2 * Atrac 3 compatible decoder
3 * Copyright (c) 2006-2008 Maxim Poliakovski
4 * Copyright (c) 2006-2008 Benjamin Larsson
6 * This file is part of Libav.
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 * Atrac 3 compatible decoder.
26 * This decoder handles Sony's ATRAC3 data.
28 * Container formats used to store atrac 3 data:
29 * RealMedia (.rm), RIFF WAV (.wav, .at3), Sony OpenMG (.oma, .aa3).
31 * To use this decoder, a calling application must supply the extradata
32 * bytes provided in the containers above.
39 #include "libavutil/float_dsp.h"
41 #include "bytestream.h"
43 #include "fmtconvert.h"
47 #include "atrac3data.h"
49 #define JOINT_STEREO 0x12
52 #define SAMPLES_PER_FRAME 1024
55 typedef struct GainInfo {
61 typedef struct GainBlock {
65 typedef struct TonalComponent {
71 typedef struct ChannelUnit {
74 float prev_frame[SAMPLES_PER_FRAME];
76 TonalComponent components[64];
77 GainBlock gain_block[2];
79 DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
80 DECLARE_ALIGNED(32, float, imdct_buf)[SAMPLES_PER_FRAME];
82 float delay_buf1[46]; ///<qmf delay buffers
87 typedef struct ATRAC3Context {
95 int samples_per_frame;
100 /** joint-stereo related variables */
101 int matrix_coeff_index_prev[4];
102 int matrix_coeff_index_now[4];
103 int matrix_coeff_index_next[4];
104 int weighting_delay[6];
108 uint8_t *decoded_bytes_buffer;
109 float temp_buf[1070];
113 int scrambled_stream;
118 FmtConvertContext fmt_conv;
119 AVFloatDSPContext fdsp;
122 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
123 static VLC spectral_coeff_tab[7];
124 static float gain_tab1[16];
125 static float gain_tab2[31];
129 * Regular 512 points IMDCT without overlapping, with the exception of the
130 * swapping of odd bands caused by the reverse spectra of the QMF.
132 * @param odd_band 1 if the band is an odd band
134 static void imlt(ATRAC3Context *q, float *input, float *output, int odd_band)
140 * Reverse the odd bands before IMDCT, this is an effect of the QMF
141 * transform or it gives better compression to do it this way.
142 * FIXME: It should be possible to handle this in imdct_calc
143 * for that to happen a modification of the prerotation step of
144 * all SIMD code and C code is needed.
145 * Or fix the functions before so they generate a pre reversed spectrum.
147 for (i = 0; i < 128; i++)
148 FFSWAP(float, input[i], input[255 - i]);
151 q->mdct_ctx.imdct_calc(&q->mdct_ctx, output, input);
153 /* Perform windowing on the output. */
154 q->fdsp.vector_fmul(output, output, mdct_window, MDCT_SIZE);
158 * indata descrambling, only used for data coming from the rm container
160 static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes)
165 uint32_t *output = (uint32_t *)out;
167 off = (intptr_t)input & 3;
168 buf = (const uint32_t *)(input - off);
169 c = av_be2ne32((0x537F6103 >> (off * 8)) | (0x537F6103 << (32 - (off * 8))));
171 for (i = 0; i < bytes / 4; i++)
172 output[i] = c ^ buf[i];
175 av_log_ask_for_sample(NULL, "Offset of %d not handled.\n", off);
180 static av_cold int init_atrac3_transforms(ATRAC3Context *q)
182 float enc_window[256];
185 /* generate the mdct window, for details see
186 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
187 for (i = 0; i < 256; i++)
188 enc_window[i] = (sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0) * 0.5;
190 if (!mdct_window[0]) {
191 for (i = 0; i < 256; i++) {
192 mdct_window[i] = enc_window[i] /
193 (enc_window[ i] * enc_window[ i] +
194 enc_window[255 - i] * enc_window[255 - i]);
195 mdct_window[511 - i] = mdct_window[i];
199 /* initialize the MDCT transform */
200 return ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768);
203 static av_cold int atrac3_decode_close(AVCodecContext *avctx)
205 ATRAC3Context *q = avctx->priv_data;
208 av_free(q->decoded_bytes_buffer);
210 ff_mdct_end(&q->mdct_ctx);
218 * @param selector which table the output values are coded with
219 * @param coding_flag constant length coding or variable length coding
220 * @param mantissas mantissa output table
221 * @param num_codes number of values to get
223 static void read_quant_spectral_coeffs(GetBitContext *gb, int selector,
224 int coding_flag, int *mantissas,
227 int i, code, huff_symb;
232 if (coding_flag != 0) {
233 /* constant length coding (CLC) */
234 int num_bits = clc_length_tab[selector];
237 for (i = 0; i < num_codes; i++) {
239 code = get_sbits(gb, num_bits);
245 for (i = 0; i < num_codes; i++) {
247 code = get_bits(gb, num_bits); // num_bits is always 4 in this case
250 mantissas[i * 2 ] = mantissa_clc_tab[code >> 2];
251 mantissas[i * 2 + 1] = mantissa_clc_tab[code & 3];
255 /* variable length coding (VLC) */
257 for (i = 0; i < num_codes; i++) {
258 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector-1].table,
259 spectral_coeff_tab[selector-1].bits, 3);
261 code = huff_symb >> 1;
267 for (i = 0; i < num_codes; i++) {
268 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table,
269 spectral_coeff_tab[selector - 1].bits, 3);
270 mantissas[i * 2 ] = mantissa_vlc_tab[huff_symb * 2 ];
271 mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1];
278 * Restore the quantized band spectrum coefficients
280 * @return subband count, fix for broken specification/files
282 static int decode_spectrum(GetBitContext *gb, float *output)
284 int num_subbands, coding_mode, i, j, first, last, subband_size;
285 int subband_vlc_index[32], sf_index[32];
289 num_subbands = get_bits(gb, 5); // number of coded subbands
290 coding_mode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
292 /* get the VLC selector table for the subbands, 0 means not coded */
293 for (i = 0; i <= num_subbands; i++)
294 subband_vlc_index[i] = get_bits(gb, 3);
296 /* read the scale factor indexes from the stream */
297 for (i = 0; i <= num_subbands; i++) {
298 if (subband_vlc_index[i] != 0)
299 sf_index[i] = get_bits(gb, 6);
302 for (i = 0; i <= num_subbands; i++) {
303 first = subband_tab[i ];
304 last = subband_tab[i + 1];
306 subband_size = last - first;
308 if (subband_vlc_index[i] != 0) {
309 /* decode spectral coefficients for this subband */
310 /* TODO: This can be done faster is several blocks share the
311 * same VLC selector (subband_vlc_index) */
312 read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode,
313 mantissas, subband_size);
315 /* decode the scale factor for this subband */
316 scale_factor = ff_atrac_sf_table[sf_index[i]] *
317 inv_max_quant[subband_vlc_index[i]];
319 /* inverse quantize the coefficients */
320 for (j = 0; first < last; first++, j++)
321 output[first] = mantissas[j] * scale_factor;
323 /* this subband was not coded, so zero the entire subband */
324 memset(output + first, 0, subband_size * sizeof(float));
328 /* clear the subbands that were not coded */
329 first = subband_tab[i];
330 memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(float));
335 * Restore the quantized tonal components
337 * @param components tonal components
338 * @param num_bands number of coded bands
340 static int decode_tonal_components(GetBitContext *gb,
341 TonalComponent *components, int num_bands)
344 int nb_components, coding_mode_selector, coding_mode;
345 int band_flags[4], mantissa[8];
346 int component_count = 0;
348 nb_components = get_bits(gb, 5);
350 /* no tonal components */
351 if (nb_components == 0)
354 coding_mode_selector = get_bits(gb, 2);
355 if (coding_mode_selector == 2)
356 return AVERROR_INVALIDDATA;
358 coding_mode = coding_mode_selector & 1;
360 for (i = 0; i < nb_components; i++) {
361 int coded_values_per_component, quant_step_index;
363 for (b = 0; b <= num_bands; b++)
364 band_flags[b] = get_bits1(gb);
366 coded_values_per_component = get_bits(gb, 3);
368 quant_step_index = get_bits(gb, 3);
369 if (quant_step_index <= 1)
370 return AVERROR_INVALIDDATA;
372 if (coding_mode_selector == 3)
373 coding_mode = get_bits1(gb);
375 for (b = 0; b < (num_bands + 1) * 4; b++) {
376 int coded_components;
378 if (band_flags[b >> 2] == 0)
381 coded_components = get_bits(gb, 3);
383 for (c = 0; c < coded_components; c++) {
384 TonalComponent *cmp = &components[component_count];
385 int sf_index, coded_values, max_coded_values;
388 sf_index = get_bits(gb, 6);
389 if (component_count >= 64)
390 return AVERROR_INVALIDDATA;
392 cmp->pos = b * 64 + get_bits(gb, 6);
394 max_coded_values = SAMPLES_PER_FRAME - cmp->pos;
395 coded_values = coded_values_per_component + 1;
396 coded_values = FFMIN(max_coded_values, coded_values);
398 scale_factor = ff_atrac_sf_table[sf_index] *
399 inv_max_quant[quant_step_index];
401 read_quant_spectral_coeffs(gb, quant_step_index, coding_mode,
402 mantissa, coded_values);
404 cmp->num_coefs = coded_values;
407 for (m = 0; m < coded_values; m++)
408 cmp->coef[m] = mantissa[m] * scale_factor;
415 return component_count;
419 * Decode gain parameters for the coded bands
421 * @param block the gainblock for the current band
422 * @param num_bands amount of coded bands
424 static int decode_gain_control(GetBitContext *gb, GainBlock *block,
430 GainInfo *gain = block->g_block;
432 for (i = 0; i <= num_bands; i++) {
433 num_data = get_bits(gb, 3);
434 gain[i].num_gain_data = num_data;
435 level = gain[i].lev_code;
436 loc = gain[i].loc_code;
438 for (cf = 0; cf < gain[i].num_gain_data; cf++) {
439 level[cf] = get_bits(gb, 4);
440 loc [cf] = get_bits(gb, 5);
441 if (cf && loc[cf] <= loc[cf - 1])
442 return AVERROR_INVALIDDATA;
446 /* Clear the unused blocks. */
448 gain[i].num_gain_data = 0;
454 * Apply gain parameters and perform the MDCT overlapping part
456 * @param input input buffer
457 * @param prev previous buffer to perform overlap against
458 * @param output output buffer
459 * @param gain1 current band gain info
460 * @param gain2 next band gain info
462 static void gain_compensate_and_overlap(float *input, float *prev,
463 float *output, GainInfo *gain1,
466 float g1, g2, gain_inc;
467 int i, j, num_data, start_loc, end_loc;
470 if (gain2->num_gain_data == 0)
473 g1 = gain_tab1[gain2->lev_code[0]];
475 if (gain1->num_gain_data == 0) {
476 for (i = 0; i < 256; i++)
477 output[i] = input[i] * g1 + prev[i];
479 num_data = gain1->num_gain_data;
480 gain1->loc_code[num_data] = 32;
481 gain1->lev_code[num_data] = 4;
483 for (i = 0, j = 0; i < num_data; i++) {
484 start_loc = gain1->loc_code[i] * 8;
485 end_loc = start_loc + 8;
487 g2 = gain_tab1[gain1->lev_code[i]];
488 gain_inc = gain_tab2[gain1->lev_code[i + 1] -
489 gain1->lev_code[i ] + 15];
492 for (; j < start_loc; j++)
493 output[j] = (input[j] * g1 + prev[j]) * g2;
495 /* interpolation is done over eight samples */
496 for (; j < end_loc; j++) {
497 output[j] = (input[j] * g1 + prev[j]) * g2;
503 output[j] = input[j] * g1 + prev[j];
506 /* Delay for the overlapping part. */
507 memcpy(prev, &input[256], 256 * sizeof(float));
511 * Combine the tonal band spectrum and regular band spectrum
513 * @param spectrum output spectrum buffer
514 * @param num_components number of tonal components
515 * @param components tonal components for this band
516 * @return position of the last tonal coefficient
518 static int add_tonal_components(float *spectrum, int num_components,
519 TonalComponent *components)
521 int i, j, last_pos = -1;
522 float *input, *output;
524 for (i = 0; i < num_components; i++) {
525 last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
526 input = components[i].coef;
527 output = &spectrum[components[i].pos];
529 for (j = 0; j < components[i].num_coefs; j++)
530 output[i] += input[i];
536 #define INTERPOLATE(old, new, nsample) \
537 ((old) + (nsample) * 0.125 * ((new) - (old)))
539 static void reverse_matrixing(float *su1, float *su2, int *prev_code,
542 int i, nsample, band;
543 float mc1_l, mc1_r, mc2_l, mc2_r;
545 for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
546 int s1 = prev_code[i];
547 int s2 = curr_code[i];
551 /* Selector value changed, interpolation needed. */
552 mc1_l = matrix_coeffs[s1 * 2 ];
553 mc1_r = matrix_coeffs[s1 * 2 + 1];
554 mc2_l = matrix_coeffs[s2 * 2 ];
555 mc2_r = matrix_coeffs[s2 * 2 + 1];
557 /* Interpolation is done over the first eight samples. */
558 for (; nsample < band + 8; nsample++) {
559 float c1 = su1[nsample];
560 float c2 = su2[nsample];
561 c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
562 c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
564 su2[nsample] = c1 * 2.0 - c2;
568 /* Apply the matrix without interpolation. */
570 case 0: /* M/S decoding */
571 for (; nsample < band + 256; nsample++) {
572 float c1 = su1[nsample];
573 float c2 = su2[nsample];
574 su1[nsample] = c2 * 2.0;
575 su2[nsample] = (c1 - c2) * 2.0;
579 for (; nsample < band + 256; nsample++) {
580 float c1 = su1[nsample];
581 float c2 = su2[nsample];
582 su1[nsample] = (c1 + c2) * 2.0;
583 su2[nsample] = c2 * -2.0;
588 for (; nsample < band + 256; nsample++) {
589 float c1 = su1[nsample];
590 float c2 = su2[nsample];
591 su1[nsample] = c1 + c2;
592 su2[nsample] = c1 - c2;
601 static void get_channel_weights(int index, int flag, float ch[2])
607 ch[0] = (index & 7) / 7.0;
608 ch[1] = sqrt(2 - ch[0] * ch[0]);
610 FFSWAP(float, ch[0], ch[1]);
614 static void channel_weighting(float *su1, float *su2, int *p3)
617 /* w[x][y] y=0 is left y=1 is right */
620 if (p3[1] != 7 || p3[3] != 7) {
621 get_channel_weights(p3[1], p3[0], w[0]);
622 get_channel_weights(p3[3], p3[2], w[1]);
624 for (band = 256; band < 4 * 256; band += 256) {
625 for (nsample = band; nsample < band + 8; nsample++) {
626 su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
627 su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
629 for(; nsample < band + 256; nsample++) {
630 su1[nsample] *= w[1][0];
631 su2[nsample] *= w[1][1];
638 * Decode a Sound Unit
640 * @param snd the channel unit to be used
641 * @param output the decoded samples before IQMF in float representation
642 * @param channel_num channel number
643 * @param coding_mode the coding mode (JOINT_STEREO or regular stereo/mono)
645 static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb,
646 ChannelUnit *snd, float *output,
647 int channel_num, int coding_mode)
649 int band, ret, num_subbands, last_tonal, num_bands;
650 GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch];
651 GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
653 if (coding_mode == JOINT_STEREO && channel_num == 1) {
654 if (get_bits(gb, 2) != 3) {
655 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
656 return AVERROR_INVALIDDATA;
659 if (get_bits(gb, 6) != 0x28) {
660 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
661 return AVERROR_INVALIDDATA;
665 /* number of coded QMF bands */
666 snd->bands_coded = get_bits(gb, 2);
668 ret = decode_gain_control(gb, gain2, snd->bands_coded);
672 snd->num_components = decode_tonal_components(gb, snd->components,
674 if (snd->num_components == -1)
677 num_subbands = decode_spectrum(gb, snd->spectrum);
679 /* Merge the decoded spectrum and tonal components. */
680 last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
684 /* calculate number of used MLT/QMF bands according to the amount of coded
686 num_bands = (subband_tab[num_subbands] - 1) >> 8;
688 num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
691 /* Reconstruct time domain samples. */
692 for (band = 0; band < 4; band++) {
693 /* Perform the IMDCT step without overlapping. */
694 if (band <= num_bands)
695 imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
697 memset(snd->imdct_buf, 0, 512 * sizeof(float));
699 /* gain compensation and overlapping */
700 gain_compensate_and_overlap(snd->imdct_buf,
701 &snd->prev_frame[band * 256],
703 &gain1->g_block[band],
704 &gain2->g_block[band]);
707 /* Swap the gain control buffers for the next frame. */
708 snd->gc_blk_switch ^= 1;
713 static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
716 ATRAC3Context *q = avctx->priv_data;
720 if (q->coding_mode == JOINT_STEREO) {
721 /* channel coupling mode */
722 /* decode Sound Unit 1 */
723 init_get_bits(&q->gb, databuf, avctx->block_align * 8);
725 ret = decode_channel_sound_unit(q, &q->gb, q->units, out_samples[0], 0,
730 /* Framedata of the su2 in the joint-stereo mode is encoded in
731 * reverse byte order so we need to swap it first. */
732 if (databuf == q->decoded_bytes_buffer) {
733 uint8_t *ptr2 = q->decoded_bytes_buffer + avctx->block_align - 1;
734 ptr1 = q->decoded_bytes_buffer;
735 for (i = 0; i < avctx->block_align / 2; i++, ptr1++, ptr2--)
736 FFSWAP(uint8_t, *ptr1, *ptr2);
738 const uint8_t *ptr2 = databuf + avctx->block_align - 1;
739 for (i = 0; i < avctx->block_align; i++)
740 q->decoded_bytes_buffer[i] = *ptr2--;
743 /* Skip the sync codes (0xF8). */
744 ptr1 = q->decoded_bytes_buffer;
745 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
746 if (i >= avctx->block_align)
747 return AVERROR_INVALIDDATA;
751 /* set the bitstream reader at the start of the second Sound Unit*/
752 init_get_bits(&q->gb, ptr1, avctx->block_align * 8);
754 /* Fill the Weighting coeffs delay buffer */
755 memmove(q->weighting_delay, &q->weighting_delay[2], 4 * sizeof(int));
756 q->weighting_delay[4] = get_bits1(&q->gb);
757 q->weighting_delay[5] = get_bits(&q->gb, 3);
759 for (i = 0; i < 4; i++) {
760 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
761 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
762 q->matrix_coeff_index_next[i] = get_bits(&q->gb, 2);
765 /* Decode Sound Unit 2. */
766 ret = decode_channel_sound_unit(q, &q->gb, &q->units[1],
767 out_samples[1], 1, JOINT_STEREO);
771 /* Reconstruct the channel coefficients. */
772 reverse_matrixing(out_samples[0], out_samples[1],
773 q->matrix_coeff_index_prev,
774 q->matrix_coeff_index_now);
776 channel_weighting(out_samples[0], out_samples[1], q->weighting_delay);
778 /* normal stereo mode or mono */
779 /* Decode the channel sound units. */
780 for (i = 0; i < avctx->channels; i++) {
781 /* Set the bitstream reader at the start of a channel sound unit. */
782 init_get_bits(&q->gb,
783 databuf + i * avctx->block_align / avctx->channels,
784 avctx->block_align * 8 / avctx->channels);
786 ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
787 out_samples[i], i, q->coding_mode);
793 /* Apply the iQMF synthesis filter. */
794 for (i = 0; i < avctx->channels; i++) {
795 float *p1 = out_samples[i];
796 float *p2 = p1 + 256;
797 float *p3 = p2 + 256;
798 float *p4 = p3 + 256;
799 ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
800 ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
801 ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
807 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
808 int *got_frame_ptr, AVPacket *avpkt)
810 const uint8_t *buf = avpkt->data;
811 int buf_size = avpkt->size;
812 ATRAC3Context *q = avctx->priv_data;
814 const uint8_t *databuf;
816 if (buf_size < avctx->block_align) {
817 av_log(avctx, AV_LOG_ERROR,
818 "Frame too small (%d bytes). Truncated file?\n", buf_size);
819 return AVERROR_INVALIDDATA;
822 /* get output buffer */
823 q->frame.nb_samples = SAMPLES_PER_FRAME;
824 if ((ret = avctx->get_buffer(avctx, &q->frame)) < 0) {
825 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
829 /* Check if we need to descramble and what buffer to pass on. */
830 if (q->scrambled_stream) {
831 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
832 databuf = q->decoded_bytes_buffer;
837 ret = decode_frame(avctx, databuf, (float **)q->frame.extended_data);
839 av_log(NULL, AV_LOG_ERROR, "Frame decoding error!\n");
844 *(AVFrame *)data = q->frame;
846 return avctx->block_align;
849 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
853 const uint8_t *edata_ptr = avctx->extradata;
854 ATRAC3Context *q = avctx->priv_data;
855 static VLC_TYPE atrac3_vlc_table[4096][2];
856 static int vlcs_initialized = 0;
858 /* Take data from the AVCodecContext (RM container). */
859 q->sample_rate = avctx->sample_rate;
860 q->bit_rate = avctx->bit_rate;
862 if (avctx->channels <= 0 || avctx->channels > 2) {
863 av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n");
864 return AVERROR(EINVAL);
867 /* Take care of the codec-specific extradata. */
868 if (avctx->extradata_size == 14) {
869 /* Parse the extradata, WAV format */
870 av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
871 bytestream_get_le16(&edata_ptr)); // Unknown value always 1
872 edata_ptr += 4; // samples per channel
873 q->coding_mode = bytestream_get_le16(&edata_ptr);
874 av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
875 bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
876 q->frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1
877 av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
878 bytestream_get_le16(&edata_ptr)); // Unknown always 0
881 q->samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
884 q->coding_mode = q->coding_mode ? JOINT_STEREO : STEREO;
885 q->scrambled_stream = 0;
887 if (avctx->block_align != 96 * avctx->channels * q->frame_factor &&
888 avctx->block_align != 152 * avctx->channels * q->frame_factor &&
889 avctx->block_align != 192 * avctx->channels * q->frame_factor) {
890 av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
891 "configuration %d/%d/%d\n", avctx->block_align,
892 avctx->channels, q->frame_factor);
893 return AVERROR_INVALIDDATA;
895 } else if (avctx->extradata_size == 10) {
896 /* Parse the extradata, RM format. */
897 version = bytestream_get_be32(&edata_ptr);
898 q->samples_per_frame = bytestream_get_be16(&edata_ptr);
899 delay = bytestream_get_be16(&edata_ptr);
900 q->coding_mode = bytestream_get_be16(&edata_ptr);
901 q->scrambled_stream = 1;
904 av_log(NULL, AV_LOG_ERROR, "Unknown extradata size %d.\n",
905 avctx->extradata_size);
908 /* Check the extradata */
911 av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
912 return AVERROR_INVALIDDATA;
915 if (q->samples_per_frame != SAMPLES_PER_FRAME &&
916 q->samples_per_frame != SAMPLES_PER_FRAME * 2) {
917 av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
918 q->samples_per_frame);
919 return AVERROR_INVALIDDATA;
922 if (delay != 0x88E) {
923 av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
925 return AVERROR_INVALIDDATA;
928 if (q->coding_mode == STEREO)
929 av_log(avctx, AV_LOG_DEBUG, "Normal stereo detected.\n");
930 else if (q->coding_mode == JOINT_STEREO)
931 av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
933 av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
935 return AVERROR_INVALIDDATA;
938 if (avctx->block_align >= UINT_MAX / 2)
939 return AVERROR(EINVAL);
941 q->decoded_bytes_buffer = av_mallocz(avctx->block_align +
942 (4 - avctx->block_align % 4) +
943 FF_INPUT_BUFFER_PADDING_SIZE);
944 if (q->decoded_bytes_buffer == NULL)
945 return AVERROR(ENOMEM);
948 /* Initialize the VLC tables. */
949 if (!vlcs_initialized) {
950 for (i = 0; i < 7; i++) {
951 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
952 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] -
954 init_vlc(&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
956 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
958 vlcs_initialized = 1;
961 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
963 if ((ret = init_atrac3_transforms(q))) {
964 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
965 av_freep(&q->decoded_bytes_buffer);
969 ff_atrac_generate_tables();
971 /* Generate gain tables */
972 for (i = 0; i < 16; i++)
973 gain_tab1[i] = powf(2.0, (4 - i));
975 for (i = -15; i < 16; i++)
976 gain_tab2[i + 15] = powf(2.0, i * -0.125);
978 /* init the joint-stereo decoding data */
979 q->weighting_delay[0] = 0;
980 q->weighting_delay[1] = 7;
981 q->weighting_delay[2] = 0;
982 q->weighting_delay[3] = 7;
983 q->weighting_delay[4] = 0;
984 q->weighting_delay[5] = 7;
986 for (i = 0; i < 4; i++) {
987 q->matrix_coeff_index_prev[i] = 3;
988 q->matrix_coeff_index_now[i] = 3;
989 q->matrix_coeff_index_next[i] = 3;
992 avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
993 ff_fmt_convert_init(&q->fmt_conv, avctx);
995 q->units = av_mallocz(sizeof(ChannelUnit) * avctx->channels);
997 atrac3_decode_close(avctx);
998 return AVERROR(ENOMEM);
1001 avcodec_get_frame_defaults(&q->frame);
1002 avctx->coded_frame = &q->frame;
1007 AVCodec ff_atrac3_decoder = {
1009 .type = AVMEDIA_TYPE_AUDIO,
1010 .id = AV_CODEC_ID_ATRAC3,
1011 .priv_data_size = sizeof(ATRAC3Context),
1012 .init = atrac3_decode_init,
1013 .close = atrac3_decode_close,
1014 .decode = atrac3_decode_frame,
1015 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1016 .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),
1017 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1018 AV_SAMPLE_FMT_NONE },