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"
42 #include "libavutil/thread.h"
45 #include "bytestream.h"
51 #include "atrac3data.h"
53 #define MIN_CHANNELS 1
54 #define MAX_CHANNELS 8
55 #define MAX_JS_PAIRS 8 / 2
57 #define JOINT_STEREO 0x12
60 #define SAMPLES_PER_FRAME 1024
63 #define ATRAC3_VLC_BITS 8
65 typedef struct GainBlock {
66 AtracGainInfo g_block[4];
69 typedef struct TonalComponent {
75 typedef struct ChannelUnit {
78 float prev_frame[SAMPLES_PER_FRAME];
80 TonalComponent components[64];
81 GainBlock gain_block[2];
83 DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
84 DECLARE_ALIGNED(32, float, imdct_buf)[SAMPLES_PER_FRAME];
86 float delay_buf1[46]; ///<qmf delay buffers
91 typedef struct ATRAC3Context {
100 /** joint-stereo related variables */
101 int matrix_coeff_index_prev[MAX_JS_PAIRS][4];
102 int matrix_coeff_index_now[MAX_JS_PAIRS][4];
103 int matrix_coeff_index_next[MAX_JS_PAIRS][4];
104 int weighting_delay[MAX_JS_PAIRS][6];
108 uint8_t *decoded_bytes_buffer;
109 float temp_buf[1070];
113 int scrambled_stream;
116 AtracGCContext gainc_ctx;
118 void (*vector_fmul)(float *dst, const float *src0, const float *src1,
122 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
123 static VLC_TYPE atrac3_vlc_table[7 * 1 << ATRAC3_VLC_BITS][2];
124 static VLC spectral_coeff_tab[7];
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->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_imdct_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_freep(&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 mantissas[i] = get_vlc2(gb, spectral_coeff_tab[selector-1].table,
255 for (i = 0; i < num_codes; i++) {
256 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table,
258 mantissas[i * 2 ] = mantissa_vlc_tab[huff_symb * 2 ];
259 mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1];
266 * Restore the quantized band spectrum coefficients
268 * @return subband count, fix for broken specification/files
270 static int decode_spectrum(GetBitContext *gb, float *output)
272 int num_subbands, coding_mode, i, j, first, last, subband_size;
273 int subband_vlc_index[32], sf_index[32];
277 num_subbands = get_bits(gb, 5); // number of coded subbands
278 coding_mode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
280 /* get the VLC selector table for the subbands, 0 means not coded */
281 for (i = 0; i <= num_subbands; i++)
282 subband_vlc_index[i] = get_bits(gb, 3);
284 /* read the scale factor indexes from the stream */
285 for (i = 0; i <= num_subbands; i++) {
286 if (subband_vlc_index[i] != 0)
287 sf_index[i] = get_bits(gb, 6);
290 for (i = 0; i <= num_subbands; i++) {
291 first = subband_tab[i ];
292 last = subband_tab[i + 1];
294 subband_size = last - first;
296 if (subband_vlc_index[i] != 0) {
297 /* decode spectral coefficients for this subband */
298 /* TODO: This can be done faster is several blocks share the
299 * same VLC selector (subband_vlc_index) */
300 read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode,
301 mantissas, subband_size);
303 /* decode the scale factor for this subband */
304 scale_factor = ff_atrac_sf_table[sf_index[i]] *
305 inv_max_quant[subband_vlc_index[i]];
307 /* inverse quantize the coefficients */
308 for (j = 0; first < last; first++, j++)
309 output[first] = mantissas[j] * scale_factor;
311 /* this subband was not coded, so zero the entire subband */
312 memset(output + first, 0, subband_size * sizeof(*output));
316 /* clear the subbands that were not coded */
317 first = subband_tab[i];
318 memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output));
323 * Restore the quantized tonal components
325 * @param components tonal components
326 * @param num_bands number of coded bands
328 static int decode_tonal_components(GetBitContext *gb,
329 TonalComponent *components, int num_bands)
332 int nb_components, coding_mode_selector, coding_mode;
333 int band_flags[4], mantissa[8];
334 int component_count = 0;
336 nb_components = get_bits(gb, 5);
338 /* no tonal components */
339 if (nb_components == 0)
342 coding_mode_selector = get_bits(gb, 2);
343 if (coding_mode_selector == 2)
344 return AVERROR_INVALIDDATA;
346 coding_mode = coding_mode_selector & 1;
348 for (i = 0; i < nb_components; i++) {
349 int coded_values_per_component, quant_step_index;
351 for (b = 0; b <= num_bands; b++)
352 band_flags[b] = get_bits1(gb);
354 coded_values_per_component = get_bits(gb, 3);
356 quant_step_index = get_bits(gb, 3);
357 if (quant_step_index <= 1)
358 return AVERROR_INVALIDDATA;
360 if (coding_mode_selector == 3)
361 coding_mode = get_bits1(gb);
363 for (b = 0; b < (num_bands + 1) * 4; b++) {
364 int coded_components;
366 if (band_flags[b >> 2] == 0)
369 coded_components = get_bits(gb, 3);
371 for (c = 0; c < coded_components; c++) {
372 TonalComponent *cmp = &components[component_count];
373 int sf_index, coded_values, max_coded_values;
376 sf_index = get_bits(gb, 6);
377 if (component_count >= 64)
378 return AVERROR_INVALIDDATA;
380 cmp->pos = b * 64 + get_bits(gb, 6);
382 max_coded_values = SAMPLES_PER_FRAME - cmp->pos;
383 coded_values = coded_values_per_component + 1;
384 coded_values = FFMIN(max_coded_values, coded_values);
386 scale_factor = ff_atrac_sf_table[sf_index] *
387 inv_max_quant[quant_step_index];
389 read_quant_spectral_coeffs(gb, quant_step_index, coding_mode,
390 mantissa, coded_values);
392 cmp->num_coefs = coded_values;
395 for (m = 0; m < coded_values; m++)
396 cmp->coef[m] = mantissa[m] * scale_factor;
403 return component_count;
407 * Decode gain parameters for the coded bands
409 * @param block the gainblock for the current band
410 * @param num_bands amount of coded bands
412 static int decode_gain_control(GetBitContext *gb, GainBlock *block,
418 AtracGainInfo *gain = block->g_block;
420 for (b = 0; b <= num_bands; b++) {
421 gain[b].num_points = get_bits(gb, 3);
422 level = gain[b].lev_code;
423 loc = gain[b].loc_code;
425 for (j = 0; j < gain[b].num_points; j++) {
426 level[j] = get_bits(gb, 4);
427 loc[j] = get_bits(gb, 5);
428 if (j && loc[j] <= loc[j - 1])
429 return AVERROR_INVALIDDATA;
433 /* Clear the unused blocks. */
435 gain[b].num_points = 0;
441 * Combine the tonal band spectrum and regular band spectrum
443 * @param spectrum output spectrum buffer
444 * @param num_components number of tonal components
445 * @param components tonal components for this band
446 * @return position of the last tonal coefficient
448 static int add_tonal_components(float *spectrum, int num_components,
449 TonalComponent *components)
451 int i, j, last_pos = -1;
452 float *input, *output;
454 for (i = 0; i < num_components; i++) {
455 last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
456 input = components[i].coef;
457 output = &spectrum[components[i].pos];
459 for (j = 0; j < components[i].num_coefs; j++)
460 output[j] += input[j];
466 #define INTERPOLATE(old, new, nsample) \
467 ((old) + (nsample) * 0.125 * ((new) - (old)))
469 static void reverse_matrixing(float *su1, float *su2, int *prev_code,
472 int i, nsample, band;
473 float mc1_l, mc1_r, mc2_l, mc2_r;
475 for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
476 int s1 = prev_code[i];
477 int s2 = curr_code[i];
481 /* Selector value changed, interpolation needed. */
482 mc1_l = matrix_coeffs[s1 * 2 ];
483 mc1_r = matrix_coeffs[s1 * 2 + 1];
484 mc2_l = matrix_coeffs[s2 * 2 ];
485 mc2_r = matrix_coeffs[s2 * 2 + 1];
487 /* Interpolation is done over the first eight samples. */
488 for (; nsample < band + 8; nsample++) {
489 float c1 = su1[nsample];
490 float c2 = su2[nsample];
491 c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
492 c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
494 su2[nsample] = c1 * 2.0 - c2;
498 /* Apply the matrix without interpolation. */
500 case 0: /* M/S decoding */
501 for (; nsample < band + 256; nsample++) {
502 float c1 = su1[nsample];
503 float c2 = su2[nsample];
504 su1[nsample] = c2 * 2.0;
505 su2[nsample] = (c1 - c2) * 2.0;
509 for (; nsample < band + 256; nsample++) {
510 float c1 = su1[nsample];
511 float c2 = su2[nsample];
512 su1[nsample] = (c1 + c2) * 2.0;
513 su2[nsample] = c2 * -2.0;
518 for (; nsample < band + 256; nsample++) {
519 float c1 = su1[nsample];
520 float c2 = su2[nsample];
521 su1[nsample] = c1 + c2;
522 su2[nsample] = c1 - c2;
531 static void get_channel_weights(int index, int flag, float ch[2])
537 ch[0] = (index & 7) / 7.0;
538 ch[1] = sqrt(2 - ch[0] * ch[0]);
540 FFSWAP(float, ch[0], ch[1]);
544 static void channel_weighting(float *su1, float *su2, int *p3)
547 /* w[x][y] y=0 is left y=1 is right */
550 if (p3[1] != 7 || p3[3] != 7) {
551 get_channel_weights(p3[1], p3[0], w[0]);
552 get_channel_weights(p3[3], p3[2], w[1]);
554 for (band = 256; band < 4 * 256; band += 256) {
555 for (nsample = band; nsample < band + 8; nsample++) {
556 su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
557 su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
559 for(; nsample < band + 256; nsample++) {
560 su1[nsample] *= w[1][0];
561 su2[nsample] *= w[1][1];
568 * Decode a Sound Unit
570 * @param snd the channel unit to be used
571 * @param output the decoded samples before IQMF in float representation
572 * @param channel_num channel number
573 * @param coding_mode the coding mode (JOINT_STEREO or single channels)
575 static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb,
576 ChannelUnit *snd, float *output,
577 int channel_num, int coding_mode)
579 int band, ret, num_subbands, last_tonal, num_bands;
580 GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch];
581 GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
583 if (coding_mode == JOINT_STEREO && (channel_num % 2) == 1) {
584 if (get_bits(gb, 2) != 3) {
585 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
586 return AVERROR_INVALIDDATA;
589 if (get_bits(gb, 6) != 0x28) {
590 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
591 return AVERROR_INVALIDDATA;
595 /* number of coded QMF bands */
596 snd->bands_coded = get_bits(gb, 2);
598 ret = decode_gain_control(gb, gain2, snd->bands_coded);
602 snd->num_components = decode_tonal_components(gb, snd->components,
604 if (snd->num_components < 0)
605 return snd->num_components;
607 num_subbands = decode_spectrum(gb, snd->spectrum);
609 /* Merge the decoded spectrum and tonal components. */
610 last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
614 /* calculate number of used MLT/QMF bands according to the amount of coded
616 num_bands = (subband_tab[num_subbands] - 1) >> 8;
618 num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
621 /* Reconstruct time domain samples. */
622 for (band = 0; band < 4; band++) {
623 /* Perform the IMDCT step without overlapping. */
624 if (band <= num_bands)
625 imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
627 memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf));
629 /* gain compensation and overlapping */
630 ff_atrac_gain_compensation(&q->gainc_ctx, snd->imdct_buf,
631 &snd->prev_frame[band * 256],
632 &gain1->g_block[band], &gain2->g_block[band],
633 256, &output[band * 256]);
636 /* Swap the gain control buffers for the next frame. */
637 snd->gc_blk_switch ^= 1;
642 static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
645 ATRAC3Context *q = avctx->priv_data;
649 if (q->coding_mode == JOINT_STEREO) {
650 /* channel coupling mode */
652 /* Decode sound unit pairs (channels are expected to be even).
653 * Multichannel joint stereo interleaves pairs (6ch: 2ch + 2ch + 2ch) */
654 const uint8_t *js_databuf;
655 int js_pair, js_block_align;
657 js_block_align = (avctx->block_align / avctx->channels) * 2; /* block pair */
659 for (ch = 0; ch < avctx->channels; ch = ch + 2) {
661 js_databuf = databuf + js_pair * js_block_align; /* align to current pair */
663 /* Set the bitstream reader at the start of first channel sound unit. */
664 init_get_bits(&q->gb,
665 js_databuf, js_block_align * 8);
667 /* decode Sound Unit 1 */
668 ret = decode_channel_sound_unit(q, &q->gb, &q->units[ch],
669 out_samples[ch], ch, JOINT_STEREO);
673 /* Framedata of the su2 in the joint-stereo mode is encoded in
674 * reverse byte order so we need to swap it first. */
675 if (js_databuf == q->decoded_bytes_buffer) {
676 uint8_t *ptr2 = q->decoded_bytes_buffer + js_block_align - 1;
677 ptr1 = q->decoded_bytes_buffer;
678 for (i = 0; i < js_block_align / 2; i++, ptr1++, ptr2--)
679 FFSWAP(uint8_t, *ptr1, *ptr2);
681 const uint8_t *ptr2 = js_databuf + js_block_align - 1;
682 for (i = 0; i < js_block_align; i++)
683 q->decoded_bytes_buffer[i] = *ptr2--;
686 /* Skip the sync codes (0xF8). */
687 ptr1 = q->decoded_bytes_buffer;
688 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
689 if (i >= js_block_align)
690 return AVERROR_INVALIDDATA;
694 /* set the bitstream reader at the start of the second Sound Unit */
695 ret = init_get_bits8(&q->gb,
696 ptr1, q->decoded_bytes_buffer + js_block_align - ptr1);
700 /* Fill the Weighting coeffs delay buffer */
701 memmove(q->weighting_delay[js_pair], &q->weighting_delay[js_pair][2],
702 4 * sizeof(*q->weighting_delay[js_pair]));
703 q->weighting_delay[js_pair][4] = get_bits1(&q->gb);
704 q->weighting_delay[js_pair][5] = get_bits(&q->gb, 3);
706 for (i = 0; i < 4; i++) {
707 q->matrix_coeff_index_prev[js_pair][i] = q->matrix_coeff_index_now[js_pair][i];
708 q->matrix_coeff_index_now[js_pair][i] = q->matrix_coeff_index_next[js_pair][i];
709 q->matrix_coeff_index_next[js_pair][i] = get_bits(&q->gb, 2);
712 /* Decode Sound Unit 2. */
713 ret = decode_channel_sound_unit(q, &q->gb, &q->units[ch+1],
714 out_samples[ch+1], ch+1, JOINT_STEREO);
718 /* Reconstruct the channel coefficients. */
719 reverse_matrixing(out_samples[ch], out_samples[ch+1],
720 q->matrix_coeff_index_prev[js_pair],
721 q->matrix_coeff_index_now[js_pair]);
723 channel_weighting(out_samples[ch], out_samples[ch+1], q->weighting_delay[js_pair]);
726 /* single channels */
727 /* Decode the channel sound units. */
728 for (i = 0; i < avctx->channels; i++) {
729 /* Set the bitstream reader at the start of a channel sound unit. */
730 init_get_bits(&q->gb,
731 databuf + i * avctx->block_align / avctx->channels,
732 avctx->block_align * 8 / avctx->channels);
734 ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
735 out_samples[i], i, q->coding_mode);
741 /* Apply the iQMF synthesis filter. */
742 for (i = 0; i < avctx->channels; i++) {
743 float *p1 = out_samples[i];
744 float *p2 = p1 + 256;
745 float *p3 = p2 + 256;
746 float *p4 = p3 + 256;
747 ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
748 ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
749 ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
755 static int al_decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
756 int size, float **out_samples)
758 ATRAC3Context *q = avctx->priv_data;
761 /* Set the bitstream reader at the start of a channel sound unit. */
762 init_get_bits(&q->gb, databuf, size * 8);
763 /* single channels */
764 /* Decode the channel sound units. */
765 for (i = 0; i < avctx->channels; i++) {
766 ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
767 out_samples[i], i, q->coding_mode);
770 while (i < avctx->channels && get_bits_left(&q->gb) > 6 && show_bits(&q->gb, 6) != 0x28) {
771 skip_bits(&q->gb, 1);
775 /* Apply the iQMF synthesis filter. */
776 for (i = 0; i < avctx->channels; i++) {
777 float *p1 = out_samples[i];
778 float *p2 = p1 + 256;
779 float *p3 = p2 + 256;
780 float *p4 = p3 + 256;
781 ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
782 ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
783 ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
789 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
790 int *got_frame_ptr, AVPacket *avpkt)
792 AVFrame *frame = data;
793 const uint8_t *buf = avpkt->data;
794 int buf_size = avpkt->size;
795 ATRAC3Context *q = avctx->priv_data;
797 const uint8_t *databuf;
799 if (buf_size < avctx->block_align) {
800 av_log(avctx, AV_LOG_ERROR,
801 "Frame too small (%d bytes). Truncated file?\n", buf_size);
802 return AVERROR_INVALIDDATA;
805 /* get output buffer */
806 frame->nb_samples = SAMPLES_PER_FRAME;
807 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
810 /* Check if we need to descramble and what buffer to pass on. */
811 if (q->scrambled_stream) {
812 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
813 databuf = q->decoded_bytes_buffer;
818 ret = decode_frame(avctx, databuf, (float **)frame->extended_data);
820 av_log(avctx, AV_LOG_ERROR, "Frame decoding error!\n");
826 return avctx->block_align;
829 static int atrac3al_decode_frame(AVCodecContext *avctx, void *data,
830 int *got_frame_ptr, AVPacket *avpkt)
832 AVFrame *frame = data;
835 frame->nb_samples = SAMPLES_PER_FRAME;
836 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
839 ret = al_decode_frame(avctx, avpkt->data, avpkt->size,
840 (float **)frame->extended_data);
842 av_log(avctx, AV_LOG_ERROR, "Frame decoding error!\n");
851 static av_cold void atrac3_init_static_data(void)
853 VLC_TYPE (*table)[2] = atrac3_vlc_table;
854 const uint8_t (*hufftabs)[2] = atrac3_hufftabs;
858 ff_atrac_generate_tables();
860 /* Initialize the VLC tables. */
861 for (i = 0; i < 7; i++) {
862 spectral_coeff_tab[i].table = table;
863 spectral_coeff_tab[i].table_allocated = 256;
864 ff_init_vlc_from_lengths(&spectral_coeff_tab[i], ATRAC3_VLC_BITS, huff_tab_sizes[i],
866 &hufftabs[0][0], 2, 1,
867 -31, INIT_VLC_USE_NEW_STATIC, NULL);
868 hufftabs += huff_tab_sizes[i];
873 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
875 static AVOnce init_static_once = AV_ONCE_INIT;
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 /* Take care of the codec-specific extradata. */
888 if (avctx->codec_id == AV_CODEC_ID_ATRAC3AL) {
890 samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
892 q->coding_mode = SINGLE;
893 } else if (avctx->extradata_size == 14) {
894 /* Parse the extradata, WAV format */
895 av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
896 bytestream_get_le16(&edata_ptr)); // Unknown value always 1
897 edata_ptr += 4; // samples per channel
898 q->coding_mode = bytestream_get_le16(&edata_ptr);
899 av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
900 bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
901 frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1
902 av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
903 bytestream_get_le16(&edata_ptr)); // Unknown always 0
906 samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
909 q->coding_mode = q->coding_mode ? JOINT_STEREO : SINGLE;
910 q->scrambled_stream = 0;
912 if (avctx->block_align != 96 * avctx->channels * frame_factor &&
913 avctx->block_align != 152 * avctx->channels * frame_factor &&
914 avctx->block_align != 192 * avctx->channels * frame_factor) {
915 av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
916 "configuration %d/%d/%d\n", avctx->block_align,
917 avctx->channels, frame_factor);
918 return AVERROR_INVALIDDATA;
920 } else if (avctx->extradata_size == 12 || avctx->extradata_size == 10) {
921 /* Parse the extradata, RM format. */
922 version = bytestream_get_be32(&edata_ptr);
923 samples_per_frame = bytestream_get_be16(&edata_ptr);
924 delay = bytestream_get_be16(&edata_ptr);
925 q->coding_mode = bytestream_get_be16(&edata_ptr);
926 q->scrambled_stream = 1;
929 av_log(avctx, AV_LOG_ERROR, "Unknown extradata size %d.\n",
930 avctx->extradata_size);
931 return AVERROR(EINVAL);
934 /* Check the extradata */
937 av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
938 return AVERROR_INVALIDDATA;
941 if (samples_per_frame != SAMPLES_PER_FRAME * avctx->channels) {
942 av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
944 return AVERROR_INVALIDDATA;
947 if (delay != 0x88E) {
948 av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
950 return AVERROR_INVALIDDATA;
953 if (q->coding_mode == SINGLE)
954 av_log(avctx, AV_LOG_DEBUG, "Single channels detected.\n");
955 else if (q->coding_mode == JOINT_STEREO) {
956 if (avctx->channels % 2 == 1) { /* Joint stereo channels must be even */
957 av_log(avctx, AV_LOG_ERROR, "Invalid joint stereo channel configuration.\n");
958 return AVERROR_INVALIDDATA;
960 av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
962 av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
964 return AVERROR_INVALIDDATA;
967 if (avctx->block_align > 4096 || avctx->block_align <= 0)
968 return AVERROR(EINVAL);
970 q->decoded_bytes_buffer = av_mallocz(FFALIGN(avctx->block_align, 4) +
971 AV_INPUT_BUFFER_PADDING_SIZE);
972 if (!q->decoded_bytes_buffer)
973 return AVERROR(ENOMEM);
975 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
977 /* initialize the MDCT transform */
978 if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) {
979 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
983 /* init the joint-stereo decoding data */
984 for (js_pair = 0; js_pair < MAX_JS_PAIRS; js_pair++) {
985 q->weighting_delay[js_pair][0] = 0;
986 q->weighting_delay[js_pair][1] = 7;
987 q->weighting_delay[js_pair][2] = 0;
988 q->weighting_delay[js_pair][3] = 7;
989 q->weighting_delay[js_pair][4] = 0;
990 q->weighting_delay[js_pair][5] = 7;
992 for (i = 0; i < 4; i++) {
993 q->matrix_coeff_index_prev[js_pair][i] = 3;
994 q->matrix_coeff_index_now[js_pair][i] = 3;
995 q->matrix_coeff_index_next[js_pair][i] = 3;
999 ff_atrac_init_gain_compensation(&q->gainc_ctx, 4, 3);
1000 fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
1002 return AVERROR(ENOMEM);
1003 q->vector_fmul = fdsp->vector_fmul;
1006 q->units = av_mallocz_array(avctx->channels, sizeof(*q->units));
1008 return AVERROR(ENOMEM);
1010 ff_thread_once(&init_static_once, atrac3_init_static_data);
1015 AVCodec ff_atrac3_decoder = {
1017 .long_name = NULL_IF_CONFIG_SMALL("ATRAC3 (Adaptive TRansform Acoustic Coding 3)"),
1018 .type = AVMEDIA_TYPE_AUDIO,
1019 .id = AV_CODEC_ID_ATRAC3,
1020 .priv_data_size = sizeof(ATRAC3Context),
1021 .init = atrac3_decode_init,
1022 .close = atrac3_decode_close,
1023 .decode = atrac3_decode_frame,
1024 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1025 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1026 AV_SAMPLE_FMT_NONE },
1027 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
1030 AVCodec ff_atrac3al_decoder = {
1032 .long_name = NULL_IF_CONFIG_SMALL("ATRAC3 AL (Adaptive TRansform Acoustic Coding 3 Advanced Lossless)"),
1033 .type = AVMEDIA_TYPE_AUDIO,
1034 .id = AV_CODEC_ID_ATRAC3AL,
1035 .priv_data_size = sizeof(ATRAC3Context),
1036 .init = atrac3_decode_init,
1037 .close = atrac3_decode_close,
1038 .decode = atrac3al_decode_frame,
1039 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1,
1040 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1041 AV_SAMPLE_FMT_NONE },
1042 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,