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/attributes.h"
40 #include "libavutil/float_dsp.h"
41 #include "libavutil/libm.h"
43 #include "bytestream.h"
45 #include "fmtconvert.h"
50 #include "atrac3data.h"
52 #define JOINT_STEREO 0x12
55 #define SAMPLES_PER_FRAME 1024
58 typedef struct GainInfo {
64 typedef struct GainBlock {
68 typedef struct TonalComponent {
74 typedef struct ChannelUnit {
77 float prev_frame[SAMPLES_PER_FRAME];
79 TonalComponent components[64];
80 GainBlock gain_block[2];
82 DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
83 DECLARE_ALIGNED(32, float, imdct_buf)[SAMPLES_PER_FRAME];
85 float delay_buf1[46]; ///<qmf delay buffers
90 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);
169 c = av_be2ne32((0x537F6103U >> (off * 8)) | (0x537F6103U << (32 - (off * 8))));
171 c = av_be2ne32(0x537F6103U);
173 for (i = 0; i < bytes / 4; i++)
174 output[i] = c ^ buf[i];
177 avpriv_request_sample(NULL, "Offset of %d", off);
182 static av_cold void init_atrac3_window(void)
186 /* generate the mdct window, for details see
187 * http://wiki.multimedia.cx/index.php?title=RealAudio_atrc#Windows */
188 for (i = 0, j = 255; i < 128; i++, j--) {
189 float wi = sin(((i + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
190 float wj = sin(((j + 0.5) / 256.0 - 0.5) * M_PI) + 1.0;
191 float w = 0.5 * (wi * wi + wj * wj);
192 mdct_window[i] = mdct_window[511 - i] = wi / w;
193 mdct_window[j] = mdct_window[511 - j] = wj / w;
197 static av_cold int atrac3_decode_close(AVCodecContext *avctx)
199 ATRAC3Context *q = avctx->priv_data;
202 av_free(q->decoded_bytes_buffer);
204 ff_mdct_end(&q->mdct_ctx);
212 * @param selector which table the output values are coded with
213 * @param coding_flag constant length coding or variable length coding
214 * @param mantissas mantissa output table
215 * @param num_codes number of values to get
217 static void read_quant_spectral_coeffs(GetBitContext *gb, int selector,
218 int coding_flag, int *mantissas,
221 int i, code, huff_symb;
226 if (coding_flag != 0) {
227 /* constant length coding (CLC) */
228 int num_bits = clc_length_tab[selector];
231 for (i = 0; i < num_codes; i++) {
233 code = get_sbits(gb, num_bits);
239 for (i = 0; i < num_codes; i++) {
241 code = get_bits(gb, num_bits); // num_bits is always 4 in this case
244 mantissas[i * 2 ] = mantissa_clc_tab[code >> 2];
245 mantissas[i * 2 + 1] = mantissa_clc_tab[code & 3];
249 /* variable length coding (VLC) */
251 for (i = 0; i < num_codes; i++) {
252 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector-1].table,
253 spectral_coeff_tab[selector-1].bits, 3);
255 code = huff_symb >> 1;
261 for (i = 0; i < num_codes; i++) {
262 huff_symb = get_vlc2(gb, spectral_coeff_tab[selector - 1].table,
263 spectral_coeff_tab[selector - 1].bits, 3);
264 mantissas[i * 2 ] = mantissa_vlc_tab[huff_symb * 2 ];
265 mantissas[i * 2 + 1] = mantissa_vlc_tab[huff_symb * 2 + 1];
272 * Restore the quantized band spectrum coefficients
274 * @return subband count, fix for broken specification/files
276 static int decode_spectrum(GetBitContext *gb, float *output)
278 int num_subbands, coding_mode, i, j, first, last, subband_size;
279 int subband_vlc_index[32], sf_index[32];
283 num_subbands = get_bits(gb, 5); // number of coded subbands
284 coding_mode = get_bits1(gb); // coding Mode: 0 - VLC/ 1-CLC
286 /* get the VLC selector table for the subbands, 0 means not coded */
287 for (i = 0; i <= num_subbands; i++)
288 subband_vlc_index[i] = get_bits(gb, 3);
290 /* read the scale factor indexes from the stream */
291 for (i = 0; i <= num_subbands; i++) {
292 if (subband_vlc_index[i] != 0)
293 sf_index[i] = get_bits(gb, 6);
296 for (i = 0; i <= num_subbands; i++) {
297 first = subband_tab[i ];
298 last = subband_tab[i + 1];
300 subband_size = last - first;
302 if (subband_vlc_index[i] != 0) {
303 /* decode spectral coefficients for this subband */
304 /* TODO: This can be done faster is several blocks share the
305 * same VLC selector (subband_vlc_index) */
306 read_quant_spectral_coeffs(gb, subband_vlc_index[i], coding_mode,
307 mantissas, subband_size);
309 /* decode the scale factor for this subband */
310 scale_factor = ff_atrac_sf_table[sf_index[i]] *
311 inv_max_quant[subband_vlc_index[i]];
313 /* inverse quantize the coefficients */
314 for (j = 0; first < last; first++, j++)
315 output[first] = mantissas[j] * scale_factor;
317 /* this subband was not coded, so zero the entire subband */
318 memset(output + first, 0, subband_size * sizeof(*output));
322 /* clear the subbands that were not coded */
323 first = subband_tab[i];
324 memset(output + first, 0, (SAMPLES_PER_FRAME - first) * sizeof(*output));
329 * Restore the quantized tonal components
331 * @param components tonal components
332 * @param num_bands number of coded bands
334 static int decode_tonal_components(GetBitContext *gb,
335 TonalComponent *components, int num_bands)
338 int nb_components, coding_mode_selector, coding_mode;
339 int band_flags[4], mantissa[8];
340 int component_count = 0;
342 nb_components = get_bits(gb, 5);
344 /* no tonal components */
345 if (nb_components == 0)
348 coding_mode_selector = get_bits(gb, 2);
349 if (coding_mode_selector == 2)
350 return AVERROR_INVALIDDATA;
352 coding_mode = coding_mode_selector & 1;
354 for (i = 0; i < nb_components; i++) {
355 int coded_values_per_component, quant_step_index;
357 for (b = 0; b <= num_bands; b++)
358 band_flags[b] = get_bits1(gb);
360 coded_values_per_component = get_bits(gb, 3);
362 quant_step_index = get_bits(gb, 3);
363 if (quant_step_index <= 1)
364 return AVERROR_INVALIDDATA;
366 if (coding_mode_selector == 3)
367 coding_mode = get_bits1(gb);
369 for (b = 0; b < (num_bands + 1) * 4; b++) {
370 int coded_components;
372 if (band_flags[b >> 2] == 0)
375 coded_components = get_bits(gb, 3);
377 for (c = 0; c < coded_components; c++) {
378 TonalComponent *cmp = &components[component_count];
379 int sf_index, coded_values, max_coded_values;
382 sf_index = get_bits(gb, 6);
383 if (component_count >= 64)
384 return AVERROR_INVALIDDATA;
386 cmp->pos = b * 64 + get_bits(gb, 6);
388 max_coded_values = SAMPLES_PER_FRAME - cmp->pos;
389 coded_values = coded_values_per_component + 1;
390 coded_values = FFMIN(max_coded_values, coded_values);
392 scale_factor = ff_atrac_sf_table[sf_index] *
393 inv_max_quant[quant_step_index];
395 read_quant_spectral_coeffs(gb, quant_step_index, coding_mode,
396 mantissa, coded_values);
398 cmp->num_coefs = coded_values;
401 for (m = 0; m < coded_values; m++)
402 cmp->coef[m] = mantissa[m] * scale_factor;
409 return component_count;
413 * Decode gain parameters for the coded bands
415 * @param block the gainblock for the current band
416 * @param num_bands amount of coded bands
418 static int decode_gain_control(GetBitContext *gb, GainBlock *block,
424 GainInfo *gain = block->g_block;
426 for (i = 0; i <= num_bands; i++) {
427 num_data = get_bits(gb, 3);
428 gain[i].num_gain_data = num_data;
429 level = gain[i].lev_code;
430 loc = gain[i].loc_code;
432 for (cf = 0; cf < gain[i].num_gain_data; cf++) {
433 level[cf] = get_bits(gb, 4);
434 loc [cf] = get_bits(gb, 5);
435 if (cf && loc[cf] <= loc[cf - 1])
436 return AVERROR_INVALIDDATA;
440 /* Clear the unused blocks. */
442 gain[i].num_gain_data = 0;
448 * Apply gain parameters and perform the MDCT overlapping part
450 * @param input input buffer
451 * @param prev previous buffer to perform overlap against
452 * @param output output buffer
453 * @param gain1 current band gain info
454 * @param gain2 next band gain info
456 static void gain_compensate_and_overlap(float *input, float *prev,
457 float *output, GainInfo *gain1,
460 float g1, g2, gain_inc;
461 int i, j, num_data, start_loc, end_loc;
464 if (gain2->num_gain_data == 0)
467 g1 = gain_tab1[gain2->lev_code[0]];
469 if (gain1->num_gain_data == 0) {
470 for (i = 0; i < 256; i++)
471 output[i] = input[i] * g1 + prev[i];
473 num_data = gain1->num_gain_data;
474 gain1->loc_code[num_data] = 32;
475 gain1->lev_code[num_data] = 4;
477 for (i = 0, j = 0; i < num_data; i++) {
478 start_loc = gain1->loc_code[i] * 8;
479 end_loc = start_loc + 8;
481 g2 = gain_tab1[gain1->lev_code[i]];
482 gain_inc = gain_tab2[gain1->lev_code[i + 1] -
483 gain1->lev_code[i ] + 15];
486 for (; j < start_loc; j++)
487 output[j] = (input[j] * g1 + prev[j]) * g2;
489 /* interpolation is done over eight samples */
490 for (; j < end_loc; j++) {
491 output[j] = (input[j] * g1 + prev[j]) * g2;
497 output[j] = input[j] * g1 + prev[j];
500 /* Delay for the overlapping part. */
501 memcpy(prev, &input[256], 256 * sizeof(*prev));
505 * Combine the tonal band spectrum and regular band spectrum
507 * @param spectrum output spectrum buffer
508 * @param num_components number of tonal components
509 * @param components tonal components for this band
510 * @return position of the last tonal coefficient
512 static int add_tonal_components(float *spectrum, int num_components,
513 TonalComponent *components)
515 int i, j, last_pos = -1;
516 float *input, *output;
518 for (i = 0; i < num_components; i++) {
519 last_pos = FFMAX(components[i].pos + components[i].num_coefs, last_pos);
520 input = components[i].coef;
521 output = &spectrum[components[i].pos];
523 for (j = 0; j < components[i].num_coefs; j++)
524 output[j] += input[j];
530 #define INTERPOLATE(old, new, nsample) \
531 ((old) + (nsample) * 0.125 * ((new) - (old)))
533 static void reverse_matrixing(float *su1, float *su2, int *prev_code,
536 int i, nsample, band;
537 float mc1_l, mc1_r, mc2_l, mc2_r;
539 for (i = 0, band = 0; band < 4 * 256; band += 256, i++) {
540 int s1 = prev_code[i];
541 int s2 = curr_code[i];
545 /* Selector value changed, interpolation needed. */
546 mc1_l = matrix_coeffs[s1 * 2 ];
547 mc1_r = matrix_coeffs[s1 * 2 + 1];
548 mc2_l = matrix_coeffs[s2 * 2 ];
549 mc2_r = matrix_coeffs[s2 * 2 + 1];
551 /* Interpolation is done over the first eight samples. */
552 for (; nsample < band + 8; nsample++) {
553 float c1 = su1[nsample];
554 float c2 = su2[nsample];
555 c2 = c1 * INTERPOLATE(mc1_l, mc2_l, nsample - band) +
556 c2 * INTERPOLATE(mc1_r, mc2_r, nsample - band);
558 su2[nsample] = c1 * 2.0 - c2;
562 /* Apply the matrix without interpolation. */
564 case 0: /* M/S decoding */
565 for (; nsample < band + 256; nsample++) {
566 float c1 = su1[nsample];
567 float c2 = su2[nsample];
568 su1[nsample] = c2 * 2.0;
569 su2[nsample] = (c1 - c2) * 2.0;
573 for (; nsample < band + 256; nsample++) {
574 float c1 = su1[nsample];
575 float c2 = su2[nsample];
576 su1[nsample] = (c1 + c2) * 2.0;
577 su2[nsample] = c2 * -2.0;
582 for (; nsample < band + 256; nsample++) {
583 float c1 = su1[nsample];
584 float c2 = su2[nsample];
585 su1[nsample] = c1 + c2;
586 su2[nsample] = c1 - c2;
595 static void get_channel_weights(int index, int flag, float ch[2])
601 ch[0] = (index & 7) / 7.0;
602 ch[1] = sqrt(2 - ch[0] * ch[0]);
604 FFSWAP(float, ch[0], ch[1]);
608 static void channel_weighting(float *su1, float *su2, int *p3)
611 /* w[x][y] y=0 is left y=1 is right */
614 if (p3[1] != 7 || p3[3] != 7) {
615 get_channel_weights(p3[1], p3[0], w[0]);
616 get_channel_weights(p3[3], p3[2], w[1]);
618 for (band = 256; band < 4 * 256; band += 256) {
619 for (nsample = band; nsample < band + 8; nsample++) {
620 su1[nsample] *= INTERPOLATE(w[0][0], w[0][1], nsample - band);
621 su2[nsample] *= INTERPOLATE(w[1][0], w[1][1], nsample - band);
623 for(; nsample < band + 256; nsample++) {
624 su1[nsample] *= w[1][0];
625 su2[nsample] *= w[1][1];
632 * Decode a Sound Unit
634 * @param snd the channel unit to be used
635 * @param output the decoded samples before IQMF in float representation
636 * @param channel_num channel number
637 * @param coding_mode the coding mode (JOINT_STEREO or regular stereo/mono)
639 static int decode_channel_sound_unit(ATRAC3Context *q, GetBitContext *gb,
640 ChannelUnit *snd, float *output,
641 int channel_num, int coding_mode)
643 int band, ret, num_subbands, last_tonal, num_bands;
644 GainBlock *gain1 = &snd->gain_block[ snd->gc_blk_switch];
645 GainBlock *gain2 = &snd->gain_block[1 - snd->gc_blk_switch];
647 if (coding_mode == JOINT_STEREO && channel_num == 1) {
648 if (get_bits(gb, 2) != 3) {
649 av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
650 return AVERROR_INVALIDDATA;
653 if (get_bits(gb, 6) != 0x28) {
654 av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
655 return AVERROR_INVALIDDATA;
659 /* number of coded QMF bands */
660 snd->bands_coded = get_bits(gb, 2);
662 ret = decode_gain_control(gb, gain2, snd->bands_coded);
666 snd->num_components = decode_tonal_components(gb, snd->components,
668 if (snd->num_components == -1)
671 num_subbands = decode_spectrum(gb, snd->spectrum);
673 /* Merge the decoded spectrum and tonal components. */
674 last_tonal = add_tonal_components(snd->spectrum, snd->num_components,
678 /* calculate number of used MLT/QMF bands according to the amount of coded
680 num_bands = (subband_tab[num_subbands] - 1) >> 8;
682 num_bands = FFMAX((last_tonal + 256) >> 8, num_bands);
685 /* Reconstruct time domain samples. */
686 for (band = 0; band < 4; band++) {
687 /* Perform the IMDCT step without overlapping. */
688 if (band <= num_bands)
689 imlt(q, &snd->spectrum[band * 256], snd->imdct_buf, band & 1);
691 memset(snd->imdct_buf, 0, 512 * sizeof(*snd->imdct_buf));
693 /* gain compensation and overlapping */
694 gain_compensate_and_overlap(snd->imdct_buf,
695 &snd->prev_frame[band * 256],
697 &gain1->g_block[band],
698 &gain2->g_block[band]);
701 /* Swap the gain control buffers for the next frame. */
702 snd->gc_blk_switch ^= 1;
707 static int decode_frame(AVCodecContext *avctx, const uint8_t *databuf,
710 ATRAC3Context *q = avctx->priv_data;
714 if (q->coding_mode == JOINT_STEREO) {
715 /* channel coupling mode */
716 /* decode Sound Unit 1 */
717 init_get_bits(&q->gb, databuf, avctx->block_align * 8);
719 ret = decode_channel_sound_unit(q, &q->gb, q->units, out_samples[0], 0,
724 /* Framedata of the su2 in the joint-stereo mode is encoded in
725 * reverse byte order so we need to swap it first. */
726 if (databuf == q->decoded_bytes_buffer) {
727 uint8_t *ptr2 = q->decoded_bytes_buffer + avctx->block_align - 1;
728 ptr1 = q->decoded_bytes_buffer;
729 for (i = 0; i < avctx->block_align / 2; i++, ptr1++, ptr2--)
730 FFSWAP(uint8_t, *ptr1, *ptr2);
732 const uint8_t *ptr2 = databuf + avctx->block_align - 1;
733 for (i = 0; i < avctx->block_align; i++)
734 q->decoded_bytes_buffer[i] = *ptr2--;
737 /* Skip the sync codes (0xF8). */
738 ptr1 = q->decoded_bytes_buffer;
739 for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
740 if (i >= avctx->block_align)
741 return AVERROR_INVALIDDATA;
745 /* set the bitstream reader at the start of the second Sound Unit*/
746 init_get_bits8(&q->gb, ptr1, q->decoded_bytes_buffer + avctx->block_align - ptr1);
748 /* Fill the Weighting coeffs delay buffer */
749 memmove(q->weighting_delay, &q->weighting_delay[2],
750 4 * sizeof(*q->weighting_delay));
751 q->weighting_delay[4] = get_bits1(&q->gb);
752 q->weighting_delay[5] = get_bits(&q->gb, 3);
754 for (i = 0; i < 4; i++) {
755 q->matrix_coeff_index_prev[i] = q->matrix_coeff_index_now[i];
756 q->matrix_coeff_index_now[i] = q->matrix_coeff_index_next[i];
757 q->matrix_coeff_index_next[i] = get_bits(&q->gb, 2);
760 /* Decode Sound Unit 2. */
761 ret = decode_channel_sound_unit(q, &q->gb, &q->units[1],
762 out_samples[1], 1, JOINT_STEREO);
766 /* Reconstruct the channel coefficients. */
767 reverse_matrixing(out_samples[0], out_samples[1],
768 q->matrix_coeff_index_prev,
769 q->matrix_coeff_index_now);
771 channel_weighting(out_samples[0], out_samples[1], q->weighting_delay);
773 /* normal stereo mode or mono */
774 /* Decode the channel sound units. */
775 for (i = 0; i < avctx->channels; i++) {
776 /* Set the bitstream reader at the start of a channel sound unit. */
777 init_get_bits(&q->gb,
778 databuf + i * avctx->block_align / avctx->channels,
779 avctx->block_align * 8 / avctx->channels);
781 ret = decode_channel_sound_unit(q, &q->gb, &q->units[i],
782 out_samples[i], i, q->coding_mode);
788 /* Apply the iQMF synthesis filter. */
789 for (i = 0; i < avctx->channels; i++) {
790 float *p1 = out_samples[i];
791 float *p2 = p1 + 256;
792 float *p3 = p2 + 256;
793 float *p4 = p3 + 256;
794 ff_atrac_iqmf(p1, p2, 256, p1, q->units[i].delay_buf1, q->temp_buf);
795 ff_atrac_iqmf(p4, p3, 256, p3, q->units[i].delay_buf2, q->temp_buf);
796 ff_atrac_iqmf(p1, p3, 512, p1, q->units[i].delay_buf3, q->temp_buf);
802 static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
803 int *got_frame_ptr, AVPacket *avpkt)
805 AVFrame *frame = data;
806 const uint8_t *buf = avpkt->data;
807 int buf_size = avpkt->size;
808 ATRAC3Context *q = avctx->priv_data;
810 const uint8_t *databuf;
812 if (buf_size < avctx->block_align) {
813 av_log(avctx, AV_LOG_ERROR,
814 "Frame too small (%d bytes). Truncated file?\n", buf_size);
815 return AVERROR_INVALIDDATA;
818 /* get output buffer */
819 frame->nb_samples = SAMPLES_PER_FRAME;
820 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
823 /* Check if we need to descramble and what buffer to pass on. */
824 if (q->scrambled_stream) {
825 decode_bytes(buf, q->decoded_bytes_buffer, avctx->block_align);
826 databuf = q->decoded_bytes_buffer;
831 ret = decode_frame(avctx, databuf, (float **)frame->extended_data);
833 av_log(NULL, AV_LOG_ERROR, "Frame decoding error!\n");
839 return avctx->block_align;
842 static av_cold void atrac3_init_static_data(void)
846 init_atrac3_window();
847 ff_atrac_generate_tables();
849 /* Initialize the VLC tables. */
850 for (i = 0; i < 7; i++) {
851 spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
852 spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] -
854 init_vlc(&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
856 huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
859 /* Generate gain tables */
860 for (i = 0; i < 16; i++)
861 gain_tab1[i] = exp2f (4 - i);
863 for (i = -15; i < 16; i++)
864 gain_tab2[i + 15] = exp2f (i * -0.125);
867 static av_cold int atrac3_decode_init(AVCodecContext *avctx)
869 static int static_init_done;
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 if (!static_init_done)
881 atrac3_init_static_data();
882 static_init_done = 1;
884 /* Take care of the codec-specific extradata. */
885 if (avctx->extradata_size == 14) {
886 /* Parse the extradata, WAV format */
887 av_log(avctx, AV_LOG_DEBUG, "[0-1] %d\n",
888 bytestream_get_le16(&edata_ptr)); // Unknown value always 1
889 edata_ptr += 4; // samples per channel
890 q->coding_mode = bytestream_get_le16(&edata_ptr);
891 av_log(avctx, AV_LOG_DEBUG,"[8-9] %d\n",
892 bytestream_get_le16(&edata_ptr)); //Dupe of coding mode
893 frame_factor = bytestream_get_le16(&edata_ptr); // Unknown always 1
894 av_log(avctx, AV_LOG_DEBUG,"[12-13] %d\n",
895 bytestream_get_le16(&edata_ptr)); // Unknown always 0
898 samples_per_frame = SAMPLES_PER_FRAME * avctx->channels;
901 q->coding_mode = q->coding_mode ? JOINT_STEREO : STEREO;
902 q->scrambled_stream = 0;
904 if (avctx->block_align != 96 * avctx->channels * frame_factor &&
905 avctx->block_align != 152 * avctx->channels * frame_factor &&
906 avctx->block_align != 192 * avctx->channels * frame_factor) {
907 av_log(avctx, AV_LOG_ERROR, "Unknown frame/channel/frame_factor "
908 "configuration %d/%d/%d\n", avctx->block_align,
909 avctx->channels, frame_factor);
910 return AVERROR_INVALIDDATA;
912 } else if (avctx->extradata_size == 12 || avctx->extradata_size == 10) {
913 /* Parse the extradata, RM format. */
914 version = bytestream_get_be32(&edata_ptr);
915 samples_per_frame = bytestream_get_be16(&edata_ptr);
916 delay = bytestream_get_be16(&edata_ptr);
917 q->coding_mode = bytestream_get_be16(&edata_ptr);
918 q->scrambled_stream = 1;
921 av_log(NULL, AV_LOG_ERROR, "Unknown extradata size %d.\n",
922 avctx->extradata_size);
923 return AVERROR(EINVAL);
926 if (q->coding_mode == JOINT_STEREO && avctx->channels < 2) {
927 av_log(avctx, AV_LOG_ERROR, "Invalid coding mode\n");
928 return AVERROR_INVALIDDATA;
931 /* Check the extradata */
934 av_log(avctx, AV_LOG_ERROR, "Version %d != 4.\n", version);
935 return AVERROR_INVALIDDATA;
938 if (samples_per_frame != SAMPLES_PER_FRAME &&
939 samples_per_frame != SAMPLES_PER_FRAME * 2) {
940 av_log(avctx, AV_LOG_ERROR, "Unknown amount of samples per frame %d.\n",
942 return AVERROR_INVALIDDATA;
945 if (delay != 0x88E) {
946 av_log(avctx, AV_LOG_ERROR, "Unknown amount of delay %x != 0x88E.\n",
948 return AVERROR_INVALIDDATA;
951 if (q->coding_mode == STEREO)
952 av_log(avctx, AV_LOG_DEBUG, "Normal stereo detected.\n");
953 else if (q->coding_mode == JOINT_STEREO)
954 av_log(avctx, AV_LOG_DEBUG, "Joint stereo detected.\n");
956 av_log(avctx, AV_LOG_ERROR, "Unknown channel coding mode %x!\n",
958 return AVERROR_INVALIDDATA;
961 if (avctx->block_align >= UINT_MAX / 2)
962 return AVERROR(EINVAL);
964 q->decoded_bytes_buffer = av_mallocz(FFALIGN(avctx->block_align, 4) +
965 FF_INPUT_BUFFER_PADDING_SIZE);
966 if (q->decoded_bytes_buffer == NULL)
967 return AVERROR(ENOMEM);
969 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
971 /* initialize the MDCT transform */
972 if ((ret = ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768)) < 0) {
973 av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
974 av_freep(&q->decoded_bytes_buffer);
978 /* init the joint-stereo decoding data */
979 q->weighting_delay[0] = 0;
980 q->weighting_delay[1] = 7;
981 q->weighting_delay[2] = 0;
982 q->weighting_delay[3] = 7;
983 q->weighting_delay[4] = 0;
984 q->weighting_delay[5] = 7;
986 for (i = 0; i < 4; i++) {
987 q->matrix_coeff_index_prev[i] = 3;
988 q->matrix_coeff_index_now[i] = 3;
989 q->matrix_coeff_index_next[i] = 3;
992 avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
993 ff_fmt_convert_init(&q->fmt_conv, avctx);
995 q->units = av_mallocz(sizeof(*q->units) * avctx->channels);
997 atrac3_decode_close(avctx);
998 return AVERROR(ENOMEM);
1004 AVCodec ff_atrac3_decoder = {
1006 .type = AVMEDIA_TYPE_AUDIO,
1007 .id = AV_CODEC_ID_ATRAC3,
1008 .priv_data_size = sizeof(ATRAC3Context),
1009 .init = atrac3_decode_init,
1010 .close = atrac3_decode_close,
1011 .decode = atrac3_decode_frame,
1012 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
1013 .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),
1014 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
1015 AV_SAMPLE_FMT_NONE },