3 * Copyright (c) 2001, 2002 Fabrice Bellard
5 * This file is part of Libav.
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 #include "libavutil/audioconvert.h"
31 #include "mpegaudiodsp.h"
36 * - test lsf / mpeg25 extensively.
39 #include "mpegaudio.h"
40 #include "mpegaudiodecheader.h"
42 #define BACKSTEP_SIZE 512
44 #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES
46 /* layer 3 "granule" */
47 typedef struct GranuleDef {
52 int scalefac_compress;
57 uint8_t scalefac_scale;
58 uint8_t count1table_select;
59 int region_size[3]; /* number of huffman codes in each region */
61 int short_start, long_end; /* long/short band indexes */
62 uint8_t scale_factors[40];
63 DECLARE_ALIGNED(16, INTFLOAT, sb_hybrid)[SBLIMIT * 18]; /* 576 samples */
66 typedef struct MPADecodeContext {
68 uint8_t last_buf[LAST_BUF_SIZE];
70 /* next header (used in free format parsing) */
71 uint32_t free_format_next_header;
74 DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
75 int synth_buf_offset[MPA_MAX_CHANNELS];
76 DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
77 INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
78 GranuleDef granules[2][2]; /* Used in Layer 3 */
79 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
82 AVCodecContext* avctx;
89 # define SHR(a,b) ((a)*(1.0f/(1<<(b))))
90 # define FIXR_OLD(a) ((int)((a) * FRAC_ONE + 0.5))
91 # define FIXR(x) ((float)(x))
92 # define FIXHR(x) ((float)(x))
93 # define MULH3(x, y, s) ((s)*(y)*(x))
94 # define MULLx(x, y, s) ((y)*(x))
95 # define RENAME(a) a ## _float
96 # define OUT_FMT AV_SAMPLE_FMT_FLT
98 # define SHR(a,b) ((a)>>(b))
99 /* WARNING: only correct for positive numbers */
100 # define FIXR_OLD(a) ((int)((a) * FRAC_ONE + 0.5))
101 # define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
102 # define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
103 # define MULH3(x, y, s) MULH((s)*(x), y)
104 # define MULLx(x, y, s) MULL(x,y,s)
105 # define RENAME(a) a ## _fixed
106 # define OUT_FMT AV_SAMPLE_FMT_S16
111 #define HEADER_SIZE 4
113 #include "mpegaudiodata.h"
114 #include "mpegaudiodectab.h"
116 /* vlc structure for decoding layer 3 huffman tables */
117 static VLC huff_vlc[16];
118 static VLC_TYPE huff_vlc_tables[
119 0 + 128 + 128 + 128 + 130 + 128 + 154 + 166 +
120 142 + 204 + 190 + 170 + 542 + 460 + 662 + 414
122 static const int huff_vlc_tables_sizes[16] = {
123 0, 128, 128, 128, 130, 128, 154, 166,
124 142, 204, 190, 170, 542, 460, 662, 414
126 static VLC huff_quad_vlc[2];
127 static VLC_TYPE huff_quad_vlc_tables[128+16][2];
128 static const int huff_quad_vlc_tables_sizes[2] = { 128, 16 };
129 /* computed from band_size_long */
130 static uint16_t band_index_long[9][23];
131 #include "mpegaudio_tablegen.h"
132 /* intensity stereo coef table */
133 static INTFLOAT is_table[2][16];
134 static INTFLOAT is_table_lsf[2][2][16];
135 static INTFLOAT csa_table[8][4];
137 static int16_t division_tab3[1<<6 ];
138 static int16_t division_tab5[1<<8 ];
139 static int16_t division_tab9[1<<11];
141 static int16_t * const division_tabs[4] = {
142 division_tab3, division_tab5, NULL, division_tab9
145 /* lower 2 bits: modulo 3, higher bits: shift */
146 static uint16_t scale_factor_modshift[64];
147 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
148 static int32_t scale_factor_mult[15][3];
149 /* mult table for layer 2 group quantization */
151 #define SCALE_GEN(v) \
152 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
154 static const int32_t scale_factor_mult2[3][3] = {
155 SCALE_GEN(4.0 / 3.0), /* 3 steps */
156 SCALE_GEN(4.0 / 5.0), /* 5 steps */
157 SCALE_GEN(4.0 / 9.0), /* 9 steps */
161 * Convert region offsets to region sizes and truncate
162 * size to big_values.
164 static void ff_region_offset2size(GranuleDef *g)
167 g->region_size[2] = 576 / 2;
168 for (i = 0; i < 3; i++) {
169 k = FFMIN(g->region_size[i], g->big_values);
170 g->region_size[i] = k - j;
175 static void ff_init_short_region(MPADecodeContext *s, GranuleDef *g)
177 if (g->block_type == 2) {
178 if (s->sample_rate_index != 8)
179 g->region_size[0] = (36 / 2);
181 g->region_size[0] = (72 / 2);
183 if (s->sample_rate_index <= 2)
184 g->region_size[0] = (36 / 2);
185 else if (s->sample_rate_index != 8)
186 g->region_size[0] = (54 / 2);
188 g->region_size[0] = (108 / 2);
190 g->region_size[1] = (576 / 2);
193 static void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2)
196 g->region_size[0] = band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
197 /* should not overflow */
198 l = FFMIN(ra1 + ra2 + 2, 22);
199 g->region_size[1] = band_index_long[s->sample_rate_index][ l] >> 1;
202 static void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g)
204 if (g->block_type == 2) {
205 if (g->switch_point) {
206 /* if switched mode, we handle the 36 first samples as
207 long blocks. For 8000Hz, we handle the 72 first
208 exponents as long blocks */
209 if (s->sample_rate_index <= 2)
214 g->short_start = 2 + (s->sample_rate_index != 8);
225 /* layer 1 unscaling */
226 /* n = number of bits of the mantissa minus 1 */
227 static inline int l1_unscale(int n, int mant, int scale_factor)
232 shift = scale_factor_modshift[scale_factor];
235 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
237 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
238 return (int)((val + (1LL << (shift - 1))) >> shift);
241 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
245 shift = scale_factor_modshift[scale_factor];
249 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
250 /* NOTE: at this point, 0 <= shift <= 21 */
252 val = (val + (1 << (shift - 1))) >> shift;
256 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
257 static inline int l3_unscale(int value, int exponent)
262 e = table_4_3_exp [4 * value + (exponent & 3)];
263 m = table_4_3_value[4 * value + (exponent & 3)];
268 m = (m + (1 << (e - 1))) >> e;
273 static av_cold void decode_init_static(void)
278 /* scale factors table for layer 1/2 */
279 for (i = 0; i < 64; i++) {
281 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
284 scale_factor_modshift[i] = mod | (shift << 2);
287 /* scale factor multiply for layer 1 */
288 for (i = 0; i < 15; i++) {
291 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
292 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
293 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
294 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
295 av_dlog(NULL, "%d: norm=%x s=%x %x %x\n", i, norm,
296 scale_factor_mult[i][0],
297 scale_factor_mult[i][1],
298 scale_factor_mult[i][2]);
301 RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
303 /* huffman decode tables */
305 for (i = 1; i < 16; i++) {
306 const HuffTable *h = &mpa_huff_tables[i];
308 uint8_t tmp_bits [512] = { 0 };
309 uint16_t tmp_codes[512] = { 0 };
314 for (x = 0; x < xsize; x++) {
315 for (y = 0; y < xsize; y++) {
316 tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ];
317 tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
322 huff_vlc[i].table = huff_vlc_tables+offset;
323 huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
324 init_vlc(&huff_vlc[i], 7, 512,
325 tmp_bits, 1, 1, tmp_codes, 2, 2,
326 INIT_VLC_USE_NEW_STATIC);
327 offset += huff_vlc_tables_sizes[i];
329 assert(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
332 for (i = 0; i < 2; i++) {
333 huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
334 huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
335 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
336 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
337 INIT_VLC_USE_NEW_STATIC);
338 offset += huff_quad_vlc_tables_sizes[i];
340 assert(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
342 for (i = 0; i < 9; i++) {
344 for (j = 0; j < 22; j++) {
345 band_index_long[i][j] = k;
346 k += band_size_long[i][j];
348 band_index_long[i][22] = k;
351 /* compute n ^ (4/3) and store it in mantissa/exp format */
353 mpegaudio_tableinit();
355 for (i = 0; i < 4; i++) {
356 if (ff_mpa_quant_bits[i] < 0) {
357 for (j = 0; j < (1 << (-ff_mpa_quant_bits[i]+1)); j++) {
358 int val1, val2, val3, steps;
360 steps = ff_mpa_quant_steps[i];
365 division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
371 for (i = 0; i < 7; i++) {
375 f = tan((double)i * M_PI / 12.0);
376 v = FIXR(f / (1.0 + f));
381 is_table[1][6 - i] = v;
384 for (i = 7; i < 16; i++)
385 is_table[0][i] = is_table[1][i] = 0.0;
387 for (i = 0; i < 16; i++) {
391 for (j = 0; j < 2; j++) {
392 e = -(j + 1) * ((i + 1) >> 1);
393 f = pow(2.0, e / 4.0);
395 is_table_lsf[j][k ^ 1][i] = FIXR(f);
396 is_table_lsf[j][k ][i] = FIXR(1.0);
397 av_dlog(NULL, "is_table_lsf %d %d: %f %f\n",
398 i, j, (float) is_table_lsf[j][0][i],
399 (float) is_table_lsf[j][1][i]);
403 for (i = 0; i < 8; i++) {
406 cs = 1.0 / sqrt(1.0 + ci * ci);
409 csa_table[i][0] = FIXHR(cs/4);
410 csa_table[i][1] = FIXHR(ca/4);
411 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
412 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
414 csa_table[i][0] = cs;
415 csa_table[i][1] = ca;
416 csa_table[i][2] = ca + cs;
417 csa_table[i][3] = ca - cs;
422 static av_cold int decode_init(AVCodecContext * avctx)
424 static int initialized_tables = 0;
425 MPADecodeContext *s = avctx->priv_data;
427 if (!initialized_tables) {
428 decode_init_static();
429 initialized_tables = 1;
434 ff_mpadsp_init(&s->mpadsp);
435 ff_dsputil_init(&s->dsp, avctx);
437 avctx->sample_fmt= OUT_FMT;
438 s->err_recognition = avctx->err_recognition;
440 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
443 avcodec_get_frame_defaults(&s->frame);
444 avctx->coded_frame = &s->frame;
449 #define C3 FIXHR(0.86602540378443864676/2)
450 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
451 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
452 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
454 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
456 static void imdct12(INTFLOAT *out, INTFLOAT *in)
458 INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
461 in1 = in[1*3] + in[0*3];
462 in2 = in[2*3] + in[1*3];
463 in3 = in[3*3] + in[2*3];
464 in4 = in[4*3] + in[3*3];
465 in5 = in[5*3] + in[4*3];
469 in2 = MULH3(in2, C3, 2);
470 in3 = MULH3(in3, C3, 4);
473 t2 = MULH3(in1 - in5, C4, 2);
483 in1 = MULH3(in5 + in3, C5, 1);
490 in5 = MULH3(in5 - in3, C6, 2);
497 /* return the number of decoded frames */
498 static int mp_decode_layer1(MPADecodeContext *s)
500 int bound, i, v, n, ch, j, mant;
501 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
502 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
504 if (s->mode == MPA_JSTEREO)
505 bound = (s->mode_ext + 1) * 4;
509 /* allocation bits */
510 for (i = 0; i < bound; i++) {
511 for (ch = 0; ch < s->nb_channels; ch++) {
512 allocation[ch][i] = get_bits(&s->gb, 4);
515 for (i = bound; i < SBLIMIT; i++)
516 allocation[0][i] = get_bits(&s->gb, 4);
519 for (i = 0; i < bound; i++) {
520 for (ch = 0; ch < s->nb_channels; ch++) {
521 if (allocation[ch][i])
522 scale_factors[ch][i] = get_bits(&s->gb, 6);
525 for (i = bound; i < SBLIMIT; i++) {
526 if (allocation[0][i]) {
527 scale_factors[0][i] = get_bits(&s->gb, 6);
528 scale_factors[1][i] = get_bits(&s->gb, 6);
532 /* compute samples */
533 for (j = 0; j < 12; j++) {
534 for (i = 0; i < bound; i++) {
535 for (ch = 0; ch < s->nb_channels; ch++) {
536 n = allocation[ch][i];
538 mant = get_bits(&s->gb, n + 1);
539 v = l1_unscale(n, mant, scale_factors[ch][i]);
543 s->sb_samples[ch][j][i] = v;
546 for (i = bound; i < SBLIMIT; i++) {
547 n = allocation[0][i];
549 mant = get_bits(&s->gb, n + 1);
550 v = l1_unscale(n, mant, scale_factors[0][i]);
551 s->sb_samples[0][j][i] = v;
552 v = l1_unscale(n, mant, scale_factors[1][i]);
553 s->sb_samples[1][j][i] = v;
555 s->sb_samples[0][j][i] = 0;
556 s->sb_samples[1][j][i] = 0;
563 static int mp_decode_layer2(MPADecodeContext *s)
565 int sblimit; /* number of used subbands */
566 const unsigned char *alloc_table;
567 int table, bit_alloc_bits, i, j, ch, bound, v;
568 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
569 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
570 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
571 int scale, qindex, bits, steps, k, l, m, b;
573 /* select decoding table */
574 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
575 s->sample_rate, s->lsf);
576 sblimit = ff_mpa_sblimit_table[table];
577 alloc_table = ff_mpa_alloc_tables[table];
579 if (s->mode == MPA_JSTEREO)
580 bound = (s->mode_ext + 1) * 4;
584 av_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
590 /* parse bit allocation */
592 for (i = 0; i < bound; i++) {
593 bit_alloc_bits = alloc_table[j];
594 for (ch = 0; ch < s->nb_channels; ch++)
595 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
596 j += 1 << bit_alloc_bits;
598 for (i = bound; i < sblimit; i++) {
599 bit_alloc_bits = alloc_table[j];
600 v = get_bits(&s->gb, bit_alloc_bits);
603 j += 1 << bit_alloc_bits;
607 for (i = 0; i < sblimit; i++) {
608 for (ch = 0; ch < s->nb_channels; ch++) {
609 if (bit_alloc[ch][i])
610 scale_code[ch][i] = get_bits(&s->gb, 2);
615 for (i = 0; i < sblimit; i++) {
616 for (ch = 0; ch < s->nb_channels; ch++) {
617 if (bit_alloc[ch][i]) {
618 sf = scale_factors[ch][i];
619 switch (scale_code[ch][i]) {
622 sf[0] = get_bits(&s->gb, 6);
623 sf[1] = get_bits(&s->gb, 6);
624 sf[2] = get_bits(&s->gb, 6);
627 sf[0] = get_bits(&s->gb, 6);
632 sf[0] = get_bits(&s->gb, 6);
633 sf[2] = get_bits(&s->gb, 6);
637 sf[0] = get_bits(&s->gb, 6);
638 sf[2] = get_bits(&s->gb, 6);
647 for (k = 0; k < 3; k++) {
648 for (l = 0; l < 12; l += 3) {
650 for (i = 0; i < bound; i++) {
651 bit_alloc_bits = alloc_table[j];
652 for (ch = 0; ch < s->nb_channels; ch++) {
653 b = bit_alloc[ch][i];
655 scale = scale_factors[ch][i][k];
656 qindex = alloc_table[j+b];
657 bits = ff_mpa_quant_bits[qindex];
660 /* 3 values at the same time */
661 v = get_bits(&s->gb, -bits);
662 v2 = division_tabs[qindex][v];
663 steps = ff_mpa_quant_steps[qindex];
665 s->sb_samples[ch][k * 12 + l + 0][i] =
666 l2_unscale_group(steps, v2 & 15, scale);
667 s->sb_samples[ch][k * 12 + l + 1][i] =
668 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
669 s->sb_samples[ch][k * 12 + l + 2][i] =
670 l2_unscale_group(steps, v2 >> 8 , scale);
672 for (m = 0; m < 3; m++) {
673 v = get_bits(&s->gb, bits);
674 v = l1_unscale(bits - 1, v, scale);
675 s->sb_samples[ch][k * 12 + l + m][i] = v;
679 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
680 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
681 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
684 /* next subband in alloc table */
685 j += 1 << bit_alloc_bits;
687 /* XXX: find a way to avoid this duplication of code */
688 for (i = bound; i < sblimit; i++) {
689 bit_alloc_bits = alloc_table[j];
692 int mant, scale0, scale1;
693 scale0 = scale_factors[0][i][k];
694 scale1 = scale_factors[1][i][k];
695 qindex = alloc_table[j+b];
696 bits = ff_mpa_quant_bits[qindex];
698 /* 3 values at the same time */
699 v = get_bits(&s->gb, -bits);
700 steps = ff_mpa_quant_steps[qindex];
703 s->sb_samples[0][k * 12 + l + 0][i] =
704 l2_unscale_group(steps, mant, scale0);
705 s->sb_samples[1][k * 12 + l + 0][i] =
706 l2_unscale_group(steps, mant, scale1);
709 s->sb_samples[0][k * 12 + l + 1][i] =
710 l2_unscale_group(steps, mant, scale0);
711 s->sb_samples[1][k * 12 + l + 1][i] =
712 l2_unscale_group(steps, mant, scale1);
713 s->sb_samples[0][k * 12 + l + 2][i] =
714 l2_unscale_group(steps, v, scale0);
715 s->sb_samples[1][k * 12 + l + 2][i] =
716 l2_unscale_group(steps, v, scale1);
718 for (m = 0; m < 3; m++) {
719 mant = get_bits(&s->gb, bits);
720 s->sb_samples[0][k * 12 + l + m][i] =
721 l1_unscale(bits - 1, mant, scale0);
722 s->sb_samples[1][k * 12 + l + m][i] =
723 l1_unscale(bits - 1, mant, scale1);
727 s->sb_samples[0][k * 12 + l + 0][i] = 0;
728 s->sb_samples[0][k * 12 + l + 1][i] = 0;
729 s->sb_samples[0][k * 12 + l + 2][i] = 0;
730 s->sb_samples[1][k * 12 + l + 0][i] = 0;
731 s->sb_samples[1][k * 12 + l + 1][i] = 0;
732 s->sb_samples[1][k * 12 + l + 2][i] = 0;
734 /* next subband in alloc table */
735 j += 1 << bit_alloc_bits;
737 /* fill remaining samples to zero */
738 for (i = sblimit; i < SBLIMIT; i++) {
739 for (ch = 0; ch < s->nb_channels; ch++) {
740 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
741 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
742 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
750 #define SPLIT(dst,sf,n) \
752 int m = (sf * 171) >> 9; \
755 } else if (n == 4) { \
758 } else if (n == 5) { \
759 int m = (sf * 205) >> 10; \
762 } else if (n == 6) { \
763 int m = (sf * 171) >> 10; \
770 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
773 SPLIT(slen[3], sf, n3)
774 SPLIT(slen[2], sf, n2)
775 SPLIT(slen[1], sf, n1)
779 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
782 const uint8_t *bstab, *pretab;
783 int len, i, j, k, l, v0, shift, gain, gains[3];
787 gain = g->global_gain - 210;
788 shift = g->scalefac_scale + 1;
790 bstab = band_size_long[s->sample_rate_index];
791 pretab = mpa_pretab[g->preflag];
792 for (i = 0; i < g->long_end; i++) {
793 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
795 for (j = len; j > 0; j--)
799 if (g->short_start < 13) {
800 bstab = band_size_short[s->sample_rate_index];
801 gains[0] = gain - (g->subblock_gain[0] << 3);
802 gains[1] = gain - (g->subblock_gain[1] << 3);
803 gains[2] = gain - (g->subblock_gain[2] << 3);
805 for (i = g->short_start; i < 13; i++) {
807 for (l = 0; l < 3; l++) {
808 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
809 for (j = len; j > 0; j--)
816 /* handle n = 0 too */
817 static inline int get_bitsz(GetBitContext *s, int n)
819 return n ? get_bits(s, n) : 0;
823 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
826 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits) {
828 s->in_gb.buffer = NULL;
829 assert((get_bits_count(&s->gb) & 7) == 0);
830 skip_bits_long(&s->gb, *pos - *end_pos);
832 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
833 *pos = get_bits_count(&s->gb);
837 /* Following is a optimized code for
839 if(get_bits1(&s->gb))
844 #define READ_FLIP_SIGN(dst,src) \
845 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
848 #define READ_FLIP_SIGN(dst,src) \
849 v = -get_bits1(&s->gb); \
850 *(dst) = (*(src) ^ v) - v;
853 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
854 int16_t *exponents, int end_pos2)
858 int last_pos, bits_left;
860 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits);
862 /* low frequencies (called big values) */
864 for (i = 0; i < 3; i++) {
865 int j, k, l, linbits;
866 j = g->region_size[i];
869 /* select vlc table */
870 k = g->table_select[i];
871 l = mpa_huff_data[k][0];
872 linbits = mpa_huff_data[k][1];
876 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
881 /* read huffcode and compute each couple */
885 int pos = get_bits_count(&s->gb);
888 // av_log(NULL, AV_LOG_ERROR, "pos: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
889 switch_buffer(s, &pos, &end_pos, &end_pos2);
890 // av_log(NULL, AV_LOG_ERROR, "new pos: %d %d\n", pos, end_pos);
894 y = get_vlc2(&s->gb, vlc->table, 7, 3);
897 g->sb_hybrid[s_index ] =
898 g->sb_hybrid[s_index+1] = 0;
903 exponent= exponents[s_index];
905 av_dlog(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
906 i, g->region_size[i] - j, x, y, exponent);
911 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
913 x += get_bitsz(&s->gb, linbits);
914 v = l3_unscale(x, exponent);
915 if (get_bits1(&s->gb))
917 g->sb_hybrid[s_index] = v;
920 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
922 y += get_bitsz(&s->gb, linbits);
923 v = l3_unscale(y, exponent);
924 if (get_bits1(&s->gb))
926 g->sb_hybrid[s_index+1] = v;
933 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
935 x += get_bitsz(&s->gb, linbits);
936 v = l3_unscale(x, exponent);
937 if (get_bits1(&s->gb))
939 g->sb_hybrid[s_index+!!y] = v;
941 g->sb_hybrid[s_index + !y] = 0;
947 /* high frequencies */
948 vlc = &huff_quad_vlc[g->count1table_select];
950 while (s_index <= 572) {
952 pos = get_bits_count(&s->gb);
953 if (pos >= end_pos) {
954 if (pos > end_pos2 && last_pos) {
955 /* some encoders generate an incorrect size for this
956 part. We must go back into the data */
958 skip_bits_long(&s->gb, last_pos - pos);
959 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
960 if(s->err_recognition & AV_EF_BITSTREAM)
964 // av_log(NULL, AV_LOG_ERROR, "pos2: %d %d %d %d\n", pos, end_pos, end_pos2, s_index);
965 switch_buffer(s, &pos, &end_pos, &end_pos2);
966 // av_log(NULL, AV_LOG_ERROR, "new pos2: %d %d %d\n", pos, end_pos, s_index);
972 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
973 av_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
974 g->sb_hybrid[s_index+0] =
975 g->sb_hybrid[s_index+1] =
976 g->sb_hybrid[s_index+2] =
977 g->sb_hybrid[s_index+3] = 0;
979 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
981 int pos = s_index + idxtab[code];
982 code ^= 8 >> idxtab[code];
983 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
987 /* skip extension bits */
988 bits_left = end_pos2 - get_bits_count(&s->gb);
989 //av_log(NULL, AV_LOG_ERROR, "left:%d buf:%p\n", bits_left, s->in_gb.buffer);
990 if (bits_left < 0 && (s->err_recognition & AV_EF_BUFFER)) {
991 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
993 } else if (bits_left > 0 && (s->err_recognition & AV_EF_BUFFER)) {
994 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
997 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
998 skip_bits_long(&s->gb, bits_left);
1000 i = get_bits_count(&s->gb);
1001 switch_buffer(s, &i, &end_pos, &end_pos2);
1006 /* Reorder short blocks from bitstream order to interleaved order. It
1007 would be faster to do it in parsing, but the code would be far more
1009 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1012 INTFLOAT *ptr, *dst, *ptr1;
1015 if (g->block_type != 2)
1018 if (g->switch_point) {
1019 if (s->sample_rate_index != 8)
1020 ptr = g->sb_hybrid + 36;
1022 ptr = g->sb_hybrid + 72;
1027 for (i = g->short_start; i < 13; i++) {
1028 len = band_size_short[s->sample_rate_index][i];
1031 for (j = len; j > 0; j--) {
1032 *dst++ = ptr[0*len];
1033 *dst++ = ptr[1*len];
1034 *dst++ = ptr[2*len];
1038 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1042 #define ISQRT2 FIXR(0.70710678118654752440)
1044 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1047 int sf_max, sf, len, non_zero_found;
1048 INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1049 int non_zero_found_short[3];
1051 /* intensity stereo */
1052 if (s->mode_ext & MODE_EXT_I_STEREO) {
1057 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1061 tab0 = g0->sb_hybrid + 576;
1062 tab1 = g1->sb_hybrid + 576;
1064 non_zero_found_short[0] = 0;
1065 non_zero_found_short[1] = 0;
1066 non_zero_found_short[2] = 0;
1067 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1068 for (i = 12; i >= g1->short_start; i--) {
1069 /* for last band, use previous scale factor */
1072 len = band_size_short[s->sample_rate_index][i];
1073 for (l = 2; l >= 0; l--) {
1076 if (!non_zero_found_short[l]) {
1077 /* test if non zero band. if so, stop doing i-stereo */
1078 for (j = 0; j < len; j++) {
1080 non_zero_found_short[l] = 1;
1084 sf = g1->scale_factors[k + l];
1090 for (j = 0; j < len; j++) {
1092 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1093 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1097 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1098 /* lower part of the spectrum : do ms stereo
1100 for (j = 0; j < len; j++) {
1103 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1104 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1111 non_zero_found = non_zero_found_short[0] |
1112 non_zero_found_short[1] |
1113 non_zero_found_short[2];
1115 for (i = g1->long_end - 1;i >= 0;i--) {
1116 len = band_size_long[s->sample_rate_index][i];
1119 /* test if non zero band. if so, stop doing i-stereo */
1120 if (!non_zero_found) {
1121 for (j = 0; j < len; j++) {
1127 /* for last band, use previous scale factor */
1128 k = (i == 21) ? 20 : i;
1129 sf = g1->scale_factors[k];
1134 for (j = 0; j < len; j++) {
1136 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1137 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1141 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1142 /* lower part of the spectrum : do ms stereo
1144 for (j = 0; j < len; j++) {
1147 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1148 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1153 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1154 /* ms stereo ONLY */
1155 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1158 s-> dsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1160 tab0 = g0->sb_hybrid;
1161 tab1 = g1->sb_hybrid;
1162 for (i = 0; i < 576; i++) {
1165 tab0[i] = tmp0 + tmp1;
1166 tab1[i] = tmp0 - tmp1;
1173 #define AA(j) do { \
1174 float tmp0 = ptr[-1-j]; \
1175 float tmp1 = ptr[ j]; \
1176 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1177 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1180 #define AA(j) do { \
1181 int tmp0 = ptr[-1-j]; \
1182 int tmp1 = ptr[ j]; \
1183 int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1184 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1185 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1189 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1194 /* we antialias only "long" bands */
1195 if (g->block_type == 2) {
1196 if (!g->switch_point)
1198 /* XXX: check this for 8000Hz case */
1204 ptr = g->sb_hybrid + 18;
1205 for (i = n; i > 0; i--) {
1219 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1220 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1222 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1224 int i, j, mdct_long_end, sblimit;
1226 /* find last non zero block */
1227 ptr = g->sb_hybrid + 576;
1228 ptr1 = g->sb_hybrid + 2 * 18;
1229 while (ptr >= ptr1) {
1233 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1236 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1238 if (g->block_type == 2) {
1239 /* XXX: check for 8000 Hz */
1240 if (g->switch_point)
1245 mdct_long_end = sblimit;
1248 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1249 mdct_long_end, g->switch_point,
1252 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1253 ptr = g->sb_hybrid + 18 * mdct_long_end;
1255 for (j = mdct_long_end; j < sblimit; j++) {
1256 /* select frequency inversion */
1257 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1258 out_ptr = sb_samples + j;
1260 for (i = 0; i < 6; i++) {
1261 *out_ptr = buf[4*i];
1264 imdct12(out2, ptr + 0);
1265 for (i = 0; i < 6; i++) {
1266 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1267 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1270 imdct12(out2, ptr + 1);
1271 for (i = 0; i < 6; i++) {
1272 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1273 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1276 imdct12(out2, ptr + 2);
1277 for (i = 0; i < 6; i++) {
1278 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1279 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1280 buf[4*(i + 6*2)] = 0;
1283 buf += (j&3) != 3 ? 1 : (4*18-3);
1286 for (j = sblimit; j < SBLIMIT; j++) {
1288 out_ptr = sb_samples + j;
1289 for (i = 0; i < 18; i++) {
1290 *out_ptr = buf[4*i];
1294 buf += (j&3) != 3 ? 1 : (4*18-3);
1298 /* main layer3 decoding function */
1299 static int mp_decode_layer3(MPADecodeContext *s)
1301 int nb_granules, main_data_begin;
1302 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1304 int16_t exponents[576]; //FIXME try INTFLOAT
1306 /* read side info */
1308 main_data_begin = get_bits(&s->gb, 8);
1309 skip_bits(&s->gb, s->nb_channels);
1312 main_data_begin = get_bits(&s->gb, 9);
1313 if (s->nb_channels == 2)
1314 skip_bits(&s->gb, 3);
1316 skip_bits(&s->gb, 5);
1318 for (ch = 0; ch < s->nb_channels; ch++) {
1319 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1320 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1324 for (gr = 0; gr < nb_granules; gr++) {
1325 for (ch = 0; ch < s->nb_channels; ch++) {
1326 av_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1327 g = &s->granules[ch][gr];
1328 g->part2_3_length = get_bits(&s->gb, 12);
1329 g->big_values = get_bits(&s->gb, 9);
1330 if (g->big_values > 288) {
1331 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1332 return AVERROR_INVALIDDATA;
1335 g->global_gain = get_bits(&s->gb, 8);
1336 /* if MS stereo only is selected, we precompute the
1337 1/sqrt(2) renormalization factor */
1338 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1340 g->global_gain -= 2;
1342 g->scalefac_compress = get_bits(&s->gb, 9);
1344 g->scalefac_compress = get_bits(&s->gb, 4);
1345 blocksplit_flag = get_bits1(&s->gb);
1346 if (blocksplit_flag) {
1347 g->block_type = get_bits(&s->gb, 2);
1348 if (g->block_type == 0) {
1349 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1350 return AVERROR_INVALIDDATA;
1352 g->switch_point = get_bits1(&s->gb);
1353 for (i = 0; i < 2; i++)
1354 g->table_select[i] = get_bits(&s->gb, 5);
1355 for (i = 0; i < 3; i++)
1356 g->subblock_gain[i] = get_bits(&s->gb, 3);
1357 ff_init_short_region(s, g);
1359 int region_address1, region_address2;
1361 g->switch_point = 0;
1362 for (i = 0; i < 3; i++)
1363 g->table_select[i] = get_bits(&s->gb, 5);
1364 /* compute huffman coded region sizes */
1365 region_address1 = get_bits(&s->gb, 4);
1366 region_address2 = get_bits(&s->gb, 3);
1367 av_dlog(s->avctx, "region1=%d region2=%d\n",
1368 region_address1, region_address2);
1369 ff_init_long_region(s, g, region_address1, region_address2);
1371 ff_region_offset2size(g);
1372 ff_compute_band_indexes(s, g);
1376 g->preflag = get_bits1(&s->gb);
1377 g->scalefac_scale = get_bits1(&s->gb);
1378 g->count1table_select = get_bits1(&s->gb);
1379 av_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1380 g->block_type, g->switch_point);
1386 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1387 int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0,
1388 FFMAX(0, LAST_BUF_SIZE - s->last_buf_size));
1389 assert((get_bits_count(&s->gb) & 7) == 0);
1390 /* now we get bits from the main_data_begin offset */
1391 av_dlog(s->avctx, "seekback: %d\n", main_data_begin);
1392 //av_log(NULL, AV_LOG_ERROR, "backstep:%d, lastbuf:%d\n", main_data_begin, s->last_buf_size);
1394 memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);
1396 init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
1397 #if !UNCHECKED_BITSTREAM_READER
1398 s->gb.size_in_bits_plus8 += extrasize * 8;
1400 s->last_buf_size <<= 3;
1401 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1402 for (ch = 0; ch < s->nb_channels; ch++) {
1403 g = &s->granules[ch][gr];
1404 s->last_buf_size += g->part2_3_length;
1405 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1408 skip = s->last_buf_size - 8 * main_data_begin;
1409 if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {
1410 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);
1412 s->in_gb.buffer = NULL;
1414 skip_bits_long(&s->gb, skip);
1420 for (; gr < nb_granules; gr++) {
1421 for (ch = 0; ch < s->nb_channels; ch++) {
1422 g = &s->granules[ch][gr];
1423 bits_pos = get_bits_count(&s->gb);
1427 int slen, slen1, slen2;
1429 /* MPEG1 scale factors */
1430 slen1 = slen_table[0][g->scalefac_compress];
1431 slen2 = slen_table[1][g->scalefac_compress];
1432 av_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1433 if (g->block_type == 2) {
1434 n = g->switch_point ? 17 : 18;
1437 for (i = 0; i < n; i++)
1438 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1440 for (i = 0; i < n; i++)
1441 g->scale_factors[j++] = 0;
1444 for (i = 0; i < 18; i++)
1445 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1446 for (i = 0; i < 3; i++)
1447 g->scale_factors[j++] = 0;
1449 for (i = 0; i < 21; i++)
1450 g->scale_factors[j++] = 0;
1453 sc = s->granules[ch][0].scale_factors;
1455 for (k = 0; k < 4; k++) {
1457 if ((g->scfsi & (0x8 >> k)) == 0) {
1458 slen = (k < 2) ? slen1 : slen2;
1460 for (i = 0; i < n; i++)
1461 g->scale_factors[j++] = get_bits(&s->gb, slen);
1463 for (i = 0; i < n; i++)
1464 g->scale_factors[j++] = 0;
1467 /* simply copy from last granule */
1468 for (i = 0; i < n; i++) {
1469 g->scale_factors[j] = sc[j];
1474 g->scale_factors[j++] = 0;
1477 int tindex, tindex2, slen[4], sl, sf;
1479 /* LSF scale factors */
1480 if (g->block_type == 2)
1481 tindex = g->switch_point ? 2 : 1;
1485 sf = g->scalefac_compress;
1486 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1487 /* intensity stereo case */
1490 lsf_sf_expand(slen, sf, 6, 6, 0);
1492 } else if (sf < 244) {
1493 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1496 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1502 lsf_sf_expand(slen, sf, 5, 4, 4);
1504 } else if (sf < 500) {
1505 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1508 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1515 for (k = 0; k < 4; k++) {
1516 n = lsf_nsf_table[tindex2][tindex][k];
1519 for (i = 0; i < n; i++)
1520 g->scale_factors[j++] = get_bits(&s->gb, sl);
1522 for (i = 0; i < n; i++)
1523 g->scale_factors[j++] = 0;
1526 /* XXX: should compute exact size */
1528 g->scale_factors[j] = 0;
1531 exponents_from_scale_factors(s, g, exponents);
1533 /* read Huffman coded residue */
1534 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1537 if (s->mode == MPA_JSTEREO)
1538 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1540 for (ch = 0; ch < s->nb_channels; ch++) {
1541 g = &s->granules[ch][gr];
1543 reorder_block(s, g);
1544 compute_antialias(s, g);
1545 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1548 if (get_bits_count(&s->gb) < 0)
1549 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1550 return nb_granules * 18;
1553 static int mp_decode_frame(MPADecodeContext *s, OUT_INT *samples,
1554 const uint8_t *buf, int buf_size)
1556 int i, nb_frames, ch, ret;
1557 OUT_INT *samples_ptr;
1559 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1561 /* skip error protection field */
1562 if (s->error_protection)
1563 skip_bits(&s->gb, 16);
1567 s->avctx->frame_size = 384;
1568 nb_frames = mp_decode_layer1(s);
1571 s->avctx->frame_size = 1152;
1572 nb_frames = mp_decode_layer2(s);
1575 s->avctx->frame_size = s->lsf ? 576 : 1152;
1577 nb_frames = mp_decode_layer3(s);
1580 if (s->in_gb.buffer) {
1581 align_get_bits(&s->gb);
1582 i = get_bits_left(&s->gb)>>3;
1583 if (i >= 0 && i <= BACKSTEP_SIZE) {
1584 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1587 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1589 s->in_gb.buffer = NULL;
1592 align_get_bits(&s->gb);
1593 assert((get_bits_count(&s->gb) & 7) == 0);
1594 i = get_bits_left(&s->gb) >> 3;
1596 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1598 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1599 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1601 assert(i <= buf_size - HEADER_SIZE && i >= 0);
1602 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1603 s->last_buf_size += i;
1606 /* get output buffer */
1608 s->frame.nb_samples = s->avctx->frame_size;
1609 if ((ret = s->avctx->get_buffer(s->avctx, &s->frame)) < 0) {
1610 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1613 samples = (OUT_INT *)s->frame.data[0];
1616 /* apply the synthesis filter */
1617 for (ch = 0; ch < s->nb_channels; ch++) {
1618 samples_ptr = samples + ch;
1619 for (i = 0; i < nb_frames; i++) {
1620 RENAME(ff_mpa_synth_filter)(
1622 s->synth_buf[ch], &(s->synth_buf_offset[ch]),
1623 RENAME(ff_mpa_synth_window), &s->dither_state,
1624 samples_ptr, s->nb_channels,
1625 s->sb_samples[ch][i]);
1626 samples_ptr += 32 * s->nb_channels;
1630 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1633 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1636 const uint8_t *buf = avpkt->data;
1637 int buf_size = avpkt->size;
1638 MPADecodeContext *s = avctx->priv_data;
1642 if (buf_size < HEADER_SIZE)
1643 return AVERROR_INVALIDDATA;
1645 header = AV_RB32(buf);
1646 if (ff_mpa_check_header(header) < 0) {
1647 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1648 return AVERROR_INVALIDDATA;
1651 if (avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
1652 /* free format: prepare to compute frame size */
1654 return AVERROR_INVALIDDATA;
1656 /* update codec info */
1657 avctx->channels = s->nb_channels;
1658 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1659 if (!avctx->bit_rate)
1660 avctx->bit_rate = s->bit_rate;
1662 if (s->frame_size <= 0 || s->frame_size > buf_size) {
1663 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1664 return AVERROR_INVALIDDATA;
1665 } else if (s->frame_size < buf_size) {
1666 buf_size= s->frame_size;
1669 out_size = mp_decode_frame(s, NULL, buf, buf_size);
1670 if (out_size >= 0) {
1672 *(AVFrame *)data = s->frame;
1673 avctx->sample_rate = s->sample_rate;
1674 //FIXME maybe move the other codec info stuff from above here too
1676 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1677 /* Only return an error if the bad frame makes up the whole packet.
1678 If there is more data in the packet, just consume the bad frame
1679 instead of returning an error, which would discard the whole
1682 if (buf_size == avpkt->size)
1689 static void flush(AVCodecContext *avctx)
1691 MPADecodeContext *s = avctx->priv_data;
1692 memset(s->synth_buf, 0, sizeof(s->synth_buf));
1693 s->last_buf_size = 0;
1696 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1697 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1698 int *got_frame_ptr, AVPacket *avpkt)
1700 const uint8_t *buf = avpkt->data;
1701 int buf_size = avpkt->size;
1702 MPADecodeContext *s = avctx->priv_data;
1708 // Discard too short frames
1709 if (buf_size < HEADER_SIZE) {
1710 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1711 return AVERROR_INVALIDDATA;
1715 if (len > MPA_MAX_CODED_FRAME_SIZE)
1716 len = MPA_MAX_CODED_FRAME_SIZE;
1718 // Get header and restore sync word
1719 header = AV_RB32(buf) | 0xffe00000;
1721 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
1722 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1723 return AVERROR_INVALIDDATA;
1726 avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1727 /* update codec info */
1728 avctx->sample_rate = s->sample_rate;
1729 avctx->channels = s->nb_channels;
1730 if (!avctx->bit_rate)
1731 avctx->bit_rate = s->bit_rate;
1733 s->frame_size = len;
1735 out_size = mp_decode_frame(s, NULL, buf, buf_size);
1737 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1738 return AVERROR_INVALIDDATA;
1742 *(AVFrame *)data = s->frame;
1746 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1748 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1751 * Context for MP3On4 decoder
1753 typedef struct MP3On4DecodeContext {
1755 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1756 int syncword; ///< syncword patch
1757 const uint8_t *coff; ///< channel offsets in output buffer
1758 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1759 OUT_INT *decoded_buf; ///< output buffer for decoded samples
1760 } MP3On4DecodeContext;
1762 #include "mpeg4audio.h"
1764 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1766 /* number of mp3 decoder instances */
1767 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1769 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1770 static const uint8_t chan_offset[8][5] = {
1775 { 2, 0, 3 }, // C FLR BS
1776 { 2, 0, 3 }, // C FLR BLRS
1777 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1778 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1781 /* mp3on4 channel layouts */
1782 static const int16_t chan_layout[8] = {
1785 AV_CH_LAYOUT_STEREO,
1786 AV_CH_LAYOUT_SURROUND,
1787 AV_CH_LAYOUT_4POINT0,
1788 AV_CH_LAYOUT_5POINT0,
1789 AV_CH_LAYOUT_5POINT1,
1790 AV_CH_LAYOUT_7POINT1
1793 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1795 MP3On4DecodeContext *s = avctx->priv_data;
1798 for (i = 0; i < s->frames; i++)
1799 av_free(s->mp3decctx[i]);
1801 av_freep(&s->decoded_buf);
1807 static int decode_init_mp3on4(AVCodecContext * avctx)
1809 MP3On4DecodeContext *s = avctx->priv_data;
1810 MPEG4AudioConfig cfg;
1813 if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
1814 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1815 return AVERROR_INVALIDDATA;
1818 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
1819 avctx->extradata_size * 8, 1);
1820 if (!cfg.chan_config || cfg.chan_config > 7) {
1821 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1822 return AVERROR_INVALIDDATA;
1824 s->frames = mp3Frames[cfg.chan_config];
1825 s->coff = chan_offset[cfg.chan_config];
1826 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1827 avctx->channel_layout = chan_layout[cfg.chan_config];
1829 if (cfg.sample_rate < 16000)
1830 s->syncword = 0xffe00000;
1832 s->syncword = 0xfff00000;
1834 /* Init the first mp3 decoder in standard way, so that all tables get builded
1835 * We replace avctx->priv_data with the context of the first decoder so that
1836 * decode_init() does not have to be changed.
1837 * Other decoders will be initialized here copying data from the first context
1839 // Allocate zeroed memory for the first decoder context
1840 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1841 if (!s->mp3decctx[0])
1843 // Put decoder context in place to make init_decode() happy
1844 avctx->priv_data = s->mp3decctx[0];
1846 s->frame = avctx->coded_frame;
1847 // Restore mp3on4 context pointer
1848 avctx->priv_data = s;
1849 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1851 /* Create a separate codec/context for each frame (first is already ok).
1852 * Each frame is 1 or 2 channels - up to 5 frames allowed
1854 for (i = 1; i < s->frames; i++) {
1855 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1856 if (!s->mp3decctx[i])
1858 s->mp3decctx[i]->adu_mode = 1;
1859 s->mp3decctx[i]->avctx = avctx;
1860 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1863 /* Allocate buffer for multi-channel output if needed */
1864 if (s->frames > 1) {
1865 s->decoded_buf = av_malloc(MPA_FRAME_SIZE * MPA_MAX_CHANNELS *
1866 sizeof(*s->decoded_buf));
1867 if (!s->decoded_buf)
1873 decode_close_mp3on4(avctx);
1874 return AVERROR(ENOMEM);
1878 static void flush_mp3on4(AVCodecContext *avctx)
1881 MP3On4DecodeContext *s = avctx->priv_data;
1883 for (i = 0; i < s->frames; i++) {
1884 MPADecodeContext *m = s->mp3decctx[i];
1885 memset(m->synth_buf, 0, sizeof(m->synth_buf));
1886 m->last_buf_size = 0;
1891 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1892 int *got_frame_ptr, AVPacket *avpkt)
1894 const uint8_t *buf = avpkt->data;
1895 int buf_size = avpkt->size;
1896 MP3On4DecodeContext *s = avctx->priv_data;
1897 MPADecodeContext *m;
1898 int fsize, len = buf_size, out_size = 0;
1900 OUT_INT *out_samples;
1901 OUT_INT *outptr, *bp;
1902 int fr, j, n, ch, ret;
1904 /* get output buffer */
1905 s->frame->nb_samples = MPA_FRAME_SIZE;
1906 if ((ret = avctx->get_buffer(avctx, s->frame)) < 0) {
1907 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1910 out_samples = (OUT_INT *)s->frame->data[0];
1912 // Discard too short frames
1913 if (buf_size < HEADER_SIZE)
1914 return AVERROR_INVALIDDATA;
1916 // If only one decoder interleave is not needed
1917 outptr = s->frames == 1 ? out_samples : s->decoded_buf;
1919 avctx->bit_rate = 0;
1922 for (fr = 0; fr < s->frames; fr++) {
1923 fsize = AV_RB16(buf) >> 4;
1924 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1925 m = s->mp3decctx[fr];
1928 if (fsize < HEADER_SIZE) {
1929 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1930 return AVERROR_INVALIDDATA;
1932 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1934 if (ff_mpa_check_header(header) < 0) // Bad header, discard block
1937 avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1939 if (ch + m->nb_channels > avctx->channels) {
1940 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1942 return AVERROR_INVALIDDATA;
1944 ch += m->nb_channels;
1946 out_size += mp_decode_frame(m, outptr, buf, fsize);
1950 if (s->frames > 1) {
1951 n = m->avctx->frame_size*m->nb_channels;
1952 /* interleave output data */
1953 bp = out_samples + s->coff[fr];
1954 if (m->nb_channels == 1) {
1955 for (j = 0; j < n; j++) {
1956 *bp = s->decoded_buf[j];
1957 bp += avctx->channels;
1960 for (j = 0; j < n; j++) {
1961 bp[0] = s->decoded_buf[j++];
1962 bp[1] = s->decoded_buf[j];
1963 bp += avctx->channels;
1967 avctx->bit_rate += m->bit_rate;
1970 /* update codec info */
1971 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1973 s->frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1975 *(AVFrame *)data = *s->frame;
1979 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */
1982 #if CONFIG_MP1_DECODER
1983 AVCodec ff_mp1_decoder = {
1985 .type = AVMEDIA_TYPE_AUDIO,
1986 .id = AV_CODEC_ID_MP1,
1987 .priv_data_size = sizeof(MPADecodeContext),
1988 .init = decode_init,
1989 .decode = decode_frame,
1990 .capabilities = CODEC_CAP_DR1,
1992 .long_name = NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),
1995 #if CONFIG_MP2_DECODER
1996 AVCodec ff_mp2_decoder = {
1998 .type = AVMEDIA_TYPE_AUDIO,
1999 .id = AV_CODEC_ID_MP2,
2000 .priv_data_size = sizeof(MPADecodeContext),
2001 .init = decode_init,
2002 .decode = decode_frame,
2003 .capabilities = CODEC_CAP_DR1,
2005 .long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
2008 #if CONFIG_MP3_DECODER
2009 AVCodec ff_mp3_decoder = {
2011 .type = AVMEDIA_TYPE_AUDIO,
2012 .id = AV_CODEC_ID_MP3,
2013 .priv_data_size = sizeof(MPADecodeContext),
2014 .init = decode_init,
2015 .decode = decode_frame,
2016 .capabilities = CODEC_CAP_DR1,
2018 .long_name = NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),
2021 #if CONFIG_MP3ADU_DECODER
2022 AVCodec ff_mp3adu_decoder = {
2024 .type = AVMEDIA_TYPE_AUDIO,
2025 .id = AV_CODEC_ID_MP3ADU,
2026 .priv_data_size = sizeof(MPADecodeContext),
2027 .init = decode_init,
2028 .decode = decode_frame_adu,
2029 .capabilities = CODEC_CAP_DR1,
2031 .long_name = NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),
2034 #if CONFIG_MP3ON4_DECODER
2035 AVCodec ff_mp3on4_decoder = {
2037 .type = AVMEDIA_TYPE_AUDIO,
2038 .id = AV_CODEC_ID_MP3ON4,
2039 .priv_data_size = sizeof(MP3On4DecodeContext),
2040 .init = decode_init_mp3on4,
2041 .close = decode_close_mp3on4,
2042 .decode = decode_frame_mp3on4,
2043 .capabilities = CODEC_CAP_DR1,
2044 .flush = flush_mp3on4,
2045 .long_name = NULL_IF_CONFIG_SMALL("MP3onMP4"),