2 * ATRAC3 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 * 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"
41 #include "libavutil/libm.h"
43 #include "bytestream.h"
49 #include "atrac3data.h"
51 #define MIN_CHANNELS 1
52 #define MAX_CHANNELS 8
53 #define MAX_JS_PAIRS 8 / 2
55 #define JOINT_STEREO 0x12
58 #define SAMPLES_PER_FRAME 1024
61 #define ATRAC3_VLC_BITS 8
63 typedef struct GainBlock {
64 AtracGainInfo g_block[4];
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[MAX_JS_PAIRS][4];
100 int matrix_coeff_index_now[MAX_JS_PAIRS][4];
101 int matrix_coeff_index_next[MAX_JS_PAIRS][4];
102 int weighting_delay[MAX_JS_PAIRS][6];
106 uint8_t *decoded_bytes_buffer;
107 float temp_buf[1070];
111 int scrambled_stream;
114 AtracGCContext gainc_ctx;
116 void (*vector_fmul)(float *dst, const float *src0, const float *src1,
120 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
121 static VLC_TYPE atrac3_vlc_table[7 * 1 << ATRAC3_VLC_BITS][2];
122 static VLC spectral_coeff_tab[7];
125 * Regular 512 points IMDCT without overlapping, with the exception of the
126 * swapping of odd bands caused by the reverse spectra of the QMF.
128 * @param odd_band 1 if the band is an odd band
130 static void imlt(ATRAC3Context *q, float *input, float *output, int odd_band)
136 * Reverse the odd bands before IMDCT, this is an effect of the QMF
137 * transform or it gives better compression to do it this way.
138 * FIXME: It should be possible to handle this in imdct_calc
139 * for that to happen a modification of the prerotation step of
140 * all SIMD code and C code is needed.
141 * Or fix the functions before so they generate a pre reversed spectrum.
143 for (i = 0; i < 128; i++)
144 FFSWAP(float, input[i], input[255 - i]);
147 q->mdct_ctx.imdct_calc(&q->mdct_ctx, output, input);
149 /* Perform windowing on the output. */
150 q->vector_fmul(output, output, mdct_window, MDCT_SIZE);
154 * indata descrambling, only used for data coming from the rm container
156 static int decode_bytes(const uint8_t *input, uint8_t *out, int bytes)
161 uint32_t *output = (uint32_t *)out;
163 off = (intptr_t)input & 3;
164 buf = (const uint32_t *)(input - off);
166 c = av_be2ne32((0x537F6103U >> (off * 8)) | (0x537F6103U << (32 - (off * 8))));
168 c = av_be2ne32(0x537F6103U);
170 for (i = 0; i < bytes / 4; i++)
171 output[i] = c ^ buf[i];
174 avpriv_request_sample(NULL, "Offset of %d", off);
179 static av_cold void init_imdct_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_freep(&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 AtracGainInfo *gain = block->g_block;
423 for (b = 0; b <= num_bands; b++) {
424 gain[b].num_points = get_bits(gb, 3);
425 level = gain[b].lev_code;
426 loc = gain[b].loc_code;
428 for (j = 0; j < gain[b].num_points; j++) {
429 level[j] = get_bits(gb, 4);
430 loc[j] = get_bits(gb, 5);
431 if (j && loc[j] <= loc[j - 1])
432 return AVERROR_INVALIDDATA;
436 /* Clear the unused blocks. */
438 gain[b].num_points = 0;
444 * Combine the tonal band spectrum and regular band spectrum
446 * @param spectrum output spectrum buffer
447 * @param num_components number of tonal components
448 * @param components tonal components for this band
449 * @return position of the last tonal coefficient
451 static int add_tonal_components(float *spectrum, int num_components,
452 TonalComponent *components)
454 int i, j, last_pos = -1;
455 float *input, *output;
457 for (i = 0; i < num_components; i++) {
458 last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
459 input = components[i].coef;
460 output = &spectrum[components[i].pos];
462 for (j = 0; j < components[i].num_coefs; j++)
463 output[j] += input[j];
469 #define INTERPOLATE(old, new, nsample) \
470 ((old) + (nsample) * 0.125 * ((new) - (old)))
472 static void reverse_matrixing(float *su1, float *su2, int *prev_code,
475 int i, nsample, band;
476 float mc1_l, mc1_r, mc2_l, mc2_r;
478 for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
479 int s1 = prev_code[i];
480 int s2 = curr_code[i];
484 /* Selector value changed, interpolation needed. */
485 mc1_l = matrix_coeffs[s1 * 2 ];
486 mc1_r = matrix_coeffs[s1 * 2 + 1];
487 mc2_l = matrix_coeffs[s2 * 2 ];
488 mc2_r = matrix_coeffs[s2 * 2 + 1];
490 /* Interpolation is done over the first eight samples. */
491 for (; nsample < band + 8; nsample++) {
492 float c1 = su1[nsample];
493 float c2 = su2[nsample];
494 c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
495 c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
497 su2[nsample] = c1 * 2.0 - c2;
501 /* Apply the matrix without interpolation. */
503 case 0: /* M/S decoding */
504 for (; nsample < band + 256; nsample++) {
505 float c1 = su1[nsample];
506 float c2 = su2[nsample];
507 su1[nsample] = c2 * 2.0;
508 su2[nsample] = (c1 - c2) * 2.0;
512 for (; nsample < band + 256; nsample++) {
513 float c1 = su1[nsample];
514 float c2 = su2[nsample];
515 su1[nsample] = (c1 + c2) * 2.0;
516 su2[nsample] = c2 * -2.0;
521 for (; nsample < band + 256; nsample++) {
522 float c1 = su1[nsample];
523 float c2 = su2[nsample];
524 su1[nsample] = c1 + c2;
525 su2[nsample] = c1 - c2;
534 static void get_channel_weights(int index, int flag, float ch[2])
540 ch[0] = (index & 7) / 7.0;
541 ch[1] = sqrt(2 - ch[0] * ch[0]);
543 FFSWAP(float, ch[0], ch[1]);
547 static void channel_weighting(float *su1, float *su2, int *p3)
550 /* w[x][y] y=0 is left y=1 is right */
553 if (p3[1] != 7 || p3[3] != 7) {
554 get_channel_weights(p3[1], p3[0], w[0]);
555 get_channel_weights(p3[3], p3[2], w[1]);
557 for (band = 256; band < 4 * 256; band += 256) {
558 for (nsample = band; nsample < band + 8; nsample++) {
559 su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
560 su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
562 for(; nsample < band + 256; nsample++) {
563 su1[nsample] *= w[1][0];
564 su2[nsample] *= w[1][1];
571 * Decode a Sound Unit
573 * @param snd the channel unit to be used
574 * @param output the decoded samples before IQMF in float representation
575 * @param channel_num channel number
576 * @param coding_mode the coding mode (JOINT_STEREO or single channels)
578 static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb,
579 ChannelUnit *snd, float *output,
580 int channel_num, int coding_mode)
582 int band, ret, num_subbands, last_tonal, num_bands;
583 GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch];
584 GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
586 if (coding_mode == JOINT_STEREO && (channel_num % 2) == 1) {
587 if (get_bits(gb, 2) != 3) {
588 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
589 return AVERROR_INVALIDDATA;
592 if (get_bits(gb, 6) != 0x28) {
593 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
594 return AVERROR_INVALIDDATA;
598 /* number of coded QMF bands */
599 snd->bands_coded = get_bits(gb, 2);
601 ret = decode_gain_control(gb, gain2, snd->bands_coded);
605 snd->num_components = decode_tonal_components(gb, snd->components,
607 if (snd->num_components < 0)
608 return snd->num_components;
610 num_subbands = decode_spectrum(gb, snd->spectrum);
612 /* Merge the decoded spectrum and tonal components. */
613 last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
617 /* calculate number of used MLT/QMF bands according to the amount of coded
619 num_bands = (subband_tab[num_subbands] - 1) >> 8;
621 num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
624 /* Reconstruct time domain samples. */
625 for (band = 0; band < 4; band++) {
626 /* Perform the IMDCT step without overlapping. */
627 if (band <= num_bands)
628 imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
630 memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf));
632 /* gain compensation and overlapping */
633 ff_atrac_gain_compensation(&q->gainc_ctx, snd->imdct_buf,
634 &snd->prev_frame[band * 256],
635 &gain1->g_block[band], &gain2->g_block[band],
636 256, &output[band * 256]);
639 /* Swap the gain control buffers for the next frame. */
640 snd->gc_blk_switch ^= 1;
645 static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
648 ATRAC3Context *q = avctx->priv_data;
652 if (q->coding_mode == JOINT_STEREO) {
653 /* channel coupling mode */
655 /* Decode sound unit pairs (channels are expected to be even).
656 * Multichannel joint stereo interleaves pairs (6ch: 2ch + 2ch + 2ch) */
657 const uint8_t *js_databuf;
658 int js_pair, js_block_align;
660 js_block_align = (avctx->block_align / avctx->channels) * 2; /* block pair */
662 for (ch = 0; ch < avctx->channels; ch = ch + 2) {
664 js_databuf = databuf + js_pair * js_block_align; /* align to current pair */
666 /* Set the bitstream reader at the start of first channel sound unit. */
667 init_get_bits(&q->gb,
668 js_databuf, js_block_align * 8);
670 /* decode Sound Unit 1 */
671 ret = decode_channel_sound_unit(q, &q->gb, &q->units[ch],
672 out_samples[ch], ch, JOINT_STEREO);
676 /* Framedata of the su2 in the joint-stereo mode is encoded in
677 * reverse byte order so we need to swap it first. */
678 if (js_databuf == q->decoded_bytes_buffer) {
679 uint8_t *ptr2 = q->decoded_bytes_buffer + js_block_align - 1;
680 ptr1 = q->decoded_bytes_buffer;
681 for (i = 0; i < js_block_align / 2; i++, ptr1++, ptr2--)
682 FFSWAP(uint8_t, *ptr1, *ptr2);
684 const uint8_t *ptr2 = js_databuf + js_block_align - 1;
685 for (i = 0; i < js_block_align; i++)
686 q->decoded_bytes_buffer[i] = *ptr2--;
689 /* Skip the sync codes (0xF8). */
690 ptr1 = q->decoded_bytes_buffer;
691 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
692 if (i >= js_block_align)
693 return AVERROR_INVALIDDATA;
697 /* set the bitstream reader at the start of the second Sound Unit */
698 ret = init_get_bits8(&q->gb,
699 ptr1, q->decoded_bytes_buffer + js_block_align - ptr1);
703 /* Fill the Weighting coeffs delay buffer */
704 memmove(q->weighting_delay[js_pair], &q->weighting_delay[js_pair][2],
705 4 * sizeof(*q->weighting_delay[js_pair]));
706 q->weighting_delay[js_pair][4] = get_bits1(&q->gb);
707 q->weighting_delay[js_pair][5] = get_bits(&q->gb, 3);
709 for (i = 0; i < 4; i++) {
710 q->matrix_coeff_index_prev[js_pair][i] = q->matrix_coeff_index_now[js_pair][i];
711 q->matrix_coeff_index_now[js_pair][i] = q->matrix_coeff_index_next[js_pair][i];
712 q->matrix_coeff_index_next[js_pair][i] = get_bits(&q->gb, 2);
715 /* Decode Sound Unit 2. */
716 ret = decode_channel_sound_unit(q, &q->gb, &q->units[ch+1],
717 out_samples[ch+1], ch+1, JOINT_STEREO);
721 /* Reconstruct the channel coefficients. */
722 reverse_matrixing(out_samples[ch], out_samples[ch+1],
723 q->matrix_coeff_index_prev[js_pair],
724 q->matrix_coeff_index_now[js_pair]);
726 channel_weighting(out_samples[ch], out_samples[ch+1], q->weighting_delay[js_pair]);
729 /* single channels */
730 /* Decode the channel sound units. */
731 for (i = 0; i < avctx->channels; i++) {
732 /* Set the bitstream reader at the start of a channel sound unit. */
733 init_get_bits(&q->gb,
734 databuf + i * avctx->block_align / avctx->channels,
735 avctx->block_align * 8 / avctx->channels);
737 ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
738 out_samples[i], i, q->coding_mode);
744 /* Apply the iQMF synthesis filter. */
745 for (i = 0; i < avctx->channels; i++) {
746 float *p1 = out_samples[i];
747 float *p2 = p1 + 256;
748 float *p3 = p2 + 256;
749 float *p4 = p3 + 256;
750 ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
751 ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
752 ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
758 static int al_decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
759 int size, float **out_samples)
761 ATRAC3Context *q = avctx->priv_data;
764 /* Set the bitstream reader at the start of a channel sound unit. */
765 init_get_bits(&q->gb, databuf, size * 8);
766 /* single channels */
767 /* Decode the channel sound units. */
768 for (i = 0; i < avctx->channels; i++) {
769 ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
770 out_samples[i], i, q->coding_mode);
773 while (i < avctx->channels && get_bits_left(&q->gb) > 6 && show_bits(&q->gb, 6) != 0x28) {
774 skip_bits(&q->gb, 1);
778 /* Apply the iQMF synthesis filter. */
779 for (i = 0; i < avctx->channels; i++) {
780 float *p1 = out_samples[i];
781 float *p2 = p1 + 256;
782 float *p3 = p2 + 256;
783 float *p4 = p3 + 256;
784 ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
785 ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
786 ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
792 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
793 int *got_frame_ptr, AVPacket *avpkt)
795 AVFrame *frame = data;
796 const uint8_t *buf = avpkt->data;
797 int buf_size = avpkt->size;
798 ATRAC3Context *q = avctx->priv_data;
800 const uint8_t *databuf;
802 if (buf_size < avctx->block_align) {
803 av_log(avctx, AV_LOG_ERROR,
804 "Frame too small (%d bytes). Truncated file?\n", buf_size);
805 return AVERROR_INVALIDDATA;
808 /* get output buffer */
809 frame->nb_samples = SAMPLES_PER_FRAME;
810 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
813 /* Check if we need to descramble and what buffer to pass on. */
814 if (q->scrambled_stream) {
815 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
816 databuf = q->decoded_bytes_buffer;
821 ret = decode_frame(avctx, databuf, (float **)frame->extended_data);
823 av_log(avctx, AV_LOG_ERROR, "Frame decoding error!\n");
829 return avctx->block_align;
832 static int atrac3al_decode_frame(AVCodecContext *avctx, void *data,
833 int *got_frame_ptr, AVPacket *avpkt)
835 AVFrame *frame = data;
838 frame->nb_samples = SAMPLES_PER_FRAME;
839 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
842 ret = al_decode_frame(avctx, avpkt->data, avpkt->size,
843 (float **)frame->extended_data);
845 av_log(avctx, AV_LOG_ERROR, "Frame decoding error!\n");
854 static av_cold void atrac3_init_static_data(void)
856 VLC_TYPE (*table)[2] = atrac3_vlc_table;
860 ff_atrac_generate_tables();
862 /* Initialize the VLC tables. */
863 for (i = 0; i < 7; i++) {
864 spectral_coeff_tab[i].table = table;
865 spectral_coeff_tab[i].table_allocated = 256;
866 init_vlc(&spectral_coeff_tab[i], ATRAC3_VLC_BITS, huff_tab_sizes[i],
868 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
873 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
875 static int static_init_done;
877 int version, delay, samples_per_frame, frame_factor;
878 const uint8_t *edata_ptr = avctx->extradata;
879 ATRAC3Context *q = avctx->priv_data;
880 AVFloatDSPContext *fdsp;
882 if (avctx->channels < MIN_CHANNELS || avctx->channels > MAX_CHANNELS) {
883 av_log(avctx, AV_LOG_ERROR, "Channel configuration error!\n");
884 return AVERROR(EINVAL);
887 if (!static_init_done)
888 atrac3_init_static_data();
889 static_init_done = 1;
891 /* Take care of the codec-specific extradata. */
892 if (avctx->codec_id == AV_CODEC_ID_ATRAC3AL) {
894 samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
896 q->coding_mode = SINGLE;
897 } else if (avctx->extradata_size == 14) {
898 /* Parse the extradata, WAV format */
899 av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
900 bytestream_get_le16(&edata_ptr)); // Unknown value always 1
901 edata_ptr += 4; // samples per channel
902 q->coding_mode = bytestream_get_le16(&edata_ptr);
903 av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
904 bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
905 frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1
906 av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
907 bytestream_get_le16(&edata_ptr)); // Unknown always 0
910 samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
913 q->coding_mode = q->coding_mode ? JOINT_STEREO : SINGLE;
914 q->scrambled_stream = 0;
916 if (avctx->block_align != 96 * avctx->channels * frame_factor &&
917 avctx->block_align != 152 * avctx->channels * frame_factor &&
918 avctx->block_align != 192 * avctx->channels * frame_factor) {
919 av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
920 "configuration %d/%d/%d\n", avctx->block_align,
921 avctx->channels, frame_factor);
922 return AVERROR_INVALIDDATA;
924 } else if (avctx->extradata_size == 12 || avctx->extradata_size == 10) {
925 /* Parse the extradata, RM format. */
926 version = bytestream_get_be32(&edata_ptr);
927 samples_per_frame = bytestream_get_be16(&edata_ptr);
928 delay = bytestream_get_be16(&edata_ptr);
929 q->coding_mode = bytestream_get_be16(&edata_ptr);
930 q->scrambled_stream = 1;
933 av_log(avctx, AV_LOG_ERROR, "Unknown extradata size %d.\n",
934 avctx->extradata_size);
935 return AVERROR(EINVAL);
938 /* Check the extradata */
941 av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
942 return AVERROR_INVALIDDATA;
945 if (samples_per_frame != SAMPLES_PER_FRAME * avctx->channels) {
946 av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
948 return AVERROR_INVALIDDATA;
951 if (delay != 0x88E) {
952 av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
954 return AVERROR_INVALIDDATA;
957 if (q->coding_mode == SINGLE)
958 av_log(avctx, AV_LOG_DEBUG, "Single channels detected.\n");
959 else if (q->coding_mode == JOINT_STEREO) {
960 if (avctx->channels % 2 == 1) { /* Joint stereo channels must be even */
961 av_log(avctx, AV_LOG_ERROR, "Invalid joint stereo channel configuration.\n");
962 return AVERROR_INVALIDDATA;
964 av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
966 av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
968 return AVERROR_INVALIDDATA;
971 if (avctx->block_align > 1024 || avctx->block_align <= 0)
972 return AVERROR(EINVAL);
974 q->decoded_bytes_buffer = av_mallocz(FFALIGN(avctx->block_align, 4) +
975 AV_INPUT_BUFFER_PADDING_SIZE);
976 if (!q->decoded_bytes_buffer)
977 return AVERROR(ENOMEM);
979 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
981 /* initialize the MDCT transform */
982 if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) {
983 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
987 /* init the joint-stereo decoding data */
988 for (js_pair = 0; js_pair < MAX_JS_PAIRS; js_pair++) {
989 q->weighting_delay[js_pair][0] = 0;
990 q->weighting_delay[js_pair][1] = 7;
991 q->weighting_delay[js_pair][2] = 0;
992 q->weighting_delay[js_pair][3] = 7;
993 q->weighting_delay[js_pair][4] = 0;
994 q->weighting_delay[js_pair][5] = 7;
996 for (i = 0; i < 4; i++) {
997 q->matrix_coeff_index_prev[js_pair][i] = 3;
998 q->matrix_coeff_index_now[js_pair][i] = 3;
999 q->matrix_coeff_index_next[js_pair][i] = 3;
1003 ff_atrac_init_gain_compensation(&q->gainc_ctx, 4, 3);
1004 fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1006 return AVERROR(ENOMEM);
1007 q->vector_fmul = fdsp->vector_fmul;
1010 q->units = av_mallocz_array(avctx->channels, sizeof(*q->units));
1012 return AVERROR(ENOMEM);
1017 AVCodec ff_atrac3_decoder = {
1019 .long_name = NULL_IF_CONFIG_SMALL("ATRAC3 (Adaptive TRansform Acoustic Coding 3)"),
1020 .type = AVMEDIA_TYPE_AUDIO,
1021 .id = AV_CODEC_ID_ATRAC3,
1022 .priv_data_size = sizeof(ATRAC3Context),
1023 .init = atrac3_decode_init,
1024 .close = atrac3_decode_close,
1025 .decode = atrac3_decode_frame,
1026 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1027 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1028 AV_SAMPLE_FMT_NONE },
1029 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
1032 AVCodec ff_atrac3al_decoder = {
1034 .long_name = NULL_IF_CONFIG_SMALL("ATRAC3 AL (Adaptive TRansform Acoustic Coding 3 Advanced Lossless)"),
1035 .type = AVMEDIA_TYPE_AUDIO,
1036 .id = AV_CODEC_ID_ATRAC3AL,
1037 .priv_data_size = sizeof(ATRAC3Context),
1038 .init = atrac3_decode_init,
1039 .close = atrac3_decode_close,
1040 .decode = atrac3al_decode_frame,
1041 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1042 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1043 AV_SAMPLE_FMT_NONE },
1044 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,