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/channel_layout.h"
28 #include "libavutil/float_dsp.h"
33 #include "mpegaudiodsp.h"
38 * - test lsf / mpeg25 extensively.
41 #include "mpegaudio.h"
42 #include "mpegaudiodecheader.h"
44 #define BACKSTEP_SIZE 512
46 #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES
48 /* layer 3 "granule" */
49 typedef struct GranuleDef {
54 int scalefac_compress;
59 uint8_t scalefac_scale;
60 uint8_t count1table_select;
61 int region_size[3]; /* number of huffman codes in each region */
63 int short_start, long_end; /* long/short band indexes */
64 uint8_t scale_factors[40];
65 DECLARE_ALIGNED(16, INTFLOAT, sb_hybrid)[SBLIMIT * 18]; /* 576 samples */
68 typedef struct MPADecodeContext {
70 uint8_t last_buf[LAST_BUF_SIZE];
72 /* next header (used in free format parsing) */
73 uint32_t free_format_next_header;
76 DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
77 int synth_buf_offset[MPA_MAX_CHANNELS];
78 DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
79 INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
80 GranuleDef granules[2][2]; /* Used in Layer 3 */
81 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
84 AVCodecContext* avctx;
86 AVFloatDSPContext fdsp;
91 # define SHR(a,b) ((a)*(1.0f/(1<<(b))))
92 # define FIXR_OLD(a) ((int)((a) * FRAC_ONE + 0.5))
93 # define FIXR(x) ((float)(x))
94 # define FIXHR(x) ((float)(x))
95 # define MULH3(x, y, s) ((s)*(y)*(x))
96 # define MULLx(x, y, s) ((y)*(x))
97 # define RENAME(a) a ## _float
98 # define OUT_FMT AV_SAMPLE_FMT_FLT
99 # define OUT_FMT_P AV_SAMPLE_FMT_FLTP
101 # define SHR(a,b) ((a)>>(b))
102 /* WARNING: only correct for positive numbers */
103 # define FIXR_OLD(a) ((int)((a) * FRAC_ONE + 0.5))
104 # define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
105 # define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
106 # define MULH3(x, y, s) MULH((s)*(x), y)
107 # define MULLx(x, y, s) MULL(x,y,s)
108 # define RENAME(a) a ## _fixed
109 # define OUT_FMT AV_SAMPLE_FMT_S16
110 # define OUT_FMT_P AV_SAMPLE_FMT_S16P
115 #define HEADER_SIZE 4
117 #include "mpegaudiodata.h"
118 #include "mpegaudiodectab.h"
120 /* vlc structure for decoding layer 3 huffman tables */
121 static VLC huff_vlc[16];
122 static VLC_TYPE huff_vlc_tables[
123 0 + 128 + 128 + 128 + 130 + 128 + 154 + 166 +
124 142 + 204 + 190 + 170 + 542 + 460 + 662 + 414
126 static const int huff_vlc_tables_sizes[16] = {
127 0, 128, 128, 128, 130, 128, 154, 166,
128 142, 204, 190, 170, 542, 460, 662, 414
130 static VLC huff_quad_vlc[2];
131 static VLC_TYPE huff_quad_vlc_tables[128+16][2];
132 static const int huff_quad_vlc_tables_sizes[2] = { 128, 16 };
133 /* computed from band_size_long */
134 static uint16_t band_index_long[9][23];
135 #include "mpegaudio_tablegen.h"
136 /* intensity stereo coef table */
137 static INTFLOAT is_table[2][16];
138 static INTFLOAT is_table_lsf[2][2][16];
139 static INTFLOAT csa_table[8][4];
141 static int16_t division_tab3[1<<6 ];
142 static int16_t division_tab5[1<<8 ];
143 static int16_t division_tab9[1<<11];
145 static int16_t * const division_tabs[4] = {
146 division_tab3, division_tab5, NULL, division_tab9
149 /* lower 2 bits: modulo 3, higher bits: shift */
150 static uint16_t scale_factor_modshift[64];
151 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
152 static int32_t scale_factor_mult[15][3];
153 /* mult table for layer 2 group quantization */
155 #define SCALE_GEN(v) \
156 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
158 static const int32_t scale_factor_mult2[3][3] = {
159 SCALE_GEN(4.0 / 3.0), /* 3 steps */
160 SCALE_GEN(4.0 / 5.0), /* 5 steps */
161 SCALE_GEN(4.0 / 9.0), /* 9 steps */
165 * Convert region offsets to region sizes and truncate
166 * size to big_values.
168 static void ff_region_offset2size(GranuleDef *g)
171 g->region_size[2] = 576 / 2;
172 for (i = 0; i < 3; i++) {
173 k = FFMIN(g->region_size[i], g->big_values);
174 g->region_size[i] = k - j;
179 static void ff_init_short_region(MPADecodeContext *s, GranuleDef *g)
181 if (g->block_type == 2) {
182 if (s->sample_rate_index != 8)
183 g->region_size[0] = (36 / 2);
185 g->region_size[0] = (72 / 2);
187 if (s->sample_rate_index <= 2)
188 g->region_size[0] = (36 / 2);
189 else if (s->sample_rate_index != 8)
190 g->region_size[0] = (54 / 2);
192 g->region_size[0] = (108 / 2);
194 g->region_size[1] = (576 / 2);
197 static void ff_init_long_region(MPADecodeContext *s, GranuleDef *g, int ra1, int ra2)
200 g->region_size[0] = band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
201 /* should not overflow */
202 l = FFMIN(ra1 + ra2 + 2, 22);
203 g->region_size[1] = band_index_long[s->sample_rate_index][ l] >> 1;
206 static void ff_compute_band_indexes(MPADecodeContext *s, GranuleDef *g)
208 if (g->block_type == 2) {
209 if (g->switch_point) {
210 /* if switched mode, we handle the 36 first samples as
211 long blocks. For 8000Hz, we handle the 72 first
212 exponents as long blocks */
213 if (s->sample_rate_index <= 2)
229 /* layer 1 unscaling */
230 /* n = number of bits of the mantissa minus 1 */
231 static inline int l1_unscale(int n, int mant, int scale_factor)
236 shift = scale_factor_modshift[scale_factor];
239 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
241 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
242 return (int)((val + (1LL << (shift - 1))) >> shift);
245 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
249 shift = scale_factor_modshift[scale_factor];
253 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
254 /* NOTE: at this point, 0 <= shift <= 21 */
256 val = (val + (1 << (shift - 1))) >> shift;
260 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
261 static inline int l3_unscale(int value, int exponent)
266 e = table_4_3_exp [4 * value + (exponent & 3)];
267 m = table_4_3_value[4 * value + (exponent & 3)];
272 m = (m + (1 << (e - 1))) >> e;
277 static av_cold void decode_init_static(void)
282 /* scale factors table for layer 1/2 */
283 for (i = 0; i < 64; i++) {
285 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
288 scale_factor_modshift[i] = mod | (shift << 2);
291 /* scale factor multiply for layer 1 */
292 for (i = 0; i < 15; i++) {
295 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
296 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
297 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
298 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
299 av_dlog(NULL, "%d: norm=%x s=%x %x %x\n", i, norm,
300 scale_factor_mult[i][0],
301 scale_factor_mult[i][1],
302 scale_factor_mult[i][2]);
305 RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
307 /* huffman decode tables */
309 for (i = 1; i < 16; i++) {
310 const HuffTable *h = &mpa_huff_tables[i];
312 uint8_t tmp_bits [512] = { 0 };
313 uint16_t tmp_codes[512] = { 0 };
318 for (x = 0; x < xsize; x++) {
319 for (y = 0; y < xsize; y++) {
320 tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ];
321 tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
326 huff_vlc[i].table = huff_vlc_tables+offset;
327 huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
328 init_vlc(&huff_vlc[i], 7, 512,
329 tmp_bits, 1, 1, tmp_codes, 2, 2,
330 INIT_VLC_USE_NEW_STATIC);
331 offset += huff_vlc_tables_sizes[i];
333 assert(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
336 for (i = 0; i < 2; i++) {
337 huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
338 huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
339 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
340 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
341 INIT_VLC_USE_NEW_STATIC);
342 offset += huff_quad_vlc_tables_sizes[i];
344 assert(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
346 for (i = 0; i < 9; i++) {
348 for (j = 0; j < 22; j++) {
349 band_index_long[i][j] = k;
350 k += band_size_long[i][j];
352 band_index_long[i][22] = k;
355 /* compute n ^ (4/3) and store it in mantissa/exp format */
357 mpegaudio_tableinit();
359 for (i = 0; i < 4; i++) {
360 if (ff_mpa_quant_bits[i] < 0) {
361 for (j = 0; j < (1 << (-ff_mpa_quant_bits[i]+1)); j++) {
362 int val1, val2, val3, steps;
364 steps = ff_mpa_quant_steps[i];
369 division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
375 for (i = 0; i < 7; i++) {
379 f = tan((double)i * M_PI / 12.0);
380 v = FIXR(f / (1.0 + f));
385 is_table[1][6 - i] = v;
388 for (i = 7; i < 16; i++)
389 is_table[0][i] = is_table[1][i] = 0.0;
391 for (i = 0; i < 16; i++) {
395 for (j = 0; j < 2; j++) {
396 e = -(j + 1) * ((i + 1) >> 1);
397 f = pow(2.0, e / 4.0);
399 is_table_lsf[j][k ^ 1][i] = FIXR(f);
400 is_table_lsf[j][k ][i] = FIXR(1.0);
401 av_dlog(NULL, "is_table_lsf %d %d: %f %f\n",
402 i, j, (float) is_table_lsf[j][0][i],
403 (float) is_table_lsf[j][1][i]);
407 for (i = 0; i < 8; i++) {
410 cs = 1.0 / sqrt(1.0 + ci * ci);
413 csa_table[i][0] = FIXHR(cs/4);
414 csa_table[i][1] = FIXHR(ca/4);
415 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
416 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
418 csa_table[i][0] = cs;
419 csa_table[i][1] = ca;
420 csa_table[i][2] = ca + cs;
421 csa_table[i][3] = ca - cs;
426 static av_cold int decode_init(AVCodecContext * avctx)
428 static int initialized_tables = 0;
429 MPADecodeContext *s = avctx->priv_data;
431 if (!initialized_tables) {
432 decode_init_static();
433 initialized_tables = 1;
438 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
439 ff_mpadsp_init(&s->mpadsp);
441 if (avctx->request_sample_fmt == OUT_FMT &&
442 avctx->codec_id != AV_CODEC_ID_MP3ON4)
443 avctx->sample_fmt = OUT_FMT;
445 avctx->sample_fmt = OUT_FMT_P;
446 s->err_recognition = avctx->err_recognition;
448 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
451 avcodec_get_frame_defaults(&s->frame);
452 avctx->coded_frame = &s->frame;
457 #define C3 FIXHR(0.86602540378443864676/2)
458 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
459 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
460 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
462 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
464 static void imdct12(INTFLOAT *out, INTFLOAT *in)
466 INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
469 in1 = in[1*3] + in[0*3];
470 in2 = in[2*3] + in[1*3];
471 in3 = in[3*3] + in[2*3];
472 in4 = in[4*3] + in[3*3];
473 in5 = in[5*3] + in[4*3];
477 in2 = MULH3(in2, C3, 2);
478 in3 = MULH3(in3, C3, 4);
481 t2 = MULH3(in1 - in5, C4, 2);
491 in1 = MULH3(in5 + in3, C5, 1);
498 in5 = MULH3(in5 - in3, C6, 2);
505 /* return the number of decoded frames */
506 static int mp_decode_layer1(MPADecodeContext *s)
508 int bound, i, v, n, ch, j, mant;
509 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
510 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
512 if (s->mode == MPA_JSTEREO)
513 bound = (s->mode_ext + 1) * 4;
517 /* allocation bits */
518 for (i = 0; i < bound; i++) {
519 for (ch = 0; ch < s->nb_channels; ch++) {
520 allocation[ch][i] = get_bits(&s->gb, 4);
523 for (i = bound; i < SBLIMIT; i++)
524 allocation[0][i] = get_bits(&s->gb, 4);
527 for (i = 0; i < bound; i++) {
528 for (ch = 0; ch < s->nb_channels; ch++) {
529 if (allocation[ch][i])
530 scale_factors[ch][i] = get_bits(&s->gb, 6);
533 for (i = bound; i < SBLIMIT; i++) {
534 if (allocation[0][i]) {
535 scale_factors[0][i] = get_bits(&s->gb, 6);
536 scale_factors[1][i] = get_bits(&s->gb, 6);
540 /* compute samples */
541 for (j = 0; j < 12; j++) {
542 for (i = 0; i < bound; i++) {
543 for (ch = 0; ch < s->nb_channels; ch++) {
544 n = allocation[ch][i];
546 mant = get_bits(&s->gb, n + 1);
547 v = l1_unscale(n, mant, scale_factors[ch][i]);
551 s->sb_samples[ch][j][i] = v;
554 for (i = bound; i < SBLIMIT; i++) {
555 n = allocation[0][i];
557 mant = get_bits(&s->gb, n + 1);
558 v = l1_unscale(n, mant, scale_factors[0][i]);
559 s->sb_samples[0][j][i] = v;
560 v = l1_unscale(n, mant, scale_factors[1][i]);
561 s->sb_samples[1][j][i] = v;
563 s->sb_samples[0][j][i] = 0;
564 s->sb_samples[1][j][i] = 0;
571 static int mp_decode_layer2(MPADecodeContext *s)
573 int sblimit; /* number of used subbands */
574 const unsigned char *alloc_table;
575 int table, bit_alloc_bits, i, j, ch, bound, v;
576 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
577 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
578 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
579 int scale, qindex, bits, steps, k, l, m, b;
581 /* select decoding table */
582 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
583 s->sample_rate, s->lsf);
584 sblimit = ff_mpa_sblimit_table[table];
585 alloc_table = ff_mpa_alloc_tables[table];
587 if (s->mode == MPA_JSTEREO)
588 bound = (s->mode_ext + 1) * 4;
592 av_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
598 /* parse bit allocation */
600 for (i = 0; i < bound; i++) {
601 bit_alloc_bits = alloc_table[j];
602 for (ch = 0; ch < s->nb_channels; ch++)
603 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
604 j += 1 << bit_alloc_bits;
606 for (i = bound; i < sblimit; i++) {
607 bit_alloc_bits = alloc_table[j];
608 v = get_bits(&s->gb, bit_alloc_bits);
611 j += 1 << bit_alloc_bits;
615 for (i = 0; i < sblimit; i++) {
616 for (ch = 0; ch < s->nb_channels; ch++) {
617 if (bit_alloc[ch][i])
618 scale_code[ch][i] = get_bits(&s->gb, 2);
623 for (i = 0; i < sblimit; i++) {
624 for (ch = 0; ch < s->nb_channels; ch++) {
625 if (bit_alloc[ch][i]) {
626 sf = scale_factors[ch][i];
627 switch (scale_code[ch][i]) {
630 sf[0] = get_bits(&s->gb, 6);
631 sf[1] = get_bits(&s->gb, 6);
632 sf[2] = get_bits(&s->gb, 6);
635 sf[0] = get_bits(&s->gb, 6);
640 sf[0] = get_bits(&s->gb, 6);
641 sf[2] = get_bits(&s->gb, 6);
645 sf[0] = get_bits(&s->gb, 6);
646 sf[2] = get_bits(&s->gb, 6);
655 for (k = 0; k < 3; k++) {
656 for (l = 0; l < 12; l += 3) {
658 for (i = 0; i < bound; i++) {
659 bit_alloc_bits = alloc_table[j];
660 for (ch = 0; ch < s->nb_channels; ch++) {
661 b = bit_alloc[ch][i];
663 scale = scale_factors[ch][i][k];
664 qindex = alloc_table[j+b];
665 bits = ff_mpa_quant_bits[qindex];
668 /* 3 values at the same time */
669 v = get_bits(&s->gb, -bits);
670 v2 = division_tabs[qindex][v];
671 steps = ff_mpa_quant_steps[qindex];
673 s->sb_samples[ch][k * 12 + l + 0][i] =
674 l2_unscale_group(steps, v2 & 15, scale);
675 s->sb_samples[ch][k * 12 + l + 1][i] =
676 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
677 s->sb_samples[ch][k * 12 + l + 2][i] =
678 l2_unscale_group(steps, v2 >> 8 , scale);
680 for (m = 0; m < 3; m++) {
681 v = get_bits(&s->gb, bits);
682 v = l1_unscale(bits - 1, v, scale);
683 s->sb_samples[ch][k * 12 + l + m][i] = v;
687 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
688 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
689 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
692 /* next subband in alloc table */
693 j += 1 << bit_alloc_bits;
695 /* XXX: find a way to avoid this duplication of code */
696 for (i = bound; i < sblimit; i++) {
697 bit_alloc_bits = alloc_table[j];
700 int mant, scale0, scale1;
701 scale0 = scale_factors[0][i][k];
702 scale1 = scale_factors[1][i][k];
703 qindex = alloc_table[j+b];
704 bits = ff_mpa_quant_bits[qindex];
706 /* 3 values at the same time */
707 v = get_bits(&s->gb, -bits);
708 steps = ff_mpa_quant_steps[qindex];
711 s->sb_samples[0][k * 12 + l + 0][i] =
712 l2_unscale_group(steps, mant, scale0);
713 s->sb_samples[1][k * 12 + l + 0][i] =
714 l2_unscale_group(steps, mant, scale1);
717 s->sb_samples[0][k * 12 + l + 1][i] =
718 l2_unscale_group(steps, mant, scale0);
719 s->sb_samples[1][k * 12 + l + 1][i] =
720 l2_unscale_group(steps, mant, scale1);
721 s->sb_samples[0][k * 12 + l + 2][i] =
722 l2_unscale_group(steps, v, scale0);
723 s->sb_samples[1][k * 12 + l + 2][i] =
724 l2_unscale_group(steps, v, scale1);
726 for (m = 0; m < 3; m++) {
727 mant = get_bits(&s->gb, bits);
728 s->sb_samples[0][k * 12 + l + m][i] =
729 l1_unscale(bits - 1, mant, scale0);
730 s->sb_samples[1][k * 12 + l + m][i] =
731 l1_unscale(bits - 1, mant, scale1);
735 s->sb_samples[0][k * 12 + l + 0][i] = 0;
736 s->sb_samples[0][k * 12 + l + 1][i] = 0;
737 s->sb_samples[0][k * 12 + l + 2][i] = 0;
738 s->sb_samples[1][k * 12 + l + 0][i] = 0;
739 s->sb_samples[1][k * 12 + l + 1][i] = 0;
740 s->sb_samples[1][k * 12 + l + 2][i] = 0;
742 /* next subband in alloc table */
743 j += 1 << bit_alloc_bits;
745 /* fill remaining samples to zero */
746 for (i = sblimit; i < SBLIMIT; i++) {
747 for (ch = 0; ch < s->nb_channels; ch++) {
748 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
749 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
750 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
758 #define SPLIT(dst,sf,n) \
760 int m = (sf * 171) >> 9; \
763 } else if (n == 4) { \
766 } else if (n == 5) { \
767 int m = (sf * 205) >> 10; \
770 } else if (n == 6) { \
771 int m = (sf * 171) >> 10; \
778 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
781 SPLIT(slen[3], sf, n3)
782 SPLIT(slen[2], sf, n2)
783 SPLIT(slen[1], sf, n1)
787 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
790 const uint8_t *bstab, *pretab;
791 int len, i, j, k, l, v0, shift, gain, gains[3];
795 gain = g->global_gain - 210;
796 shift = g->scalefac_scale + 1;
798 bstab = band_size_long[s->sample_rate_index];
799 pretab = mpa_pretab[g->preflag];
800 for (i = 0; i < g->long_end; i++) {
801 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
803 for (j = len; j > 0; j--)
807 if (g->short_start < 13) {
808 bstab = band_size_short[s->sample_rate_index];
809 gains[0] = gain - (g->subblock_gain[0] << 3);
810 gains[1] = gain - (g->subblock_gain[1] << 3);
811 gains[2] = gain - (g->subblock_gain[2] << 3);
813 for (i = g->short_start; i < 13; i++) {
815 for (l = 0; l < 3; l++) {
816 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
817 for (j = len; j > 0; j--)
824 /* handle n = 0 too */
825 static inline int get_bitsz(GetBitContext *s, int n)
827 return n ? get_bits(s, n) : 0;
831 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
834 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits) {
836 s->in_gb.buffer = NULL;
837 assert((get_bits_count(&s->gb) & 7) == 0);
838 skip_bits_long(&s->gb, *pos - *end_pos);
840 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
841 *pos = get_bits_count(&s->gb);
845 /* Following is a optimized code for
847 if(get_bits1(&s->gb))
852 #define READ_FLIP_SIGN(dst,src) \
853 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
856 #define READ_FLIP_SIGN(dst,src) \
857 v = -get_bits1(&s->gb); \
858 *(dst) = (*(src) ^ v) - v;
861 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
862 int16_t *exponents, int end_pos2)
866 int last_pos, bits_left;
868 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits);
870 /* low frequencies (called big values) */
872 for (i = 0; i < 3; i++) {
873 int j, k, l, linbits;
874 j = g->region_size[i];
877 /* select vlc table */
878 k = g->table_select[i];
879 l = mpa_huff_data[k][0];
880 linbits = mpa_huff_data[k][1];
884 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
889 /* read huffcode and compute each couple */
893 int pos = get_bits_count(&s->gb);
896 switch_buffer(s, &pos, &end_pos, &end_pos2);
900 y = get_vlc2(&s->gb, vlc->table, 7, 3);
903 g->sb_hybrid[s_index ] =
904 g->sb_hybrid[s_index+1] = 0;
909 exponent= exponents[s_index];
911 av_dlog(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
912 i, g->region_size[i] - j, x, y, exponent);
917 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
919 x += get_bitsz(&s->gb, linbits);
920 v = l3_unscale(x, exponent);
921 if (get_bits1(&s->gb))
923 g->sb_hybrid[s_index] = v;
926 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
928 y += get_bitsz(&s->gb, linbits);
929 v = l3_unscale(y, exponent);
930 if (get_bits1(&s->gb))
932 g->sb_hybrid[s_index+1] = v;
939 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
941 x += get_bitsz(&s->gb, linbits);
942 v = l3_unscale(x, exponent);
943 if (get_bits1(&s->gb))
945 g->sb_hybrid[s_index+!!y] = v;
947 g->sb_hybrid[s_index + !y] = 0;
953 /* high frequencies */
954 vlc = &huff_quad_vlc[g->count1table_select];
956 while (s_index <= 572) {
958 pos = get_bits_count(&s->gb);
959 if (pos >= end_pos) {
960 if (pos > end_pos2 && last_pos) {
961 /* some encoders generate an incorrect size for this
962 part. We must go back into the data */
964 skip_bits_long(&s->gb, last_pos - pos);
965 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
966 if(s->err_recognition & AV_EF_BITSTREAM)
970 switch_buffer(s, &pos, &end_pos, &end_pos2);
976 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
977 av_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
978 g->sb_hybrid[s_index+0] =
979 g->sb_hybrid[s_index+1] =
980 g->sb_hybrid[s_index+2] =
981 g->sb_hybrid[s_index+3] = 0;
983 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
985 int pos = s_index + idxtab[code];
986 code ^= 8 >> idxtab[code];
987 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
991 /* skip extension bits */
992 bits_left = end_pos2 - get_bits_count(&s->gb);
993 if (bits_left < 0 && (s->err_recognition & AV_EF_BUFFER)) {
994 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
996 } else if (bits_left > 0 && (s->err_recognition & AV_EF_BUFFER)) {
997 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
1000 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
1001 skip_bits_long(&s->gb, bits_left);
1003 i = get_bits_count(&s->gb);
1004 switch_buffer(s, &i, &end_pos, &end_pos2);
1009 /* Reorder short blocks from bitstream order to interleaved order. It
1010 would be faster to do it in parsing, but the code would be far more
1012 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1015 INTFLOAT *ptr, *dst, *ptr1;
1018 if (g->block_type != 2)
1021 if (g->switch_point) {
1022 if (s->sample_rate_index != 8)
1023 ptr = g->sb_hybrid + 36;
1025 ptr = g->sb_hybrid + 72;
1030 for (i = g->short_start; i < 13; i++) {
1031 len = band_size_short[s->sample_rate_index][i];
1034 for (j = len; j > 0; j--) {
1035 *dst++ = ptr[0*len];
1036 *dst++ = ptr[1*len];
1037 *dst++ = ptr[2*len];
1041 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1045 #define ISQRT2 FIXR(0.70710678118654752440)
1047 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1050 int sf_max, sf, len, non_zero_found;
1051 INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1052 int non_zero_found_short[3];
1054 /* intensity stereo */
1055 if (s->mode_ext & MODE_EXT_I_STEREO) {
1060 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1064 tab0 = g0->sb_hybrid + 576;
1065 tab1 = g1->sb_hybrid + 576;
1067 non_zero_found_short[0] = 0;
1068 non_zero_found_short[1] = 0;
1069 non_zero_found_short[2] = 0;
1070 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1071 for (i = 12; i >= g1->short_start; i--) {
1072 /* for last band, use previous scale factor */
1075 len = band_size_short[s->sample_rate_index][i];
1076 for (l = 2; l >= 0; l--) {
1079 if (!non_zero_found_short[l]) {
1080 /* test if non zero band. if so, stop doing i-stereo */
1081 for (j = 0; j < len; j++) {
1083 non_zero_found_short[l] = 1;
1087 sf = g1->scale_factors[k + l];
1093 for (j = 0; j < len; j++) {
1095 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1096 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1100 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1101 /* lower part of the spectrum : do ms stereo
1103 for (j = 0; j < len; j++) {
1106 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1107 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1114 non_zero_found = non_zero_found_short[0] |
1115 non_zero_found_short[1] |
1116 non_zero_found_short[2];
1118 for (i = g1->long_end - 1;i >= 0;i--) {
1119 len = band_size_long[s->sample_rate_index][i];
1122 /* test if non zero band. if so, stop doing i-stereo */
1123 if (!non_zero_found) {
1124 for (j = 0; j < len; j++) {
1130 /* for last band, use previous scale factor */
1131 k = (i == 21) ? 20 : i;
1132 sf = g1->scale_factors[k];
1137 for (j = 0; j < len; j++) {
1139 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1140 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1144 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1145 /* lower part of the spectrum : do ms stereo
1147 for (j = 0; j < len; j++) {
1150 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1151 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1156 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1157 /* ms stereo ONLY */
1158 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1161 s->fdsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1163 tab0 = g0->sb_hybrid;
1164 tab1 = g1->sb_hybrid;
1165 for (i = 0; i < 576; i++) {
1168 tab0[i] = tmp0 + tmp1;
1169 tab1[i] = tmp0 - tmp1;
1176 #define AA(j) do { \
1177 float tmp0 = ptr[-1-j]; \
1178 float tmp1 = ptr[ j]; \
1179 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1180 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1183 #define AA(j) do { \
1184 int tmp0 = ptr[-1-j]; \
1185 int tmp1 = ptr[ j]; \
1186 int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1187 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1188 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1192 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1197 /* we antialias only "long" bands */
1198 if (g->block_type == 2) {
1199 if (!g->switch_point)
1201 /* XXX: check this for 8000Hz case */
1207 ptr = g->sb_hybrid + 18;
1208 for (i = n; i > 0; i--) {
1222 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1223 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1225 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1227 int i, j, mdct_long_end, sblimit;
1229 /* find last non zero block */
1230 ptr = g->sb_hybrid + 576;
1231 ptr1 = g->sb_hybrid + 2 * 18;
1232 while (ptr >= ptr1) {
1236 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1239 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1241 if (g->block_type == 2) {
1242 /* XXX: check for 8000 Hz */
1243 if (g->switch_point)
1248 mdct_long_end = sblimit;
1251 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1252 mdct_long_end, g->switch_point,
1255 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1256 ptr = g->sb_hybrid + 18 * mdct_long_end;
1258 for (j = mdct_long_end; j < sblimit; j++) {
1259 /* select frequency inversion */
1260 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1261 out_ptr = sb_samples + j;
1263 for (i = 0; i < 6; i++) {
1264 *out_ptr = buf[4*i];
1267 imdct12(out2, ptr + 0);
1268 for (i = 0; i < 6; i++) {
1269 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1270 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1273 imdct12(out2, ptr + 1);
1274 for (i = 0; i < 6; i++) {
1275 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1276 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1279 imdct12(out2, ptr + 2);
1280 for (i = 0; i < 6; i++) {
1281 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1282 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1283 buf[4*(i + 6*2)] = 0;
1286 buf += (j&3) != 3 ? 1 : (4*18-3);
1289 for (j = sblimit; j < SBLIMIT; j++) {
1291 out_ptr = sb_samples + j;
1292 for (i = 0; i < 18; i++) {
1293 *out_ptr = buf[4*i];
1297 buf += (j&3) != 3 ? 1 : (4*18-3);
1301 /* main layer3 decoding function */
1302 static int mp_decode_layer3(MPADecodeContext *s)
1304 int nb_granules, main_data_begin;
1305 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1307 int16_t exponents[576]; //FIXME try INTFLOAT
1309 /* read side info */
1311 main_data_begin = get_bits(&s->gb, 8);
1312 skip_bits(&s->gb, s->nb_channels);
1315 main_data_begin = get_bits(&s->gb, 9);
1316 if (s->nb_channels == 2)
1317 skip_bits(&s->gb, 3);
1319 skip_bits(&s->gb, 5);
1321 for (ch = 0; ch < s->nb_channels; ch++) {
1322 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1323 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1327 for (gr = 0; gr < nb_granules; gr++) {
1328 for (ch = 0; ch < s->nb_channels; ch++) {
1329 av_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1330 g = &s->granules[ch][gr];
1331 g->part2_3_length = get_bits(&s->gb, 12);
1332 g->big_values = get_bits(&s->gb, 9);
1333 if (g->big_values > 288) {
1334 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1335 return AVERROR_INVALIDDATA;
1338 g->global_gain = get_bits(&s->gb, 8);
1339 /* if MS stereo only is selected, we precompute the
1340 1/sqrt(2) renormalization factor */
1341 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1343 g->global_gain -= 2;
1345 g->scalefac_compress = get_bits(&s->gb, 9);
1347 g->scalefac_compress = get_bits(&s->gb, 4);
1348 blocksplit_flag = get_bits1(&s->gb);
1349 if (blocksplit_flag) {
1350 g->block_type = get_bits(&s->gb, 2);
1351 if (g->block_type == 0) {
1352 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1353 return AVERROR_INVALIDDATA;
1355 g->switch_point = get_bits1(&s->gb);
1356 for (i = 0; i < 2; i++)
1357 g->table_select[i] = get_bits(&s->gb, 5);
1358 for (i = 0; i < 3; i++)
1359 g->subblock_gain[i] = get_bits(&s->gb, 3);
1360 ff_init_short_region(s, g);
1362 int region_address1, region_address2;
1364 g->switch_point = 0;
1365 for (i = 0; i < 3; i++)
1366 g->table_select[i] = get_bits(&s->gb, 5);
1367 /* compute huffman coded region sizes */
1368 region_address1 = get_bits(&s->gb, 4);
1369 region_address2 = get_bits(&s->gb, 3);
1370 av_dlog(s->avctx, "region1=%d region2=%d\n",
1371 region_address1, region_address2);
1372 ff_init_long_region(s, g, region_address1, region_address2);
1374 ff_region_offset2size(g);
1375 ff_compute_band_indexes(s, g);
1379 g->preflag = get_bits1(&s->gb);
1380 g->scalefac_scale = get_bits1(&s->gb);
1381 g->count1table_select = get_bits1(&s->gb);
1382 av_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1383 g->block_type, g->switch_point);
1389 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1390 int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0,
1391 FFMAX(0, LAST_BUF_SIZE - s->last_buf_size));
1392 assert((get_bits_count(&s->gb) & 7) == 0);
1393 /* now we get bits from the main_data_begin offset */
1394 av_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1395 main_data_begin, s->last_buf_size);
1397 memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);
1399 init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
1400 #if !UNCHECKED_BITSTREAM_READER
1401 s->gb.size_in_bits_plus8 += extrasize * 8;
1403 s->last_buf_size <<= 3;
1404 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1405 for (ch = 0; ch < s->nb_channels; ch++) {
1406 g = &s->granules[ch][gr];
1407 s->last_buf_size += g->part2_3_length;
1408 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1409 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1412 skip = s->last_buf_size - 8 * main_data_begin;
1413 if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {
1414 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);
1416 s->in_gb.buffer = NULL;
1418 skip_bits_long(&s->gb, skip);
1424 for (; gr < nb_granules; gr++) {
1425 for (ch = 0; ch < s->nb_channels; ch++) {
1426 g = &s->granules[ch][gr];
1427 bits_pos = get_bits_count(&s->gb);
1431 int slen, slen1, slen2;
1433 /* MPEG1 scale factors */
1434 slen1 = slen_table[0][g->scalefac_compress];
1435 slen2 = slen_table[1][g->scalefac_compress];
1436 av_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1437 if (g->block_type == 2) {
1438 n = g->switch_point ? 17 : 18;
1441 for (i = 0; i < n; i++)
1442 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1444 for (i = 0; i < n; i++)
1445 g->scale_factors[j++] = 0;
1448 for (i = 0; i < 18; i++)
1449 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1450 for (i = 0; i < 3; i++)
1451 g->scale_factors[j++] = 0;
1453 for (i = 0; i < 21; i++)
1454 g->scale_factors[j++] = 0;
1457 sc = s->granules[ch][0].scale_factors;
1459 for (k = 0; k < 4; k++) {
1461 if ((g->scfsi & (0x8 >> k)) == 0) {
1462 slen = (k < 2) ? slen1 : slen2;
1464 for (i = 0; i < n; i++)
1465 g->scale_factors[j++] = get_bits(&s->gb, slen);
1467 for (i = 0; i < n; i++)
1468 g->scale_factors[j++] = 0;
1471 /* simply copy from last granule */
1472 for (i = 0; i < n; i++) {
1473 g->scale_factors[j] = sc[j];
1478 g->scale_factors[j++] = 0;
1481 int tindex, tindex2, slen[4], sl, sf;
1483 /* LSF scale factors */
1484 if (g->block_type == 2)
1485 tindex = g->switch_point ? 2 : 1;
1489 sf = g->scalefac_compress;
1490 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1491 /* intensity stereo case */
1494 lsf_sf_expand(slen, sf, 6, 6, 0);
1496 } else if (sf < 244) {
1497 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1500 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1506 lsf_sf_expand(slen, sf, 5, 4, 4);
1508 } else if (sf < 500) {
1509 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1512 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1519 for (k = 0; k < 4; k++) {
1520 n = lsf_nsf_table[tindex2][tindex][k];
1523 for (i = 0; i < n; i++)
1524 g->scale_factors[j++] = get_bits(&s->gb, sl);
1526 for (i = 0; i < n; i++)
1527 g->scale_factors[j++] = 0;
1530 /* XXX: should compute exact size */
1532 g->scale_factors[j] = 0;
1535 exponents_from_scale_factors(s, g, exponents);
1537 /* read Huffman coded residue */
1538 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1541 if (s->mode == MPA_JSTEREO)
1542 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1544 for (ch = 0; ch < s->nb_channels; ch++) {
1545 g = &s->granules[ch][gr];
1547 reorder_block(s, g);
1548 compute_antialias(s, g);
1549 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1552 if (get_bits_count(&s->gb) < 0)
1553 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1554 return nb_granules * 18;
1557 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1558 const uint8_t *buf, int buf_size)
1560 int i, nb_frames, ch, ret;
1561 OUT_INT *samples_ptr;
1563 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1565 /* skip error protection field */
1566 if (s->error_protection)
1567 skip_bits(&s->gb, 16);
1571 s->avctx->frame_size = 384;
1572 nb_frames = mp_decode_layer1(s);
1575 s->avctx->frame_size = 1152;
1576 nb_frames = mp_decode_layer2(s);
1579 s->avctx->frame_size = s->lsf ? 576 : 1152;
1581 nb_frames = mp_decode_layer3(s);
1587 if (s->in_gb.buffer) {
1588 align_get_bits(&s->gb);
1589 i = get_bits_left(&s->gb)>>3;
1590 if (i >= 0 && i <= BACKSTEP_SIZE) {
1591 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1594 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1596 s->in_gb.buffer = NULL;
1599 align_get_bits(&s->gb);
1600 assert((get_bits_count(&s->gb) & 7) == 0);
1601 i = get_bits_left(&s->gb) >> 3;
1603 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1605 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1606 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1608 assert(i <= buf_size - HEADER_SIZE && i >= 0);
1609 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1610 s->last_buf_size += i;
1613 /* get output buffer */
1615 s->frame.nb_samples = s->avctx->frame_size;
1616 if ((ret = ff_get_buffer(s->avctx, &s->frame)) < 0) {
1617 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1620 samples = (OUT_INT **)s->frame.extended_data;
1623 /* apply the synthesis filter */
1624 for (ch = 0; ch < s->nb_channels; ch++) {
1626 if (s->avctx->sample_fmt == OUT_FMT_P) {
1627 samples_ptr = samples[ch];
1630 samples_ptr = samples[0] + ch;
1631 sample_stride = s->nb_channels;
1633 for (i = 0; i < nb_frames; i++) {
1634 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1635 &(s->synth_buf_offset[ch]),
1636 RENAME(ff_mpa_synth_window),
1637 &s->dither_state, samples_ptr,
1638 sample_stride, s->sb_samples[ch][i]);
1639 samples_ptr += 32 * sample_stride;
1643 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1646 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1649 const uint8_t *buf = avpkt->data;
1650 int buf_size = avpkt->size;
1651 MPADecodeContext *s = avctx->priv_data;
1655 if (buf_size < HEADER_SIZE)
1656 return AVERROR_INVALIDDATA;
1658 header = AV_RB32(buf);
1659 if (ff_mpa_check_header(header) < 0) {
1660 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1661 return AVERROR_INVALIDDATA;
1664 if (avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
1665 /* free format: prepare to compute frame size */
1667 return AVERROR_INVALIDDATA;
1669 /* update codec info */
1670 avctx->channels = s->nb_channels;
1671 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1672 if (!avctx->bit_rate)
1673 avctx->bit_rate = s->bit_rate;
1675 if (s->frame_size <= 0 || s->frame_size > buf_size) {
1676 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1677 return AVERROR_INVALIDDATA;
1678 } else if (s->frame_size < buf_size) {
1679 buf_size= s->frame_size;
1682 ret = mp_decode_frame(s, NULL, buf, buf_size);
1685 *(AVFrame *)data = s->frame;
1686 avctx->sample_rate = s->sample_rate;
1687 //FIXME maybe move the other codec info stuff from above here too
1689 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1690 /* Only return an error if the bad frame makes up the whole packet or
1691 * the error is related to buffer management.
1692 * If there is more data in the packet, just consume the bad frame
1693 * instead of returning an error, which would discard the whole
1696 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1703 static void mp_flush(MPADecodeContext *ctx)
1705 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1706 ctx->last_buf_size = 0;
1709 static void flush(AVCodecContext *avctx)
1711 mp_flush(avctx->priv_data);
1714 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1715 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1716 int *got_frame_ptr, AVPacket *avpkt)
1718 const uint8_t *buf = avpkt->data;
1719 int buf_size = avpkt->size;
1720 MPADecodeContext *s = avctx->priv_data;
1726 // Discard too short frames
1727 if (buf_size < HEADER_SIZE) {
1728 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1729 return AVERROR_INVALIDDATA;
1733 if (len > MPA_MAX_CODED_FRAME_SIZE)
1734 len = MPA_MAX_CODED_FRAME_SIZE;
1736 // Get header and restore sync word
1737 header = AV_RB32(buf) | 0xffe00000;
1739 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
1740 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1741 return AVERROR_INVALIDDATA;
1744 avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1745 /* update codec info */
1746 avctx->sample_rate = s->sample_rate;
1747 avctx->channels = s->nb_channels;
1748 if (!avctx->bit_rate)
1749 avctx->bit_rate = s->bit_rate;
1751 s->frame_size = len;
1753 ret = mp_decode_frame(s, NULL, buf, buf_size);
1755 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1760 *(AVFrame *)data = s->frame;
1764 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1766 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1769 * Context for MP3On4 decoder
1771 typedef struct MP3On4DecodeContext {
1773 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1774 int syncword; ///< syncword patch
1775 const uint8_t *coff; ///< channel offsets in output buffer
1776 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1777 } MP3On4DecodeContext;
1779 #include "mpeg4audio.h"
1781 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1783 /* number of mp3 decoder instances */
1784 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1786 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1787 static const uint8_t chan_offset[8][5] = {
1792 { 2, 0, 3 }, // C FLR BS
1793 { 2, 0, 3 }, // C FLR BLRS
1794 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1795 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1798 /* mp3on4 channel layouts */
1799 static const int16_t chan_layout[8] = {
1802 AV_CH_LAYOUT_STEREO,
1803 AV_CH_LAYOUT_SURROUND,
1804 AV_CH_LAYOUT_4POINT0,
1805 AV_CH_LAYOUT_5POINT0,
1806 AV_CH_LAYOUT_5POINT1,
1807 AV_CH_LAYOUT_7POINT1
1810 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1812 MP3On4DecodeContext *s = avctx->priv_data;
1815 for (i = 0; i < s->frames; i++)
1816 av_free(s->mp3decctx[i]);
1822 static int decode_init_mp3on4(AVCodecContext * avctx)
1824 MP3On4DecodeContext *s = avctx->priv_data;
1825 MPEG4AudioConfig cfg;
1828 if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
1829 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1830 return AVERROR_INVALIDDATA;
1833 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
1834 avctx->extradata_size * 8, 1);
1835 if (!cfg.chan_config || cfg.chan_config > 7) {
1836 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1837 return AVERROR_INVALIDDATA;
1839 s->frames = mp3Frames[cfg.chan_config];
1840 s->coff = chan_offset[cfg.chan_config];
1841 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1842 avctx->channel_layout = chan_layout[cfg.chan_config];
1844 if (cfg.sample_rate < 16000)
1845 s->syncword = 0xffe00000;
1847 s->syncword = 0xfff00000;
1849 /* Init the first mp3 decoder in standard way, so that all tables get builded
1850 * We replace avctx->priv_data with the context of the first decoder so that
1851 * decode_init() does not have to be changed.
1852 * Other decoders will be initialized here copying data from the first context
1854 // Allocate zeroed memory for the first decoder context
1855 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1856 if (!s->mp3decctx[0])
1858 // Put decoder context in place to make init_decode() happy
1859 avctx->priv_data = s->mp3decctx[0];
1861 s->frame = avctx->coded_frame;
1862 // Restore mp3on4 context pointer
1863 avctx->priv_data = s;
1864 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1866 /* Create a separate codec/context for each frame (first is already ok).
1867 * Each frame is 1 or 2 channels - up to 5 frames allowed
1869 for (i = 1; i < s->frames; i++) {
1870 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1871 if (!s->mp3decctx[i])
1873 s->mp3decctx[i]->adu_mode = 1;
1874 s->mp3decctx[i]->avctx = avctx;
1875 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1880 decode_close_mp3on4(avctx);
1881 return AVERROR(ENOMEM);
1885 static void flush_mp3on4(AVCodecContext *avctx)
1888 MP3On4DecodeContext *s = avctx->priv_data;
1890 for (i = 0; i < s->frames; i++)
1891 mp_flush(s->mp3decctx[i]);
1895 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1896 int *got_frame_ptr, AVPacket *avpkt)
1898 const uint8_t *buf = avpkt->data;
1899 int buf_size = avpkt->size;
1900 MP3On4DecodeContext *s = avctx->priv_data;
1901 MPADecodeContext *m;
1902 int fsize, len = buf_size, out_size = 0;
1904 OUT_INT **out_samples;
1908 /* get output buffer */
1909 s->frame->nb_samples = MPA_FRAME_SIZE;
1910 if ((ret = ff_get_buffer(avctx, s->frame)) < 0) {
1911 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1914 out_samples = (OUT_INT **)s->frame->extended_data;
1916 // Discard too short frames
1917 if (buf_size < HEADER_SIZE)
1918 return AVERROR_INVALIDDATA;
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 outptr[0] = out_samples[s->coff[fr]];
1948 if (m->nb_channels > 1)
1949 outptr[1] = out_samples[s->coff[fr] + 1];
1951 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0)
1958 avctx->bit_rate += m->bit_rate;
1961 /* update codec info */
1962 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1964 s->frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1966 *(AVFrame *)data = *s->frame;
1970 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */
1973 #if CONFIG_MP1_DECODER
1974 AVCodec ff_mp1_decoder = {
1976 .type = AVMEDIA_TYPE_AUDIO,
1977 .id = AV_CODEC_ID_MP1,
1978 .priv_data_size = sizeof(MPADecodeContext),
1979 .init = decode_init,
1980 .decode = decode_frame,
1981 .capabilities = CODEC_CAP_DR1,
1983 .long_name = NULL_IF_CONFIG_SMALL("MP1 (MPEG audio layer 1)"),
1984 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
1986 AV_SAMPLE_FMT_NONE },
1989 #if CONFIG_MP2_DECODER
1990 AVCodec ff_mp2_decoder = {
1992 .type = AVMEDIA_TYPE_AUDIO,
1993 .id = AV_CODEC_ID_MP2,
1994 .priv_data_size = sizeof(MPADecodeContext),
1995 .init = decode_init,
1996 .decode = decode_frame,
1997 .capabilities = CODEC_CAP_DR1,
1999 .long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
2000 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
2002 AV_SAMPLE_FMT_NONE },
2005 #if CONFIG_MP3_DECODER
2006 AVCodec ff_mp3_decoder = {
2008 .type = AVMEDIA_TYPE_AUDIO,
2009 .id = AV_CODEC_ID_MP3,
2010 .priv_data_size = sizeof(MPADecodeContext),
2011 .init = decode_init,
2012 .decode = decode_frame,
2013 .capabilities = CODEC_CAP_DR1,
2015 .long_name = NULL_IF_CONFIG_SMALL("MP3 (MPEG audio layer 3)"),
2016 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
2018 AV_SAMPLE_FMT_NONE },
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)"),
2032 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
2034 AV_SAMPLE_FMT_NONE },
2037 #if CONFIG_MP3ON4_DECODER
2038 AVCodec ff_mp3on4_decoder = {
2040 .type = AVMEDIA_TYPE_AUDIO,
2041 .id = AV_CODEC_ID_MP3ON4,
2042 .priv_data_size = sizeof(MP3On4DecodeContext),
2043 .init = decode_init_mp3on4,
2044 .close = decode_close_mp3on4,
2045 .decode = decode_frame_mp3on4,
2046 .capabilities = CODEC_CAP_DR1,
2047 .flush = flush_mp3on4,
2048 .long_name = NULL_IF_CONFIG_SMALL("MP3onMP4"),
2049 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
2050 AV_SAMPLE_FMT_NONE },