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 FFmpeg.
8 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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"
40 #include "libavutil/libm.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 {
99 /** joint-stereo related variables */
100 int matrix_coeff_index_prev[4];
101 int matrix_coeff_index_now[4];
102 int matrix_coeff_index_next[4];
103 int weighting_delay[6];
107 uint8_t *decoded_bytes_buffer;
108 float temp_buf[1070];
112 int scrambled_stream;
116 FmtConvertContext fmt_conv;
117 AVFloatDSPContext fdsp;
120 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
121 static VLC_TYPE atrac3_vlc_table[4096][2];
122 static VLC spectral_coeff_tab[7];
123 static float gain_tab1[16];
124 static float gain_tab2[31];
128 * Regular 512 points IMDCT without overlapping, with the exception of the
129 * swapping of odd bands caused by the reverse spectra of the QMF.
131 * @param odd_band 1 if the band is an odd band
133 static void imlt(ATRAC3Context *q, float *input, float *output, int odd_band)
139 * Reverse the odd bands before IMDCT, this is an effect of the QMF
140 * transform or it gives better compression to do it this way.
141 * FIXME: It should be possible to handle this in imdct_calc
142 * for that to happen a modification of the prerotation step of
143 * all SIMD code and C code is needed.
144 * Or fix the functions before so they generate a pre reversed spectrum.
146 for (i = 0; i < 128; i++)
147 FFSWAP(float, input[i], input[255 - i]);
150 q->mdct_ctx.imdct_calc(&q->mdct_ctx, output, input);
152 /* Perform windowing on the output. */
153 q->fdsp.vector_fmul(output, output, mdct_window, MDCT_SIZE);
157 * indata descrambling, only used for data coming from the rm container
159 static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes)
164 uint32_t *output = (uint32_t *)out;
166 off = (intptr_t)input & 3;
167 buf = (const uint32_t *)(input - off);
168 c = av_be2ne32((0x537F6103 >> (off * 8)) | (0x537F6103 << (32 - (off * 8))));
170 for (i = 0; i < bytes / 4; i++)
171 output[i] = c ^ buf[i];
174 av_log_ask_for_sample(NULL, "Offset of %d not handled.\n", off);
179 static av_cold void init_atrac3_window(void)
183 /* generate the mdct window, for details see
184 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
185 for (i = 0, j = 255; i < 128; i++, j--) {
186 float wi = sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
187 float wj = sin(((j + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
188 float w = 0.5 * (wi * wi + wj * wj);
189 mdct_window[i] = mdct_window[511 - i] = wi / w;
190 mdct_window[j] = mdct_window[511 - j] = wj / w;
194 static av_cold int atrac3_decode_close(AVCodecContext *avctx)
196 ATRAC3Context *q = avctx->priv_data;
199 av_free(q->decoded_bytes_buffer);
201 ff_mdct_end(&q->mdct_ctx);
209 * @param selector which table the output values are coded with
210 * @param coding_flag constant length coding or variable length coding
211 * @param mantissas mantissa output table
212 * @param num_codes number of values to get
214 static void read_quant_spectral_coeffs(GetBitContext *gb, int selector,
215 int coding_flag, int *mantissas,
218 int i, code, huff_symb;
223 if (coding_flag != 0) {
224 /* constant length coding (CLC) */
225 int num_bits = clc_length_tab[selector];
228 for (i = 0; i < num_codes; i++) {
230 code = get_sbits(gb, num_bits);
236 for (i = 0; i < num_codes; i++) {
238 code = get_bits(gb, num_bits); // num_bits is always 4 in this case
241 mantissas[i * 2 ] = mantissa_clc_tab[code >> 2];
242 mantissas[i * 2 + 1] = mantissa_clc_tab[code & 3];
246 /* variable length coding (VLC) */
248 for (i = 0; i < num_codes; i++) {
249 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector-1].table,
250 spectral_coeff_tab[selector-1].bits, 3);
252 code = huff_symb >> 1;
258 for (i = 0; i < num_codes; i++) {
259 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table,
260 spectral_coeff_tab[selector - 1].bits, 3);
261 mantissas[i * 2 ] = mantissa_vlc_tab[huff_symb * 2 ];
262 mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1];
269 * Restore the quantized band spectrum coefficients
271 * @return subband count, fix for broken specification/files
273 static int decode_spectrum(GetBitContext *gb, float *output)
275 int num_subbands, coding_mode, i, j, first, last, subband_size;
276 int subband_vlc_index[32], sf_index[32];
280 num_subbands = get_bits(gb, 5); // number of coded subbands
281 coding_mode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
283 /* get the VLC selector table for the subbands, 0 means not coded */
284 for (i = 0; i <= num_subbands; i++)
285 subband_vlc_index[i] = get_bits(gb, 3);
287 /* read the scale factor indexes from the stream */
288 for (i = 0; i <= num_subbands; i++) {
289 if (subband_vlc_index[i] != 0)
290 sf_index[i] = get_bits(gb, 6);
293 for (i = 0; i <= num_subbands; i++) {
294 first = subband_tab[i ];
295 last = subband_tab[i + 1];
297 subband_size = last - first;
299 if (subband_vlc_index[i] != 0) {
300 /* decode spectral coefficients for this subband */
301 /* TODO: This can be done faster is several blocks share the
302 * same VLC selector (subband_vlc_index) */
303 read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode,
304 mantissas, subband_size);
306 /* decode the scale factor for this subband */
307 scale_factor = ff_atrac_sf_table[sf_index[i]] *
308 inv_max_quant[subband_vlc_index[i]];
310 /* inverse quantize the coefficients */
311 for (j = 0; first < last; first++, j++)
312 output[first] = mantissas[j] * scale_factor;
314 /* this subband was not coded, so zero the entire subband */
315 memset(output + first, 0, subband_size * sizeof(*output));
319 /* clear the subbands that were not coded */
320 first = subband_tab[i];
321 memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output));
326 * Restore the quantized tonal components
328 * @param components tonal components
329 * @param num_bands number of coded bands
331 static int decode_tonal_components(GetBitContext *gb,
332 TonalComponent *components, int num_bands)
335 int nb_components, coding_mode_selector, coding_mode;
336 int band_flags[4], mantissa[8];
337 int component_count = 0;
339 nb_components = get_bits(gb, 5);
341 /* no tonal components */
342 if (nb_components == 0)
345 coding_mode_selector = get_bits(gb, 2);
346 if (coding_mode_selector == 2)
347 return AVERROR_INVALIDDATA;
349 coding_mode = coding_mode_selector & 1;
351 for (i = 0; i < nb_components; i++) {
352 int coded_values_per_component, quant_step_index;
354 for (b = 0; b <= num_bands; b++)
355 band_flags[b] = get_bits1(gb);
357 coded_values_per_component = get_bits(gb, 3);
359 quant_step_index = get_bits(gb, 3);
360 if (quant_step_index <= 1)
361 return AVERROR_INVALIDDATA;
363 if (coding_mode_selector == 3)
364 coding_mode = get_bits1(gb);
366 for (b = 0; b < (num_bands + 1) * 4; b++) {
367 int coded_components;
369 if (band_flags[b >> 2] == 0)
372 coded_components = get_bits(gb, 3);
374 for (c = 0; c < coded_components; c++) {
375 TonalComponent *cmp = &components[component_count];
376 int sf_index, coded_values, max_coded_values;
379 sf_index = get_bits(gb, 6);
380 if (component_count >= 64)
381 return AVERROR_INVALIDDATA;
383 cmp->pos = b * 64 + get_bits(gb, 6);
385 max_coded_values = SAMPLES_PER_FRAME - cmp->pos;
386 coded_values = coded_values_per_component + 1;
387 coded_values = FFMIN(max_coded_values, coded_values);
389 scale_factor = ff_atrac_sf_table[sf_index] *
390 inv_max_quant[quant_step_index];
392 read_quant_spectral_coeffs(gb, quant_step_index, coding_mode,
393 mantissa, coded_values);
395 cmp->num_coefs = coded_values;
398 for (m = 0; m < coded_values; m++)
399 cmp->coef[m] = mantissa[m] * scale_factor;
406 return component_count;
410 * Decode gain parameters for the coded bands
412 * @param block the gainblock for the current band
413 * @param num_bands amount of coded bands
415 static int decode_gain_control(GetBitContext *gb, GainBlock *block,
421 GainInfo *gain = block->g_block;
423 for (i = 0; i <= num_bands; i++) {
424 num_data = get_bits(gb, 3);
425 gain[i].num_gain_data = num_data;
426 level = gain[i].lev_code;
427 loc = gain[i].loc_code;
429 for (cf = 0; cf < gain[i].num_gain_data; cf++) {
430 level[cf] = get_bits(gb, 4);
431 loc [cf] = get_bits(gb, 5);
432 if (cf && loc[cf] <= loc[cf - 1])
433 return AVERROR_INVALIDDATA;
437 /* Clear the unused blocks. */
439 gain[i].num_gain_data = 0;
445 * Apply gain parameters and perform the MDCT overlapping part
447 * @param input input buffer
448 * @param prev previous buffer to perform overlap against
449 * @param output output buffer
450 * @param gain1 current band gain info
451 * @param gain2 next band gain info
453 static void gain_compensate_and_overlap(float *input, float *prev,
454 float *output, GainInfo *gain1,
457 float g1, g2, gain_inc;
458 int i, j, num_data, start_loc, end_loc;
461 if (gain2->num_gain_data == 0)
464 g1 = gain_tab1[gain2->lev_code[0]];
466 if (gain1->num_gain_data == 0) {
467 for (i = 0; i < 256; i++)
468 output[i] = input[i] * g1 + prev[i];
470 num_data = gain1->num_gain_data;
471 gain1->loc_code[num_data] = 32;
472 gain1->lev_code[num_data] = 4;
474 for (i = 0, j = 0; i < num_data; i++) {
475 start_loc = gain1->loc_code[i] * 8;
476 end_loc = start_loc + 8;
478 g2 = gain_tab1[gain1->lev_code[i]];
479 gain_inc = gain_tab2[gain1->lev_code[i + 1] -
480 gain1->lev_code[i ] + 15];
483 for (; j < start_loc; j++)
484 output[j] = (input[j] * g1 + prev[j]) * g2;
486 /* interpolation is done over eight samples */
487 for (; j < end_loc; j++) {
488 output[j] = (input[j] * g1 + prev[j]) * g2;
494 output[j] = input[j] * g1 + prev[j];
497 /* Delay for the overlapping part. */
498 memcpy(prev, &input[256], 256 * sizeof(*prev));
502 * Combine the tonal band spectrum and regular band spectrum
504 * @param spectrum output spectrum buffer
505 * @param num_components number of tonal components
506 * @param components tonal components for this band
507 * @return position of the last tonal coefficient
509 static int add_tonal_components(float *spectrum, int num_components,
510 TonalComponent *components)
512 int i, j, last_pos = -1;
513 float *input, *output;
515 for (i = 0; i < num_components; i++) {
516 last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
517 input = components[i].coef;
518 output = &spectrum[components[i].pos];
520 for (j = 0; j < components[i].num_coefs; j++)
521 output[i] += input[i];
527 #define INTERPOLATE(old, new, nsample) \
528 ((old) + (nsample) * 0.125 * ((new) - (old)))
530 static void reverse_matrixing(float *su1, float *su2, int *prev_code,
533 int i, nsample, band;
534 float mc1_l, mc1_r, mc2_l, mc2_r;
536 for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
537 int s1 = prev_code[i];
538 int s2 = curr_code[i];
542 /* Selector value changed, interpolation needed. */
543 mc1_l = matrix_coeffs[s1 * 2 ];
544 mc1_r = matrix_coeffs[s1 * 2 + 1];
545 mc2_l = matrix_coeffs[s2 * 2 ];
546 mc2_r = matrix_coeffs[s2 * 2 + 1];
548 /* Interpolation is done over the first eight samples. */
549 for (; nsample < band + 8; nsample++) {
550 float c1 = su1[nsample];
551 float c2 = su2[nsample];
552 c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
553 c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
555 su2[nsample] = c1 * 2.0 - c2;
559 /* Apply the matrix without interpolation. */
561 case 0: /* M/S decoding */
562 for (; nsample < band + 256; nsample++) {
563 float c1 = su1[nsample];
564 float c2 = su2[nsample];
565 su1[nsample] = c2 * 2.0;
566 su2[nsample] = (c1 - c2) * 2.0;
570 for (; nsample < band + 256; nsample++) {
571 float c1 = su1[nsample];
572 float c2 = su2[nsample];
573 su1[nsample] = (c1 + c2) * 2.0;
574 su2[nsample] = c2 * -2.0;
579 for (; nsample < band + 256; nsample++) {
580 float c1 = su1[nsample];
581 float c2 = su2[nsample];
582 su1[nsample] = c1 + c2;
583 su2[nsample] = c1 - c2;
592 static void get_channel_weights(int index, int flag, float ch[2])
598 ch[0] = (index & 7) / 7.0;
599 ch[1] = sqrt(2 - ch[0] * ch[0]);
601 FFSWAP(float, ch[0], ch[1]);
605 static void channel_weighting(float *su1, float *su2, int *p3)
608 /* w[x][y] y=0 is left y=1 is right */
611 if (p3[1] != 7 || p3[3] != 7) {
612 get_channel_weights(p3[1], p3[0], w[0]);
613 get_channel_weights(p3[3], p3[2], w[1]);
615 for (band = 256; band < 4 * 256; band += 256) {
616 for (nsample = band; nsample < band + 8; nsample++) {
617 su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
618 su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
620 for(; nsample < band + 256; nsample++) {
621 su1[nsample] *= w[1][0];
622 su2[nsample] *= w[1][1];
629 * Decode a Sound Unit
631 * @param snd the channel unit to be used
632 * @param output the decoded samples before IQMF in float representation
633 * @param channel_num channel number
634 * @param coding_mode the coding mode (JOINT_STEREO or regular stereo/mono)
636 static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb,
637 ChannelUnit *snd, float *output,
638 int channel_num, int coding_mode)
640 int band, ret, num_subbands, last_tonal, num_bands;
641 GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch];
642 GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
644 if (coding_mode == JOINT_STEREO && channel_num == 1) {
645 if (get_bits(gb, 2) != 3) {
646 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
647 return AVERROR_INVALIDDATA;
650 if (get_bits(gb, 6) != 0x28) {
651 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
652 return AVERROR_INVALIDDATA;
656 /* number of coded QMF bands */
657 snd->bands_coded = get_bits(gb, 2);
659 ret = decode_gain_control(gb, gain2, snd->bands_coded);
663 snd->num_components = decode_tonal_components(gb, snd->components,
665 if (snd->num_components == -1)
668 num_subbands = decode_spectrum(gb, snd->spectrum);
670 /* Merge the decoded spectrum and tonal components. */
671 last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
675 /* calculate number of used MLT/QMF bands according to the amount of coded
677 num_bands = (subband_tab[num_subbands] - 1) >> 8;
679 num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
682 /* Reconstruct time domain samples. */
683 for (band = 0; band < 4; band++) {
684 /* Perform the IMDCT step without overlapping. */
685 if (band <= num_bands)
686 imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
688 memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf));
690 /* gain compensation and overlapping */
691 gain_compensate_and_overlap(snd->imdct_buf,
692 &snd->prev_frame[band * 256],
694 &gain1->g_block[band],
695 &gain2->g_block[band]);
698 /* Swap the gain control buffers for the next frame. */
699 snd->gc_blk_switch ^= 1;
704 static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
707 ATRAC3Context *q = avctx->priv_data;
711 if (q->coding_mode == JOINT_STEREO) {
712 /* channel coupling mode */
713 /* decode Sound Unit 1 */
714 init_get_bits(&q->gb, databuf, avctx->block_align * 8);
716 ret = decode_channel_sound_unit(q, &q->gb, q->units, out_samples[0], 0,
721 /* Framedata of the su2 in the joint-stereo mode is encoded in
722 * reverse byte order so we need to swap it first. */
723 if (databuf == q->decoded_bytes_buffer) {
724 uint8_t *ptr2 = q->decoded_bytes_buffer + avctx->block_align - 1;
725 ptr1 = q->decoded_bytes_buffer;
726 for (i = 0; i < avctx->block_align / 2; i++, ptr1++, ptr2--)
727 FFSWAP(uint8_t, *ptr1, *ptr2);
729 const uint8_t *ptr2 = databuf + avctx->block_align - 1;
730 for (i = 0; i < avctx->block_align; i++)
731 q->decoded_bytes_buffer[i] = *ptr2--;
734 /* Skip the sync codes (0xF8). */
735 ptr1 = q->decoded_bytes_buffer;
736 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
737 if (i >= avctx->block_align)
738 return AVERROR_INVALIDDATA;
742 /* set the bitstream reader at the start of the second Sound Unit*/
743 init_get_bits(&q->gb, ptr1, avctx->block_align * 8);
745 /* Fill the Weighting coeffs delay buffer */
746 memmove(q->weighting_delay, &q->weighting_delay[2],
747 4 * sizeof(*q->weighting_delay));
748 q->weighting_delay[4] = get_bits1(&q->gb);
749 q->weighting_delay[5] = get_bits(&q->gb, 3);
751 for (i = 0; i < 4; i++) {
752 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
753 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
754 q->matrix_coeff_index_next[i] = get_bits(&q->gb, 2);
757 /* Decode Sound Unit 2. */
758 ret = decode_channel_sound_unit(q, &q->gb, &q->units[1],
759 out_samples[1], 1, JOINT_STEREO);
763 /* Reconstruct the channel coefficients. */
764 reverse_matrixing(out_samples[0], out_samples[1],
765 q->matrix_coeff_index_prev,
766 q->matrix_coeff_index_now);
768 channel_weighting(out_samples[0], out_samples[1], q->weighting_delay);
770 /* normal stereo mode or mono */
771 /* Decode the channel sound units. */
772 for (i = 0; i < avctx->channels; i++) {
773 /* Set the bitstream reader at the start of a channel sound unit. */
774 init_get_bits(&q->gb,
775 databuf + i * avctx->block_align / avctx->channels,
776 avctx->block_align * 8 / avctx->channels);
778 ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
779 out_samples[i], i, q->coding_mode);
785 /* Apply the iQMF synthesis filter. */
786 for (i = 0; i < avctx->channels; i++) {
787 float *p1 = out_samples[i];
788 float *p2 = p1 + 256;
789 float *p3 = p2 + 256;
790 float *p4 = p3 + 256;
791 ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
792 ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
793 ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
799 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
800 int *got_frame_ptr, AVPacket *avpkt)
802 const uint8_t *buf = avpkt->data;
803 int buf_size = avpkt->size;
804 ATRAC3Context *q = avctx->priv_data;
806 const uint8_t *databuf;
808 if (buf_size < avctx->block_align) {
809 av_log(avctx, AV_LOG_ERROR,
810 "Frame too small (%d bytes). Truncated file?\n", buf_size);
811 return AVERROR_INVALIDDATA;
814 /* get output buffer */
815 q->frame.nb_samples = SAMPLES_PER_FRAME;
816 if ((ret = ff_get_buffer(avctx, &q->frame)) < 0) {
817 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
821 /* Check if we need to descramble and what buffer to pass on. */
822 if (q->scrambled_stream) {
823 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
824 databuf = q->decoded_bytes_buffer;
829 ret = decode_frame(avctx, databuf, (float **)q->frame.extended_data);
831 av_log(NULL, AV_LOG_ERROR, "Frame decoding error!\n");
836 *(AVFrame *)data = q->frame;
838 return avctx->block_align;
841 static void atrac3_init_static_data(void)
845 init_atrac3_window();
846 ff_atrac_generate_tables();
848 /* Initialize the VLC tables. */
849 for (i = 0; i < 7; i++) {
850 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
851 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] -
853 init_vlc(&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
855 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
858 /* Generate gain tables */
859 for (i = 0; i < 16; i++)
860 gain_tab1[i] = exp2f (4 - i);
862 for (i = -15; i < 16; i++)
863 gain_tab2[i + 15] = exp2f (i * -0.125);
866 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
868 static int static_init_done;
870 int version, delay, samples_per_frame, frame_factor;
871 const uint8_t *edata_ptr = avctx->extradata;
872 ATRAC3Context *q = avctx->priv_data;
874 if (avctx->channels <= 0 || avctx->channels > 2) {
875 av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n");
876 return AVERROR(EINVAL);
879 if (!static_init_done)
880 atrac3_init_static_data();
881 static_init_done = 1;
883 /* Take care of the codec-specific extradata. */
884 if (avctx->extradata_size == 14) {
885 /* Parse the extradata, WAV format */
886 av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
887 bytestream_get_le16(&edata_ptr)); // Unknown value always 1
888 edata_ptr += 4; // samples per channel
889 q->coding_mode = bytestream_get_le16(&edata_ptr);
890 av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
891 bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
892 frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1
893 av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
894 bytestream_get_le16(&edata_ptr)); // Unknown always 0
897 samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
900 q->coding_mode = q->coding_mode ? JOINT_STEREO : STEREO;
901 q->scrambled_stream = 0;
903 if (avctx->block_align != 96 * avctx->channels * frame_factor &&
904 avctx->block_align != 152 * avctx->channels * frame_factor &&
905 avctx->block_align != 192 * avctx->channels * frame_factor) {
906 av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
907 "configuration %d/%d/%d\n", avctx->block_align,
908 avctx->channels, frame_factor);
909 return AVERROR_INVALIDDATA;
911 } else if (avctx->extradata_size == 10) {
912 /* Parse the extradata, RM format. */
913 version = bytestream_get_be32(&edata_ptr);
914 samples_per_frame = bytestream_get_be16(&edata_ptr);
915 delay = bytestream_get_be16(&edata_ptr);
916 q->coding_mode = bytestream_get_be16(&edata_ptr);
917 q->scrambled_stream = 1;
920 av_log(NULL, AV_LOG_ERROR, "Unknown extradata size %d.\n",
921 avctx->extradata_size);
922 return AVERROR(EINVAL);
925 if (q->coding_mode == JOINT_STEREO && avctx->channels < 2) {
926 av_log(avctx, AV_LOG_ERROR, "Invalid coding mode\n");
927 return AVERROR_INVALIDDATA;
930 /* Check the extradata */
933 av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
934 return AVERROR_INVALIDDATA;
937 if (samples_per_frame != SAMPLES_PER_FRAME &&
938 samples_per_frame != SAMPLES_PER_FRAME * 2) {
939 av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
941 return AVERROR_INVALIDDATA;
944 if (delay != 0x88E) {
945 av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
947 return AVERROR_INVALIDDATA;
950 if (q->coding_mode == STEREO)
951 av_log(avctx, AV_LOG_DEBUG, "Normal stereo detected.\n");
952 else if (q->coding_mode == JOINT_STEREO)
953 av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
955 av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
957 return AVERROR_INVALIDDATA;
960 if (avctx->block_align >= UINT_MAX / 2)
961 return AVERROR(EINVAL);
963 q->decoded_bytes_buffer = av_mallocz(FFALIGN(avctx->block_align, 4) +
964 FF_INPUT_BUFFER_PADDING_SIZE);
965 if (q->decoded_bytes_buffer == NULL)
966 return AVERROR(ENOMEM);
968 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
970 /* initialize the MDCT transform */
971 if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) {
972 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
973 av_freep(&q->decoded_bytes_buffer);
977 /* init the joint-stereo decoding data */
978 q->weighting_delay[0] = 0;
979 q->weighting_delay[1] = 7;
980 q->weighting_delay[2] = 0;
981 q->weighting_delay[3] = 7;
982 q->weighting_delay[4] = 0;
983 q->weighting_delay[5] = 7;
985 for (i = 0; i < 4; i++) {
986 q->matrix_coeff_index_prev[i] = 3;
987 q->matrix_coeff_index_now[i] = 3;
988 q->matrix_coeff_index_next[i] = 3;
991 avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
992 ff_fmt_convert_init(&q->fmt_conv, avctx);
994 q->units = av_mallocz(sizeof(*q->units) * avctx->channels);
996 atrac3_decode_close(avctx);
997 return AVERROR(ENOMEM);
1000 avcodec_get_frame_defaults(&q->frame);
1001 avctx->coded_frame = &q->frame;
1006 AVCodec ff_atrac3_decoder = {
1008 .type = AVMEDIA_TYPE_AUDIO,
1009 .id = AV_CODEC_ID_ATRAC3,
1010 .priv_data_size = sizeof(ATRAC3Context),
1011 .init = atrac3_decode_init,
1012 .close = atrac3_decode_close,
1013 .decode = atrac3_decode_frame,
1014 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1015 .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),
1016 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1017 AV_SAMPLE_FMT_NONE },