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/attributes.h"
28 #include "libavutil/avassert.h"
29 #include "libavutil/channel_layout.h"
30 #include "libavutil/float_dsp.h"
35 #include "mpegaudiodsp.h"
39 * - test lsf / mpeg25 extensively.
42 #include "mpegaudio.h"
43 #include "mpegaudiodecheader.h"
45 #define BACKSTEP_SIZE 512
47 #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES
49 /* layer 3 "granule" */
50 typedef struct GranuleDef {
55 int scalefac_compress;
60 uint8_t scalefac_scale;
61 uint8_t count1table_select;
62 int region_size[3]; /* number of huffman codes in each region */
64 int short_start, long_end; /* long/short band indexes */
65 uint8_t scale_factors[40];
66 DECLARE_ALIGNED(16, INTFLOAT, sb_hybrid)[SBLIMIT * 18]; /* 576 samples */
69 typedef struct MPADecodeContext {
71 uint8_t last_buf[LAST_BUF_SIZE];
73 /* next header (used in free format parsing) */
74 uint32_t free_format_next_header;
77 DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
78 int synth_buf_offset[MPA_MAX_CHANNELS];
79 DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
80 INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
81 GranuleDef granules[2][2]; /* Used in Layer 3 */
82 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
85 AVCodecContext* avctx;
87 AVFloatDSPContext fdsp;
93 #include "mpegaudiodata.h"
94 #include "mpegaudiodectab.h"
96 /* vlc structure for decoding layer 3 huffman tables */
97 static VLC huff_vlc[16];
98 static VLC_TYPE huff_vlc_tables[
99 0 + 128 + 128 + 128 + 130 + 128 + 154 + 166 +
100 142 + 204 + 190 + 170 + 542 + 460 + 662 + 414
102 static const int huff_vlc_tables_sizes[16] = {
103 0, 128, 128, 128, 130, 128, 154, 166,
104 142, 204, 190, 170, 542, 460, 662, 414
106 static VLC huff_quad_vlc[2];
107 static VLC_TYPE huff_quad_vlc_tables[128+16][2];
108 static const int huff_quad_vlc_tables_sizes[2] = { 128, 16 };
109 /* computed from band_size_long */
110 static uint16_t band_index_long[9][23];
111 #include "mpegaudio_tablegen.h"
112 /* intensity stereo coef table */
113 static INTFLOAT is_table[2][16];
114 static INTFLOAT is_table_lsf[2][2][16];
115 static INTFLOAT csa_table[8][4];
117 static int16_t division_tab3[1<<6 ];
118 static int16_t division_tab5[1<<8 ];
119 static int16_t division_tab9[1<<11];
121 static int16_t * const division_tabs[4] = {
122 division_tab3, division_tab5, NULL, division_tab9
125 /* lower 2 bits: modulo 3, higher bits: shift */
126 static uint16_t scale_factor_modshift[64];
127 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
128 static int32_t scale_factor_mult[15][3];
129 /* mult table for layer 2 group quantization */
131 #define SCALE_GEN(v) \
132 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
134 static const int32_t scale_factor_mult2[3][3] = {
135 SCALE_GEN(4.0 / 3.0), /* 3 steps */
136 SCALE_GEN(4.0 / 5.0), /* 5 steps */
137 SCALE_GEN(4.0 / 9.0), /* 9 steps */
141 * Convert region offsets to region sizes and truncate
142 * size to big_values.
144 static void region_offset2size(GranuleDef *g)
147 g->region_size[2] = 576 / 2;
148 for (i = 0; i < 3; i++) {
149 k = FFMIN(g->region_size[i], g->big_values);
150 g->region_size[i] = k - j;
155 static void init_short_region(MPADecodeContext *s, GranuleDef *g)
157 if (g->block_type == 2) {
158 if (s->sample_rate_index != 8)
159 g->region_size[0] = (36 / 2);
161 g->region_size[0] = (72 / 2);
163 if (s->sample_rate_index <= 2)
164 g->region_size[0] = (36 / 2);
165 else if (s->sample_rate_index != 8)
166 g->region_size[0] = (54 / 2);
168 g->region_size[0] = (108 / 2);
170 g->region_size[1] = (576 / 2);
173 static void init_long_region(MPADecodeContext *s, GranuleDef *g,
177 g->region_size[0] = band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
178 /* should not overflow */
179 l = FFMIN(ra1 + ra2 + 2, 22);
180 g->region_size[1] = band_index_long[s->sample_rate_index][ l] >> 1;
183 static void compute_band_indexes(MPADecodeContext *s, GranuleDef *g)
185 if (g->block_type == 2) {
186 if (g->switch_point) {
187 /* if switched mode, we handle the 36 first samples as
188 long blocks. For 8000Hz, we handle the 72 first
189 exponents as long blocks */
190 if (s->sample_rate_index <= 2)
206 /* layer 1 unscaling */
207 /* n = number of bits of the mantissa minus 1 */
208 static inline int l1_unscale(int n, int mant, int scale_factor)
213 shift = scale_factor_modshift[scale_factor];
216 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
218 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
219 return (int)((val + (1LL << (shift - 1))) >> shift);
222 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
226 shift = scale_factor_modshift[scale_factor];
230 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
231 /* NOTE: at this point, 0 <= shift <= 21 */
233 val = (val + (1 << (shift - 1))) >> shift;
237 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
238 static inline int l3_unscale(int value, int exponent)
243 e = table_4_3_exp [4 * value + (exponent & 3)];
244 m = table_4_3_value[4 * value + (exponent & 3)];
249 m = (m + (1 << (e - 1))) >> e;
254 static av_cold void decode_init_static(void)
259 /* scale factors table for layer 1/2 */
260 for (i = 0; i < 64; i++) {
262 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
265 scale_factor_modshift[i] = mod | (shift << 2);
268 /* scale factor multiply for layer 1 */
269 for (i = 0; i < 15; i++) {
272 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
273 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
274 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
275 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
276 ff_dlog(NULL, "%d: norm=%x s=%x %x %x\n", i, norm,
277 scale_factor_mult[i][0],
278 scale_factor_mult[i][1],
279 scale_factor_mult[i][2]);
282 RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
284 /* huffman decode tables */
286 for (i = 1; i < 16; i++) {
287 const HuffTable *h = &mpa_huff_tables[i];
289 uint8_t tmp_bits [512] = { 0 };
290 uint16_t tmp_codes[512] = { 0 };
295 for (x = 0; x < xsize; x++) {
296 for (y = 0; y < xsize; y++) {
297 tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ];
298 tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
303 huff_vlc[i].table = huff_vlc_tables+offset;
304 huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
305 init_vlc(&huff_vlc[i], 7, 512,
306 tmp_bits, 1, 1, tmp_codes, 2, 2,
307 INIT_VLC_USE_NEW_STATIC);
308 offset += huff_vlc_tables_sizes[i];
310 assert(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
313 for (i = 0; i < 2; i++) {
314 huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
315 huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
316 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
317 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
318 INIT_VLC_USE_NEW_STATIC);
319 offset += huff_quad_vlc_tables_sizes[i];
321 assert(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
323 for (i = 0; i < 9; i++) {
325 for (j = 0; j < 22; j++) {
326 band_index_long[i][j] = k;
327 k += band_size_long[i][j];
329 band_index_long[i][22] = k;
332 /* compute n ^ (4/3) and store it in mantissa/exp format */
334 mpegaudio_tableinit();
336 for (i = 0; i < 4; i++) {
337 if (ff_mpa_quant_bits[i] < 0) {
338 for (j = 0; j < (1 << (-ff_mpa_quant_bits[i]+1)); j++) {
339 int val1, val2, val3, steps;
341 steps = ff_mpa_quant_steps[i];
346 division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
352 for (i = 0; i < 7; i++) {
356 f = tan((double)i * M_PI / 12.0);
357 v = FIXR(f / (1.0 + f));
362 is_table[1][6 - i] = v;
365 for (i = 7; i < 16; i++)
366 is_table[0][i] = is_table[1][i] = 0.0;
368 for (i = 0; i < 16; i++) {
372 for (j = 0; j < 2; j++) {
373 e = -(j + 1) * ((i + 1) >> 1);
374 f = pow(2.0, e / 4.0);
376 is_table_lsf[j][k ^ 1][i] = FIXR(f);
377 is_table_lsf[j][k ][i] = FIXR(1.0);
378 ff_dlog(NULL, "is_table_lsf %d %d: %f %f\n",
379 i, j, (float) is_table_lsf[j][0][i],
380 (float) is_table_lsf[j][1][i]);
384 for (i = 0; i < 8; i++) {
387 cs = 1.0 / sqrt(1.0 + ci * ci);
390 csa_table[i][0] = FIXHR(cs/4);
391 csa_table[i][1] = FIXHR(ca/4);
392 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
393 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
395 csa_table[i][0] = cs;
396 csa_table[i][1] = ca;
397 csa_table[i][2] = ca + cs;
398 csa_table[i][3] = ca - cs;
403 static av_cold int decode_init(AVCodecContext * avctx)
405 static int initialized_tables = 0;
406 MPADecodeContext *s = avctx->priv_data;
408 if (!initialized_tables) {
409 decode_init_static();
410 initialized_tables = 1;
415 avpriv_float_dsp_init(&s->fdsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
416 ff_mpadsp_init(&s->mpadsp);
418 if (avctx->request_sample_fmt == OUT_FMT &&
419 avctx->codec_id != AV_CODEC_ID_MP3ON4)
420 avctx->sample_fmt = OUT_FMT;
422 avctx->sample_fmt = OUT_FMT_P;
423 s->err_recognition = avctx->err_recognition;
425 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
431 #define C3 FIXHR(0.86602540378443864676/2)
432 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
433 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
434 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
436 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
438 static void imdct12(INTFLOAT *out, INTFLOAT *in)
440 INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
443 in1 = in[1*3] + in[0*3];
444 in2 = in[2*3] + in[1*3];
445 in3 = in[3*3] + in[2*3];
446 in4 = in[4*3] + in[3*3];
447 in5 = in[5*3] + in[4*3];
451 in2 = MULH3(in2, C3, 2);
452 in3 = MULH3(in3, C3, 4);
455 t2 = MULH3(in1 - in5, C4, 2);
465 in1 = MULH3(in5 + in3, C5, 1);
472 in5 = MULH3(in5 - in3, C6, 2);
479 /* return the number of decoded frames */
480 static int mp_decode_layer1(MPADecodeContext *s)
482 int bound, i, v, n, ch, j, mant;
483 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
484 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
486 if (s->mode == MPA_JSTEREO)
487 bound = (s->mode_ext + 1) * 4;
491 /* allocation bits */
492 for (i = 0; i < bound; i++) {
493 for (ch = 0; ch < s->nb_channels; ch++) {
494 allocation[ch][i] = get_bits(&s->gb, 4);
497 for (i = bound; i < SBLIMIT; i++)
498 allocation[0][i] = get_bits(&s->gb, 4);
501 for (i = 0; i < bound; i++) {
502 for (ch = 0; ch < s->nb_channels; ch++) {
503 if (allocation[ch][i])
504 scale_factors[ch][i] = get_bits(&s->gb, 6);
507 for (i = bound; i < SBLIMIT; i++) {
508 if (allocation[0][i]) {
509 scale_factors[0][i] = get_bits(&s->gb, 6);
510 scale_factors[1][i] = get_bits(&s->gb, 6);
514 /* compute samples */
515 for (j = 0; j < 12; j++) {
516 for (i = 0; i < bound; i++) {
517 for (ch = 0; ch < s->nb_channels; ch++) {
518 n = allocation[ch][i];
520 mant = get_bits(&s->gb, n + 1);
521 v = l1_unscale(n, mant, scale_factors[ch][i]);
525 s->sb_samples[ch][j][i] = v;
528 for (i = bound; i < SBLIMIT; i++) {
529 n = allocation[0][i];
531 mant = get_bits(&s->gb, n + 1);
532 v = l1_unscale(n, mant, scale_factors[0][i]);
533 s->sb_samples[0][j][i] = v;
534 v = l1_unscale(n, mant, scale_factors[1][i]);
535 s->sb_samples[1][j][i] = v;
537 s->sb_samples[0][j][i] = 0;
538 s->sb_samples[1][j][i] = 0;
545 static int mp_decode_layer2(MPADecodeContext *s)
547 int sblimit; /* number of used subbands */
548 const unsigned char *alloc_table;
549 int table, bit_alloc_bits, i, j, ch, bound, v;
550 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
551 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
552 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
553 int scale, qindex, bits, steps, k, l, m, b;
555 /* select decoding table */
556 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
557 s->sample_rate, s->lsf);
558 sblimit = ff_mpa_sblimit_table[table];
559 alloc_table = ff_mpa_alloc_tables[table];
561 if (s->mode == MPA_JSTEREO)
562 bound = (s->mode_ext + 1) * 4;
566 ff_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
572 /* parse bit allocation */
574 for (i = 0; i < bound; i++) {
575 bit_alloc_bits = alloc_table[j];
576 for (ch = 0; ch < s->nb_channels; ch++)
577 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
578 j += 1 << bit_alloc_bits;
580 for (i = bound; i < sblimit; i++) {
581 bit_alloc_bits = alloc_table[j];
582 v = get_bits(&s->gb, bit_alloc_bits);
585 j += 1 << bit_alloc_bits;
589 for (i = 0; i < sblimit; i++) {
590 for (ch = 0; ch < s->nb_channels; ch++) {
591 if (bit_alloc[ch][i])
592 scale_code[ch][i] = get_bits(&s->gb, 2);
597 for (i = 0; i < sblimit; i++) {
598 for (ch = 0; ch < s->nb_channels; ch++) {
599 if (bit_alloc[ch][i]) {
600 sf = scale_factors[ch][i];
601 switch (scale_code[ch][i]) {
604 sf[0] = get_bits(&s->gb, 6);
605 sf[1] = get_bits(&s->gb, 6);
606 sf[2] = get_bits(&s->gb, 6);
609 sf[0] = get_bits(&s->gb, 6);
614 sf[0] = get_bits(&s->gb, 6);
615 sf[2] = get_bits(&s->gb, 6);
619 sf[0] = get_bits(&s->gb, 6);
620 sf[2] = get_bits(&s->gb, 6);
629 for (k = 0; k < 3; k++) {
630 for (l = 0; l < 12; l += 3) {
632 for (i = 0; i < bound; i++) {
633 bit_alloc_bits = alloc_table[j];
634 for (ch = 0; ch < s->nb_channels; ch++) {
635 b = bit_alloc[ch][i];
637 scale = scale_factors[ch][i][k];
638 qindex = alloc_table[j+b];
639 bits = ff_mpa_quant_bits[qindex];
642 /* 3 values at the same time */
643 v = get_bits(&s->gb, -bits);
644 v2 = division_tabs[qindex][v];
645 steps = ff_mpa_quant_steps[qindex];
647 s->sb_samples[ch][k * 12 + l + 0][i] =
648 l2_unscale_group(steps, v2 & 15, scale);
649 s->sb_samples[ch][k * 12 + l + 1][i] =
650 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
651 s->sb_samples[ch][k * 12 + l + 2][i] =
652 l2_unscale_group(steps, v2 >> 8 , scale);
654 for (m = 0; m < 3; m++) {
655 v = get_bits(&s->gb, bits);
656 v = l1_unscale(bits - 1, v, scale);
657 s->sb_samples[ch][k * 12 + l + m][i] = v;
661 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
662 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
663 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
666 /* next subband in alloc table */
667 j += 1 << bit_alloc_bits;
669 /* XXX: find a way to avoid this duplication of code */
670 for (i = bound; i < sblimit; i++) {
671 bit_alloc_bits = alloc_table[j];
674 int mant, scale0, scale1;
675 scale0 = scale_factors[0][i][k];
676 scale1 = scale_factors[1][i][k];
677 qindex = alloc_table[j+b];
678 bits = ff_mpa_quant_bits[qindex];
680 /* 3 values at the same time */
681 v = get_bits(&s->gb, -bits);
682 steps = ff_mpa_quant_steps[qindex];
685 s->sb_samples[0][k * 12 + l + 0][i] =
686 l2_unscale_group(steps, mant, scale0);
687 s->sb_samples[1][k * 12 + l + 0][i] =
688 l2_unscale_group(steps, mant, scale1);
691 s->sb_samples[0][k * 12 + l + 1][i] =
692 l2_unscale_group(steps, mant, scale0);
693 s->sb_samples[1][k * 12 + l + 1][i] =
694 l2_unscale_group(steps, mant, scale1);
695 s->sb_samples[0][k * 12 + l + 2][i] =
696 l2_unscale_group(steps, v, scale0);
697 s->sb_samples[1][k * 12 + l + 2][i] =
698 l2_unscale_group(steps, v, scale1);
700 for (m = 0; m < 3; m++) {
701 mant = get_bits(&s->gb, bits);
702 s->sb_samples[0][k * 12 + l + m][i] =
703 l1_unscale(bits - 1, mant, scale0);
704 s->sb_samples[1][k * 12 + l + m][i] =
705 l1_unscale(bits - 1, mant, scale1);
709 s->sb_samples[0][k * 12 + l + 0][i] = 0;
710 s->sb_samples[0][k * 12 + l + 1][i] = 0;
711 s->sb_samples[0][k * 12 + l + 2][i] = 0;
712 s->sb_samples[1][k * 12 + l + 0][i] = 0;
713 s->sb_samples[1][k * 12 + l + 1][i] = 0;
714 s->sb_samples[1][k * 12 + l + 2][i] = 0;
716 /* next subband in alloc table */
717 j += 1 << bit_alloc_bits;
719 /* fill remaining samples to zero */
720 for (i = sblimit; i < SBLIMIT; i++) {
721 for (ch = 0; ch < s->nb_channels; ch++) {
722 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
723 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
724 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
732 #define SPLIT(dst,sf,n) \
734 int m = (sf * 171) >> 9; \
737 } else if (n == 4) { \
740 } else if (n == 5) { \
741 int m = (sf * 205) >> 10; \
744 } else if (n == 6) { \
745 int m = (sf * 171) >> 10; \
752 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
755 SPLIT(slen[3], sf, n3)
756 SPLIT(slen[2], sf, n2)
757 SPLIT(slen[1], sf, n1)
761 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
764 const uint8_t *bstab, *pretab;
765 int len, i, j, k, l, v0, shift, gain, gains[3];
769 gain = g->global_gain - 210;
770 shift = g->scalefac_scale + 1;
772 bstab = band_size_long[s->sample_rate_index];
773 pretab = mpa_pretab[g->preflag];
774 for (i = 0; i < g->long_end; i++) {
775 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
777 for (j = len; j > 0; j--)
781 if (g->short_start < 13) {
782 bstab = band_size_short[s->sample_rate_index];
783 gains[0] = gain - (g->subblock_gain[0] << 3);
784 gains[1] = gain - (g->subblock_gain[1] << 3);
785 gains[2] = gain - (g->subblock_gain[2] << 3);
787 for (i = g->short_start; i < 13; i++) {
789 for (l = 0; l < 3; l++) {
790 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
791 for (j = len; j > 0; j--)
798 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
801 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits) {
803 s->in_gb.buffer = NULL;
804 assert((get_bits_count(&s->gb) & 7) == 0);
805 skip_bits_long(&s->gb, *pos - *end_pos);
807 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
808 *pos = get_bits_count(&s->gb);
812 /* Following is a optimized code for
814 if(get_bits1(&s->gb))
819 #define READ_FLIP_SIGN(dst,src) \
820 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
823 #define READ_FLIP_SIGN(dst,src) \
824 v = -get_bits1(&s->gb); \
825 *(dst) = (*(src) ^ v) - v;
828 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
829 int16_t *exponents, int end_pos2)
833 int last_pos, bits_left;
835 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits);
837 /* low frequencies (called big values) */
839 for (i = 0; i < 3; i++) {
840 int j, k, l, linbits;
841 j = g->region_size[i];
844 /* select vlc table */
845 k = g->table_select[i];
846 l = mpa_huff_data[k][0];
847 linbits = mpa_huff_data[k][1];
851 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
856 /* read huffcode and compute each couple */
860 int pos = get_bits_count(&s->gb);
863 switch_buffer(s, &pos, &end_pos, &end_pos2);
867 y = get_vlc2(&s->gb, vlc->table, 7, 3);
870 g->sb_hybrid[s_index ] =
871 g->sb_hybrid[s_index+1] = 0;
876 exponent= exponents[s_index];
878 ff_dlog(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
879 i, g->region_size[i] - j, x, y, exponent);
884 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
886 x += get_bitsz(&s->gb, linbits);
887 v = l3_unscale(x, exponent);
888 if (get_bits1(&s->gb))
890 g->sb_hybrid[s_index] = v;
893 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
895 y += get_bitsz(&s->gb, linbits);
896 v = l3_unscale(y, exponent);
897 if (get_bits1(&s->gb))
899 g->sb_hybrid[s_index+1] = v;
906 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
908 x += get_bitsz(&s->gb, linbits);
909 v = l3_unscale(x, exponent);
910 if (get_bits1(&s->gb))
912 g->sb_hybrid[s_index+!!y] = v;
914 g->sb_hybrid[s_index + !y] = 0;
920 /* high frequencies */
921 vlc = &huff_quad_vlc[g->count1table_select];
923 while (s_index <= 572) {
925 pos = get_bits_count(&s->gb);
926 if (pos >= end_pos) {
927 if (pos > end_pos2 && last_pos) {
928 /* some encoders generate an incorrect size for this
929 part. We must go back into the data */
931 skip_bits_long(&s->gb, last_pos - pos);
932 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
933 if(s->err_recognition & AV_EF_BITSTREAM)
937 switch_buffer(s, &pos, &end_pos, &end_pos2);
943 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
944 ff_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
945 g->sb_hybrid[s_index+0] =
946 g->sb_hybrid[s_index+1] =
947 g->sb_hybrid[s_index+2] =
948 g->sb_hybrid[s_index+3] = 0;
950 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
952 int pos = s_index + idxtab[code];
953 code ^= 8 >> idxtab[code];
954 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
958 /* skip extension bits */
959 bits_left = end_pos2 - get_bits_count(&s->gb);
960 if (bits_left < 0 && (s->err_recognition & AV_EF_BUFFER)) {
961 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
963 } else if (bits_left > 0 && (s->err_recognition & AV_EF_BUFFER)) {
964 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
967 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
968 skip_bits_long(&s->gb, bits_left);
970 i = get_bits_count(&s->gb);
971 switch_buffer(s, &i, &end_pos, &end_pos2);
976 /* Reorder short blocks from bitstream order to interleaved order. It
977 would be faster to do it in parsing, but the code would be far more
979 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
982 INTFLOAT *ptr, *dst, *ptr1;
985 if (g->block_type != 2)
988 if (g->switch_point) {
989 if (s->sample_rate_index != 8)
990 ptr = g->sb_hybrid + 36;
992 ptr = g->sb_hybrid + 72;
997 for (i = g->short_start; i < 13; i++) {
998 len = band_size_short[s->sample_rate_index][i];
1001 for (j = len; j > 0; j--) {
1002 *dst++ = ptr[0*len];
1003 *dst++ = ptr[1*len];
1004 *dst++ = ptr[2*len];
1008 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1012 #define ISQRT2 FIXR(0.70710678118654752440)
1014 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1017 int sf_max, sf, len, non_zero_found;
1018 INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1019 int non_zero_found_short[3];
1021 /* intensity stereo */
1022 if (s->mode_ext & MODE_EXT_I_STEREO) {
1027 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1031 tab0 = g0->sb_hybrid + 576;
1032 tab1 = g1->sb_hybrid + 576;
1034 non_zero_found_short[0] = 0;
1035 non_zero_found_short[1] = 0;
1036 non_zero_found_short[2] = 0;
1037 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1038 for (i = 12; i >= g1->short_start; i--) {
1039 /* for last band, use previous scale factor */
1042 len = band_size_short[s->sample_rate_index][i];
1043 for (l = 2; l >= 0; l--) {
1046 if (!non_zero_found_short[l]) {
1047 /* test if non zero band. if so, stop doing i-stereo */
1048 for (j = 0; j < len; j++) {
1050 non_zero_found_short[l] = 1;
1054 sf = g1->scale_factors[k + l];
1060 for (j = 0; j < len; j++) {
1062 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1063 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1067 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1068 /* lower part of the spectrum : do ms stereo
1070 for (j = 0; j < len; j++) {
1073 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1074 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1081 non_zero_found = non_zero_found_short[0] |
1082 non_zero_found_short[1] |
1083 non_zero_found_short[2];
1085 for (i = g1->long_end - 1;i >= 0;i--) {
1086 len = band_size_long[s->sample_rate_index][i];
1089 /* test if non zero band. if so, stop doing i-stereo */
1090 if (!non_zero_found) {
1091 for (j = 0; j < len; j++) {
1097 /* for last band, use previous scale factor */
1098 k = (i == 21) ? 20 : i;
1099 sf = g1->scale_factors[k];
1104 for (j = 0; j < len; j++) {
1106 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1107 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1111 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1112 /* lower part of the spectrum : do ms stereo
1114 for (j = 0; j < len; j++) {
1117 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1118 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1123 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1124 /* ms stereo ONLY */
1125 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1128 s->fdsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1130 tab0 = g0->sb_hybrid;
1131 tab1 = g1->sb_hybrid;
1132 for (i = 0; i < 576; i++) {
1135 tab0[i] = tmp0 + tmp1;
1136 tab1[i] = tmp0 - tmp1;
1143 #define AA(j) do { \
1144 float tmp0 = ptr[-1-j]; \
1145 float tmp1 = ptr[ j]; \
1146 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1147 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1150 #define AA(j) do { \
1151 int tmp0 = ptr[-1-j]; \
1152 int tmp1 = ptr[ j]; \
1153 int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1154 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1155 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1159 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1164 /* we antialias only "long" bands */
1165 if (g->block_type == 2) {
1166 if (!g->switch_point)
1168 /* XXX: check this for 8000Hz case */
1174 ptr = g->sb_hybrid + 18;
1175 for (i = n; i > 0; i--) {
1189 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1190 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1192 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1194 int i, j, mdct_long_end, sblimit;
1196 /* find last non zero block */
1197 ptr = g->sb_hybrid + 576;
1198 ptr1 = g->sb_hybrid + 2 * 18;
1199 while (ptr >= ptr1) {
1203 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1206 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1208 if (g->block_type == 2) {
1209 /* XXX: check for 8000 Hz */
1210 if (g->switch_point)
1215 mdct_long_end = sblimit;
1218 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1219 mdct_long_end, g->switch_point,
1222 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1223 ptr = g->sb_hybrid + 18 * mdct_long_end;
1225 for (j = mdct_long_end; j < sblimit; j++) {
1226 /* select frequency inversion */
1227 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1228 out_ptr = sb_samples + j;
1230 for (i = 0; i < 6; i++) {
1231 *out_ptr = buf[4*i];
1234 imdct12(out2, ptr + 0);
1235 for (i = 0; i < 6; i++) {
1236 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1237 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1240 imdct12(out2, ptr + 1);
1241 for (i = 0; i < 6; i++) {
1242 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1243 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1246 imdct12(out2, ptr + 2);
1247 for (i = 0; i < 6; i++) {
1248 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1249 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1250 buf[4*(i + 6*2)] = 0;
1253 buf += (j&3) != 3 ? 1 : (4*18-3);
1256 for (j = sblimit; j < SBLIMIT; j++) {
1258 out_ptr = sb_samples + j;
1259 for (i = 0; i < 18; i++) {
1260 *out_ptr = buf[4*i];
1264 buf += (j&3) != 3 ? 1 : (4*18-3);
1268 /* main layer3 decoding function */
1269 static int mp_decode_layer3(MPADecodeContext *s)
1271 int nb_granules, main_data_begin;
1272 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1274 int16_t exponents[576]; //FIXME try INTFLOAT
1276 /* read side info */
1278 main_data_begin = get_bits(&s->gb, 8);
1279 skip_bits(&s->gb, s->nb_channels);
1282 main_data_begin = get_bits(&s->gb, 9);
1283 if (s->nb_channels == 2)
1284 skip_bits(&s->gb, 3);
1286 skip_bits(&s->gb, 5);
1288 for (ch = 0; ch < s->nb_channels; ch++) {
1289 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1290 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1294 for (gr = 0; gr < nb_granules; gr++) {
1295 for (ch = 0; ch < s->nb_channels; ch++) {
1296 ff_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1297 g = &s->granules[ch][gr];
1298 g->part2_3_length = get_bits(&s->gb, 12);
1299 g->big_values = get_bits(&s->gb, 9);
1300 if (g->big_values > 288) {
1301 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1302 return AVERROR_INVALIDDATA;
1305 g->global_gain = get_bits(&s->gb, 8);
1306 /* if MS stereo only is selected, we precompute the
1307 1/sqrt(2) renormalization factor */
1308 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1310 g->global_gain -= 2;
1312 g->scalefac_compress = get_bits(&s->gb, 9);
1314 g->scalefac_compress = get_bits(&s->gb, 4);
1315 blocksplit_flag = get_bits1(&s->gb);
1316 if (blocksplit_flag) {
1317 g->block_type = get_bits(&s->gb, 2);
1318 if (g->block_type == 0) {
1319 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1320 return AVERROR_INVALIDDATA;
1322 g->switch_point = get_bits1(&s->gb);
1323 for (i = 0; i < 2; i++)
1324 g->table_select[i] = get_bits(&s->gb, 5);
1325 for (i = 0; i < 3; i++)
1326 g->subblock_gain[i] = get_bits(&s->gb, 3);
1327 init_short_region(s, g);
1329 int region_address1, region_address2;
1331 g->switch_point = 0;
1332 for (i = 0; i < 3; i++)
1333 g->table_select[i] = get_bits(&s->gb, 5);
1334 /* compute huffman coded region sizes */
1335 region_address1 = get_bits(&s->gb, 4);
1336 region_address2 = get_bits(&s->gb, 3);
1337 ff_dlog(s->avctx, "region1=%d region2=%d\n",
1338 region_address1, region_address2);
1339 init_long_region(s, g, region_address1, region_address2);
1341 region_offset2size(g);
1342 compute_band_indexes(s, g);
1346 g->preflag = get_bits1(&s->gb);
1347 g->scalefac_scale = get_bits1(&s->gb);
1348 g->count1table_select = get_bits1(&s->gb);
1349 ff_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1350 g->block_type, g->switch_point);
1356 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1357 int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0,
1358 FFMAX(0, LAST_BUF_SIZE - s->last_buf_size));
1359 assert((get_bits_count(&s->gb) & 7) == 0);
1360 /* now we get bits from the main_data_begin offset */
1361 ff_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1362 main_data_begin, s->last_buf_size);
1364 memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);
1366 init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
1367 #if !UNCHECKED_BITSTREAM_READER
1368 s->gb.size_in_bits_plus8 += extrasize * 8;
1370 s->last_buf_size <<= 3;
1371 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1372 for (ch = 0; ch < s->nb_channels; ch++) {
1373 g = &s->granules[ch][gr];
1374 s->last_buf_size += g->part2_3_length;
1375 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1376 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1379 skip = s->last_buf_size - 8 * main_data_begin;
1380 if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {
1381 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);
1383 s->in_gb.buffer = NULL;
1385 skip_bits_long(&s->gb, skip);
1391 for (; gr < nb_granules; gr++) {
1392 for (ch = 0; ch < s->nb_channels; ch++) {
1393 g = &s->granules[ch][gr];
1394 bits_pos = get_bits_count(&s->gb);
1398 int slen, slen1, slen2;
1400 /* MPEG-1 scale factors */
1401 slen1 = slen_table[0][g->scalefac_compress];
1402 slen2 = slen_table[1][g->scalefac_compress];
1403 ff_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1404 if (g->block_type == 2) {
1405 n = g->switch_point ? 17 : 18;
1408 for (i = 0; i < n; i++)
1409 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1411 for (i = 0; i < n; i++)
1412 g->scale_factors[j++] = 0;
1415 for (i = 0; i < 18; i++)
1416 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1417 for (i = 0; i < 3; i++)
1418 g->scale_factors[j++] = 0;
1420 for (i = 0; i < 21; i++)
1421 g->scale_factors[j++] = 0;
1424 sc = s->granules[ch][0].scale_factors;
1426 for (k = 0; k < 4; k++) {
1428 if ((g->scfsi & (0x8 >> k)) == 0) {
1429 slen = (k < 2) ? slen1 : slen2;
1431 for (i = 0; i < n; i++)
1432 g->scale_factors[j++] = get_bits(&s->gb, slen);
1434 for (i = 0; i < n; i++)
1435 g->scale_factors[j++] = 0;
1438 /* simply copy from last granule */
1439 for (i = 0; i < n; i++) {
1440 g->scale_factors[j] = sc[j];
1445 g->scale_factors[j++] = 0;
1448 int tindex, tindex2, slen[4], sl, sf;
1450 /* LSF scale factors */
1451 if (g->block_type == 2)
1452 tindex = g->switch_point ? 2 : 1;
1456 sf = g->scalefac_compress;
1457 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1458 /* intensity stereo case */
1461 lsf_sf_expand(slen, sf, 6, 6, 0);
1463 } else if (sf < 244) {
1464 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1467 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1473 lsf_sf_expand(slen, sf, 5, 4, 4);
1475 } else if (sf < 500) {
1476 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1479 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1486 for (k = 0; k < 4; k++) {
1487 n = lsf_nsf_table[tindex2][tindex][k];
1490 for (i = 0; i < n; i++)
1491 g->scale_factors[j++] = get_bits(&s->gb, sl);
1493 for (i = 0; i < n; i++)
1494 g->scale_factors[j++] = 0;
1497 /* XXX: should compute exact size */
1499 g->scale_factors[j] = 0;
1502 exponents_from_scale_factors(s, g, exponents);
1504 /* read Huffman coded residue */
1505 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1508 if (s->mode == MPA_JSTEREO)
1509 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1511 for (ch = 0; ch < s->nb_channels; ch++) {
1512 g = &s->granules[ch][gr];
1514 reorder_block(s, g);
1515 compute_antialias(s, g);
1516 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1519 if (get_bits_count(&s->gb) < 0)
1520 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1521 return nb_granules * 18;
1524 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1525 const uint8_t *buf, int buf_size)
1527 int i, nb_frames, ch, ret;
1528 OUT_INT *samples_ptr;
1530 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1532 /* skip error protection field */
1533 if (s->error_protection)
1534 skip_bits(&s->gb, 16);
1538 s->avctx->frame_size = 384;
1539 nb_frames = mp_decode_layer1(s);
1542 s->avctx->frame_size = 1152;
1543 nb_frames = mp_decode_layer2(s);
1546 s->avctx->frame_size = s->lsf ? 576 : 1152;
1548 nb_frames = mp_decode_layer3(s);
1554 if (s->in_gb.buffer) {
1555 align_get_bits(&s->gb);
1556 i = get_bits_left(&s->gb)>>3;
1557 if (i >= 0 && i <= BACKSTEP_SIZE) {
1558 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1561 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1563 s->in_gb.buffer = NULL;
1566 align_get_bits(&s->gb);
1567 assert((get_bits_count(&s->gb) & 7) == 0);
1568 i = get_bits_left(&s->gb) >> 3;
1570 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1572 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1573 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1575 assert(i <= buf_size - HEADER_SIZE && i >= 0);
1576 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1577 s->last_buf_size += i;
1580 /* get output buffer */
1582 av_assert0(s->frame != NULL);
1583 s->frame->nb_samples = s->avctx->frame_size;
1584 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0) {
1585 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1588 samples = (OUT_INT **)s->frame->extended_data;
1591 /* apply the synthesis filter */
1592 for (ch = 0; ch < s->nb_channels; ch++) {
1594 if (s->avctx->sample_fmt == OUT_FMT_P) {
1595 samples_ptr = samples[ch];
1598 samples_ptr = samples[0] + ch;
1599 sample_stride = s->nb_channels;
1601 for (i = 0; i < nb_frames; i++) {
1602 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1603 &(s->synth_buf_offset[ch]),
1604 RENAME(ff_mpa_synth_window),
1605 &s->dither_state, samples_ptr,
1606 sample_stride, s->sb_samples[ch][i]);
1607 samples_ptr += 32 * sample_stride;
1611 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1614 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1617 const uint8_t *buf = avpkt->data;
1618 int buf_size = avpkt->size;
1619 MPADecodeContext *s = avctx->priv_data;
1623 if (buf_size < HEADER_SIZE)
1624 return AVERROR_INVALIDDATA;
1626 header = AV_RB32(buf);
1628 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1630 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1631 return AVERROR_INVALIDDATA;
1632 } else if (ret == 1) {
1633 /* free format: prepare to compute frame size */
1635 return AVERROR_INVALIDDATA;
1637 /* update codec info */
1638 avctx->channels = s->nb_channels;
1639 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1640 if (!avctx->bit_rate)
1641 avctx->bit_rate = s->bit_rate;
1645 ret = mp_decode_frame(s, NULL, buf, buf_size);
1647 s->frame->nb_samples = avctx->frame_size;
1649 avctx->sample_rate = s->sample_rate;
1650 //FIXME maybe move the other codec info stuff from above here too
1652 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1653 /* Only return an error if the bad frame makes up the whole packet or
1654 * the error is related to buffer management.
1655 * If there is more data in the packet, just consume the bad frame
1656 * instead of returning an error, which would discard the whole
1659 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1666 static void mp_flush(MPADecodeContext *ctx)
1668 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1669 ctx->last_buf_size = 0;
1672 static void flush(AVCodecContext *avctx)
1674 mp_flush(avctx->priv_data);
1677 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1678 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1679 int *got_frame_ptr, AVPacket *avpkt)
1681 const uint8_t *buf = avpkt->data;
1682 int buf_size = avpkt->size;
1683 MPADecodeContext *s = avctx->priv_data;
1689 // Discard too short frames
1690 if (buf_size < HEADER_SIZE) {
1691 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1692 return AVERROR_INVALIDDATA;
1696 if (len > MPA_MAX_CODED_FRAME_SIZE)
1697 len = MPA_MAX_CODED_FRAME_SIZE;
1699 // Get header and restore sync word
1700 header = AV_RB32(buf) | 0xffe00000;
1702 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1704 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1707 /* update codec info */
1708 avctx->sample_rate = s->sample_rate;
1709 avctx->channels = s->nb_channels;
1710 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1711 if (!avctx->bit_rate)
1712 avctx->bit_rate = s->bit_rate;
1714 s->frame_size = len;
1718 ret = mp_decode_frame(s, NULL, buf, buf_size);
1720 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1728 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1730 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1733 * Context for MP3On4 decoder
1735 typedef struct MP3On4DecodeContext {
1736 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1737 int syncword; ///< syncword patch
1738 const uint8_t *coff; ///< channel offsets in output buffer
1739 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1740 } MP3On4DecodeContext;
1742 #include "mpeg4audio.h"
1744 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1746 /* number of mp3 decoder instances */
1747 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1749 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1750 static const uint8_t chan_offset[8][5] = {
1755 { 2, 0, 3 }, // C FLR BS
1756 { 2, 0, 3 }, // C FLR BLRS
1757 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1758 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1761 /* mp3on4 channel layouts */
1762 static const int16_t chan_layout[8] = {
1765 AV_CH_LAYOUT_STEREO,
1766 AV_CH_LAYOUT_SURROUND,
1767 AV_CH_LAYOUT_4POINT0,
1768 AV_CH_LAYOUT_5POINT0,
1769 AV_CH_LAYOUT_5POINT1,
1770 AV_CH_LAYOUT_7POINT1
1773 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1775 MP3On4DecodeContext *s = avctx->priv_data;
1778 for (i = 0; i < s->frames; i++)
1779 av_free(s->mp3decctx[i]);
1785 static av_cold int decode_init_mp3on4(AVCodecContext * avctx)
1787 MP3On4DecodeContext *s = avctx->priv_data;
1788 MPEG4AudioConfig cfg;
1791 if ((avctx->extradata_size < 2) || !avctx->extradata) {
1792 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1793 return AVERROR_INVALIDDATA;
1796 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
1797 avctx->extradata_size * 8, 1);
1798 if (!cfg.chan_config || cfg.chan_config > 7) {
1799 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1800 return AVERROR_INVALIDDATA;
1802 s->frames = mp3Frames[cfg.chan_config];
1803 s->coff = chan_offset[cfg.chan_config];
1804 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1805 avctx->channel_layout = chan_layout[cfg.chan_config];
1807 if (cfg.sample_rate < 16000)
1808 s->syncword = 0xffe00000;
1810 s->syncword = 0xfff00000;
1812 /* Init the first mp3 decoder in standard way, so that all tables get builded
1813 * We replace avctx->priv_data with the context of the first decoder so that
1814 * decode_init() does not have to be changed.
1815 * Other decoders will be initialized here copying data from the first context
1817 // Allocate zeroed memory for the first decoder context
1818 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1819 if (!s->mp3decctx[0])
1821 // Put decoder context in place to make init_decode() happy
1822 avctx->priv_data = s->mp3decctx[0];
1824 // Restore mp3on4 context pointer
1825 avctx->priv_data = s;
1826 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1828 /* Create a separate codec/context for each frame (first is already ok).
1829 * Each frame is 1 or 2 channels - up to 5 frames allowed
1831 for (i = 1; i < s->frames; i++) {
1832 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1833 if (!s->mp3decctx[i])
1835 s->mp3decctx[i]->adu_mode = 1;
1836 s->mp3decctx[i]->avctx = avctx;
1837 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1842 decode_close_mp3on4(avctx);
1843 return AVERROR(ENOMEM);
1847 static void flush_mp3on4(AVCodecContext *avctx)
1850 MP3On4DecodeContext *s = avctx->priv_data;
1852 for (i = 0; i < s->frames; i++)
1853 mp_flush(s->mp3decctx[i]);
1857 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1858 int *got_frame_ptr, AVPacket *avpkt)
1860 AVFrame *frame = data;
1861 const uint8_t *buf = avpkt->data;
1862 int buf_size = avpkt->size;
1863 MP3On4DecodeContext *s = avctx->priv_data;
1864 MPADecodeContext *m;
1865 int fsize, len = buf_size, out_size = 0;
1867 OUT_INT **out_samples;
1871 /* get output buffer */
1872 frame->nb_samples = MPA_FRAME_SIZE;
1873 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1874 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1877 out_samples = (OUT_INT **)frame->extended_data;
1879 // Discard too short frames
1880 if (buf_size < HEADER_SIZE)
1881 return AVERROR_INVALIDDATA;
1883 avctx->bit_rate = 0;
1886 for (fr = 0; fr < s->frames; fr++) {
1887 fsize = AV_RB16(buf) >> 4;
1888 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1889 m = s->mp3decctx[fr];
1892 if (fsize < HEADER_SIZE) {
1893 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1894 return AVERROR_INVALIDDATA;
1896 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1898 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1899 if (ret < 0) // Bad header, discard block
1902 if (ch + m->nb_channels > avctx->channels ||
1903 s->coff[fr] + m->nb_channels > avctx->channels) {
1904 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1906 return AVERROR_INVALIDDATA;
1908 ch += m->nb_channels;
1910 outptr[0] = out_samples[s->coff[fr]];
1911 if (m->nb_channels > 1)
1912 outptr[1] = out_samples[s->coff[fr] + 1];
1914 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0)
1921 avctx->bit_rate += m->bit_rate;
1924 /* update codec info */
1925 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1927 frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1932 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */