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)
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 switch_buffer(s, &pos, &end_pos, &end_pos2);
892 y = get_vlc2(&s->gb, vlc->table, 7, 3);
895 g->sb_hybrid[s_index ] =
896 g->sb_hybrid[s_index+1] = 0;
901 exponent= exponents[s_index];
903 av_dlog(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
904 i, g->region_size[i] - j, x, y, exponent);
909 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
911 x += get_bitsz(&s->gb, linbits);
912 v = l3_unscale(x, exponent);
913 if (get_bits1(&s->gb))
915 g->sb_hybrid[s_index] = v;
918 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
920 y += get_bitsz(&s->gb, linbits);
921 v = l3_unscale(y, exponent);
922 if (get_bits1(&s->gb))
924 g->sb_hybrid[s_index+1] = v;
931 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
933 x += get_bitsz(&s->gb, linbits);
934 v = l3_unscale(x, exponent);
935 if (get_bits1(&s->gb))
937 g->sb_hybrid[s_index+!!y] = v;
939 g->sb_hybrid[s_index + !y] = 0;
945 /* high frequencies */
946 vlc = &huff_quad_vlc[g->count1table_select];
948 while (s_index <= 572) {
950 pos = get_bits_count(&s->gb);
951 if (pos >= end_pos) {
952 if (pos > end_pos2 && last_pos) {
953 /* some encoders generate an incorrect size for this
954 part. We must go back into the data */
956 skip_bits_long(&s->gb, last_pos - pos);
957 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
958 if(s->err_recognition & AV_EF_BITSTREAM)
962 switch_buffer(s, &pos, &end_pos, &end_pos2);
968 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
969 av_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
970 g->sb_hybrid[s_index+0] =
971 g->sb_hybrid[s_index+1] =
972 g->sb_hybrid[s_index+2] =
973 g->sb_hybrid[s_index+3] = 0;
975 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
977 int pos = s_index + idxtab[code];
978 code ^= 8 >> idxtab[code];
979 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
983 /* skip extension bits */
984 bits_left = end_pos2 - get_bits_count(&s->gb);
985 if (bits_left < 0 && (s->err_recognition & AV_EF_BUFFER)) {
986 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
988 } else if (bits_left > 0 && (s->err_recognition & AV_EF_BUFFER)) {
989 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
992 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
993 skip_bits_long(&s->gb, bits_left);
995 i = get_bits_count(&s->gb);
996 switch_buffer(s, &i, &end_pos, &end_pos2);
1001 /* Reorder short blocks from bitstream order to interleaved order. It
1002 would be faster to do it in parsing, but the code would be far more
1004 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1007 INTFLOAT *ptr, *dst, *ptr1;
1010 if (g->block_type != 2)
1013 if (g->switch_point) {
1014 if (s->sample_rate_index != 8)
1015 ptr = g->sb_hybrid + 36;
1017 ptr = g->sb_hybrid + 72;
1022 for (i = g->short_start; i < 13; i++) {
1023 len = band_size_short[s->sample_rate_index][i];
1026 for (j = len; j > 0; j--) {
1027 *dst++ = ptr[0*len];
1028 *dst++ = ptr[1*len];
1029 *dst++ = ptr[2*len];
1033 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1037 #define ISQRT2 FIXR(0.70710678118654752440)
1039 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1042 int sf_max, sf, len, non_zero_found;
1043 INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1044 int non_zero_found_short[3];
1046 /* intensity stereo */
1047 if (s->mode_ext & MODE_EXT_I_STEREO) {
1052 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1056 tab0 = g0->sb_hybrid + 576;
1057 tab1 = g1->sb_hybrid + 576;
1059 non_zero_found_short[0] = 0;
1060 non_zero_found_short[1] = 0;
1061 non_zero_found_short[2] = 0;
1062 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1063 for (i = 12; i >= g1->short_start; i--) {
1064 /* for last band, use previous scale factor */
1067 len = band_size_short[s->sample_rate_index][i];
1068 for (l = 2; l >= 0; l--) {
1071 if (!non_zero_found_short[l]) {
1072 /* test if non zero band. if so, stop doing i-stereo */
1073 for (j = 0; j < len; j++) {
1075 non_zero_found_short[l] = 1;
1079 sf = g1->scale_factors[k + l];
1085 for (j = 0; j < len; j++) {
1087 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1088 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1092 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1093 /* lower part of the spectrum : do ms stereo
1095 for (j = 0; j < len; j++) {
1098 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1099 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1106 non_zero_found = non_zero_found_short[0] |
1107 non_zero_found_short[1] |
1108 non_zero_found_short[2];
1110 for (i = g1->long_end - 1;i >= 0;i--) {
1111 len = band_size_long[s->sample_rate_index][i];
1114 /* test if non zero band. if so, stop doing i-stereo */
1115 if (!non_zero_found) {
1116 for (j = 0; j < len; j++) {
1122 /* for last band, use previous scale factor */
1123 k = (i == 21) ? 20 : i;
1124 sf = g1->scale_factors[k];
1129 for (j = 0; j < len; j++) {
1131 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1132 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1136 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1137 /* lower part of the spectrum : do ms stereo
1139 for (j = 0; j < len; j++) {
1142 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1143 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1148 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1149 /* ms stereo ONLY */
1150 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1153 s-> dsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1155 tab0 = g0->sb_hybrid;
1156 tab1 = g1->sb_hybrid;
1157 for (i = 0; i < 576; i++) {
1160 tab0[i] = tmp0 + tmp1;
1161 tab1[i] = tmp0 - tmp1;
1168 #define AA(j) do { \
1169 float tmp0 = ptr[-1-j]; \
1170 float tmp1 = ptr[ j]; \
1171 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1172 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1175 #define AA(j) do { \
1176 int tmp0 = ptr[-1-j]; \
1177 int tmp1 = ptr[ j]; \
1178 int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1179 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1180 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1184 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1189 /* we antialias only "long" bands */
1190 if (g->block_type == 2) {
1191 if (!g->switch_point)
1193 /* XXX: check this for 8000Hz case */
1199 ptr = g->sb_hybrid + 18;
1200 for (i = n; i > 0; i--) {
1214 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1215 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1217 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1219 int i, j, mdct_long_end, sblimit;
1221 /* find last non zero block */
1222 ptr = g->sb_hybrid + 576;
1223 ptr1 = g->sb_hybrid + 2 * 18;
1224 while (ptr >= ptr1) {
1228 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1231 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1233 if (g->block_type == 2) {
1234 /* XXX: check for 8000 Hz */
1235 if (g->switch_point)
1240 mdct_long_end = sblimit;
1243 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1244 mdct_long_end, g->switch_point,
1247 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1248 ptr = g->sb_hybrid + 18 * mdct_long_end;
1250 for (j = mdct_long_end; j < sblimit; j++) {
1251 /* select frequency inversion */
1252 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1253 out_ptr = sb_samples + j;
1255 for (i = 0; i < 6; i++) {
1256 *out_ptr = buf[4*i];
1259 imdct12(out2, ptr + 0);
1260 for (i = 0; i < 6; i++) {
1261 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1262 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1265 imdct12(out2, ptr + 1);
1266 for (i = 0; i < 6; i++) {
1267 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1268 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1271 imdct12(out2, ptr + 2);
1272 for (i = 0; i < 6; i++) {
1273 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1274 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1275 buf[4*(i + 6*2)] = 0;
1278 buf += (j&3) != 3 ? 1 : (4*18-3);
1281 for (j = sblimit; j < SBLIMIT; j++) {
1283 out_ptr = sb_samples + j;
1284 for (i = 0; i < 18; i++) {
1285 *out_ptr = buf[4*i];
1289 buf += (j&3) != 3 ? 1 : (4*18-3);
1293 /* main layer3 decoding function */
1294 static int mp_decode_layer3(MPADecodeContext *s)
1296 int nb_granules, main_data_begin;
1297 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1299 int16_t exponents[576]; //FIXME try INTFLOAT
1301 /* read side info */
1303 main_data_begin = get_bits(&s->gb, 8);
1304 skip_bits(&s->gb, s->nb_channels);
1307 main_data_begin = get_bits(&s->gb, 9);
1308 if (s->nb_channels == 2)
1309 skip_bits(&s->gb, 3);
1311 skip_bits(&s->gb, 5);
1313 for (ch = 0; ch < s->nb_channels; ch++) {
1314 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1315 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1319 for (gr = 0; gr < nb_granules; gr++) {
1320 for (ch = 0; ch < s->nb_channels; ch++) {
1321 av_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1322 g = &s->granules[ch][gr];
1323 g->part2_3_length = get_bits(&s->gb, 12);
1324 g->big_values = get_bits(&s->gb, 9);
1325 if (g->big_values > 288) {
1326 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1327 return AVERROR_INVALIDDATA;
1330 g->global_gain = get_bits(&s->gb, 8);
1331 /* if MS stereo only is selected, we precompute the
1332 1/sqrt(2) renormalization factor */
1333 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1335 g->global_gain -= 2;
1337 g->scalefac_compress = get_bits(&s->gb, 9);
1339 g->scalefac_compress = get_bits(&s->gb, 4);
1340 blocksplit_flag = get_bits1(&s->gb);
1341 if (blocksplit_flag) {
1342 g->block_type = get_bits(&s->gb, 2);
1343 if (g->block_type == 0) {
1344 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1345 return AVERROR_INVALIDDATA;
1347 g->switch_point = get_bits1(&s->gb);
1348 for (i = 0; i < 2; i++)
1349 g->table_select[i] = get_bits(&s->gb, 5);
1350 for (i = 0; i < 3; i++)
1351 g->subblock_gain[i] = get_bits(&s->gb, 3);
1352 ff_init_short_region(s, g);
1354 int region_address1, region_address2;
1356 g->switch_point = 0;
1357 for (i = 0; i < 3; i++)
1358 g->table_select[i] = get_bits(&s->gb, 5);
1359 /* compute huffman coded region sizes */
1360 region_address1 = get_bits(&s->gb, 4);
1361 region_address2 = get_bits(&s->gb, 3);
1362 av_dlog(s->avctx, "region1=%d region2=%d\n",
1363 region_address1, region_address2);
1364 ff_init_long_region(s, g, region_address1, region_address2);
1366 ff_region_offset2size(g);
1367 ff_compute_band_indexes(s, g);
1371 g->preflag = get_bits1(&s->gb);
1372 g->scalefac_scale = get_bits1(&s->gb);
1373 g->count1table_select = get_bits1(&s->gb);
1374 av_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1375 g->block_type, g->switch_point);
1381 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1382 int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0,
1383 FFMAX(0, LAST_BUF_SIZE - s->last_buf_size));
1384 assert((get_bits_count(&s->gb) & 7) == 0);
1385 /* now we get bits from the main_data_begin offset */
1386 av_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1387 main_data_begin, s->last_buf_size);
1389 memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);
1391 init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
1392 #if !UNCHECKED_BITSTREAM_READER
1393 s->gb.size_in_bits_plus8 += extrasize * 8;
1395 s->last_buf_size <<= 3;
1396 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1397 for (ch = 0; ch < s->nb_channels; ch++) {
1398 g = &s->granules[ch][gr];
1399 s->last_buf_size += g->part2_3_length;
1400 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1401 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1404 skip = s->last_buf_size - 8 * main_data_begin;
1405 if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {
1406 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);
1408 s->in_gb.buffer = NULL;
1410 skip_bits_long(&s->gb, skip);
1416 for (; gr < nb_granules; gr++) {
1417 for (ch = 0; ch < s->nb_channels; ch++) {
1418 g = &s->granules[ch][gr];
1419 bits_pos = get_bits_count(&s->gb);
1423 int slen, slen1, slen2;
1425 /* MPEG1 scale factors */
1426 slen1 = slen_table[0][g->scalefac_compress];
1427 slen2 = slen_table[1][g->scalefac_compress];
1428 av_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1429 if (g->block_type == 2) {
1430 n = g->switch_point ? 17 : 18;
1433 for (i = 0; i < n; i++)
1434 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1436 for (i = 0; i < n; i++)
1437 g->scale_factors[j++] = 0;
1440 for (i = 0; i < 18; i++)
1441 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1442 for (i = 0; i < 3; i++)
1443 g->scale_factors[j++] = 0;
1445 for (i = 0; i < 21; i++)
1446 g->scale_factors[j++] = 0;
1449 sc = s->granules[ch][0].scale_factors;
1451 for (k = 0; k < 4; k++) {
1453 if ((g->scfsi & (0x8 >> k)) == 0) {
1454 slen = (k < 2) ? slen1 : slen2;
1456 for (i = 0; i < n; i++)
1457 g->scale_factors[j++] = get_bits(&s->gb, slen);
1459 for (i = 0; i < n; i++)
1460 g->scale_factors[j++] = 0;
1463 /* simply copy from last granule */
1464 for (i = 0; i < n; i++) {
1465 g->scale_factors[j] = sc[j];
1470 g->scale_factors[j++] = 0;
1473 int tindex, tindex2, slen[4], sl, sf;
1475 /* LSF scale factors */
1476 if (g->block_type == 2)
1477 tindex = g->switch_point ? 2 : 1;
1481 sf = g->scalefac_compress;
1482 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1483 /* intensity stereo case */
1486 lsf_sf_expand(slen, sf, 6, 6, 0);
1488 } else if (sf < 244) {
1489 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1492 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1498 lsf_sf_expand(slen, sf, 5, 4, 4);
1500 } else if (sf < 500) {
1501 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1504 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1511 for (k = 0; k < 4; k++) {
1512 n = lsf_nsf_table[tindex2][tindex][k];
1515 for (i = 0; i < n; i++)
1516 g->scale_factors[j++] = get_bits(&s->gb, sl);
1518 for (i = 0; i < n; i++)
1519 g->scale_factors[j++] = 0;
1522 /* XXX: should compute exact size */
1524 g->scale_factors[j] = 0;
1527 exponents_from_scale_factors(s, g, exponents);
1529 /* read Huffman coded residue */
1530 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1533 if (s->mode == MPA_JSTEREO)
1534 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1536 for (ch = 0; ch < s->nb_channels; ch++) {
1537 g = &s->granules[ch][gr];
1539 reorder_block(s, g);
1540 compute_antialias(s, g);
1541 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1544 if (get_bits_count(&s->gb) < 0)
1545 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1546 return nb_granules * 18;
1549 static int mp_decode_frame(MPADecodeContext *s, OUT_INT *samples,
1550 const uint8_t *buf, int buf_size)
1552 int i, nb_frames, ch, ret;
1553 OUT_INT *samples_ptr;
1555 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1557 /* skip error protection field */
1558 if (s->error_protection)
1559 skip_bits(&s->gb, 16);
1563 s->avctx->frame_size = 384;
1564 nb_frames = mp_decode_layer1(s);
1567 s->avctx->frame_size = 1152;
1568 nb_frames = mp_decode_layer2(s);
1571 s->avctx->frame_size = s->lsf ? 576 : 1152;
1573 nb_frames = mp_decode_layer3(s);
1579 if (s->in_gb.buffer) {
1580 align_get_bits(&s->gb);
1581 i = get_bits_left(&s->gb)>>3;
1582 if (i >= 0 && i <= BACKSTEP_SIZE) {
1583 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1586 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1588 s->in_gb.buffer = NULL;
1591 align_get_bits(&s->gb);
1592 assert((get_bits_count(&s->gb) & 7) == 0);
1593 i = get_bits_left(&s->gb) >> 3;
1595 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1597 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1598 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1600 assert(i <= buf_size - HEADER_SIZE && i >= 0);
1601 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1602 s->last_buf_size += i;
1605 /* get output buffer */
1607 s->frame.nb_samples = s->avctx->frame_size;
1608 if ((ret = s->avctx->get_buffer(s->avctx, &s->frame)) < 0) {
1609 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1612 samples = (OUT_INT *)s->frame.data[0];
1615 /* apply the synthesis filter */
1616 for (ch = 0; ch < s->nb_channels; ch++) {
1617 samples_ptr = samples + ch;
1618 for (i = 0; i < nb_frames; i++) {
1619 RENAME(ff_mpa_synth_filter)(
1621 s->synth_buf[ch], &(s->synth_buf_offset[ch]),
1622 RENAME(ff_mpa_synth_window), &s->dither_state,
1623 samples_ptr, s->nb_channels,
1624 s->sb_samples[ch][i]);
1625 samples_ptr += 32 * s->nb_channels;
1629 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1632 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1635 const uint8_t *buf = avpkt->data;
1636 int buf_size = avpkt->size;
1637 MPADecodeContext *s = avctx->priv_data;
1641 if (buf_size < HEADER_SIZE)
1642 return AVERROR_INVALIDDATA;
1644 header = AV_RB32(buf);
1645 if (ff_mpa_check_header(header) < 0) {
1646 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1647 return AVERROR_INVALIDDATA;
1650 if (avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
1651 /* free format: prepare to compute frame size */
1653 return AVERROR_INVALIDDATA;
1655 /* update codec info */
1656 avctx->channels = s->nb_channels;
1657 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1658 if (!avctx->bit_rate)
1659 avctx->bit_rate = s->bit_rate;
1661 if (s->frame_size <= 0 || s->frame_size > buf_size) {
1662 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1663 return AVERROR_INVALIDDATA;
1664 } else if (s->frame_size < buf_size) {
1665 buf_size= s->frame_size;
1668 ret = mp_decode_frame(s, NULL, buf, buf_size);
1671 *(AVFrame *)data = s->frame;
1672 avctx->sample_rate = s->sample_rate;
1673 //FIXME maybe move the other codec info stuff from above here too
1675 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1676 /* Only return an error if the bad frame makes up the whole packet or
1677 * the error is related to buffer management.
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 || ret != AVERROR_INVALIDDATA)
1689 static void mp_flush(MPADecodeContext *ctx)
1691 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1692 ctx->last_buf_size = 0;
1695 static void flush(AVCodecContext *avctx)
1697 mp_flush(avctx->priv_data);
1700 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1701 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1702 int *got_frame_ptr, AVPacket *avpkt)
1704 const uint8_t *buf = avpkt->data;
1705 int buf_size = avpkt->size;
1706 MPADecodeContext *s = avctx->priv_data;
1712 // Discard too short frames
1713 if (buf_size < HEADER_SIZE) {
1714 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1715 return AVERROR_INVALIDDATA;
1719 if (len > MPA_MAX_CODED_FRAME_SIZE)
1720 len = MPA_MAX_CODED_FRAME_SIZE;
1722 // Get header and restore sync word
1723 header = AV_RB32(buf) | 0xffe00000;
1725 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
1726 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1727 return AVERROR_INVALIDDATA;
1730 avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1731 /* update codec info */
1732 avctx->sample_rate = s->sample_rate;
1733 avctx->channels = s->nb_channels;
1734 if (!avctx->bit_rate)
1735 avctx->bit_rate = s->bit_rate;
1737 s->frame_size = len;
1739 ret = mp_decode_frame(s, NULL, buf, buf_size);
1741 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1746 *(AVFrame *)data = s->frame;
1750 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1752 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1755 * Context for MP3On4 decoder
1757 typedef struct MP3On4DecodeContext {
1759 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1760 int syncword; ///< syncword patch
1761 const uint8_t *coff; ///< channel offsets in output buffer
1762 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1763 OUT_INT *decoded_buf; ///< output buffer for decoded samples
1764 } MP3On4DecodeContext;
1766 #include "mpeg4audio.h"
1768 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1770 /* number of mp3 decoder instances */
1771 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1773 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1774 static const uint8_t chan_offset[8][5] = {
1779 { 2, 0, 3 }, // C FLR BS
1780 { 2, 0, 3 }, // C FLR BLRS
1781 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1782 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1785 /* mp3on4 channel layouts */
1786 static const int16_t chan_layout[8] = {
1789 AV_CH_LAYOUT_STEREO,
1790 AV_CH_LAYOUT_SURROUND,
1791 AV_CH_LAYOUT_4POINT0,
1792 AV_CH_LAYOUT_5POINT0,
1793 AV_CH_LAYOUT_5POINT1,
1794 AV_CH_LAYOUT_7POINT1
1797 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1799 MP3On4DecodeContext *s = avctx->priv_data;
1802 for (i = 0; i < s->frames; i++)
1803 av_free(s->mp3decctx[i]);
1805 av_freep(&s->decoded_buf);
1811 static int decode_init_mp3on4(AVCodecContext * avctx)
1813 MP3On4DecodeContext *s = avctx->priv_data;
1814 MPEG4AudioConfig cfg;
1817 if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
1818 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1819 return AVERROR_INVALIDDATA;
1822 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
1823 avctx->extradata_size * 8, 1);
1824 if (!cfg.chan_config || cfg.chan_config > 7) {
1825 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1826 return AVERROR_INVALIDDATA;
1828 s->frames = mp3Frames[cfg.chan_config];
1829 s->coff = chan_offset[cfg.chan_config];
1830 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1831 avctx->channel_layout = chan_layout[cfg.chan_config];
1833 if (cfg.sample_rate < 16000)
1834 s->syncword = 0xffe00000;
1836 s->syncword = 0xfff00000;
1838 /* Init the first mp3 decoder in standard way, so that all tables get builded
1839 * We replace avctx->priv_data with the context of the first decoder so that
1840 * decode_init() does not have to be changed.
1841 * Other decoders will be initialized here copying data from the first context
1843 // Allocate zeroed memory for the first decoder context
1844 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1845 if (!s->mp3decctx[0])
1847 // Put decoder context in place to make init_decode() happy
1848 avctx->priv_data = s->mp3decctx[0];
1850 s->frame = avctx->coded_frame;
1851 // Restore mp3on4 context pointer
1852 avctx->priv_data = s;
1853 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1855 /* Create a separate codec/context for each frame (first is already ok).
1856 * Each frame is 1 or 2 channels - up to 5 frames allowed
1858 for (i = 1; i < s->frames; i++) {
1859 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1860 if (!s->mp3decctx[i])
1862 s->mp3decctx[i]->adu_mode = 1;
1863 s->mp3decctx[i]->avctx = avctx;
1864 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1867 /* Allocate buffer for multi-channel output if needed */
1868 if (s->frames > 1) {
1869 s->decoded_buf = av_malloc(MPA_FRAME_SIZE * MPA_MAX_CHANNELS *
1870 sizeof(*s->decoded_buf));
1871 if (!s->decoded_buf)
1877 decode_close_mp3on4(avctx);
1878 return AVERROR(ENOMEM);
1882 static void flush_mp3on4(AVCodecContext *avctx)
1885 MP3On4DecodeContext *s = avctx->priv_data;
1887 for (i = 0; i < s->frames; i++)
1888 mp_flush(s->mp3decctx[i]);
1892 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1893 int *got_frame_ptr, AVPacket *avpkt)
1895 const uint8_t *buf = avpkt->data;
1896 int buf_size = avpkt->size;
1897 MP3On4DecodeContext *s = avctx->priv_data;
1898 MPADecodeContext *m;
1899 int fsize, len = buf_size, out_size = 0;
1901 OUT_INT *out_samples;
1902 OUT_INT *outptr, *bp;
1903 int fr, j, n, ch, ret;
1905 /* get output buffer */
1906 s->frame->nb_samples = MPA_FRAME_SIZE;
1907 if ((ret = avctx->get_buffer(avctx, s->frame)) < 0) {
1908 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1911 out_samples = (OUT_INT *)s->frame->data[0];
1913 // Discard too short frames
1914 if (buf_size < HEADER_SIZE)
1915 return AVERROR_INVALIDDATA;
1917 // If only one decoder interleave is not needed
1918 outptr = s->frames == 1 ? out_samples : s->decoded_buf;
1920 avctx->bit_rate = 0;
1923 for (fr = 0; fr < s->frames; fr++) {
1924 fsize = AV_RB16(buf) >> 4;
1925 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1926 m = s->mp3decctx[fr];
1929 if (fsize < HEADER_SIZE) {
1930 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1931 return AVERROR_INVALIDDATA;
1933 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1935 if (ff_mpa_check_header(header) < 0) // Bad header, discard block
1938 avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1940 if (ch + m->nb_channels > avctx->channels) {
1941 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1943 return AVERROR_INVALIDDATA;
1945 ch += m->nb_channels;
1947 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0)
1954 if (s->frames > 1) {
1955 n = m->avctx->frame_size*m->nb_channels;
1956 /* interleave output data */
1957 bp = out_samples + s->coff[fr];
1958 if (m->nb_channels == 1) {
1959 for (j = 0; j < n; j++) {
1960 *bp = s->decoded_buf[j];
1961 bp += avctx->channels;
1964 for (j = 0; j < n; j++) {
1965 bp[0] = s->decoded_buf[j++];
1966 bp[1] = s->decoded_buf[j];
1967 bp += avctx->channels;
1971 avctx->bit_rate += m->bit_rate;
1974 /* update codec info */
1975 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1977 s->frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1979 *(AVFrame *)data = *s->frame;
1983 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */
1986 #if CONFIG_MP1_DECODER
1987 AVCodec ff_mp1_decoder = {
1989 .type = AVMEDIA_TYPE_AUDIO,
1990 .id = AV_CODEC_ID_MP1,
1991 .priv_data_size = sizeof(MPADecodeContext),
1992 .init = decode_init,
1993 .decode = decode_frame,
1994 .capabilities = CODEC_CAP_DR1,
1996 .long_name = NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),
1999 #if CONFIG_MP2_DECODER
2000 AVCodec ff_mp2_decoder = {
2002 .type = AVMEDIA_TYPE_AUDIO,
2003 .id = AV_CODEC_ID_MP2,
2004 .priv_data_size = sizeof(MPADecodeContext),
2005 .init = decode_init,
2006 .decode = decode_frame,
2007 .capabilities = CODEC_CAP_DR1,
2009 .long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
2012 #if CONFIG_MP3_DECODER
2013 AVCodec ff_mp3_decoder = {
2015 .type = AVMEDIA_TYPE_AUDIO,
2016 .id = AV_CODEC_ID_MP3,
2017 .priv_data_size = sizeof(MPADecodeContext),
2018 .init = decode_init,
2019 .decode = decode_frame,
2020 .capabilities = CODEC_CAP_DR1,
2022 .long_name = NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),
2025 #if CONFIG_MP3ADU_DECODER
2026 AVCodec ff_mp3adu_decoder = {
2028 .type = AVMEDIA_TYPE_AUDIO,
2029 .id = AV_CODEC_ID_MP3ADU,
2030 .priv_data_size = sizeof(MPADecodeContext),
2031 .init = decode_init,
2032 .decode = decode_frame_adu,
2033 .capabilities = CODEC_CAP_DR1,
2035 .long_name = NULL_IF_CONFIG_SMALL("ADU (Application Data Unit) MP3 (MPEG audio layer 3)"),
2038 #if CONFIG_MP3ON4_DECODER
2039 AVCodec ff_mp3on4_decoder = {
2041 .type = AVMEDIA_TYPE_AUDIO,
2042 .id = AV_CODEC_ID_MP3ON4,
2043 .priv_data_size = sizeof(MP3On4DecodeContext),
2044 .init = decode_init_mp3on4,
2045 .close = decode_close_mp3on4,
2046 .decode = decode_frame_mp3on4,
2047 .capabilities = CODEC_CAP_DR1,
2048 .flush = flush_mp3on4,
2049 .long_name = NULL_IF_CONFIG_SMALL("MP3onMP4"),