2 * ATRAC3 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 * ATRAC3 compatible decoder.
26 * This decoder handles Sony's ATRAC3 data.
28 * Container formats used to store ATRAC3 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/attributes.h"
40 #include "libavutil/float_dsp.h"
42 #include "bytestream.h"
44 #include "fmtconvert.h"
49 #include "atrac3data.h"
51 #define JOINT_STEREO 0x12
54 #define SAMPLES_PER_FRAME 1024
57 typedef struct GainInfo {
63 typedef struct GainBlock {
67 typedef struct TonalComponent {
73 typedef struct ChannelUnit {
76 float prev_frame[SAMPLES_PER_FRAME];
78 TonalComponent components[64];
79 GainBlock gain_block[2];
81 DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
82 DECLARE_ALIGNED(32, float, imdct_buf)[SAMPLES_PER_FRAME];
84 float delay_buf1[46]; ///<qmf delay buffers
89 typedef struct ATRAC3Context {
98 /** joint-stereo related variables */
99 int matrix_coeff_index_prev[4];
100 int matrix_coeff_index_now[4];
101 int matrix_coeff_index_next[4];
102 int weighting_delay[6];
106 uint8_t *decoded_bytes_buffer;
107 float temp_buf[1070];
111 int scrambled_stream;
115 FmtConvertContext fmt_conv;
116 AVFloatDSPContext fdsp;
119 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
120 static VLC_TYPE atrac3_vlc_table[4096][2];
121 static VLC spectral_coeff_tab[7];
122 static float gain_tab1[16];
123 static float gain_tab2[31];
127 * Regular 512 points IMDCT without overlapping, with the exception of the
128 * swapping of odd bands caused by the reverse spectra of the QMF.
130 * @param odd_band 1 if the band is an odd band
132 static void imlt(ATRAC3Context *q, float *input, float *output, int odd_band)
138 * Reverse the odd bands before IMDCT, this is an effect of the QMF
139 * transform or it gives better compression to do it this way.
140 * FIXME: It should be possible to handle this in imdct_calc
141 * for that to happen a modification of the prerotation step of
142 * all SIMD code and C code is needed.
143 * Or fix the functions before so they generate a pre reversed spectrum.
145 for (i = 0; i < 128; i++)
146 FFSWAP(float, input[i], input[255 - i]);
149 q->mdct_ctx.imdct_calc(&q->mdct_ctx, output, input);
151 /* Perform windowing on the output. */
152 q->fdsp.vector_fmul(output, output, mdct_window, MDCT_SIZE);
156 * indata descrambling, only used for data coming from the rm container
158 static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes)
163 uint32_t *output = (uint32_t *)out;
165 off = (intptr_t)input & 3;
166 buf = (const uint32_t *)(input - off);
168 c = av_be2ne32((0x537F6103U >> (off * 8)) | (0x537F6103U << (32 - (off * 8))));
170 c = av_be2ne32(0x537F6103U);
172 for (i = 0; i < bytes / 4; i++)
173 output[i] = c ^ buf[i];
176 avpriv_request_sample(NULL, "Offset of %d", off);
181 static av_cold void init_atrac3_window(void)
185 /* generate the mdct window, for details see
186 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
187 for (i = 0, j = 255; i < 128; i++, j--) {
188 float wi = sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
189 float wj = sin(((j + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
190 float w = 0.5 * (wi * wi + wj * wj);
191 mdct_window[i] = mdct_window[511 - i] = wi / w;
192 mdct_window[j] = mdct_window[511 - j] = wj / w;
196 static av_cold int atrac3_decode_close(AVCodecContext *avctx)
198 ATRAC3Context *q = avctx->priv_data;
201 av_free(q->decoded_bytes_buffer);
203 ff_mdct_end(&q->mdct_ctx);
211 * @param selector which table the output values are coded with
212 * @param coding_flag constant length coding or variable length coding
213 * @param mantissas mantissa output table
214 * @param num_codes number of values to get
216 static void read_quant_spectral_coeffs(GetBitContext *gb, int selector,
217 int coding_flag, int *mantissas,
220 int i, code, huff_symb;
225 if (coding_flag != 0) {
226 /* constant length coding (CLC) */
227 int num_bits = clc_length_tab[selector];
230 for (i = 0; i < num_codes; i++) {
232 code = get_sbits(gb, num_bits);
238 for (i = 0; i < num_codes; i++) {
240 code = get_bits(gb, num_bits); // num_bits is always 4 in this case
243 mantissas[i * 2 ] = mantissa_clc_tab[code >> 2];
244 mantissas[i * 2 + 1] = mantissa_clc_tab[code & 3];
248 /* variable length coding (VLC) */
250 for (i = 0; i < num_codes; i++) {
251 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector-1].table,
252 spectral_coeff_tab[selector-1].bits, 3);
254 code = huff_symb >> 1;
260 for (i = 0; i < num_codes; i++) {
261 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table,
262 spectral_coeff_tab[selector - 1].bits, 3);
263 mantissas[i * 2 ] = mantissa_vlc_tab[huff_symb * 2 ];
264 mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1];
271 * Restore the quantized band spectrum coefficients
273 * @return subband count, fix for broken specification/files
275 static int decode_spectrum(GetBitContext *gb, float *output)
277 int num_subbands, coding_mode, i, j, first, last, subband_size;
278 int subband_vlc_index[32], sf_index[32];
282 num_subbands = get_bits(gb, 5); // number of coded subbands
283 coding_mode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
285 /* get the VLC selector table for the subbands, 0 means not coded */
286 for (i = 0; i <= num_subbands; i++)
287 subband_vlc_index[i] = get_bits(gb, 3);
289 /* read the scale factor indexes from the stream */
290 for (i = 0; i <= num_subbands; i++) {
291 if (subband_vlc_index[i] != 0)
292 sf_index[i] = get_bits(gb, 6);
295 for (i = 0; i <= num_subbands; i++) {
296 first = subband_tab[i ];
297 last = subband_tab[i + 1];
299 subband_size = last - first;
301 if (subband_vlc_index[i] != 0) {
302 /* decode spectral coefficients for this subband */
303 /* TODO: This can be done faster is several blocks share the
304 * same VLC selector (subband_vlc_index) */
305 read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode,
306 mantissas, subband_size);
308 /* decode the scale factor for this subband */
309 scale_factor = ff_atrac_sf_table[sf_index[i]] *
310 inv_max_quant[subband_vlc_index[i]];
312 /* inverse quantize the coefficients */
313 for (j = 0; first < last; first++, j++)
314 output[first] = mantissas[j] * scale_factor;
316 /* this subband was not coded, so zero the entire subband */
317 memset(output + first, 0, subband_size * sizeof(*output));
321 /* clear the subbands that were not coded */
322 first = subband_tab[i];
323 memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output));
328 * Restore the quantized tonal components
330 * @param components tonal components
331 * @param num_bands number of coded bands
333 static int decode_tonal_components(GetBitContext *gb,
334 TonalComponent *components, int num_bands)
337 int nb_components, coding_mode_selector, coding_mode;
338 int band_flags[4], mantissa[8];
339 int component_count = 0;
341 nb_components = get_bits(gb, 5);
343 /* no tonal components */
344 if (nb_components == 0)
347 coding_mode_selector = get_bits(gb, 2);
348 if (coding_mode_selector == 2)
349 return AVERROR_INVALIDDATA;
351 coding_mode = coding_mode_selector & 1;
353 for (i = 0; i < nb_components; i++) {
354 int coded_values_per_component, quant_step_index;
356 for (b = 0; b <= num_bands; b++)
357 band_flags[b] = get_bits1(gb);
359 coded_values_per_component = get_bits(gb, 3);
361 quant_step_index = get_bits(gb, 3);
362 if (quant_step_index <= 1)
363 return AVERROR_INVALIDDATA;
365 if (coding_mode_selector == 3)
366 coding_mode = get_bits1(gb);
368 for (b = 0; b < (num_bands + 1) * 4; b++) {
369 int coded_components;
371 if (band_flags[b >> 2] == 0)
374 coded_components = get_bits(gb, 3);
376 for (c = 0; c < coded_components; c++) {
377 TonalComponent *cmp = &components[component_count];
378 int sf_index, coded_values, max_coded_values;
381 sf_index = get_bits(gb, 6);
382 if (component_count >= 64)
383 return AVERROR_INVALIDDATA;
385 cmp->pos = b * 64 + get_bits(gb, 6);
387 max_coded_values = SAMPLES_PER_FRAME - cmp->pos;
388 coded_values = coded_values_per_component + 1;
389 coded_values = FFMIN(max_coded_values, coded_values);
391 scale_factor = ff_atrac_sf_table[sf_index] *
392 inv_max_quant[quant_step_index];
394 read_quant_spectral_coeffs(gb, quant_step_index, coding_mode,
395 mantissa, coded_values);
397 cmp->num_coefs = coded_values;
400 for (m = 0; m < coded_values; m++)
401 cmp->coef[m] = mantissa[m] * scale_factor;
408 return component_count;
412 * Decode gain parameters for the coded bands
414 * @param block the gainblock for the current band
415 * @param num_bands amount of coded bands
417 static int decode_gain_control(GetBitContext *gb, GainBlock *block,
423 GainInfo *gain = block->g_block;
425 for (i = 0; i <= num_bands; i++) {
426 num_data = get_bits(gb, 3);
427 gain[i].num_gain_data = num_data;
428 level = gain[i].lev_code;
429 loc = gain[i].loc_code;
431 for (j = 0; j < gain[i].num_gain_data; j++) {
432 level[j] = get_bits(gb, 4);
433 loc[j] = get_bits(gb, 5);
434 if (j && loc[j] <= loc[j - 1])
435 return AVERROR_INVALIDDATA;
439 /* Clear the unused blocks. */
441 gain[i].num_gain_data = 0;
447 * Apply gain parameters and perform the MDCT overlapping part
449 * @param input input buffer
450 * @param prev previous buffer to perform overlap against
451 * @param output output buffer
452 * @param gain1 current band gain info
453 * @param gain2 next band gain info
455 static void gain_compensate_and_overlap(float *input, float *prev,
456 float *output, GainInfo *gain1,
459 float g1, g2, gain_inc;
460 int i, j, num_data, start_loc, end_loc;
463 if (gain2->num_gain_data == 0)
466 g1 = gain_tab1[gain2->lev_code[0]];
468 if (gain1->num_gain_data == 0) {
469 for (i = 0; i < 256; i++)
470 output[i] = input[i] * g1 + prev[i];
472 num_data = gain1->num_gain_data;
473 gain1->loc_code[num_data] = 32;
474 gain1->lev_code[num_data] = 4;
476 for (i = 0, j = 0; i < num_data; i++) {
477 start_loc = gain1->loc_code[i] * 8;
478 end_loc = start_loc + 8;
480 g2 = gain_tab1[gain1->lev_code[i]];
481 gain_inc = gain_tab2[gain1->lev_code[i + 1] -
482 gain1->lev_code[i ] + 15];
485 for (; j < start_loc; j++)
486 output[j] = (input[j] * g1 + prev[j]) * g2;
488 /* interpolation is done over eight samples */
489 for (; j < end_loc; j++) {
490 output[j] = (input[j] * g1 + prev[j]) * g2;
496 output[j] = input[j] * g1 + prev[j];
499 /* Delay for the overlapping part. */
500 memcpy(prev, &input[256], 256 * sizeof(*prev));
504 * Combine the tonal band spectrum and regular band spectrum
506 * @param spectrum output spectrum buffer
507 * @param num_components number of tonal components
508 * @param components tonal components for this band
509 * @return position of the last tonal coefficient
511 static int add_tonal_components(float *spectrum, int num_components,
512 TonalComponent *components)
514 int i, j, last_pos = -1;
515 float *input, *output;
517 for (i = 0; i < num_components; i++) {
518 last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
519 input = components[i].coef;
520 output = &spectrum[components[i].pos];
522 for (j = 0; j < components[i].num_coefs; j++)
523 output[j] += input[j];
529 #define INTERPOLATE(old, new, nsample) \
530 ((old) + (nsample) * 0.125 * ((new) - (old)))
532 static void reverse_matrixing(float *su1, float *su2, int *prev_code,
535 int i, nsample, band;
536 float mc1_l, mc1_r, mc2_l, mc2_r;
538 for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
539 int s1 = prev_code[i];
540 int s2 = curr_code[i];
544 /* Selector value changed, interpolation needed. */
545 mc1_l = matrix_coeffs[s1 * 2 ];
546 mc1_r = matrix_coeffs[s1 * 2 + 1];
547 mc2_l = matrix_coeffs[s2 * 2 ];
548 mc2_r = matrix_coeffs[s2 * 2 + 1];
550 /* Interpolation is done over the first eight samples. */
551 for (; nsample < band + 8; nsample++) {
552 float c1 = su1[nsample];
553 float c2 = su2[nsample];
554 c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
555 c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
557 su2[nsample] = c1 * 2.0 - c2;
561 /* Apply the matrix without interpolation. */
563 case 0: /* M/S decoding */
564 for (; nsample < band + 256; nsample++) {
565 float c1 = su1[nsample];
566 float c2 = su2[nsample];
567 su1[nsample] = c2 * 2.0;
568 su2[nsample] = (c1 - c2) * 2.0;
572 for (; nsample < band + 256; nsample++) {
573 float c1 = su1[nsample];
574 float c2 = su2[nsample];
575 su1[nsample] = (c1 + c2) * 2.0;
576 su2[nsample] = c2 * -2.0;
581 for (; nsample < band + 256; nsample++) {
582 float c1 = su1[nsample];
583 float c2 = su2[nsample];
584 su1[nsample] = c1 + c2;
585 su2[nsample] = c1 - c2;
594 static void get_channel_weights(int index, int flag, float ch[2])
600 ch[0] = (index & 7) / 7.0;
601 ch[1] = sqrt(2 - ch[0] * ch[0]);
603 FFSWAP(float, ch[0], ch[1]);
607 static void channel_weighting(float *su1, float *su2, int *p3)
610 /* w[x][y] y=0 is left y=1 is right */
613 if (p3[1] != 7 || p3[3] != 7) {
614 get_channel_weights(p3[1], p3[0], w[0]);
615 get_channel_weights(p3[3], p3[2], w[1]);
617 for (band = 256; band < 4 * 256; band += 256) {
618 for (nsample = band; nsample < band + 8; nsample++) {
619 su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
620 su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
622 for(; nsample < band + 256; nsample++) {
623 su1[nsample] *= w[1][0];
624 su2[nsample] *= w[1][1];
631 * Decode a Sound Unit
633 * @param snd the channel unit to be used
634 * @param output the decoded samples before IQMF in float representation
635 * @param channel_num channel number
636 * @param coding_mode the coding mode (JOINT_STEREO or regular stereo/mono)
638 static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb,
639 ChannelUnit *snd, float *output,
640 int channel_num, int coding_mode)
642 int band, ret, num_subbands, last_tonal, num_bands;
643 GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch];
644 GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
646 if (coding_mode == JOINT_STEREO && channel_num == 1) {
647 if (get_bits(gb, 2) != 3) {
648 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
649 return AVERROR_INVALIDDATA;
652 if (get_bits(gb, 6) != 0x28) {
653 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
654 return AVERROR_INVALIDDATA;
658 /* number of coded QMF bands */
659 snd->bands_coded = get_bits(gb, 2);
661 ret = decode_gain_control(gb, gain2, snd->bands_coded);
665 snd->num_components = decode_tonal_components(gb, snd->components,
667 if (snd->num_components < 0)
668 return snd->num_components;
670 num_subbands = decode_spectrum(gb, snd->spectrum);
672 /* Merge the decoded spectrum and tonal components. */
673 last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
677 /* calculate number of used MLT/QMF bands according to the amount of coded
679 num_bands = (subband_tab[num_subbands] - 1) >> 8;
681 num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
684 /* Reconstruct time domain samples. */
685 for (band = 0; band < 4; band++) {
686 /* Perform the IMDCT step without overlapping. */
687 if (band <= num_bands)
688 imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
690 memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf));
692 /* gain compensation and overlapping */
693 gain_compensate_and_overlap(snd->imdct_buf,
694 &snd->prev_frame[band * 256],
696 &gain1->g_block[band],
697 &gain2->g_block[band]);
700 /* Swap the gain control buffers for the next frame. */
701 snd->gc_blk_switch ^= 1;
706 static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
709 ATRAC3Context *q = avctx->priv_data;
713 if (q->coding_mode == JOINT_STEREO) {
714 /* channel coupling mode */
715 /* decode Sound Unit 1 */
716 init_get_bits(&q->gb, databuf, avctx->block_align * 8);
718 ret = decode_channel_sound_unit(q, &q->gb, q->units, out_samples[0], 0,
723 /* Framedata of the su2 in the joint-stereo mode is encoded in
724 * reverse byte order so we need to swap it first. */
725 if (databuf == q->decoded_bytes_buffer) {
726 uint8_t *ptr2 = q->decoded_bytes_buffer + avctx->block_align - 1;
727 ptr1 = q->decoded_bytes_buffer;
728 for (i = 0; i < avctx->block_align / 2; i++, ptr1++, ptr2--)
729 FFSWAP(uint8_t, *ptr1, *ptr2);
731 const uint8_t *ptr2 = databuf + avctx->block_align - 1;
732 for (i = 0; i < avctx->block_align; i++)
733 q->decoded_bytes_buffer[i] = *ptr2--;
736 /* Skip the sync codes (0xF8). */
737 ptr1 = q->decoded_bytes_buffer;
738 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
739 if (i >= avctx->block_align)
740 return AVERROR_INVALIDDATA;
744 /* set the bitstream reader at the start of the second Sound Unit*/
745 init_get_bits(&q->gb, ptr1, (avctx->block_align - i) * 8);
747 /* Fill the Weighting coeffs delay buffer */
748 memmove(q->weighting_delay, &q->weighting_delay[2],
749 4 * sizeof(*q->weighting_delay));
750 q->weighting_delay[4] = get_bits1(&q->gb);
751 q->weighting_delay[5] = get_bits(&q->gb, 3);
753 for (i = 0; i < 4; i++) {
754 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
755 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
756 q->matrix_coeff_index_next[i] = get_bits(&q->gb, 2);
759 /* Decode Sound Unit 2. */
760 ret = decode_channel_sound_unit(q, &q->gb, &q->units[1],
761 out_samples[1], 1, JOINT_STEREO);
765 /* Reconstruct the channel coefficients. */
766 reverse_matrixing(out_samples[0], out_samples[1],
767 q->matrix_coeff_index_prev,
768 q->matrix_coeff_index_now);
770 channel_weighting(out_samples[0], out_samples[1], q->weighting_delay);
772 /* normal stereo mode or mono */
773 /* Decode the channel sound units. */
774 for (i = 0; i < avctx->channels; i++) {
775 /* Set the bitstream reader at the start of a channel sound unit. */
776 init_get_bits(&q->gb,
777 databuf + i * avctx->block_align / avctx->channels,
778 avctx->block_align * 8 / avctx->channels);
780 ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
781 out_samples[i], i, q->coding_mode);
787 /* Apply the iQMF synthesis filter. */
788 for (i = 0; i < avctx->channels; i++) {
789 float *p1 = out_samples[i];
790 float *p2 = p1 + 256;
791 float *p3 = p2 + 256;
792 float *p4 = p3 + 256;
793 ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
794 ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
795 ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
801 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
802 int *got_frame_ptr, AVPacket *avpkt)
804 AVFrame *frame = data;
805 const uint8_t *buf = avpkt->data;
806 int buf_size = avpkt->size;
807 ATRAC3Context *q = avctx->priv_data;
809 const uint8_t *databuf;
811 if (buf_size < avctx->block_align) {
812 av_log(avctx, AV_LOG_ERROR,
813 "Frame too small (%d bytes). Truncated file?\n", buf_size);
814 return AVERROR_INVALIDDATA;
817 /* get output buffer */
818 frame->nb_samples = SAMPLES_PER_FRAME;
819 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
820 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
824 /* Check if we need to descramble and what buffer to pass on. */
825 if (q->scrambled_stream) {
826 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
827 databuf = q->decoded_bytes_buffer;
832 ret = decode_frame(avctx, databuf, (float **)frame->extended_data);
834 av_log(NULL, AV_LOG_ERROR, "Frame decoding error!\n");
840 return avctx->block_align;
843 static av_cold void atrac3_init_static_data(AVCodec *codec)
847 init_atrac3_window();
848 ff_atrac_generate_tables();
850 /* Initialize the VLC tables. */
851 for (i = 0; i < 7; i++) {
852 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
853 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] -
855 init_vlc(&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
857 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
860 /* Generate gain tables */
861 for (i = 0; i < 16; i++)
862 gain_tab1[i] = powf(2.0, (4 - i));
864 for (i = -15; i < 16; i++)
865 gain_tab2[i + 15] = powf(2.0, i * -0.125);
868 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
871 int version, delay, samples_per_frame, frame_factor;
872 const uint8_t *edata_ptr = avctx->extradata;
873 ATRAC3Context *q = avctx->priv_data;
875 if (avctx->channels <= 0 || avctx->channels > 2) {
876 av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n");
877 return AVERROR(EINVAL);
880 /* Take care of the codec-specific extradata. */
881 if (avctx->extradata_size == 14) {
882 /* Parse the extradata, WAV format */
883 av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
884 bytestream_get_le16(&edata_ptr)); // Unknown value always 1
885 edata_ptr += 4; // samples per channel
886 q->coding_mode = bytestream_get_le16(&edata_ptr);
887 av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
888 bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
889 frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1
890 av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
891 bytestream_get_le16(&edata_ptr)); // Unknown always 0
894 samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
897 q->coding_mode = q->coding_mode ? JOINT_STEREO : STEREO;
898 q->scrambled_stream = 0;
900 if (avctx->block_align != 96 * avctx->channels * frame_factor &&
901 avctx->block_align != 152 * avctx->channels * frame_factor &&
902 avctx->block_align != 192 * avctx->channels * frame_factor) {
903 av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
904 "configuration %d/%d/%d\n", avctx->block_align,
905 avctx->channels, frame_factor);
906 return AVERROR_INVALIDDATA;
908 } else if (avctx->extradata_size == 10) {
909 /* Parse the extradata, RM format. */
910 version = bytestream_get_be32(&edata_ptr);
911 samples_per_frame = bytestream_get_be16(&edata_ptr);
912 delay = bytestream_get_be16(&edata_ptr);
913 q->coding_mode = bytestream_get_be16(&edata_ptr);
914 q->scrambled_stream = 1;
917 av_log(NULL, AV_LOG_ERROR, "Unknown extradata size %d.\n",
918 avctx->extradata_size);
919 return AVERROR(EINVAL);
922 /* Check the extradata */
925 av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
926 return AVERROR_INVALIDDATA;
929 if (samples_per_frame != SAMPLES_PER_FRAME &&
930 samples_per_frame != SAMPLES_PER_FRAME * 2) {
931 av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
933 return AVERROR_INVALIDDATA;
936 if (delay != 0x88E) {
937 av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
939 return AVERROR_INVALIDDATA;
942 if (q->coding_mode == STEREO)
943 av_log(avctx, AV_LOG_DEBUG, "Normal stereo detected.\n");
944 else if (q->coding_mode == JOINT_STEREO) {
945 if (avctx->channels != 2)
946 return AVERROR_INVALIDDATA;
947 av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
949 av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
951 return AVERROR_INVALIDDATA;
954 if (avctx->block_align >= UINT_MAX / 2)
955 return AVERROR(EINVAL);
957 q->decoded_bytes_buffer = av_mallocz(FFALIGN(avctx->block_align, 4) +
958 FF_INPUT_BUFFER_PADDING_SIZE);
959 if (q->decoded_bytes_buffer == NULL)
960 return AVERROR(ENOMEM);
962 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
964 /* initialize the MDCT transform */
965 if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) {
966 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
967 av_freep(&q->decoded_bytes_buffer);
971 /* init the joint-stereo decoding data */
972 q->weighting_delay[0] = 0;
973 q->weighting_delay[1] = 7;
974 q->weighting_delay[2] = 0;
975 q->weighting_delay[3] = 7;
976 q->weighting_delay[4] = 0;
977 q->weighting_delay[5] = 7;
979 for (i = 0; i < 4; i++) {
980 q->matrix_coeff_index_prev[i] = 3;
981 q->matrix_coeff_index_now[i] = 3;
982 q->matrix_coeff_index_next[i] = 3;
985 avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
986 ff_fmt_convert_init(&q->fmt_conv, avctx);
988 q->units = av_mallocz(sizeof(*q->units) * avctx->channels);
990 atrac3_decode_close(avctx);
991 return AVERROR(ENOMEM);
997 AVCodec ff_atrac3_decoder = {
999 .long_name = NULL_IF_CONFIG_SMALL("ATRAC3 (Adaptive TRansform Acoustic Coding 3)"),
1000 .type = AVMEDIA_TYPE_AUDIO,
1001 .id = AV_CODEC_ID_ATRAC3,
1002 .priv_data_size = sizeof(ATRAC3Context),
1003 .init = atrac3_decode_init,
1004 .init_static_data = atrac3_init_static_data,
1005 .close = atrac3_decode_close,
1006 .decode = atrac3_decode_frame,
1007 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1008 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1009 AV_SAMPLE_FMT_NONE },