blob: 88b6b19d1123716a9dac520b6ef6079a8b20bac2
1 | /* |
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 |
7 | * |
8 | * This file is part of FFmpeg. |
9 | * |
10 | * FFmpeg 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. |
14 | * |
15 | * FFmpeg 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. |
19 | * |
20 | * You should have received a copy of the GNU Lesser General Public |
21 | * License along with FFmpeg; if not, write to the Free Software |
22 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
23 | */ |
24 | |
25 | /** |
26 | * @file |
27 | * QDM2 decoder |
28 | * @author Ewald Snel, Benjamin Larsson, Alex Beregszaszi, Roberto Togni |
29 | * |
30 | * The decoder is not perfect yet, there are still some distortions |
31 | * especially on files encoded with 16 or 8 subbands. |
32 | */ |
33 | |
34 | #include <math.h> |
35 | #include <stddef.h> |
36 | #include <stdio.h> |
37 | |
38 | #include "libavutil/channel_layout.h" |
39 | |
40 | #define BITSTREAM_READER_LE |
41 | #include "avcodec.h" |
42 | #include "get_bits.h" |
43 | #include "bytestream.h" |
44 | #include "internal.h" |
45 | #include "mpegaudio.h" |
46 | #include "mpegaudiodsp.h" |
47 | #include "rdft.h" |
48 | |
49 | #include "qdm2_tablegen.h" |
50 | |
51 | #define QDM2_LIST_ADD(list, size, packet) \ |
52 | do { \ |
53 | if (size > 0) { \ |
54 | list[size - 1].next = &list[size]; \ |
55 | } \ |
56 | list[size].packet = packet; \ |
57 | list[size].next = NULL; \ |
58 | size++; \ |
59 | } while(0) |
60 | |
61 | // Result is 8, 16 or 30 |
62 | #define QDM2_SB_USED(sub_sampling) (((sub_sampling) >= 2) ? 30 : 8 << (sub_sampling)) |
63 | |
64 | #define FIX_NOISE_IDX(noise_idx) \ |
65 | if ((noise_idx) >= 3840) \ |
66 | (noise_idx) -= 3840; \ |
67 | |
68 | #define SB_DITHERING_NOISE(sb,noise_idx) (noise_table[(noise_idx)++] * sb_noise_attenuation[(sb)]) |
69 | |
70 | #define SAMPLES_NEEDED \ |
71 | av_log (NULL,AV_LOG_INFO,"This file triggers some untested code. Please contact the developers.\n"); |
72 | |
73 | #define SAMPLES_NEEDED_2(why) \ |
74 | av_log (NULL,AV_LOG_INFO,"This file triggers some missing code. Please contact the developers.\nPosition: %s\n",why); |
75 | |
76 | #define QDM2_MAX_FRAME_SIZE 512 |
77 | |
78 | typedef int8_t sb_int8_array[2][30][64]; |
79 | |
80 | /** |
81 | * Subpacket |
82 | */ |
83 | typedef struct QDM2SubPacket { |
84 | int type; ///< subpacket type |
85 | unsigned int size; ///< subpacket size |
86 | const uint8_t *data; ///< pointer to subpacket data (points to input data buffer, it's not a private copy) |
87 | } QDM2SubPacket; |
88 | |
89 | /** |
90 | * A node in the subpacket list |
91 | */ |
92 | typedef struct QDM2SubPNode { |
93 | QDM2SubPacket *packet; ///< packet |
94 | struct QDM2SubPNode *next; ///< pointer to next packet in the list, NULL if leaf node |
95 | } QDM2SubPNode; |
96 | |
97 | typedef struct QDM2Complex { |
98 | float re; |
99 | float im; |
100 | } QDM2Complex; |
101 | |
102 | typedef struct FFTTone { |
103 | float level; |
104 | QDM2Complex *complex; |
105 | const float *table; |
106 | int phase; |
107 | int phase_shift; |
108 | int duration; |
109 | short time_index; |
110 | short cutoff; |
111 | } FFTTone; |
112 | |
113 | typedef struct FFTCoefficient { |
114 | int16_t sub_packet; |
115 | uint8_t channel; |
116 | int16_t offset; |
117 | int16_t exp; |
118 | uint8_t phase; |
119 | } FFTCoefficient; |
120 | |
121 | typedef struct QDM2FFT { |
122 | DECLARE_ALIGNED(32, QDM2Complex, complex)[MPA_MAX_CHANNELS][256]; |
123 | } QDM2FFT; |
124 | |
125 | /** |
126 | * QDM2 decoder context |
127 | */ |
128 | typedef struct QDM2Context { |
129 | /// Parameters from codec header, do not change during playback |
130 | int nb_channels; ///< number of channels |
131 | int channels; ///< number of channels |
132 | int group_size; ///< size of frame group (16 frames per group) |
133 | int fft_size; ///< size of FFT, in complex numbers |
134 | int checksum_size; ///< size of data block, used also for checksum |
135 | |
136 | /// Parameters built from header parameters, do not change during playback |
137 | int group_order; ///< order of frame group |
138 | int fft_order; ///< order of FFT (actually fftorder+1) |
139 | int frame_size; ///< size of data frame |
140 | int frequency_range; |
141 | int sub_sampling; ///< subsampling: 0=25%, 1=50%, 2=100% */ |
142 | int coeff_per_sb_select; ///< selector for "num. of coeffs. per subband" tables. Can be 0, 1, 2 |
143 | int cm_table_select; ///< selector for "coding method" tables. Can be 0, 1 (from init: 0-4) |
144 | |
145 | /// Packets and packet lists |
146 | QDM2SubPacket sub_packets[16]; ///< the packets themselves |
147 | QDM2SubPNode sub_packet_list_A[16]; ///< list of all packets |
148 | QDM2SubPNode sub_packet_list_B[16]; ///< FFT packets B are on list |
149 | int sub_packets_B; ///< number of packets on 'B' list |
150 | QDM2SubPNode sub_packet_list_C[16]; ///< packets with errors? |
151 | QDM2SubPNode sub_packet_list_D[16]; ///< DCT packets |
152 | |
153 | /// FFT and tones |
154 | FFTTone fft_tones[1000]; |
155 | int fft_tone_start; |
156 | int fft_tone_end; |
157 | FFTCoefficient fft_coefs[1000]; |
158 | int fft_coefs_index; |
159 | int fft_coefs_min_index[5]; |
160 | int fft_coefs_max_index[5]; |
161 | int fft_level_exp[6]; |
162 | RDFTContext rdft_ctx; |
163 | QDM2FFT fft; |
164 | |
165 | /// I/O data |
166 | const uint8_t *compressed_data; |
167 | int compressed_size; |
168 | float output_buffer[QDM2_MAX_FRAME_SIZE * MPA_MAX_CHANNELS * 2]; |
169 | |
170 | /// Synthesis filter |
171 | MPADSPContext mpadsp; |
172 | DECLARE_ALIGNED(32, float, synth_buf)[MPA_MAX_CHANNELS][512*2]; |
173 | int synth_buf_offset[MPA_MAX_CHANNELS]; |
174 | DECLARE_ALIGNED(32, float, sb_samples)[MPA_MAX_CHANNELS][128][SBLIMIT]; |
175 | DECLARE_ALIGNED(32, float, samples)[MPA_MAX_CHANNELS * MPA_FRAME_SIZE]; |
176 | |
177 | /// Mixed temporary data used in decoding |
178 | float tone_level[MPA_MAX_CHANNELS][30][64]; |
179 | int8_t coding_method[MPA_MAX_CHANNELS][30][64]; |
180 | int8_t quantized_coeffs[MPA_MAX_CHANNELS][10][8]; |
181 | int8_t tone_level_idx_base[MPA_MAX_CHANNELS][30][8]; |
182 | int8_t tone_level_idx_hi1[MPA_MAX_CHANNELS][3][8][8]; |
183 | int8_t tone_level_idx_mid[MPA_MAX_CHANNELS][26][8]; |
184 | int8_t tone_level_idx_hi2[MPA_MAX_CHANNELS][26]; |
185 | int8_t tone_level_idx[MPA_MAX_CHANNELS][30][64]; |
186 | int8_t tone_level_idx_temp[MPA_MAX_CHANNELS][30][64]; |
187 | |
188 | // Flags |
189 | int has_errors; ///< packet has errors |
190 | int superblocktype_2_3; ///< select fft tables and some algorithm based on superblock type |
191 | int do_synth_filter; ///< used to perform or skip synthesis filter |
192 | |
193 | int sub_packet; |
194 | int noise_idx; ///< index for dithering noise table |
195 | } QDM2Context; |
196 | |
197 | static const int switchtable[23] = { |
198 | 0, 5, 1, 5, 5, 5, 5, 5, 2, 5, 5, 5, 5, 5, 5, 5, 3, 5, 5, 5, 5, 5, 4 |
199 | }; |
200 | |
201 | static int qdm2_get_vlc(GetBitContext *gb, const VLC *vlc, int flag, int depth) |
202 | { |
203 | int value; |
204 | |
205 | value = get_vlc2(gb, vlc->table, vlc->bits, depth); |
206 | |
207 | /* stage-2, 3 bits exponent escape sequence */ |
208 | if (value-- == 0) |
209 | value = get_bits(gb, get_bits(gb, 3) + 1); |
210 | |
211 | /* stage-3, optional */ |
212 | if (flag) { |
213 | int tmp; |
214 | |
215 | if (value >= 60) { |
216 | av_log(NULL, AV_LOG_ERROR, "value %d in qdm2_get_vlc too large\n", value); |
217 | return 0; |
218 | } |
219 | |
220 | tmp= vlc_stage3_values[value]; |
221 | |
222 | if ((value & ~3) > 0) |
223 | tmp += get_bits(gb, (value >> 2)); |
224 | value = tmp; |
225 | } |
226 | |
227 | return value; |
228 | } |
229 | |
230 | static int qdm2_get_se_vlc(const VLC *vlc, GetBitContext *gb, int depth) |
231 | { |
232 | int value = qdm2_get_vlc(gb, vlc, 0, depth); |
233 | |
234 | return (value & 1) ? ((value + 1) >> 1) : -(value >> 1); |
235 | } |
236 | |
237 | /** |
238 | * QDM2 checksum |
239 | * |
240 | * @param data pointer to data to be checksummed |
241 | * @param length data length |
242 | * @param value checksum value |
243 | * |
244 | * @return 0 if checksum is OK |
245 | */ |
246 | static uint16_t qdm2_packet_checksum(const uint8_t *data, int length, int value) |
247 | { |
248 | int i; |
249 | |
250 | for (i = 0; i < length; i++) |
251 | value -= data[i]; |
252 | |
253 | return (uint16_t)(value & 0xffff); |
254 | } |
255 | |
256 | /** |
257 | * Fill a QDM2SubPacket structure with packet type, size, and data pointer. |
258 | * |
259 | * @param gb bitreader context |
260 | * @param sub_packet packet under analysis |
261 | */ |
262 | static void qdm2_decode_sub_packet_header(GetBitContext *gb, |
263 | QDM2SubPacket *sub_packet) |
264 | { |
265 | sub_packet->type = get_bits(gb, 8); |
266 | |
267 | if (sub_packet->type == 0) { |
268 | sub_packet->size = 0; |
269 | sub_packet->data = NULL; |
270 | } else { |
271 | sub_packet->size = get_bits(gb, 8); |
272 | |
273 | if (sub_packet->type & 0x80) { |
274 | sub_packet->size <<= 8; |
275 | sub_packet->size |= get_bits(gb, 8); |
276 | sub_packet->type &= 0x7f; |
277 | } |
278 | |
279 | if (sub_packet->type == 0x7f) |
280 | sub_packet->type |= (get_bits(gb, 8) << 8); |
281 | |
282 | // FIXME: this depends on bitreader-internal data |
283 | sub_packet->data = &gb->buffer[get_bits_count(gb) / 8]; |
284 | } |
285 | |
286 | av_log(NULL, AV_LOG_DEBUG, "Subpacket: type=%d size=%d start_offs=%x\n", |
287 | sub_packet->type, sub_packet->size, get_bits_count(gb) / 8); |
288 | } |
289 | |
290 | /** |
291 | * Return node pointer to first packet of requested type in list. |
292 | * |
293 | * @param list list of subpackets to be scanned |
294 | * @param type type of searched subpacket |
295 | * @return node pointer for subpacket if found, else NULL |
296 | */ |
297 | static QDM2SubPNode *qdm2_search_subpacket_type_in_list(QDM2SubPNode *list, |
298 | int type) |
299 | { |
300 | while (list && list->packet) { |
301 | if (list->packet->type == type) |
302 | return list; |
303 | list = list->next; |
304 | } |
305 | return NULL; |
306 | } |
307 | |
308 | /** |
309 | * Replace 8 elements with their average value. |
310 | * Called by qdm2_decode_superblock before starting subblock decoding. |
311 | * |
312 | * @param q context |
313 | */ |
314 | static void average_quantized_coeffs(QDM2Context *q) |
315 | { |
316 | int i, j, n, ch, sum; |
317 | |
318 | n = coeff_per_sb_for_avg[q->coeff_per_sb_select][QDM2_SB_USED(q->sub_sampling) - 1] + 1; |
319 | |
320 | for (ch = 0; ch < q->nb_channels; ch++) |
321 | for (i = 0; i < n; i++) { |
322 | sum = 0; |
323 | |
324 | for (j = 0; j < 8; j++) |
325 | sum += q->quantized_coeffs[ch][i][j]; |
326 | |
327 | sum /= 8; |
328 | if (sum > 0) |
329 | sum--; |
330 | |
331 | for (j = 0; j < 8; j++) |
332 | q->quantized_coeffs[ch][i][j] = sum; |
333 | } |
334 | } |
335 | |
336 | /** |
337 | * Build subband samples with noise weighted by q->tone_level. |
338 | * Called by synthfilt_build_sb_samples. |
339 | * |
340 | * @param q context |
341 | * @param sb subband index |
342 | */ |
343 | static void build_sb_samples_from_noise(QDM2Context *q, int sb) |
344 | { |
345 | int ch, j; |
346 | |
347 | FIX_NOISE_IDX(q->noise_idx); |
348 | |
349 | if (!q->nb_channels) |
350 | return; |
351 | |
352 | for (ch = 0; ch < q->nb_channels; ch++) { |
353 | for (j = 0; j < 64; j++) { |
354 | q->sb_samples[ch][j * 2][sb] = |
355 | SB_DITHERING_NOISE(sb, q->noise_idx) * q->tone_level[ch][sb][j]; |
356 | q->sb_samples[ch][j * 2 + 1][sb] = |
357 | SB_DITHERING_NOISE(sb, q->noise_idx) * q->tone_level[ch][sb][j]; |
358 | } |
359 | } |
360 | } |
361 | |
362 | /** |
363 | * Called while processing data from subpackets 11 and 12. |
364 | * Used after making changes to coding_method array. |
365 | * |
366 | * @param sb subband index |
367 | * @param channels number of channels |
368 | * @param coding_method q->coding_method[0][0][0] |
369 | */ |
370 | static int fix_coding_method_array(int sb, int channels, |
371 | sb_int8_array coding_method) |
372 | { |
373 | int j, k; |
374 | int ch; |
375 | int run, case_val; |
376 | |
377 | for (ch = 0; ch < channels; ch++) { |
378 | for (j = 0; j < 64; ) { |
379 | if (coding_method[ch][sb][j] < 8) |
380 | return -1; |
381 | if ((coding_method[ch][sb][j] - 8) > 22) { |
382 | run = 1; |
383 | case_val = 8; |
384 | } else { |
385 | switch (switchtable[coding_method[ch][sb][j] - 8]) { |
386 | case 0: run = 10; |
387 | case_val = 10; |
388 | break; |
389 | case 1: run = 1; |
390 | case_val = 16; |
391 | break; |
392 | case 2: run = 5; |
393 | case_val = 24; |
394 | break; |
395 | case 3: run = 3; |
396 | case_val = 30; |
397 | break; |
398 | case 4: run = 1; |
399 | case_val = 30; |
400 | break; |
401 | case 5: run = 1; |
402 | case_val = 8; |
403 | break; |
404 | default: run = 1; |
405 | case_val = 8; |
406 | break; |
407 | } |
408 | } |
409 | for (k = 0; k < run; k++) { |
410 | if (j + k < 128) { |
411 | if (coding_method[ch][sb + (j + k) / 64][(j + k) % 64] > coding_method[ch][sb][j]) { |
412 | if (k > 0) { |
413 | SAMPLES_NEEDED |
414 | //not debugged, almost never used |
415 | memset(&coding_method[ch][sb][j + k], case_val, |
416 | k *sizeof(int8_t)); |
417 | memset(&coding_method[ch][sb][j + k], case_val, |
418 | 3 * sizeof(int8_t)); |
419 | } |
420 | } |
421 | } |
422 | } |
423 | j += run; |
424 | } |
425 | } |
426 | return 0; |
427 | } |
428 | |
429 | /** |
430 | * Related to synthesis filter |
431 | * Called by process_subpacket_10 |
432 | * |
433 | * @param q context |
434 | * @param flag 1 if called after getting data from subpacket 10, 0 if no subpacket 10 |
435 | */ |
436 | static void fill_tone_level_array(QDM2Context *q, int flag) |
437 | { |
438 | int i, sb, ch, sb_used; |
439 | int tmp, tab; |
440 | |
441 | for (ch = 0; ch < q->nb_channels; ch++) |
442 | for (sb = 0; sb < 30; sb++) |
443 | for (i = 0; i < 8; i++) { |
444 | if ((tab=coeff_per_sb_for_dequant[q->coeff_per_sb_select][sb]) < (last_coeff[q->coeff_per_sb_select] - 1)) |
445 | tmp = q->quantized_coeffs[ch][tab + 1][i] * dequant_table[q->coeff_per_sb_select][tab + 1][sb]+ |
446 | q->quantized_coeffs[ch][tab][i] * dequant_table[q->coeff_per_sb_select][tab][sb]; |
447 | else |
448 | tmp = q->quantized_coeffs[ch][tab][i] * dequant_table[q->coeff_per_sb_select][tab][sb]; |
449 | if(tmp < 0) |
450 | tmp += 0xff; |
451 | q->tone_level_idx_base[ch][sb][i] = (tmp / 256) & 0xff; |
452 | } |
453 | |
454 | sb_used = QDM2_SB_USED(q->sub_sampling); |
455 | |
456 | if ((q->superblocktype_2_3 != 0) && !flag) { |
457 | for (sb = 0; sb < sb_used; sb++) |
458 | for (ch = 0; ch < q->nb_channels; ch++) |
459 | for (i = 0; i < 64; i++) { |
460 | q->tone_level_idx[ch][sb][i] = q->tone_level_idx_base[ch][sb][i / 8]; |
461 | if (q->tone_level_idx[ch][sb][i] < 0) |
462 | q->tone_level[ch][sb][i] = 0; |
463 | else |
464 | q->tone_level[ch][sb][i] = fft_tone_level_table[0][q->tone_level_idx[ch][sb][i] & 0x3f]; |
465 | } |
466 | } else { |
467 | tab = q->superblocktype_2_3 ? 0 : 1; |
468 | for (sb = 0; sb < sb_used; sb++) { |
469 | if ((sb >= 4) && (sb <= 23)) { |
470 | for (ch = 0; ch < q->nb_channels; ch++) |
471 | for (i = 0; i < 64; i++) { |
472 | tmp = q->tone_level_idx_base[ch][sb][i / 8] - |
473 | q->tone_level_idx_hi1[ch][sb / 8][i / 8][i % 8] - |
474 | q->tone_level_idx_mid[ch][sb - 4][i / 8] - |
475 | q->tone_level_idx_hi2[ch][sb - 4]; |
476 | q->tone_level_idx[ch][sb][i] = tmp & 0xff; |
477 | if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp)) |
478 | q->tone_level[ch][sb][i] = 0; |
479 | else |
480 | q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f]; |
481 | } |
482 | } else { |
483 | if (sb > 4) { |
484 | for (ch = 0; ch < q->nb_channels; ch++) |
485 | for (i = 0; i < 64; i++) { |
486 | tmp = q->tone_level_idx_base[ch][sb][i / 8] - |
487 | q->tone_level_idx_hi1[ch][2][i / 8][i % 8] - |
488 | q->tone_level_idx_hi2[ch][sb - 4]; |
489 | q->tone_level_idx[ch][sb][i] = tmp & 0xff; |
490 | if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp)) |
491 | q->tone_level[ch][sb][i] = 0; |
492 | else |
493 | q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f]; |
494 | } |
495 | } else { |
496 | for (ch = 0; ch < q->nb_channels; ch++) |
497 | for (i = 0; i < 64; i++) { |
498 | tmp = q->tone_level_idx[ch][sb][i] = q->tone_level_idx_base[ch][sb][i / 8]; |
499 | if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp)) |
500 | q->tone_level[ch][sb][i] = 0; |
501 | else |
502 | q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f]; |
503 | } |
504 | } |
505 | } |
506 | } |
507 | } |
508 | } |
509 | |
510 | /** |
511 | * Related to synthesis filter |
512 | * Called by process_subpacket_11 |
513 | * c is built with data from subpacket 11 |
514 | * Most of this function is used only if superblock_type_2_3 == 0, |
515 | * never seen it in samples. |
516 | * |
517 | * @param tone_level_idx |
518 | * @param tone_level_idx_temp |
519 | * @param coding_method q->coding_method[0][0][0] |
520 | * @param nb_channels number of channels |
521 | * @param c coming from subpacket 11, passed as 8*c |
522 | * @param superblocktype_2_3 flag based on superblock packet type |
523 | * @param cm_table_select q->cm_table_select |
524 | */ |
525 | static void fill_coding_method_array(sb_int8_array tone_level_idx, |
526 | sb_int8_array tone_level_idx_temp, |
527 | sb_int8_array coding_method, |
528 | int nb_channels, |
529 | int c, int superblocktype_2_3, |
530 | int cm_table_select) |
531 | { |
532 | int ch, sb, j; |
533 | int tmp, acc, esp_40, comp; |
534 | int add1, add2, add3, add4; |
535 | int64_t multres; |
536 | |
537 | if (!superblocktype_2_3) { |
538 | /* This case is untested, no samples available */ |
539 | avpriv_request_sample(NULL, "!superblocktype_2_3"); |
540 | return; |
541 | for (ch = 0; ch < nb_channels; ch++) { |
542 | for (sb = 0; sb < 30; sb++) { |
543 | for (j = 1; j < 63; j++) { // The loop only iterates to 63 so the code doesn't overflow the buffer |
544 | add1 = tone_level_idx[ch][sb][j] - 10; |
545 | if (add1 < 0) |
546 | add1 = 0; |
547 | add2 = add3 = add4 = 0; |
548 | if (sb > 1) { |
549 | add2 = tone_level_idx[ch][sb - 2][j] + tone_level_idx_offset_table[sb][0] - 6; |
550 | if (add2 < 0) |
551 | add2 = 0; |
552 | } |
553 | if (sb > 0) { |
554 | add3 = tone_level_idx[ch][sb - 1][j] + tone_level_idx_offset_table[sb][1] - 6; |
555 | if (add3 < 0) |
556 | add3 = 0; |
557 | } |
558 | if (sb < 29) { |
559 | add4 = tone_level_idx[ch][sb + 1][j] + tone_level_idx_offset_table[sb][3] - 6; |
560 | if (add4 < 0) |
561 | add4 = 0; |
562 | } |
563 | tmp = tone_level_idx[ch][sb][j + 1] * 2 - add4 - add3 - add2 - add1; |
564 | if (tmp < 0) |
565 | tmp = 0; |
566 | tone_level_idx_temp[ch][sb][j + 1] = tmp & 0xff; |
567 | } |
568 | tone_level_idx_temp[ch][sb][0] = tone_level_idx_temp[ch][sb][1]; |
569 | } |
570 | } |
571 | acc = 0; |
572 | for (ch = 0; ch < nb_channels; ch++) |
573 | for (sb = 0; sb < 30; sb++) |
574 | for (j = 0; j < 64; j++) |
575 | acc += tone_level_idx_temp[ch][sb][j]; |
576 | |
577 | multres = 0x66666667LL * (acc * 10); |
578 | esp_40 = (multres >> 32) / 8 + ((multres & 0xffffffff) >> 31); |
579 | for (ch = 0; ch < nb_channels; ch++) |
580 | for (sb = 0; sb < 30; sb++) |
581 | for (j = 0; j < 64; j++) { |
582 | comp = tone_level_idx_temp[ch][sb][j]* esp_40 * 10; |
583 | if (comp < 0) |
584 | comp += 0xff; |
585 | comp /= 256; // signed shift |
586 | switch(sb) { |
587 | case 0: |
588 | if (comp < 30) |
589 | comp = 30; |
590 | comp += 15; |
591 | break; |
592 | case 1: |
593 | if (comp < 24) |
594 | comp = 24; |
595 | comp += 10; |
596 | break; |
597 | case 2: |
598 | case 3: |
599 | case 4: |
600 | if (comp < 16) |
601 | comp = 16; |
602 | } |
603 | if (comp <= 5) |
604 | tmp = 0; |
605 | else if (comp <= 10) |
606 | tmp = 10; |
607 | else if (comp <= 16) |
608 | tmp = 16; |
609 | else if (comp <= 24) |
610 | tmp = -1; |
611 | else |
612 | tmp = 0; |
613 | coding_method[ch][sb][j] = ((tmp & 0xfffa) + 30 )& 0xff; |
614 | } |
615 | for (sb = 0; sb < 30; sb++) |
616 | fix_coding_method_array(sb, nb_channels, coding_method); |
617 | for (ch = 0; ch < nb_channels; ch++) |
618 | for (sb = 0; sb < 30; sb++) |
619 | for (j = 0; j < 64; j++) |
620 | if (sb >= 10) { |
621 | if (coding_method[ch][sb][j] < 10) |
622 | coding_method[ch][sb][j] = 10; |
623 | } else { |
624 | if (sb >= 2) { |
625 | if (coding_method[ch][sb][j] < 16) |
626 | coding_method[ch][sb][j] = 16; |
627 | } else { |
628 | if (coding_method[ch][sb][j] < 30) |
629 | coding_method[ch][sb][j] = 30; |
630 | } |
631 | } |
632 | } else { // superblocktype_2_3 != 0 |
633 | for (ch = 0; ch < nb_channels; ch++) |
634 | for (sb = 0; sb < 30; sb++) |
635 | for (j = 0; j < 64; j++) |
636 | coding_method[ch][sb][j] = coding_method_table[cm_table_select][sb]; |
637 | } |
638 | } |
639 | |
640 | /** |
641 | * Called by process_subpacket_11 to process more data from subpacket 11 |
642 | * with sb 0-8. |
643 | * Called by process_subpacket_12 to process data from subpacket 12 with |
644 | * sb 8-sb_used. |
645 | * |
646 | * @param q context |
647 | * @param gb bitreader context |
648 | * @param length packet length in bits |
649 | * @param sb_min lower subband processed (sb_min included) |
650 | * @param sb_max higher subband processed (sb_max excluded) |
651 | */ |
652 | static int synthfilt_build_sb_samples(QDM2Context *q, GetBitContext *gb, |
653 | int length, int sb_min, int sb_max) |
654 | { |
655 | int sb, j, k, n, ch, run, channels; |
656 | int joined_stereo, zero_encoding; |
657 | int type34_first; |
658 | float type34_div = 0; |
659 | float type34_predictor; |
660 | float samples[10]; |
661 | int sign_bits[16] = {0}; |
662 | |
663 | if (length == 0) { |
664 | // If no data use noise |
665 | for (sb=sb_min; sb < sb_max; sb++) |
666 | build_sb_samples_from_noise(q, sb); |
667 | |
668 | return 0; |
669 | } |
670 | |
671 | for (sb = sb_min; sb < sb_max; sb++) { |
672 | channels = q->nb_channels; |
673 | |
674 | if (q->nb_channels <= 1 || sb < 12) |
675 | joined_stereo = 0; |
676 | else if (sb >= 24) |
677 | joined_stereo = 1; |
678 | else |
679 | joined_stereo = (get_bits_left(gb) >= 1) ? get_bits1(gb) : 0; |
680 | |
681 | if (joined_stereo) { |
682 | if (get_bits_left(gb) >= 16) |
683 | for (j = 0; j < 16; j++) |
684 | sign_bits[j] = get_bits1(gb); |
685 | |
686 | for (j = 0; j < 64; j++) |
687 | if (q->coding_method[1][sb][j] > q->coding_method[0][sb][j]) |
688 | q->coding_method[0][sb][j] = q->coding_method[1][sb][j]; |
689 | |
690 | if (fix_coding_method_array(sb, q->nb_channels, |
691 | q->coding_method)) { |
692 | av_log(NULL, AV_LOG_ERROR, "coding method invalid\n"); |
693 | build_sb_samples_from_noise(q, sb); |
694 | continue; |
695 | } |
696 | channels = 1; |
697 | } |
698 | |
699 | for (ch = 0; ch < channels; ch++) { |
700 | FIX_NOISE_IDX(q->noise_idx); |
701 | zero_encoding = (get_bits_left(gb) >= 1) ? get_bits1(gb) : 0; |
702 | type34_predictor = 0.0; |
703 | type34_first = 1; |
704 | |
705 | for (j = 0; j < 128; ) { |
706 | switch (q->coding_method[ch][sb][j / 2]) { |
707 | case 8: |
708 | if (get_bits_left(gb) >= 10) { |
709 | if (zero_encoding) { |
710 | for (k = 0; k < 5; k++) { |
711 | if ((j + 2 * k) >= 128) |
712 | break; |
713 | samples[2 * k] = get_bits1(gb) ? dequant_1bit[joined_stereo][2 * get_bits1(gb)] : 0; |
714 | } |
715 | } else { |
716 | n = get_bits(gb, 8); |
717 | if (n >= 243) { |
718 | av_log(NULL, AV_LOG_ERROR, "Invalid 8bit codeword\n"); |
719 | return AVERROR_INVALIDDATA; |
720 | } |
721 | |
722 | for (k = 0; k < 5; k++) |
723 | samples[2 * k] = dequant_1bit[joined_stereo][random_dequant_index[n][k]]; |
724 | } |
725 | for (k = 0; k < 5; k++) |
726 | samples[2 * k + 1] = SB_DITHERING_NOISE(sb,q->noise_idx); |
727 | } else { |
728 | for (k = 0; k < 10; k++) |
729 | samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx); |
730 | } |
731 | run = 10; |
732 | break; |
733 | |
734 | case 10: |
735 | if (get_bits_left(gb) >= 1) { |
736 | float f = 0.81; |
737 | |
738 | if (get_bits1(gb)) |
739 | f = -f; |
740 | f -= noise_samples[((sb + 1) * (j +5 * ch + 1)) & 127] * 9.0 / 40.0; |
741 | samples[0] = f; |
742 | } else { |
743 | samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx); |
744 | } |
745 | run = 1; |
746 | break; |
747 | |
748 | case 16: |
749 | if (get_bits_left(gb) >= 10) { |
750 | if (zero_encoding) { |
751 | for (k = 0; k < 5; k++) { |
752 | if ((j + k) >= 128) |
753 | break; |
754 | samples[k] = (get_bits1(gb) == 0) ? 0 : dequant_1bit[joined_stereo][2 * get_bits1(gb)]; |
755 | } |
756 | } else { |
757 | n = get_bits (gb, 8); |
758 | if (n >= 243) { |
759 | av_log(NULL, AV_LOG_ERROR, "Invalid 8bit codeword\n"); |
760 | return AVERROR_INVALIDDATA; |
761 | } |
762 | |
763 | for (k = 0; k < 5; k++) |
764 | samples[k] = dequant_1bit[joined_stereo][random_dequant_index[n][k]]; |
765 | } |
766 | } else { |
767 | for (k = 0; k < 5; k++) |
768 | samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx); |
769 | } |
770 | run = 5; |
771 | break; |
772 | |
773 | case 24: |
774 | if (get_bits_left(gb) >= 7) { |
775 | n = get_bits(gb, 7); |
776 | if (n >= 125) { |
777 | av_log(NULL, AV_LOG_ERROR, "Invalid 7bit codeword\n"); |
778 | return AVERROR_INVALIDDATA; |
779 | } |
780 | |
781 | for (k = 0; k < 3; k++) |
782 | samples[k] = (random_dequant_type24[n][k] - 2.0) * 0.5; |
783 | } else { |
784 | for (k = 0; k < 3; k++) |
785 | samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx); |
786 | } |
787 | run = 3; |
788 | break; |
789 | |
790 | case 30: |
791 | if (get_bits_left(gb) >= 4) { |
792 | unsigned index = qdm2_get_vlc(gb, &vlc_tab_type30, 0, 1); |
793 | if (index >= FF_ARRAY_ELEMS(type30_dequant)) { |
794 | av_log(NULL, AV_LOG_ERROR, "index %d out of type30_dequant array\n", index); |
795 | return AVERROR_INVALIDDATA; |
796 | } |
797 | samples[0] = type30_dequant[index]; |
798 | } else |
799 | samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx); |
800 | |
801 | run = 1; |
802 | break; |
803 | |
804 | case 34: |
805 | if (get_bits_left(gb) >= 7) { |
806 | if (type34_first) { |
807 | type34_div = (float)(1 << get_bits(gb, 2)); |
808 | samples[0] = ((float)get_bits(gb, 5) - 16.0) / 15.0; |
809 | type34_predictor = samples[0]; |
810 | type34_first = 0; |
811 | } else { |
812 | unsigned index = qdm2_get_vlc(gb, &vlc_tab_type34, 0, 1); |
813 | if (index >= FF_ARRAY_ELEMS(type34_delta)) { |
814 | av_log(NULL, AV_LOG_ERROR, "index %d out of type34_delta array\n", index); |
815 | return AVERROR_INVALIDDATA; |
816 | } |
817 | samples[0] = type34_delta[index] / type34_div + type34_predictor; |
818 | type34_predictor = samples[0]; |
819 | } |
820 | } else { |
821 | samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx); |
822 | } |
823 | run = 1; |
824 | break; |
825 | |
826 | default: |
827 | samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx); |
828 | run = 1; |
829 | break; |
830 | } |
831 | |
832 | if (joined_stereo) { |
833 | for (k = 0; k < run && j + k < 128; k++) { |
834 | q->sb_samples[0][j + k][sb] = |
835 | q->tone_level[0][sb][(j + k) / 2] * samples[k]; |
836 | if (q->nb_channels == 2) { |
837 | if (sign_bits[(j + k) / 8]) |
838 | q->sb_samples[1][j + k][sb] = |
839 | q->tone_level[1][sb][(j + k) / 2] * -samples[k]; |
840 | else |
841 | q->sb_samples[1][j + k][sb] = |
842 | q->tone_level[1][sb][(j + k) / 2] * samples[k]; |
843 | } |
844 | } |
845 | } else { |
846 | for (k = 0; k < run; k++) |
847 | if ((j + k) < 128) |
848 | q->sb_samples[ch][j + k][sb] = q->tone_level[ch][sb][(j + k)/2] * samples[k]; |
849 | } |
850 | |
851 | j += run; |
852 | } // j loop |
853 | } // channel loop |
854 | } // subband loop |
855 | return 0; |
856 | } |
857 | |
858 | /** |
859 | * Init the first element of a channel in quantized_coeffs with data |
860 | * from packet 10 (quantized_coeffs[ch][0]). |
861 | * This is similar to process_subpacket_9, but for a single channel |
862 | * and for element [0] |
863 | * same VLC tables as process_subpacket_9 are used. |
864 | * |
865 | * @param quantized_coeffs pointer to quantized_coeffs[ch][0] |
866 | * @param gb bitreader context |
867 | */ |
868 | static int init_quantized_coeffs_elem0(int8_t *quantized_coeffs, |
869 | GetBitContext *gb) |
870 | { |
871 | int i, k, run, level, diff; |
872 | |
873 | if (get_bits_left(gb) < 16) |
874 | return -1; |
875 | level = qdm2_get_vlc(gb, &vlc_tab_level, 0, 2); |
876 | |
877 | quantized_coeffs[0] = level; |
878 | |
879 | for (i = 0; i < 7; ) { |
880 | if (get_bits_left(gb) < 16) |
881 | return -1; |
882 | run = qdm2_get_vlc(gb, &vlc_tab_run, 0, 1) + 1; |
883 | |
884 | if (i + run >= 8) |
885 | return -1; |
886 | |
887 | if (get_bits_left(gb) < 16) |
888 | return -1; |
889 | diff = qdm2_get_se_vlc(&vlc_tab_diff, gb, 2); |
890 | |
891 | for (k = 1; k <= run; k++) |
892 | quantized_coeffs[i + k] = (level + ((k * diff) / run)); |
893 | |
894 | level += diff; |
895 | i += run; |
896 | } |
897 | return 0; |
898 | } |
899 | |
900 | /** |
901 | * Related to synthesis filter, process data from packet 10 |
902 | * Init part of quantized_coeffs via function init_quantized_coeffs_elem0 |
903 | * Init tone_level_idx_hi1, tone_level_idx_hi2, tone_level_idx_mid with |
904 | * data from packet 10 |
905 | * |
906 | * @param q context |
907 | * @param gb bitreader context |
908 | */ |
909 | static void init_tone_level_dequantization(QDM2Context *q, GetBitContext *gb) |
910 | { |
911 | int sb, j, k, n, ch; |
912 | |
913 | for (ch = 0; ch < q->nb_channels; ch++) { |
914 | init_quantized_coeffs_elem0(q->quantized_coeffs[ch][0], gb); |
915 | |
916 | if (get_bits_left(gb) < 16) { |
917 | memset(q->quantized_coeffs[ch][0], 0, 8); |
918 | break; |
919 | } |
920 | } |
921 | |
922 | n = q->sub_sampling + 1; |
923 | |
924 | for (sb = 0; sb < n; sb++) |
925 | for (ch = 0; ch < q->nb_channels; ch++) |
926 | for (j = 0; j < 8; j++) { |
927 | if (get_bits_left(gb) < 1) |
928 | break; |
929 | if (get_bits1(gb)) { |
930 | for (k=0; k < 8; k++) { |
931 | if (get_bits_left(gb) < 16) |
932 | break; |
933 | q->tone_level_idx_hi1[ch][sb][j][k] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_hi1, 0, 2); |
934 | } |
935 | } else { |
936 | for (k=0; k < 8; k++) |
937 | q->tone_level_idx_hi1[ch][sb][j][k] = 0; |
938 | } |
939 | } |
940 | |
941 | n = QDM2_SB_USED(q->sub_sampling) - 4; |
942 | |
943 | for (sb = 0; sb < n; sb++) |
944 | for (ch = 0; ch < q->nb_channels; ch++) { |
945 | if (get_bits_left(gb) < 16) |
946 | break; |
947 | q->tone_level_idx_hi2[ch][sb] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_hi2, 0, 2); |
948 | if (sb > 19) |
949 | q->tone_level_idx_hi2[ch][sb] -= 16; |
950 | else |
951 | for (j = 0; j < 8; j++) |
952 | q->tone_level_idx_mid[ch][sb][j] = -16; |
953 | } |
954 | |
955 | n = QDM2_SB_USED(q->sub_sampling) - 5; |
956 | |
957 | for (sb = 0; sb < n; sb++) |
958 | for (ch = 0; ch < q->nb_channels; ch++) |
959 | for (j = 0; j < 8; j++) { |
960 | if (get_bits_left(gb) < 16) |
961 | break; |
962 | q->tone_level_idx_mid[ch][sb][j] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_mid, 0, 2) - 32; |
963 | } |
964 | } |
965 | |
966 | /** |
967 | * Process subpacket 9, init quantized_coeffs with data from it |
968 | * |
969 | * @param q context |
970 | * @param node pointer to node with packet |
971 | */ |
972 | static int process_subpacket_9(QDM2Context *q, QDM2SubPNode *node) |
973 | { |
974 | GetBitContext gb; |
975 | int i, j, k, n, ch, run, level, diff; |
976 | |
977 | init_get_bits(&gb, node->packet->data, node->packet->size * 8); |
978 | |
979 | n = coeff_per_sb_for_avg[q->coeff_per_sb_select][QDM2_SB_USED(q->sub_sampling) - 1] + 1; |
980 | |
981 | for (i = 1; i < n; i++) |
982 | for (ch = 0; ch < q->nb_channels; ch++) { |
983 | level = qdm2_get_vlc(&gb, &vlc_tab_level, 0, 2); |
984 | q->quantized_coeffs[ch][i][0] = level; |
985 | |
986 | for (j = 0; j < (8 - 1); ) { |
987 | run = qdm2_get_vlc(&gb, &vlc_tab_run, 0, 1) + 1; |
988 | diff = qdm2_get_se_vlc(&vlc_tab_diff, &gb, 2); |
989 | |
990 | if (j + run >= 8) |
991 | return -1; |
992 | |
993 | for (k = 1; k <= run; k++) |
994 | q->quantized_coeffs[ch][i][j + k] = (level + ((k * diff) / run)); |
995 | |
996 | level += diff; |
997 | j += run; |
998 | } |
999 | } |
1000 | |
1001 | for (ch = 0; ch < q->nb_channels; ch++) |
1002 | for (i = 0; i < 8; i++) |
1003 | q->quantized_coeffs[ch][0][i] = 0; |
1004 | |
1005 | return 0; |
1006 | } |
1007 | |
1008 | /** |
1009 | * Process subpacket 10 if not null, else |
1010 | * |
1011 | * @param q context |
1012 | * @param node pointer to node with packet |
1013 | */ |
1014 | static void process_subpacket_10(QDM2Context *q, QDM2SubPNode *node) |
1015 | { |
1016 | GetBitContext gb; |
1017 | |
1018 | if (node) { |
1019 | init_get_bits(&gb, node->packet->data, node->packet->size * 8); |
1020 | init_tone_level_dequantization(q, &gb); |
1021 | fill_tone_level_array(q, 1); |
1022 | } else { |
1023 | fill_tone_level_array(q, 0); |
1024 | } |
1025 | } |
1026 | |
1027 | /** |
1028 | * Process subpacket 11 |
1029 | * |
1030 | * @param q context |
1031 | * @param node pointer to node with packet |
1032 | */ |
1033 | static void process_subpacket_11(QDM2Context *q, QDM2SubPNode *node) |
1034 | { |
1035 | GetBitContext gb; |
1036 | int length = 0; |
1037 | |
1038 | if (node) { |
1039 | length = node->packet->size * 8; |
1040 | init_get_bits(&gb, node->packet->data, length); |
1041 | } |
1042 | |
1043 | if (length >= 32) { |
1044 | int c = get_bits(&gb, 13); |
1045 | |
1046 | if (c > 3) |
1047 | fill_coding_method_array(q->tone_level_idx, |
1048 | q->tone_level_idx_temp, q->coding_method, |
1049 | q->nb_channels, 8 * c, |
1050 | q->superblocktype_2_3, q->cm_table_select); |
1051 | } |
1052 | |
1053 | synthfilt_build_sb_samples(q, &gb, length, 0, 8); |
1054 | } |
1055 | |
1056 | /** |
1057 | * Process subpacket 12 |
1058 | * |
1059 | * @param q context |
1060 | * @param node pointer to node with packet |
1061 | */ |
1062 | static void process_subpacket_12(QDM2Context *q, QDM2SubPNode *node) |
1063 | { |
1064 | GetBitContext gb; |
1065 | int length = 0; |
1066 | |
1067 | if (node) { |
1068 | length = node->packet->size * 8; |
1069 | init_get_bits(&gb, node->packet->data, length); |
1070 | } |
1071 | |
1072 | synthfilt_build_sb_samples(q, &gb, length, 8, QDM2_SB_USED(q->sub_sampling)); |
1073 | } |
1074 | |
1075 | /** |
1076 | * Process new subpackets for synthesis filter |
1077 | * |
1078 | * @param q context |
1079 | * @param list list with synthesis filter packets (list D) |
1080 | */ |
1081 | static void process_synthesis_subpackets(QDM2Context *q, QDM2SubPNode *list) |
1082 | { |
1083 | QDM2SubPNode *nodes[4]; |
1084 | |
1085 | nodes[0] = qdm2_search_subpacket_type_in_list(list, 9); |
1086 | if (nodes[0]) |
1087 | process_subpacket_9(q, nodes[0]); |
1088 | |
1089 | nodes[1] = qdm2_search_subpacket_type_in_list(list, 10); |
1090 | if (nodes[1]) |
1091 | process_subpacket_10(q, nodes[1]); |
1092 | else |
1093 | process_subpacket_10(q, NULL); |
1094 | |
1095 | nodes[2] = qdm2_search_subpacket_type_in_list(list, 11); |
1096 | if (nodes[0] && nodes[1] && nodes[2]) |
1097 | process_subpacket_11(q, nodes[2]); |
1098 | else |
1099 | process_subpacket_11(q, NULL); |
1100 | |
1101 | nodes[3] = qdm2_search_subpacket_type_in_list(list, 12); |
1102 | if (nodes[0] && nodes[1] && nodes[3]) |
1103 | process_subpacket_12(q, nodes[3]); |
1104 | else |
1105 | process_subpacket_12(q, NULL); |
1106 | } |
1107 | |
1108 | /** |
1109 | * Decode superblock, fill packet lists. |
1110 | * |
1111 | * @param q context |
1112 | */ |
1113 | static void qdm2_decode_super_block(QDM2Context *q) |
1114 | { |
1115 | GetBitContext gb; |
1116 | QDM2SubPacket header, *packet; |
1117 | int i, packet_bytes, sub_packet_size, sub_packets_D; |
1118 | unsigned int next_index = 0; |
1119 | |
1120 | memset(q->tone_level_idx_hi1, 0, sizeof(q->tone_level_idx_hi1)); |
1121 | memset(q->tone_level_idx_mid, 0, sizeof(q->tone_level_idx_mid)); |
1122 | memset(q->tone_level_idx_hi2, 0, sizeof(q->tone_level_idx_hi2)); |
1123 | |
1124 | q->sub_packets_B = 0; |
1125 | sub_packets_D = 0; |
1126 | |
1127 | average_quantized_coeffs(q); // average elements in quantized_coeffs[max_ch][10][8] |
1128 | |
1129 | init_get_bits(&gb, q->compressed_data, q->compressed_size * 8); |
1130 | qdm2_decode_sub_packet_header(&gb, &header); |
1131 | |
1132 | if (header.type < 2 || header.type >= 8) { |
1133 | q->has_errors = 1; |
1134 | av_log(NULL, AV_LOG_ERROR, "bad superblock type\n"); |
1135 | return; |
1136 | } |
1137 | |
1138 | q->superblocktype_2_3 = (header.type == 2 || header.type == 3); |
1139 | packet_bytes = (q->compressed_size - get_bits_count(&gb) / 8); |
1140 | |
1141 | init_get_bits(&gb, header.data, header.size * 8); |
1142 | |
1143 | if (header.type == 2 || header.type == 4 || header.type == 5) { |
1144 | int csum = 257 * get_bits(&gb, 8); |
1145 | csum += 2 * get_bits(&gb, 8); |
1146 | |
1147 | csum = qdm2_packet_checksum(q->compressed_data, q->checksum_size, csum); |
1148 | |
1149 | if (csum != 0) { |
1150 | q->has_errors = 1; |
1151 | av_log(NULL, AV_LOG_ERROR, "bad packet checksum\n"); |
1152 | return; |
1153 | } |
1154 | } |
1155 | |
1156 | q->sub_packet_list_B[0].packet = NULL; |
1157 | q->sub_packet_list_D[0].packet = NULL; |
1158 | |
1159 | for (i = 0; i < 6; i++) |
1160 | if (--q->fft_level_exp[i] < 0) |
1161 | q->fft_level_exp[i] = 0; |
1162 | |
1163 | for (i = 0; packet_bytes > 0; i++) { |
1164 | int j; |
1165 | |
1166 | if (i >= FF_ARRAY_ELEMS(q->sub_packet_list_A)) { |
1167 | SAMPLES_NEEDED_2("too many packet bytes"); |
1168 | return; |
1169 | } |
1170 | |
1171 | q->sub_packet_list_A[i].next = NULL; |
1172 | |
1173 | if (i > 0) { |
1174 | q->sub_packet_list_A[i - 1].next = &q->sub_packet_list_A[i]; |
1175 | |
1176 | /* seek to next block */ |
1177 | init_get_bits(&gb, header.data, header.size * 8); |
1178 | skip_bits(&gb, next_index * 8); |
1179 | |
1180 | if (next_index >= header.size) |
1181 | break; |
1182 | } |
1183 | |
1184 | /* decode subpacket */ |
1185 | packet = &q->sub_packets[i]; |
1186 | qdm2_decode_sub_packet_header(&gb, packet); |
1187 | next_index = packet->size + get_bits_count(&gb) / 8; |
1188 | sub_packet_size = ((packet->size > 0xff) ? 1 : 0) + packet->size + 2; |
1189 | |
1190 | if (packet->type == 0) |
1191 | break; |
1192 | |
1193 | if (sub_packet_size > packet_bytes) { |
1194 | if (packet->type != 10 && packet->type != 11 && packet->type != 12) |
1195 | break; |
1196 | packet->size += packet_bytes - sub_packet_size; |
1197 | } |
1198 | |
1199 | packet_bytes -= sub_packet_size; |
1200 | |
1201 | /* add subpacket to 'all subpackets' list */ |
1202 | q->sub_packet_list_A[i].packet = packet; |
1203 | |
1204 | /* add subpacket to related list */ |
1205 | if (packet->type == 8) { |
1206 | SAMPLES_NEEDED_2("packet type 8"); |
1207 | return; |
1208 | } else if (packet->type >= 9 && packet->type <= 12) { |
1209 | /* packets for MPEG Audio like Synthesis Filter */ |
1210 | QDM2_LIST_ADD(q->sub_packet_list_D, sub_packets_D, packet); |
1211 | } else if (packet->type == 13) { |
1212 | for (j = 0; j < 6; j++) |
1213 | q->fft_level_exp[j] = get_bits(&gb, 6); |
1214 | } else if (packet->type == 14) { |
1215 | for (j = 0; j < 6; j++) |
1216 | q->fft_level_exp[j] = qdm2_get_vlc(&gb, &fft_level_exp_vlc, 0, 2); |
1217 | } else if (packet->type == 15) { |
1218 | SAMPLES_NEEDED_2("packet type 15") |
1219 | return; |
1220 | } else if (packet->type >= 16 && packet->type < 48 && |
1221 | !fft_subpackets[packet->type - 16]) { |
1222 | /* packets for FFT */ |
1223 | QDM2_LIST_ADD(q->sub_packet_list_B, q->sub_packets_B, packet); |
1224 | } |
1225 | } // Packet bytes loop |
1226 | |
1227 | if (q->sub_packet_list_D[0].packet) { |
1228 | process_synthesis_subpackets(q, q->sub_packet_list_D); |
1229 | q->do_synth_filter = 1; |
1230 | } else if (q->do_synth_filter) { |
1231 | process_subpacket_10(q, NULL); |
1232 | process_subpacket_11(q, NULL); |
1233 | process_subpacket_12(q, NULL); |
1234 | } |
1235 | } |
1236 | |
1237 | static void qdm2_fft_init_coefficient(QDM2Context *q, int sub_packet, |
1238 | int offset, int duration, int channel, |
1239 | int exp, int phase) |
1240 | { |
1241 | if (q->fft_coefs_min_index[duration] < 0) |
1242 | q->fft_coefs_min_index[duration] = q->fft_coefs_index; |
1243 | |
1244 | q->fft_coefs[q->fft_coefs_index].sub_packet = |
1245 | ((sub_packet >= 16) ? (sub_packet - 16) : sub_packet); |
1246 | q->fft_coefs[q->fft_coefs_index].channel = channel; |
1247 | q->fft_coefs[q->fft_coefs_index].offset = offset; |
1248 | q->fft_coefs[q->fft_coefs_index].exp = exp; |
1249 | q->fft_coefs[q->fft_coefs_index].phase = phase; |
1250 | q->fft_coefs_index++; |
1251 | } |
1252 | |
1253 | static void qdm2_fft_decode_tones(QDM2Context *q, int duration, |
1254 | GetBitContext *gb, int b) |
1255 | { |
1256 | int channel, stereo, phase, exp; |
1257 | int local_int_4, local_int_8, stereo_phase, local_int_10; |
1258 | int local_int_14, stereo_exp, local_int_20, local_int_28; |
1259 | int n, offset; |
1260 | |
1261 | local_int_4 = 0; |
1262 | local_int_28 = 0; |
1263 | local_int_20 = 2; |
1264 | local_int_8 = (4 - duration); |
1265 | local_int_10 = 1 << (q->group_order - duration - 1); |
1266 | offset = 1; |
1267 | |
1268 | while (get_bits_left(gb)>0) { |
1269 | if (q->superblocktype_2_3) { |
1270 | while ((n = qdm2_get_vlc(gb, &vlc_tab_fft_tone_offset[local_int_8], 1, 2)) < 2) { |
1271 | if (get_bits_left(gb)<0) { |
1272 | if(local_int_4 < q->group_size) |
1273 | av_log(NULL, AV_LOG_ERROR, "overread in qdm2_fft_decode_tones()\n"); |
1274 | return; |
1275 | } |
1276 | offset = 1; |
1277 | if (n == 0) { |
1278 | local_int_4 += local_int_10; |
1279 | local_int_28 += (1 << local_int_8); |
1280 | } else { |
1281 | local_int_4 += 8 * local_int_10; |
1282 | local_int_28 += (8 << local_int_8); |
1283 | } |
1284 | } |
1285 | offset += (n - 2); |
1286 | } else { |
1287 | offset += qdm2_get_vlc(gb, &vlc_tab_fft_tone_offset[local_int_8], 1, 2); |
1288 | while (offset >= (local_int_10 - 1)) { |
1289 | offset += (1 - (local_int_10 - 1)); |
1290 | local_int_4 += local_int_10; |
1291 | local_int_28 += (1 << local_int_8); |
1292 | } |
1293 | } |
1294 | |
1295 | if (local_int_4 >= q->group_size) |
1296 | return; |
1297 | |
1298 | local_int_14 = (offset >> local_int_8); |
1299 | if (local_int_14 >= FF_ARRAY_ELEMS(fft_level_index_table)) |
1300 | return; |
1301 | |
1302 | if (q->nb_channels > 1) { |
1303 | channel = get_bits1(gb); |
1304 | stereo = get_bits1(gb); |
1305 | } else { |
1306 | channel = 0; |
1307 | stereo = 0; |
1308 | } |
1309 | |
1310 | exp = qdm2_get_vlc(gb, (b ? &fft_level_exp_vlc : &fft_level_exp_alt_vlc), 0, 2); |
1311 | exp += q->fft_level_exp[fft_level_index_table[local_int_14]]; |
1312 | exp = (exp < 0) ? 0 : exp; |
1313 | |
1314 | phase = get_bits(gb, 3); |
1315 | stereo_exp = 0; |
1316 | stereo_phase = 0; |
1317 | |
1318 | if (stereo) { |
1319 | stereo_exp = (exp - qdm2_get_vlc(gb, &fft_stereo_exp_vlc, 0, 1)); |
1320 | stereo_phase = (phase - qdm2_get_vlc(gb, &fft_stereo_phase_vlc, 0, 1)); |
1321 | if (stereo_phase < 0) |
1322 | stereo_phase += 8; |
1323 | } |
1324 | |
1325 | if (q->frequency_range > (local_int_14 + 1)) { |
1326 | int sub_packet = (local_int_20 + local_int_28); |
1327 | |
1328 | qdm2_fft_init_coefficient(q, sub_packet, offset, duration, |
1329 | channel, exp, phase); |
1330 | if (stereo) |
1331 | qdm2_fft_init_coefficient(q, sub_packet, offset, duration, |
1332 | 1 - channel, |
1333 | stereo_exp, stereo_phase); |
1334 | } |
1335 | offset++; |
1336 | } |
1337 | } |
1338 | |
1339 | static void qdm2_decode_fft_packets(QDM2Context *q) |
1340 | { |
1341 | int i, j, min, max, value, type, unknown_flag; |
1342 | GetBitContext gb; |
1343 | |
1344 | if (!q->sub_packet_list_B[0].packet) |
1345 | return; |
1346 | |
1347 | /* reset minimum indexes for FFT coefficients */ |
1348 | q->fft_coefs_index = 0; |
1349 | for (i = 0; i < 5; i++) |
1350 | q->fft_coefs_min_index[i] = -1; |
1351 | |
1352 | /* process subpackets ordered by type, largest type first */ |
1353 | for (i = 0, max = 256; i < q->sub_packets_B; i++) { |
1354 | QDM2SubPacket *packet = NULL; |
1355 | |
1356 | /* find subpacket with largest type less than max */ |
1357 | for (j = 0, min = 0; j < q->sub_packets_B; j++) { |
1358 | value = q->sub_packet_list_B[j].packet->type; |
1359 | if (value > min && value < max) { |
1360 | min = value; |
1361 | packet = q->sub_packet_list_B[j].packet; |
1362 | } |
1363 | } |
1364 | |
1365 | max = min; |
1366 | |
1367 | /* check for errors (?) */ |
1368 | if (!packet) |
1369 | return; |
1370 | |
1371 | if (i == 0 && |
1372 | (packet->type < 16 || packet->type >= 48 || |
1373 | fft_subpackets[packet->type - 16])) |
1374 | return; |
1375 | |
1376 | /* decode FFT tones */ |
1377 | init_get_bits(&gb, packet->data, packet->size * 8); |
1378 | |
1379 | if (packet->type >= 32 && packet->type < 48 && !fft_subpackets[packet->type - 16]) |
1380 | unknown_flag = 1; |
1381 | else |
1382 | unknown_flag = 0; |
1383 | |
1384 | type = packet->type; |
1385 | |
1386 | if ((type >= 17 && type < 24) || (type >= 33 && type < 40)) { |
1387 | int duration = q->sub_sampling + 5 - (type & 15); |
1388 | |
1389 | if (duration >= 0 && duration < 4) |
1390 | qdm2_fft_decode_tones(q, duration, &gb, unknown_flag); |
1391 | } else if (type == 31) { |
1392 | for (j = 0; j < 4; j++) |
1393 | qdm2_fft_decode_tones(q, j, &gb, unknown_flag); |
1394 | } else if (type == 46) { |
1395 | for (j = 0; j < 6; j++) |
1396 | q->fft_level_exp[j] = get_bits(&gb, 6); |
1397 | for (j = 0; j < 4; j++) |
1398 | qdm2_fft_decode_tones(q, j, &gb, unknown_flag); |
1399 | } |
1400 | } // Loop on B packets |
1401 | |
1402 | /* calculate maximum indexes for FFT coefficients */ |
1403 | for (i = 0, j = -1; i < 5; i++) |
1404 | if (q->fft_coefs_min_index[i] >= 0) { |
1405 | if (j >= 0) |
1406 | q->fft_coefs_max_index[j] = q->fft_coefs_min_index[i]; |
1407 | j = i; |
1408 | } |
1409 | if (j >= 0) |
1410 | q->fft_coefs_max_index[j] = q->fft_coefs_index; |
1411 | } |
1412 | |
1413 | static void qdm2_fft_generate_tone(QDM2Context *q, FFTTone *tone) |
1414 | { |
1415 | float level, f[6]; |
1416 | int i; |
1417 | QDM2Complex c; |
1418 | const double iscale = 2.0 * M_PI / 512.0; |
1419 | |
1420 | tone->phase += tone->phase_shift; |
1421 | |
1422 | /* calculate current level (maximum amplitude) of tone */ |
1423 | level = fft_tone_envelope_table[tone->duration][tone->time_index] * tone->level; |
1424 | c.im = level * sin(tone->phase * iscale); |
1425 | c.re = level * cos(tone->phase * iscale); |
1426 | |
1427 | /* generate FFT coefficients for tone */ |
1428 | if (tone->duration >= 3 || tone->cutoff >= 3) { |
1429 | tone->complex[0].im += c.im; |
1430 | tone->complex[0].re += c.re; |
1431 | tone->complex[1].im -= c.im; |
1432 | tone->complex[1].re -= c.re; |
1433 | } else { |
1434 | f[1] = -tone->table[4]; |
1435 | f[0] = tone->table[3] - tone->table[0]; |
1436 | f[2] = 1.0 - tone->table[2] - tone->table[3]; |
1437 | f[3] = tone->table[1] + tone->table[4] - 1.0; |
1438 | f[4] = tone->table[0] - tone->table[1]; |
1439 | f[5] = tone->table[2]; |
1440 | for (i = 0; i < 2; i++) { |
1441 | tone->complex[fft_cutoff_index_table[tone->cutoff][i]].re += |
1442 | c.re * f[i]; |
1443 | tone->complex[fft_cutoff_index_table[tone->cutoff][i]].im += |
1444 | c.im * ((tone->cutoff <= i) ? -f[i] : f[i]); |
1445 | } |
1446 | for (i = 0; i < 4; i++) { |
1447 | tone->complex[i].re += c.re * f[i + 2]; |
1448 | tone->complex[i].im += c.im * f[i + 2]; |
1449 | } |
1450 | } |
1451 | |
1452 | /* copy the tone if it has not yet died out */ |
1453 | if (++tone->time_index < ((1 << (5 - tone->duration)) - 1)) { |
1454 | memcpy(&q->fft_tones[q->fft_tone_end], tone, sizeof(FFTTone)); |
1455 | q->fft_tone_end = (q->fft_tone_end + 1) % 1000; |
1456 | } |
1457 | } |
1458 | |
1459 | static void qdm2_fft_tone_synthesizer(QDM2Context *q, int sub_packet) |
1460 | { |
1461 | int i, j, ch; |
1462 | const double iscale = 0.25 * M_PI; |
1463 | |
1464 | for (ch = 0; ch < q->channels; ch++) { |
1465 | memset(q->fft.complex[ch], 0, q->fft_size * sizeof(QDM2Complex)); |
1466 | } |
1467 | |
1468 | |
1469 | /* apply FFT tones with duration 4 (1 FFT period) */ |
1470 | if (q->fft_coefs_min_index[4] >= 0) |
1471 | for (i = q->fft_coefs_min_index[4]; i < q->fft_coefs_max_index[4]; i++) { |
1472 | float level; |
1473 | QDM2Complex c; |
1474 | |
1475 | if (q->fft_coefs[i].sub_packet != sub_packet) |
1476 | break; |
1477 | |
1478 | ch = (q->channels == 1) ? 0 : q->fft_coefs[i].channel; |
1479 | 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]; |
1480 | |
1481 | c.re = level * cos(q->fft_coefs[i].phase * iscale); |
1482 | c.im = level * sin(q->fft_coefs[i].phase * iscale); |
1483 | q->fft.complex[ch][q->fft_coefs[i].offset + 0].re += c.re; |
1484 | q->fft.complex[ch][q->fft_coefs[i].offset + 0].im += c.im; |
1485 | q->fft.complex[ch][q->fft_coefs[i].offset + 1].re -= c.re; |
1486 | q->fft.complex[ch][q->fft_coefs[i].offset + 1].im -= c.im; |
1487 | } |
1488 | |
1489 | /* generate existing FFT tones */ |
1490 | for (i = q->fft_tone_end; i != q->fft_tone_start; ) { |
1491 | qdm2_fft_generate_tone(q, &q->fft_tones[q->fft_tone_start]); |
1492 | q->fft_tone_start = (q->fft_tone_start + 1) % 1000; |
1493 | } |
1494 | |
1495 | /* create and generate new FFT tones with duration 0 (long) to 3 (short) */ |
1496 | for (i = 0; i < 4; i++) |
1497 | if (q->fft_coefs_min_index[i] >= 0) { |
1498 | for (j = q->fft_coefs_min_index[i]; j < q->fft_coefs_max_index[i]; j++) { |
1499 | int offset, four_i; |
1500 | FFTTone tone; |
1501 | |
1502 | if (q->fft_coefs[j].sub_packet != sub_packet) |
1503 | break; |
1504 | |
1505 | four_i = (4 - i); |
1506 | offset = q->fft_coefs[j].offset >> four_i; |
1507 | ch = (q->channels == 1) ? 0 : q->fft_coefs[j].channel; |
1508 | |
1509 | if (offset < q->frequency_range) { |
1510 | if (offset < 2) |
1511 | tone.cutoff = offset; |
1512 | else |
1513 | tone.cutoff = (offset >= 60) ? 3 : 2; |
1514 | |
1515 | 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]; |
1516 | tone.complex = &q->fft.complex[ch][offset]; |
1517 | tone.table = fft_tone_sample_table[i][q->fft_coefs[j].offset - (offset << four_i)]; |
1518 | tone.phase = 64 * q->fft_coefs[j].phase - (offset << 8) - 128; |
1519 | tone.phase_shift = (2 * q->fft_coefs[j].offset + 1) << (7 - four_i); |
1520 | tone.duration = i; |
1521 | tone.time_index = 0; |
1522 | |
1523 | qdm2_fft_generate_tone(q, &tone); |
1524 | } |
1525 | } |
1526 | q->fft_coefs_min_index[i] = j; |
1527 | } |
1528 | } |
1529 | |
1530 | static void qdm2_calculate_fft(QDM2Context *q, int channel, int sub_packet) |
1531 | { |
1532 | const float gain = (q->channels == 1 && q->nb_channels == 2) ? 0.5f : 1.0f; |
1533 | float *out = q->output_buffer + channel; |
1534 | int i; |
1535 | q->fft.complex[channel][0].re *= 2.0f; |
1536 | q->fft.complex[channel][0].im = 0.0f; |
1537 | q->rdft_ctx.rdft_calc(&q->rdft_ctx, (FFTSample *)q->fft.complex[channel]); |
1538 | /* add samples to output buffer */ |
1539 | for (i = 0; i < FFALIGN(q->fft_size, 8); i++) { |
1540 | out[0] += q->fft.complex[channel][i].re * gain; |
1541 | out[q->channels] += q->fft.complex[channel][i].im * gain; |
1542 | out += 2 * q->channels; |
1543 | } |
1544 | } |
1545 | |
1546 | /** |
1547 | * @param q context |
1548 | * @param index subpacket number |
1549 | */ |
1550 | static void qdm2_synthesis_filter(QDM2Context *q, int index) |
1551 | { |
1552 | int i, k, ch, sb_used, sub_sampling, dither_state = 0; |
1553 | |
1554 | /* copy sb_samples */ |
1555 | sb_used = QDM2_SB_USED(q->sub_sampling); |
1556 | |
1557 | for (ch = 0; ch < q->channels; ch++) |
1558 | for (i = 0; i < 8; i++) |
1559 | for (k = sb_used; k < SBLIMIT; k++) |
1560 | q->sb_samples[ch][(8 * index) + i][k] = 0; |
1561 | |
1562 | for (ch = 0; ch < q->nb_channels; ch++) { |
1563 | float *samples_ptr = q->samples + ch; |
1564 | |
1565 | for (i = 0; i < 8; i++) { |
1566 | ff_mpa_synth_filter_float(&q->mpadsp, |
1567 | q->synth_buf[ch], &(q->synth_buf_offset[ch]), |
1568 | ff_mpa_synth_window_float, &dither_state, |
1569 | samples_ptr, q->nb_channels, |
1570 | q->sb_samples[ch][(8 * index) + i]); |
1571 | samples_ptr += 32 * q->nb_channels; |
1572 | } |
1573 | } |
1574 | |
1575 | /* add samples to output buffer */ |
1576 | sub_sampling = (4 >> q->sub_sampling); |
1577 | |
1578 | for (ch = 0; ch < q->channels; ch++) |
1579 | for (i = 0; i < q->frame_size; i++) |
1580 | q->output_buffer[q->channels * i + ch] += (1 << 23) * q->samples[q->nb_channels * sub_sampling * i + ch]; |
1581 | } |
1582 | |
1583 | /** |
1584 | * Init static data (does not depend on specific file) |
1585 | * |
1586 | * @param q context |
1587 | */ |
1588 | static av_cold void qdm2_init_static_data(void) { |
1589 | static int done; |
1590 | |
1591 | if(done) |
1592 | return; |
1593 | |
1594 | qdm2_init_vlc(); |
1595 | ff_mpa_synth_init_float(ff_mpa_synth_window_float); |
1596 | softclip_table_init(); |
1597 | rnd_table_init(); |
1598 | init_noise_samples(); |
1599 | |
1600 | done = 1; |
1601 | } |
1602 | |
1603 | /** |
1604 | * Init parameters from codec extradata |
1605 | */ |
1606 | static av_cold int qdm2_decode_init(AVCodecContext *avctx) |
1607 | { |
1608 | QDM2Context *s = avctx->priv_data; |
1609 | int tmp_val, tmp, size; |
1610 | GetByteContext gb; |
1611 | |
1612 | qdm2_init_static_data(); |
1613 | |
1614 | /* extradata parsing |
1615 | |
1616 | Structure: |
1617 | wave { |
1618 | frma (QDM2) |
1619 | QDCA |
1620 | QDCP |
1621 | } |
1622 | |
1623 | 32 size (including this field) |
1624 | 32 tag (=frma) |
1625 | 32 type (=QDM2 or QDMC) |
1626 | |
1627 | 32 size (including this field, in bytes) |
1628 | 32 tag (=QDCA) // maybe mandatory parameters |
1629 | 32 unknown (=1) |
1630 | 32 channels (=2) |
1631 | 32 samplerate (=44100) |
1632 | 32 bitrate (=96000) |
1633 | 32 block size (=4096) |
1634 | 32 frame size (=256) (for one channel) |
1635 | 32 packet size (=1300) |
1636 | |
1637 | 32 size (including this field, in bytes) |
1638 | 32 tag (=QDCP) // maybe some tuneable parameters |
1639 | 32 float1 (=1.0) |
1640 | 32 zero ? |
1641 | 32 float2 (=1.0) |
1642 | 32 float3 (=1.0) |
1643 | 32 unknown (27) |
1644 | 32 unknown (8) |
1645 | 32 zero ? |
1646 | */ |
1647 | |
1648 | if (!avctx->extradata || (avctx->extradata_size < 48)) { |
1649 | av_log(avctx, AV_LOG_ERROR, "extradata missing or truncated\n"); |
1650 | return AVERROR_INVALIDDATA; |
1651 | } |
1652 | |
1653 | bytestream2_init(&gb, avctx->extradata, avctx->extradata_size); |
1654 | |
1655 | while (bytestream2_get_bytes_left(&gb) > 8) { |
1656 | if (bytestream2_peek_be64(&gb) == (((uint64_t)MKBETAG('f','r','m','a') << 32) | |
1657 | (uint64_t)MKBETAG('Q','D','M','2'))) |
1658 | break; |
1659 | bytestream2_skip(&gb, 1); |
1660 | } |
1661 | |
1662 | if (bytestream2_get_bytes_left(&gb) < 12) { |
1663 | av_log(avctx, AV_LOG_ERROR, "not enough extradata (%i)\n", |
1664 | bytestream2_get_bytes_left(&gb)); |
1665 | return AVERROR_INVALIDDATA; |
1666 | } |
1667 | |
1668 | bytestream2_skip(&gb, 8); |
1669 | size = bytestream2_get_be32(&gb); |
1670 | |
1671 | if (size > bytestream2_get_bytes_left(&gb)) { |
1672 | av_log(avctx, AV_LOG_ERROR, "extradata size too small, %i < %i\n", |
1673 | bytestream2_get_bytes_left(&gb), size); |
1674 | return AVERROR_INVALIDDATA; |
1675 | } |
1676 | |
1677 | av_log(avctx, AV_LOG_DEBUG, "size: %d\n", size); |
1678 | if (bytestream2_get_be32(&gb) != MKBETAG('Q','D','C','A')) { |
1679 | av_log(avctx, AV_LOG_ERROR, "invalid extradata, expecting QDCA\n"); |
1680 | return AVERROR_INVALIDDATA; |
1681 | } |
1682 | |
1683 | bytestream2_skip(&gb, 4); |
1684 | |
1685 | avctx->channels = s->nb_channels = s->channels = bytestream2_get_be32(&gb); |
1686 | if (s->channels <= 0 || s->channels > MPA_MAX_CHANNELS) { |
1687 | av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n"); |
1688 | return AVERROR_INVALIDDATA; |
1689 | } |
1690 | avctx->channel_layout = avctx->channels == 2 ? AV_CH_LAYOUT_STEREO : |
1691 | AV_CH_LAYOUT_MONO; |
1692 | |
1693 | avctx->sample_rate = bytestream2_get_be32(&gb); |
1694 | avctx->bit_rate = bytestream2_get_be32(&gb); |
1695 | s->group_size = bytestream2_get_be32(&gb); |
1696 | s->fft_size = bytestream2_get_be32(&gb); |
1697 | s->checksum_size = bytestream2_get_be32(&gb); |
1698 | if (s->checksum_size >= 1U << 28) { |
1699 | av_log(avctx, AV_LOG_ERROR, "data block size too large (%u)\n", s->checksum_size); |
1700 | return AVERROR_INVALIDDATA; |
1701 | } |
1702 | |
1703 | s->fft_order = av_log2(s->fft_size) + 1; |
1704 | |
1705 | // something like max decodable tones |
1706 | s->group_order = av_log2(s->group_size) + 1; |
1707 | s->frame_size = s->group_size / 16; // 16 iterations per super block |
1708 | |
1709 | if (s->frame_size > QDM2_MAX_FRAME_SIZE) |
1710 | return AVERROR_INVALIDDATA; |
1711 | |
1712 | s->sub_sampling = s->fft_order - 7; |
1713 | s->frequency_range = 255 / (1 << (2 - s->sub_sampling)); |
1714 | |
1715 | switch ((s->sub_sampling * 2 + s->channels - 1)) { |
1716 | case 0: tmp = 40; break; |
1717 | case 1: tmp = 48; break; |
1718 | case 2: tmp = 56; break; |
1719 | case 3: tmp = 72; break; |
1720 | case 4: tmp = 80; break; |
1721 | case 5: tmp = 100;break; |
1722 | default: tmp=s->sub_sampling; break; |
1723 | } |
1724 | tmp_val = 0; |
1725 | if ((tmp * 1000) < avctx->bit_rate) tmp_val = 1; |
1726 | if ((tmp * 1440) < avctx->bit_rate) tmp_val = 2; |
1727 | if ((tmp * 1760) < avctx->bit_rate) tmp_val = 3; |
1728 | if ((tmp * 2240) < avctx->bit_rate) tmp_val = 4; |
1729 | s->cm_table_select = tmp_val; |
1730 | |
1731 | if (avctx->bit_rate <= 8000) |
1732 | s->coeff_per_sb_select = 0; |
1733 | else if (avctx->bit_rate < 16000) |
1734 | s->coeff_per_sb_select = 1; |
1735 | else |
1736 | s->coeff_per_sb_select = 2; |
1737 | |
1738 | // Fail on unknown fft order |
1739 | if ((s->fft_order < 7) || (s->fft_order > 9)) { |
1740 | avpriv_request_sample(avctx, "Unknown FFT order %d", s->fft_order); |
1741 | return AVERROR_PATCHWELCOME; |
1742 | } |
1743 | if (s->fft_size != (1 << (s->fft_order - 1))) { |
1744 | av_log(avctx, AV_LOG_ERROR, "FFT size %d not power of 2.\n", s->fft_size); |
1745 | return AVERROR_INVALIDDATA; |
1746 | } |
1747 | |
1748 | ff_rdft_init(&s->rdft_ctx, s->fft_order, IDFT_C2R); |
1749 | ff_mpadsp_init(&s->mpadsp); |
1750 | |
1751 | avctx->sample_fmt = AV_SAMPLE_FMT_S16; |
1752 | |
1753 | return 0; |
1754 | } |
1755 | |
1756 | static av_cold int qdm2_decode_close(AVCodecContext *avctx) |
1757 | { |
1758 | QDM2Context *s = avctx->priv_data; |
1759 | |
1760 | ff_rdft_end(&s->rdft_ctx); |
1761 | |
1762 | return 0; |
1763 | } |
1764 | |
1765 | static int qdm2_decode(QDM2Context *q, const uint8_t *in, int16_t *out) |
1766 | { |
1767 | int ch, i; |
1768 | const int frame_size = (q->frame_size * q->channels); |
1769 | |
1770 | if((unsigned)frame_size > FF_ARRAY_ELEMS(q->output_buffer)/2) |
1771 | return -1; |
1772 | |
1773 | /* select input buffer */ |
1774 | q->compressed_data = in; |
1775 | q->compressed_size = q->checksum_size; |
1776 | |
1777 | /* copy old block, clear new block of output samples */ |
1778 | memmove(q->output_buffer, &q->output_buffer[frame_size], frame_size * sizeof(float)); |
1779 | memset(&q->output_buffer[frame_size], 0, frame_size * sizeof(float)); |
1780 | |
1781 | /* decode block of QDM2 compressed data */ |
1782 | if (q->sub_packet == 0) { |
1783 | q->has_errors = 0; // zero it for a new super block |
1784 | av_log(NULL,AV_LOG_DEBUG,"Superblock follows\n"); |
1785 | qdm2_decode_super_block(q); |
1786 | } |
1787 | |
1788 | /* parse subpackets */ |
1789 | if (!q->has_errors) { |
1790 | if (q->sub_packet == 2) |
1791 | qdm2_decode_fft_packets(q); |
1792 | |
1793 | qdm2_fft_tone_synthesizer(q, q->sub_packet); |
1794 | } |
1795 | |
1796 | /* sound synthesis stage 1 (FFT) */ |
1797 | for (ch = 0; ch < q->channels; ch++) { |
1798 | qdm2_calculate_fft(q, ch, q->sub_packet); |
1799 | |
1800 | if (!q->has_errors && q->sub_packet_list_C[0].packet) { |
1801 | SAMPLES_NEEDED_2("has errors, and C list is not empty") |
1802 | return -1; |
1803 | } |
1804 | } |
1805 | |
1806 | /* sound synthesis stage 2 (MPEG audio like synthesis filter) */ |
1807 | if (!q->has_errors && q->do_synth_filter) |
1808 | qdm2_synthesis_filter(q, q->sub_packet); |
1809 | |
1810 | q->sub_packet = (q->sub_packet + 1) % 16; |
1811 | |
1812 | /* clip and convert output float[] to 16-bit signed samples */ |
1813 | for (i = 0; i < frame_size; i++) { |
1814 | int value = (int)q->output_buffer[i]; |
1815 | |
1816 | if (value > SOFTCLIP_THRESHOLD) |
1817 | value = (value > HARDCLIP_THRESHOLD) ? 32767 : softclip_table[ value - SOFTCLIP_THRESHOLD]; |
1818 | else if (value < -SOFTCLIP_THRESHOLD) |
1819 | value = (value < -HARDCLIP_THRESHOLD) ? -32767 : -softclip_table[-value - SOFTCLIP_THRESHOLD]; |
1820 | |
1821 | out[i] = value; |
1822 | } |
1823 | |
1824 | return 0; |
1825 | } |
1826 | |
1827 | static int qdm2_decode_frame(AVCodecContext *avctx, void *data, |
1828 | int *got_frame_ptr, AVPacket *avpkt) |
1829 | { |
1830 | AVFrame *frame = data; |
1831 | const uint8_t *buf = avpkt->data; |
1832 | int buf_size = avpkt->size; |
1833 | QDM2Context *s = avctx->priv_data; |
1834 | int16_t *out; |
1835 | int i, ret; |
1836 | |
1837 | if(!buf) |
1838 | return 0; |
1839 | if(buf_size < s->checksum_size) |
1840 | return -1; |
1841 | |
1842 | /* get output buffer */ |
1843 | frame->nb_samples = 16 * s->frame_size; |
1844 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
1845 | return ret; |
1846 | out = (int16_t *)frame->data[0]; |
1847 | |
1848 | for (i = 0; i < 16; i++) { |
1849 | if ((ret = qdm2_decode(s, buf, out)) < 0) |
1850 | return ret; |
1851 | out += s->channels * s->frame_size; |
1852 | } |
1853 | |
1854 | *got_frame_ptr = 1; |
1855 | |
1856 | return s->checksum_size; |
1857 | } |
1858 | |
1859 | AVCodec ff_qdm2_decoder = { |
1860 | .name = "qdm2", |
1861 | .long_name = NULL_IF_CONFIG_SMALL("QDesign Music Codec 2"), |
1862 | .type = AVMEDIA_TYPE_AUDIO, |
1863 | .id = AV_CODEC_ID_QDM2, |
1864 | .priv_data_size = sizeof(QDM2Context), |
1865 | .init = qdm2_decode_init, |
1866 | .close = qdm2_decode_close, |
1867 | .decode = qdm2_decode_frame, |
1868 | .capabilities = AV_CODEC_CAP_DR1, |
1869 | }; |
1870 |