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 FFmpeg.
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.
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.
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
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 #include "libavutil/channel_layout.h"
39 #include "libavutil/thread.h"
41 #define BITSTREAM_READER_LE
44 #include "bytestream.h"
46 #include "mpegaudio.h"
47 #include "mpegaudiodsp.h"
50 #include "qdm2_tablegen.h"
52 #define QDM2_LIST_ADD(list, size, packet) \
55 list[size - 1].next = &list[size]; \
57 list[size].packet = packet; \
58 list[size].next = NULL; \
62 // Result is 8, 16 or 30
63 #define QDM2_SB_USED(sub_sampling) (((sub_sampling) >= 2) ? 30 : 8 << (sub_sampling))
65 #define FIX_NOISE_IDX(noise_idx) \
66 if ((noise_idx) >= 3840) \
67 (noise_idx) -= 3840; \
69 #define SB_DITHERING_NOISE(sb,noise_idx) (noise_table[(noise_idx)++] * sb_noise_attenuation[(sb)])
71 #define SAMPLES_NEEDED \
72 av_log (NULL,AV_LOG_INFO,"This file triggers some untested code. Please contact the developers.\n");
74 #define SAMPLES_NEEDED_2(why) \
75 av_log (NULL,AV_LOG_INFO,"This file triggers some missing code. Please contact the developers.\nPosition: %s\n",why);
77 #define QDM2_MAX_FRAME_SIZE 512
79 typedef int8_t sb_int8_array[2][30][64];
84 typedef struct QDM2SubPacket {
85 int type; ///< subpacket type
86 unsigned int size; ///< subpacket size
87 const uint8_t *data; ///< pointer to subpacket data (points to input data buffer, it's not a private copy)
91 * A node in the subpacket list
93 typedef struct QDM2SubPNode {
94 QDM2SubPacket *packet; ///< packet
95 struct QDM2SubPNode *next; ///< pointer to next packet in the list, NULL if leaf node
98 typedef struct QDM2Complex {
103 typedef struct FFTTone {
105 QDM2Complex *complex;
114 typedef struct FFTCoefficient {
122 typedef struct QDM2FFT {
123 DECLARE_ALIGNED(32, QDM2Complex, complex)[MPA_MAX_CHANNELS][256];
127 * QDM2 decoder context
129 typedef struct QDM2Context {
130 /// Parameters from codec header, do not change during playback
131 int nb_channels; ///< number of channels
132 int channels; ///< number of channels
133 int group_size; ///< size of frame group (16 frames per group)
134 int fft_size; ///< size of FFT, in complex numbers
135 int checksum_size; ///< size of data block, used also for checksum
137 /// Parameters built from header parameters, do not change during playback
138 int group_order; ///< order of frame group
139 int fft_order; ///< order of FFT (actually fftorder+1)
140 int frame_size; ///< size of data frame
142 int sub_sampling; ///< subsampling: 0=25%, 1=50%, 2=100% */
143 int coeff_per_sb_select; ///< selector for "num. of coeffs. per subband" tables. Can be 0, 1, 2
144 int cm_table_select; ///< selector for "coding method" tables. Can be 0, 1 (from init: 0-4)
146 /// Packets and packet lists
147 QDM2SubPacket sub_packets[16]; ///< the packets themselves
148 QDM2SubPNode sub_packet_list_A[16]; ///< list of all packets
149 QDM2SubPNode sub_packet_list_B[16]; ///< FFT packets B are on list
150 int sub_packets_B; ///< number of packets on 'B' list
151 QDM2SubPNode sub_packet_list_C[16]; ///< packets with errors?
152 QDM2SubPNode sub_packet_list_D[16]; ///< DCT packets
155 FFTTone fft_tones[1000];
158 FFTCoefficient fft_coefs[1000];
160 int fft_coefs_min_index[5];
161 int fft_coefs_max_index[5];
162 int fft_level_exp[6];
163 RDFTContext rdft_ctx;
167 const uint8_t *compressed_data;
169 float output_buffer[QDM2_MAX_FRAME_SIZE * MPA_MAX_CHANNELS * 2];
172 MPADSPContext mpadsp;
173 DECLARE_ALIGNED(32, float, synth_buf)[MPA_MAX_CHANNELS][512*2];
174 int synth_buf_offset[MPA_MAX_CHANNELS];
175 DECLARE_ALIGNED(32, float, sb_samples)[MPA_MAX_CHANNELS][128][SBLIMIT];
176 DECLARE_ALIGNED(32, float, samples)[MPA_MAX_CHANNELS * MPA_FRAME_SIZE];
178 /// Mixed temporary data used in decoding
179 float tone_level[MPA_MAX_CHANNELS][30][64];
180 int8_t coding_method[MPA_MAX_CHANNELS][30][64];
181 int8_t quantized_coeffs[MPA_MAX_CHANNELS][10][8];
182 int8_t tone_level_idx_base[MPA_MAX_CHANNELS][30][8];
183 int8_t tone_level_idx_hi1[MPA_MAX_CHANNELS][3][8][8];
184 int8_t tone_level_idx_mid[MPA_MAX_CHANNELS][26][8];
185 int8_t tone_level_idx_hi2[MPA_MAX_CHANNELS][26];
186 int8_t tone_level_idx[MPA_MAX_CHANNELS][30][64];
187 int8_t tone_level_idx_temp[MPA_MAX_CHANNELS][30][64];
190 int has_errors; ///< packet has errors
191 int superblocktype_2_3; ///< select fft tables and some algorithm based on superblock type
192 int do_synth_filter; ///< used to perform or skip synthesis filter
195 int noise_idx; ///< index for dithering noise table
198 static const int switchtable[23] = {
199 0, 5, 1, 5, 5, 5, 5, 5, 2, 5, 5, 5, 5, 5, 5, 5, 3, 5, 5, 5, 5, 5, 4
202 static int qdm2_get_vlc(GetBitContext *gb, const VLC *vlc, int flag, int depth)
206 value = get_vlc2(gb, vlc->table, vlc->bits, depth);
208 /* stage-2, 3 bits exponent escape sequence */
210 value = get_bits(gb, get_bits(gb, 3) + 1);
212 /* stage-3, optional */
217 av_log(NULL, AV_LOG_ERROR, "value %d in qdm2_get_vlc too large\n", value);
221 tmp= vlc_stage3_values[value];
223 if ((value & ~3) > 0)
224 tmp += get_bits(gb, (value >> 2));
231 static int qdm2_get_se_vlc(const VLC *vlc, GetBitContext *gb, int depth)
233 int value = qdm2_get_vlc(gb, vlc, 0, depth);
235 return (value & 1) ? ((value + 1) >> 1) : -(value >> 1);
241 * @param data pointer to data to be checksummed
242 * @param length data length
243 * @param value checksum value
245 * @return 0 if checksum is OK
247 static uint16_t qdm2_packet_checksum(const uint8_t *data, int length, int value)
251 for (i = 0; i < length; i++)
254 return (uint16_t)(value & 0xffff);
258 * Fill a QDM2SubPacket structure with packet type, size, and data pointer.
260 * @param gb bitreader context
261 * @param sub_packet packet under analysis
263 static void qdm2_decode_sub_packet_header(GetBitContext *gb,
264 QDM2SubPacket *sub_packet)
266 sub_packet->type = get_bits(gb, 8);
268 if (sub_packet->type == 0) {
269 sub_packet->size = 0;
270 sub_packet->data = NULL;
272 sub_packet->size = get_bits(gb, 8);
274 if (sub_packet->type & 0x80) {
275 sub_packet->size <<= 8;
276 sub_packet->size |= get_bits(gb, 8);
277 sub_packet->type &= 0x7f;
280 if (sub_packet->type == 0x7f)
281 sub_packet->type |= (get_bits(gb, 8) << 8);
283 // FIXME: this depends on bitreader-internal data
284 sub_packet->data = &gb->buffer[get_bits_count(gb) / 8];
287 av_log(NULL, AV_LOG_DEBUG, "Subpacket: type=%d size=%d start_offs=%x\n",
288 sub_packet->type, sub_packet->size, get_bits_count(gb) / 8);
292 * Return node pointer to first packet of requested type in list.
294 * @param list list of subpackets to be scanned
295 * @param type type of searched subpacket
296 * @return node pointer for subpacket if found, else NULL
298 static QDM2SubPNode *qdm2_search_subpacket_type_in_list(QDM2SubPNode *list,
301 while (list && list->packet) {
302 if (list->packet->type == type)
310 * Replace 8 elements with their average value.
311 * Called by qdm2_decode_superblock before starting subblock decoding.
315 static void average_quantized_coeffs(QDM2Context *q)
317 int i, j, n, ch, sum;
319 n = coeff_per_sb_for_avg[q->coeff_per_sb_select][QDM2_SB_USED(q->sub_sampling) - 1] + 1;
321 for (ch = 0; ch < q->nb_channels; ch++)
322 for (i = 0; i < n; i++) {
325 for (j = 0; j < 8; j++)
326 sum += q->quantized_coeffs[ch][i][j];
332 for (j = 0; j < 8; j++)
333 q->quantized_coeffs[ch][i][j] = sum;
338 * Build subband samples with noise weighted by q->tone_level.
339 * Called by synthfilt_build_sb_samples.
342 * @param sb subband index
344 static void build_sb_samples_from_noise(QDM2Context *q, int sb)
348 FIX_NOISE_IDX(q->noise_idx);
353 for (ch = 0; ch < q->nb_channels; ch++) {
354 for (j = 0; j < 64; j++) {
355 q->sb_samples[ch][j * 2][sb] =
356 SB_DITHERING_NOISE(sb, q->noise_idx) * q->tone_level[ch][sb][j];
357 q->sb_samples[ch][j * 2 + 1][sb] =
358 SB_DITHERING_NOISE(sb, q->noise_idx) * q->tone_level[ch][sb][j];
364 * Called while processing data from subpackets 11 and 12.
365 * Used after making changes to coding_method array.
367 * @param sb subband index
368 * @param channels number of channels
369 * @param coding_method q->coding_method[0][0][0]
371 static int fix_coding_method_array(int sb, int channels,
372 sb_int8_array coding_method)
378 for (ch = 0; ch < channels; ch++) {
379 for (j = 0; j < 64; ) {
380 if (coding_method[ch][sb][j] < 8)
382 if ((coding_method[ch][sb][j] - 8) > 22) {
386 switch (switchtable[coding_method[ch][sb][j] - 8]) {
410 for (k = 0; k < run; k++) {
412 int sbjk = sb + (j + k) / 64;
417 if (coding_method[ch][sbjk][(j + k) % 64] > coding_method[ch][sb][j]) {
420 //not debugged, almost never used
421 memset(&coding_method[ch][sb][j + k], case_val,
423 memset(&coding_method[ch][sb][j + k], case_val,
436 * Related to synthesis filter
437 * Called by process_subpacket_10
440 * @param flag 1 if called after getting data from subpacket 10, 0 if no subpacket 10
442 static void fill_tone_level_array(QDM2Context *q, int flag)
444 int i, sb, ch, sb_used;
447 for (ch = 0; ch < q->nb_channels; ch++)
448 for (sb = 0; sb < 30; sb++)
449 for (i = 0; i < 8; i++) {
450 if ((tab=coeff_per_sb_for_dequant[q->coeff_per_sb_select][sb]) < (last_coeff[q->coeff_per_sb_select] - 1))
451 tmp = q->quantized_coeffs[ch][tab + 1][i] * dequant_table[q->coeff_per_sb_select][tab + 1][sb]+
452 q->quantized_coeffs[ch][tab][i] * dequant_table[q->coeff_per_sb_select][tab][sb];
454 tmp = q->quantized_coeffs[ch][tab][i] * dequant_table[q->coeff_per_sb_select][tab][sb];
457 q->tone_level_idx_base[ch][sb][i] = (tmp / 256) & 0xff;
460 sb_used = QDM2_SB_USED(q->sub_sampling);
462 if ((q->superblocktype_2_3 != 0) && !flag) {
463 for (sb = 0; sb < sb_used; sb++)
464 for (ch = 0; ch < q->nb_channels; ch++)
465 for (i = 0; i < 64; i++) {
466 q->tone_level_idx[ch][sb][i] = q->tone_level_idx_base[ch][sb][i / 8];
467 if (q->tone_level_idx[ch][sb][i] < 0)
468 q->tone_level[ch][sb][i] = 0;
470 q->tone_level[ch][sb][i] = fft_tone_level_table[0][q->tone_level_idx[ch][sb][i] & 0x3f];
473 tab = q->superblocktype_2_3 ? 0 : 1;
474 for (sb = 0; sb < sb_used; sb++) {
475 if ((sb >= 4) && (sb <= 23)) {
476 for (ch = 0; ch < q->nb_channels; ch++)
477 for (i = 0; i < 64; i++) {
478 tmp = q->tone_level_idx_base[ch][sb][i / 8] -
479 q->tone_level_idx_hi1[ch][sb / 8][i / 8][i % 8] -
480 q->tone_level_idx_mid[ch][sb - 4][i / 8] -
481 q->tone_level_idx_hi2[ch][sb - 4];
482 q->tone_level_idx[ch][sb][i] = tmp & 0xff;
483 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
484 q->tone_level[ch][sb][i] = 0;
486 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
490 for (ch = 0; ch < q->nb_channels; ch++)
491 for (i = 0; i < 64; i++) {
492 tmp = q->tone_level_idx_base[ch][sb][i / 8] -
493 q->tone_level_idx_hi1[ch][2][i / 8][i % 8] -
494 q->tone_level_idx_hi2[ch][sb - 4];
495 q->tone_level_idx[ch][sb][i] = tmp & 0xff;
496 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
497 q->tone_level[ch][sb][i] = 0;
499 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
502 for (ch = 0; ch < q->nb_channels; ch++)
503 for (i = 0; i < 64; i++) {
504 tmp = q->tone_level_idx[ch][sb][i] = q->tone_level_idx_base[ch][sb][i / 8];
505 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
506 q->tone_level[ch][sb][i] = 0;
508 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
517 * Related to synthesis filter
518 * Called by process_subpacket_11
519 * c is built with data from subpacket 11
520 * Most of this function is used only if superblock_type_2_3 == 0,
521 * never seen it in samples.
523 * @param tone_level_idx
524 * @param tone_level_idx_temp
525 * @param coding_method q->coding_method[0][0][0]
526 * @param nb_channels number of channels
527 * @param c coming from subpacket 11, passed as 8*c
528 * @param superblocktype_2_3 flag based on superblock packet type
529 * @param cm_table_select q->cm_table_select
531 static void fill_coding_method_array(sb_int8_array tone_level_idx,
532 sb_int8_array tone_level_idx_temp,
533 sb_int8_array coding_method,
535 int c, int superblocktype_2_3,
539 int tmp, acc, esp_40, comp;
540 int add1, add2, add3, add4;
543 if (!superblocktype_2_3) {
544 /* This case is untested, no samples available */
545 avpriv_request_sample(NULL, "!superblocktype_2_3");
547 for (ch = 0; ch < nb_channels; ch++) {
548 for (sb = 0; sb < 30; sb++) {
549 for (j = 1; j < 63; j++) { // The loop only iterates to 63 so the code doesn't overflow the buffer
550 add1 = tone_level_idx[ch][sb][j] - 10;
553 add2 = add3 = add4 = 0;
555 add2 = tone_level_idx[ch][sb - 2][j] + tone_level_idx_offset_table[sb][0] - 6;
560 add3 = tone_level_idx[ch][sb - 1][j] + tone_level_idx_offset_table[sb][1] - 6;
565 add4 = tone_level_idx[ch][sb + 1][j] + tone_level_idx_offset_table[sb][3] - 6;
569 tmp = tone_level_idx[ch][sb][j + 1] * 2 - add4 - add3 - add2 - add1;
572 tone_level_idx_temp[ch][sb][j + 1] = tmp & 0xff;
574 tone_level_idx_temp[ch][sb][0] = tone_level_idx_temp[ch][sb][1];
578 for (ch = 0; ch < nb_channels; ch++)
579 for (sb = 0; sb < 30; sb++)
580 for (j = 0; j < 64; j++)
581 acc += tone_level_idx_temp[ch][sb][j];
583 multres = 0x66666667LL * (acc * 10);
584 esp_40 = (multres >> 32) / 8 + ((multres & 0xffffffff) >> 31);
585 for (ch = 0; ch < nb_channels; ch++)
586 for (sb = 0; sb < 30; sb++)
587 for (j = 0; j < 64; j++) {
588 comp = tone_level_idx_temp[ch][sb][j]* esp_40 * 10;
591 comp /= 256; // signed shift
619 coding_method[ch][sb][j] = ((tmp & 0xfffa) + 30 )& 0xff;
621 for (sb = 0; sb < 30; sb++)
622 fix_coding_method_array(sb, nb_channels, coding_method);
623 for (ch = 0; ch < nb_channels; ch++)
624 for (sb = 0; sb < 30; sb++)
625 for (j = 0; j < 64; j++)
627 if (coding_method[ch][sb][j] < 10)
628 coding_method[ch][sb][j] = 10;
631 if (coding_method[ch][sb][j] < 16)
632 coding_method[ch][sb][j] = 16;
634 if (coding_method[ch][sb][j] < 30)
635 coding_method[ch][sb][j] = 30;
638 } else { // superblocktype_2_3 != 0
639 for (ch = 0; ch < nb_channels; ch++)
640 for (sb = 0; sb < 30; sb++)
641 for (j = 0; j < 64; j++)
642 coding_method[ch][sb][j] = coding_method_table[cm_table_select][sb];
647 * Called by process_subpacket_11 to process more data from subpacket 11
649 * Called by process_subpacket_12 to process data from subpacket 12 with
653 * @param gb bitreader context
654 * @param length packet length in bits
655 * @param sb_min lower subband processed (sb_min included)
656 * @param sb_max higher subband processed (sb_max excluded)
658 static int synthfilt_build_sb_samples(QDM2Context *q, GetBitContext *gb,
659 int length, int sb_min, int sb_max)
661 int sb, j, k, n, ch, run, channels;
662 int joined_stereo, zero_encoding;
664 float type34_div = 0;
665 float type34_predictor;
667 int sign_bits[16] = {0};
670 // If no data use noise
671 for (sb=sb_min; sb < sb_max; sb++)
672 build_sb_samples_from_noise(q, sb);
677 for (sb = sb_min; sb < sb_max; sb++) {
678 channels = q->nb_channels;
680 if (q->nb_channels <= 1 || sb < 12)
685 joined_stereo = (get_bits_left(gb) >= 1) ? get_bits1(gb) : 0;
688 if (get_bits_left(gb) >= 16)
689 for (j = 0; j < 16; j++)
690 sign_bits[j] = get_bits1(gb);
692 for (j = 0; j < 64; j++)
693 if (q->coding_method[1][sb][j] > q->coding_method[0][sb][j])
694 q->coding_method[0][sb][j] = q->coding_method[1][sb][j];
696 if (fix_coding_method_array(sb, q->nb_channels,
698 av_log(NULL, AV_LOG_ERROR, "coding method invalid\n");
699 build_sb_samples_from_noise(q, sb);
705 for (ch = 0; ch < channels; ch++) {
706 FIX_NOISE_IDX(q->noise_idx);
707 zero_encoding = (get_bits_left(gb) >= 1) ? get_bits1(gb) : 0;
708 type34_predictor = 0.0;
711 for (j = 0; j < 128; ) {
712 switch (q->coding_method[ch][sb][j / 2]) {
714 if (get_bits_left(gb) >= 10) {
716 for (k = 0; k < 5; k++) {
717 if ((j + 2 * k) >= 128)
719 samples[2 * k] = get_bits1(gb) ? dequant_1bit[joined_stereo][2 * get_bits1(gb)] : 0;
724 av_log(NULL, AV_LOG_ERROR, "Invalid 8bit codeword\n");
725 return AVERROR_INVALIDDATA;
728 for (k = 0; k < 5; k++)
729 samples[2 * k] = dequant_1bit[joined_stereo][random_dequant_index[n][k]];
731 for (k = 0; k < 5; k++)
732 samples[2 * k + 1] = SB_DITHERING_NOISE(sb,q->noise_idx);
734 for (k = 0; k < 10; k++)
735 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
741 if (get_bits_left(gb) >= 1) {
746 f -= noise_samples[((sb + 1) * (j +5 * ch + 1)) & 127] * 9.0 / 40.0;
749 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
755 if (get_bits_left(gb) >= 10) {
757 for (k = 0; k < 5; k++) {
760 samples[k] = (get_bits1(gb) == 0) ? 0 : dequant_1bit[joined_stereo][2 * get_bits1(gb)];
763 n = get_bits (gb, 8);
765 av_log(NULL, AV_LOG_ERROR, "Invalid 8bit codeword\n");
766 return AVERROR_INVALIDDATA;
769 for (k = 0; k < 5; k++)
770 samples[k] = dequant_1bit[joined_stereo][random_dequant_index[n][k]];
773 for (k = 0; k < 5; k++)
774 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
780 if (get_bits_left(gb) >= 7) {
783 av_log(NULL, AV_LOG_ERROR, "Invalid 7bit codeword\n");
784 return AVERROR_INVALIDDATA;
787 for (k = 0; k < 3; k++)
788 samples[k] = (random_dequant_type24[n][k] - 2.0) * 0.5;
790 for (k = 0; k < 3; k++)
791 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
797 if (get_bits_left(gb) >= 4) {
798 unsigned index = qdm2_get_vlc(gb, &vlc_tab_type30, 0, 1);
799 if (index >= FF_ARRAY_ELEMS(type30_dequant)) {
800 av_log(NULL, AV_LOG_ERROR, "index %d out of type30_dequant array\n", index);
801 return AVERROR_INVALIDDATA;
803 samples[0] = type30_dequant[index];
805 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
811 if (get_bits_left(gb) >= 7) {
813 type34_div = (float)(1 << get_bits(gb, 2));
814 samples[0] = ((float)get_bits(gb, 5) - 16.0) / 15.0;
815 type34_predictor = samples[0];
818 unsigned index = qdm2_get_vlc(gb, &vlc_tab_type34, 0, 1);
819 if (index >= FF_ARRAY_ELEMS(type34_delta)) {
820 av_log(NULL, AV_LOG_ERROR, "index %d out of type34_delta array\n", index);
821 return AVERROR_INVALIDDATA;
823 samples[0] = type34_delta[index] / type34_div + type34_predictor;
824 type34_predictor = samples[0];
827 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
833 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
839 for (k = 0; k < run && j + k < 128; k++) {
840 q->sb_samples[0][j + k][sb] =
841 q->tone_level[0][sb][(j + k) / 2] * samples[k];
842 if (q->nb_channels == 2) {
843 if (sign_bits[(j + k) / 8])
844 q->sb_samples[1][j + k][sb] =
845 q->tone_level[1][sb][(j + k) / 2] * -samples[k];
847 q->sb_samples[1][j + k][sb] =
848 q->tone_level[1][sb][(j + k) / 2] * samples[k];
852 for (k = 0; k < run; k++)
854 q->sb_samples[ch][j + k][sb] = q->tone_level[ch][sb][(j + k)/2] * samples[k];
865 * Init the first element of a channel in quantized_coeffs with data
866 * from packet 10 (quantized_coeffs[ch][0]).
867 * This is similar to process_subpacket_9, but for a single channel
868 * and for element [0]
869 * same VLC tables as process_subpacket_9 are used.
871 * @param quantized_coeffs pointer to quantized_coeffs[ch][0]
872 * @param gb bitreader context
874 static int init_quantized_coeffs_elem0(int8_t *quantized_coeffs,
877 int i, k, run, level, diff;
879 if (get_bits_left(gb) < 16)
881 level = qdm2_get_vlc(gb, &vlc_tab_level, 0, 2);
883 quantized_coeffs[0] = level;
885 for (i = 0; i < 7; ) {
886 if (get_bits_left(gb) < 16)
888 run = qdm2_get_vlc(gb, &vlc_tab_run, 0, 1) + 1;
893 if (get_bits_left(gb) < 16)
895 diff = qdm2_get_se_vlc(&vlc_tab_diff, gb, 2);
897 for (k = 1; k <= run; k++)
898 quantized_coeffs[i + k] = (level + ((k * diff) / run));
907 * Related to synthesis filter, process data from packet 10
908 * Init part of quantized_coeffs via function init_quantized_coeffs_elem0
909 * Init tone_level_idx_hi1, tone_level_idx_hi2, tone_level_idx_mid with
910 * data from packet 10
913 * @param gb bitreader context
915 static void init_tone_level_dequantization(QDM2Context *q, GetBitContext *gb)
919 for (ch = 0; ch < q->nb_channels; ch++) {
920 init_quantized_coeffs_elem0(q->quantized_coeffs[ch][0], gb);
922 if (get_bits_left(gb) < 16) {
923 memset(q->quantized_coeffs[ch][0], 0, 8);
928 n = q->sub_sampling + 1;
930 for (sb = 0; sb < n; sb++)
931 for (ch = 0; ch < q->nb_channels; ch++)
932 for (j = 0; j < 8; j++) {
933 if (get_bits_left(gb) < 1)
936 for (k=0; k < 8; k++) {
937 if (get_bits_left(gb) < 16)
939 q->tone_level_idx_hi1[ch][sb][j][k] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_hi1, 0, 2);
942 for (k=0; k < 8; k++)
943 q->tone_level_idx_hi1[ch][sb][j][k] = 0;
947 n = QDM2_SB_USED(q->sub_sampling) - 4;
949 for (sb = 0; sb < n; sb++)
950 for (ch = 0; ch < q->nb_channels; ch++) {
951 if (get_bits_left(gb) < 16)
953 q->tone_level_idx_hi2[ch][sb] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_hi2, 0, 2);
955 q->tone_level_idx_hi2[ch][sb] -= 16;
957 for (j = 0; j < 8; j++)
958 q->tone_level_idx_mid[ch][sb][j] = -16;
961 n = QDM2_SB_USED(q->sub_sampling) - 5;
963 for (sb = 0; sb < n; sb++)
964 for (ch = 0; ch < q->nb_channels; ch++)
965 for (j = 0; j < 8; j++) {
966 if (get_bits_left(gb) < 16)
968 q->tone_level_idx_mid[ch][sb][j] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_mid, 0, 2) - 32;
973 * Process subpacket 9, init quantized_coeffs with data from it
976 * @param node pointer to node with packet
978 static int process_subpacket_9(QDM2Context *q, QDM2SubPNode *node)
981 int i, j, k, n, ch, run, level, diff;
983 init_get_bits(&gb, node->packet->data, node->packet->size * 8);
985 n = coeff_per_sb_for_avg[q->coeff_per_sb_select][QDM2_SB_USED(q->sub_sampling) - 1] + 1;
987 for (i = 1; i < n; i++)
988 for (ch = 0; ch < q->nb_channels; ch++) {
989 level = qdm2_get_vlc(&gb, &vlc_tab_level, 0, 2);
990 q->quantized_coeffs[ch][i][0] = level;
992 for (j = 0; j < (8 - 1); ) {
993 run = qdm2_get_vlc(&gb, &vlc_tab_run, 0, 1) + 1;
994 diff = qdm2_get_se_vlc(&vlc_tab_diff, &gb, 2);
999 for (k = 1; k <= run; k++)
1000 q->quantized_coeffs[ch][i][j + k] = (level + ((k * diff) / run));
1007 for (ch = 0; ch < q->nb_channels; ch++)
1008 for (i = 0; i < 8; i++)
1009 q->quantized_coeffs[ch][0][i] = 0;
1015 * Process subpacket 10 if not null, else
1018 * @param node pointer to node with packet
1020 static void process_subpacket_10(QDM2Context *q, QDM2SubPNode *node)
1025 init_get_bits(&gb, node->packet->data, node->packet->size * 8);
1026 init_tone_level_dequantization(q, &gb);
1027 fill_tone_level_array(q, 1);
1029 fill_tone_level_array(q, 0);
1034 * Process subpacket 11
1037 * @param node pointer to node with packet
1039 static void process_subpacket_11(QDM2Context *q, QDM2SubPNode *node)
1045 length = node->packet->size * 8;
1046 init_get_bits(&gb, node->packet->data, length);
1050 int c = get_bits(&gb, 13);
1053 fill_coding_method_array(q->tone_level_idx,
1054 q->tone_level_idx_temp, q->coding_method,
1055 q->nb_channels, 8 * c,
1056 q->superblocktype_2_3, q->cm_table_select);
1059 synthfilt_build_sb_samples(q, &gb, length, 0, 8);
1063 * Process subpacket 12
1066 * @param node pointer to node with packet
1068 static void process_subpacket_12(QDM2Context *q, QDM2SubPNode *node)
1074 length = node->packet->size * 8;
1075 init_get_bits(&gb, node->packet->data, length);
1078 synthfilt_build_sb_samples(q, &gb, length, 8, QDM2_SB_USED(q->sub_sampling));
1082 * Process new subpackets for synthesis filter
1085 * @param list list with synthesis filter packets (list D)
1087 static void process_synthesis_subpackets(QDM2Context *q, QDM2SubPNode *list)
1089 QDM2SubPNode *nodes[4];
1091 nodes[0] = qdm2_search_subpacket_type_in_list(list, 9);
1093 process_subpacket_9(q, nodes[0]);
1095 nodes[1] = qdm2_search_subpacket_type_in_list(list, 10);
1097 process_subpacket_10(q, nodes[1]);
1099 process_subpacket_10(q, NULL);
1101 nodes[2] = qdm2_search_subpacket_type_in_list(list, 11);
1102 if (nodes[0] && nodes[1] && nodes[2])
1103 process_subpacket_11(q, nodes[2]);
1105 process_subpacket_11(q, NULL);
1107 nodes[3] = qdm2_search_subpacket_type_in_list(list, 12);
1108 if (nodes[0] && nodes[1] && nodes[3])
1109 process_subpacket_12(q, nodes[3]);
1111 process_subpacket_12(q, NULL);
1115 * Decode superblock, fill packet lists.
1119 static void qdm2_decode_super_block(QDM2Context *q)
1122 QDM2SubPacket header, *packet;
1123 int i, packet_bytes, sub_packet_size, sub_packets_D;
1124 unsigned int next_index = 0;
1126 memset(q->tone_level_idx_hi1, 0, sizeof(q->tone_level_idx_hi1));
1127 memset(q->tone_level_idx_mid, 0, sizeof(q->tone_level_idx_mid));
1128 memset(q->tone_level_idx_hi2, 0, sizeof(q->tone_level_idx_hi2));
1130 q->sub_packets_B = 0;
1133 average_quantized_coeffs(q); // average elements in quantized_coeffs[max_ch][10][8]
1135 init_get_bits(&gb, q->compressed_data, q->compressed_size * 8);
1136 qdm2_decode_sub_packet_header(&gb, &header);
1138 if (header.type < 2 || header.type >= 8) {
1140 av_log(NULL, AV_LOG_ERROR, "bad superblock type\n");
1144 q->superblocktype_2_3 = (header.type == 2 || header.type == 3);
1145 packet_bytes = (q->compressed_size - get_bits_count(&gb) / 8);
1147 init_get_bits(&gb, header.data, header.size * 8);
1149 if (header.type == 2 || header.type == 4 || header.type == 5) {
1150 int csum = 257 * get_bits(&gb, 8);
1151 csum += 2 * get_bits(&gb, 8);
1153 csum = qdm2_packet_checksum(q->compressed_data, q->checksum_size, csum);
1157 av_log(NULL, AV_LOG_ERROR, "bad packet checksum\n");
1162 q->sub_packet_list_B[0].packet = NULL;
1163 q->sub_packet_list_D[0].packet = NULL;
1165 for (i = 0; i < 6; i++)
1166 if (--q->fft_level_exp[i] < 0)
1167 q->fft_level_exp[i] = 0;
1169 for (i = 0; packet_bytes > 0; i++) {
1172 if (i >= FF_ARRAY_ELEMS(q->sub_packet_list_A)) {
1173 SAMPLES_NEEDED_2("too many packet bytes");
1177 q->sub_packet_list_A[i].next = NULL;
1180 q->sub_packet_list_A[i - 1].next = &q->sub_packet_list_A[i];
1182 /* seek to next block */
1183 init_get_bits(&gb, header.data, header.size * 8);
1184 skip_bits(&gb, next_index * 8);
1186 if (next_index >= header.size)
1190 /* decode subpacket */
1191 packet = &q->sub_packets[i];
1192 qdm2_decode_sub_packet_header(&gb, packet);
1193 next_index = packet->size + get_bits_count(&gb) / 8;
1194 sub_packet_size = ((packet->size > 0xff) ? 1 : 0) + packet->size + 2;
1196 if (packet->type == 0)
1199 if (sub_packet_size > packet_bytes) {
1200 if (packet->type != 10 && packet->type != 11 && packet->type != 12)
1202 packet->size += packet_bytes - sub_packet_size;
1205 packet_bytes -= sub_packet_size;
1207 /* add subpacket to 'all subpackets' list */
1208 q->sub_packet_list_A[i].packet = packet;
1210 /* add subpacket to related list */
1211 if (packet->type == 8) {
1212 SAMPLES_NEEDED_2("packet type 8");
1214 } else if (packet->type >= 9 && packet->type <= 12) {
1215 /* packets for MPEG Audio like Synthesis Filter */
1216 QDM2_LIST_ADD(q->sub_packet_list_D, sub_packets_D, packet);
1217 } else if (packet->type == 13) {
1218 for (j = 0; j < 6; j++)
1219 q->fft_level_exp[j] = get_bits(&gb, 6);
1220 } else if (packet->type == 14) {
1221 for (j = 0; j < 6; j++)
1222 q->fft_level_exp[j] = qdm2_get_vlc(&gb, &fft_level_exp_vlc, 0, 2);
1223 } else if (packet->type == 15) {
1224 SAMPLES_NEEDED_2("packet type 15")
1226 } else if (packet->type >= 16 && packet->type < 48 &&
1227 !fft_subpackets[packet->type - 16]) {
1228 /* packets for FFT */
1229 QDM2_LIST_ADD(q->sub_packet_list_B, q->sub_packets_B, packet);
1231 } // Packet bytes loop
1233 if (q->sub_packet_list_D[0].packet) {
1234 process_synthesis_subpackets(q, q->sub_packet_list_D);
1235 q->do_synth_filter = 1;
1236 } else if (q->do_synth_filter) {
1237 process_subpacket_10(q, NULL);
1238 process_subpacket_11(q, NULL);
1239 process_subpacket_12(q, NULL);
1243 static void qdm2_fft_init_coefficient(QDM2Context *q, int sub_packet,
1244 int offset, int duration, int channel,
1247 if (q->fft_coefs_min_index[duration] < 0)
1248 q->fft_coefs_min_index[duration] = q->fft_coefs_index;
1250 q->fft_coefs[q->fft_coefs_index].sub_packet =
1251 ((sub_packet >= 16) ? (sub_packet - 16) : sub_packet);
1252 q->fft_coefs[q->fft_coefs_index].channel = channel;
1253 q->fft_coefs[q->fft_coefs_index].offset = offset;
1254 q->fft_coefs[q->fft_coefs_index].exp = exp;
1255 q->fft_coefs[q->fft_coefs_index].phase = phase;
1256 q->fft_coefs_index++;
1259 static void qdm2_fft_decode_tones(QDM2Context *q, int duration,
1260 GetBitContext *gb, int b)
1262 int channel, stereo, phase, exp;
1263 int local_int_4, local_int_8, stereo_phase, local_int_10;
1264 int local_int_14, stereo_exp, local_int_20, local_int_28;
1270 local_int_8 = (4 - duration);
1271 local_int_10 = 1 << (q->group_order - duration - 1);
1274 while (get_bits_left(gb)>0) {
1275 if (q->superblocktype_2_3) {
1276 while ((n = qdm2_get_vlc(gb, &vlc_tab_fft_tone_offset[local_int_8], 1, 2)) < 2) {
1277 if (get_bits_left(gb)<0) {
1278 if(local_int_4 < q->group_size)
1279 av_log(NULL, AV_LOG_ERROR, "overread in qdm2_fft_decode_tones()\n");
1284 local_int_4 += local_int_10;
1285 local_int_28 += (1 << local_int_8);
1287 local_int_4 += 8 * local_int_10;
1288 local_int_28 += (8 << local_int_8);
1293 if (local_int_10 <= 2) {
1294 av_log(NULL, AV_LOG_ERROR, "qdm2_fft_decode_tones() stuck\n");
1297 offset += qdm2_get_vlc(gb, &vlc_tab_fft_tone_offset[local_int_8], 1, 2);
1298 while (offset >= (local_int_10 - 1)) {
1299 offset += (1 - (local_int_10 - 1));
1300 local_int_4 += local_int_10;
1301 local_int_28 += (1 << local_int_8);
1305 if (local_int_4 >= q->group_size)
1308 local_int_14 = (offset >> local_int_8);
1309 if (local_int_14 >= FF_ARRAY_ELEMS(fft_level_index_table))
1312 if (q->nb_channels > 1) {
1313 channel = get_bits1(gb);
1314 stereo = get_bits1(gb);
1320 exp = qdm2_get_vlc(gb, (b ? &fft_level_exp_vlc : &fft_level_exp_alt_vlc), 0, 2);
1321 exp += q->fft_level_exp[fft_level_index_table[local_int_14]];
1322 exp = (exp < 0) ? 0 : exp;
1324 phase = get_bits(gb, 3);
1329 stereo_exp = (exp - qdm2_get_vlc(gb, &fft_stereo_exp_vlc, 0, 1));
1330 stereo_phase = (phase - qdm2_get_vlc(gb, &fft_stereo_phase_vlc, 0, 1));
1331 if (stereo_phase < 0)
1335 if (q->frequency_range > (local_int_14 + 1)) {
1336 int sub_packet = (local_int_20 + local_int_28);
1338 if (q->fft_coefs_index + stereo >= FF_ARRAY_ELEMS(q->fft_coefs))
1341 qdm2_fft_init_coefficient(q, sub_packet, offset, duration,
1342 channel, exp, phase);
1344 qdm2_fft_init_coefficient(q, sub_packet, offset, duration,
1346 stereo_exp, stereo_phase);
1352 static void qdm2_decode_fft_packets(QDM2Context *q)
1354 int i, j, min, max, value, type, unknown_flag;
1357 if (!q->sub_packet_list_B[0].packet)
1360 /* reset minimum indexes for FFT coefficients */
1361 q->fft_coefs_index = 0;
1362 for (i = 0; i < 5; i++)
1363 q->fft_coefs_min_index[i] = -1;
1365 /* process subpackets ordered by type, largest type first */
1366 for (i = 0, max = 256; i < q->sub_packets_B; i++) {
1367 QDM2SubPacket *packet = NULL;
1369 /* find subpacket with largest type less than max */
1370 for (j = 0, min = 0; j < q->sub_packets_B; j++) {
1371 value = q->sub_packet_list_B[j].packet->type;
1372 if (value > min && value < max) {
1374 packet = q->sub_packet_list_B[j].packet;
1380 /* check for errors (?) */
1385 (packet->type < 16 || packet->type >= 48 ||
1386 fft_subpackets[packet->type - 16]))
1389 /* decode FFT tones */
1390 init_get_bits(&gb, packet->data, packet->size * 8);
1392 if (packet->type >= 32 && packet->type < 48 && !fft_subpackets[packet->type - 16])
1397 type = packet->type;
1399 if ((type >= 17 && type < 24) || (type >= 33 && type < 40)) {
1400 int duration = q->sub_sampling + 5 - (type & 15);
1402 if (duration >= 0 && duration < 4)
1403 qdm2_fft_decode_tones(q, duration, &gb, unknown_flag);
1404 } else if (type == 31) {
1405 for (j = 0; j < 4; j++)
1406 qdm2_fft_decode_tones(q, j, &gb, unknown_flag);
1407 } else if (type == 46) {
1408 for (j = 0; j < 6; j++)
1409 q->fft_level_exp[j] = get_bits(&gb, 6);
1410 for (j = 0; j < 4; j++)
1411 qdm2_fft_decode_tones(q, j, &gb, unknown_flag);
1413 } // Loop on B packets
1415 /* calculate maximum indexes for FFT coefficients */
1416 for (i = 0, j = -1; i < 5; i++)
1417 if (q->fft_coefs_min_index[i] >= 0) {
1419 q->fft_coefs_max_index[j] = q->fft_coefs_min_index[i];
1423 q->fft_coefs_max_index[j] = q->fft_coefs_index;
1426 static void qdm2_fft_generate_tone(QDM2Context *q, FFTTone *tone)
1431 const double iscale = 2.0 * M_PI / 512.0;
1433 tone->phase += tone->phase_shift;
1435 /* calculate current level (maximum amplitude) of tone */
1436 level = fft_tone_envelope_table[tone->duration][tone->time_index] * tone->level;
1437 c.im = level * sin(tone->phase * iscale);
1438 c.re = level * cos(tone->phase * iscale);
1440 /* generate FFT coefficients for tone */
1441 if (tone->duration >= 3 || tone->cutoff >= 3) {
1442 tone->complex[0].im += c.im;
1443 tone->complex[0].re += c.re;
1444 tone->complex[1].im -= c.im;
1445 tone->complex[1].re -= c.re;
1447 f[1] = -tone->table[4];
1448 f[0] = tone->table[3] - tone->table[0];
1449 f[2] = 1.0 - tone->table[2] - tone->table[3];
1450 f[3] = tone->table[1] + tone->table[4] - 1.0;
1451 f[4] = tone->table[0] - tone->table[1];
1452 f[5] = tone->table[2];
1453 for (i = 0; i < 2; i++) {
1454 tone->complex[fft_cutoff_index_table[tone->cutoff][i]].re +=
1456 tone->complex[fft_cutoff_index_table[tone->cutoff][i]].im +=
1457 c.im * ((tone->cutoff <= i) ? -f[i] : f[i]);
1459 for (i = 0; i < 4; i++) {
1460 tone->complex[i].re += c.re * f[i + 2];
1461 tone->complex[i].im += c.im * f[i + 2];
1465 /* copy the tone if it has not yet died out */
1466 if (++tone->time_index < ((1 << (5 - tone->duration)) - 1)) {
1467 memcpy(&q->fft_tones[q->fft_tone_end], tone, sizeof(FFTTone));
1468 q->fft_tone_end = (q->fft_tone_end + 1) % 1000;
1472 static void qdm2_fft_tone_synthesizer(QDM2Context *q, int sub_packet)
1475 const double iscale = 0.25 * M_PI;
1477 for (ch = 0; ch < q->channels; ch++) {
1478 memset(q->fft.complex[ch], 0, q->fft_size * sizeof(QDM2Complex));
1482 /* apply FFT tones with duration 4 (1 FFT period) */
1483 if (q->fft_coefs_min_index[4] >= 0)
1484 for (i = q->fft_coefs_min_index[4]; i < q->fft_coefs_max_index[4]; i++) {
1488 if (q->fft_coefs[i].sub_packet != sub_packet)
1491 ch = (q->channels == 1) ? 0 : q->fft_coefs[i].channel;
1492 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];
1494 c.re = level * cos(q->fft_coefs[i].phase * iscale);
1495 c.im = level * sin(q->fft_coefs[i].phase * iscale);
1496 q->fft.complex[ch][q->fft_coefs[i].offset + 0].re += c.re;
1497 q->fft.complex[ch][q->fft_coefs[i].offset + 0].im += c.im;
1498 q->fft.complex[ch][q->fft_coefs[i].offset + 1].re -= c.re;
1499 q->fft.complex[ch][q->fft_coefs[i].offset + 1].im -= c.im;
1502 /* generate existing FFT tones */
1503 for (i = q->fft_tone_end; i != q->fft_tone_start; ) {
1504 qdm2_fft_generate_tone(q, &q->fft_tones[q->fft_tone_start]);
1505 q->fft_tone_start = (q->fft_tone_start + 1) % 1000;
1508 /* create and generate new FFT tones with duration 0 (long) to 3 (short) */
1509 for (i = 0; i < 4; i++)
1510 if (q->fft_coefs_min_index[i] >= 0) {
1511 for (j = q->fft_coefs_min_index[i]; j < q->fft_coefs_max_index[i]; j++) {
1515 if (q->fft_coefs[j].sub_packet != sub_packet)
1519 offset = q->fft_coefs[j].offset >> four_i;
1520 ch = (q->channels == 1) ? 0 : q->fft_coefs[j].channel;
1522 if (offset < q->frequency_range) {
1524 tone.cutoff = offset;
1526 tone.cutoff = (offset >= 60) ? 3 : 2;
1528 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];
1529 tone.complex = &q->fft.complex[ch][offset];
1530 tone.table = fft_tone_sample_table[i][q->fft_coefs[j].offset - (offset << four_i)];
1531 tone.phase = 64 * q->fft_coefs[j].phase - (offset << 8) - 128;
1532 tone.phase_shift = (2 * q->fft_coefs[j].offset + 1) << (7 - four_i);
1534 tone.time_index = 0;
1536 qdm2_fft_generate_tone(q, &tone);
1539 q->fft_coefs_min_index[i] = j;
1543 static void qdm2_calculate_fft(QDM2Context *q, int channel, int sub_packet)
1545 const float gain = (q->channels == 1 && q->nb_channels == 2) ? 0.5f : 1.0f;
1546 float *out = q->output_buffer + channel;
1548 q->fft.complex[channel][0].re *= 2.0f;
1549 q->fft.complex[channel][0].im = 0.0f;
1550 q->rdft_ctx.rdft_calc(&q->rdft_ctx, (FFTSample *)q->fft.complex[channel]);
1551 /* add samples to output buffer */
1552 for (i = 0; i < FFALIGN(q->fft_size, 8); i++) {
1553 out[0] += q->fft.complex[channel][i].re * gain;
1554 out[q->channels] += q->fft.complex[channel][i].im * gain;
1555 out += 2 * q->channels;
1561 * @param index subpacket number
1563 static void qdm2_synthesis_filter(QDM2Context *q, int index)
1565 int i, k, ch, sb_used, sub_sampling, dither_state = 0;
1567 /* copy sb_samples */
1568 sb_used = QDM2_SB_USED(q->sub_sampling);
1570 for (ch = 0; ch < q->channels; ch++)
1571 for (i = 0; i < 8; i++)
1572 for (k = sb_used; k < SBLIMIT; k++)
1573 q->sb_samples[ch][(8 * index) + i][k] = 0;
1575 for (ch = 0; ch < q->nb_channels; ch++) {
1576 float *samples_ptr = q->samples + ch;
1578 for (i = 0; i < 8; i++) {
1579 ff_mpa_synth_filter_float(&q->mpadsp,
1580 q->synth_buf[ch], &(q->synth_buf_offset[ch]),
1581 ff_mpa_synth_window_float, &dither_state,
1582 samples_ptr, q->nb_channels,
1583 q->sb_samples[ch][(8 * index) + i]);
1584 samples_ptr += 32 * q->nb_channels;
1588 /* add samples to output buffer */
1589 sub_sampling = (4 >> q->sub_sampling);
1591 for (ch = 0; ch < q->channels; ch++)
1592 for (i = 0; i < q->frame_size; i++)
1593 q->output_buffer[q->channels * i + ch] += (1 << 23) * q->samples[q->nb_channels * sub_sampling * i + ch];
1597 * Init static data (does not depend on specific file)
1599 static av_cold void qdm2_init_static_data(void) {
1601 softclip_table_init();
1603 init_noise_samples();
1605 ff_mpa_synth_init_float();
1609 * Init parameters from codec extradata
1611 static av_cold int qdm2_decode_init(AVCodecContext *avctx)
1613 static AVOnce init_static_once = AV_ONCE_INIT;
1614 QDM2Context *s = avctx->priv_data;
1615 int tmp_val, tmp, size;
1618 /* extradata parsing
1627 32 size (including this field)
1629 32 type (=QDM2 or QDMC)
1631 32 size (including this field, in bytes)
1632 32 tag (=QDCA) // maybe mandatory parameters
1635 32 samplerate (=44100)
1637 32 block size (=4096)
1638 32 frame size (=256) (for one channel)
1639 32 packet size (=1300)
1641 32 size (including this field, in bytes)
1642 32 tag (=QDCP) // maybe some tuneable parameters
1652 if (!avctx->extradata || (avctx->extradata_size < 48)) {
1653 av_log(avctx, AV_LOG_ERROR, "extradata missing or truncated\n");
1654 return AVERROR_INVALIDDATA;
1657 bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
1659 while (bytestream2_get_bytes_left(&gb) > 8) {
1660 if (bytestream2_peek_be64(&gb) == (((uint64_t)MKBETAG('f','r','m','a') << 32) |
1661 (uint64_t)MKBETAG('Q','D','M','2')))
1663 bytestream2_skip(&gb, 1);
1666 if (bytestream2_get_bytes_left(&gb) < 12) {
1667 av_log(avctx, AV_LOG_ERROR, "not enough extradata (%i)\n",
1668 bytestream2_get_bytes_left(&gb));
1669 return AVERROR_INVALIDDATA;
1672 bytestream2_skip(&gb, 8);
1673 size = bytestream2_get_be32(&gb);
1675 if (size > bytestream2_get_bytes_left(&gb)) {
1676 av_log(avctx, AV_LOG_ERROR, "extradata size too small, %i < %i\n",
1677 bytestream2_get_bytes_left(&gb), size);
1678 return AVERROR_INVALIDDATA;
1681 av_log(avctx, AV_LOG_DEBUG, "size: %d\n", size);
1682 if (bytestream2_get_be32(&gb) != MKBETAG('Q','D','C','A')) {
1683 av_log(avctx, AV_LOG_ERROR, "invalid extradata, expecting QDCA\n");
1684 return AVERROR_INVALIDDATA;
1687 bytestream2_skip(&gb, 4);
1689 avctx->channels = s->nb_channels = s->channels = bytestream2_get_be32(&gb);
1690 if (s->channels <= 0 || s->channels > MPA_MAX_CHANNELS) {
1691 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1692 return AVERROR_INVALIDDATA;
1694 avctx->channel_layout = avctx->channels == 2 ? AV_CH_LAYOUT_STEREO :
1697 avctx->sample_rate = bytestream2_get_be32(&gb);
1698 avctx->bit_rate = bytestream2_get_be32(&gb);
1699 s->group_size = bytestream2_get_be32(&gb);
1700 s->fft_size = bytestream2_get_be32(&gb);
1701 s->checksum_size = bytestream2_get_be32(&gb);
1702 if (s->checksum_size >= 1U << 28 || s->checksum_size <= 1) {
1703 av_log(avctx, AV_LOG_ERROR, "data block size invalid (%u)\n", s->checksum_size);
1704 return AVERROR_INVALIDDATA;
1707 s->fft_order = av_log2(s->fft_size) + 1;
1709 // Fail on unknown fft order
1710 if ((s->fft_order < 7) || (s->fft_order > 9)) {
1711 avpriv_request_sample(avctx, "Unknown FFT order %d", s->fft_order);
1712 return AVERROR_PATCHWELCOME;
1715 // something like max decodable tones
1716 s->group_order = av_log2(s->group_size) + 1;
1717 s->frame_size = s->group_size / 16; // 16 iterations per super block
1719 if (s->frame_size > QDM2_MAX_FRAME_SIZE)
1720 return AVERROR_INVALIDDATA;
1722 s->sub_sampling = s->fft_order - 7;
1723 s->frequency_range = 255 / (1 << (2 - s->sub_sampling));
1725 if (s->frame_size * 4 >> s->sub_sampling > MPA_FRAME_SIZE) {
1726 avpriv_request_sample(avctx, "large frames");
1727 return AVERROR_PATCHWELCOME;
1730 switch ((s->sub_sampling * 2 + s->channels - 1)) {
1731 case 0: tmp = 40; break;
1732 case 1: tmp = 48; break;
1733 case 2: tmp = 56; break;
1734 case 3: tmp = 72; break;
1735 case 4: tmp = 80; break;
1736 case 5: tmp = 100;break;
1737 default: tmp=s->sub_sampling; break;
1740 if ((tmp * 1000) < avctx->bit_rate) tmp_val = 1;
1741 if ((tmp * 1440) < avctx->bit_rate) tmp_val = 2;
1742 if ((tmp * 1760) < avctx->bit_rate) tmp_val = 3;
1743 if ((tmp * 2240) < avctx->bit_rate) tmp_val = 4;
1744 s->cm_table_select = tmp_val;
1746 if (avctx->bit_rate <= 8000)
1747 s->coeff_per_sb_select = 0;
1748 else if (avctx->bit_rate < 16000)
1749 s->coeff_per_sb_select = 1;
1751 s->coeff_per_sb_select = 2;
1753 if (s->fft_size != (1 << (s->fft_order - 1))) {
1754 av_log(avctx, AV_LOG_ERROR, "FFT size %d not power of 2.\n", s->fft_size);
1755 return AVERROR_INVALIDDATA;
1758 ff_rdft_init(&s->rdft_ctx, s->fft_order, IDFT_C2R);
1759 ff_mpadsp_init(&s->mpadsp);
1761 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
1763 ff_thread_once(&init_static_once, qdm2_init_static_data);
1768 static av_cold int qdm2_decode_close(AVCodecContext *avctx)
1770 QDM2Context *s = avctx->priv_data;
1772 ff_rdft_end(&s->rdft_ctx);
1777 static int qdm2_decode(QDM2Context *q, const uint8_t *in, int16_t *out)
1780 const int frame_size = (q->frame_size * q->channels);
1782 if((unsigned)frame_size > FF_ARRAY_ELEMS(q->output_buffer)/2)
1785 /* select input buffer */
1786 q->compressed_data = in;
1787 q->compressed_size = q->checksum_size;
1789 /* copy old block, clear new block of output samples */
1790 memmove(q->output_buffer, &q->output_buffer[frame_size], frame_size * sizeof(float));
1791 memset(&q->output_buffer[frame_size], 0, frame_size * sizeof(float));
1793 /* decode block of QDM2 compressed data */
1794 if (q->sub_packet == 0) {
1795 q->has_errors = 0; // zero it for a new super block
1796 av_log(NULL,AV_LOG_DEBUG,"Superblock follows\n");
1797 qdm2_decode_super_block(q);
1800 /* parse subpackets */
1801 if (!q->has_errors) {
1802 if (q->sub_packet == 2)
1803 qdm2_decode_fft_packets(q);
1805 qdm2_fft_tone_synthesizer(q, q->sub_packet);
1808 /* sound synthesis stage 1 (FFT) */
1809 for (ch = 0; ch < q->channels; ch++) {
1810 qdm2_calculate_fft(q, ch, q->sub_packet);
1812 if (!q->has_errors && q->sub_packet_list_C[0].packet) {
1813 SAMPLES_NEEDED_2("has errors, and C list is not empty")
1818 /* sound synthesis stage 2 (MPEG audio like synthesis filter) */
1819 if (!q->has_errors && q->do_synth_filter)
1820 qdm2_synthesis_filter(q, q->sub_packet);
1822 q->sub_packet = (q->sub_packet + 1) % 16;
1824 /* clip and convert output float[] to 16-bit signed samples */
1825 for (i = 0; i < frame_size; i++) {
1826 int value = (int)q->output_buffer[i];
1828 if (value > SOFTCLIP_THRESHOLD)
1829 value = (value > HARDCLIP_THRESHOLD) ? 32767 : softclip_table[ value - SOFTCLIP_THRESHOLD];
1830 else if (value < -SOFTCLIP_THRESHOLD)
1831 value = (value < -HARDCLIP_THRESHOLD) ? -32767 : -softclip_table[-value - SOFTCLIP_THRESHOLD];
1839 static int qdm2_decode_frame(AVCodecContext *avctx, void *data,
1840 int *got_frame_ptr, AVPacket *avpkt)
1842 AVFrame *frame = data;
1843 const uint8_t *buf = avpkt->data;
1844 int buf_size = avpkt->size;
1845 QDM2Context *s = avctx->priv_data;
1851 if(buf_size < s->checksum_size)
1854 /* get output buffer */
1855 frame->nb_samples = 16 * s->frame_size;
1856 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1858 out = (int16_t *)frame->data[0];
1860 for (i = 0; i < 16; i++) {
1861 if ((ret = qdm2_decode(s, buf, out)) < 0)
1863 out += s->channels * s->frame_size;
1868 return s->checksum_size;
1871 AVCodec ff_qdm2_decoder = {
1873 .long_name = NULL_IF_CONFIG_SMALL("QDesign Music Codec 2"),
1874 .type = AVMEDIA_TYPE_AUDIO,
1875 .id = AV_CODEC_ID_QDM2,
1876 .priv_data_size = sizeof(QDM2Context),
1877 .init = qdm2_decode_init,
1878 .close = qdm2_decode_close,
1879 .decode = qdm2_decode_frame,
1880 .capabilities = AV_CODEC_CAP_DR1,
1881 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,