3 * MIPS Technologies, Inc., California.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14 * contributors may be used to endorse or promote products derived from
15 * this software without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * AAC Spectral Band Replication decoding functions (fixed-point)
30 * Copyright (c) 2008-2009 Robert Swain ( rob opendot cl )
31 * Copyright (c) 2009-2010 Alex Converse <alex.converse@gmail.com>
33 * This file is part of FFmpeg.
35 * FFmpeg is free software; you can redistribute it and/or
36 * modify it under the terms of the GNU Lesser General Public
37 * License as published by the Free Software Foundation; either
38 * version 2.1 of the License, or (at your option) any later version.
40 * FFmpeg is distributed in the hope that it will be useful,
41 * but WITHOUT ANY WARRANTY; without even the implied warranty of
42 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
43 * Lesser General Public License for more details.
45 * You should have received a copy of the GNU Lesser General Public
46 * License along with FFmpeg; if not, write to the Free Software
47 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
52 * AAC Spectral Band Replication decoding functions (fixed-point)
53 * Note: Rounding-to-nearest used unless otherwise stated
54 * @author Robert Swain ( rob opendot cl )
55 * @author Stanislav Ocovaj ( stanislav.ocovaj imgtec com )
62 #include "aacsbrdata.h"
63 #include "aacsbr_fixed_tablegen.h"
67 #include "libavutil/internal.h"
68 #include "libavutil/libm.h"
69 #include "libavutil/avassert.h"
75 static VLC vlc_sbr[10];
76 static void aacsbr_func_ptr_init(AACSBRContext *c);
77 static const int CONST_LN2 = Q31(0.6931471806/256); // ln(2)/256
78 static const int CONST_RECIP_LN2 = Q31(0.7213475204); // 0.5/ln(2)
79 static const int CONST_076923 = Q31(0.76923076923076923077f);
81 static const int fixed_log_table[10] =
83 Q31(1.0/2), Q31(1.0/3), Q31(1.0/4), Q31(1.0/5), Q31(1.0/6),
84 Q31(1.0/7), Q31(1.0/8), Q31(1.0/9), Q31(1.0/10), Q31(1.0/11)
87 static int fixed_log(int x)
89 int i, ret, xpow, tmp;
93 for (i=0; i<10; i+=2){
94 xpow = (int)(((int64_t)xpow * x + 0x40000000) >> 31);
95 tmp = (int)(((int64_t)xpow * fixed_log_table[i] + 0x40000000) >> 31);
98 xpow = (int)(((int64_t)xpow * x + 0x40000000) >> 31);
99 tmp = (int)(((int64_t)xpow * fixed_log_table[i+1] + 0x40000000) >> 31);
106 static const int fixed_exp_table[7] =
108 Q31(1.0/2), Q31(1.0/6), Q31(1.0/24), Q31(1.0/120),
109 Q31(1.0/720), Q31(1.0/5040), Q31(1.0/40320)
112 static int fixed_exp(int x)
114 int i, ret, xpow, tmp;
119 xpow = (int)(((int64_t)xpow * x + 0x400000) >> 23);
120 tmp = (int)(((int64_t)xpow * fixed_exp_table[i] + 0x40000000) >> 31);
127 static void make_bands(int16_t* bands, int start, int stop, int num_bands)
129 int k, previous, present;
130 int base, prod, nz = 0;
132 base = (stop << 23) / start;
133 while (base < 0x40000000){
137 base = fixed_log(base - 0x80000000);
138 base = (((base + 0x80) >> 8) + (8-nz)*CONST_LN2) / num_bands;
139 base = fixed_exp(base);
144 for (k = 0; k < num_bands-1; k++) {
145 prod = (int)(((int64_t)prod * base + 0x400000) >> 23);
146 present = (prod + 0x400000) >> 23;
147 bands[k] = present - previous;
150 bands[num_bands-1] = stop - previous;
153 /// Dequantization and stereo decoding (14496-3 sp04 p203)
154 static void sbr_dequant(SpectralBandReplication *sbr, int id_aac)
159 if (id_aac == TYPE_CPE && sbr->bs_coupling) {
160 int alpha = sbr->data[0].bs_amp_res ? 2 : 1;
161 int pan_offset = sbr->data[0].bs_amp_res ? 12 : 24;
162 for (e = 1; e <= sbr->data[0].bs_num_env; e++) {
163 for (k = 0; k < sbr->n[sbr->data[0].bs_freq_res[e]]; k++) {
164 SoftFloat temp1, temp2, fac;
166 temp1.exp = sbr->data[0].env_facs_q[e][k] * alpha + 14;
168 temp1.mant = 759250125;
170 temp1.mant = 0x20000000;
171 temp1.exp = (temp1.exp >> 1) + 1;
172 if (temp1.exp > 66) { // temp1 > 1E20
173 av_log(NULL, AV_LOG_ERROR, "envelope scalefactor overflow in dequant\n");
177 temp2.exp = (pan_offset - sbr->data[1].env_facs_q[e][k]) * alpha;
179 temp2.mant = 759250125;
181 temp2.mant = 0x20000000;
182 temp2.exp = (temp2.exp >> 1) + 1;
183 fac = av_div_sf(temp1, av_add_sf(FLOAT_1, temp2));
184 sbr->data[0].env_facs[e][k] = fac;
185 sbr->data[1].env_facs[e][k] = av_mul_sf(fac, temp2);
188 for (e = 1; e <= sbr->data[0].bs_num_noise; e++) {
189 for (k = 0; k < sbr->n_q; k++) {
190 SoftFloat temp1, temp2, fac;
192 temp1.exp = NOISE_FLOOR_OFFSET - \
193 sbr->data[0].noise_facs_q[e][k] + 2;
194 temp1.mant = 0x20000000;
195 av_assert0(temp1.exp <= 66);
196 temp2.exp = 12 - sbr->data[1].noise_facs_q[e][k] + 1;
197 temp2.mant = 0x20000000;
198 fac = av_div_sf(temp1, av_add_sf(FLOAT_1, temp2));
199 sbr->data[0].noise_facs[e][k] = fac;
200 sbr->data[1].noise_facs[e][k] = av_mul_sf(fac, temp2);
203 } else { // SCE or one non-coupled CPE
204 for (ch = 0; ch < (id_aac == TYPE_CPE) + 1; ch++) {
205 int alpha = sbr->data[ch].bs_amp_res ? 2 : 1;
206 for (e = 1; e <= sbr->data[ch].bs_num_env; e++)
207 for (k = 0; k < sbr->n[sbr->data[ch].bs_freq_res[e]]; k++){
210 temp1.exp = alpha * sbr->data[ch].env_facs_q[e][k] + 12;
212 temp1.mant = 759250125;
214 temp1.mant = 0x20000000;
215 temp1.exp = (temp1.exp >> 1) + 1;
216 if (temp1.exp > 66) { // temp1 > 1E20
217 av_log(NULL, AV_LOG_ERROR, "envelope scalefactor overflow in dequant\n");
220 sbr->data[ch].env_facs[e][k] = temp1;
222 for (e = 1; e <= sbr->data[ch].bs_num_noise; e++)
223 for (k = 0; k < sbr->n_q; k++){
224 sbr->data[ch].noise_facs[e][k].exp = NOISE_FLOOR_OFFSET - \
225 sbr->data[ch].noise_facs_q[e][k] + 1;
226 sbr->data[ch].noise_facs[e][k].mant = 0x20000000;
232 /** High Frequency Generation (14496-3 sp04 p214+) and Inverse Filtering
233 * (14496-3 sp04 p214)
234 * Warning: This routine does not seem numerically stable.
236 static void sbr_hf_inverse_filter(SBRDSPContext *dsp,
237 int (*alpha0)[2], int (*alpha1)[2],
238 const int X_low[32][40][2], int k0)
243 for (k = 0; k < k0; k++) {
244 SoftFloat phi[3][2][2];
245 SoftFloat a00, a01, a10, a11;
248 dsp->autocorrelate(X_low[k], phi);
250 dk = av_sub_sf(av_mul_sf(phi[2][1][0], phi[1][0][0]),
251 av_mul_sf(av_add_sf(av_mul_sf(phi[1][1][0], phi[1][1][0]),
252 av_mul_sf(phi[1][1][1], phi[1][1][1])), FLOAT_0999999));
258 SoftFloat temp_real, temp_im;
259 temp_real = av_sub_sf(av_sub_sf(av_mul_sf(phi[0][0][0], phi[1][1][0]),
260 av_mul_sf(phi[0][0][1], phi[1][1][1])),
261 av_mul_sf(phi[0][1][0], phi[1][0][0]));
262 temp_im = av_sub_sf(av_add_sf(av_mul_sf(phi[0][0][0], phi[1][1][1]),
263 av_mul_sf(phi[0][0][1], phi[1][1][0])),
264 av_mul_sf(phi[0][1][1], phi[1][0][0]));
266 a10 = av_div_sf(temp_real, dk);
267 a11 = av_div_sf(temp_im, dk);
270 if (!phi[1][0][0].mant) {
274 SoftFloat temp_real, temp_im;
275 temp_real = av_add_sf(phi[0][0][0],
276 av_add_sf(av_mul_sf(a10, phi[1][1][0]),
277 av_mul_sf(a11, phi[1][1][1])));
278 temp_im = av_add_sf(phi[0][0][1],
279 av_sub_sf(av_mul_sf(a11, phi[1][1][0]),
280 av_mul_sf(a10, phi[1][1][1])));
282 temp_real.mant = -temp_real.mant;
283 temp_im.mant = -temp_im.mant;
284 a00 = av_div_sf(temp_real, phi[1][0][0]);
285 a01 = av_div_sf(temp_im, phi[1][0][0]);
290 alpha0[k][0] = 0x7fffffff;
291 else if (shift <= -30)
296 alpha0[k][0] = a00.mant * (1<<-shift);
298 round = 1 << (shift-1);
299 alpha0[k][0] = (a00.mant + round) >> shift;
305 alpha0[k][1] = 0x7fffffff;
306 else if (shift <= -30)
311 alpha0[k][1] = a01.mant * (1<<-shift);
313 round = 1 << (shift-1);
314 alpha0[k][1] = (a01.mant + round) >> shift;
319 alpha1[k][0] = 0x7fffffff;
320 else if (shift <= -30)
325 alpha1[k][0] = a10.mant * (1<<-shift);
327 round = 1 << (shift-1);
328 alpha1[k][0] = (a10.mant + round) >> shift;
334 alpha1[k][1] = 0x7fffffff;
335 else if (shift <= -30)
340 alpha1[k][1] = a11.mant * (1<<-shift);
342 round = 1 << (shift-1);
343 alpha1[k][1] = (a11.mant + round) >> shift;
347 shift = (int)(((int64_t)(alpha1[k][0]>>1) * (alpha1[k][0]>>1) + \
348 (int64_t)(alpha1[k][1]>>1) * (alpha1[k][1]>>1) + \
350 if (shift >= 0x20000000){
357 shift = (int)(((int64_t)(alpha0[k][0]>>1) * (alpha0[k][0]>>1) + \
358 (int64_t)(alpha0[k][1]>>1) * (alpha0[k][1]>>1) + \
360 if (shift >= 0x20000000){
369 /// Chirp Factors (14496-3 sp04 p214)
370 static void sbr_chirp(SpectralBandReplication *sbr, SBRData *ch_data)
374 static const int bw_tab[] = { 0, 1610612736, 1932735283, 2104533975 };
377 for (i = 0; i < sbr->n_q; i++) {
378 if (ch_data->bs_invf_mode[0][i] + ch_data->bs_invf_mode[1][i] == 1)
381 new_bw = bw_tab[ch_data->bs_invf_mode[0][i]];
383 if (new_bw < ch_data->bw_array[i]){
384 accu = (int64_t)new_bw * 1610612736;
385 accu += (int64_t)ch_data->bw_array[i] * 0x20000000;
386 new_bw = (int)((accu + 0x40000000) >> 31);
388 accu = (int64_t)new_bw * 1946157056;
389 accu += (int64_t)ch_data->bw_array[i] * 201326592;
390 new_bw = (int)((accu + 0x40000000) >> 31);
392 ch_data->bw_array[i] = new_bw < 0x2000000 ? 0 : new_bw;
397 * Calculation of levels of additional HF signal components (14496-3 sp04 p219)
398 * and Calculation of gain (14496-3 sp04 p219)
400 static void sbr_gain_calc(AACContext *ac, SpectralBandReplication *sbr,
401 SBRData *ch_data, const int e_a[2])
404 // max gain limits : -3dB, 0dB, 3dB, inf dB (limiter off)
405 static const SoftFloat limgain[4] = { { 760155524, 0 }, { 0x20000000, 1 },
406 { 758351638, 1 }, { 625000000, 34 } };
408 for (e = 0; e < ch_data->bs_num_env; e++) {
409 int delta = !((e == e_a[1]) || (e == e_a[0]));
410 for (k = 0; k < sbr->n_lim; k++) {
411 SoftFloat gain_boost, gain_max;
413 sum[0] = sum[1] = FLOAT_0;
414 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
415 const SoftFloat temp = av_div_sf(sbr->e_origmapped[e][m],
416 av_add_sf(FLOAT_1, sbr->q_mapped[e][m]));
417 sbr->q_m[e][m] = av_sqrt_sf(av_mul_sf(temp, sbr->q_mapped[e][m]));
418 sbr->s_m[e][m] = av_sqrt_sf(av_mul_sf(temp, av_int2sf(ch_data->s_indexmapped[e + 1][m], 0)));
419 if (!sbr->s_mapped[e][m]) {
421 sbr->gain[e][m] = av_sqrt_sf(av_div_sf(sbr->e_origmapped[e][m],
422 av_mul_sf(av_add_sf(FLOAT_1, sbr->e_curr[e][m]),
423 av_add_sf(FLOAT_1, sbr->q_mapped[e][m]))));
425 sbr->gain[e][m] = av_sqrt_sf(av_div_sf(sbr->e_origmapped[e][m],
426 av_add_sf(FLOAT_1, sbr->e_curr[e][m])));
429 sbr->gain[e][m] = av_sqrt_sf(
431 av_mul_sf(sbr->e_origmapped[e][m], sbr->q_mapped[e][m]),
433 av_add_sf(FLOAT_1, sbr->e_curr[e][m]),
434 av_add_sf(FLOAT_1, sbr->q_mapped[e][m]))));
436 sbr->gain[e][m] = av_add_sf(sbr->gain[e][m], FLOAT_MIN);
438 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
439 sum[0] = av_add_sf(sum[0], sbr->e_origmapped[e][m]);
440 sum[1] = av_add_sf(sum[1], sbr->e_curr[e][m]);
442 gain_max = av_mul_sf(limgain[sbr->bs_limiter_gains],
445 av_add_sf(FLOAT_EPSILON, sum[0]),
446 av_add_sf(FLOAT_EPSILON, sum[1]))));
447 if (av_gt_sf(gain_max, FLOAT_100000))
448 gain_max = FLOAT_100000;
449 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
450 SoftFloat q_m_max = av_div_sf(
451 av_mul_sf(sbr->q_m[e][m], gain_max),
453 if (av_gt_sf(sbr->q_m[e][m], q_m_max))
454 sbr->q_m[e][m] = q_m_max;
455 if (av_gt_sf(sbr->gain[e][m], gain_max))
456 sbr->gain[e][m] = gain_max;
458 sum[0] = sum[1] = FLOAT_0;
459 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
460 sum[0] = av_add_sf(sum[0], sbr->e_origmapped[e][m]);
461 sum[1] = av_add_sf(sum[1],
463 av_mul_sf(sbr->e_curr[e][m],
466 sum[1] = av_add_sf(sum[1],
467 av_mul_sf(sbr->s_m[e][m], sbr->s_m[e][m]));
468 if (delta && !sbr->s_m[e][m].mant)
469 sum[1] = av_add_sf(sum[1],
470 av_mul_sf(sbr->q_m[e][m], sbr->q_m[e][m]));
472 gain_boost = av_sqrt_sf(
474 av_add_sf(FLOAT_EPSILON, sum[0]),
475 av_add_sf(FLOAT_EPSILON, sum[1])));
476 if (av_gt_sf(gain_boost, FLOAT_1584893192))
477 gain_boost = FLOAT_1584893192;
479 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
480 sbr->gain[e][m] = av_mul_sf(sbr->gain[e][m], gain_boost);
481 sbr->q_m[e][m] = av_mul_sf(sbr->q_m[e][m], gain_boost);
482 sbr->s_m[e][m] = av_mul_sf(sbr->s_m[e][m], gain_boost);
488 /// Assembling HF Signals (14496-3 sp04 p220)
489 static void sbr_hf_assemble(int Y1[38][64][2],
490 const int X_high[64][40][2],
491 SpectralBandReplication *sbr, SBRData *ch_data,
495 const int h_SL = 4 * !sbr->bs_smoothing_mode;
496 const int kx = sbr->kx[1];
497 const int m_max = sbr->m[1];
498 static const SoftFloat h_smooth[5] = {
505 SoftFloat (*g_temp)[48] = ch_data->g_temp, (*q_temp)[48] = ch_data->q_temp;
506 int indexnoise = ch_data->f_indexnoise;
507 int indexsine = ch_data->f_indexsine;
510 for (i = 0; i < h_SL; i++) {
511 memcpy(g_temp[i + 2*ch_data->t_env[0]], sbr->gain[0], m_max * sizeof(sbr->gain[0][0]));
512 memcpy(q_temp[i + 2*ch_data->t_env[0]], sbr->q_m[0], m_max * sizeof(sbr->q_m[0][0]));
515 for (i = 0; i < 4; i++) {
516 memcpy(g_temp[i + 2 * ch_data->t_env[0]],
517 g_temp[i + 2 * ch_data->t_env_num_env_old],
519 memcpy(q_temp[i + 2 * ch_data->t_env[0]],
520 q_temp[i + 2 * ch_data->t_env_num_env_old],
525 for (e = 0; e < ch_data->bs_num_env; e++) {
526 for (i = 2 * ch_data->t_env[e]; i < 2 * ch_data->t_env[e + 1]; i++) {
527 memcpy(g_temp[h_SL + i], sbr->gain[e], m_max * sizeof(sbr->gain[0][0]));
528 memcpy(q_temp[h_SL + i], sbr->q_m[e], m_max * sizeof(sbr->q_m[0][0]));
532 for (e = 0; e < ch_data->bs_num_env; e++) {
533 for (i = 2 * ch_data->t_env[e]; i < 2 * ch_data->t_env[e + 1]; i++) {
534 SoftFloat g_filt_tab[48];
535 SoftFloat q_filt_tab[48];
536 SoftFloat *g_filt, *q_filt;
538 if (h_SL && e != e_a[0] && e != e_a[1]) {
541 for (m = 0; m < m_max; m++) {
542 const int idx1 = i + h_SL;
543 g_filt[m].mant = g_filt[m].exp = 0;
544 q_filt[m].mant = q_filt[m].exp = 0;
545 for (j = 0; j <= h_SL; j++) {
546 g_filt[m] = av_add_sf(g_filt[m],
547 av_mul_sf(g_temp[idx1 - j][m],
549 q_filt[m] = av_add_sf(q_filt[m],
550 av_mul_sf(q_temp[idx1 - j][m],
555 g_filt = g_temp[i + h_SL];
559 sbr->dsp.hf_g_filt(Y1[i] + kx, X_high + kx, g_filt, m_max,
560 i + ENVELOPE_ADJUSTMENT_OFFSET);
562 if (e != e_a[0] && e != e_a[1]) {
563 sbr->dsp.hf_apply_noise[indexsine](Y1[i] + kx, sbr->s_m[e],
567 int idx = indexsine&1;
568 int A = (1-((indexsine+(kx & 1))&2));
569 int B = (A^(-idx)) + idx;
570 int *out = &Y1[i][kx][idx];
574 SoftFloat *in = sbr->s_m[e];
575 for (m = 0; m+1 < m_max; m+=2) {
577 shift = 22 - in[m ].exp;
578 shift2= 22 - in[m+1].exp;
579 if (shift < 1 || shift2 < 1) {
580 av_log(NULL, AV_LOG_ERROR, "Overflow in sbr_hf_assemble, shift=%d,%d\n", shift, shift2);
584 round = 1 << (shift-1);
585 out[2*m ] += (int)(in[m ].mant * A + round) >> shift;
589 round = 1 << (shift2-1);
590 out[2*m+2] += (int)(in[m+1].mant * B + round) >> shift2;
595 shift = 22 - in[m ].exp;
597 av_log(NULL, AV_LOG_ERROR, "Overflow in sbr_hf_assemble, shift=%d\n", shift);
599 } else if (shift < 32) {
600 round = 1 << (shift-1);
601 out[2*m ] += (int)(in[m ].mant * A + round) >> shift;
605 indexnoise = (indexnoise + m_max) & 0x1ff;
606 indexsine = (indexsine + 1) & 3;
609 ch_data->f_indexnoise = indexnoise;
610 ch_data->f_indexsine = indexsine;
613 #include "aacsbr_template.c"