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
23 #include "libavutil/common.h"
24 #include "libavutil/mathematics.h"
28 #include "aacps_tablegen.h"
29 #include "aacpsdata.c"
32 #define PS_BASELINE 0 ///< Operate in Baseline PS mode
33 ///< Baseline implies 10 or 20 stereo bands,
34 ///< mixing mode A, and no ipd/opd
36 #define numQMFSlots 32 //numTimeSlots * RATE
38 static const int8_t num_env_tab[2][4] = {
43 static const int8_t nr_iidicc_par_tab[] = {
44 10, 20, 34, 10, 20, 34,
47 static const int8_t nr_iidopd_par_tab[] = {
64 static const int huff_iid[] = {
71 static VLC vlc_ps[10];
73 #define READ_PAR_DATA(PAR, OFFSET, MASK, ERR_CONDITION) \
75 * Read Inter-channel Intensity Difference/Inter-Channel Coherence/ \
76 * Inter-channel Phase Difference/Overall Phase Difference parameters from the \
79 * @param avctx contains the current codec context \
80 * @param gb pointer to the input bitstream \
81 * @param ps pointer to the Parametric Stereo context \
82 * @param PAR pointer to the parameter to be read \
83 * @param e envelope to decode \
84 * @param dt 1: time delta-coded, 0: frequency delta-coded \
86 static int read_ ## PAR ## _data(AVCodecContext *avctx, GetBitContext *gb, PSContext *ps, \
87 int8_t (*PAR)[PS_MAX_NR_IIDICC], int table_idx, int e, int dt) \
89 int b, num = ps->nr_ ## PAR ## _par; \
90 VLC_TYPE (*vlc_table)[2] = vlc_ps[table_idx].table; \
92 int e_prev = e ? e - 1 : ps->num_env_old - 1; \
93 e_prev = FFMAX(e_prev, 0); \
94 for (b = 0; b < num; b++) { \
95 int val = PAR[e_prev][b] + get_vlc2(gb, vlc_table, 9, 3) - OFFSET; \
96 if (MASK) val &= MASK; \
103 for (b = 0; b < num; b++) { \
104 val += get_vlc2(gb, vlc_table, 9, 3) - OFFSET; \
105 if (MASK) val &= MASK; \
113 av_log(avctx, AV_LOG_ERROR, "illegal "#PAR"\n"); \
117 READ_PAR_DATA(iid, huff_offset[table_idx], 0, FFABS(ps->iid_par[e][b]) > 7 + 8 * ps->iid_quant)
118 READ_PAR_DATA(icc, huff_offset[table_idx], 0, ps->icc_par[e][b] > 7U)
119 READ_PAR_DATA(ipdopd, 0, 0x07, 0)
121 static int ps_read_extension_data(GetBitContext *gb, PSContext *ps, int ps_extension_id)
124 int count = get_bits_count(gb);
129 ps->enable_ipdopd = get_bits1(gb);
130 if (ps->enable_ipdopd) {
131 for (e = 0; e < ps->num_env; e++) {
132 int dt = get_bits1(gb);
133 read_ipdopd_data(NULL, gb, ps, ps->ipd_par, dt ? huff_ipd_dt : huff_ipd_df, e, dt);
135 read_ipdopd_data(NULL, gb, ps, ps->opd_par, dt ? huff_opd_dt : huff_opd_df, e, dt);
138 skip_bits1(gb); //reserved_ps
139 return get_bits_count(gb) - count;
142 static void ipdopd_reset(int8_t *ipd_hist, int8_t *opd_hist)
145 for (i = 0; i < PS_MAX_NR_IPDOPD; i++) {
151 int ff_ps_read_data(AVCodecContext *avctx, GetBitContext *gb_host, PSContext *ps, int bits_left)
154 int bit_count_start = get_bits_count(gb_host);
157 GetBitContext gbc = *gb_host, *gb = &gbc;
159 header = get_bits1(gb);
160 if (header) { //enable_ps_header
161 ps->enable_iid = get_bits1(gb);
162 if (ps->enable_iid) {
163 int iid_mode = get_bits(gb, 3);
165 av_log(avctx, AV_LOG_ERROR, "iid_mode %d is reserved.\n",
169 ps->nr_iid_par = nr_iidicc_par_tab[iid_mode];
170 ps->iid_quant = iid_mode > 2;
171 ps->nr_ipdopd_par = nr_iidopd_par_tab[iid_mode];
173 ps->enable_icc = get_bits1(gb);
174 if (ps->enable_icc) {
175 ps->icc_mode = get_bits(gb, 3);
176 if (ps->icc_mode > 5) {
177 av_log(avctx, AV_LOG_ERROR, "icc_mode %d is reserved.\n",
181 ps->nr_icc_par = nr_iidicc_par_tab[ps->icc_mode];
183 ps->enable_ext = get_bits1(gb);
186 ps->frame_class = get_bits1(gb);
187 ps->num_env_old = ps->num_env;
188 ps->num_env = num_env_tab[ps->frame_class][get_bits(gb, 2)];
190 ps->border_position[0] = -1;
191 if (ps->frame_class) {
192 for (e = 1; e <= ps->num_env; e++)
193 ps->border_position[e] = get_bits(gb, 5);
195 for (e = 1; e <= ps->num_env; e++)
196 ps->border_position[e] = (e * numQMFSlots >> ff_log2_tab[ps->num_env]) - 1;
198 if (ps->enable_iid) {
199 for (e = 0; e < ps->num_env; e++) {
200 int dt = get_bits1(gb);
201 if (read_iid_data(avctx, gb, ps, ps->iid_par, huff_iid[2*dt+ps->iid_quant], e, dt))
205 memset(ps->iid_par, 0, sizeof(ps->iid_par));
208 for (e = 0; e < ps->num_env; e++) {
209 int dt = get_bits1(gb);
210 if (read_icc_data(avctx, gb, ps, ps->icc_par, dt ? huff_icc_dt : huff_icc_df, e, dt))
214 memset(ps->icc_par, 0, sizeof(ps->icc_par));
216 if (ps->enable_ext) {
217 int cnt = get_bits(gb, 4);
219 cnt += get_bits(gb, 8);
223 int ps_extension_id = get_bits(gb, 2);
224 cnt -= 2 + ps_read_extension_data(gb, ps, ps_extension_id);
227 av_log(avctx, AV_LOG_ERROR, "ps extension overflow %d\n", cnt);
233 ps->enable_ipdopd &= !PS_BASELINE;
236 if (!ps->num_env || ps->border_position[ps->num_env] < numQMFSlots - 1) {
237 //Create a fake envelope
238 int source = ps->num_env ? ps->num_env - 1 : ps->num_env_old - 1;
240 if (source >= 0 && source != ps->num_env) {
241 if (ps->enable_iid) {
242 memcpy(ps->iid_par+ps->num_env, ps->iid_par+source, sizeof(ps->iid_par[0]));
244 if (ps->enable_icc) {
245 memcpy(ps->icc_par+ps->num_env, ps->icc_par+source, sizeof(ps->icc_par[0]));
247 if (ps->enable_ipdopd) {
248 memcpy(ps->ipd_par+ps->num_env, ps->ipd_par+source, sizeof(ps->ipd_par[0]));
249 memcpy(ps->opd_par+ps->num_env, ps->opd_par+source, sizeof(ps->opd_par[0]));
253 for (b = 0; b < ps->nr_iid_par; b++) {
254 if (FFABS(ps->iid_par[ps->num_env][b]) > 7 + 8 * ps->iid_quant) {
255 av_log(avctx, AV_LOG_ERROR, "iid_par invalid\n");
261 for (b = 0; b < ps->nr_iid_par; b++) {
262 if (ps->icc_par[ps->num_env][b] > 7U) {
263 av_log(avctx, AV_LOG_ERROR, "icc_par invalid\n");
269 ps->border_position[ps->num_env] = numQMFSlots - 1;
273 ps->is34bands_old = ps->is34bands;
274 if (!PS_BASELINE && (ps->enable_iid || ps->enable_icc))
275 ps->is34bands = (ps->enable_iid && ps->nr_iid_par == 34) ||
276 (ps->enable_icc && ps->nr_icc_par == 34);
279 if (!ps->enable_ipdopd) {
280 memset(ps->ipd_par, 0, sizeof(ps->ipd_par));
281 memset(ps->opd_par, 0, sizeof(ps->opd_par));
287 bits_consumed = get_bits_count(gb) - bit_count_start;
288 if (bits_consumed <= bits_left) {
289 skip_bits_long(gb_host, bits_consumed);
290 return bits_consumed;
292 av_log(avctx, AV_LOG_ERROR, "Expected to read %d PS bits actually read %d.\n", bits_left, bits_consumed);
295 skip_bits_long(gb_host, bits_left);
296 memset(ps->iid_par, 0, sizeof(ps->iid_par));
297 memset(ps->icc_par, 0, sizeof(ps->icc_par));
298 memset(ps->ipd_par, 0, sizeof(ps->ipd_par));
299 memset(ps->opd_par, 0, sizeof(ps->opd_par));
303 /** Split one subband into 2 subsubbands with a symmetric real filter.
304 * The filter must have its non-center even coefficients equal to zero. */
305 static void hybrid2_re(float (*in)[2], float (*out)[32][2], const float filter[8], int len, int reverse)
308 for (i = 0; i < len; i++, in++) {
309 float re_in = filter[6] * in[6][0]; //real inphase
310 float re_op = 0.0f; //real out of phase
311 float im_in = filter[6] * in[6][1]; //imag inphase
312 float im_op = 0.0f; //imag out of phase
313 for (j = 0; j < 6; j += 2) {
314 re_op += filter[j+1] * (in[j+1][0] + in[12-j-1][0]);
315 im_op += filter[j+1] * (in[j+1][1] + in[12-j-1][1]);
317 out[ reverse][i][0] = re_in + re_op;
318 out[ reverse][i][1] = im_in + im_op;
319 out[!reverse][i][0] = re_in - re_op;
320 out[!reverse][i][1] = im_in - im_op;
324 /** Split one subband into 6 subsubbands with a complex filter */
325 static void hybrid6_cx(PSDSPContext *dsp, float (*in)[2], float (*out)[32][2], const float (*filter)[8][2], int len)
329 LOCAL_ALIGNED_16(float, temp, [8], [2]);
331 for (i = 0; i < len; i++, in++) {
332 dsp->hybrid_analysis(temp, in, filter, 1, N);
333 out[0][i][0] = temp[6][0];
334 out[0][i][1] = temp[6][1];
335 out[1][i][0] = temp[7][0];
336 out[1][i][1] = temp[7][1];
337 out[2][i][0] = temp[0][0];
338 out[2][i][1] = temp[0][1];
339 out[3][i][0] = temp[1][0];
340 out[3][i][1] = temp[1][1];
341 out[4][i][0] = temp[2][0] + temp[5][0];
342 out[4][i][1] = temp[2][1] + temp[5][1];
343 out[5][i][0] = temp[3][0] + temp[4][0];
344 out[5][i][1] = temp[3][1] + temp[4][1];
348 static void hybrid4_8_12_cx(PSDSPContext *dsp, float (*in)[2], float (*out)[32][2], const float (*filter)[8][2], int N, int len)
352 for (i = 0; i < len; i++, in++) {
353 dsp->hybrid_analysis(out[0] + i, in, filter, 32, N);
357 static void hybrid_analysis(PSDSPContext *dsp, float out[91][32][2],
358 float in[5][44][2], float L[2][38][64],
362 for (i = 0; i < 5; i++) {
363 for (j = 0; j < 38; j++) {
364 in[i][j+6][0] = L[0][j][i];
365 in[i][j+6][1] = L[1][j][i];
369 hybrid4_8_12_cx(dsp, in[0], out, f34_0_12, 12, len);
370 hybrid4_8_12_cx(dsp, in[1], out+12, f34_1_8, 8, len);
371 hybrid4_8_12_cx(dsp, in[2], out+20, f34_2_4, 4, len);
372 hybrid4_8_12_cx(dsp, in[3], out+24, f34_2_4, 4, len);
373 hybrid4_8_12_cx(dsp, in[4], out+28, f34_2_4, 4, len);
374 dsp->hybrid_analysis_ileave(out + 27, L, 5, len);
376 hybrid6_cx(dsp, in[0], out, f20_0_8, len);
377 hybrid2_re(in[1], out+6, g1_Q2, len, 1);
378 hybrid2_re(in[2], out+8, g1_Q2, len, 0);
379 dsp->hybrid_analysis_ileave(out + 7, L, 3, len);
382 for (i = 0; i < 5; i++) {
383 memcpy(in[i], in[i]+32, 6 * sizeof(in[i][0]));
387 static void hybrid_synthesis(PSDSPContext *dsp, float out[2][38][64],
388 float in[91][32][2], int is34, int len)
392 for (n = 0; n < len; n++) {
393 memset(out[0][n], 0, 5*sizeof(out[0][n][0]));
394 memset(out[1][n], 0, 5*sizeof(out[1][n][0]));
395 for (i = 0; i < 12; i++) {
396 out[0][n][0] += in[ i][n][0];
397 out[1][n][0] += in[ i][n][1];
399 for (i = 0; i < 8; i++) {
400 out[0][n][1] += in[12+i][n][0];
401 out[1][n][1] += in[12+i][n][1];
403 for (i = 0; i < 4; i++) {
404 out[0][n][2] += in[20+i][n][0];
405 out[1][n][2] += in[20+i][n][1];
406 out[0][n][3] += in[24+i][n][0];
407 out[1][n][3] += in[24+i][n][1];
408 out[0][n][4] += in[28+i][n][0];
409 out[1][n][4] += in[28+i][n][1];
412 dsp->hybrid_synthesis_deint(out, in + 27, 5, len);
414 for (n = 0; n < len; n++) {
415 out[0][n][0] = in[0][n][0] + in[1][n][0] + in[2][n][0] +
416 in[3][n][0] + in[4][n][0] + in[5][n][0];
417 out[1][n][0] = in[0][n][1] + in[1][n][1] + in[2][n][1] +
418 in[3][n][1] + in[4][n][1] + in[5][n][1];
419 out[0][n][1] = in[6][n][0] + in[7][n][0];
420 out[1][n][1] = in[6][n][1] + in[7][n][1];
421 out[0][n][2] = in[8][n][0] + in[9][n][0];
422 out[1][n][2] = in[8][n][1] + in[9][n][1];
424 dsp->hybrid_synthesis_deint(out, in + 7, 3, len);
428 /// All-pass filter decay slope
429 #define DECAY_SLOPE 0.05f
430 /// Number of frequency bands that can be addressed by the parameter index, b(k)
431 static const int NR_PAR_BANDS[] = { 20, 34 };
432 /// Number of frequency bands that can be addressed by the sub subband index, k
433 static const int NR_BANDS[] = { 71, 91 };
434 /// Start frequency band for the all-pass filter decay slope
435 static const int DECAY_CUTOFF[] = { 10, 32 };
436 /// Number of all-pass filer bands
437 static const int NR_ALLPASS_BANDS[] = { 30, 50 };
438 /// First stereo band using the short one sample delay
439 static const int SHORT_DELAY_BAND[] = { 42, 62 };
442 static void map_idx_10_to_20(int8_t *par_mapped, const int8_t *par, int full)
451 for (; b >= 0; b--) {
452 par_mapped[2*b+1] = par_mapped[2*b] = par[b];
456 static void map_idx_34_to_20(int8_t *par_mapped, const int8_t *par, int full)
458 par_mapped[ 0] = (2*par[ 0] + par[ 1]) / 3;
459 par_mapped[ 1] = ( par[ 1] + 2*par[ 2]) / 3;
460 par_mapped[ 2] = (2*par[ 3] + par[ 4]) / 3;
461 par_mapped[ 3] = ( par[ 4] + 2*par[ 5]) / 3;
462 par_mapped[ 4] = ( par[ 6] + par[ 7]) / 2;
463 par_mapped[ 5] = ( par[ 8] + par[ 9]) / 2;
464 par_mapped[ 6] = par[10];
465 par_mapped[ 7] = par[11];
466 par_mapped[ 8] = ( par[12] + par[13]) / 2;
467 par_mapped[ 9] = ( par[14] + par[15]) / 2;
468 par_mapped[10] = par[16];
470 par_mapped[11] = par[17];
471 par_mapped[12] = par[18];
472 par_mapped[13] = par[19];
473 par_mapped[14] = ( par[20] + par[21]) / 2;
474 par_mapped[15] = ( par[22] + par[23]) / 2;
475 par_mapped[16] = ( par[24] + par[25]) / 2;
476 par_mapped[17] = ( par[26] + par[27]) / 2;
477 par_mapped[18] = ( par[28] + par[29] + par[30] + par[31]) / 4;
478 par_mapped[19] = ( par[32] + par[33]) / 2;
482 static void map_val_34_to_20(float par[PS_MAX_NR_IIDICC])
484 par[ 0] = (2*par[ 0] + par[ 1]) * 0.33333333f;
485 par[ 1] = ( par[ 1] + 2*par[ 2]) * 0.33333333f;
486 par[ 2] = (2*par[ 3] + par[ 4]) * 0.33333333f;
487 par[ 3] = ( par[ 4] + 2*par[ 5]) * 0.33333333f;
488 par[ 4] = ( par[ 6] + par[ 7]) * 0.5f;
489 par[ 5] = ( par[ 8] + par[ 9]) * 0.5f;
492 par[ 8] = ( par[12] + par[13]) * 0.5f;
493 par[ 9] = ( par[14] + par[15]) * 0.5f;
498 par[14] = ( par[20] + par[21]) * 0.5f;
499 par[15] = ( par[22] + par[23]) * 0.5f;
500 par[16] = ( par[24] + par[25]) * 0.5f;
501 par[17] = ( par[26] + par[27]) * 0.5f;
502 par[18] = ( par[28] + par[29] + par[30] + par[31]) * 0.25f;
503 par[19] = ( par[32] + par[33]) * 0.5f;
506 static void map_idx_10_to_34(int8_t *par_mapped, const int8_t *par, int full)
509 par_mapped[33] = par[9];
510 par_mapped[32] = par[9];
511 par_mapped[31] = par[9];
512 par_mapped[30] = par[9];
513 par_mapped[29] = par[9];
514 par_mapped[28] = par[9];
515 par_mapped[27] = par[8];
516 par_mapped[26] = par[8];
517 par_mapped[25] = par[8];
518 par_mapped[24] = par[8];
519 par_mapped[23] = par[7];
520 par_mapped[22] = par[7];
521 par_mapped[21] = par[7];
522 par_mapped[20] = par[7];
523 par_mapped[19] = par[6];
524 par_mapped[18] = par[6];
525 par_mapped[17] = par[5];
526 par_mapped[16] = par[5];
530 par_mapped[15] = par[4];
531 par_mapped[14] = par[4];
532 par_mapped[13] = par[4];
533 par_mapped[12] = par[4];
534 par_mapped[11] = par[3];
535 par_mapped[10] = par[3];
536 par_mapped[ 9] = par[2];
537 par_mapped[ 8] = par[2];
538 par_mapped[ 7] = par[2];
539 par_mapped[ 6] = par[2];
540 par_mapped[ 5] = par[1];
541 par_mapped[ 4] = par[1];
542 par_mapped[ 3] = par[1];
543 par_mapped[ 2] = par[0];
544 par_mapped[ 1] = par[0];
545 par_mapped[ 0] = par[0];
548 static void map_idx_20_to_34(int8_t *par_mapped, const int8_t *par, int full)
551 par_mapped[33] = par[19];
552 par_mapped[32] = par[19];
553 par_mapped[31] = par[18];
554 par_mapped[30] = par[18];
555 par_mapped[29] = par[18];
556 par_mapped[28] = par[18];
557 par_mapped[27] = par[17];
558 par_mapped[26] = par[17];
559 par_mapped[25] = par[16];
560 par_mapped[24] = par[16];
561 par_mapped[23] = par[15];
562 par_mapped[22] = par[15];
563 par_mapped[21] = par[14];
564 par_mapped[20] = par[14];
565 par_mapped[19] = par[13];
566 par_mapped[18] = par[12];
567 par_mapped[17] = par[11];
569 par_mapped[16] = par[10];
570 par_mapped[15] = par[ 9];
571 par_mapped[14] = par[ 9];
572 par_mapped[13] = par[ 8];
573 par_mapped[12] = par[ 8];
574 par_mapped[11] = par[ 7];
575 par_mapped[10] = par[ 6];
576 par_mapped[ 9] = par[ 5];
577 par_mapped[ 8] = par[ 5];
578 par_mapped[ 7] = par[ 4];
579 par_mapped[ 6] = par[ 4];
580 par_mapped[ 5] = par[ 3];
581 par_mapped[ 4] = (par[ 2] + par[ 3]) / 2;
582 par_mapped[ 3] = par[ 2];
583 par_mapped[ 2] = par[ 1];
584 par_mapped[ 1] = (par[ 0] + par[ 1]) / 2;
585 par_mapped[ 0] = par[ 0];
588 static void map_val_20_to_34(float par[PS_MAX_NR_IIDICC])
619 par[ 4] = (par[ 2] + par[ 3]) * 0.5f;
622 par[ 1] = (par[ 0] + par[ 1]) * 0.5f;
625 static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[32][2], int is34)
627 LOCAL_ALIGNED_16(float, power, [34], [PS_QMF_TIME_SLOTS]);
628 LOCAL_ALIGNED_16(float, transient_gain, [34], [PS_QMF_TIME_SLOTS]);
629 float *peak_decay_nrg = ps->peak_decay_nrg;
630 float *power_smooth = ps->power_smooth;
631 float *peak_decay_diff_smooth = ps->peak_decay_diff_smooth;
632 float (*delay)[PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2] = ps->delay;
633 float (*ap_delay)[PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2] = ps->ap_delay;
634 const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20;
635 const float peak_decay_factor = 0.76592833836465f;
636 const float transient_impact = 1.5f;
637 const float a_smooth = 0.25f; ///< Smoothing coefficient
641 memset(power, 0, 34 * sizeof(*power));
643 if (is34 != ps->is34bands_old) {
644 memset(ps->peak_decay_nrg, 0, sizeof(ps->peak_decay_nrg));
645 memset(ps->power_smooth, 0, sizeof(ps->power_smooth));
646 memset(ps->peak_decay_diff_smooth, 0, sizeof(ps->peak_decay_diff_smooth));
647 memset(ps->delay, 0, sizeof(ps->delay));
648 memset(ps->ap_delay, 0, sizeof(ps->ap_delay));
651 for (k = 0; k < NR_BANDS[is34]; k++) {
653 ps->dsp.add_squares(power[i], s[k], nL - n0);
656 //Transient detection
657 for (i = 0; i < NR_PAR_BANDS[is34]; i++) {
658 for (n = n0; n < nL; n++) {
659 float decayed_peak = peak_decay_factor * peak_decay_nrg[i];
661 peak_decay_nrg[i] = FFMAX(decayed_peak, power[i][n]);
662 power_smooth[i] += a_smooth * (power[i][n] - power_smooth[i]);
663 peak_decay_diff_smooth[i] += a_smooth * (peak_decay_nrg[i] - power[i][n] - peak_decay_diff_smooth[i]);
664 denom = transient_impact * peak_decay_diff_smooth[i];
665 transient_gain[i][n] = (denom > power_smooth[i]) ?
666 power_smooth[i] / denom : 1.0f;
670 //Decorrelation and transient reduction
673 // | | Q_fract_allpass[k][m]*z^-link_delay[m] - a[m]*g_decay_slope[k]
674 //H[k][z] = z^-2 * phi_fract[k] * | | ----------------------------------------------------------------
675 // | | 1 - a[m]*g_decay_slope[k]*Q_fract_allpass[k][m]*z^-link_delay[m]
677 //d[k][z] (out) = transient_gain_mapped[k][z] * H[k][z] * s[k][z]
678 for (k = 0; k < NR_ALLPASS_BANDS[is34]; k++) {
680 float g_decay_slope = 1.f - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]);
681 g_decay_slope = av_clipf(g_decay_slope, 0.f, 1.f);
682 memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
683 memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
684 for (m = 0; m < PS_AP_LINKS; m++) {
685 memcpy(ap_delay[k][m], ap_delay[k][m]+numQMFSlots, 5*sizeof(ap_delay[k][m][0]));
687 ps->dsp.decorrelate(out[k], delay[k] + PS_MAX_DELAY - 2, ap_delay[k],
688 phi_fract[is34][k], Q_fract_allpass[is34][k],
689 transient_gain[b], g_decay_slope, nL - n0);
691 for (; k < SHORT_DELAY_BAND[is34]; k++) {
693 memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
694 memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
696 ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 14,
697 transient_gain[i], nL - n0);
699 for (; k < NR_BANDS[is34]; k++) {
701 memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
702 memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
704 ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 1,
705 transient_gain[i], nL - n0);
709 static void remap34(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC],
710 int8_t (*par)[PS_MAX_NR_IIDICC],
711 int num_par, int num_env, int full)
713 int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped;
715 if (num_par == 20 || num_par == 11) {
716 for (e = 0; e < num_env; e++) {
717 map_idx_20_to_34(par_mapped[e], par[e], full);
719 } else if (num_par == 10 || num_par == 5) {
720 for (e = 0; e < num_env; e++) {
721 map_idx_10_to_34(par_mapped[e], par[e], full);
728 static void remap20(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC],
729 int8_t (*par)[PS_MAX_NR_IIDICC],
730 int num_par, int num_env, int full)
732 int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped;
734 if (num_par == 34 || num_par == 17) {
735 for (e = 0; e < num_env; e++) {
736 map_idx_34_to_20(par_mapped[e], par[e], full);
738 } else if (num_par == 10 || num_par == 5) {
739 for (e = 0; e < num_env; e++) {
740 map_idx_10_to_20(par_mapped[e], par[e], full);
747 static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2], int is34)
751 float (*H11)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H11;
752 float (*H12)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H12;
753 float (*H21)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H21;
754 float (*H22)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H22;
755 int8_t *opd_hist = ps->opd_hist;
756 int8_t *ipd_hist = ps->ipd_hist;
757 int8_t iid_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
758 int8_t icc_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
759 int8_t ipd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
760 int8_t opd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
761 int8_t (*iid_mapped)[PS_MAX_NR_IIDICC] = iid_mapped_buf;
762 int8_t (*icc_mapped)[PS_MAX_NR_IIDICC] = icc_mapped_buf;
763 int8_t (*ipd_mapped)[PS_MAX_NR_IIDICC] = ipd_mapped_buf;
764 int8_t (*opd_mapped)[PS_MAX_NR_IIDICC] = opd_mapped_buf;
765 const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20;
766 const float (*H_LUT)[8][4] = (PS_BASELINE || ps->icc_mode < 3) ? HA : HB;
769 if (ps->num_env_old) {
770 memcpy(H11[0][0], H11[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H11[0][0][0]));
771 memcpy(H11[1][0], H11[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H11[1][0][0]));
772 memcpy(H12[0][0], H12[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H12[0][0][0]));
773 memcpy(H12[1][0], H12[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H12[1][0][0]));
774 memcpy(H21[0][0], H21[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H21[0][0][0]));
775 memcpy(H21[1][0], H21[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H21[1][0][0]));
776 memcpy(H22[0][0], H22[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H22[0][0][0]));
777 memcpy(H22[1][0], H22[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H22[1][0][0]));
781 remap34(&iid_mapped, ps->iid_par, ps->nr_iid_par, ps->num_env, 1);
782 remap34(&icc_mapped, ps->icc_par, ps->nr_icc_par, ps->num_env, 1);
783 if (ps->enable_ipdopd) {
784 remap34(&ipd_mapped, ps->ipd_par, ps->nr_ipdopd_par, ps->num_env, 0);
785 remap34(&opd_mapped, ps->opd_par, ps->nr_ipdopd_par, ps->num_env, 0);
787 if (!ps->is34bands_old) {
788 map_val_20_to_34(H11[0][0]);
789 map_val_20_to_34(H11[1][0]);
790 map_val_20_to_34(H12[0][0]);
791 map_val_20_to_34(H12[1][0]);
792 map_val_20_to_34(H21[0][0]);
793 map_val_20_to_34(H21[1][0]);
794 map_val_20_to_34(H22[0][0]);
795 map_val_20_to_34(H22[1][0]);
796 ipdopd_reset(ipd_hist, opd_hist);
799 remap20(&iid_mapped, ps->iid_par, ps->nr_iid_par, ps->num_env, 1);
800 remap20(&icc_mapped, ps->icc_par, ps->nr_icc_par, ps->num_env, 1);
801 if (ps->enable_ipdopd) {
802 remap20(&ipd_mapped, ps->ipd_par, ps->nr_ipdopd_par, ps->num_env, 0);
803 remap20(&opd_mapped, ps->opd_par, ps->nr_ipdopd_par, ps->num_env, 0);
805 if (ps->is34bands_old) {
806 map_val_34_to_20(H11[0][0]);
807 map_val_34_to_20(H11[1][0]);
808 map_val_34_to_20(H12[0][0]);
809 map_val_34_to_20(H12[1][0]);
810 map_val_34_to_20(H21[0][0]);
811 map_val_34_to_20(H21[1][0]);
812 map_val_34_to_20(H22[0][0]);
813 map_val_34_to_20(H22[1][0]);
814 ipdopd_reset(ipd_hist, opd_hist);
819 for (e = 0; e < ps->num_env; e++) {
820 for (b = 0; b < NR_PAR_BANDS[is34]; b++) {
821 float h11, h12, h21, h22;
822 h11 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][0];
823 h12 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][1];
824 h21 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][2];
825 h22 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][3];
826 if (!PS_BASELINE && ps->enable_ipdopd && b < ps->nr_ipdopd_par) {
827 //The spec say says to only run this smoother when enable_ipdopd
828 //is set but the reference decoder appears to run it constantly
829 float h11i, h12i, h21i, h22i;
830 float ipd_adj_re, ipd_adj_im;
831 int opd_idx = opd_hist[b] * 8 + opd_mapped[e][b];
832 int ipd_idx = ipd_hist[b] * 8 + ipd_mapped[e][b];
833 float opd_re = pd_re_smooth[opd_idx];
834 float opd_im = pd_im_smooth[opd_idx];
835 float ipd_re = pd_re_smooth[ipd_idx];
836 float ipd_im = pd_im_smooth[ipd_idx];
837 opd_hist[b] = opd_idx & 0x3F;
838 ipd_hist[b] = ipd_idx & 0x3F;
840 ipd_adj_re = opd_re*ipd_re + opd_im*ipd_im;
841 ipd_adj_im = opd_im*ipd_re - opd_re*ipd_im;
844 h12i = h12 * ipd_adj_im;
845 h12 = h12 * ipd_adj_re;
848 h22i = h22 * ipd_adj_im;
849 h22 = h22 * ipd_adj_re;
850 H11[1][e+1][b] = h11i;
851 H12[1][e+1][b] = h12i;
852 H21[1][e+1][b] = h21i;
853 H22[1][e+1][b] = h22i;
855 H11[0][e+1][b] = h11;
856 H12[0][e+1][b] = h12;
857 H21[0][e+1][b] = h21;
858 H22[0][e+1][b] = h22;
860 for (k = 0; k < NR_BANDS[is34]; k++) {
863 int start = ps->border_position[e];
864 int stop = ps->border_position[e+1];
865 float width = 1.f / (stop - start);
867 h[0][0] = H11[0][e][b];
868 h[0][1] = H12[0][e][b];
869 h[0][2] = H21[0][e][b];
870 h[0][3] = H22[0][e][b];
871 if (!PS_BASELINE && ps->enable_ipdopd) {
872 //Is this necessary? ps_04_new seems unchanged
873 if ((is34 && k <= 13 && k >= 9) || (!is34 && k <= 1)) {
874 h[1][0] = -H11[1][e][b];
875 h[1][1] = -H12[1][e][b];
876 h[1][2] = -H21[1][e][b];
877 h[1][3] = -H22[1][e][b];
879 h[1][0] = H11[1][e][b];
880 h[1][1] = H12[1][e][b];
881 h[1][2] = H21[1][e][b];
882 h[1][3] = H22[1][e][b];
886 h_step[0][0] = (H11[0][e+1][b] - h[0][0]) * width;
887 h_step[0][1] = (H12[0][e+1][b] - h[0][1]) * width;
888 h_step[0][2] = (H21[0][e+1][b] - h[0][2]) * width;
889 h_step[0][3] = (H22[0][e+1][b] - h[0][3]) * width;
890 if (!PS_BASELINE && ps->enable_ipdopd) {
891 h_step[1][0] = (H11[1][e+1][b] - h[1][0]) * width;
892 h_step[1][1] = (H12[1][e+1][b] - h[1][1]) * width;
893 h_step[1][2] = (H21[1][e+1][b] - h[1][2]) * width;
894 h_step[1][3] = (H22[1][e+1][b] - h[1][3]) * width;
896 ps->dsp.stereo_interpolate[!PS_BASELINE && ps->enable_ipdopd](
897 l[k] + start + 1, r[k] + start + 1,
898 h, h_step, stop - start);
903 int ff_ps_apply(AVCodecContext *avctx, PSContext *ps, float L[2][38][64], float R[2][38][64], int top)
905 LOCAL_ALIGNED_16(float, Lbuf, [91], [32][2]);
906 LOCAL_ALIGNED_16(float, Rbuf, [91], [32][2]);
908 int is34 = ps->is34bands;
910 top += NR_BANDS[is34] - 64;
911 memset(ps->delay+top, 0, (NR_BANDS[is34] - top)*sizeof(ps->delay[0]));
912 if (top < NR_ALLPASS_BANDS[is34])
913 memset(ps->ap_delay + top, 0, (NR_ALLPASS_BANDS[is34] - top)*sizeof(ps->ap_delay[0]));
915 hybrid_analysis(&ps->dsp, Lbuf, ps->in_buf, L, is34, len);
916 decorrelation(ps, Rbuf, Lbuf, is34);
917 stereo_processing(ps, Lbuf, Rbuf, is34);
918 hybrid_synthesis(&ps->dsp, L, Lbuf, is34, len);
919 hybrid_synthesis(&ps->dsp, R, Rbuf, is34, len);
924 #define PS_INIT_VLC_STATIC(num, size) \
925 INIT_VLC_STATIC(&vlc_ps[num], 9, ps_tmp[num].table_size / ps_tmp[num].elem_size, \
926 ps_tmp[num].ps_bits, 1, 1, \
927 ps_tmp[num].ps_codes, ps_tmp[num].elem_size, ps_tmp[num].elem_size, \
930 #define PS_VLC_ROW(name) \
931 { name ## _codes, name ## _bits, sizeof(name ## _codes), sizeof(name ## _codes[0]) }
933 av_cold void ff_ps_init(void) {
934 // Syntax initialization
935 static const struct {
936 const void *ps_codes, *ps_bits;
937 const unsigned int table_size, elem_size;
939 PS_VLC_ROW(huff_iid_df1),
940 PS_VLC_ROW(huff_iid_dt1),
941 PS_VLC_ROW(huff_iid_df0),
942 PS_VLC_ROW(huff_iid_dt0),
943 PS_VLC_ROW(huff_icc_df),
944 PS_VLC_ROW(huff_icc_dt),
945 PS_VLC_ROW(huff_ipd_df),
946 PS_VLC_ROW(huff_ipd_dt),
947 PS_VLC_ROW(huff_opd_df),
948 PS_VLC_ROW(huff_opd_dt),
951 PS_INIT_VLC_STATIC(0, 1544);
952 PS_INIT_VLC_STATIC(1, 832);
953 PS_INIT_VLC_STATIC(2, 1024);
954 PS_INIT_VLC_STATIC(3, 1036);
955 PS_INIT_VLC_STATIC(4, 544);
956 PS_INIT_VLC_STATIC(5, 544);
957 PS_INIT_VLC_STATIC(6, 512);
958 PS_INIT_VLC_STATIC(7, 512);
959 PS_INIT_VLC_STATIC(8, 512);
960 PS_INIT_VLC_STATIC(9, 512);
965 av_cold void ff_ps_ctx_init(PSContext *ps)
967 ff_psdsp_init(&ps->dsp);