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];
74 /* next header (used in free format parsing) */
75 uint32_t free_format_next_header;
78 DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
79 int synth_buf_offset[MPA_MAX_CHANNELS];
80 DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
81 INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
82 GranuleDef granules[2][2]; /* Used in Layer 3 */
83 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
86 AVCodecContext* avctx;
88 AVFloatDSPContext fdsp;
94 #include "mpegaudiodata.h"
95 #include "mpegaudiodectab.h"
97 /* vlc structure for decoding layer 3 huffman tables */
98 static VLC huff_vlc[16];
99 static VLC_TYPE huff_vlc_tables[
100 0 + 128 + 128 + 128 + 130 + 128 + 154 + 166 +
101 142 + 204 + 190 + 170 + 542 + 460 + 662 + 414
103 static const int huff_vlc_tables_sizes[16] = {
104 0, 128, 128, 128, 130, 128, 154, 166,
105 142, 204, 190, 170, 542, 460, 662, 414
107 static VLC huff_quad_vlc[2];
108 static VLC_TYPE huff_quad_vlc_tables[128+16][2];
109 static const int huff_quad_vlc_tables_sizes[2] = { 128, 16 };
110 /* computed from band_size_long */
111 static uint16_t band_index_long[9][23];
112 #include "mpegaudio_tablegen.h"
113 /* intensity stereo coef table */
114 static INTFLOAT is_table[2][16];
115 static INTFLOAT is_table_lsf[2][2][16];
116 static INTFLOAT csa_table[8][4];
118 static int16_t division_tab3[1<<6 ];
119 static int16_t division_tab5[1<<8 ];
120 static int16_t division_tab9[1<<11];
122 static int16_t * const division_tabs[4] = {
123 division_tab3, division_tab5, NULL, division_tab9
126 /* lower 2 bits: modulo 3, higher bits: shift */
127 static uint16_t scale_factor_modshift[64];
128 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
129 static int32_t scale_factor_mult[15][3];
130 /* mult table for layer 2 group quantization */
132 #define SCALE_GEN(v) \
133 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
135 static const int32_t scale_factor_mult2[3][3] = {
136 SCALE_GEN(4.0 / 3.0), /* 3 steps */
137 SCALE_GEN(4.0 / 5.0), /* 5 steps */
138 SCALE_GEN(4.0 / 9.0), /* 9 steps */
142 * Convert region offsets to region sizes and truncate
143 * size to big_values.
145 static void region_offset2size(GranuleDef *g)
148 g->region_size[2] = 576 / 2;
149 for (i = 0; i < 3; i++) {
150 k = FFMIN(g->region_size[i], g->big_values);
151 g->region_size[i] = k - j;
156 static void init_short_region(MPADecodeContext *s, GranuleDef *g)
158 if (g->block_type == 2) {
159 if (s->sample_rate_index != 8)
160 g->region_size[0] = (36 / 2);
162 g->region_size[0] = (72 / 2);
164 if (s->sample_rate_index <= 2)
165 g->region_size[0] = (36 / 2);
166 else if (s->sample_rate_index != 8)
167 g->region_size[0] = (54 / 2);
169 g->region_size[0] = (108 / 2);
171 g->region_size[1] = (576 / 2);
174 static void init_long_region(MPADecodeContext *s, GranuleDef *g,
178 g->region_size[0] = band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
179 /* should not overflow */
180 l = FFMIN(ra1 + ra2 + 2, 22);
181 g->region_size[1] = band_index_long[s->sample_rate_index][ l] >> 1;
184 static void compute_band_indexes(MPADecodeContext *s, GranuleDef *g)
186 if (g->block_type == 2) {
187 if (g->switch_point) {
188 /* if switched mode, we handle the 36 first samples as
189 long blocks. For 8000Hz, we handle the 72 first
190 exponents as long blocks */
191 if (s->sample_rate_index <= 2)
207 /* layer 1 unscaling */
208 /* n = number of bits of the mantissa minus 1 */
209 static inline int l1_unscale(int n, int mant, int scale_factor)
214 shift = scale_factor_modshift[scale_factor];
217 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
219 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
220 return (int)((val + (1LL << (shift - 1))) >> shift);
223 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
227 shift = scale_factor_modshift[scale_factor];
231 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
232 /* NOTE: at this point, 0 <= shift <= 21 */
234 val = (val + (1 << (shift - 1))) >> shift;
238 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
239 static inline int l3_unscale(int value, int exponent)
244 e = table_4_3_exp [4 * value + (exponent & 3)];
245 m = table_4_3_value[4 * value + (exponent & 3)];
250 m = (m + (1 << (e - 1))) >> e;
255 static av_cold void decode_init_static(void)
260 /* scale factors table for layer 1/2 */
261 for (i = 0; i < 64; i++) {
263 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
266 scale_factor_modshift[i] = mod | (shift << 2);
269 /* scale factor multiply for layer 1 */
270 for (i = 0; i < 15; i++) {
273 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
274 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
275 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
276 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
277 ff_dlog(NULL, "%d: norm=%x s=%"PRIx32" %"PRIx32" %"PRIx32"\n", i, norm,
278 scale_factor_mult[i][0],
279 scale_factor_mult[i][1],
280 scale_factor_mult[i][2]);
283 RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
285 /* huffman decode tables */
287 for (i = 1; i < 16; i++) {
288 const HuffTable *h = &mpa_huff_tables[i];
290 uint8_t tmp_bits [512] = { 0 };
291 uint16_t tmp_codes[512] = { 0 };
296 for (x = 0; x < xsize; x++) {
297 for (y = 0; y < xsize; y++) {
298 tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ];
299 tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
304 huff_vlc[i].table = huff_vlc_tables+offset;
305 huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
306 init_vlc(&huff_vlc[i], 7, 512,
307 tmp_bits, 1, 1, tmp_codes, 2, 2,
308 INIT_VLC_USE_NEW_STATIC);
309 offset += huff_vlc_tables_sizes[i];
311 assert(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
314 for (i = 0; i < 2; i++) {
315 huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
316 huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
317 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
318 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
319 INIT_VLC_USE_NEW_STATIC);
320 offset += huff_quad_vlc_tables_sizes[i];
322 assert(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
324 for (i = 0; i < 9; i++) {
326 for (j = 0; j < 22; j++) {
327 band_index_long[i][j] = k;
328 k += band_size_long[i][j];
330 band_index_long[i][22] = k;
333 /* compute n ^ (4/3) and store it in mantissa/exp format */
335 mpegaudio_tableinit();
337 for (i = 0; i < 4; i++) {
338 if (ff_mpa_quant_bits[i] < 0) {
339 for (j = 0; j < (1 << (-ff_mpa_quant_bits[i]+1)); j++) {
340 int val1, val2, val3, steps;
342 steps = ff_mpa_quant_steps[i];
347 division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
353 for (i = 0; i < 7; i++) {
357 f = tan((double)i * M_PI / 12.0);
358 v = FIXR(f / (1.0 + f));
363 is_table[1][6 - i] = v;
366 for (i = 7; i < 16; i++)
367 is_table[0][i] = is_table[1][i] = 0.0;
369 for (i = 0; i < 16; i++) {
373 for (j = 0; j < 2; j++) {
374 e = -(j + 1) * ((i + 1) >> 1);
375 f = pow(2.0, e / 4.0);
377 is_table_lsf[j][k ^ 1][i] = FIXR(f);
378 is_table_lsf[j][k ][i] = FIXR(1.0);
379 ff_dlog(NULL, "is_table_lsf %d %d: %f %f\n",
380 i, j, (float) is_table_lsf[j][0][i],
381 (float) is_table_lsf[j][1][i]);
385 for (i = 0; i < 8; i++) {
388 cs = 1.0 / sqrt(1.0 + ci * ci);
391 csa_table[i][0] = FIXHR(cs/4);
392 csa_table[i][1] = FIXHR(ca/4);
393 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
394 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
396 csa_table[i][0] = cs;
397 csa_table[i][1] = ca;
398 csa_table[i][2] = ca + cs;
399 csa_table[i][3] = ca - cs;
404 static av_cold int decode_init(AVCodecContext * avctx)
406 static int initialized_tables = 0;
407 MPADecodeContext *s = avctx->priv_data;
409 if (!initialized_tables) {
410 decode_init_static();
411 initialized_tables = 1;
416 avpriv_float_dsp_init(&s->fdsp, avctx->flags & AV_CODEC_FLAG_BITEXACT);
417 ff_mpadsp_init(&s->mpadsp);
419 if (avctx->request_sample_fmt == OUT_FMT &&
420 avctx->codec_id != AV_CODEC_ID_MP3ON4)
421 avctx->sample_fmt = OUT_FMT;
423 avctx->sample_fmt = OUT_FMT_P;
424 s->err_recognition = avctx->err_recognition;
426 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
432 #define C3 FIXHR(0.86602540378443864676/2)
433 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
434 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
435 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
437 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
439 static void imdct12(INTFLOAT *out, INTFLOAT *in)
441 INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
444 in1 = in[1*3] + in[0*3];
445 in2 = in[2*3] + in[1*3];
446 in3 = in[3*3] + in[2*3];
447 in4 = in[4*3] + in[3*3];
448 in5 = in[5*3] + in[4*3];
452 in2 = MULH3(in2, C3, 2);
453 in3 = MULH3(in3, C3, 4);
456 t2 = MULH3(in1 - in5, C4, 2);
466 in1 = MULH3(in5 + in3, C5, 1);
473 in5 = MULH3(in5 - in3, C6, 2);
480 /* return the number of decoded frames */
481 static int mp_decode_layer1(MPADecodeContext *s)
483 int bound, i, v, n, ch, j, mant;
484 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
485 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
487 if (s->mode == MPA_JSTEREO)
488 bound = (s->mode_ext + 1) * 4;
492 /* allocation bits */
493 for (i = 0; i < bound; i++) {
494 for (ch = 0; ch < s->nb_channels; ch++) {
495 allocation[ch][i] = get_bits(&s->gb, 4);
498 for (i = bound; i < SBLIMIT; i++)
499 allocation[0][i] = get_bits(&s->gb, 4);
502 for (i = 0; i < bound; i++) {
503 for (ch = 0; ch < s->nb_channels; ch++) {
504 if (allocation[ch][i])
505 scale_factors[ch][i] = get_bits(&s->gb, 6);
508 for (i = bound; i < SBLIMIT; i++) {
509 if (allocation[0][i]) {
510 scale_factors[0][i] = get_bits(&s->gb, 6);
511 scale_factors[1][i] = get_bits(&s->gb, 6);
515 /* compute samples */
516 for (j = 0; j < 12; j++) {
517 for (i = 0; i < bound; i++) {
518 for (ch = 0; ch < s->nb_channels; ch++) {
519 n = allocation[ch][i];
521 mant = get_bits(&s->gb, n + 1);
522 v = l1_unscale(n, mant, scale_factors[ch][i]);
526 s->sb_samples[ch][j][i] = v;
529 for (i = bound; i < SBLIMIT; i++) {
530 n = allocation[0][i];
532 mant = get_bits(&s->gb, n + 1);
533 v = l1_unscale(n, mant, scale_factors[0][i]);
534 s->sb_samples[0][j][i] = v;
535 v = l1_unscale(n, mant, scale_factors[1][i]);
536 s->sb_samples[1][j][i] = v;
538 s->sb_samples[0][j][i] = 0;
539 s->sb_samples[1][j][i] = 0;
546 static int mp_decode_layer2(MPADecodeContext *s)
548 int sblimit; /* number of used subbands */
549 const unsigned char *alloc_table;
550 int table, bit_alloc_bits, i, j, ch, bound, v;
551 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
552 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
553 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
554 int scale, qindex, bits, steps, k, l, m, b;
556 /* select decoding table */
557 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
558 s->sample_rate, s->lsf);
559 sblimit = ff_mpa_sblimit_table[table];
560 alloc_table = ff_mpa_alloc_tables[table];
562 if (s->mode == MPA_JSTEREO)
563 bound = (s->mode_ext + 1) * 4;
567 ff_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
573 /* parse bit allocation */
575 for (i = 0; i < bound; i++) {
576 bit_alloc_bits = alloc_table[j];
577 for (ch = 0; ch < s->nb_channels; ch++)
578 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
579 j += 1 << bit_alloc_bits;
581 for (i = bound; i < sblimit; i++) {
582 bit_alloc_bits = alloc_table[j];
583 v = get_bits(&s->gb, bit_alloc_bits);
586 j += 1 << bit_alloc_bits;
590 for (i = 0; i < sblimit; i++) {
591 for (ch = 0; ch < s->nb_channels; ch++) {
592 if (bit_alloc[ch][i])
593 scale_code[ch][i] = get_bits(&s->gb, 2);
598 for (i = 0; i < sblimit; i++) {
599 for (ch = 0; ch < s->nb_channels; ch++) {
600 if (bit_alloc[ch][i]) {
601 sf = scale_factors[ch][i];
602 switch (scale_code[ch][i]) {
605 sf[0] = get_bits(&s->gb, 6);
606 sf[1] = get_bits(&s->gb, 6);
607 sf[2] = get_bits(&s->gb, 6);
610 sf[0] = get_bits(&s->gb, 6);
615 sf[0] = get_bits(&s->gb, 6);
616 sf[2] = get_bits(&s->gb, 6);
620 sf[0] = get_bits(&s->gb, 6);
621 sf[2] = get_bits(&s->gb, 6);
630 for (k = 0; k < 3; k++) {
631 for (l = 0; l < 12; l += 3) {
633 for (i = 0; i < bound; i++) {
634 bit_alloc_bits = alloc_table[j];
635 for (ch = 0; ch < s->nb_channels; ch++) {
636 b = bit_alloc[ch][i];
638 scale = scale_factors[ch][i][k];
639 qindex = alloc_table[j+b];
640 bits = ff_mpa_quant_bits[qindex];
643 /* 3 values at the same time */
644 v = get_bits(&s->gb, -bits);
645 v2 = division_tabs[qindex][v];
646 steps = ff_mpa_quant_steps[qindex];
648 s->sb_samples[ch][k * 12 + l + 0][i] =
649 l2_unscale_group(steps, v2 & 15, scale);
650 s->sb_samples[ch][k * 12 + l + 1][i] =
651 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
652 s->sb_samples[ch][k * 12 + l + 2][i] =
653 l2_unscale_group(steps, v2 >> 8 , scale);
655 for (m = 0; m < 3; m++) {
656 v = get_bits(&s->gb, bits);
657 v = l1_unscale(bits - 1, v, scale);
658 s->sb_samples[ch][k * 12 + l + m][i] = v;
662 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
663 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
664 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
667 /* next subband in alloc table */
668 j += 1 << bit_alloc_bits;
670 /* XXX: find a way to avoid this duplication of code */
671 for (i = bound; i < sblimit; i++) {
672 bit_alloc_bits = alloc_table[j];
675 int mant, scale0, scale1;
676 scale0 = scale_factors[0][i][k];
677 scale1 = scale_factors[1][i][k];
678 qindex = alloc_table[j+b];
679 bits = ff_mpa_quant_bits[qindex];
681 /* 3 values at the same time */
682 v = get_bits(&s->gb, -bits);
683 steps = ff_mpa_quant_steps[qindex];
686 s->sb_samples[0][k * 12 + l + 0][i] =
687 l2_unscale_group(steps, mant, scale0);
688 s->sb_samples[1][k * 12 + l + 0][i] =
689 l2_unscale_group(steps, mant, scale1);
692 s->sb_samples[0][k * 12 + l + 1][i] =
693 l2_unscale_group(steps, mant, scale0);
694 s->sb_samples[1][k * 12 + l + 1][i] =
695 l2_unscale_group(steps, mant, scale1);
696 s->sb_samples[0][k * 12 + l + 2][i] =
697 l2_unscale_group(steps, v, scale0);
698 s->sb_samples[1][k * 12 + l + 2][i] =
699 l2_unscale_group(steps, v, scale1);
701 for (m = 0; m < 3; m++) {
702 mant = get_bits(&s->gb, bits);
703 s->sb_samples[0][k * 12 + l + m][i] =
704 l1_unscale(bits - 1, mant, scale0);
705 s->sb_samples[1][k * 12 + l + m][i] =
706 l1_unscale(bits - 1, mant, scale1);
710 s->sb_samples[0][k * 12 + l + 0][i] = 0;
711 s->sb_samples[0][k * 12 + l + 1][i] = 0;
712 s->sb_samples[0][k * 12 + l + 2][i] = 0;
713 s->sb_samples[1][k * 12 + l + 0][i] = 0;
714 s->sb_samples[1][k * 12 + l + 1][i] = 0;
715 s->sb_samples[1][k * 12 + l + 2][i] = 0;
717 /* next subband in alloc table */
718 j += 1 << bit_alloc_bits;
720 /* fill remaining samples to zero */
721 for (i = sblimit; i < SBLIMIT; i++) {
722 for (ch = 0; ch < s->nb_channels; ch++) {
723 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
724 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
725 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
733 #define SPLIT(dst,sf,n) \
735 int m = (sf * 171) >> 9; \
738 } else if (n == 4) { \
741 } else if (n == 5) { \
742 int m = (sf * 205) >> 10; \
745 } else if (n == 6) { \
746 int m = (sf * 171) >> 10; \
753 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
756 SPLIT(slen[3], sf, n3)
757 SPLIT(slen[2], sf, n2)
758 SPLIT(slen[1], sf, n1)
762 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
765 const uint8_t *bstab, *pretab;
766 int len, i, j, k, l, v0, shift, gain, gains[3];
770 gain = g->global_gain - 210;
771 shift = g->scalefac_scale + 1;
773 bstab = band_size_long[s->sample_rate_index];
774 pretab = mpa_pretab[g->preflag];
775 for (i = 0; i < g->long_end; i++) {
776 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
778 for (j = len; j > 0; j--)
782 if (g->short_start < 13) {
783 bstab = band_size_short[s->sample_rate_index];
784 gains[0] = gain - (g->subblock_gain[0] << 3);
785 gains[1] = gain - (g->subblock_gain[1] << 3);
786 gains[2] = gain - (g->subblock_gain[2] << 3);
788 for (i = g->short_start; i < 13; i++) {
790 for (l = 0; l < 3; l++) {
791 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
792 for (j = len; j > 0; j--)
799 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
802 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits - s->extrasize * 8) {
804 s->in_gb.buffer = NULL;
806 assert((get_bits_count(&s->gb) & 7) == 0);
807 skip_bits_long(&s->gb, *pos - *end_pos);
809 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
810 *pos = get_bits_count(&s->gb);
814 /* Following is a optimized code for
816 if(get_bits1(&s->gb))
821 #define READ_FLIP_SIGN(dst,src) \
822 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
825 #define READ_FLIP_SIGN(dst,src) \
826 v = -get_bits1(&s->gb); \
827 *(dst) = (*(src) ^ v) - v;
830 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
831 int16_t *exponents, int end_pos2)
835 int last_pos, bits_left;
837 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits - s->extrasize * 8);
839 /* low frequencies (called big values) */
841 for (i = 0; i < 3; i++) {
842 int j, k, l, linbits;
843 j = g->region_size[i];
846 /* select vlc table */
847 k = g->table_select[i];
848 l = mpa_huff_data[k][0];
849 linbits = mpa_huff_data[k][1];
853 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
858 /* read huffcode and compute each couple */
862 int pos = get_bits_count(&s->gb);
865 switch_buffer(s, &pos, &end_pos, &end_pos2);
869 y = get_vlc2(&s->gb, vlc->table, 7, 3);
872 g->sb_hybrid[s_index ] =
873 g->sb_hybrid[s_index+1] = 0;
878 exponent= exponents[s_index];
880 ff_dlog(s->avctx, "region=%d n=%d y=%d exp=%d\n",
881 i, g->region_size[i] - j, y, exponent);
886 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
888 x += get_bitsz(&s->gb, linbits);
889 v = l3_unscale(x, exponent);
890 if (get_bits1(&s->gb))
892 g->sb_hybrid[s_index] = v;
895 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
897 y += get_bitsz(&s->gb, linbits);
898 v = l3_unscale(y, exponent);
899 if (get_bits1(&s->gb))
901 g->sb_hybrid[s_index+1] = v;
908 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
910 x += get_bitsz(&s->gb, linbits);
911 v = l3_unscale(x, exponent);
912 if (get_bits1(&s->gb))
914 g->sb_hybrid[s_index+!!y] = v;
916 g->sb_hybrid[s_index + !y] = 0;
922 /* high frequencies */
923 vlc = &huff_quad_vlc[g->count1table_select];
925 while (s_index <= 572) {
927 pos = get_bits_count(&s->gb);
928 if (pos >= end_pos) {
929 if (pos > end_pos2 && last_pos) {
930 /* some encoders generate an incorrect size for this
931 part. We must go back into the data */
933 skip_bits_long(&s->gb, last_pos - pos);
934 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
935 if(s->err_recognition & AV_EF_BITSTREAM)
939 switch_buffer(s, &pos, &end_pos, &end_pos2);
945 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
946 ff_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
947 g->sb_hybrid[s_index+0] =
948 g->sb_hybrid[s_index+1] =
949 g->sb_hybrid[s_index+2] =
950 g->sb_hybrid[s_index+3] = 0;
952 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
954 int pos = s_index + idxtab[code];
955 code ^= 8 >> idxtab[code];
956 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
960 /* skip extension bits */
961 bits_left = end_pos2 - get_bits_count(&s->gb);
962 if (bits_left < 0 && (s->err_recognition & AV_EF_BUFFER)) {
963 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
965 } else if (bits_left > 0 && (s->err_recognition & AV_EF_BUFFER)) {
966 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
969 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
970 skip_bits_long(&s->gb, bits_left);
972 i = get_bits_count(&s->gb);
973 switch_buffer(s, &i, &end_pos, &end_pos2);
978 /* Reorder short blocks from bitstream order to interleaved order. It
979 would be faster to do it in parsing, but the code would be far more
981 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
984 INTFLOAT *ptr, *dst, *ptr1;
987 if (g->block_type != 2)
990 if (g->switch_point) {
991 if (s->sample_rate_index != 8)
992 ptr = g->sb_hybrid + 36;
994 ptr = g->sb_hybrid + 72;
999 for (i = g->short_start; i < 13; i++) {
1000 len = band_size_short[s->sample_rate_index][i];
1003 for (j = len; j > 0; j--) {
1004 *dst++ = ptr[0*len];
1005 *dst++ = ptr[1*len];
1006 *dst++ = ptr[2*len];
1010 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1014 #define ISQRT2 FIXR(0.70710678118654752440)
1016 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1019 int sf_max, sf, len, non_zero_found;
1020 INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1021 int non_zero_found_short[3];
1023 /* intensity stereo */
1024 if (s->mode_ext & MODE_EXT_I_STEREO) {
1029 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1033 tab0 = g0->sb_hybrid + 576;
1034 tab1 = g1->sb_hybrid + 576;
1036 non_zero_found_short[0] = 0;
1037 non_zero_found_short[1] = 0;
1038 non_zero_found_short[2] = 0;
1039 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1040 for (i = 12; i >= g1->short_start; i--) {
1041 /* for last band, use previous scale factor */
1044 len = band_size_short[s->sample_rate_index][i];
1045 for (l = 2; l >= 0; l--) {
1048 if (!non_zero_found_short[l]) {
1049 /* test if non zero band. if so, stop doing i-stereo */
1050 for (j = 0; j < len; j++) {
1052 non_zero_found_short[l] = 1;
1056 sf = g1->scale_factors[k + l];
1062 for (j = 0; j < len; j++) {
1064 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1065 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1069 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1070 /* lower part of the spectrum : do ms stereo
1072 for (j = 0; j < len; j++) {
1075 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1076 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1083 non_zero_found = non_zero_found_short[0] |
1084 non_zero_found_short[1] |
1085 non_zero_found_short[2];
1087 for (i = g1->long_end - 1;i >= 0;i--) {
1088 len = band_size_long[s->sample_rate_index][i];
1091 /* test if non zero band. if so, stop doing i-stereo */
1092 if (!non_zero_found) {
1093 for (j = 0; j < len; j++) {
1099 /* for last band, use previous scale factor */
1100 k = (i == 21) ? 20 : i;
1101 sf = g1->scale_factors[k];
1106 for (j = 0; j < len; j++) {
1108 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1109 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1113 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1114 /* lower part of the spectrum : do ms stereo
1116 for (j = 0; j < len; j++) {
1119 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1120 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1125 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1126 /* ms stereo ONLY */
1127 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1130 s->fdsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1132 tab0 = g0->sb_hybrid;
1133 tab1 = g1->sb_hybrid;
1134 for (i = 0; i < 576; i++) {
1137 tab0[i] = tmp0 + tmp1;
1138 tab1[i] = tmp0 - tmp1;
1145 #define AA(j) do { \
1146 float tmp0 = ptr[-1-j]; \
1147 float tmp1 = ptr[ j]; \
1148 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1149 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1152 #define AA(j) do { \
1153 int tmp0 = ptr[-1-j]; \
1154 int tmp1 = ptr[ j]; \
1155 int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1156 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1157 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1161 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1166 /* we antialias only "long" bands */
1167 if (g->block_type == 2) {
1168 if (!g->switch_point)
1170 /* XXX: check this for 8000Hz case */
1176 ptr = g->sb_hybrid + 18;
1177 for (i = n; i > 0; i--) {
1191 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1192 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1194 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1196 int i, j, mdct_long_end, sblimit;
1198 /* find last non zero block */
1199 ptr = g->sb_hybrid + 576;
1200 ptr1 = g->sb_hybrid + 2 * 18;
1201 while (ptr >= ptr1) {
1205 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1208 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1210 if (g->block_type == 2) {
1211 /* XXX: check for 8000 Hz */
1212 if (g->switch_point)
1217 mdct_long_end = sblimit;
1220 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1221 mdct_long_end, g->switch_point,
1224 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1225 ptr = g->sb_hybrid + 18 * mdct_long_end;
1227 for (j = mdct_long_end; j < sblimit; j++) {
1228 /* select frequency inversion */
1229 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1230 out_ptr = sb_samples + j;
1232 for (i = 0; i < 6; i++) {
1233 *out_ptr = buf[4*i];
1236 imdct12(out2, ptr + 0);
1237 for (i = 0; i < 6; i++) {
1238 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1239 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1242 imdct12(out2, ptr + 1);
1243 for (i = 0; i < 6; i++) {
1244 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1245 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1248 imdct12(out2, ptr + 2);
1249 for (i = 0; i < 6; i++) {
1250 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1251 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1252 buf[4*(i + 6*2)] = 0;
1255 buf += (j&3) != 3 ? 1 : (4*18-3);
1258 for (j = sblimit; j < SBLIMIT; j++) {
1260 out_ptr = sb_samples + j;
1261 for (i = 0; i < 18; i++) {
1262 *out_ptr = buf[4*i];
1266 buf += (j&3) != 3 ? 1 : (4*18-3);
1270 /* main layer3 decoding function */
1271 static int mp_decode_layer3(MPADecodeContext *s)
1273 int nb_granules, main_data_begin;
1274 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1276 int16_t exponents[576]; //FIXME try INTFLOAT
1278 /* read side info */
1280 main_data_begin = get_bits(&s->gb, 8);
1281 skip_bits(&s->gb, s->nb_channels);
1284 main_data_begin = get_bits(&s->gb, 9);
1285 if (s->nb_channels == 2)
1286 skip_bits(&s->gb, 3);
1288 skip_bits(&s->gb, 5);
1290 for (ch = 0; ch < s->nb_channels; ch++) {
1291 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1292 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1296 for (gr = 0; gr < nb_granules; gr++) {
1297 for (ch = 0; ch < s->nb_channels; ch++) {
1298 ff_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1299 g = &s->granules[ch][gr];
1300 g->part2_3_length = get_bits(&s->gb, 12);
1301 g->big_values = get_bits(&s->gb, 9);
1302 if (g->big_values > 288) {
1303 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1304 return AVERROR_INVALIDDATA;
1307 g->global_gain = get_bits(&s->gb, 8);
1308 /* if MS stereo only is selected, we precompute the
1309 1/sqrt(2) renormalization factor */
1310 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1312 g->global_gain -= 2;
1314 g->scalefac_compress = get_bits(&s->gb, 9);
1316 g->scalefac_compress = get_bits(&s->gb, 4);
1317 blocksplit_flag = get_bits1(&s->gb);
1318 if (blocksplit_flag) {
1319 g->block_type = get_bits(&s->gb, 2);
1320 if (g->block_type == 0) {
1321 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1322 return AVERROR_INVALIDDATA;
1324 g->switch_point = get_bits1(&s->gb);
1325 for (i = 0; i < 2; i++)
1326 g->table_select[i] = get_bits(&s->gb, 5);
1327 for (i = 0; i < 3; i++)
1328 g->subblock_gain[i] = get_bits(&s->gb, 3);
1329 init_short_region(s, g);
1331 int region_address1, region_address2;
1333 g->switch_point = 0;
1334 for (i = 0; i < 3; i++)
1335 g->table_select[i] = get_bits(&s->gb, 5);
1336 /* compute huffman coded region sizes */
1337 region_address1 = get_bits(&s->gb, 4);
1338 region_address2 = get_bits(&s->gb, 3);
1339 ff_dlog(s->avctx, "region1=%d region2=%d\n",
1340 region_address1, region_address2);
1341 init_long_region(s, g, region_address1, region_address2);
1343 region_offset2size(g);
1344 compute_band_indexes(s, g);
1348 g->preflag = get_bits1(&s->gb);
1349 g->scalefac_scale = get_bits1(&s->gb);
1350 g->count1table_select = get_bits1(&s->gb);
1351 ff_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1352 g->block_type, g->switch_point);
1358 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1359 s->extrasize = av_clip((get_bits_left(&s->gb) >> 3) - s->extrasize, 0,
1360 FFMAX(0, LAST_BUF_SIZE - s->last_buf_size));
1361 assert((get_bits_count(&s->gb) & 7) == 0);
1362 /* now we get bits from the main_data_begin offset */
1363 ff_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1364 main_data_begin, s->last_buf_size);
1366 memcpy(s->last_buf + s->last_buf_size, ptr, s->extrasize);
1368 init_get_bits(&s->gb, s->last_buf, (s->last_buf_size + s->extrasize) * 8);
1369 s->last_buf_size <<= 3;
1370 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1371 for (ch = 0; ch < s->nb_channels; ch++) {
1372 g = &s->granules[ch][gr];
1373 s->last_buf_size += g->part2_3_length;
1374 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1375 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1378 skip = s->last_buf_size - 8 * main_data_begin;
1379 if (skip >= s->gb.size_in_bits - s->extrasize * 8 && s->in_gb.buffer) {
1380 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits + s->extrasize * 8);
1382 s->in_gb.buffer = NULL;
1385 skip_bits_long(&s->gb, skip);
1392 for (; gr < nb_granules; gr++) {
1393 for (ch = 0; ch < s->nb_channels; ch++) {
1394 g = &s->granules[ch][gr];
1395 bits_pos = get_bits_count(&s->gb);
1399 int slen, slen1, slen2;
1401 /* MPEG-1 scale factors */
1402 slen1 = slen_table[0][g->scalefac_compress];
1403 slen2 = slen_table[1][g->scalefac_compress];
1404 ff_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1405 if (g->block_type == 2) {
1406 n = g->switch_point ? 17 : 18;
1409 for (i = 0; i < n; i++)
1410 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1412 for (i = 0; i < n; i++)
1413 g->scale_factors[j++] = 0;
1416 for (i = 0; i < 18; i++)
1417 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1418 for (i = 0; i < 3; i++)
1419 g->scale_factors[j++] = 0;
1421 for (i = 0; i < 21; i++)
1422 g->scale_factors[j++] = 0;
1425 sc = s->granules[ch][0].scale_factors;
1427 for (k = 0; k < 4; k++) {
1429 if ((g->scfsi & (0x8 >> k)) == 0) {
1430 slen = (k < 2) ? slen1 : slen2;
1432 for (i = 0; i < n; i++)
1433 g->scale_factors[j++] = get_bits(&s->gb, slen);
1435 for (i = 0; i < n; i++)
1436 g->scale_factors[j++] = 0;
1439 /* simply copy from last granule */
1440 for (i = 0; i < n; i++) {
1441 g->scale_factors[j] = sc[j];
1446 g->scale_factors[j++] = 0;
1449 int tindex, tindex2, slen[4], sl, sf;
1451 /* LSF scale factors */
1452 if (g->block_type == 2)
1453 tindex = g->switch_point ? 2 : 1;
1457 sf = g->scalefac_compress;
1458 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1459 /* intensity stereo case */
1462 lsf_sf_expand(slen, sf, 6, 6, 0);
1464 } else if (sf < 244) {
1465 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1468 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1474 lsf_sf_expand(slen, sf, 5, 4, 4);
1476 } else if (sf < 500) {
1477 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1480 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1487 for (k = 0; k < 4; k++) {
1488 n = lsf_nsf_table[tindex2][tindex][k];
1491 for (i = 0; i < n; i++)
1492 g->scale_factors[j++] = get_bits(&s->gb, sl);
1494 for (i = 0; i < n; i++)
1495 g->scale_factors[j++] = 0;
1498 /* XXX: should compute exact size */
1500 g->scale_factors[j] = 0;
1503 exponents_from_scale_factors(s, g, exponents);
1505 /* read Huffman coded residue */
1506 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1509 if (s->mode == MPA_JSTEREO)
1510 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1512 for (ch = 0; ch < s->nb_channels; ch++) {
1513 g = &s->granules[ch][gr];
1515 reorder_block(s, g);
1516 compute_antialias(s, g);
1517 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1520 if (get_bits_count(&s->gb) < 0)
1521 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1522 return nb_granules * 18;
1525 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1526 const uint8_t *buf, int buf_size)
1528 int i, nb_frames, ch, ret;
1529 OUT_INT *samples_ptr;
1531 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1533 /* skip error protection field */
1534 if (s->error_protection)
1535 skip_bits(&s->gb, 16);
1539 s->avctx->frame_size = 384;
1540 nb_frames = mp_decode_layer1(s);
1543 s->avctx->frame_size = 1152;
1544 nb_frames = mp_decode_layer2(s);
1547 s->avctx->frame_size = s->lsf ? 576 : 1152;
1549 nb_frames = mp_decode_layer3(s);
1555 if (s->in_gb.buffer) {
1556 align_get_bits(&s->gb);
1557 i = (get_bits_left(&s->gb) >> 3) - s->extrasize;
1558 if (i >= 0 && i <= BACKSTEP_SIZE) {
1559 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1562 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1564 s->in_gb.buffer = NULL;
1568 align_get_bits(&s->gb);
1569 assert((get_bits_count(&s->gb) & 7) == 0);
1570 i = (get_bits_left(&s->gb) >> 3) - s->extrasize;
1571 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1573 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1574 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1576 assert(i <= buf_size - HEADER_SIZE && i >= 0);
1577 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1578 s->last_buf_size += i;
1581 /* get output buffer */
1583 av_assert0(s->frame != NULL);
1584 s->frame->nb_samples = s->avctx->frame_size;
1585 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0) {
1586 av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1589 samples = (OUT_INT **)s->frame->extended_data;
1592 /* apply the synthesis filter */
1593 for (ch = 0; ch < s->nb_channels; ch++) {
1595 if (s->avctx->sample_fmt == OUT_FMT_P) {
1596 samples_ptr = samples[ch];
1599 samples_ptr = samples[0] + ch;
1600 sample_stride = s->nb_channels;
1602 for (i = 0; i < nb_frames; i++) {
1603 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1604 &(s->synth_buf_offset[ch]),
1605 RENAME(ff_mpa_synth_window),
1606 &s->dither_state, samples_ptr,
1607 sample_stride, s->sb_samples[ch][i]);
1608 samples_ptr += 32 * sample_stride;
1612 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1615 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1618 const uint8_t *buf = avpkt->data;
1619 int buf_size = avpkt->size;
1620 MPADecodeContext *s = avctx->priv_data;
1624 if (buf_size < HEADER_SIZE)
1625 return AVERROR_INVALIDDATA;
1627 header = AV_RB32(buf);
1629 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1631 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1632 return AVERROR_INVALIDDATA;
1633 } else if (ret == 1) {
1634 /* free format: prepare to compute frame size */
1636 return AVERROR_INVALIDDATA;
1638 /* update codec info */
1639 avctx->channels = s->nb_channels;
1640 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1641 if (!avctx->bit_rate)
1642 avctx->bit_rate = s->bit_rate;
1646 ret = mp_decode_frame(s, NULL, buf, buf_size);
1648 s->frame->nb_samples = avctx->frame_size;
1650 avctx->sample_rate = s->sample_rate;
1651 //FIXME maybe move the other codec info stuff from above here too
1653 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1654 /* Only return an error if the bad frame makes up the whole packet or
1655 * the error is related to buffer management.
1656 * If there is more data in the packet, just consume the bad frame
1657 * instead of returning an error, which would discard the whole
1660 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1667 static void mp_flush(MPADecodeContext *ctx)
1669 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1670 ctx->last_buf_size = 0;
1673 static void flush(AVCodecContext *avctx)
1675 mp_flush(avctx->priv_data);
1678 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1679 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1680 int *got_frame_ptr, AVPacket *avpkt)
1682 const uint8_t *buf = avpkt->data;
1683 int buf_size = avpkt->size;
1684 MPADecodeContext *s = avctx->priv_data;
1690 // Discard too short frames
1691 if (buf_size < HEADER_SIZE) {
1692 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1693 return AVERROR_INVALIDDATA;
1697 if (len > MPA_MAX_CODED_FRAME_SIZE)
1698 len = MPA_MAX_CODED_FRAME_SIZE;
1700 // Get header and restore sync word
1701 header = AV_RB32(buf) | 0xffe00000;
1703 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1705 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1708 /* update codec info */
1709 avctx->sample_rate = s->sample_rate;
1710 avctx->channels = s->nb_channels;
1711 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1712 if (!avctx->bit_rate)
1713 avctx->bit_rate = s->bit_rate;
1715 s->frame_size = len;
1719 ret = mp_decode_frame(s, NULL, buf, buf_size);
1721 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1729 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1731 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1734 * Context for MP3On4 decoder
1736 typedef struct MP3On4DecodeContext {
1737 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1738 int syncword; ///< syncword patch
1739 const uint8_t *coff; ///< channel offsets in output buffer
1740 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1741 } MP3On4DecodeContext;
1743 #include "mpeg4audio.h"
1745 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1747 /* number of mp3 decoder instances */
1748 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1750 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1751 static const uint8_t chan_offset[8][5] = {
1756 { 2, 0, 3 }, // C FLR BS
1757 { 2, 0, 3 }, // C FLR BLRS
1758 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1759 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1762 /* mp3on4 channel layouts */
1763 static const int16_t chan_layout[8] = {
1766 AV_CH_LAYOUT_STEREO,
1767 AV_CH_LAYOUT_SURROUND,
1768 AV_CH_LAYOUT_4POINT0,
1769 AV_CH_LAYOUT_5POINT0,
1770 AV_CH_LAYOUT_5POINT1,
1771 AV_CH_LAYOUT_7POINT1
1774 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1776 MP3On4DecodeContext *s = avctx->priv_data;
1779 for (i = 0; i < s->frames; i++)
1780 av_free(s->mp3decctx[i]);
1786 static av_cold int decode_init_mp3on4(AVCodecContext * avctx)
1788 MP3On4DecodeContext *s = avctx->priv_data;
1789 MPEG4AudioConfig cfg;
1792 if ((avctx->extradata_size < 2) || !avctx->extradata) {
1793 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1794 return AVERROR_INVALIDDATA;
1797 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
1798 avctx->extradata_size * 8, 1);
1799 if (!cfg.chan_config || cfg.chan_config > 7) {
1800 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1801 return AVERROR_INVALIDDATA;
1803 s->frames = mp3Frames[cfg.chan_config];
1804 s->coff = chan_offset[cfg.chan_config];
1805 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1806 avctx->channel_layout = chan_layout[cfg.chan_config];
1808 if (cfg.sample_rate < 16000)
1809 s->syncword = 0xffe00000;
1811 s->syncword = 0xfff00000;
1813 /* Init the first mp3 decoder in standard way, so that all tables get builded
1814 * We replace avctx->priv_data with the context of the first decoder so that
1815 * decode_init() does not have to be changed.
1816 * Other decoders will be initialized here copying data from the first context
1818 // Allocate zeroed memory for the first decoder context
1819 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1820 if (!s->mp3decctx[0])
1822 // Put decoder context in place to make init_decode() happy
1823 avctx->priv_data = s->mp3decctx[0];
1825 // Restore mp3on4 context pointer
1826 avctx->priv_data = s;
1827 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1829 /* Create a separate codec/context for each frame (first is already ok).
1830 * Each frame is 1 or 2 channels - up to 5 frames allowed
1832 for (i = 1; i < s->frames; i++) {
1833 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1834 if (!s->mp3decctx[i])
1836 s->mp3decctx[i]->adu_mode = 1;
1837 s->mp3decctx[i]->avctx = avctx;
1838 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1843 decode_close_mp3on4(avctx);
1844 return AVERROR(ENOMEM);
1848 static void flush_mp3on4(AVCodecContext *avctx)
1851 MP3On4DecodeContext *s = avctx->priv_data;
1853 for (i = 0; i < s->frames; i++)
1854 mp_flush(s->mp3decctx[i]);
1858 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1859 int *got_frame_ptr, AVPacket *avpkt)
1861 AVFrame *frame = data;
1862 const uint8_t *buf = avpkt->data;
1863 int buf_size = avpkt->size;
1864 MP3On4DecodeContext *s = avctx->priv_data;
1865 MPADecodeContext *m;
1866 int fsize, len = buf_size, out_size = 0;
1868 OUT_INT **out_samples;
1872 /* get output buffer */
1873 frame->nb_samples = MPA_FRAME_SIZE;
1874 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1875 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
1878 out_samples = (OUT_INT **)frame->extended_data;
1880 // Discard too short frames
1881 if (buf_size < HEADER_SIZE)
1882 return AVERROR_INVALIDDATA;
1884 avctx->bit_rate = 0;
1887 for (fr = 0; fr < s->frames; fr++) {
1888 fsize = AV_RB16(buf) >> 4;
1889 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1890 m = s->mp3decctx[fr];
1893 if (fsize < HEADER_SIZE) {
1894 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1895 return AVERROR_INVALIDDATA;
1897 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1899 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1900 if (ret < 0) // Bad header, discard block
1903 if (ch + m->nb_channels > avctx->channels ||
1904 s->coff[fr] + m->nb_channels > avctx->channels) {
1905 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1907 return AVERROR_INVALIDDATA;
1909 ch += m->nb_channels;
1911 outptr[0] = out_samples[s->coff[fr]];
1912 if (m->nb_channels > 1)
1913 outptr[1] = out_samples[s->coff[fr] + 1];
1915 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0)
1922 avctx->bit_rate += m->bit_rate;
1925 /* update codec info */
1926 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1928 frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1933 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */