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"
40 #define BITSTREAM_READER_LE
43 #include "bytestream.h"
45 #include "mpegaudio.h"
46 #include "mpegaudiodsp.h"
49 #include "qdm2_tablegen.h"
51 #define QDM2_LIST_ADD(list, size, packet) \
54 list[size - 1].next = &list[size]; \
56 list[size].packet = packet; \
57 list[size].next = NULL; \
61 // Result is 8, 16 or 30
62 #define QDM2_SB_USED(sub_sampling) (((sub_sampling) >= 2) ? 30 : 8 << (sub_sampling))
64 #define FIX_NOISE_IDX(noise_idx) \
65 if ((noise_idx) >= 3840) \
66 (noise_idx) -= 3840; \
68 #define SB_DITHERING_NOISE(sb,noise_idx) (noise_table[(noise_idx)++] * sb_noise_attenuation[(sb)])
70 #define SAMPLES_NEEDED \
71 av_log (NULL,AV_LOG_INFO,"This file triggers some untested code. Please contact the developers.\n");
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);
76 #define QDM2_MAX_FRAME_SIZE 512
78 typedef int8_t sb_int8_array[2][30][64];
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)
90 * A node in the subpacket list
92 typedef struct QDM2SubPNode {
93 QDM2SubPacket *packet; ///< packet
94 struct QDM2SubPNode *next; ///< pointer to next packet in the list, NULL if leaf node
97 typedef struct QDM2Complex {
102 typedef struct FFTTone {
104 QDM2Complex *complex;
113 typedef struct FFTCoefficient {
121 typedef struct QDM2FFT {
122 DECLARE_ALIGNED(32, QDM2Complex, complex)[MPA_MAX_CHANNELS][256];
126 * QDM2 decoder context
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
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
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)
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
154 FFTTone fft_tones[1000];
157 FFTCoefficient fft_coefs[1000];
159 int fft_coefs_min_index[5];
160 int fft_coefs_max_index[5];
161 int fft_level_exp[6];
162 RDFTContext rdft_ctx;
166 const uint8_t *compressed_data;
168 float output_buffer[QDM2_MAX_FRAME_SIZE * MPA_MAX_CHANNELS * 2];
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];
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];
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
194 int noise_idx; ///< index for dithering noise table
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
201 static int qdm2_get_vlc(GetBitContext *gb, const VLC *vlc, int flag, int depth)
205 value = get_vlc2(gb, vlc->table, vlc->bits, depth);
207 /* stage-2, 3 bits exponent escape sequence */
209 value = get_bits(gb, get_bits(gb, 3) + 1);
211 /* stage-3, optional */
216 av_log(NULL, AV_LOG_ERROR, "value %d in qdm2_get_vlc too large\n", value);
220 tmp= vlc_stage3_values[value];
222 if ((value & ~3) > 0)
223 tmp += get_bits(gb, (value >> 2));
230 static int qdm2_get_se_vlc(const VLC *vlc, GetBitContext *gb, int depth)
232 int value = qdm2_get_vlc(gb, vlc, 0, depth);
234 return (value & 1) ? ((value + 1) >> 1) : -(value >> 1);
240 * @param data pointer to data to be checksummed
241 * @param length data length
242 * @param value checksum value
244 * @return 0 if checksum is OK
246 static uint16_t qdm2_packet_checksum(const uint8_t *data, int length, int value)
250 for (i = 0; i < length; i++)
253 return (uint16_t)(value & 0xffff);
257 * Fill a QDM2SubPacket structure with packet type, size, and data pointer.
259 * @param gb bitreader context
260 * @param sub_packet packet under analysis
262 static void qdm2_decode_sub_packet_header(GetBitContext *gb,
263 QDM2SubPacket *sub_packet)
265 sub_packet->type = get_bits(gb, 8);
267 if (sub_packet->type == 0) {
268 sub_packet->size = 0;
269 sub_packet->data = NULL;
271 sub_packet->size = get_bits(gb, 8);
273 if (sub_packet->type & 0x80) {
274 sub_packet->size <<= 8;
275 sub_packet->size |= get_bits(gb, 8);
276 sub_packet->type &= 0x7f;
279 if (sub_packet->type == 0x7f)
280 sub_packet->type |= (get_bits(gb, 8) << 8);
282 // FIXME: this depends on bitreader-internal data
283 sub_packet->data = &gb->buffer[get_bits_count(gb) / 8];
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);
291 * Return node pointer to first packet of requested type in list.
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
297 static QDM2SubPNode *qdm2_search_subpacket_type_in_list(QDM2SubPNode *list,
300 while (list && list->packet) {
301 if (list->packet->type == type)
309 * Replace 8 elements with their average value.
310 * Called by qdm2_decode_superblock before starting subblock decoding.
314 static void average_quantized_coeffs(QDM2Context *q)
316 int i, j, n, ch, sum;
318 n = coeff_per_sb_for_avg[q->coeff_per_sb_select][QDM2_SB_USED(q->sub_sampling) - 1] + 1;
320 for (ch = 0; ch < q->nb_channels; ch++)
321 for (i = 0; i < n; i++) {
324 for (j = 0; j < 8; j++)
325 sum += q->quantized_coeffs[ch][i][j];
331 for (j = 0; j < 8; j++)
332 q->quantized_coeffs[ch][i][j] = sum;
337 * Build subband samples with noise weighted by q->tone_level.
338 * Called by synthfilt_build_sb_samples.
341 * @param sb subband index
343 static void build_sb_samples_from_noise(QDM2Context *q, int sb)
347 FIX_NOISE_IDX(q->noise_idx);
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];
363 * Called while processing data from subpackets 11 and 12.
364 * Used after making changes to coding_method array.
366 * @param sb subband index
367 * @param channels number of channels
368 * @param coding_method q->coding_method[0][0][0]
370 static int fix_coding_method_array(int sb, int channels,
371 sb_int8_array coding_method)
377 for (ch = 0; ch < channels; ch++) {
378 for (j = 0; j < 64; ) {
379 if (coding_method[ch][sb][j] < 8)
381 if ((coding_method[ch][sb][j] - 8) > 22) {
385 switch (switchtable[coding_method[ch][sb][j] - 8]) {
409 for (k = 0; k < run; k++) {
411 int sbjk = sb + (j + k) / 64;
416 if (coding_method[ch][sbjk][(j + k) % 64] > coding_method[ch][sb][j]) {
419 //not debugged, almost never used
420 memset(&coding_method[ch][sb][j + k], case_val,
422 memset(&coding_method[ch][sb][j + k], case_val,
435 * Related to synthesis filter
436 * Called by process_subpacket_10
439 * @param flag 1 if called after getting data from subpacket 10, 0 if no subpacket 10
441 static void fill_tone_level_array(QDM2Context *q, int flag)
443 int i, sb, ch, sb_used;
446 for (ch = 0; ch < q->nb_channels; ch++)
447 for (sb = 0; sb < 30; sb++)
448 for (i = 0; i < 8; i++) {
449 if ((tab=coeff_per_sb_for_dequant[q->coeff_per_sb_select][sb]) < (last_coeff[q->coeff_per_sb_select] - 1))
450 tmp = q->quantized_coeffs[ch][tab + 1][i] * dequant_table[q->coeff_per_sb_select][tab + 1][sb]+
451 q->quantized_coeffs[ch][tab][i] * dequant_table[q->coeff_per_sb_select][tab][sb];
453 tmp = q->quantized_coeffs[ch][tab][i] * dequant_table[q->coeff_per_sb_select][tab][sb];
456 q->tone_level_idx_base[ch][sb][i] = (tmp / 256) & 0xff;
459 sb_used = QDM2_SB_USED(q->sub_sampling);
461 if ((q->superblocktype_2_3 != 0) && !flag) {
462 for (sb = 0; sb < sb_used; sb++)
463 for (ch = 0; ch < q->nb_channels; ch++)
464 for (i = 0; i < 64; i++) {
465 q->tone_level_idx[ch][sb][i] = q->tone_level_idx_base[ch][sb][i / 8];
466 if (q->tone_level_idx[ch][sb][i] < 0)
467 q->tone_level[ch][sb][i] = 0;
469 q->tone_level[ch][sb][i] = fft_tone_level_table[0][q->tone_level_idx[ch][sb][i] & 0x3f];
472 tab = q->superblocktype_2_3 ? 0 : 1;
473 for (sb = 0; sb < sb_used; sb++) {
474 if ((sb >= 4) && (sb <= 23)) {
475 for (ch = 0; ch < q->nb_channels; ch++)
476 for (i = 0; i < 64; i++) {
477 tmp = q->tone_level_idx_base[ch][sb][i / 8] -
478 q->tone_level_idx_hi1[ch][sb / 8][i / 8][i % 8] -
479 q->tone_level_idx_mid[ch][sb - 4][i / 8] -
480 q->tone_level_idx_hi2[ch][sb - 4];
481 q->tone_level_idx[ch][sb][i] = tmp & 0xff;
482 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
483 q->tone_level[ch][sb][i] = 0;
485 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
489 for (ch = 0; ch < q->nb_channels; ch++)
490 for (i = 0; i < 64; i++) {
491 tmp = q->tone_level_idx_base[ch][sb][i / 8] -
492 q->tone_level_idx_hi1[ch][2][i / 8][i % 8] -
493 q->tone_level_idx_hi2[ch][sb - 4];
494 q->tone_level_idx[ch][sb][i] = tmp & 0xff;
495 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
496 q->tone_level[ch][sb][i] = 0;
498 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
501 for (ch = 0; ch < q->nb_channels; ch++)
502 for (i = 0; i < 64; i++) {
503 tmp = q->tone_level_idx[ch][sb][i] = q->tone_level_idx_base[ch][sb][i / 8];
504 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
505 q->tone_level[ch][sb][i] = 0;
507 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
516 * Related to synthesis filter
517 * Called by process_subpacket_11
518 * c is built with data from subpacket 11
519 * Most of this function is used only if superblock_type_2_3 == 0,
520 * never seen it in samples.
522 * @param tone_level_idx
523 * @param tone_level_idx_temp
524 * @param coding_method q->coding_method[0][0][0]
525 * @param nb_channels number of channels
526 * @param c coming from subpacket 11, passed as 8*c
527 * @param superblocktype_2_3 flag based on superblock packet type
528 * @param cm_table_select q->cm_table_select
530 static void fill_coding_method_array(sb_int8_array tone_level_idx,
531 sb_int8_array tone_level_idx_temp,
532 sb_int8_array coding_method,
534 int c, int superblocktype_2_3,
538 int tmp, acc, esp_40, comp;
539 int add1, add2, add3, add4;
542 if (!superblocktype_2_3) {
543 /* This case is untested, no samples available */
544 avpriv_request_sample(NULL, "!superblocktype_2_3");
546 for (ch = 0; ch < nb_channels; ch++) {
547 for (sb = 0; sb < 30; sb++) {
548 for (j = 1; j < 63; j++) { // The loop only iterates to 63 so the code doesn't overflow the buffer
549 add1 = tone_level_idx[ch][sb][j] - 10;
552 add2 = add3 = add4 = 0;
554 add2 = tone_level_idx[ch][sb - 2][j] + tone_level_idx_offset_table[sb][0] - 6;
559 add3 = tone_level_idx[ch][sb - 1][j] + tone_level_idx_offset_table[sb][1] - 6;
564 add4 = tone_level_idx[ch][sb + 1][j] + tone_level_idx_offset_table[sb][3] - 6;
568 tmp = tone_level_idx[ch][sb][j + 1] * 2 - add4 - add3 - add2 - add1;
571 tone_level_idx_temp[ch][sb][j + 1] = tmp & 0xff;
573 tone_level_idx_temp[ch][sb][0] = tone_level_idx_temp[ch][sb][1];
577 for (ch = 0; ch < nb_channels; ch++)
578 for (sb = 0; sb < 30; sb++)
579 for (j = 0; j < 64; j++)
580 acc += tone_level_idx_temp[ch][sb][j];
582 multres = 0x66666667LL * (acc * 10);
583 esp_40 = (multres >> 32) / 8 + ((multres & 0xffffffff) >> 31);
584 for (ch = 0; ch < nb_channels; ch++)
585 for (sb = 0; sb < 30; sb++)
586 for (j = 0; j < 64; j++) {
587 comp = tone_level_idx_temp[ch][sb][j]* esp_40 * 10;
590 comp /= 256; // signed shift
618 coding_method[ch][sb][j] = ((tmp & 0xfffa) + 30 )& 0xff;
620 for (sb = 0; sb < 30; sb++)
621 fix_coding_method_array(sb, nb_channels, coding_method);
622 for (ch = 0; ch < nb_channels; ch++)
623 for (sb = 0; sb < 30; sb++)
624 for (j = 0; j < 64; j++)
626 if (coding_method[ch][sb][j] < 10)
627 coding_method[ch][sb][j] = 10;
630 if (coding_method[ch][sb][j] < 16)
631 coding_method[ch][sb][j] = 16;
633 if (coding_method[ch][sb][j] < 30)
634 coding_method[ch][sb][j] = 30;
637 } else { // superblocktype_2_3 != 0
638 for (ch = 0; ch < nb_channels; ch++)
639 for (sb = 0; sb < 30; sb++)
640 for (j = 0; j < 64; j++)
641 coding_method[ch][sb][j] = coding_method_table[cm_table_select][sb];
646 * Called by process_subpacket_11 to process more data from subpacket 11
648 * Called by process_subpacket_12 to process data from subpacket 12 with
652 * @param gb bitreader context
653 * @param length packet length in bits
654 * @param sb_min lower subband processed (sb_min included)
655 * @param sb_max higher subband processed (sb_max excluded)
657 static int synthfilt_build_sb_samples(QDM2Context *q, GetBitContext *gb,
658 int length, int sb_min, int sb_max)
660 int sb, j, k, n, ch, run, channels;
661 int joined_stereo, zero_encoding;
663 float type34_div = 0;
664 float type34_predictor;
666 int sign_bits[16] = {0};
669 // If no data use noise
670 for (sb=sb_min; sb < sb_max; sb++)
671 build_sb_samples_from_noise(q, sb);
676 for (sb = sb_min; sb < sb_max; sb++) {
677 channels = q->nb_channels;
679 if (q->nb_channels <= 1 || sb < 12)
684 joined_stereo = (get_bits_left(gb) >= 1) ? get_bits1(gb) : 0;
687 if (get_bits_left(gb) >= 16)
688 for (j = 0; j < 16; j++)
689 sign_bits[j] = get_bits1(gb);
691 for (j = 0; j < 64; j++)
692 if (q->coding_method[1][sb][j] > q->coding_method[0][sb][j])
693 q->coding_method[0][sb][j] = q->coding_method[1][sb][j];
695 if (fix_coding_method_array(sb, q->nb_channels,
697 av_log(NULL, AV_LOG_ERROR, "coding method invalid\n");
698 build_sb_samples_from_noise(q, sb);
704 for (ch = 0; ch < channels; ch++) {
705 FIX_NOISE_IDX(q->noise_idx);
706 zero_encoding = (get_bits_left(gb) >= 1) ? get_bits1(gb) : 0;
707 type34_predictor = 0.0;
710 for (j = 0; j < 128; ) {
711 switch (q->coding_method[ch][sb][j / 2]) {
713 if (get_bits_left(gb) >= 10) {
715 for (k = 0; k < 5; k++) {
716 if ((j + 2 * k) >= 128)
718 samples[2 * k] = get_bits1(gb) ? dequant_1bit[joined_stereo][2 * get_bits1(gb)] : 0;
723 av_log(NULL, AV_LOG_ERROR, "Invalid 8bit codeword\n");
724 return AVERROR_INVALIDDATA;
727 for (k = 0; k < 5; k++)
728 samples[2 * k] = dequant_1bit[joined_stereo][random_dequant_index[n][k]];
730 for (k = 0; k < 5; k++)
731 samples[2 * k + 1] = SB_DITHERING_NOISE(sb,q->noise_idx);
733 for (k = 0; k < 10; k++)
734 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
740 if (get_bits_left(gb) >= 1) {
745 f -= noise_samples[((sb + 1) * (j +5 * ch + 1)) & 127] * 9.0 / 40.0;
748 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
754 if (get_bits_left(gb) >= 10) {
756 for (k = 0; k < 5; k++) {
759 samples[k] = (get_bits1(gb) == 0) ? 0 : dequant_1bit[joined_stereo][2 * get_bits1(gb)];
762 n = get_bits (gb, 8);
764 av_log(NULL, AV_LOG_ERROR, "Invalid 8bit codeword\n");
765 return AVERROR_INVALIDDATA;
768 for (k = 0; k < 5; k++)
769 samples[k] = dequant_1bit[joined_stereo][random_dequant_index[n][k]];
772 for (k = 0; k < 5; k++)
773 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
779 if (get_bits_left(gb) >= 7) {
782 av_log(NULL, AV_LOG_ERROR, "Invalid 7bit codeword\n");
783 return AVERROR_INVALIDDATA;
786 for (k = 0; k < 3; k++)
787 samples[k] = (random_dequant_type24[n][k] - 2.0) * 0.5;
789 for (k = 0; k < 3; k++)
790 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
796 if (get_bits_left(gb) >= 4) {
797 unsigned index = qdm2_get_vlc(gb, &vlc_tab_type30, 0, 1);
798 if (index >= FF_ARRAY_ELEMS(type30_dequant)) {
799 av_log(NULL, AV_LOG_ERROR, "index %d out of type30_dequant array\n", index);
800 return AVERROR_INVALIDDATA;
802 samples[0] = type30_dequant[index];
804 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
810 if (get_bits_left(gb) >= 7) {
812 type34_div = (float)(1 << get_bits(gb, 2));
813 samples[0] = ((float)get_bits(gb, 5) - 16.0) / 15.0;
814 type34_predictor = samples[0];
817 unsigned index = qdm2_get_vlc(gb, &vlc_tab_type34, 0, 1);
818 if (index >= FF_ARRAY_ELEMS(type34_delta)) {
819 av_log(NULL, AV_LOG_ERROR, "index %d out of type34_delta array\n", index);
820 return AVERROR_INVALIDDATA;
822 samples[0] = type34_delta[index] / type34_div + type34_predictor;
823 type34_predictor = samples[0];
826 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
832 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
838 for (k = 0; k < run && j + k < 128; k++) {
839 q->sb_samples[0][j + k][sb] =
840 q->tone_level[0][sb][(j + k) / 2] * samples[k];
841 if (q->nb_channels == 2) {
842 if (sign_bits[(j + k) / 8])
843 q->sb_samples[1][j + k][sb] =
844 q->tone_level[1][sb][(j + k) / 2] * -samples[k];
846 q->sb_samples[1][j + k][sb] =
847 q->tone_level[1][sb][(j + k) / 2] * samples[k];
851 for (k = 0; k < run; k++)
853 q->sb_samples[ch][j + k][sb] = q->tone_level[ch][sb][(j + k)/2] * samples[k];
864 * Init the first element of a channel in quantized_coeffs with data
865 * from packet 10 (quantized_coeffs[ch][0]).
866 * This is similar to process_subpacket_9, but for a single channel
867 * and for element [0]
868 * same VLC tables as process_subpacket_9 are used.
870 * @param quantized_coeffs pointer to quantized_coeffs[ch][0]
871 * @param gb bitreader context
873 static int init_quantized_coeffs_elem0(int8_t *quantized_coeffs,
876 int i, k, run, level, diff;
878 if (get_bits_left(gb) < 16)
880 level = qdm2_get_vlc(gb, &vlc_tab_level, 0, 2);
882 quantized_coeffs[0] = level;
884 for (i = 0; i < 7; ) {
885 if (get_bits_left(gb) < 16)
887 run = qdm2_get_vlc(gb, &vlc_tab_run, 0, 1) + 1;
892 if (get_bits_left(gb) < 16)
894 diff = qdm2_get_se_vlc(&vlc_tab_diff, gb, 2);
896 for (k = 1; k <= run; k++)
897 quantized_coeffs[i + k] = (level + ((k * diff) / run));
906 * Related to synthesis filter, process data from packet 10
907 * Init part of quantized_coeffs via function init_quantized_coeffs_elem0
908 * Init tone_level_idx_hi1, tone_level_idx_hi2, tone_level_idx_mid with
909 * data from packet 10
912 * @param gb bitreader context
914 static void init_tone_level_dequantization(QDM2Context *q, GetBitContext *gb)
918 for (ch = 0; ch < q->nb_channels; ch++) {
919 init_quantized_coeffs_elem0(q->quantized_coeffs[ch][0], gb);
921 if (get_bits_left(gb) < 16) {
922 memset(q->quantized_coeffs[ch][0], 0, 8);
927 n = q->sub_sampling + 1;
929 for (sb = 0; sb < n; sb++)
930 for (ch = 0; ch < q->nb_channels; ch++)
931 for (j = 0; j < 8; j++) {
932 if (get_bits_left(gb) < 1)
935 for (k=0; k < 8; k++) {
936 if (get_bits_left(gb) < 16)
938 q->tone_level_idx_hi1[ch][sb][j][k] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_hi1, 0, 2);
941 for (k=0; k < 8; k++)
942 q->tone_level_idx_hi1[ch][sb][j][k] = 0;
946 n = QDM2_SB_USED(q->sub_sampling) - 4;
948 for (sb = 0; sb < n; sb++)
949 for (ch = 0; ch < q->nb_channels; ch++) {
950 if (get_bits_left(gb) < 16)
952 q->tone_level_idx_hi2[ch][sb] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_hi2, 0, 2);
954 q->tone_level_idx_hi2[ch][sb] -= 16;
956 for (j = 0; j < 8; j++)
957 q->tone_level_idx_mid[ch][sb][j] = -16;
960 n = QDM2_SB_USED(q->sub_sampling) - 5;
962 for (sb = 0; sb < n; sb++)
963 for (ch = 0; ch < q->nb_channels; ch++)
964 for (j = 0; j < 8; j++) {
965 if (get_bits_left(gb) < 16)
967 q->tone_level_idx_mid[ch][sb][j] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_mid, 0, 2) - 32;
972 * Process subpacket 9, init quantized_coeffs with data from it
975 * @param node pointer to node with packet
977 static int process_subpacket_9(QDM2Context *q, QDM2SubPNode *node)
980 int i, j, k, n, ch, run, level, diff;
982 init_get_bits(&gb, node->packet->data, node->packet->size * 8);
984 n = coeff_per_sb_for_avg[q->coeff_per_sb_select][QDM2_SB_USED(q->sub_sampling) - 1] + 1;
986 for (i = 1; i < n; i++)
987 for (ch = 0; ch < q->nb_channels; ch++) {
988 level = qdm2_get_vlc(&gb, &vlc_tab_level, 0, 2);
989 q->quantized_coeffs[ch][i][0] = level;
991 for (j = 0; j < (8 - 1); ) {
992 run = qdm2_get_vlc(&gb, &vlc_tab_run, 0, 1) + 1;
993 diff = qdm2_get_se_vlc(&vlc_tab_diff, &gb, 2);
998 for (k = 1; k <= run; k++)
999 q->quantized_coeffs[ch][i][j + k] = (level + ((k * diff) / run));
1006 for (ch = 0; ch < q->nb_channels; ch++)
1007 for (i = 0; i < 8; i++)
1008 q->quantized_coeffs[ch][0][i] = 0;
1014 * Process subpacket 10 if not null, else
1017 * @param node pointer to node with packet
1019 static void process_subpacket_10(QDM2Context *q, QDM2SubPNode *node)
1024 init_get_bits(&gb, node->packet->data, node->packet->size * 8);
1025 init_tone_level_dequantization(q, &gb);
1026 fill_tone_level_array(q, 1);
1028 fill_tone_level_array(q, 0);
1033 * Process subpacket 11
1036 * @param node pointer to node with packet
1038 static void process_subpacket_11(QDM2Context *q, QDM2SubPNode *node)
1044 length = node->packet->size * 8;
1045 init_get_bits(&gb, node->packet->data, length);
1049 int c = get_bits(&gb, 13);
1052 fill_coding_method_array(q->tone_level_idx,
1053 q->tone_level_idx_temp, q->coding_method,
1054 q->nb_channels, 8 * c,
1055 q->superblocktype_2_3, q->cm_table_select);
1058 synthfilt_build_sb_samples(q, &gb, length, 0, 8);
1062 * Process subpacket 12
1065 * @param node pointer to node with packet
1067 static void process_subpacket_12(QDM2Context *q, QDM2SubPNode *node)
1073 length = node->packet->size * 8;
1074 init_get_bits(&gb, node->packet->data, length);
1077 synthfilt_build_sb_samples(q, &gb, length, 8, QDM2_SB_USED(q->sub_sampling));
1081 * Process new subpackets for synthesis filter
1084 * @param list list with synthesis filter packets (list D)
1086 static void process_synthesis_subpackets(QDM2Context *q, QDM2SubPNode *list)
1088 QDM2SubPNode *nodes[4];
1090 nodes[0] = qdm2_search_subpacket_type_in_list(list, 9);
1092 process_subpacket_9(q, nodes[0]);
1094 nodes[1] = qdm2_search_subpacket_type_in_list(list, 10);
1096 process_subpacket_10(q, nodes[1]);
1098 process_subpacket_10(q, NULL);
1100 nodes[2] = qdm2_search_subpacket_type_in_list(list, 11);
1101 if (nodes[0] && nodes[1] && nodes[2])
1102 process_subpacket_11(q, nodes[2]);
1104 process_subpacket_11(q, NULL);
1106 nodes[3] = qdm2_search_subpacket_type_in_list(list, 12);
1107 if (nodes[0] && nodes[1] && nodes[3])
1108 process_subpacket_12(q, nodes[3]);
1110 process_subpacket_12(q, NULL);
1114 * Decode superblock, fill packet lists.
1118 static void qdm2_decode_super_block(QDM2Context *q)
1121 QDM2SubPacket header, *packet;
1122 int i, packet_bytes, sub_packet_size, sub_packets_D;
1123 unsigned int next_index = 0;
1125 memset(q->tone_level_idx_hi1, 0, sizeof(q->tone_level_idx_hi1));
1126 memset(q->tone_level_idx_mid, 0, sizeof(q->tone_level_idx_mid));
1127 memset(q->tone_level_idx_hi2, 0, sizeof(q->tone_level_idx_hi2));
1129 q->sub_packets_B = 0;
1132 average_quantized_coeffs(q); // average elements in quantized_coeffs[max_ch][10][8]
1134 init_get_bits(&gb, q->compressed_data, q->compressed_size * 8);
1135 qdm2_decode_sub_packet_header(&gb, &header);
1137 if (header.type < 2 || header.type >= 8) {
1139 av_log(NULL, AV_LOG_ERROR, "bad superblock type\n");
1143 q->superblocktype_2_3 = (header.type == 2 || header.type == 3);
1144 packet_bytes = (q->compressed_size - get_bits_count(&gb) / 8);
1146 init_get_bits(&gb, header.data, header.size * 8);
1148 if (header.type == 2 || header.type == 4 || header.type == 5) {
1149 int csum = 257 * get_bits(&gb, 8);
1150 csum += 2 * get_bits(&gb, 8);
1152 csum = qdm2_packet_checksum(q->compressed_data, q->checksum_size, csum);
1156 av_log(NULL, AV_LOG_ERROR, "bad packet checksum\n");
1161 q->sub_packet_list_B[0].packet = NULL;
1162 q->sub_packet_list_D[0].packet = NULL;
1164 for (i = 0; i < 6; i++)
1165 if (--q->fft_level_exp[i] < 0)
1166 q->fft_level_exp[i] = 0;
1168 for (i = 0; packet_bytes > 0; i++) {
1171 if (i >= FF_ARRAY_ELEMS(q->sub_packet_list_A)) {
1172 SAMPLES_NEEDED_2("too many packet bytes");
1176 q->sub_packet_list_A[i].next = NULL;
1179 q->sub_packet_list_A[i - 1].next = &q->sub_packet_list_A[i];
1181 /* seek to next block */
1182 init_get_bits(&gb, header.data, header.size * 8);
1183 skip_bits(&gb, next_index * 8);
1185 if (next_index >= header.size)
1189 /* decode subpacket */
1190 packet = &q->sub_packets[i];
1191 qdm2_decode_sub_packet_header(&gb, packet);
1192 next_index = packet->size + get_bits_count(&gb) / 8;
1193 sub_packet_size = ((packet->size > 0xff) ? 1 : 0) + packet->size + 2;
1195 if (packet->type == 0)
1198 if (sub_packet_size > packet_bytes) {
1199 if (packet->type != 10 && packet->type != 11 && packet->type != 12)
1201 packet->size += packet_bytes - sub_packet_size;
1204 packet_bytes -= sub_packet_size;
1206 /* add subpacket to 'all subpackets' list */
1207 q->sub_packet_list_A[i].packet = packet;
1209 /* add subpacket to related list */
1210 if (packet->type == 8) {
1211 SAMPLES_NEEDED_2("packet type 8");
1213 } else if (packet->type >= 9 && packet->type <= 12) {
1214 /* packets for MPEG Audio like Synthesis Filter */
1215 QDM2_LIST_ADD(q->sub_packet_list_D, sub_packets_D, packet);
1216 } else if (packet->type == 13) {
1217 for (j = 0; j < 6; j++)
1218 q->fft_level_exp[j] = get_bits(&gb, 6);
1219 } else if (packet->type == 14) {
1220 for (j = 0; j < 6; j++)
1221 q->fft_level_exp[j] = qdm2_get_vlc(&gb, &fft_level_exp_vlc, 0, 2);
1222 } else if (packet->type == 15) {
1223 SAMPLES_NEEDED_2("packet type 15")
1225 } else if (packet->type >= 16 && packet->type < 48 &&
1226 !fft_subpackets[packet->type - 16]) {
1227 /* packets for FFT */
1228 QDM2_LIST_ADD(q->sub_packet_list_B, q->sub_packets_B, packet);
1230 } // Packet bytes loop
1232 if (q->sub_packet_list_D[0].packet) {
1233 process_synthesis_subpackets(q, q->sub_packet_list_D);
1234 q->do_synth_filter = 1;
1235 } else if (q->do_synth_filter) {
1236 process_subpacket_10(q, NULL);
1237 process_subpacket_11(q, NULL);
1238 process_subpacket_12(q, NULL);
1242 static void qdm2_fft_init_coefficient(QDM2Context *q, int sub_packet,
1243 int offset, int duration, int channel,
1246 if (q->fft_coefs_min_index[duration] < 0)
1247 q->fft_coefs_min_index[duration] = q->fft_coefs_index;
1249 q->fft_coefs[q->fft_coefs_index].sub_packet =
1250 ((sub_packet >= 16) ? (sub_packet - 16) : sub_packet);
1251 q->fft_coefs[q->fft_coefs_index].channel = channel;
1252 q->fft_coefs[q->fft_coefs_index].offset = offset;
1253 q->fft_coefs[q->fft_coefs_index].exp = exp;
1254 q->fft_coefs[q->fft_coefs_index].phase = phase;
1255 q->fft_coefs_index++;
1258 static void qdm2_fft_decode_tones(QDM2Context *q, int duration,
1259 GetBitContext *gb, int b)
1261 int channel, stereo, phase, exp;
1262 int local_int_4, local_int_8, stereo_phase, local_int_10;
1263 int local_int_14, stereo_exp, local_int_20, local_int_28;
1269 local_int_8 = (4 - duration);
1270 local_int_10 = 1 << (q->group_order - duration - 1);
1273 while (get_bits_left(gb)>0) {
1274 if (q->superblocktype_2_3) {
1275 while ((n = qdm2_get_vlc(gb, &vlc_tab_fft_tone_offset[local_int_8], 1, 2)) < 2) {
1276 if (get_bits_left(gb)<0) {
1277 if(local_int_4 < q->group_size)
1278 av_log(NULL, AV_LOG_ERROR, "overread in qdm2_fft_decode_tones()\n");
1283 local_int_4 += local_int_10;
1284 local_int_28 += (1 << local_int_8);
1286 local_int_4 += 8 * local_int_10;
1287 local_int_28 += (8 << local_int_8);
1292 if (local_int_10 <= 2) {
1293 av_log(NULL, AV_LOG_ERROR, "qdm2_fft_decode_tones() stuck\n");
1296 offset += qdm2_get_vlc(gb, &vlc_tab_fft_tone_offset[local_int_8], 1, 2);
1297 while (offset >= (local_int_10 - 1)) {
1298 offset += (1 - (local_int_10 - 1));
1299 local_int_4 += local_int_10;
1300 local_int_28 += (1 << local_int_8);
1304 if (local_int_4 >= q->group_size)
1307 local_int_14 = (offset >> local_int_8);
1308 if (local_int_14 >= FF_ARRAY_ELEMS(fft_level_index_table))
1311 if (q->nb_channels > 1) {
1312 channel = get_bits1(gb);
1313 stereo = get_bits1(gb);
1319 exp = qdm2_get_vlc(gb, (b ? &fft_level_exp_vlc : &fft_level_exp_alt_vlc), 0, 2);
1320 exp += q->fft_level_exp[fft_level_index_table[local_int_14]];
1321 exp = (exp < 0) ? 0 : exp;
1323 phase = get_bits(gb, 3);
1328 stereo_exp = (exp - qdm2_get_vlc(gb, &fft_stereo_exp_vlc, 0, 1));
1329 stereo_phase = (phase - qdm2_get_vlc(gb, &fft_stereo_phase_vlc, 0, 1));
1330 if (stereo_phase < 0)
1334 if (q->frequency_range > (local_int_14 + 1)) {
1335 int sub_packet = (local_int_20 + local_int_28);
1337 if (q->fft_coefs_index + stereo >= FF_ARRAY_ELEMS(q->fft_coefs))
1340 qdm2_fft_init_coefficient(q, sub_packet, offset, duration,
1341 channel, exp, phase);
1343 qdm2_fft_init_coefficient(q, sub_packet, offset, duration,
1345 stereo_exp, stereo_phase);
1351 static void qdm2_decode_fft_packets(QDM2Context *q)
1353 int i, j, min, max, value, type, unknown_flag;
1356 if (!q->sub_packet_list_B[0].packet)
1359 /* reset minimum indexes for FFT coefficients */
1360 q->fft_coefs_index = 0;
1361 for (i = 0; i < 5; i++)
1362 q->fft_coefs_min_index[i] = -1;
1364 /* process subpackets ordered by type, largest type first */
1365 for (i = 0, max = 256; i < q->sub_packets_B; i++) {
1366 QDM2SubPacket *packet = NULL;
1368 /* find subpacket with largest type less than max */
1369 for (j = 0, min = 0; j < q->sub_packets_B; j++) {
1370 value = q->sub_packet_list_B[j].packet->type;
1371 if (value > min && value < max) {
1373 packet = q->sub_packet_list_B[j].packet;
1379 /* check for errors (?) */
1384 (packet->type < 16 || packet->type >= 48 ||
1385 fft_subpackets[packet->type - 16]))
1388 /* decode FFT tones */
1389 init_get_bits(&gb, packet->data, packet->size * 8);
1391 if (packet->type >= 32 && packet->type < 48 && !fft_subpackets[packet->type - 16])
1396 type = packet->type;
1398 if ((type >= 17 && type < 24) || (type >= 33 && type < 40)) {
1399 int duration = q->sub_sampling + 5 - (type & 15);
1401 if (duration >= 0 && duration < 4)
1402 qdm2_fft_decode_tones(q, duration, &gb, unknown_flag);
1403 } else if (type == 31) {
1404 for (j = 0; j < 4; j++)
1405 qdm2_fft_decode_tones(q, j, &gb, unknown_flag);
1406 } else if (type == 46) {
1407 for (j = 0; j < 6; j++)
1408 q->fft_level_exp[j] = get_bits(&gb, 6);
1409 for (j = 0; j < 4; j++)
1410 qdm2_fft_decode_tones(q, j, &gb, unknown_flag);
1412 } // Loop on B packets
1414 /* calculate maximum indexes for FFT coefficients */
1415 for (i = 0, j = -1; i < 5; i++)
1416 if (q->fft_coefs_min_index[i] >= 0) {
1418 q->fft_coefs_max_index[j] = q->fft_coefs_min_index[i];
1422 q->fft_coefs_max_index[j] = q->fft_coefs_index;
1425 static void qdm2_fft_generate_tone(QDM2Context *q, FFTTone *tone)
1430 const double iscale = 2.0 * M_PI / 512.0;
1432 tone->phase += tone->phase_shift;
1434 /* calculate current level (maximum amplitude) of tone */
1435 level = fft_tone_envelope_table[tone->duration][tone->time_index] * tone->level;
1436 c.im = level * sin(tone->phase * iscale);
1437 c.re = level * cos(tone->phase * iscale);
1439 /* generate FFT coefficients for tone */
1440 if (tone->duration >= 3 || tone->cutoff >= 3) {
1441 tone->complex[0].im += c.im;
1442 tone->complex[0].re += c.re;
1443 tone->complex[1].im -= c.im;
1444 tone->complex[1].re -= c.re;
1446 f[1] = -tone->table[4];
1447 f[0] = tone->table[3] - tone->table[0];
1448 f[2] = 1.0 - tone->table[2] - tone->table[3];
1449 f[3] = tone->table[1] + tone->table[4] - 1.0;
1450 f[4] = tone->table[0] - tone->table[1];
1451 f[5] = tone->table[2];
1452 for (i = 0; i < 2; i++) {
1453 tone->complex[fft_cutoff_index_table[tone->cutoff][i]].re +=
1455 tone->complex[fft_cutoff_index_table[tone->cutoff][i]].im +=
1456 c.im * ((tone->cutoff <= i) ? -f[i] : f[i]);
1458 for (i = 0; i < 4; i++) {
1459 tone->complex[i].re += c.re * f[i + 2];
1460 tone->complex[i].im += c.im * f[i + 2];
1464 /* copy the tone if it has not yet died out */
1465 if (++tone->time_index < ((1 << (5 - tone->duration)) - 1)) {
1466 memcpy(&q->fft_tones[q->fft_tone_end], tone, sizeof(FFTTone));
1467 q->fft_tone_end = (q->fft_tone_end + 1) % 1000;
1471 static void qdm2_fft_tone_synthesizer(QDM2Context *q, int sub_packet)
1474 const double iscale = 0.25 * M_PI;
1476 for (ch = 0; ch < q->channels; ch++) {
1477 memset(q->fft.complex[ch], 0, q->fft_size * sizeof(QDM2Complex));
1481 /* apply FFT tones with duration 4 (1 FFT period) */
1482 if (q->fft_coefs_min_index[4] >= 0)
1483 for (i = q->fft_coefs_min_index[4]; i < q->fft_coefs_max_index[4]; i++) {
1487 if (q->fft_coefs[i].sub_packet != sub_packet)
1490 ch = (q->channels == 1) ? 0 : q->fft_coefs[i].channel;
1491 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];
1493 c.re = level * cos(q->fft_coefs[i].phase * iscale);
1494 c.im = level * sin(q->fft_coefs[i].phase * iscale);
1495 q->fft.complex[ch][q->fft_coefs[i].offset + 0].re += c.re;
1496 q->fft.complex[ch][q->fft_coefs[i].offset + 0].im += c.im;
1497 q->fft.complex[ch][q->fft_coefs[i].offset + 1].re -= c.re;
1498 q->fft.complex[ch][q->fft_coefs[i].offset + 1].im -= c.im;
1501 /* generate existing FFT tones */
1502 for (i = q->fft_tone_end; i != q->fft_tone_start; ) {
1503 qdm2_fft_generate_tone(q, &q->fft_tones[q->fft_tone_start]);
1504 q->fft_tone_start = (q->fft_tone_start + 1) % 1000;
1507 /* create and generate new FFT tones with duration 0 (long) to 3 (short) */
1508 for (i = 0; i < 4; i++)
1509 if (q->fft_coefs_min_index[i] >= 0) {
1510 for (j = q->fft_coefs_min_index[i]; j < q->fft_coefs_max_index[i]; j++) {
1514 if (q->fft_coefs[j].sub_packet != sub_packet)
1518 offset = q->fft_coefs[j].offset >> four_i;
1519 ch = (q->channels == 1) ? 0 : q->fft_coefs[j].channel;
1521 if (offset < q->frequency_range) {
1523 tone.cutoff = offset;
1525 tone.cutoff = (offset >= 60) ? 3 : 2;
1527 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];
1528 tone.complex = &q->fft.complex[ch][offset];
1529 tone.table = fft_tone_sample_table[i][q->fft_coefs[j].offset - (offset << four_i)];
1530 tone.phase = 64 * q->fft_coefs[j].phase - (offset << 8) - 128;
1531 tone.phase_shift = (2 * q->fft_coefs[j].offset + 1) << (7 - four_i);
1533 tone.time_index = 0;
1535 qdm2_fft_generate_tone(q, &tone);
1538 q->fft_coefs_min_index[i] = j;
1542 static void qdm2_calculate_fft(QDM2Context *q, int channel, int sub_packet)
1544 const float gain = (q->channels == 1 && q->nb_channels == 2) ? 0.5f : 1.0f;
1545 float *out = q->output_buffer + channel;
1547 q->fft.complex[channel][0].re *= 2.0f;
1548 q->fft.complex[channel][0].im = 0.0f;
1549 q->rdft_ctx.rdft_calc(&q->rdft_ctx, (FFTSample *)q->fft.complex[channel]);
1550 /* add samples to output buffer */
1551 for (i = 0; i < FFALIGN(q->fft_size, 8); i++) {
1552 out[0] += q->fft.complex[channel][i].re * gain;
1553 out[q->channels] += q->fft.complex[channel][i].im * gain;
1554 out += 2 * q->channels;
1560 * @param index subpacket number
1562 static void qdm2_synthesis_filter(QDM2Context *q, int index)
1564 int i, k, ch, sb_used, sub_sampling, dither_state = 0;
1566 /* copy sb_samples */
1567 sb_used = QDM2_SB_USED(q->sub_sampling);
1569 for (ch = 0; ch < q->channels; ch++)
1570 for (i = 0; i < 8; i++)
1571 for (k = sb_used; k < SBLIMIT; k++)
1572 q->sb_samples[ch][(8 * index) + i][k] = 0;
1574 for (ch = 0; ch < q->nb_channels; ch++) {
1575 float *samples_ptr = q->samples + ch;
1577 for (i = 0; i < 8; i++) {
1578 ff_mpa_synth_filter_float(&q->mpadsp,
1579 q->synth_buf[ch], &(q->synth_buf_offset[ch]),
1580 ff_mpa_synth_window_float, &dither_state,
1581 samples_ptr, q->nb_channels,
1582 q->sb_samples[ch][(8 * index) + i]);
1583 samples_ptr += 32 * q->nb_channels;
1587 /* add samples to output buffer */
1588 sub_sampling = (4 >> q->sub_sampling);
1590 for (ch = 0; ch < q->channels; ch++)
1591 for (i = 0; i < q->frame_size; i++)
1592 q->output_buffer[q->channels * i + ch] += (1 << 23) * q->samples[q->nb_channels * sub_sampling * i + ch];
1596 * Init static data (does not depend on specific file)
1598 static av_cold void qdm2_init_static_data(void) {
1605 softclip_table_init();
1607 init_noise_samples();
1609 ff_mpa_synth_init_float();
1615 * Init parameters from codec extradata
1617 static av_cold int qdm2_decode_init(AVCodecContext *avctx)
1619 QDM2Context *s = avctx->priv_data;
1620 int tmp_val, tmp, size;
1623 qdm2_init_static_data();
1625 /* extradata parsing
1634 32 size (including this field)
1636 32 type (=QDM2 or QDMC)
1638 32 size (including this field, in bytes)
1639 32 tag (=QDCA) // maybe mandatory parameters
1642 32 samplerate (=44100)
1644 32 block size (=4096)
1645 32 frame size (=256) (for one channel)
1646 32 packet size (=1300)
1648 32 size (including this field, in bytes)
1649 32 tag (=QDCP) // maybe some tuneable parameters
1659 if (!avctx->extradata || (avctx->extradata_size < 48)) {
1660 av_log(avctx, AV_LOG_ERROR, "extradata missing or truncated\n");
1661 return AVERROR_INVALIDDATA;
1664 bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
1666 while (bytestream2_get_bytes_left(&gb) > 8) {
1667 if (bytestream2_peek_be64(&gb) == (((uint64_t)MKBETAG('f','r','m','a') << 32) |
1668 (uint64_t)MKBETAG('Q','D','M','2')))
1670 bytestream2_skip(&gb, 1);
1673 if (bytestream2_get_bytes_left(&gb) < 12) {
1674 av_log(avctx, AV_LOG_ERROR, "not enough extradata (%i)\n",
1675 bytestream2_get_bytes_left(&gb));
1676 return AVERROR_INVALIDDATA;
1679 bytestream2_skip(&gb, 8);
1680 size = bytestream2_get_be32(&gb);
1682 if (size > bytestream2_get_bytes_left(&gb)) {
1683 av_log(avctx, AV_LOG_ERROR, "extradata size too small, %i < %i\n",
1684 bytestream2_get_bytes_left(&gb), size);
1685 return AVERROR_INVALIDDATA;
1688 av_log(avctx, AV_LOG_DEBUG, "size: %d\n", size);
1689 if (bytestream2_get_be32(&gb) != MKBETAG('Q','D','C','A')) {
1690 av_log(avctx, AV_LOG_ERROR, "invalid extradata, expecting QDCA\n");
1691 return AVERROR_INVALIDDATA;
1694 bytestream2_skip(&gb, 4);
1696 avctx->channels = s->nb_channels = s->channels = bytestream2_get_be32(&gb);
1697 if (s->channels <= 0 || s->channels > MPA_MAX_CHANNELS) {
1698 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1699 return AVERROR_INVALIDDATA;
1701 avctx->channel_layout = avctx->channels == 2 ? AV_CH_LAYOUT_STEREO :
1704 avctx->sample_rate = bytestream2_get_be32(&gb);
1705 avctx->bit_rate = bytestream2_get_be32(&gb);
1706 s->group_size = bytestream2_get_be32(&gb);
1707 s->fft_size = bytestream2_get_be32(&gb);
1708 s->checksum_size = bytestream2_get_be32(&gb);
1709 if (s->checksum_size >= 1U << 28 || s->checksum_size <= 1) {
1710 av_log(avctx, AV_LOG_ERROR, "data block size invalid (%u)\n", s->checksum_size);
1711 return AVERROR_INVALIDDATA;
1714 s->fft_order = av_log2(s->fft_size) + 1;
1716 // Fail on unknown fft order
1717 if ((s->fft_order < 7) || (s->fft_order > 9)) {
1718 avpriv_request_sample(avctx, "Unknown FFT order %d", s->fft_order);
1719 return AVERROR_PATCHWELCOME;
1722 // something like max decodable tones
1723 s->group_order = av_log2(s->group_size) + 1;
1724 s->frame_size = s->group_size / 16; // 16 iterations per super block
1726 if (s->frame_size > QDM2_MAX_FRAME_SIZE)
1727 return AVERROR_INVALIDDATA;
1729 s->sub_sampling = s->fft_order - 7;
1730 s->frequency_range = 255 / (1 << (2 - s->sub_sampling));
1732 if (s->frame_size * 4 >> s->sub_sampling > MPA_FRAME_SIZE) {
1733 avpriv_request_sample(avctx, "large frames");
1734 return AVERROR_PATCHWELCOME;
1737 switch ((s->sub_sampling * 2 + s->channels - 1)) {
1738 case 0: tmp = 40; break;
1739 case 1: tmp = 48; break;
1740 case 2: tmp = 56; break;
1741 case 3: tmp = 72; break;
1742 case 4: tmp = 80; break;
1743 case 5: tmp = 100;break;
1744 default: tmp=s->sub_sampling; break;
1747 if ((tmp * 1000) < avctx->bit_rate) tmp_val = 1;
1748 if ((tmp * 1440) < avctx->bit_rate) tmp_val = 2;
1749 if ((tmp * 1760) < avctx->bit_rate) tmp_val = 3;
1750 if ((tmp * 2240) < avctx->bit_rate) tmp_val = 4;
1751 s->cm_table_select = tmp_val;
1753 if (avctx->bit_rate <= 8000)
1754 s->coeff_per_sb_select = 0;
1755 else if (avctx->bit_rate < 16000)
1756 s->coeff_per_sb_select = 1;
1758 s->coeff_per_sb_select = 2;
1760 if (s->fft_size != (1 << (s->fft_order - 1))) {
1761 av_log(avctx, AV_LOG_ERROR, "FFT size %d not power of 2.\n", s->fft_size);
1762 return AVERROR_INVALIDDATA;
1765 ff_rdft_init(&s->rdft_ctx, s->fft_order, IDFT_C2R);
1766 ff_mpadsp_init(&s->mpadsp);
1768 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
1773 static av_cold int qdm2_decode_close(AVCodecContext *avctx)
1775 QDM2Context *s = avctx->priv_data;
1777 ff_rdft_end(&s->rdft_ctx);
1782 static int qdm2_decode(QDM2Context *q, const uint8_t *in, int16_t *out)
1785 const int frame_size = (q->frame_size * q->channels);
1787 if((unsigned)frame_size > FF_ARRAY_ELEMS(q->output_buffer)/2)
1790 /* select input buffer */
1791 q->compressed_data = in;
1792 q->compressed_size = q->checksum_size;
1794 /* copy old block, clear new block of output samples */
1795 memmove(q->output_buffer, &q->output_buffer[frame_size], frame_size * sizeof(float));
1796 memset(&q->output_buffer[frame_size], 0, frame_size * sizeof(float));
1798 /* decode block of QDM2 compressed data */
1799 if (q->sub_packet == 0) {
1800 q->has_errors = 0; // zero it for a new super block
1801 av_log(NULL,AV_LOG_DEBUG,"Superblock follows\n");
1802 qdm2_decode_super_block(q);
1805 /* parse subpackets */
1806 if (!q->has_errors) {
1807 if (q->sub_packet == 2)
1808 qdm2_decode_fft_packets(q);
1810 qdm2_fft_tone_synthesizer(q, q->sub_packet);
1813 /* sound synthesis stage 1 (FFT) */
1814 for (ch = 0; ch < q->channels; ch++) {
1815 qdm2_calculate_fft(q, ch, q->sub_packet);
1817 if (!q->has_errors && q->sub_packet_list_C[0].packet) {
1818 SAMPLES_NEEDED_2("has errors, and C list is not empty")
1823 /* sound synthesis stage 2 (MPEG audio like synthesis filter) */
1824 if (!q->has_errors && q->do_synth_filter)
1825 qdm2_synthesis_filter(q, q->sub_packet);
1827 q->sub_packet = (q->sub_packet + 1) % 16;
1829 /* clip and convert output float[] to 16-bit signed samples */
1830 for (i = 0; i < frame_size; i++) {
1831 int value = (int)q->output_buffer[i];
1833 if (value > SOFTCLIP_THRESHOLD)
1834 value = (value > HARDCLIP_THRESHOLD) ? 32767 : softclip_table[ value - SOFTCLIP_THRESHOLD];
1835 else if (value < -SOFTCLIP_THRESHOLD)
1836 value = (value < -HARDCLIP_THRESHOLD) ? -32767 : -softclip_table[-value - SOFTCLIP_THRESHOLD];
1844 static int qdm2_decode_frame(AVCodecContext *avctx, void *data,
1845 int *got_frame_ptr, AVPacket *avpkt)
1847 AVFrame *frame = data;
1848 const uint8_t *buf = avpkt->data;
1849 int buf_size = avpkt->size;
1850 QDM2Context *s = avctx->priv_data;
1856 if(buf_size < s->checksum_size)
1859 /* get output buffer */
1860 frame->nb_samples = 16 * s->frame_size;
1861 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1863 out = (int16_t *)frame->data[0];
1865 for (i = 0; i < 16; i++) {
1866 if ((ret = qdm2_decode(s, buf, out)) < 0)
1868 out += s->channels * s->frame_size;
1873 return s->checksum_size;
1876 AVCodec ff_qdm2_decoder = {
1878 .long_name = NULL_IF_CONFIG_SMALL("QDesign Music Codec 2"),
1879 .type = AVMEDIA_TYPE_AUDIO,
1880 .id = AV_CODEC_ID_QDM2,
1881 .priv_data_size = sizeof(QDM2Context),
1882 .init = qdm2_decode_init,
1883 .close = qdm2_decode_close,
1884 .decode = qdm2_decode_frame,
1885 .capabilities = AV_CODEC_CAP_DR1,