2 * AAC Spectral Band Replication decoding functions
3 * Copyright (c) 2008-2009 Robert Swain ( rob opendot cl )
4 * Copyright (c) 2009-2010 Alex Converse <alex.converse@gmail.com>
8 * MIPS Technologies, Inc., California.
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
29 * AAC Spectral Band Replication decoding functions
30 * @author Robert Swain ( rob opendot cl )
31 * @author Stanislav Ocovaj ( stanislav.ocovaj@imgtec.com )
32 * @author Zoran Basaric ( zoran.basaric@imgtec.com )
35 #include "libavutil/qsort.h"
37 av_cold void AAC_RENAME(ff_aac_sbr_init)(void)
40 const void *sbr_codes, *sbr_bits;
41 const unsigned int table_size, elem_size;
43 SBR_VLC_ROW(t_huffman_env_1_5dB),
44 SBR_VLC_ROW(f_huffman_env_1_5dB),
45 SBR_VLC_ROW(t_huffman_env_bal_1_5dB),
46 SBR_VLC_ROW(f_huffman_env_bal_1_5dB),
47 SBR_VLC_ROW(t_huffman_env_3_0dB),
48 SBR_VLC_ROW(f_huffman_env_3_0dB),
49 SBR_VLC_ROW(t_huffman_env_bal_3_0dB),
50 SBR_VLC_ROW(f_huffman_env_bal_3_0dB),
51 SBR_VLC_ROW(t_huffman_noise_3_0dB),
52 SBR_VLC_ROW(t_huffman_noise_bal_3_0dB),
55 // SBR VLC table initialization
56 SBR_INIT_VLC_STATIC(0, 1098);
57 SBR_INIT_VLC_STATIC(1, 1092);
58 SBR_INIT_VLC_STATIC(2, 768);
59 SBR_INIT_VLC_STATIC(3, 1026);
60 SBR_INIT_VLC_STATIC(4, 1058);
61 SBR_INIT_VLC_STATIC(5, 1052);
62 SBR_INIT_VLC_STATIC(6, 544);
63 SBR_INIT_VLC_STATIC(7, 544);
64 SBR_INIT_VLC_STATIC(8, 592);
65 SBR_INIT_VLC_STATIC(9, 512);
69 AAC_RENAME(ff_ps_init)();
72 /** Places SBR in pure upsampling mode. */
73 static void sbr_turnoff(SpectralBandReplication *sbr) {
75 sbr->ready_for_dequant = 0;
76 // Init defults used in pure upsampling mode
77 sbr->kx[1] = 32; //Typo in spec, kx' inits to 32
79 // Reset values for first SBR header
80 sbr->data[0].e_a[1] = sbr->data[1].e_a[1] = -1;
81 memset(&sbr->spectrum_params, -1, sizeof(SpectrumParameters));
84 av_cold void AAC_RENAME(ff_aac_sbr_ctx_init)(AACContext *ac, SpectralBandReplication *sbr, int id_aac)
86 if(sbr->mdct.mdct_bits)
88 sbr->kx[0] = sbr->kx[1];
91 sbr->data[0].synthesis_filterbank_samples_offset = SBR_SYNTHESIS_BUF_SIZE - (1280 - 128);
92 sbr->data[1].synthesis_filterbank_samples_offset = SBR_SYNTHESIS_BUF_SIZE - (1280 - 128);
93 /* SBR requires samples to be scaled to +/-32768.0 to work correctly.
94 * mdct scale factors are adjusted to scale up from +/-1.0 at analysis
95 * and scale back down at synthesis. */
96 AAC_RENAME_32(ff_mdct_init)(&sbr->mdct, 7, 1, 1.0 / (64 * 32768.0));
97 AAC_RENAME_32(ff_mdct_init)(&sbr->mdct_ana, 7, 1, -2.0 * 32768.0);
98 AAC_RENAME(ff_ps_ctx_init)(&sbr->ps);
99 AAC_RENAME(ff_sbrdsp_init)(&sbr->dsp);
100 aacsbr_func_ptr_init(&sbr->c);
103 av_cold void AAC_RENAME(ff_aac_sbr_ctx_close)(SpectralBandReplication *sbr)
105 AAC_RENAME_32(ff_mdct_end)(&sbr->mdct);
106 AAC_RENAME_32(ff_mdct_end)(&sbr->mdct_ana);
109 static int qsort_comparison_function_int16(const void *a, const void *b)
111 return *(const int16_t *)a - *(const int16_t *)b;
114 static inline int in_table_int16(const int16_t *table, int last_el, int16_t needle)
117 for (i = 0; i <= last_el; i++)
118 if (table[i] == needle)
123 /// Limiter Frequency Band Table (14496-3 sp04 p198)
124 static void sbr_make_f_tablelim(SpectralBandReplication *sbr)
127 if (sbr->bs_limiter_bands > 0) {
128 static const INTFLOAT bands_warped[3] = { Q23(1.32715174233856803909f), //2^(0.49/1.2)
129 Q23(1.18509277094158210129f), //2^(0.49/2)
130 Q23(1.11987160404675912501f) }; //2^(0.49/3)
131 const INTFLOAT lim_bands_per_octave_warped = bands_warped[sbr->bs_limiter_bands - 1];
132 int16_t patch_borders[7];
133 uint16_t *in = sbr->f_tablelim + 1, *out = sbr->f_tablelim;
135 patch_borders[0] = sbr->kx[1];
136 for (k = 1; k <= sbr->num_patches; k++)
137 patch_borders[k] = patch_borders[k-1] + sbr->patch_num_subbands[k-1];
139 memcpy(sbr->f_tablelim, sbr->f_tablelow,
140 (sbr->n[0] + 1) * sizeof(sbr->f_tablelow[0]));
141 if (sbr->num_patches > 1)
142 memcpy(sbr->f_tablelim + sbr->n[0] + 1, patch_borders + 1,
143 (sbr->num_patches - 1) * sizeof(patch_borders[0]));
145 AV_QSORT(sbr->f_tablelim, sbr->num_patches + sbr->n[0],
147 qsort_comparison_function_int16);
149 sbr->n_lim = sbr->n[0] + sbr->num_patches - 1;
150 while (out < sbr->f_tablelim + sbr->n_lim) {
152 if ((*in << 23) >= *out * lim_bands_per_octave_warped) {
154 if (*in >= *out * lim_bands_per_octave_warped) {
155 #endif /* USE_FIXED */
157 } else if (*in == *out ||
158 !in_table_int16(patch_borders, sbr->num_patches, *in)) {
161 } else if (!in_table_int16(patch_borders, sbr->num_patches, *out)) {
169 sbr->f_tablelim[0] = sbr->f_tablelow[0];
170 sbr->f_tablelim[1] = sbr->f_tablelow[sbr->n[0]];
175 static unsigned int read_sbr_header(SpectralBandReplication *sbr, GetBitContext *gb)
177 unsigned int cnt = get_bits_count(gb);
178 uint8_t bs_header_extra_1;
179 uint8_t bs_header_extra_2;
180 int old_bs_limiter_bands = sbr->bs_limiter_bands;
181 SpectrumParameters old_spectrum_params;
184 sbr->ready_for_dequant = 0;
186 // Save last spectrum parameters variables to compare to new ones
187 memcpy(&old_spectrum_params, &sbr->spectrum_params, sizeof(SpectrumParameters));
189 sbr->bs_amp_res_header = get_bits1(gb);
190 sbr->spectrum_params.bs_start_freq = get_bits(gb, 4);
191 sbr->spectrum_params.bs_stop_freq = get_bits(gb, 4);
192 sbr->spectrum_params.bs_xover_band = get_bits(gb, 3);
193 skip_bits(gb, 2); // bs_reserved
195 bs_header_extra_1 = get_bits1(gb);
196 bs_header_extra_2 = get_bits1(gb);
198 if (bs_header_extra_1) {
199 sbr->spectrum_params.bs_freq_scale = get_bits(gb, 2);
200 sbr->spectrum_params.bs_alter_scale = get_bits1(gb);
201 sbr->spectrum_params.bs_noise_bands = get_bits(gb, 2);
203 sbr->spectrum_params.bs_freq_scale = 2;
204 sbr->spectrum_params.bs_alter_scale = 1;
205 sbr->spectrum_params.bs_noise_bands = 2;
208 // Check if spectrum parameters changed
209 if (memcmp(&old_spectrum_params, &sbr->spectrum_params, sizeof(SpectrumParameters)))
212 if (bs_header_extra_2) {
213 sbr->bs_limiter_bands = get_bits(gb, 2);
214 sbr->bs_limiter_gains = get_bits(gb, 2);
215 sbr->bs_interpol_freq = get_bits1(gb);
216 sbr->bs_smoothing_mode = get_bits1(gb);
218 sbr->bs_limiter_bands = 2;
219 sbr->bs_limiter_gains = 2;
220 sbr->bs_interpol_freq = 1;
221 sbr->bs_smoothing_mode = 1;
224 if (sbr->bs_limiter_bands != old_bs_limiter_bands && !sbr->reset)
225 sbr_make_f_tablelim(sbr);
227 return get_bits_count(gb) - cnt;
230 static int array_min_int16(const int16_t *array, int nel)
232 int i, min = array[0];
233 for (i = 1; i < nel; i++)
234 min = FFMIN(array[i], min);
238 static int check_n_master(AVCodecContext *avctx, int n_master, int bs_xover_band)
240 // Requirements (14496-3 sp04 p205)
242 av_log(avctx, AV_LOG_ERROR, "Invalid n_master: %d\n", n_master);
245 if (bs_xover_band >= n_master) {
246 av_log(avctx, AV_LOG_ERROR,
247 "Invalid bitstream, crossover band index beyond array bounds: %d\n",
254 /// Master Frequency Band Table (14496-3 sp04 p194)
255 static int sbr_make_f_master(AACContext *ac, SpectralBandReplication *sbr,
256 SpectrumParameters *spectrum)
258 unsigned int temp, max_qmf_subbands = 0;
259 unsigned int start_min, stop_min;
261 const int8_t *sbr_offset_ptr;
264 if (sbr->sample_rate < 32000) {
266 } else if (sbr->sample_rate < 64000) {
271 switch (sbr->sample_rate) {
273 sbr_offset_ptr = sbr_offset[0];
276 sbr_offset_ptr = sbr_offset[1];
279 sbr_offset_ptr = sbr_offset[2];
282 sbr_offset_ptr = sbr_offset[3];
284 case 44100: case 48000: case 64000:
285 sbr_offset_ptr = sbr_offset[4];
287 case 88200: case 96000: case 128000: case 176400: case 192000:
288 sbr_offset_ptr = sbr_offset[5];
291 av_log(ac->avctx, AV_LOG_ERROR,
292 "Unsupported sample rate for SBR: %d\n", sbr->sample_rate);
296 start_min = ((temp << 7) + (sbr->sample_rate >> 1)) / sbr->sample_rate;
297 stop_min = ((temp << 8) + (sbr->sample_rate >> 1)) / sbr->sample_rate;
299 sbr->k[0] = start_min + sbr_offset_ptr[spectrum->bs_start_freq];
301 if (spectrum->bs_stop_freq < 14) {
302 sbr->k[2] = stop_min;
303 make_bands(stop_dk, stop_min, 64, 13);
304 AV_QSORT(stop_dk, 13, int16_t, qsort_comparison_function_int16);
305 for (k = 0; k < spectrum->bs_stop_freq; k++)
306 sbr->k[2] += stop_dk[k];
307 } else if (spectrum->bs_stop_freq == 14) {
308 sbr->k[2] = 2*sbr->k[0];
309 } else if (spectrum->bs_stop_freq == 15) {
310 sbr->k[2] = 3*sbr->k[0];
312 av_log(ac->avctx, AV_LOG_ERROR,
313 "Invalid bs_stop_freq: %d\n", spectrum->bs_stop_freq);
316 sbr->k[2] = FFMIN(64, sbr->k[2]);
318 // Requirements (14496-3 sp04 p205)
319 if (sbr->sample_rate <= 32000) {
320 max_qmf_subbands = 48;
321 } else if (sbr->sample_rate == 44100) {
322 max_qmf_subbands = 35;
323 } else if (sbr->sample_rate >= 48000)
324 max_qmf_subbands = 32;
328 if (sbr->k[2] - sbr->k[0] > max_qmf_subbands) {
329 av_log(ac->avctx, AV_LOG_ERROR,
330 "Invalid bitstream, too many QMF subbands: %d\n", sbr->k[2] - sbr->k[0]);
334 if (!spectrum->bs_freq_scale) {
337 dk = spectrum->bs_alter_scale + 1;
338 sbr->n_master = ((sbr->k[2] - sbr->k[0] + (dk&2)) >> dk) << 1;
339 if (check_n_master(ac->avctx, sbr->n_master, sbr->spectrum_params.bs_xover_band))
342 for (k = 1; k <= sbr->n_master; k++)
343 sbr->f_master[k] = dk;
345 k2diff = sbr->k[2] - sbr->k[0] - sbr->n_master * dk;
348 sbr->f_master[2]-= (k2diff < -1);
350 sbr->f_master[sbr->n_master]++;
353 sbr->f_master[0] = sbr->k[0];
354 for (k = 1; k <= sbr->n_master; k++)
355 sbr->f_master[k] += sbr->f_master[k - 1];
358 int half_bands = 7 - spectrum->bs_freq_scale; // bs_freq_scale = {1,2,3}
359 int two_regions, num_bands_0;
360 int vdk0_max, vdk1_min;
364 #endif /* USE_FIXED */
366 if (49 * sbr->k[2] > 110 * sbr->k[0]) {
368 sbr->k[1] = 2 * sbr->k[0];
371 sbr->k[1] = sbr->k[2];
375 tmp = (sbr->k[1] << 23) / sbr->k[0];
376 while (tmp < 0x40000000) {
380 tmp = fixed_log(tmp - 0x80000000);
381 tmp = (int)(((int64_t)tmp * CONST_RECIP_LN2 + 0x20000000) >> 30);
382 tmp = (((tmp + 0x80) >> 8) + ((8 - nz) << 23)) * half_bands;
383 num_bands_0 = ((tmp + 0x400000) >> 23) * 2;
385 num_bands_0 = lrintf(half_bands * log2f(sbr->k[1] / (float)sbr->k[0])) * 2;
386 #endif /* USE_FIXED */
388 if (num_bands_0 <= 0) { // Requirements (14496-3 sp04 p205)
389 av_log(ac->avctx, AV_LOG_ERROR, "Invalid num_bands_0: %d\n", num_bands_0);
395 make_bands(vk0+1, sbr->k[0], sbr->k[1], num_bands_0);
397 AV_QSORT(vk0 + 1, num_bands_0, int16_t, qsort_comparison_function_int16);
398 vdk0_max = vk0[num_bands_0];
401 for (k = 1; k <= num_bands_0; k++) {
402 if (vk0[k] <= 0) { // Requirements (14496-3 sp04 p205)
403 av_log(ac->avctx, AV_LOG_ERROR, "Invalid vDk0[%d]: %d\n", k, vk0[k]);
414 tmp = (sbr->k[2] << 23) / sbr->k[1];
416 while (tmp < 0x40000000) {
420 tmp = fixed_log(tmp - 0x80000000);
421 tmp = (int)(((int64_t)tmp * CONST_RECIP_LN2 + 0x20000000) >> 30);
422 tmp = (((tmp + 0x80) >> 8) + ((8 - nz) << 23)) * half_bands;
423 if (spectrum->bs_alter_scale)
424 tmp = (int)(((int64_t)tmp * CONST_076923 + 0x40000000) >> 31);
425 num_bands_1 = ((tmp + 0x400000) >> 23) * 2;
427 float invwarp = spectrum->bs_alter_scale ? 0.76923076923076923077f
428 : 1.0f; // bs_alter_scale = {0,1}
429 int num_bands_1 = lrintf(half_bands * invwarp *
430 log2f(sbr->k[2] / (float)sbr->k[1])) * 2;
431 #endif /* USE_FIXED */
432 make_bands(vk1+1, sbr->k[1], sbr->k[2], num_bands_1);
434 vdk1_min = array_min_int16(vk1 + 1, num_bands_1);
436 if (vdk1_min < vdk0_max) {
438 AV_QSORT(vk1 + 1, num_bands_1, int16_t, qsort_comparison_function_int16);
439 change = FFMIN(vdk0_max - vk1[1], (vk1[num_bands_1] - vk1[1]) >> 1);
441 vk1[num_bands_1] -= change;
444 AV_QSORT(vk1 + 1, num_bands_1, int16_t, qsort_comparison_function_int16);
447 for (k = 1; k <= num_bands_1; k++) {
448 if (vk1[k] <= 0) { // Requirements (14496-3 sp04 p205)
449 av_log(ac->avctx, AV_LOG_ERROR, "Invalid vDk1[%d]: %d\n", k, vk1[k]);
455 sbr->n_master = num_bands_0 + num_bands_1;
456 if (check_n_master(ac->avctx, sbr->n_master, sbr->spectrum_params.bs_xover_band))
458 memcpy(&sbr->f_master[0], vk0,
459 (num_bands_0 + 1) * sizeof(sbr->f_master[0]));
460 memcpy(&sbr->f_master[num_bands_0 + 1], vk1 + 1,
461 num_bands_1 * sizeof(sbr->f_master[0]));
464 sbr->n_master = num_bands_0;
465 if (check_n_master(ac->avctx, sbr->n_master, sbr->spectrum_params.bs_xover_band))
467 memcpy(sbr->f_master, vk0, (num_bands_0 + 1) * sizeof(sbr->f_master[0]));
474 /// High Frequency Generation - Patch Construction (14496-3 sp04 p216 fig. 4.46)
475 static int sbr_hf_calc_npatches(AACContext *ac, SpectralBandReplication *sbr)
477 int i, k, last_k = -1, last_msb = -1, sb = 0;
479 int usb = sbr->kx[1];
480 int goal_sb = ((1000 << 11) + (sbr->sample_rate >> 1)) / sbr->sample_rate;
482 sbr->num_patches = 0;
484 if (goal_sb < sbr->kx[1] + sbr->m[1]) {
485 for (k = 0; sbr->f_master[k] < goal_sb; k++) ;
491 if (k == last_k && msb == last_msb) {
492 av_log(ac->avctx, AV_LOG_ERROR, "patch construction failed\n");
493 return AVERROR_INVALIDDATA;
497 for (i = k; i == k || sb > (sbr->k[0] - 1 + msb - odd); i--) {
498 sb = sbr->f_master[i];
499 odd = (sb + sbr->k[0]) & 1;
502 // Requirements (14496-3 sp04 p205) sets the maximum number of patches to 5.
503 // After this check the final number of patches can still be six which is
504 // illegal however the Coding Technologies decoder check stream has a final
505 // count of 6 patches
506 if (sbr->num_patches > 5) {
507 av_log(ac->avctx, AV_LOG_ERROR, "Too many patches: %d\n", sbr->num_patches);
511 sbr->patch_num_subbands[sbr->num_patches] = FFMAX(sb - usb, 0);
512 sbr->patch_start_subband[sbr->num_patches] = sbr->k[0] - odd - sbr->patch_num_subbands[sbr->num_patches];
514 if (sbr->patch_num_subbands[sbr->num_patches] > 0) {
521 if (sbr->f_master[k] - sb < 3)
523 } while (sb != sbr->kx[1] + sbr->m[1]);
525 if (sbr->num_patches > 1 &&
526 sbr->patch_num_subbands[sbr->num_patches - 1] < 3)
532 /// Derived Frequency Band Tables (14496-3 sp04 p197)
533 static int sbr_make_f_derived(AACContext *ac, SpectralBandReplication *sbr)
538 #endif /* USE_FIXED */
540 sbr->n[1] = sbr->n_master - sbr->spectrum_params.bs_xover_band;
541 sbr->n[0] = (sbr->n[1] + 1) >> 1;
543 memcpy(sbr->f_tablehigh, &sbr->f_master[sbr->spectrum_params.bs_xover_band],
544 (sbr->n[1] + 1) * sizeof(sbr->f_master[0]));
545 sbr->m[1] = sbr->f_tablehigh[sbr->n[1]] - sbr->f_tablehigh[0];
546 sbr->kx[1] = sbr->f_tablehigh[0];
548 // Requirements (14496-3 sp04 p205)
549 if (sbr->kx[1] + sbr->m[1] > 64) {
550 av_log(ac->avctx, AV_LOG_ERROR,
551 "Stop frequency border too high: %d\n", sbr->kx[1] + sbr->m[1]);
554 if (sbr->kx[1] > 32) {
555 av_log(ac->avctx, AV_LOG_ERROR, "Start frequency border too high: %d\n", sbr->kx[1]);
559 sbr->f_tablelow[0] = sbr->f_tablehigh[0];
560 temp = sbr->n[1] & 1;
561 for (k = 1; k <= sbr->n[0]; k++)
562 sbr->f_tablelow[k] = sbr->f_tablehigh[2 * k - temp];
564 temp = (sbr->k[2] << 23) / sbr->kx[1];
565 while (temp < 0x40000000) {
569 temp = fixed_log(temp - 0x80000000);
570 temp = (int)(((int64_t)temp * CONST_RECIP_LN2 + 0x20000000) >> 30);
571 temp = (((temp + 0x80) >> 8) + ((8 - nz) << 23)) * sbr->spectrum_params.bs_noise_bands;
573 sbr->n_q = (temp + 0x400000) >> 23;
577 sbr->n_q = FFMAX(1, lrintf(sbr->spectrum_params.bs_noise_bands *
578 log2f(sbr->k[2] / (float)sbr->kx[1]))); // 0 <= bs_noise_bands <= 3
579 #endif /* USE_FIXED */
582 av_log(ac->avctx, AV_LOG_ERROR, "Too many noise floor scale factors: %d\n", sbr->n_q);
586 sbr->f_tablenoise[0] = sbr->f_tablelow[0];
588 for (k = 1; k <= sbr->n_q; k++) {
589 temp += (sbr->n[0] - temp) / (sbr->n_q + 1 - k);
590 sbr->f_tablenoise[k] = sbr->f_tablelow[temp];
593 if (sbr_hf_calc_npatches(ac, sbr) < 0)
596 sbr_make_f_tablelim(sbr);
598 sbr->data[0].f_indexnoise = 0;
599 sbr->data[1].f_indexnoise = 0;
604 static av_always_inline void get_bits1_vector(GetBitContext *gb, uint8_t *vec,
608 for (i = 0; i < elements; i++) {
609 vec[i] = get_bits1(gb);
613 /** ceil(log2(index+1)) */
614 static const int8_t ceil_log2[] = {
618 static int read_sbr_grid(AACContext *ac, SpectralBandReplication *sbr,
619 GetBitContext *gb, SBRData *ch_data)
623 // frameLengthFlag ? 15 : 16; 960 sample length frames unsupported; this value is numTimeSlots
624 int abs_bord_trail = 16;
625 int num_rel_lead, num_rel_trail;
626 unsigned bs_num_env_old = ch_data->bs_num_env;
627 int bs_frame_class, bs_num_env;
629 ch_data->bs_freq_res[0] = ch_data->bs_freq_res[ch_data->bs_num_env];
630 ch_data->bs_amp_res = sbr->bs_amp_res_header;
631 ch_data->t_env_num_env_old = ch_data->t_env[bs_num_env_old];
633 switch (bs_frame_class = get_bits(gb, 2)) {
635 bs_num_env = 1 << get_bits(gb, 2);
636 if (bs_num_env > 4) {
637 av_log(ac->avctx, AV_LOG_ERROR,
638 "Invalid bitstream, too many SBR envelopes in FIXFIX type SBR frame: %d\n",
642 ch_data->bs_num_env = bs_num_env;
643 num_rel_lead = ch_data->bs_num_env - 1;
644 if (ch_data->bs_num_env == 1)
645 ch_data->bs_amp_res = 0;
648 ch_data->t_env[0] = 0;
649 ch_data->t_env[ch_data->bs_num_env] = abs_bord_trail;
651 abs_bord_trail = (abs_bord_trail + (ch_data->bs_num_env >> 1)) /
653 for (i = 0; i < num_rel_lead; i++)
654 ch_data->t_env[i + 1] = ch_data->t_env[i] + abs_bord_trail;
656 ch_data->bs_freq_res[1] = get_bits1(gb);
657 for (i = 1; i < ch_data->bs_num_env; i++)
658 ch_data->bs_freq_res[i + 1] = ch_data->bs_freq_res[1];
661 abs_bord_trail += get_bits(gb, 2);
662 num_rel_trail = get_bits(gb, 2);
663 ch_data->bs_num_env = num_rel_trail + 1;
664 ch_data->t_env[0] = 0;
665 ch_data->t_env[ch_data->bs_num_env] = abs_bord_trail;
667 for (i = 0; i < num_rel_trail; i++)
668 ch_data->t_env[ch_data->bs_num_env - 1 - i] =
669 ch_data->t_env[ch_data->bs_num_env - i] - 2 * get_bits(gb, 2) - 2;
671 bs_pointer = get_bits(gb, ceil_log2[ch_data->bs_num_env]);
673 for (i = 0; i < ch_data->bs_num_env; i++)
674 ch_data->bs_freq_res[ch_data->bs_num_env - i] = get_bits1(gb);
677 ch_data->t_env[0] = get_bits(gb, 2);
678 num_rel_lead = get_bits(gb, 2);
679 ch_data->bs_num_env = num_rel_lead + 1;
680 ch_data->t_env[ch_data->bs_num_env] = abs_bord_trail;
682 for (i = 0; i < num_rel_lead; i++)
683 ch_data->t_env[i + 1] = ch_data->t_env[i] + 2 * get_bits(gb, 2) + 2;
685 bs_pointer = get_bits(gb, ceil_log2[ch_data->bs_num_env]);
687 get_bits1_vector(gb, ch_data->bs_freq_res + 1, ch_data->bs_num_env);
690 ch_data->t_env[0] = get_bits(gb, 2);
691 abs_bord_trail += get_bits(gb, 2);
692 num_rel_lead = get_bits(gb, 2);
693 num_rel_trail = get_bits(gb, 2);
694 bs_num_env = num_rel_lead + num_rel_trail + 1;
696 if (bs_num_env > 5) {
697 av_log(ac->avctx, AV_LOG_ERROR,
698 "Invalid bitstream, too many SBR envelopes in VARVAR type SBR frame: %d\n",
702 ch_data->bs_num_env = bs_num_env;
704 ch_data->t_env[ch_data->bs_num_env] = abs_bord_trail;
706 for (i = 0; i < num_rel_lead; i++)
707 ch_data->t_env[i + 1] = ch_data->t_env[i] + 2 * get_bits(gb, 2) + 2;
708 for (i = 0; i < num_rel_trail; i++)
709 ch_data->t_env[ch_data->bs_num_env - 1 - i] =
710 ch_data->t_env[ch_data->bs_num_env - i] - 2 * get_bits(gb, 2) - 2;
712 bs_pointer = get_bits(gb, ceil_log2[ch_data->bs_num_env]);
714 get_bits1_vector(gb, ch_data->bs_freq_res + 1, ch_data->bs_num_env);
717 ch_data->bs_frame_class = bs_frame_class;
719 av_assert0(bs_pointer >= 0);
720 if (bs_pointer > ch_data->bs_num_env + 1) {
721 av_log(ac->avctx, AV_LOG_ERROR,
722 "Invalid bitstream, bs_pointer points to a middle noise border outside the time borders table: %d\n",
727 for (i = 1; i <= ch_data->bs_num_env; i++) {
728 if (ch_data->t_env[i-1] >= ch_data->t_env[i]) {
729 av_log(ac->avctx, AV_LOG_ERROR, "Not strictly monotone time borders\n");
734 ch_data->bs_num_noise = (ch_data->bs_num_env > 1) + 1;
736 ch_data->t_q[0] = ch_data->t_env[0];
737 ch_data->t_q[ch_data->bs_num_noise] = ch_data->t_env[ch_data->bs_num_env];
738 if (ch_data->bs_num_noise > 1) {
740 if (ch_data->bs_frame_class == FIXFIX) {
741 idx = ch_data->bs_num_env >> 1;
742 } else if (ch_data->bs_frame_class & 1) { // FIXVAR or VARVAR
743 idx = ch_data->bs_num_env - FFMAX(bs_pointer - 1, 1);
747 else if (bs_pointer == 1)
748 idx = ch_data->bs_num_env - 1;
749 else // bs_pointer > 1
750 idx = bs_pointer - 1;
752 ch_data->t_q[1] = ch_data->t_env[idx];
755 ch_data->e_a[0] = -(ch_data->e_a[1] != bs_num_env_old); // l_APrev
756 ch_data->e_a[1] = -1;
757 if ((ch_data->bs_frame_class & 1) && bs_pointer) { // FIXVAR or VARVAR and bs_pointer != 0
758 ch_data->e_a[1] = ch_data->bs_num_env + 1 - bs_pointer;
759 } else if ((ch_data->bs_frame_class == 2) && (bs_pointer > 1)) // VARFIX and bs_pointer > 1
760 ch_data->e_a[1] = bs_pointer - 1;
765 static void copy_sbr_grid(SBRData *dst, const SBRData *src) {
766 //These variables are saved from the previous frame rather than copied
767 dst->bs_freq_res[0] = dst->bs_freq_res[dst->bs_num_env];
768 dst->t_env_num_env_old = dst->t_env[dst->bs_num_env];
769 dst->e_a[0] = -(dst->e_a[1] != dst->bs_num_env);
771 //These variables are read from the bitstream and therefore copied
772 memcpy(dst->bs_freq_res+1, src->bs_freq_res+1, sizeof(dst->bs_freq_res)-sizeof(*dst->bs_freq_res));
773 memcpy(dst->t_env, src->t_env, sizeof(dst->t_env));
774 memcpy(dst->t_q, src->t_q, sizeof(dst->t_q));
775 dst->bs_num_env = src->bs_num_env;
776 dst->bs_amp_res = src->bs_amp_res;
777 dst->bs_num_noise = src->bs_num_noise;
778 dst->bs_frame_class = src->bs_frame_class;
779 dst->e_a[1] = src->e_a[1];
782 /// Read how the envelope and noise floor data is delta coded
783 static void read_sbr_dtdf(SpectralBandReplication *sbr, GetBitContext *gb,
786 get_bits1_vector(gb, ch_data->bs_df_env, ch_data->bs_num_env);
787 get_bits1_vector(gb, ch_data->bs_df_noise, ch_data->bs_num_noise);
790 /// Read inverse filtering data
791 static void read_sbr_invf(SpectralBandReplication *sbr, GetBitContext *gb,
796 memcpy(ch_data->bs_invf_mode[1], ch_data->bs_invf_mode[0], 5 * sizeof(uint8_t));
797 for (i = 0; i < sbr->n_q; i++)
798 ch_data->bs_invf_mode[0][i] = get_bits(gb, 2);
801 static int read_sbr_envelope(AACContext *ac, SpectralBandReplication *sbr, GetBitContext *gb,
802 SBRData *ch_data, int ch)
806 VLC_TYPE (*t_huff)[2], (*f_huff)[2];
808 const int delta = (ch == 1 && sbr->bs_coupling == 1) + 1;
809 const int odd = sbr->n[1] & 1;
811 if (sbr->bs_coupling && ch) {
812 if (ch_data->bs_amp_res) {
814 t_huff = vlc_sbr[T_HUFFMAN_ENV_BAL_3_0DB].table;
815 t_lav = vlc_sbr_lav[T_HUFFMAN_ENV_BAL_3_0DB];
816 f_huff = vlc_sbr[F_HUFFMAN_ENV_BAL_3_0DB].table;
817 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_BAL_3_0DB];
820 t_huff = vlc_sbr[T_HUFFMAN_ENV_BAL_1_5DB].table;
821 t_lav = vlc_sbr_lav[T_HUFFMAN_ENV_BAL_1_5DB];
822 f_huff = vlc_sbr[F_HUFFMAN_ENV_BAL_1_5DB].table;
823 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_BAL_1_5DB];
826 if (ch_data->bs_amp_res) {
828 t_huff = vlc_sbr[T_HUFFMAN_ENV_3_0DB].table;
829 t_lav = vlc_sbr_lav[T_HUFFMAN_ENV_3_0DB];
830 f_huff = vlc_sbr[F_HUFFMAN_ENV_3_0DB].table;
831 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_3_0DB];
834 t_huff = vlc_sbr[T_HUFFMAN_ENV_1_5DB].table;
835 t_lav = vlc_sbr_lav[T_HUFFMAN_ENV_1_5DB];
836 f_huff = vlc_sbr[F_HUFFMAN_ENV_1_5DB].table;
837 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_1_5DB];
841 for (i = 0; i < ch_data->bs_num_env; i++) {
842 if (ch_data->bs_df_env[i]) {
843 // bs_freq_res[0] == bs_freq_res[bs_num_env] from prev frame
844 if (ch_data->bs_freq_res[i + 1] == ch_data->bs_freq_res[i]) {
845 for (j = 0; j < sbr->n[ch_data->bs_freq_res[i + 1]]; j++) {
846 ch_data->env_facs_q[i + 1][j] = ch_data->env_facs_q[i][j] + delta * (get_vlc2(gb, t_huff, 9, 3) - t_lav);
847 if (ch_data->env_facs_q[i + 1][j] > 127U) {
848 av_log(ac->avctx, AV_LOG_ERROR, "env_facs_q %d is invalid\n", ch_data->env_facs_q[i + 1][j]);
849 return AVERROR_INVALIDDATA;
852 } else if (ch_data->bs_freq_res[i + 1]) {
853 for (j = 0; j < sbr->n[ch_data->bs_freq_res[i + 1]]; j++) {
854 k = (j + odd) >> 1; // find k such that f_tablelow[k] <= f_tablehigh[j] < f_tablelow[k + 1]
855 ch_data->env_facs_q[i + 1][j] = ch_data->env_facs_q[i][k] + delta * (get_vlc2(gb, t_huff, 9, 3) - t_lav);
856 if (ch_data->env_facs_q[i + 1][j] > 127U) {
857 av_log(ac->avctx, AV_LOG_ERROR, "env_facs_q %d is invalid\n", ch_data->env_facs_q[i + 1][j]);
858 return AVERROR_INVALIDDATA;
862 for (j = 0; j < sbr->n[ch_data->bs_freq_res[i + 1]]; j++) {
863 k = j ? 2*j - odd : 0; // find k such that f_tablehigh[k] == f_tablelow[j]
864 ch_data->env_facs_q[i + 1][j] = ch_data->env_facs_q[i][k] + delta * (get_vlc2(gb, t_huff, 9, 3) - t_lav);
865 if (ch_data->env_facs_q[i + 1][j] > 127U) {
866 av_log(ac->avctx, AV_LOG_ERROR, "env_facs_q %d is invalid\n", ch_data->env_facs_q[i + 1][j]);
867 return AVERROR_INVALIDDATA;
872 ch_data->env_facs_q[i + 1][0] = delta * get_bits(gb, bits); // bs_env_start_value_balance
873 for (j = 1; j < sbr->n[ch_data->bs_freq_res[i + 1]]; j++) {
874 ch_data->env_facs_q[i + 1][j] = ch_data->env_facs_q[i + 1][j - 1] + delta * (get_vlc2(gb, f_huff, 9, 3) - f_lav);
875 if (ch_data->env_facs_q[i + 1][j] > 127U) {
876 av_log(ac->avctx, AV_LOG_ERROR, "env_facs_q %d is invalid\n", ch_data->env_facs_q[i + 1][j]);
877 return AVERROR_INVALIDDATA;
883 //assign 0th elements of env_facs_q from last elements
884 memcpy(ch_data->env_facs_q[0], ch_data->env_facs_q[ch_data->bs_num_env],
885 sizeof(ch_data->env_facs_q[0]));
890 static int read_sbr_noise(AACContext *ac, SpectralBandReplication *sbr, GetBitContext *gb,
891 SBRData *ch_data, int ch)
894 VLC_TYPE (*t_huff)[2], (*f_huff)[2];
896 int delta = (ch == 1 && sbr->bs_coupling == 1) + 1;
898 if (sbr->bs_coupling && ch) {
899 t_huff = vlc_sbr[T_HUFFMAN_NOISE_BAL_3_0DB].table;
900 t_lav = vlc_sbr_lav[T_HUFFMAN_NOISE_BAL_3_0DB];
901 f_huff = vlc_sbr[F_HUFFMAN_ENV_BAL_3_0DB].table;
902 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_BAL_3_0DB];
904 t_huff = vlc_sbr[T_HUFFMAN_NOISE_3_0DB].table;
905 t_lav = vlc_sbr_lav[T_HUFFMAN_NOISE_3_0DB];
906 f_huff = vlc_sbr[F_HUFFMAN_ENV_3_0DB].table;
907 f_lav = vlc_sbr_lav[F_HUFFMAN_ENV_3_0DB];
910 for (i = 0; i < ch_data->bs_num_noise; i++) {
911 if (ch_data->bs_df_noise[i]) {
912 for (j = 0; j < sbr->n_q; j++) {
913 ch_data->noise_facs_q[i + 1][j] = ch_data->noise_facs_q[i][j] + delta * (get_vlc2(gb, t_huff, 9, 2) - t_lav);
914 if (ch_data->noise_facs_q[i + 1][j] > 30U) {
915 av_log(ac->avctx, AV_LOG_ERROR, "noise_facs_q %d is invalid\n", ch_data->noise_facs_q[i + 1][j]);
916 return AVERROR_INVALIDDATA;
920 ch_data->noise_facs_q[i + 1][0] = delta * get_bits(gb, 5); // bs_noise_start_value_balance or bs_noise_start_value_level
921 for (j = 1; j < sbr->n_q; j++) {
922 ch_data->noise_facs_q[i + 1][j] = ch_data->noise_facs_q[i + 1][j - 1] + delta * (get_vlc2(gb, f_huff, 9, 3) - f_lav);
923 if (ch_data->noise_facs_q[i + 1][j] > 30U) {
924 av_log(ac->avctx, AV_LOG_ERROR, "noise_facs_q %d is invalid\n", ch_data->noise_facs_q[i + 1][j]);
925 return AVERROR_INVALIDDATA;
931 //assign 0th elements of noise_facs_q from last elements
932 memcpy(ch_data->noise_facs_q[0], ch_data->noise_facs_q[ch_data->bs_num_noise],
933 sizeof(ch_data->noise_facs_q[0]));
937 static void read_sbr_extension(AACContext *ac, SpectralBandReplication *sbr,
939 int bs_extension_id, int *num_bits_left)
941 switch (bs_extension_id) {
942 case EXTENSION_ID_PS:
943 if (!ac->oc[1].m4ac.ps) {
944 av_log(ac->avctx, AV_LOG_ERROR, "Parametric Stereo signaled to be not-present but was found in the bitstream.\n");
945 skip_bits_long(gb, *num_bits_left); // bs_fill_bits
948 *num_bits_left -= AAC_RENAME(ff_ps_read_data)(ac->avctx, gb, &sbr->ps, *num_bits_left);
949 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
953 // some files contain 0-padding
954 if (bs_extension_id || *num_bits_left > 16 || show_bits(gb, *num_bits_left))
955 avpriv_request_sample(ac->avctx, "Reserved SBR extensions");
956 skip_bits_long(gb, *num_bits_left); // bs_fill_bits
962 static int read_sbr_single_channel_element(AACContext *ac,
963 SpectralBandReplication *sbr,
968 if (get_bits1(gb)) // bs_data_extra
969 skip_bits(gb, 4); // bs_reserved
971 if (read_sbr_grid(ac, sbr, gb, &sbr->data[0]))
973 read_sbr_dtdf(sbr, gb, &sbr->data[0]);
974 read_sbr_invf(sbr, gb, &sbr->data[0]);
975 if((ret = read_sbr_envelope(ac, sbr, gb, &sbr->data[0], 0)) < 0)
977 if((ret = read_sbr_noise(ac, sbr, gb, &sbr->data[0], 0)) < 0)
980 if ((sbr->data[0].bs_add_harmonic_flag = get_bits1(gb)))
981 get_bits1_vector(gb, sbr->data[0].bs_add_harmonic, sbr->n[1]);
986 static int read_sbr_channel_pair_element(AACContext *ac,
987 SpectralBandReplication *sbr,
992 if (get_bits1(gb)) // bs_data_extra
993 skip_bits(gb, 8); // bs_reserved
995 if ((sbr->bs_coupling = get_bits1(gb))) {
996 if (read_sbr_grid(ac, sbr, gb, &sbr->data[0]))
998 copy_sbr_grid(&sbr->data[1], &sbr->data[0]);
999 read_sbr_dtdf(sbr, gb, &sbr->data[0]);
1000 read_sbr_dtdf(sbr, gb, &sbr->data[1]);
1001 read_sbr_invf(sbr, gb, &sbr->data[0]);
1002 memcpy(sbr->data[1].bs_invf_mode[1], sbr->data[1].bs_invf_mode[0], sizeof(sbr->data[1].bs_invf_mode[0]));
1003 memcpy(sbr->data[1].bs_invf_mode[0], sbr->data[0].bs_invf_mode[0], sizeof(sbr->data[1].bs_invf_mode[0]));
1004 if((ret = read_sbr_envelope(ac, sbr, gb, &sbr->data[0], 0)) < 0)
1006 if((ret = read_sbr_noise(ac, sbr, gb, &sbr->data[0], 0)) < 0)
1008 if((ret = read_sbr_envelope(ac, sbr, gb, &sbr->data[1], 1)) < 0)
1010 if((ret = read_sbr_noise(ac, sbr, gb, &sbr->data[1], 1)) < 0)
1013 if (read_sbr_grid(ac, sbr, gb, &sbr->data[0]) ||
1014 read_sbr_grid(ac, sbr, gb, &sbr->data[1]))
1016 read_sbr_dtdf(sbr, gb, &sbr->data[0]);
1017 read_sbr_dtdf(sbr, gb, &sbr->data[1]);
1018 read_sbr_invf(sbr, gb, &sbr->data[0]);
1019 read_sbr_invf(sbr, gb, &sbr->data[1]);
1020 if((ret = read_sbr_envelope(ac, sbr, gb, &sbr->data[0], 0)) < 0)
1022 if((ret = read_sbr_envelope(ac, sbr, gb, &sbr->data[1], 1)) < 0)
1024 if((ret = read_sbr_noise(ac, sbr, gb, &sbr->data[0], 0)) < 0)
1026 if((ret = read_sbr_noise(ac, sbr, gb, &sbr->data[1], 1)) < 0)
1030 if ((sbr->data[0].bs_add_harmonic_flag = get_bits1(gb)))
1031 get_bits1_vector(gb, sbr->data[0].bs_add_harmonic, sbr->n[1]);
1032 if ((sbr->data[1].bs_add_harmonic_flag = get_bits1(gb)))
1033 get_bits1_vector(gb, sbr->data[1].bs_add_harmonic, sbr->n[1]);
1038 static unsigned int read_sbr_data(AACContext *ac, SpectralBandReplication *sbr,
1039 GetBitContext *gb, int id_aac)
1041 unsigned int cnt = get_bits_count(gb);
1043 sbr->id_aac = id_aac;
1044 sbr->ready_for_dequant = 1;
1046 if (id_aac == TYPE_SCE || id_aac == TYPE_CCE) {
1047 if (read_sbr_single_channel_element(ac, sbr, gb)) {
1049 return get_bits_count(gb) - cnt;
1051 } else if (id_aac == TYPE_CPE) {
1052 if (read_sbr_channel_pair_element(ac, sbr, gb)) {
1054 return get_bits_count(gb) - cnt;
1057 av_log(ac->avctx, AV_LOG_ERROR,
1058 "Invalid bitstream - cannot apply SBR to element type %d\n", id_aac);
1060 return get_bits_count(gb) - cnt;
1062 if (get_bits1(gb)) { // bs_extended_data
1063 int num_bits_left = get_bits(gb, 4); // bs_extension_size
1064 if (num_bits_left == 15)
1065 num_bits_left += get_bits(gb, 8); // bs_esc_count
1067 num_bits_left <<= 3;
1068 while (num_bits_left > 7) {
1070 read_sbr_extension(ac, sbr, gb, get_bits(gb, 2), &num_bits_left); // bs_extension_id
1072 if (num_bits_left < 0) {
1073 av_log(ac->avctx, AV_LOG_ERROR, "SBR Extension over read.\n");
1075 if (num_bits_left > 0)
1076 skip_bits(gb, num_bits_left);
1079 return get_bits_count(gb) - cnt;
1082 static void sbr_reset(AACContext *ac, SpectralBandReplication *sbr)
1085 err = sbr_make_f_master(ac, sbr, &sbr->spectrum_params);
1087 err = sbr_make_f_derived(ac, sbr);
1089 av_log(ac->avctx, AV_LOG_ERROR,
1090 "SBR reset failed. Switching SBR to pure upsampling mode.\n");
1096 * Decode Spectral Band Replication extension data; reference: table 4.55.
1098 * @param crc flag indicating the presence of CRC checksum
1099 * @param cnt length of TYPE_FIL syntactic element in bytes
1101 * @return Returns number of bytes consumed from the TYPE_FIL element.
1103 int AAC_RENAME(ff_decode_sbr_extension)(AACContext *ac, SpectralBandReplication *sbr,
1104 GetBitContext *gb_host, int crc, int cnt, int id_aac)
1106 unsigned int num_sbr_bits = 0, num_align_bits;
1107 unsigned bytes_read;
1108 GetBitContext gbc = *gb_host, *gb = &gbc;
1109 skip_bits_long(gb_host, cnt*8 - 4);
1113 if (!sbr->sample_rate)
1114 sbr->sample_rate = 2 * ac->oc[1].m4ac.sample_rate; //TODO use the nominal sample rate for arbitrary sample rate support
1115 if (!ac->oc[1].m4ac.ext_sample_rate)
1116 ac->oc[1].m4ac.ext_sample_rate = 2 * ac->oc[1].m4ac.sample_rate;
1119 skip_bits(gb, 10); // bs_sbr_crc_bits; TODO - implement CRC check
1123 //Save some state from the previous frame.
1124 sbr->kx[0] = sbr->kx[1];
1125 sbr->m[0] = sbr->m[1];
1126 sbr->kx_and_m_pushed = 1;
1129 if (get_bits1(gb)) // bs_header_flag
1130 num_sbr_bits += read_sbr_header(sbr, gb);
1136 num_sbr_bits += read_sbr_data(ac, sbr, gb, id_aac);
1138 num_align_bits = ((cnt << 3) - 4 - num_sbr_bits) & 7;
1139 bytes_read = ((num_sbr_bits + num_align_bits + 4) >> 3);
1141 if (bytes_read > cnt) {
1142 av_log(ac->avctx, AV_LOG_ERROR,
1143 "Expected to read %d SBR bytes actually read %d.\n", cnt, bytes_read);
1150 * Analysis QMF Bank (14496-3 sp04 p206)
1152 * @param x pointer to the beginning of the first sample window
1153 * @param W array of complex-valued samples split into subbands
1155 #ifndef sbr_qmf_analysis
1157 static void sbr_qmf_analysis(AVFixedDSPContext *dsp, FFTContext *mdct,
1159 static void sbr_qmf_analysis(AVFloatDSPContext *dsp, FFTContext *mdct,
1160 #endif /* USE_FIXED */
1161 SBRDSPContext *sbrdsp, const INTFLOAT *in, INTFLOAT *x,
1162 INTFLOAT z[320], INTFLOAT W[2][32][32][2], int buf_idx)
1168 memcpy(x , x+1024, (320-32)*sizeof(x[0]));
1169 memcpy(x+288, in, 1024*sizeof(x[0]));
1170 for (i = 0; i < 32; i++) { // numTimeSlots*RATE = 16*2 as 960 sample frames
1171 // are not supported
1172 dsp->vector_fmul_reverse(z, sbr_qmf_window_ds, x, 320);
1174 sbrdsp->qmf_pre_shuffle(z);
1176 for (j = 64; j < 128; j++) {
1178 av_log(NULL, AV_LOG_WARNING,
1179 "sbr_qmf_analysis: value %09d too large, setting to %09d\n",
1182 } else if (z[j] < -(1<<24)) {
1183 av_log(NULL, AV_LOG_WARNING,
1184 "sbr_qmf_analysis: value %09d too small, setting to %09d\n",
1190 mdct->imdct_half(mdct, z, z+64);
1191 sbrdsp->qmf_post_shuffle(W[buf_idx][i], z);
1198 * Synthesis QMF Bank (14496-3 sp04 p206) and Downsampled Synthesis QMF Bank
1199 * (14496-3 sp04 p206)
1201 #ifndef sbr_qmf_synthesis
1202 static void sbr_qmf_synthesis(FFTContext *mdct,
1204 SBRDSPContext *sbrdsp, AVFixedDSPContext *dsp,
1206 SBRDSPContext *sbrdsp, AVFloatDSPContext *dsp,
1207 #endif /* USE_FIXED */
1208 INTFLOAT *out, INTFLOAT X[2][38][64],
1209 INTFLOAT mdct_buf[2][64],
1210 INTFLOAT *v0, int *v_off, const unsigned int div)
1213 const INTFLOAT *sbr_qmf_window = div ? sbr_qmf_window_ds : sbr_qmf_window_us;
1214 const int step = 128 >> div;
1216 for (i = 0; i < 32; i++) {
1217 if (*v_off < step) {
1218 int saved_samples = (1280 - 128) >> div;
1219 memcpy(&v0[SBR_SYNTHESIS_BUF_SIZE - saved_samples], v0, saved_samples * sizeof(INTFLOAT));
1220 *v_off = SBR_SYNTHESIS_BUF_SIZE - saved_samples - step;
1226 for (n = 0; n < 32; n++) {
1227 X[0][i][ n] = -X[0][i][n];
1228 X[0][i][32+n] = X[1][i][31-n];
1230 mdct->imdct_half(mdct, mdct_buf[0], X[0][i]);
1231 sbrdsp->qmf_deint_neg(v, mdct_buf[0]);
1233 sbrdsp->neg_odd_64(X[1][i]);
1234 mdct->imdct_half(mdct, mdct_buf[0], X[0][i]);
1235 mdct->imdct_half(mdct, mdct_buf[1], X[1][i]);
1236 sbrdsp->qmf_deint_bfly(v, mdct_buf[1], mdct_buf[0]);
1238 dsp->vector_fmul (out, v , sbr_qmf_window , 64 >> div);
1239 dsp->vector_fmul_add(out, v + ( 192 >> div), sbr_qmf_window + ( 64 >> div), out , 64 >> div);
1240 dsp->vector_fmul_add(out, v + ( 256 >> div), sbr_qmf_window + (128 >> div), out , 64 >> div);
1241 dsp->vector_fmul_add(out, v + ( 448 >> div), sbr_qmf_window + (192 >> div), out , 64 >> div);
1242 dsp->vector_fmul_add(out, v + ( 512 >> div), sbr_qmf_window + (256 >> div), out , 64 >> div);
1243 dsp->vector_fmul_add(out, v + ( 704 >> div), sbr_qmf_window + (320 >> div), out , 64 >> div);
1244 dsp->vector_fmul_add(out, v + ( 768 >> div), sbr_qmf_window + (384 >> div), out , 64 >> div);
1245 dsp->vector_fmul_add(out, v + ( 960 >> div), sbr_qmf_window + (448 >> div), out , 64 >> div);
1246 dsp->vector_fmul_add(out, v + (1024 >> div), sbr_qmf_window + (512 >> div), out , 64 >> div);
1247 dsp->vector_fmul_add(out, v + (1216 >> div), sbr_qmf_window + (576 >> div), out , 64 >> div);
1253 /// Generate the subband filtered lowband
1254 static int sbr_lf_gen(AACContext *ac, SpectralBandReplication *sbr,
1255 INTFLOAT X_low[32][40][2], const INTFLOAT W[2][32][32][2],
1259 const int t_HFGen = 8;
1261 memset(X_low, 0, 32*sizeof(*X_low));
1262 for (k = 0; k < sbr->kx[1]; k++) {
1263 for (i = t_HFGen; i < i_f + t_HFGen; i++) {
1264 X_low[k][i][0] = W[buf_idx][i - t_HFGen][k][0];
1265 X_low[k][i][1] = W[buf_idx][i - t_HFGen][k][1];
1268 buf_idx = 1-buf_idx;
1269 for (k = 0; k < sbr->kx[0]; k++) {
1270 for (i = 0; i < t_HFGen; i++) {
1271 X_low[k][i][0] = W[buf_idx][i + i_f - t_HFGen][k][0];
1272 X_low[k][i][1] = W[buf_idx][i + i_f - t_HFGen][k][1];
1278 /// High Frequency Generator (14496-3 sp04 p215)
1279 static int sbr_hf_gen(AACContext *ac, SpectralBandReplication *sbr,
1280 INTFLOAT X_high[64][40][2], const INTFLOAT X_low[32][40][2],
1281 const INTFLOAT (*alpha0)[2], const INTFLOAT (*alpha1)[2],
1282 const INTFLOAT bw_array[5], const uint8_t *t_env,
1288 for (j = 0; j < sbr->num_patches; j++) {
1289 for (x = 0; x < sbr->patch_num_subbands[j]; x++, k++) {
1290 const int p = sbr->patch_start_subband[j] + x;
1291 while (g <= sbr->n_q && k >= sbr->f_tablenoise[g])
1296 av_log(ac->avctx, AV_LOG_ERROR,
1297 "ERROR : no subband found for frequency %d\n", k);
1301 sbr->dsp.hf_gen(X_high[k] + ENVELOPE_ADJUSTMENT_OFFSET,
1302 X_low[p] + ENVELOPE_ADJUSTMENT_OFFSET,
1303 alpha0[p], alpha1[p], bw_array[g],
1304 2 * t_env[0], 2 * t_env[bs_num_env]);
1307 if (k < sbr->m[1] + sbr->kx[1])
1308 memset(X_high + k, 0, (sbr->m[1] + sbr->kx[1] - k) * sizeof(*X_high));
1313 /// Generate the subband filtered lowband
1314 static int sbr_x_gen(SpectralBandReplication *sbr, INTFLOAT X[2][38][64],
1315 const INTFLOAT Y0[38][64][2], const INTFLOAT Y1[38][64][2],
1316 const INTFLOAT X_low[32][40][2], int ch)
1320 const int i_Temp = FFMAX(2*sbr->data[ch].t_env_num_env_old - i_f, 0);
1321 memset(X, 0, 2*sizeof(*X));
1322 for (k = 0; k < sbr->kx[0]; k++) {
1323 for (i = 0; i < i_Temp; i++) {
1324 X[0][i][k] = X_low[k][i + ENVELOPE_ADJUSTMENT_OFFSET][0];
1325 X[1][i][k] = X_low[k][i + ENVELOPE_ADJUSTMENT_OFFSET][1];
1328 for (; k < sbr->kx[0] + sbr->m[0]; k++) {
1329 for (i = 0; i < i_Temp; i++) {
1330 X[0][i][k] = Y0[i + i_f][k][0];
1331 X[1][i][k] = Y0[i + i_f][k][1];
1335 for (k = 0; k < sbr->kx[1]; k++) {
1336 for (i = i_Temp; i < 38; i++) {
1337 X[0][i][k] = X_low[k][i + ENVELOPE_ADJUSTMENT_OFFSET][0];
1338 X[1][i][k] = X_low[k][i + ENVELOPE_ADJUSTMENT_OFFSET][1];
1341 for (; k < sbr->kx[1] + sbr->m[1]; k++) {
1342 for (i = i_Temp; i < i_f; i++) {
1343 X[0][i][k] = Y1[i][k][0];
1344 X[1][i][k] = Y1[i][k][1];
1350 /** High Frequency Adjustment (14496-3 sp04 p217) and Mapping
1351 * (14496-3 sp04 p217)
1353 static int sbr_mapping(AACContext *ac, SpectralBandReplication *sbr,
1354 SBRData *ch_data, int e_a[2])
1358 memset(ch_data->s_indexmapped[1], 0, 7*sizeof(ch_data->s_indexmapped[1]));
1359 for (e = 0; e < ch_data->bs_num_env; e++) {
1360 const unsigned int ilim = sbr->n[ch_data->bs_freq_res[e + 1]];
1361 uint16_t *table = ch_data->bs_freq_res[e + 1] ? sbr->f_tablehigh : sbr->f_tablelow;
1364 if (sbr->kx[1] != table[0]) {
1365 av_log(ac->avctx, AV_LOG_ERROR, "kx != f_table{high,low}[0]. "
1366 "Derived frequency tables were not regenerated.\n");
1370 for (i = 0; i < ilim; i++)
1371 for (m = table[i]; m < table[i + 1]; m++)
1372 sbr->e_origmapped[e][m - sbr->kx[1]] = ch_data->env_facs[e+1][i];
1374 // ch_data->bs_num_noise > 1 => 2 noise floors
1375 k = (ch_data->bs_num_noise > 1) && (ch_data->t_env[e] >= ch_data->t_q[1]);
1376 for (i = 0; i < sbr->n_q; i++)
1377 for (m = sbr->f_tablenoise[i]; m < sbr->f_tablenoise[i + 1]; m++)
1378 sbr->q_mapped[e][m - sbr->kx[1]] = ch_data->noise_facs[k+1][i];
1380 for (i = 0; i < sbr->n[1]; i++) {
1381 if (ch_data->bs_add_harmonic_flag) {
1382 const unsigned int m_midpoint =
1383 (sbr->f_tablehigh[i] + sbr->f_tablehigh[i + 1]) >> 1;
1385 ch_data->s_indexmapped[e + 1][m_midpoint - sbr->kx[1]] = ch_data->bs_add_harmonic[i] *
1386 (e >= e_a[1] || (ch_data->s_indexmapped[0][m_midpoint - sbr->kx[1]] == 1));
1390 for (i = 0; i < ilim; i++) {
1391 int additional_sinusoid_present = 0;
1392 for (m = table[i]; m < table[i + 1]; m++) {
1393 if (ch_data->s_indexmapped[e + 1][m - sbr->kx[1]]) {
1394 additional_sinusoid_present = 1;
1398 memset(&sbr->s_mapped[e][table[i] - sbr->kx[1]], additional_sinusoid_present,
1399 (table[i + 1] - table[i]) * sizeof(sbr->s_mapped[e][0]));
1403 memcpy(ch_data->s_indexmapped[0], ch_data->s_indexmapped[ch_data->bs_num_env], sizeof(ch_data->s_indexmapped[0]));
1407 /// Estimation of current envelope (14496-3 sp04 p218)
1408 static void sbr_env_estimate(AAC_FLOAT (*e_curr)[48], INTFLOAT X_high[64][40][2],
1409 SpectralBandReplication *sbr, SBRData *ch_data)
1412 int kx1 = sbr->kx[1];
1414 if (sbr->bs_interpol_freq) {
1415 for (e = 0; e < ch_data->bs_num_env; e++) {
1417 const SoftFloat recip_env_size = av_int2sf(0x20000000 / (ch_data->t_env[e + 1] - ch_data->t_env[e]), 30);
1419 const float recip_env_size = 0.5f / (ch_data->t_env[e + 1] - ch_data->t_env[e]);
1420 #endif /* USE_FIXED */
1421 int ilb = ch_data->t_env[e] * 2 + ENVELOPE_ADJUSTMENT_OFFSET;
1422 int iub = ch_data->t_env[e + 1] * 2 + ENVELOPE_ADJUSTMENT_OFFSET;
1424 for (m = 0; m < sbr->m[1]; m++) {
1425 AAC_FLOAT sum = sbr->dsp.sum_square(X_high[m+kx1] + ilb, iub - ilb);
1427 e_curr[e][m] = av_mul_sf(sum, recip_env_size);
1429 e_curr[e][m] = sum * recip_env_size;
1430 #endif /* USE_FIXED */
1436 for (e = 0; e < ch_data->bs_num_env; e++) {
1437 const int env_size = 2 * (ch_data->t_env[e + 1] - ch_data->t_env[e]);
1438 int ilb = ch_data->t_env[e] * 2 + ENVELOPE_ADJUSTMENT_OFFSET;
1439 int iub = ch_data->t_env[e + 1] * 2 + ENVELOPE_ADJUSTMENT_OFFSET;
1440 const uint16_t *table = ch_data->bs_freq_res[e + 1] ? sbr->f_tablehigh : sbr->f_tablelow;
1442 for (p = 0; p < sbr->n[ch_data->bs_freq_res[e + 1]]; p++) {
1444 SoftFloat sum = FLOAT_0;
1445 const SoftFloat den = av_int2sf(0x20000000 / (env_size * (table[p + 1] - table[p])), 29);
1446 for (k = table[p]; k < table[p + 1]; k++) {
1447 sum = av_add_sf(sum, sbr->dsp.sum_square(X_high[k] + ilb, iub - ilb));
1449 sum = av_mul_sf(sum, den);
1452 const int den = env_size * (table[p + 1] - table[p]);
1454 for (k = table[p]; k < table[p + 1]; k++) {
1455 sum += sbr->dsp.sum_square(X_high[k] + ilb, iub - ilb);
1458 #endif /* USE_FIXED */
1459 for (k = table[p]; k < table[p + 1]; k++) {
1460 e_curr[e][k - kx1] = sum;
1467 void AAC_RENAME(ff_sbr_apply)(AACContext *ac, SpectralBandReplication *sbr, int id_aac,
1468 INTFLOAT* L, INTFLOAT* R)
1470 int downsampled = ac->oc[1].m4ac.ext_sample_rate < sbr->sample_rate;
1472 int nch = (id_aac == TYPE_CPE) ? 2 : 1;
1475 if (id_aac != sbr->id_aac) {
1476 av_log(ac->avctx, id_aac == TYPE_LFE ? AV_LOG_VERBOSE : AV_LOG_WARNING,
1477 "element type mismatch %d != %d\n", id_aac, sbr->id_aac);
1481 if (sbr->start && !sbr->ready_for_dequant) {
1482 av_log(ac->avctx, AV_LOG_ERROR,
1483 "No quantized data read for sbr_dequant.\n");
1487 if (!sbr->kx_and_m_pushed) {
1488 sbr->kx[0] = sbr->kx[1];
1489 sbr->m[0] = sbr->m[1];
1491 sbr->kx_and_m_pushed = 0;
1495 sbr_dequant(sbr, id_aac);
1496 sbr->ready_for_dequant = 0;
1498 for (ch = 0; ch < nch; ch++) {
1499 /* decode channel */
1500 sbr_qmf_analysis(ac->fdsp, &sbr->mdct_ana, &sbr->dsp, ch ? R : L, sbr->data[ch].analysis_filterbank_samples,
1501 (INTFLOAT*)sbr->qmf_filter_scratch,
1502 sbr->data[ch].W, sbr->data[ch].Ypos);
1503 sbr->c.sbr_lf_gen(ac, sbr, sbr->X_low,
1504 (const INTFLOAT (*)[32][32][2]) sbr->data[ch].W,
1505 sbr->data[ch].Ypos);
1506 sbr->data[ch].Ypos ^= 1;
1508 sbr->c.sbr_hf_inverse_filter(&sbr->dsp, sbr->alpha0, sbr->alpha1,
1509 (const INTFLOAT (*)[40][2]) sbr->X_low, sbr->k[0]);
1510 sbr_chirp(sbr, &sbr->data[ch]);
1511 av_assert0(sbr->data[ch].bs_num_env > 0);
1512 sbr_hf_gen(ac, sbr, sbr->X_high,
1513 (const INTFLOAT (*)[40][2]) sbr->X_low,
1514 (const INTFLOAT (*)[2]) sbr->alpha0,
1515 (const INTFLOAT (*)[2]) sbr->alpha1,
1516 sbr->data[ch].bw_array, sbr->data[ch].t_env,
1517 sbr->data[ch].bs_num_env);
1520 err = sbr_mapping(ac, sbr, &sbr->data[ch], sbr->data[ch].e_a);
1522 sbr_env_estimate(sbr->e_curr, sbr->X_high, sbr, &sbr->data[ch]);
1523 sbr_gain_calc(ac, sbr, &sbr->data[ch], sbr->data[ch].e_a);
1524 sbr->c.sbr_hf_assemble(sbr->data[ch].Y[sbr->data[ch].Ypos],
1525 (const INTFLOAT (*)[40][2]) sbr->X_high,
1526 sbr, &sbr->data[ch],
1532 sbr->c.sbr_x_gen(sbr, sbr->X[ch],
1533 (const INTFLOAT (*)[64][2]) sbr->data[ch].Y[1-sbr->data[ch].Ypos],
1534 (const INTFLOAT (*)[64][2]) sbr->data[ch].Y[ sbr->data[ch].Ypos],
1535 (const INTFLOAT (*)[40][2]) sbr->X_low, ch);
1538 if (ac->oc[1].m4ac.ps == 1) {
1539 if (sbr->ps.start) {
1540 AAC_RENAME(ff_ps_apply)(ac->avctx, &sbr->ps, sbr->X[0], sbr->X[1], sbr->kx[1] + sbr->m[1]);
1542 memcpy(sbr->X[1], sbr->X[0], sizeof(sbr->X[0]));
1547 sbr_qmf_synthesis(&sbr->mdct, &sbr->dsp, ac->fdsp,
1548 L, sbr->X[0], sbr->qmf_filter_scratch,
1549 sbr->data[0].synthesis_filterbank_samples,
1550 &sbr->data[0].synthesis_filterbank_samples_offset,
1553 sbr_qmf_synthesis(&sbr->mdct, &sbr->dsp, ac->fdsp,
1554 R, sbr->X[1], sbr->qmf_filter_scratch,
1555 sbr->data[1].synthesis_filterbank_samples,
1556 &sbr->data[1].synthesis_filterbank_samples_offset,
1560 static void aacsbr_func_ptr_init(AACSBRContext *c)
1562 c->sbr_lf_gen = sbr_lf_gen;
1563 c->sbr_hf_assemble = sbr_hf_assemble;
1564 c->sbr_x_gen = sbr_x_gen;
1565 c->sbr_hf_inverse_filter = sbr_hf_inverse_filter;
1569 ff_aacsbr_func_ptr_init_mips(c);