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"
66 #include "libavutil/internal.h"
67 #include "libavutil/libm.h"
68 #include "libavutil/avassert.h"
74 static VLC vlc_sbr[10];
75 static void aacsbr_func_ptr_init(AACSBRContext *c);
76 static const int CONST_LN2 = Q31(0.6931471806/256); // ln(2)/256
77 static const int CONST_RECIP_LN2 = Q31(0.7213475204); // 0.5/ln(2)
78 static const int CONST_076923 = Q31(0.76923076923076923077f);
80 static const int fixed_log_table[10] =
82 Q31(1.0/2), Q31(1.0/3), Q31(1.0/4), Q31(1.0/5), Q31(1.0/6),
83 Q31(1.0/7), Q31(1.0/8), Q31(1.0/9), Q31(1.0/10), Q31(1.0/11)
86 static int fixed_log(int x)
88 int i, ret, xpow, tmp;
92 for (i=0; i<10; i+=2){
93 xpow = (int)(((int64_t)xpow * x + 0x40000000) >> 31);
94 tmp = (int)(((int64_t)xpow * fixed_log_table[i] + 0x40000000) >> 31);
97 xpow = (int)(((int64_t)xpow * x + 0x40000000) >> 31);
98 tmp = (int)(((int64_t)xpow * fixed_log_table[i+1] + 0x40000000) >> 31);
105 static const int fixed_exp_table[7] =
107 Q31(1.0/2), Q31(1.0/6), Q31(1.0/24), Q31(1.0/120),
108 Q31(1.0/720), Q31(1.0/5040), Q31(1.0/40320)
111 static int fixed_exp(int x)
113 int i, ret, xpow, tmp;
118 xpow = (int)(((int64_t)xpow * x + 0x400000) >> 23);
119 tmp = (int)(((int64_t)xpow * fixed_exp_table[i] + 0x40000000) >> 31);
126 static void make_bands(int16_t* bands, int start, int stop, int num_bands)
128 int k, previous, present;
129 int base, prod, nz = 0;
131 base = (stop << 23) / start;
132 while (base < 0x40000000){
136 base = fixed_log(base - 0x80000000);
137 base = (((base + 0x80) >> 8) + (8-nz)*CONST_LN2) / num_bands;
138 base = fixed_exp(base);
143 for (k = 0; k < num_bands-1; k++) {
144 prod = (int)(((int64_t)prod * base + 0x400000) >> 23);
145 present = (prod + 0x400000) >> 23;
146 bands[k] = present - previous;
149 bands[num_bands-1] = stop - previous;
152 /// Dequantization and stereo decoding (14496-3 sp04 p203)
153 static void sbr_dequant(SpectralBandReplication *sbr, int id_aac)
158 if (id_aac == TYPE_CPE && sbr->bs_coupling) {
159 int alpha = sbr->data[0].bs_amp_res ? 2 : 1;
160 int pan_offset = sbr->data[0].bs_amp_res ? 12 : 24;
161 for (e = 1; e <= sbr->data[0].bs_num_env; e++) {
162 for (k = 0; k < sbr->n[sbr->data[0].bs_freq_res[e]]; k++) {
163 SoftFloat temp1, temp2, fac;
165 temp1.exp = sbr->data[0].env_facs_q[e][k] * alpha + 14;
167 temp1.mant = 759250125;
169 temp1.mant = 0x20000000;
170 temp1.exp = (temp1.exp >> 1) + 1;
171 if (temp1.exp > 66) { // temp1 > 1E20
172 av_log(NULL, AV_LOG_ERROR, "envelope scalefactor overflow in dequant\n");
176 temp2.exp = (pan_offset - sbr->data[1].env_facs_q[e][k]) * alpha;
178 temp2.mant = 759250125;
180 temp2.mant = 0x20000000;
181 temp2.exp = (temp2.exp >> 1) + 1;
182 fac = av_div_sf(temp1, av_add_sf(FLOAT_1, temp2));
183 sbr->data[0].env_facs[e][k] = fac;
184 sbr->data[1].env_facs[e][k] = av_mul_sf(fac, temp2);
187 for (e = 1; e <= sbr->data[0].bs_num_noise; e++) {
188 for (k = 0; k < sbr->n_q; k++) {
189 SoftFloat temp1, temp2, fac;
191 temp1.exp = NOISE_FLOOR_OFFSET - \
192 sbr->data[0].noise_facs_q[e][k] + 2;
193 temp1.mant = 0x20000000;
194 av_assert0(temp1.exp <= 66);
195 temp2.exp = 12 - sbr->data[1].noise_facs_q[e][k] + 1;
196 temp2.mant = 0x20000000;
197 fac = av_div_sf(temp1, av_add_sf(FLOAT_1, temp2));
198 sbr->data[0].noise_facs[e][k] = fac;
199 sbr->data[1].noise_facs[e][k] = av_mul_sf(fac, temp2);
202 } else { // SCE or one non-coupled CPE
203 for (ch = 0; ch < (id_aac == TYPE_CPE) + 1; ch++) {
204 int alpha = sbr->data[ch].bs_amp_res ? 2 : 1;
205 for (e = 1; e <= sbr->data[ch].bs_num_env; e++)
206 for (k = 0; k < sbr->n[sbr->data[ch].bs_freq_res[e]]; k++){
209 temp1.exp = alpha * sbr->data[ch].env_facs_q[e][k] + 12;
211 temp1.mant = 759250125;
213 temp1.mant = 0x20000000;
214 temp1.exp = (temp1.exp >> 1) + 1;
215 if (temp1.exp > 66) { // temp1 > 1E20
216 av_log(NULL, AV_LOG_ERROR, "envelope scalefactor overflow in dequant\n");
219 sbr->data[ch].env_facs[e][k] = temp1;
221 for (e = 1; e <= sbr->data[ch].bs_num_noise; e++)
222 for (k = 0; k < sbr->n_q; k++){
223 sbr->data[ch].noise_facs[e][k].exp = NOISE_FLOOR_OFFSET - \
224 sbr->data[ch].noise_facs_q[e][k] + 1;
225 sbr->data[ch].noise_facs[e][k].mant = 0x20000000;
231 /** High Frequency Generation (14496-3 sp04 p214+) and Inverse Filtering
232 * (14496-3 sp04 p214)
233 * Warning: This routine does not seem numerically stable.
235 static void sbr_hf_inverse_filter(SBRDSPContext *dsp,
236 int (*alpha0)[2], int (*alpha1)[2],
237 const int X_low[32][40][2], int k0)
242 for (k = 0; k < k0; k++) {
243 SoftFloat phi[3][2][2];
244 SoftFloat a00, a01, a10, a11;
247 dsp->autocorrelate(X_low[k], phi);
249 dk = av_sub_sf(av_mul_sf(phi[2][1][0], phi[1][0][0]),
250 av_mul_sf(av_add_sf(av_mul_sf(phi[1][1][0], phi[1][1][0]),
251 av_mul_sf(phi[1][1][1], phi[1][1][1])), FLOAT_0999999));
257 SoftFloat temp_real, temp_im;
258 temp_real = av_sub_sf(av_sub_sf(av_mul_sf(phi[0][0][0], phi[1][1][0]),
259 av_mul_sf(phi[0][0][1], phi[1][1][1])),
260 av_mul_sf(phi[0][1][0], phi[1][0][0]));
261 temp_im = av_sub_sf(av_add_sf(av_mul_sf(phi[0][0][0], phi[1][1][1]),
262 av_mul_sf(phi[0][0][1], phi[1][1][0])),
263 av_mul_sf(phi[0][1][1], phi[1][0][0]));
265 a10 = av_div_sf(temp_real, dk);
266 a11 = av_div_sf(temp_im, dk);
269 if (!phi[1][0][0].mant) {
273 SoftFloat temp_real, temp_im;
274 temp_real = av_add_sf(phi[0][0][0],
275 av_add_sf(av_mul_sf(a10, phi[1][1][0]),
276 av_mul_sf(a11, phi[1][1][1])));
277 temp_im = av_add_sf(phi[0][0][1],
278 av_sub_sf(av_mul_sf(a11, phi[1][1][0]),
279 av_mul_sf(a10, phi[1][1][1])));
281 temp_real.mant = -temp_real.mant;
282 temp_im.mant = -temp_im.mant;
283 a00 = av_div_sf(temp_real, phi[1][0][0]);
284 a01 = av_div_sf(temp_im, phi[1][0][0]);
289 alpha0[k][0] = 0x7fffffff;
290 else if (shift <= -30)
295 alpha0[k][0] = a00.mant * (1<<-shift);
297 round = 1 << (shift-1);
298 alpha0[k][0] = (a00.mant + round) >> shift;
304 alpha0[k][1] = 0x7fffffff;
305 else if (shift <= -30)
310 alpha0[k][1] = a01.mant * (1<<-shift);
312 round = 1 << (shift-1);
313 alpha0[k][1] = (a01.mant + round) >> shift;
318 alpha1[k][0] = 0x7fffffff;
319 else if (shift <= -30)
324 alpha1[k][0] = a10.mant * (1<<-shift);
326 round = 1 << (shift-1);
327 alpha1[k][0] = (a10.mant + round) >> shift;
333 alpha1[k][1] = 0x7fffffff;
334 else if (shift <= -30)
339 alpha1[k][1] = a11.mant * (1<<-shift);
341 round = 1 << (shift-1);
342 alpha1[k][1] = (a11.mant + round) >> shift;
346 shift = (int)(((int64_t)(alpha1[k][0]>>1) * (alpha1[k][0]>>1) + \
347 (int64_t)(alpha1[k][1]>>1) * (alpha1[k][1]>>1) + \
349 if (shift >= 0x20000000){
356 shift = (int)(((int64_t)(alpha0[k][0]>>1) * (alpha0[k][0]>>1) + \
357 (int64_t)(alpha0[k][1]>>1) * (alpha0[k][1]>>1) + \
359 if (shift >= 0x20000000){
368 /// Chirp Factors (14496-3 sp04 p214)
369 static void sbr_chirp(SpectralBandReplication *sbr, SBRData *ch_data)
373 static const int bw_tab[] = { 0, 1610612736, 1932735283, 2104533975 };
376 for (i = 0; i < sbr->n_q; i++) {
377 if (ch_data->bs_invf_mode[0][i] + ch_data->bs_invf_mode[1][i] == 1)
380 new_bw = bw_tab[ch_data->bs_invf_mode[0][i]];
382 if (new_bw < ch_data->bw_array[i]){
383 accu = (int64_t)new_bw * 1610612736;
384 accu += (int64_t)ch_data->bw_array[i] * 0x20000000;
385 new_bw = (int)((accu + 0x40000000) >> 31);
387 accu = (int64_t)new_bw * 1946157056;
388 accu += (int64_t)ch_data->bw_array[i] * 201326592;
389 new_bw = (int)((accu + 0x40000000) >> 31);
391 ch_data->bw_array[i] = new_bw < 0x2000000 ? 0 : new_bw;
396 * Calculation of levels of additional HF signal components (14496-3 sp04 p219)
397 * and Calculation of gain (14496-3 sp04 p219)
399 static void sbr_gain_calc(AACContext *ac, SpectralBandReplication *sbr,
400 SBRData *ch_data, const int e_a[2])
403 // max gain limits : -3dB, 0dB, 3dB, inf dB (limiter off)
404 static const SoftFloat limgain[4] = { { 760155524, 0 }, { 0x20000000, 1 },
405 { 758351638, 1 }, { 625000000, 34 } };
407 for (e = 0; e < ch_data->bs_num_env; e++) {
408 int delta = !((e == e_a[1]) || (e == e_a[0]));
409 for (k = 0; k < sbr->n_lim; k++) {
410 SoftFloat gain_boost, gain_max;
412 sum[0] = sum[1] = FLOAT_0;
413 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
414 const SoftFloat temp = av_div_sf(sbr->e_origmapped[e][m],
415 av_add_sf(FLOAT_1, sbr->q_mapped[e][m]));
416 sbr->q_m[e][m] = av_sqrt_sf(av_mul_sf(temp, sbr->q_mapped[e][m]));
417 sbr->s_m[e][m] = av_sqrt_sf(av_mul_sf(temp, av_int2sf(ch_data->s_indexmapped[e + 1][m], 0)));
418 if (!sbr->s_mapped[e][m]) {
420 sbr->gain[e][m] = av_sqrt_sf(av_div_sf(sbr->e_origmapped[e][m],
421 av_mul_sf(av_add_sf(FLOAT_1, sbr->e_curr[e][m]),
422 av_add_sf(FLOAT_1, sbr->q_mapped[e][m]))));
424 sbr->gain[e][m] = av_sqrt_sf(av_div_sf(sbr->e_origmapped[e][m],
425 av_add_sf(FLOAT_1, sbr->e_curr[e][m])));
428 sbr->gain[e][m] = av_sqrt_sf(
430 av_mul_sf(sbr->e_origmapped[e][m], sbr->q_mapped[e][m]),
432 av_add_sf(FLOAT_1, sbr->e_curr[e][m]),
433 av_add_sf(FLOAT_1, sbr->q_mapped[e][m]))));
435 sbr->gain[e][m] = av_add_sf(sbr->gain[e][m], FLOAT_MIN);
437 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
438 sum[0] = av_add_sf(sum[0], sbr->e_origmapped[e][m]);
439 sum[1] = av_add_sf(sum[1], sbr->e_curr[e][m]);
441 gain_max = av_mul_sf(limgain[sbr->bs_limiter_gains],
444 av_add_sf(FLOAT_EPSILON, sum[0]),
445 av_add_sf(FLOAT_EPSILON, sum[1]))));
446 if (av_gt_sf(gain_max, FLOAT_100000))
447 gain_max = FLOAT_100000;
448 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
449 SoftFloat q_m_max = av_div_sf(
450 av_mul_sf(sbr->q_m[e][m], gain_max),
452 if (av_gt_sf(sbr->q_m[e][m], q_m_max))
453 sbr->q_m[e][m] = q_m_max;
454 if (av_gt_sf(sbr->gain[e][m], gain_max))
455 sbr->gain[e][m] = gain_max;
457 sum[0] = sum[1] = FLOAT_0;
458 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
459 sum[0] = av_add_sf(sum[0], sbr->e_origmapped[e][m]);
460 sum[1] = av_add_sf(sum[1],
462 av_mul_sf(sbr->e_curr[e][m],
465 sum[1] = av_add_sf(sum[1],
466 av_mul_sf(sbr->s_m[e][m], sbr->s_m[e][m]));
467 if (delta && !sbr->s_m[e][m].mant)
468 sum[1] = av_add_sf(sum[1],
469 av_mul_sf(sbr->q_m[e][m], sbr->q_m[e][m]));
471 gain_boost = av_sqrt_sf(
473 av_add_sf(FLOAT_EPSILON, sum[0]),
474 av_add_sf(FLOAT_EPSILON, sum[1])));
475 if (av_gt_sf(gain_boost, FLOAT_1584893192))
476 gain_boost = FLOAT_1584893192;
478 for (m = sbr->f_tablelim[k] - sbr->kx[1]; m < sbr->f_tablelim[k + 1] - sbr->kx[1]; m++) {
479 sbr->gain[e][m] = av_mul_sf(sbr->gain[e][m], gain_boost);
480 sbr->q_m[e][m] = av_mul_sf(sbr->q_m[e][m], gain_boost);
481 sbr->s_m[e][m] = av_mul_sf(sbr->s_m[e][m], gain_boost);
487 /// Assembling HF Signals (14496-3 sp04 p220)
488 static void sbr_hf_assemble(int Y1[38][64][2],
489 const int X_high[64][40][2],
490 SpectralBandReplication *sbr, SBRData *ch_data,
494 const int h_SL = 4 * !sbr->bs_smoothing_mode;
495 const int kx = sbr->kx[1];
496 const int m_max = sbr->m[1];
497 static const SoftFloat h_smooth[5] = {
504 SoftFloat (*g_temp)[48] = ch_data->g_temp, (*q_temp)[48] = ch_data->q_temp;
505 int indexnoise = ch_data->f_indexnoise;
506 int indexsine = ch_data->f_indexsine;
509 for (i = 0; i < h_SL; i++) {
510 memcpy(g_temp[i + 2*ch_data->t_env[0]], sbr->gain[0], m_max * sizeof(sbr->gain[0][0]));
511 memcpy(q_temp[i + 2*ch_data->t_env[0]], sbr->q_m[0], m_max * sizeof(sbr->q_m[0][0]));
514 for (i = 0; i < 4; i++) {
515 memcpy(g_temp[i + 2 * ch_data->t_env[0]],
516 g_temp[i + 2 * ch_data->t_env_num_env_old],
518 memcpy(q_temp[i + 2 * ch_data->t_env[0]],
519 q_temp[i + 2 * ch_data->t_env_num_env_old],
524 for (e = 0; e < ch_data->bs_num_env; e++) {
525 for (i = 2 * ch_data->t_env[e]; i < 2 * ch_data->t_env[e + 1]; i++) {
526 memcpy(g_temp[h_SL + i], sbr->gain[e], m_max * sizeof(sbr->gain[0][0]));
527 memcpy(q_temp[h_SL + i], sbr->q_m[e], m_max * sizeof(sbr->q_m[0][0]));
531 for (e = 0; e < ch_data->bs_num_env; e++) {
532 for (i = 2 * ch_data->t_env[e]; i < 2 * ch_data->t_env[e + 1]; i++) {
533 SoftFloat g_filt_tab[48];
534 SoftFloat q_filt_tab[48];
535 SoftFloat *g_filt, *q_filt;
537 if (h_SL && e != e_a[0] && e != e_a[1]) {
540 for (m = 0; m < m_max; m++) {
541 const int idx1 = i + h_SL;
542 g_filt[m].mant = g_filt[m].exp = 0;
543 q_filt[m].mant = q_filt[m].exp = 0;
544 for (j = 0; j <= h_SL; j++) {
545 g_filt[m] = av_add_sf(g_filt[m],
546 av_mul_sf(g_temp[idx1 - j][m],
548 q_filt[m] = av_add_sf(q_filt[m],
549 av_mul_sf(q_temp[idx1 - j][m],
554 g_filt = g_temp[i + h_SL];
558 sbr->dsp.hf_g_filt(Y1[i] + kx, X_high + kx, g_filt, m_max,
559 i + ENVELOPE_ADJUSTMENT_OFFSET);
561 if (e != e_a[0] && e != e_a[1]) {
562 sbr->dsp.hf_apply_noise[indexsine](Y1[i] + kx, sbr->s_m[e],
566 int idx = indexsine&1;
567 int A = (1-((indexsine+(kx & 1))&2));
568 int B = (A^(-idx)) + idx;
569 unsigned *out = &Y1[i][kx][idx];
573 SoftFloat *in = sbr->s_m[e];
574 for (m = 0; m+1 < m_max; m+=2) {
576 shift = 22 - in[m ].exp;
577 shift2= 22 - in[m+1].exp;
578 if (shift < 1 || shift2 < 1) {
579 av_log(NULL, AV_LOG_ERROR, "Overflow in sbr_hf_assemble, shift=%d,%d\n", shift, shift2);
583 round = 1 << (shift-1);
584 out[2*m ] += (int)(in[m ].mant * A + round) >> shift;
588 round = 1 << (shift2-1);
589 out[2*m+2] += (int)(in[m+1].mant * B + round) >> shift2;
594 shift = 22 - in[m ].exp;
596 av_log(NULL, AV_LOG_ERROR, "Overflow in sbr_hf_assemble, shift=%d\n", shift);
598 } else if (shift < 32) {
599 round = 1 << (shift-1);
600 out[2*m ] += (int)(in[m ].mant * A + round) >> shift;
604 indexnoise = (indexnoise + m_max) & 0x1ff;
605 indexsine = (indexsine + 1) & 3;
608 ch_data->f_indexnoise = indexnoise;
609 ch_data->f_indexsine = indexsine;
612 #include "aacsbr_template.c"