2 * QDM2 compatible decoder
3 * Copyright (c) 2003 Ewald Snel
4 * Copyright (c) 2005 Benjamin Larsson
5 * Copyright (c) 2005 Alex Beregszaszi
6 * Copyright (c) 2005 Roberto Togni
8 * This file is part of Libav.
10 * Libav is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * Libav is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with Libav; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
28 * @author Ewald Snel, Benjamin Larsson, Alex Beregszaszi, Roberto Togni
30 * The decoder is not perfect yet, there are still some distortions
31 * especially on files encoded with 16 or 8 subbands.
38 #define BITSTREAM_READER_LE
39 #include "libavutil/channel_layout.h"
44 #include "mpegaudiodsp.h"
45 #include "mpegaudio.h"
48 #include "qdm2_tablegen.h"
54 #define QDM2_LIST_ADD(list, size, packet) \
57 list[size - 1].next = &list[size]; \
59 list[size].packet = packet; \
60 list[size].next = NULL; \
64 // Result is 8, 16 or 30
65 #define QDM2_SB_USED(sub_sampling) (((sub_sampling) >= 2) ? 30 : 8 << (sub_sampling))
67 #define FIX_NOISE_IDX(noise_idx) \
68 if ((noise_idx) >= 3840) \
69 (noise_idx) -= 3840; \
71 #define SB_DITHERING_NOISE(sb,noise_idx) (noise_table[(noise_idx)++] * sb_noise_attenuation[(sb)])
73 #define SAMPLES_NEEDED \
74 av_log (NULL,AV_LOG_INFO,"This file triggers some untested code. Please contact the developers.\n");
76 #define SAMPLES_NEEDED_2(why) \
77 av_log (NULL,AV_LOG_INFO,"This file triggers some missing code. Please contact the developers.\nPosition: %s\n",why);
79 #define QDM2_MAX_FRAME_SIZE 512
81 typedef int8_t sb_int8_array[2][30][64];
86 typedef struct QDM2SubPacket {
87 int type; ///< subpacket type
88 unsigned int size; ///< subpacket size
89 const uint8_t *data; ///< pointer to subpacket data (points to input data buffer, it's not a private copy)
93 * A node in the subpacket list
95 typedef struct QDM2SubPNode {
96 QDM2SubPacket *packet; ///< packet
97 struct QDM2SubPNode *next; ///< pointer to next packet in the list, NULL if leaf node
100 typedef struct QDM2Complex {
105 typedef struct FFTTone {
107 QDM2Complex *complex;
116 typedef struct FFTCoefficient {
124 typedef struct QDM2FFT {
125 DECLARE_ALIGNED(32, QDM2Complex, complex)[MPA_MAX_CHANNELS][256];
129 * QDM2 decoder context
131 typedef struct QDM2Context {
132 /// Parameters from codec header, do not change during playback
133 int nb_channels; ///< number of channels
134 int channels; ///< number of channels
135 int group_size; ///< size of frame group (16 frames per group)
136 int fft_size; ///< size of FFT, in complex numbers
137 int checksum_size; ///< size of data block, used also for checksum
139 /// Parameters built from header parameters, do not change during playback
140 int group_order; ///< order of frame group
141 int fft_order; ///< order of FFT (actually fftorder+1)
142 int frame_size; ///< size of data frame
144 int sub_sampling; ///< subsampling: 0=25%, 1=50%, 2=100% */
145 int coeff_per_sb_select; ///< selector for "num. of coeffs. per subband" tables. Can be 0, 1, 2
146 int cm_table_select; ///< selector for "coding method" tables. Can be 0, 1 (from init: 0-4)
148 /// Packets and packet lists
149 QDM2SubPacket sub_packets[16]; ///< the packets themselves
150 QDM2SubPNode sub_packet_list_A[16]; ///< list of all packets
151 QDM2SubPNode sub_packet_list_B[16]; ///< FFT packets B are on list
152 int sub_packets_B; ///< number of packets on 'B' list
153 QDM2SubPNode sub_packet_list_C[16]; ///< packets with errors?
154 QDM2SubPNode sub_packet_list_D[16]; ///< DCT packets
157 FFTTone fft_tones[1000];
160 FFTCoefficient fft_coefs[1000];
162 int fft_coefs_min_index[5];
163 int fft_coefs_max_index[5];
164 int fft_level_exp[6];
165 RDFTContext rdft_ctx;
169 const uint8_t *compressed_data;
171 float output_buffer[QDM2_MAX_FRAME_SIZE * 2];
174 MPADSPContext mpadsp;
175 DECLARE_ALIGNED(32, float, synth_buf)[MPA_MAX_CHANNELS][512*2];
176 int synth_buf_offset[MPA_MAX_CHANNELS];
177 DECLARE_ALIGNED(32, float, sb_samples)[MPA_MAX_CHANNELS][128][SBLIMIT];
178 DECLARE_ALIGNED(32, float, samples)[MPA_MAX_CHANNELS * MPA_FRAME_SIZE];
180 /// Mixed temporary data used in decoding
181 float tone_level[MPA_MAX_CHANNELS][30][64];
182 int8_t coding_method[MPA_MAX_CHANNELS][30][64];
183 int8_t quantized_coeffs[MPA_MAX_CHANNELS][10][8];
184 int8_t tone_level_idx_base[MPA_MAX_CHANNELS][30][8];
185 int8_t tone_level_idx_hi1[MPA_MAX_CHANNELS][3][8][8];
186 int8_t tone_level_idx_mid[MPA_MAX_CHANNELS][26][8];
187 int8_t tone_level_idx_hi2[MPA_MAX_CHANNELS][26];
188 int8_t tone_level_idx[MPA_MAX_CHANNELS][30][64];
189 int8_t tone_level_idx_temp[MPA_MAX_CHANNELS][30][64];
192 int has_errors; ///< packet has errors
193 int superblocktype_2_3; ///< select fft tables and some algorithm based on superblock type
194 int do_synth_filter; ///< used to perform or skip synthesis filter
197 int noise_idx; ///< index for dithering noise table
201 static VLC vlc_tab_level;
202 static VLC vlc_tab_diff;
203 static VLC vlc_tab_run;
204 static VLC fft_level_exp_alt_vlc;
205 static VLC fft_level_exp_vlc;
206 static VLC fft_stereo_exp_vlc;
207 static VLC fft_stereo_phase_vlc;
208 static VLC vlc_tab_tone_level_idx_hi1;
209 static VLC vlc_tab_tone_level_idx_mid;
210 static VLC vlc_tab_tone_level_idx_hi2;
211 static VLC vlc_tab_type30;
212 static VLC vlc_tab_type34;
213 static VLC vlc_tab_fft_tone_offset[5];
215 static const uint16_t qdm2_vlc_offs[] = {
216 0,260,566,598,894,1166,1230,1294,1678,1950,2214,2278,2310,2570,2834,3124,3448,3838,
219 static const int switchtable[23] = {
220 0, 5, 1, 5, 5, 5, 5, 5, 2, 5, 5, 5, 5, 5, 5, 5, 3, 5, 5, 5, 5, 5, 4
223 static av_cold void qdm2_init_vlc(void)
225 static VLC_TYPE qdm2_table[3838][2];
227 vlc_tab_level.table = &qdm2_table[qdm2_vlc_offs[0]];
228 vlc_tab_level.table_allocated = qdm2_vlc_offs[1] - qdm2_vlc_offs[0];
229 init_vlc(&vlc_tab_level, 8, 24,
230 vlc_tab_level_huffbits, 1, 1,
231 vlc_tab_level_huffcodes, 2, 2,
232 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
234 vlc_tab_diff.table = &qdm2_table[qdm2_vlc_offs[1]];
235 vlc_tab_diff.table_allocated = qdm2_vlc_offs[2] - qdm2_vlc_offs[1];
236 init_vlc(&vlc_tab_diff, 8, 37,
237 vlc_tab_diff_huffbits, 1, 1,
238 vlc_tab_diff_huffcodes, 2, 2,
239 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
241 vlc_tab_run.table = &qdm2_table[qdm2_vlc_offs[2]];
242 vlc_tab_run.table_allocated = qdm2_vlc_offs[3] - qdm2_vlc_offs[2];
243 init_vlc(&vlc_tab_run, 5, 6,
244 vlc_tab_run_huffbits, 1, 1,
245 vlc_tab_run_huffcodes, 1, 1,
246 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
248 fft_level_exp_alt_vlc.table = &qdm2_table[qdm2_vlc_offs[3]];
249 fft_level_exp_alt_vlc.table_allocated = qdm2_vlc_offs[4] -
251 init_vlc(&fft_level_exp_alt_vlc, 8, 28,
252 fft_level_exp_alt_huffbits, 1, 1,
253 fft_level_exp_alt_huffcodes, 2, 2,
254 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
256 fft_level_exp_vlc.table = &qdm2_table[qdm2_vlc_offs[4]];
257 fft_level_exp_vlc.table_allocated = qdm2_vlc_offs[5] - qdm2_vlc_offs[4];
258 init_vlc(&fft_level_exp_vlc, 8, 20,
259 fft_level_exp_huffbits, 1, 1,
260 fft_level_exp_huffcodes, 2, 2,
261 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
263 fft_stereo_exp_vlc.table = &qdm2_table[qdm2_vlc_offs[5]];
264 fft_stereo_exp_vlc.table_allocated = qdm2_vlc_offs[6] -
266 init_vlc(&fft_stereo_exp_vlc, 6, 7,
267 fft_stereo_exp_huffbits, 1, 1,
268 fft_stereo_exp_huffcodes, 1, 1,
269 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
271 fft_stereo_phase_vlc.table = &qdm2_table[qdm2_vlc_offs[6]];
272 fft_stereo_phase_vlc.table_allocated = qdm2_vlc_offs[7] -
274 init_vlc(&fft_stereo_phase_vlc, 6, 9,
275 fft_stereo_phase_huffbits, 1, 1,
276 fft_stereo_phase_huffcodes, 1, 1,
277 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
279 vlc_tab_tone_level_idx_hi1.table =
280 &qdm2_table[qdm2_vlc_offs[7]];
281 vlc_tab_tone_level_idx_hi1.table_allocated = qdm2_vlc_offs[8] -
283 init_vlc(&vlc_tab_tone_level_idx_hi1, 8, 20,
284 vlc_tab_tone_level_idx_hi1_huffbits, 1, 1,
285 vlc_tab_tone_level_idx_hi1_huffcodes, 2, 2,
286 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
288 vlc_tab_tone_level_idx_mid.table =
289 &qdm2_table[qdm2_vlc_offs[8]];
290 vlc_tab_tone_level_idx_mid.table_allocated = qdm2_vlc_offs[9] -
292 init_vlc(&vlc_tab_tone_level_idx_mid, 8, 24,
293 vlc_tab_tone_level_idx_mid_huffbits, 1, 1,
294 vlc_tab_tone_level_idx_mid_huffcodes, 2, 2,
295 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
297 vlc_tab_tone_level_idx_hi2.table =
298 &qdm2_table[qdm2_vlc_offs[9]];
299 vlc_tab_tone_level_idx_hi2.table_allocated = qdm2_vlc_offs[10] -
301 init_vlc(&vlc_tab_tone_level_idx_hi2, 8, 24,
302 vlc_tab_tone_level_idx_hi2_huffbits, 1, 1,
303 vlc_tab_tone_level_idx_hi2_huffcodes, 2, 2,
304 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
306 vlc_tab_type30.table = &qdm2_table[qdm2_vlc_offs[10]];
307 vlc_tab_type30.table_allocated = qdm2_vlc_offs[11] - qdm2_vlc_offs[10];
308 init_vlc(&vlc_tab_type30, 6, 9,
309 vlc_tab_type30_huffbits, 1, 1,
310 vlc_tab_type30_huffcodes, 1, 1,
311 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
313 vlc_tab_type34.table = &qdm2_table[qdm2_vlc_offs[11]];
314 vlc_tab_type34.table_allocated = qdm2_vlc_offs[12] - qdm2_vlc_offs[11];
315 init_vlc(&vlc_tab_type34, 5, 10,
316 vlc_tab_type34_huffbits, 1, 1,
317 vlc_tab_type34_huffcodes, 1, 1,
318 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
320 vlc_tab_fft_tone_offset[0].table =
321 &qdm2_table[qdm2_vlc_offs[12]];
322 vlc_tab_fft_tone_offset[0].table_allocated = qdm2_vlc_offs[13] -
324 init_vlc(&vlc_tab_fft_tone_offset[0], 8, 23,
325 vlc_tab_fft_tone_offset_0_huffbits, 1, 1,
326 vlc_tab_fft_tone_offset_0_huffcodes, 2, 2,
327 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
329 vlc_tab_fft_tone_offset[1].table =
330 &qdm2_table[qdm2_vlc_offs[13]];
331 vlc_tab_fft_tone_offset[1].table_allocated = qdm2_vlc_offs[14] -
333 init_vlc(&vlc_tab_fft_tone_offset[1], 8, 28,
334 vlc_tab_fft_tone_offset_1_huffbits, 1, 1,
335 vlc_tab_fft_tone_offset_1_huffcodes, 2, 2,
336 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
338 vlc_tab_fft_tone_offset[2].table =
339 &qdm2_table[qdm2_vlc_offs[14]];
340 vlc_tab_fft_tone_offset[2].table_allocated = qdm2_vlc_offs[15] -
342 init_vlc(&vlc_tab_fft_tone_offset[2], 8, 32,
343 vlc_tab_fft_tone_offset_2_huffbits, 1, 1,
344 vlc_tab_fft_tone_offset_2_huffcodes, 2, 2,
345 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
347 vlc_tab_fft_tone_offset[3].table =
348 &qdm2_table[qdm2_vlc_offs[15]];
349 vlc_tab_fft_tone_offset[3].table_allocated = qdm2_vlc_offs[16] -
351 init_vlc(&vlc_tab_fft_tone_offset[3], 8, 35,
352 vlc_tab_fft_tone_offset_3_huffbits, 1, 1,
353 vlc_tab_fft_tone_offset_3_huffcodes, 2, 2,
354 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
356 vlc_tab_fft_tone_offset[4].table =
357 &qdm2_table[qdm2_vlc_offs[16]];
358 vlc_tab_fft_tone_offset[4].table_allocated = qdm2_vlc_offs[17] -
360 init_vlc(&vlc_tab_fft_tone_offset[4], 8, 38,
361 vlc_tab_fft_tone_offset_4_huffbits, 1, 1,
362 vlc_tab_fft_tone_offset_4_huffcodes, 2, 2,
363 INIT_VLC_USE_NEW_STATIC | INIT_VLC_LE);
366 static int qdm2_get_vlc(GetBitContext *gb, VLC *vlc, int flag, int depth)
370 value = get_vlc2(gb, vlc->table, vlc->bits, depth);
372 /* stage-2, 3 bits exponent escape sequence */
374 value = get_bits(gb, get_bits(gb, 3) + 1);
376 /* stage-3, optional */
378 int tmp = vlc_stage3_values[value];
380 if ((value & ~3) > 0)
381 tmp += get_bits(gb, (value >> 2));
388 static int qdm2_get_se_vlc(VLC *vlc, GetBitContext *gb, int depth)
390 int value = qdm2_get_vlc(gb, vlc, 0, depth);
392 return (value & 1) ? ((value + 1) >> 1) : -(value >> 1);
398 * @param data pointer to data to be checksum'ed
399 * @param length data length
400 * @param value checksum value
402 * @return 0 if checksum is OK
404 static uint16_t qdm2_packet_checksum(const uint8_t *data, int length, int value)
408 for (i = 0; i < length; i++)
411 return (uint16_t)(value & 0xffff);
415 * Fill a QDM2SubPacket structure with packet type, size, and data pointer.
417 * @param gb bitreader context
418 * @param sub_packet packet under analysis
420 static void qdm2_decode_sub_packet_header(GetBitContext *gb,
421 QDM2SubPacket *sub_packet)
423 sub_packet->type = get_bits(gb, 8);
425 if (sub_packet->type == 0) {
426 sub_packet->size = 0;
427 sub_packet->data = NULL;
429 sub_packet->size = get_bits(gb, 8);
431 if (sub_packet->type & 0x80) {
432 sub_packet->size <<= 8;
433 sub_packet->size |= get_bits(gb, 8);
434 sub_packet->type &= 0x7f;
437 if (sub_packet->type == 0x7f)
438 sub_packet->type |= (get_bits(gb, 8) << 8);
440 // FIXME: this depends on bitreader-internal data
441 sub_packet->data = &gb->buffer[get_bits_count(gb) / 8];
444 av_log(NULL, AV_LOG_DEBUG, "Subpacket: type=%d size=%d start_offs=%x\n",
445 sub_packet->type, sub_packet->size, get_bits_count(gb) / 8);
449 * Return node pointer to first packet of requested type in list.
451 * @param list list of subpackets to be scanned
452 * @param type type of searched subpacket
453 * @return node pointer for subpacket if found, else NULL
455 static QDM2SubPNode *qdm2_search_subpacket_type_in_list(QDM2SubPNode *list,
458 while (list && list->packet) {
459 if (list->packet->type == type)
467 * Replace 8 elements with their average value.
468 * Called by qdm2_decode_superblock before starting subblock decoding.
472 static void average_quantized_coeffs(QDM2Context *q)
474 int i, j, n, ch, sum;
476 n = coeff_per_sb_for_avg[q->coeff_per_sb_select][QDM2_SB_USED(q->sub_sampling) - 1] + 1;
478 for (ch = 0; ch < q->nb_channels; ch++)
479 for (i = 0; i < n; i++) {
482 for (j = 0; j < 8; j++)
483 sum += q->quantized_coeffs[ch][i][j];
489 for (j = 0; j < 8; j++)
490 q->quantized_coeffs[ch][i][j] = sum;
495 * Build subband samples with noise weighted by q->tone_level.
496 * Called by synthfilt_build_sb_samples.
499 * @param sb subband index
501 static void build_sb_samples_from_noise(QDM2Context *q, int sb)
505 FIX_NOISE_IDX(q->noise_idx);
510 for (ch = 0; ch < q->nb_channels; ch++) {
511 for (j = 0; j < 64; j++) {
512 q->sb_samples[ch][j * 2][sb] =
513 SB_DITHERING_NOISE(sb, q->noise_idx) * q->tone_level[ch][sb][j];
514 q->sb_samples[ch][j * 2 + 1][sb] =
515 SB_DITHERING_NOISE(sb, q->noise_idx) * q->tone_level[ch][sb][j];
521 * Called while processing data from subpackets 11 and 12.
522 * Used after making changes to coding_method array.
524 * @param sb subband index
525 * @param channels number of channels
526 * @param coding_method q->coding_method[0][0][0]
528 static int fix_coding_method_array(int sb, int channels,
529 sb_int8_array coding_method)
535 for (ch = 0; ch < channels; ch++) {
536 for (j = 0; j < 64; ) {
537 if (coding_method[ch][sb][j] < 8)
539 if ((coding_method[ch][sb][j] - 8) > 22) {
543 switch (switchtable[coding_method[ch][sb][j] - 8]) {
567 for (k = 0; k < run; k++) {
569 if (coding_method[ch][sb + (j + k) / 64][(j + k) % 64] > coding_method[ch][sb][j]) {
572 //not debugged, almost never used
573 memset(&coding_method[ch][sb][j + k], case_val,
575 memset(&coding_method[ch][sb][j + k], case_val,
588 * Related to synthesis filter
589 * Called by process_subpacket_10
592 * @param flag 1 if called after getting data from subpacket 10, 0 if no subpacket 10
594 static void fill_tone_level_array(QDM2Context *q, int flag)
596 int i, sb, ch, sb_used;
599 for (ch = 0; ch < q->nb_channels; ch++)
600 for (sb = 0; sb < 30; sb++)
601 for (i = 0; i < 8; i++) {
602 if ((tab=coeff_per_sb_for_dequant[q->coeff_per_sb_select][sb]) < (last_coeff[q->coeff_per_sb_select] - 1))
603 tmp = q->quantized_coeffs[ch][tab + 1][i] * dequant_table[q->coeff_per_sb_select][tab + 1][sb]+
604 q->quantized_coeffs[ch][tab][i] * dequant_table[q->coeff_per_sb_select][tab][sb];
606 tmp = q->quantized_coeffs[ch][tab][i] * dequant_table[q->coeff_per_sb_select][tab][sb];
609 q->tone_level_idx_base[ch][sb][i] = (tmp / 256) & 0xff;
612 sb_used = QDM2_SB_USED(q->sub_sampling);
614 if ((q->superblocktype_2_3 != 0) && !flag) {
615 for (sb = 0; sb < sb_used; sb++)
616 for (ch = 0; ch < q->nb_channels; ch++)
617 for (i = 0; i < 64; i++) {
618 q->tone_level_idx[ch][sb][i] = q->tone_level_idx_base[ch][sb][i / 8];
619 if (q->tone_level_idx[ch][sb][i] < 0)
620 q->tone_level[ch][sb][i] = 0;
622 q->tone_level[ch][sb][i] = fft_tone_level_table[0][q->tone_level_idx[ch][sb][i] & 0x3f];
625 tab = q->superblocktype_2_3 ? 0 : 1;
626 for (sb = 0; sb < sb_used; sb++) {
627 if ((sb >= 4) && (sb <= 23)) {
628 for (ch = 0; ch < q->nb_channels; ch++)
629 for (i = 0; i < 64; i++) {
630 tmp = q->tone_level_idx_base[ch][sb][i / 8] -
631 q->tone_level_idx_hi1[ch][sb / 8][i / 8][i % 8] -
632 q->tone_level_idx_mid[ch][sb - 4][i / 8] -
633 q->tone_level_idx_hi2[ch][sb - 4];
634 q->tone_level_idx[ch][sb][i] = tmp & 0xff;
635 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
636 q->tone_level[ch][sb][i] = 0;
638 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
642 for (ch = 0; ch < q->nb_channels; ch++)
643 for (i = 0; i < 64; i++) {
644 tmp = q->tone_level_idx_base[ch][sb][i / 8] -
645 q->tone_level_idx_hi1[ch][2][i / 8][i % 8] -
646 q->tone_level_idx_hi2[ch][sb - 4];
647 q->tone_level_idx[ch][sb][i] = tmp & 0xff;
648 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
649 q->tone_level[ch][sb][i] = 0;
651 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
654 for (ch = 0; ch < q->nb_channels; ch++)
655 for (i = 0; i < 64; i++) {
656 tmp = q->tone_level_idx[ch][sb][i] = q->tone_level_idx_base[ch][sb][i / 8];
657 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
658 q->tone_level[ch][sb][i] = 0;
660 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
669 * Related to synthesis filter
670 * Called by process_subpacket_11
671 * c is built with data from subpacket 11
672 * Most of this function is used only if superblock_type_2_3 == 0,
673 * never seen it in samples.
675 * @param tone_level_idx
676 * @param tone_level_idx_temp
677 * @param coding_method q->coding_method[0][0][0]
678 * @param nb_channels number of channels
679 * @param c coming from subpacket 11, passed as 8*c
680 * @param superblocktype_2_3 flag based on superblock packet type
681 * @param cm_table_select q->cm_table_select
683 static void fill_coding_method_array(sb_int8_array tone_level_idx,
684 sb_int8_array tone_level_idx_temp,
685 sb_int8_array coding_method,
687 int c, int superblocktype_2_3,
691 int tmp, acc, esp_40, comp;
692 int add1, add2, add3, add4;
695 if (!superblocktype_2_3) {
696 /* This case is untested, no samples available */
698 for (ch = 0; ch < nb_channels; ch++)
699 for (sb = 0; sb < 30; sb++) {
700 for (j = 1; j < 63; j++) { // The loop only iterates to 63 so the code doesn't overflow the buffer
701 add1 = tone_level_idx[ch][sb][j] - 10;
704 add2 = add3 = add4 = 0;
706 add2 = tone_level_idx[ch][sb - 2][j] + tone_level_idx_offset_table[sb][0] - 6;
711 add3 = tone_level_idx[ch][sb - 1][j] + tone_level_idx_offset_table[sb][1] - 6;
716 add4 = tone_level_idx[ch][sb + 1][j] + tone_level_idx_offset_table[sb][3] - 6;
720 tmp = tone_level_idx[ch][sb][j + 1] * 2 - add4 - add3 - add2 - add1;
723 tone_level_idx_temp[ch][sb][j + 1] = tmp & 0xff;
725 tone_level_idx_temp[ch][sb][0] = tone_level_idx_temp[ch][sb][1];
728 for (ch = 0; ch < nb_channels; ch++)
729 for (sb = 0; sb < 30; sb++)
730 for (j = 0; j < 64; j++)
731 acc += tone_level_idx_temp[ch][sb][j];
733 multres = 0x66666667LL * (acc * 10);
734 esp_40 = (multres >> 32) / 8 + ((multres & 0xffffffff) >> 31);
735 for (ch = 0; ch < nb_channels; ch++)
736 for (sb = 0; sb < 30; sb++)
737 for (j = 0; j < 64; j++) {
738 comp = tone_level_idx_temp[ch][sb][j]* esp_40 * 10;
741 comp /= 256; // signed shift
769 coding_method[ch][sb][j] = ((tmp & 0xfffa) + 30 )& 0xff;
771 for (sb = 0; sb < 30; sb++)
772 fix_coding_method_array(sb, nb_channels, coding_method);
773 for (ch = 0; ch < nb_channels; ch++)
774 for (sb = 0; sb < 30; sb++)
775 for (j = 0; j < 64; j++)
777 if (coding_method[ch][sb][j] < 10)
778 coding_method[ch][sb][j] = 10;
781 if (coding_method[ch][sb][j] < 16)
782 coding_method[ch][sb][j] = 16;
784 if (coding_method[ch][sb][j] < 30)
785 coding_method[ch][sb][j] = 30;
788 } else { // superblocktype_2_3 != 0
789 for (ch = 0; ch < nb_channels; ch++)
790 for (sb = 0; sb < 30; sb++)
791 for (j = 0; j < 64; j++)
792 coding_method[ch][sb][j] = coding_method_table[cm_table_select][sb];
797 * Called by process_subpacket_11 to process more data from subpacket 11
799 * Called by process_subpacket_12 to process data from subpacket 12 with
803 * @param gb bitreader context
804 * @param length packet length in bits
805 * @param sb_min lower subband processed (sb_min included)
806 * @param sb_max higher subband processed (sb_max excluded)
808 static void synthfilt_build_sb_samples(QDM2Context *q, GetBitContext *gb,
809 int length, int sb_min, int sb_max)
811 int sb, j, k, n, ch, run, channels;
812 int joined_stereo, zero_encoding;
814 float type34_div = 0;
815 float type34_predictor;
816 float samples[10], sign_bits[16];
819 // If no data use noise
820 for (sb=sb_min; sb < sb_max; sb++)
821 build_sb_samples_from_noise(q, sb);
826 for (sb = sb_min; sb < sb_max; sb++) {
827 channels = q->nb_channels;
829 if (q->nb_channels <= 1 || sb < 12)
834 joined_stereo = (get_bits_left(gb) >= 1) ? get_bits1(gb) : 0;
837 if (get_bits_left(gb) >= 16)
838 for (j = 0; j < 16; j++)
839 sign_bits[j] = get_bits1(gb);
841 for (j = 0; j < 64; j++)
842 if (q->coding_method[1][sb][j] > q->coding_method[0][sb][j])
843 q->coding_method[0][sb][j] = q->coding_method[1][sb][j];
845 if (fix_coding_method_array(sb, q->nb_channels,
847 build_sb_samples_from_noise(q, sb);
853 for (ch = 0; ch < channels; ch++) {
854 FIX_NOISE_IDX(q->noise_idx);
855 zero_encoding = (get_bits_left(gb) >= 1) ? get_bits1(gb) : 0;
856 type34_predictor = 0.0;
859 for (j = 0; j < 128; ) {
860 switch (q->coding_method[ch][sb][j / 2]) {
862 if (get_bits_left(gb) >= 10) {
864 for (k = 0; k < 5; k++) {
865 if ((j + 2 * k) >= 128)
867 samples[2 * k] = get_bits1(gb) ? dequant_1bit[joined_stereo][2 * get_bits1(gb)] : 0;
871 for (k = 0; k < 5; k++)
872 samples[2 * k] = dequant_1bit[joined_stereo][random_dequant_index[n][k]];
874 for (k = 0; k < 5; k++)
875 samples[2 * k + 1] = SB_DITHERING_NOISE(sb,q->noise_idx);
877 for (k = 0; k < 10; k++)
878 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
884 if (get_bits_left(gb) >= 1) {
889 f -= noise_samples[((sb + 1) * (j +5 * ch + 1)) & 127] * 9.0 / 40.0;
892 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
898 if (get_bits_left(gb) >= 10) {
900 for (k = 0; k < 5; k++) {
903 samples[k] = (get_bits1(gb) == 0) ? 0 : dequant_1bit[joined_stereo][2 * get_bits1(gb)];
906 n = get_bits (gb, 8);
907 for (k = 0; k < 5; k++)
908 samples[k] = dequant_1bit[joined_stereo][random_dequant_index[n][k]];
911 for (k = 0; k < 5; k++)
912 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
918 if (get_bits_left(gb) >= 7) {
920 for (k = 0; k < 3; k++)
921 samples[k] = (random_dequant_type24[n][k] - 2.0) * 0.5;
923 for (k = 0; k < 3; k++)
924 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
930 if (get_bits_left(gb) >= 4) {
931 unsigned index = qdm2_get_vlc(gb, &vlc_tab_type30, 0, 1);
932 if (index < FF_ARRAY_ELEMS(type30_dequant)) {
933 samples[0] = type30_dequant[index];
935 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
937 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
943 if (get_bits_left(gb) >= 7) {
945 type34_div = (float)(1 << get_bits(gb, 2));
946 samples[0] = ((float)get_bits(gb, 5) - 16.0) / 15.0;
947 type34_predictor = samples[0];
950 unsigned index = qdm2_get_vlc(gb, &vlc_tab_type34, 0, 1);
951 if (index < FF_ARRAY_ELEMS(type34_delta)) {
952 samples[0] = type34_delta[index] / type34_div + type34_predictor;
953 type34_predictor = samples[0];
955 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
958 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
964 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
970 for (k = 0; k < run && j + k < 128; k++) {
971 q->sb_samples[0][j + k][sb] =
972 q->tone_level[0][sb][(j + k) / 2] * samples[k];
973 if (q->nb_channels == 2) {
974 if (sign_bits[(j + k) / 8])
975 q->sb_samples[1][j + k][sb] =
976 q->tone_level[1][sb][(j + k) / 2] * -samples[k];
978 q->sb_samples[1][j + k][sb] =
979 q->tone_level[1][sb][(j + k) / 2] * samples[k];
983 for (k = 0; k < run; k++)
985 q->sb_samples[ch][j + k][sb] = q->tone_level[ch][sb][(j + k)/2] * samples[k];
995 * Init the first element of a channel in quantized_coeffs with data
996 * from packet 10 (quantized_coeffs[ch][0]).
997 * This is similar to process_subpacket_9, but for a single channel
998 * and for element [0]
999 * same VLC tables as process_subpacket_9 are used.
1001 * @param quantized_coeffs pointer to quantized_coeffs[ch][0]
1002 * @param gb bitreader context
1004 static void init_quantized_coeffs_elem0(int8_t *quantized_coeffs,
1007 int i, k, run, level, diff;
1009 if (get_bits_left(gb) < 16)
1011 level = qdm2_get_vlc(gb, &vlc_tab_level, 0, 2);
1013 quantized_coeffs[0] = level;
1015 for (i = 0; i < 7; ) {
1016 if (get_bits_left(gb) < 16)
1018 run = qdm2_get_vlc(gb, &vlc_tab_run, 0, 1) + 1;
1020 if (get_bits_left(gb) < 16)
1022 diff = qdm2_get_se_vlc(&vlc_tab_diff, gb, 2);
1024 for (k = 1; k <= run; k++)
1025 quantized_coeffs[i + k] = (level + ((k * diff) / run));
1033 * Related to synthesis filter, process data from packet 10
1034 * Init part of quantized_coeffs via function init_quantized_coeffs_elem0
1035 * Init tone_level_idx_hi1, tone_level_idx_hi2, tone_level_idx_mid with
1036 * data from packet 10
1039 * @param gb bitreader context
1041 static void init_tone_level_dequantization(QDM2Context *q, GetBitContext *gb)
1043 int sb, j, k, n, ch;
1045 for (ch = 0; ch < q->nb_channels; ch++) {
1046 init_quantized_coeffs_elem0(q->quantized_coeffs[ch][0], gb);
1048 if (get_bits_left(gb) < 16) {
1049 memset(q->quantized_coeffs[ch][0], 0, 8);
1054 n = q->sub_sampling + 1;
1056 for (sb = 0; sb < n; sb++)
1057 for (ch = 0; ch < q->nb_channels; ch++)
1058 for (j = 0; j < 8; j++) {
1059 if (get_bits_left(gb) < 1)
1061 if (get_bits1(gb)) {
1062 for (k=0; k < 8; k++) {
1063 if (get_bits_left(gb) < 16)
1065 q->tone_level_idx_hi1[ch][sb][j][k] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_hi1, 0, 2);
1068 for (k=0; k < 8; k++)
1069 q->tone_level_idx_hi1[ch][sb][j][k] = 0;
1073 n = QDM2_SB_USED(q->sub_sampling) - 4;
1075 for (sb = 0; sb < n; sb++)
1076 for (ch = 0; ch < q->nb_channels; ch++) {
1077 if (get_bits_left(gb) < 16)
1079 q->tone_level_idx_hi2[ch][sb] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_hi2, 0, 2);
1081 q->tone_level_idx_hi2[ch][sb] -= 16;
1083 for (j = 0; j < 8; j++)
1084 q->tone_level_idx_mid[ch][sb][j] = -16;
1087 n = QDM2_SB_USED(q->sub_sampling) - 5;
1089 for (sb = 0; sb < n; sb++)
1090 for (ch = 0; ch < q->nb_channels; ch++)
1091 for (j = 0; j < 8; j++) {
1092 if (get_bits_left(gb) < 16)
1094 q->tone_level_idx_mid[ch][sb][j] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_mid, 0, 2) - 32;
1099 * Process subpacket 9, init quantized_coeffs with data from it
1102 * @param node pointer to node with packet
1104 static void process_subpacket_9(QDM2Context *q, QDM2SubPNode *node)
1107 int i, j, k, n, ch, run, level, diff;
1109 init_get_bits(&gb, node->packet->data, node->packet->size * 8);
1111 n = coeff_per_sb_for_avg[q->coeff_per_sb_select][QDM2_SB_USED(q->sub_sampling) - 1] + 1;
1113 for (i = 1; i < n; i++)
1114 for (ch = 0; ch < q->nb_channels; ch++) {
1115 level = qdm2_get_vlc(&gb, &vlc_tab_level, 0, 2);
1116 q->quantized_coeffs[ch][i][0] = level;
1118 for (j = 0; j < (8 - 1); ) {
1119 run = qdm2_get_vlc(&gb, &vlc_tab_run, 0, 1) + 1;
1120 diff = qdm2_get_se_vlc(&vlc_tab_diff, &gb, 2);
1122 for (k = 1; k <= run; k++)
1123 q->quantized_coeffs[ch][i][j + k] = (level + ((k * diff) / run));
1130 for (ch = 0; ch < q->nb_channels; ch++)
1131 for (i = 0; i < 8; i++)
1132 q->quantized_coeffs[ch][0][i] = 0;
1136 * Process subpacket 10 if not null, else
1139 * @param node pointer to node with packet
1141 static void process_subpacket_10(QDM2Context *q, QDM2SubPNode *node)
1146 init_get_bits(&gb, node->packet->data, node->packet->size * 8);
1147 init_tone_level_dequantization(q, &gb);
1148 fill_tone_level_array(q, 1);
1150 fill_tone_level_array(q, 0);
1155 * Process subpacket 11
1158 * @param node pointer to node with packet
1160 static void process_subpacket_11(QDM2Context *q, QDM2SubPNode *node)
1166 length = node->packet->size * 8;
1167 init_get_bits(&gb, node->packet->data, length);
1171 int c = get_bits(&gb, 13);
1174 fill_coding_method_array(q->tone_level_idx,
1175 q->tone_level_idx_temp, q->coding_method,
1176 q->nb_channels, 8 * c,
1177 q->superblocktype_2_3, q->cm_table_select);
1180 synthfilt_build_sb_samples(q, &gb, length, 0, 8);
1184 * Process subpacket 12
1187 * @param node pointer to node with packet
1189 static void process_subpacket_12(QDM2Context *q, QDM2SubPNode *node)
1195 length = node->packet->size * 8;
1196 init_get_bits(&gb, node->packet->data, length);
1199 synthfilt_build_sb_samples(q, &gb, length, 8, QDM2_SB_USED(q->sub_sampling));
1203 * Process new subpackets for synthesis filter
1206 * @param list list with synthesis filter packets (list D)
1208 static void process_synthesis_subpackets(QDM2Context *q, QDM2SubPNode *list)
1210 QDM2SubPNode *nodes[4];
1212 nodes[0] = qdm2_search_subpacket_type_in_list(list, 9);
1214 process_subpacket_9(q, nodes[0]);
1216 nodes[1] = qdm2_search_subpacket_type_in_list(list, 10);
1218 process_subpacket_10(q, nodes[1]);
1220 process_subpacket_10(q, NULL);
1222 nodes[2] = qdm2_search_subpacket_type_in_list(list, 11);
1223 if (nodes[0] && nodes[1] && nodes[2])
1224 process_subpacket_11(q, nodes[2]);
1226 process_subpacket_11(q, NULL);
1228 nodes[3] = qdm2_search_subpacket_type_in_list(list, 12);
1229 if (nodes[0] && nodes[1] && nodes[3])
1230 process_subpacket_12(q, nodes[3]);
1232 process_subpacket_12(q, NULL);
1236 * Decode superblock, fill packet lists.
1240 static void qdm2_decode_super_block(QDM2Context *q)
1243 QDM2SubPacket header, *packet;
1244 int i, packet_bytes, sub_packet_size, sub_packets_D;
1245 unsigned int next_index = 0;
1247 memset(q->tone_level_idx_hi1, 0, sizeof(q->tone_level_idx_hi1));
1248 memset(q->tone_level_idx_mid, 0, sizeof(q->tone_level_idx_mid));
1249 memset(q->tone_level_idx_hi2, 0, sizeof(q->tone_level_idx_hi2));
1251 q->sub_packets_B = 0;
1254 average_quantized_coeffs(q); // average elements in quantized_coeffs[max_ch][10][8]
1256 init_get_bits(&gb, q->compressed_data, q->compressed_size * 8);
1257 qdm2_decode_sub_packet_header(&gb, &header);
1259 if (header.type < 2 || header.type >= 8) {
1261 av_log(NULL, AV_LOG_ERROR, "bad superblock type\n");
1265 q->superblocktype_2_3 = (header.type == 2 || header.type == 3);
1266 packet_bytes = (q->compressed_size - get_bits_count(&gb) / 8);
1268 init_get_bits(&gb, header.data, header.size * 8);
1270 if (header.type == 2 || header.type == 4 || header.type == 5) {
1271 int csum = 257 * get_bits(&gb, 8);
1272 csum += 2 * get_bits(&gb, 8);
1274 csum = qdm2_packet_checksum(q->compressed_data, q->checksum_size, csum);
1278 av_log(NULL, AV_LOG_ERROR, "bad packet checksum\n");
1283 q->sub_packet_list_B[0].packet = NULL;
1284 q->sub_packet_list_D[0].packet = NULL;
1286 for (i = 0; i < 6; i++)
1287 if (--q->fft_level_exp[i] < 0)
1288 q->fft_level_exp[i] = 0;
1290 for (i = 0; packet_bytes > 0; i++) {
1293 if (i >= FF_ARRAY_ELEMS(q->sub_packet_list_A)) {
1294 SAMPLES_NEEDED_2("too many packet bytes");
1298 q->sub_packet_list_A[i].next = NULL;
1301 q->sub_packet_list_A[i - 1].next = &q->sub_packet_list_A[i];
1303 /* seek to next block */
1304 init_get_bits(&gb, header.data, header.size * 8);
1305 skip_bits(&gb, next_index * 8);
1307 if (next_index >= header.size)
1311 /* decode subpacket */
1312 packet = &q->sub_packets[i];
1313 qdm2_decode_sub_packet_header(&gb, packet);
1314 next_index = packet->size + get_bits_count(&gb) / 8;
1315 sub_packet_size = ((packet->size > 0xff) ? 1 : 0) + packet->size + 2;
1317 if (packet->type == 0)
1320 if (sub_packet_size > packet_bytes) {
1321 if (packet->type != 10 && packet->type != 11 && packet->type != 12)
1323 packet->size += packet_bytes - sub_packet_size;
1326 packet_bytes -= sub_packet_size;
1328 /* add subpacket to 'all subpackets' list */
1329 q->sub_packet_list_A[i].packet = packet;
1331 /* add subpacket to related list */
1332 if (packet->type == 8) {
1333 SAMPLES_NEEDED_2("packet type 8");
1335 } else if (packet->type >= 9 && packet->type <= 12) {
1336 /* packets for MPEG Audio like Synthesis Filter */
1337 QDM2_LIST_ADD(q->sub_packet_list_D, sub_packets_D, packet);
1338 } else if (packet->type == 13) {
1339 for (j = 0; j < 6; j++)
1340 q->fft_level_exp[j] = get_bits(&gb, 6);
1341 } else if (packet->type == 14) {
1342 for (j = 0; j < 6; j++)
1343 q->fft_level_exp[j] = qdm2_get_vlc(&gb, &fft_level_exp_vlc, 0, 2);
1344 } else if (packet->type == 15) {
1345 SAMPLES_NEEDED_2("packet type 15")
1347 } else if (packet->type >= 16 && packet->type < 48 &&
1348 !fft_subpackets[packet->type - 16]) {
1349 /* packets for FFT */
1350 QDM2_LIST_ADD(q->sub_packet_list_B, q->sub_packets_B, packet);
1352 } // Packet bytes loop
1354 if (q->sub_packet_list_D[0].packet) {
1355 process_synthesis_subpackets(q, q->sub_packet_list_D);
1356 q->do_synth_filter = 1;
1357 } else if (q->do_synth_filter) {
1358 process_subpacket_10(q, NULL);
1359 process_subpacket_11(q, NULL);
1360 process_subpacket_12(q, NULL);
1364 static void qdm2_fft_init_coefficient(QDM2Context *q, int sub_packet,
1365 int offset, int duration, int channel,
1368 if (q->fft_coefs_min_index[duration] < 0)
1369 q->fft_coefs_min_index[duration] = q->fft_coefs_index;
1371 q->fft_coefs[q->fft_coefs_index].sub_packet =
1372 ((sub_packet >= 16) ? (sub_packet - 16) : sub_packet);
1373 q->fft_coefs[q->fft_coefs_index].channel = channel;
1374 q->fft_coefs[q->fft_coefs_index].offset = offset;
1375 q->fft_coefs[q->fft_coefs_index].exp = exp;
1376 q->fft_coefs[q->fft_coefs_index].phase = phase;
1377 q->fft_coefs_index++;
1380 static void qdm2_fft_decode_tones(QDM2Context *q, int duration,
1381 GetBitContext *gb, int b)
1383 int channel, stereo, phase, exp;
1384 int local_int_4, local_int_8, stereo_phase, local_int_10;
1385 int local_int_14, stereo_exp, local_int_20, local_int_28;
1391 local_int_8 = (4 - duration);
1392 local_int_10 = 1 << (q->group_order - duration - 1);
1396 if (q->superblocktype_2_3) {
1397 while ((n = qdm2_get_vlc(gb, &vlc_tab_fft_tone_offset[local_int_8], 1, 2)) < 2) {
1400 local_int_4 += local_int_10;
1401 local_int_28 += (1 << local_int_8);
1403 local_int_4 += 8 * local_int_10;
1404 local_int_28 += (8 << local_int_8);
1409 offset += qdm2_get_vlc(gb, &vlc_tab_fft_tone_offset[local_int_8], 1, 2);
1410 while (offset >= (local_int_10 - 1)) {
1411 offset += (1 - (local_int_10 - 1));
1412 local_int_4 += local_int_10;
1413 local_int_28 += (1 << local_int_8);
1417 if (local_int_4 >= q->group_size)
1420 local_int_14 = (offset >> local_int_8);
1421 if (local_int_14 >= FF_ARRAY_ELEMS(fft_level_index_table))
1424 if (q->nb_channels > 1) {
1425 channel = get_bits1(gb);
1426 stereo = get_bits1(gb);
1432 exp = qdm2_get_vlc(gb, (b ? &fft_level_exp_vlc : &fft_level_exp_alt_vlc), 0, 2);
1433 exp += q->fft_level_exp[fft_level_index_table[local_int_14]];
1434 exp = (exp < 0) ? 0 : exp;
1436 phase = get_bits(gb, 3);
1441 stereo_exp = (exp - qdm2_get_vlc(gb, &fft_stereo_exp_vlc, 0, 1));
1442 stereo_phase = (phase - qdm2_get_vlc(gb, &fft_stereo_phase_vlc, 0, 1));
1443 if (stereo_phase < 0)
1447 if (q->frequency_range > (local_int_14 + 1)) {
1448 int sub_packet = (local_int_20 + local_int_28);
1450 qdm2_fft_init_coefficient(q, sub_packet, offset, duration,
1451 channel, exp, phase);
1453 qdm2_fft_init_coefficient(q, sub_packet, offset, duration,
1455 stereo_exp, stereo_phase);
1461 static void qdm2_decode_fft_packets(QDM2Context *q)
1463 int i, j, min, max, value, type, unknown_flag;
1466 if (!q->sub_packet_list_B[0].packet)
1469 /* reset minimum indexes for FFT coefficients */
1470 q->fft_coefs_index = 0;
1471 for (i = 0; i < 5; i++)
1472 q->fft_coefs_min_index[i] = -1;
1474 /* process subpackets ordered by type, largest type first */
1475 for (i = 0, max = 256; i < q->sub_packets_B; i++) {
1476 QDM2SubPacket *packet = NULL;
1478 /* find subpacket with largest type less than max */
1479 for (j = 0, min = 0; j < q->sub_packets_B; j++) {
1480 value = q->sub_packet_list_B[j].packet->type;
1481 if (value > min && value < max) {
1483 packet = q->sub_packet_list_B[j].packet;
1489 /* check for errors (?) */
1494 (packet->type < 16 || packet->type >= 48 ||
1495 fft_subpackets[packet->type - 16]))
1498 /* decode FFT tones */
1499 init_get_bits(&gb, packet->data, packet->size * 8);
1501 if (packet->type >= 32 && packet->type < 48 && !fft_subpackets[packet->type - 16])
1506 type = packet->type;
1508 if ((type >= 17 && type < 24) || (type >= 33 && type < 40)) {
1509 int duration = q->sub_sampling + 5 - (type & 15);
1511 if (duration >= 0 && duration < 4)
1512 qdm2_fft_decode_tones(q, duration, &gb, unknown_flag);
1513 } else if (type == 31) {
1514 for (j = 0; j < 4; j++)
1515 qdm2_fft_decode_tones(q, j, &gb, unknown_flag);
1516 } else if (type == 46) {
1517 for (j = 0; j < 6; j++)
1518 q->fft_level_exp[j] = get_bits(&gb, 6);
1519 for (j = 0; j < 4; j++)
1520 qdm2_fft_decode_tones(q, j, &gb, unknown_flag);
1522 } // Loop on B packets
1524 /* calculate maximum indexes for FFT coefficients */
1525 for (i = 0, j = -1; i < 5; i++)
1526 if (q->fft_coefs_min_index[i] >= 0) {
1528 q->fft_coefs_max_index[j] = q->fft_coefs_min_index[i];
1532 q->fft_coefs_max_index[j] = q->fft_coefs_index;
1535 static void qdm2_fft_generate_tone(QDM2Context *q, FFTTone *tone)
1540 const double iscale = 2.0 * M_PI / 512.0;
1542 tone->phase += tone->phase_shift;
1544 /* calculate current level (maximum amplitude) of tone */
1545 level = fft_tone_envelope_table[tone->duration][tone->time_index] * tone->level;
1546 c.im = level * sin(tone->phase * iscale);
1547 c.re = level * cos(tone->phase * iscale);
1549 /* generate FFT coefficients for tone */
1550 if (tone->duration >= 3 || tone->cutoff >= 3) {
1551 tone->complex[0].im += c.im;
1552 tone->complex[0].re += c.re;
1553 tone->complex[1].im -= c.im;
1554 tone->complex[1].re -= c.re;
1556 f[1] = -tone->table[4];
1557 f[0] = tone->table[3] - tone->table[0];
1558 f[2] = 1.0 - tone->table[2] - tone->table[3];
1559 f[3] = tone->table[1] + tone->table[4] - 1.0;
1560 f[4] = tone->table[0] - tone->table[1];
1561 f[5] = tone->table[2];
1562 for (i = 0; i < 2; i++) {
1563 tone->complex[fft_cutoff_index_table[tone->cutoff][i]].re +=
1565 tone->complex[fft_cutoff_index_table[tone->cutoff][i]].im +=
1566 c.im * ((tone->cutoff <= i) ? -f[i] : f[i]);
1568 for (i = 0; i < 4; i++) {
1569 tone->complex[i].re += c.re * f[i + 2];
1570 tone->complex[i].im += c.im * f[i + 2];
1574 /* copy the tone if it has not yet died out */
1575 if (++tone->time_index < ((1 << (5 - tone->duration)) - 1)) {
1576 memcpy(&q->fft_tones[q->fft_tone_end], tone, sizeof(FFTTone));
1577 q->fft_tone_end = (q->fft_tone_end + 1) % 1000;
1581 static void qdm2_fft_tone_synthesizer(QDM2Context *q, int sub_packet)
1584 const double iscale = 0.25 * M_PI;
1586 for (ch = 0; ch < q->channels; ch++) {
1587 memset(q->fft.complex[ch], 0, q->fft_size * sizeof(QDM2Complex));
1591 /* apply FFT tones with duration 4 (1 FFT period) */
1592 if (q->fft_coefs_min_index[4] >= 0)
1593 for (i = q->fft_coefs_min_index[4]; i < q->fft_coefs_max_index[4]; i++) {
1597 if (q->fft_coefs[i].sub_packet != sub_packet)
1600 ch = (q->channels == 1) ? 0 : q->fft_coefs[i].channel;
1601 level = (q->fft_coefs[i].exp < 0) ? 0.0 : fft_tone_level_table[q->superblocktype_2_3 ? 0 : 1][q->fft_coefs[i].exp & 63];
1603 c.re = level * cos(q->fft_coefs[i].phase * iscale);
1604 c.im = level * sin(q->fft_coefs[i].phase * iscale);
1605 q->fft.complex[ch][q->fft_coefs[i].offset + 0].re += c.re;
1606 q->fft.complex[ch][q->fft_coefs[i].offset + 0].im += c.im;
1607 q->fft.complex[ch][q->fft_coefs[i].offset + 1].re -= c.re;
1608 q->fft.complex[ch][q->fft_coefs[i].offset + 1].im -= c.im;
1611 /* generate existing FFT tones */
1612 for (i = q->fft_tone_end; i != q->fft_tone_start; ) {
1613 qdm2_fft_generate_tone(q, &q->fft_tones[q->fft_tone_start]);
1614 q->fft_tone_start = (q->fft_tone_start + 1) % 1000;
1617 /* create and generate new FFT tones with duration 0 (long) to 3 (short) */
1618 for (i = 0; i < 4; i++)
1619 if (q->fft_coefs_min_index[i] >= 0) {
1620 for (j = q->fft_coefs_min_index[i]; j < q->fft_coefs_max_index[i]; j++) {
1624 if (q->fft_coefs[j].sub_packet != sub_packet)
1628 offset = q->fft_coefs[j].offset >> four_i;
1629 ch = (q->channels == 1) ? 0 : q->fft_coefs[j].channel;
1631 if (offset < q->frequency_range) {
1633 tone.cutoff = offset;
1635 tone.cutoff = (offset >= 60) ? 3 : 2;
1637 tone.level = (q->fft_coefs[j].exp < 0) ? 0.0 : fft_tone_level_table[q->superblocktype_2_3 ? 0 : 1][q->fft_coefs[j].exp & 63];
1638 tone.complex = &q->fft.complex[ch][offset];
1639 tone.table = fft_tone_sample_table[i][q->fft_coefs[j].offset - (offset << four_i)];
1640 tone.phase = 64 * q->fft_coefs[j].phase - (offset << 8) - 128;
1641 tone.phase_shift = (2 * q->fft_coefs[j].offset + 1) << (7 - four_i);
1643 tone.time_index = 0;
1645 qdm2_fft_generate_tone(q, &tone);
1648 q->fft_coefs_min_index[i] = j;
1652 static void qdm2_calculate_fft(QDM2Context *q, int channel, int sub_packet)
1654 const float gain = (q->channels == 1 && q->nb_channels == 2) ? 0.5f : 1.0f;
1655 float *out = q->output_buffer + channel;
1657 q->fft.complex[channel][0].re *= 2.0f;
1658 q->fft.complex[channel][0].im = 0.0f;
1659 q->rdft_ctx.rdft_calc(&q->rdft_ctx, (FFTSample *)q->fft.complex[channel]);
1660 /* add samples to output buffer */
1661 for (i = 0; i < FFALIGN(q->fft_size, 8); i++) {
1662 out[0] += q->fft.complex[channel][i].re * gain;
1663 out[q->channels] += q->fft.complex[channel][i].im * gain;
1664 out += 2 * q->channels;
1670 * @param index subpacket number
1672 static void qdm2_synthesis_filter(QDM2Context *q, int index)
1674 int i, k, ch, sb_used, sub_sampling, dither_state = 0;
1676 /* copy sb_samples */
1677 sb_used = QDM2_SB_USED(q->sub_sampling);
1679 for (ch = 0; ch < q->channels; ch++)
1680 for (i = 0; i < 8; i++)
1681 for (k = sb_used; k < SBLIMIT; k++)
1682 q->sb_samples[ch][(8 * index) + i][k] = 0;
1684 for (ch = 0; ch < q->nb_channels; ch++) {
1685 float *samples_ptr = q->samples + ch;
1687 for (i = 0; i < 8; i++) {
1688 ff_mpa_synth_filter_float(&q->mpadsp,
1689 q->synth_buf[ch], &(q->synth_buf_offset[ch]),
1690 ff_mpa_synth_window_float, &dither_state,
1691 samples_ptr, q->nb_channels,
1692 q->sb_samples[ch][(8 * index) + i]);
1693 samples_ptr += 32 * q->nb_channels;
1697 /* add samples to output buffer */
1698 sub_sampling = (4 >> q->sub_sampling);
1700 for (ch = 0; ch < q->channels; ch++)
1701 for (i = 0; i < q->frame_size; i++)
1702 q->output_buffer[q->channels * i + ch] += (1 << 23) * q->samples[q->nb_channels * sub_sampling * i + ch];
1706 * Init static data (does not depend on specific file)
1710 static av_cold void qdm2_init_static_data(AVCodec *codec) {
1712 ff_mpa_synth_init_float(ff_mpa_synth_window_float);
1713 softclip_table_init();
1715 init_noise_samples();
1719 * Init parameters from codec extradata
1721 static av_cold int qdm2_decode_init(AVCodecContext *avctx)
1723 QDM2Context *s = avctx->priv_data;
1726 int tmp_val, tmp, size;
1728 /* extradata parsing
1737 32 size (including this field)
1739 32 type (=QDM2 or QDMC)
1741 32 size (including this field, in bytes)
1742 32 tag (=QDCA) // maybe mandatory parameters
1745 32 samplerate (=44100)
1747 32 block size (=4096)
1748 32 frame size (=256) (for one channel)
1749 32 packet size (=1300)
1751 32 size (including this field, in bytes)
1752 32 tag (=QDCP) // maybe some tuneable parameters
1762 if (!avctx->extradata || (avctx->extradata_size < 48)) {
1763 av_log(avctx, AV_LOG_ERROR, "extradata missing or truncated\n");
1764 return AVERROR_INVALIDDATA;
1767 extradata = avctx->extradata;
1768 extradata_size = avctx->extradata_size;
1770 while (extradata_size > 7) {
1771 if (!memcmp(extradata, "frmaQDM", 7))
1777 if (extradata_size < 12) {
1778 av_log(avctx, AV_LOG_ERROR, "not enough extradata (%i)\n",
1780 return AVERROR_INVALIDDATA;
1783 if (memcmp(extradata, "frmaQDM", 7)) {
1784 av_log(avctx, AV_LOG_ERROR, "invalid headers, QDM? not found\n");
1785 return AVERROR_INVALIDDATA;
1788 if (extradata[7] == 'C') {
1790 avpriv_report_missing_feature(avctx, "QDMC version 1");
1791 return AVERROR_PATCHWELCOME;
1795 extradata_size -= 8;
1797 size = AV_RB32(extradata);
1799 if(size > extradata_size){
1800 av_log(avctx, AV_LOG_ERROR, "extradata size too small, %i < %i\n",
1801 extradata_size, size);
1802 return AVERROR_INVALIDDATA;
1806 av_log(avctx, AV_LOG_DEBUG, "size: %d\n", size);
1807 if (AV_RB32(extradata) != MKBETAG('Q','D','C','A')) {
1808 av_log(avctx, AV_LOG_ERROR, "invalid extradata, expecting QDCA\n");
1809 return AVERROR_INVALIDDATA;
1814 avctx->channels = s->nb_channels = s->channels = AV_RB32(extradata);
1816 if (s->channels <= 0 || s->channels > MPA_MAX_CHANNELS)
1817 return AVERROR_INVALIDDATA;
1818 avctx->channel_layout = avctx->channels == 2 ? AV_CH_LAYOUT_STEREO :
1821 avctx->sample_rate = AV_RB32(extradata);
1824 avctx->bit_rate = AV_RB32(extradata);
1827 s->group_size = AV_RB32(extradata);
1830 s->fft_size = AV_RB32(extradata);
1833 s->checksum_size = AV_RB32(extradata);
1834 if (s->checksum_size >= 1U << 28) {
1835 av_log(avctx, AV_LOG_ERROR, "data block size too large (%u)\n", s->checksum_size);
1836 return AVERROR_INVALIDDATA;
1839 s->fft_order = av_log2(s->fft_size) + 1;
1841 // something like max decodable tones
1842 s->group_order = av_log2(s->group_size) + 1;
1843 s->frame_size = s->group_size / 16; // 16 iterations per super block
1844 if (s->frame_size > QDM2_MAX_FRAME_SIZE)
1845 return AVERROR_INVALIDDATA;
1847 s->sub_sampling = s->fft_order - 7;
1848 s->frequency_range = 255 / (1 << (2 - s->sub_sampling));
1850 switch ((s->sub_sampling * 2 + s->channels - 1)) {
1851 case 0: tmp = 40; break;
1852 case 1: tmp = 48; break;
1853 case 2: tmp = 56; break;
1854 case 3: tmp = 72; break;
1855 case 4: tmp = 80; break;
1856 case 5: tmp = 100;break;
1857 default: tmp=s->sub_sampling; break;
1860 if ((tmp * 1000) < avctx->bit_rate) tmp_val = 1;
1861 if ((tmp * 1440) < avctx->bit_rate) tmp_val = 2;
1862 if ((tmp * 1760) < avctx->bit_rate) tmp_val = 3;
1863 if ((tmp * 2240) < avctx->bit_rate) tmp_val = 4;
1864 s->cm_table_select = tmp_val;
1866 if (s->sub_sampling == 0)
1869 tmp = ((-(s->sub_sampling -1)) & 8000) + 20000;
1876 s->coeff_per_sb_select = 0;
1877 else if (tmp <= 16000)
1878 s->coeff_per_sb_select = 1;
1880 s->coeff_per_sb_select = 2;
1882 // Fail on unknown fft order
1883 if ((s->fft_order < 7) || (s->fft_order > 9)) {
1884 avpriv_request_sample(avctx, "Unknown FFT order %d", s->fft_order);
1885 return AVERROR_PATCHWELCOME;
1887 if (s->fft_size != (1 << (s->fft_order - 1))) {
1888 av_log(avctx, AV_LOG_ERROR, "FFT size %d not power of 2.\n", s->fft_size);
1889 return AVERROR_INVALIDDATA;
1892 ff_rdft_init(&s->rdft_ctx, s->fft_order, IDFT_C2R);
1893 ff_mpadsp_init(&s->mpadsp);
1895 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
1900 static av_cold int qdm2_decode_close(AVCodecContext *avctx)
1902 QDM2Context *s = avctx->priv_data;
1904 ff_rdft_end(&s->rdft_ctx);
1909 static int qdm2_decode(QDM2Context *q, const uint8_t *in, int16_t *out)
1912 const int frame_size = (q->frame_size * q->channels);
1914 /* select input buffer */
1915 q->compressed_data = in;
1916 q->compressed_size = q->checksum_size;
1918 /* copy old block, clear new block of output samples */
1919 memmove(q->output_buffer, &q->output_buffer[frame_size], frame_size * sizeof(float));
1920 memset(&q->output_buffer[frame_size], 0, frame_size * sizeof(float));
1922 /* decode block of QDM2 compressed data */
1923 if (q->sub_packet == 0) {
1924 q->has_errors = 0; // zero it for a new super block
1925 av_log(NULL,AV_LOG_DEBUG,"Superblock follows\n");
1926 qdm2_decode_super_block(q);
1929 /* parse subpackets */
1930 if (!q->has_errors) {
1931 if (q->sub_packet == 2)
1932 qdm2_decode_fft_packets(q);
1934 qdm2_fft_tone_synthesizer(q, q->sub_packet);
1937 /* sound synthesis stage 1 (FFT) */
1938 for (ch = 0; ch < q->channels; ch++) {
1939 qdm2_calculate_fft(q, ch, q->sub_packet);
1941 if (!q->has_errors && q->sub_packet_list_C[0].packet) {
1942 SAMPLES_NEEDED_2("has errors, and C list is not empty")
1947 /* sound synthesis stage 2 (MPEG audio like synthesis filter) */
1948 if (!q->has_errors && q->do_synth_filter)
1949 qdm2_synthesis_filter(q, q->sub_packet);
1951 q->sub_packet = (q->sub_packet + 1) % 16;
1953 /* clip and convert output float[] to 16bit signed samples */
1954 for (i = 0; i < frame_size; i++) {
1955 int value = (int)q->output_buffer[i];
1957 if (value > SOFTCLIP_THRESHOLD)
1958 value = (value > HARDCLIP_THRESHOLD) ? 32767 : softclip_table[ value - SOFTCLIP_THRESHOLD];
1959 else if (value < -SOFTCLIP_THRESHOLD)
1960 value = (value < -HARDCLIP_THRESHOLD) ? -32767 : -softclip_table[-value - SOFTCLIP_THRESHOLD];
1968 static int qdm2_decode_frame(AVCodecContext *avctx, void *data,
1969 int *got_frame_ptr, AVPacket *avpkt)
1971 AVFrame *frame = data;
1972 const uint8_t *buf = avpkt->data;
1973 int buf_size = avpkt->size;
1974 QDM2Context *s = avctx->priv_data;
1980 if(buf_size < s->checksum_size)
1983 /* get output buffer */
1984 frame->nb_samples = 16 * s->frame_size;
1985 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1986 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1989 out = (int16_t *)frame->data[0];
1991 for (i = 0; i < 16; i++) {
1992 if ((ret = qdm2_decode(s, buf, out)) < 0)
1994 out += s->channels * s->frame_size;
1999 return s->checksum_size;
2002 AVCodec ff_qdm2_decoder = {
2004 .long_name = NULL_IF_CONFIG_SMALL("QDesign Music Codec 2"),
2005 .type = AVMEDIA_TYPE_AUDIO,
2006 .id = AV_CODEC_ID_QDM2,
2007 .priv_data_size = sizeof(QDM2Context),
2008 .init = qdm2_decode_init,
2009 .init_static_data = qdm2_init_static_data,
2010 .close = qdm2_decode_close,
2011 .decode = qdm2_decode_frame,
2012 .capabilities = AV_CODEC_CAP_DR1,