof(ps->ap_delay[0]));
996 hybrid_analysis(&ps->dsp, Lbuf, ps->in_buf, L, is34, len);
997 decorrelation(ps, Rbuf, (const INTFLOAT (*)[32][2]) Lbuf, is34);
998 stereo_processing(ps, Lbuf, Rbuf, is34);
999 hybrid_synthesis(&ps->dsp, L, Lbuf, is34, len);
1000 hybrid_synthesis(&ps->dsp, R, Rbuf, is34, len);
1005 #define PS_INIT_VLC_STATIC(num, size) \
1006 INIT_VLC_STATIC(&vlc_ps[num], 9, ps_tmp[num].table_size / ps_tmp[num].elem_size, \
1007 ps_tmp[num].ps_bits, 1, 1, \
1008 ps_tmp[num].ps_codes, ps_tmp[num].elem_size, ps_tmp[num].elem_size, \
1011 #define PS_VLC_ROW(name) \
1012 { name ## _codes, name ## _bits, sizeof(name ## _codes), sizeof(name ## _codes[0]) }
1014 av_cold void AAC_RENAME(ff_ps_init)(void) {
1015 // Syntax initialization
1016 static const struct {
1017 const void *ps_codes, *ps_bits;
1018 const unsigned int table_size, elem_size;
1020 PS_VLC_ROW(huff_iid_df1),
1021 PS_VLC_ROW(huff_iid_dt1),
1022 PS_VLC_ROW(huff_iid_df0),
1023 PS_VLC_ROW(huff_iid_dt0),
1024 PS_VLC_ROW(huff_icc_df),
1025 PS_VLC_ROW(huff_icc_dt),
1026 PS_VLC_ROW(huff_ipd_df),
1027 PS_VLC_ROW(huff_ipd_dt),
1028 PS_VLC_ROW(huff_opd_df),
1029 PS_VLC_ROW(huff_opd_dt),
1032 PS_INIT_VLC_STATIC(0, 1544);
1033 PS_INIT_VLC_STATIC(1, 832);
1034 PS_INIT_VLC_STATIC(2, 1024);
1035 PS_INIT_VLC_STATIC(3, 1036);
1036 PS_INIT_VLC_STATIC(4, 544);
1037 PS_INIT_VLC_STATIC(5, 544);
1038 PS_INIT_VLC_STATIC(6, 512);
1039 PS_INIT_VLC_STATIC(7, 512);
1040 PS_INIT_VLC_STATIC(8, 512);
1041 PS_INIT_VLC_STATIC(9, 512);
1046 av_cold void AAC_RENAME(ff_ps_ctx_init)(PSContext *ps)
1048 AAC_RENAME(ff_psdsp_init)(&ps->dsp);