3 * Copyright (c) 2001, 2002 Fabrice Bellard
5 * This file is part of FFmpeg.
7 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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"
31 #include "libavutil/libm.h"
36 #include "mpegaudiodsp.h"
40 * - test lsf / mpeg25 extensively.
43 #include "mpegaudio.h"
44 #include "mpegaudiodecheader.h"
46 #define BACKSTEP_SIZE 512
48 #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES
50 /* layer 3 "granule" */
51 typedef struct GranuleDef {
56 int scalefac_compress;
61 uint8_t scalefac_scale;
62 uint8_t count1table_select;
63 int region_size[3]; /* number of huffman codes in each region */
65 int short_start, long_end; /* long/short band indexes */
66 uint8_t scale_factors[40];
67 DECLARE_ALIGNED(16, INTFLOAT, sb_hybrid)[SBLIMIT * 18]; /* 576 samples */
70 typedef struct MPADecodeContext {
72 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(s->sample_rate_index == 8)
189 avpriv_request_sample(s->avctx, "switch point in 8khz");
190 /* if switched mode, we handle the 36 first samples as
191 long blocks. For 8000Hz, we handle the 72 first
192 exponents as long blocks */
193 if (s->sample_rate_index <= 2)
209 /* layer 1 unscaling */
210 /* n = number of bits of the mantissa minus 1 */
211 static inline int l1_unscale(int n, int mant, int scale_factor)
216 shift = scale_factor_modshift[scale_factor];
219 val = MUL64(mant + (-1 << n) + 1, scale_factor_mult[n-1][mod]);
221 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
222 return (int)((val + (1LL << (shift - 1))) >> shift);
225 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
229 shift = scale_factor_modshift[scale_factor];
233 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
234 /* NOTE: at this point, 0 <= shift <= 21 */
236 val = (val + (1 << (shift - 1))) >> shift;
240 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
241 static inline int l3_unscale(int value, int exponent)
246 e = table_4_3_exp [4 * value + (exponent & 3)];
247 m = table_4_3_value[4 * value + (exponent & 3)];
251 av_log(NULL, AV_LOG_WARNING, "l3_unscale: e is %d\n", e);
255 m = (m + (1 << (e - 1))) >> e;
260 static av_cold void decode_init_static(void)
265 /* scale factors table for layer 1/2 */
266 for (i = 0; i < 64; i++) {
268 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
271 scale_factor_modshift[i] = mod | (shift << 2);
274 /* scale factor multiply for layer 1 */
275 for (i = 0; i < 15; i++) {
278 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
279 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
280 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
281 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
282 av_dlog(NULL, "%d: norm=%x s=%x %x %x\n", i, norm,
283 scale_factor_mult[i][0],
284 scale_factor_mult[i][1],
285 scale_factor_mult[i][2]);
288 RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
290 /* huffman decode tables */
292 for (i = 1; i < 16; i++) {
293 const HuffTable *h = &mpa_huff_tables[i];
295 uint8_t tmp_bits [512] = { 0 };
296 uint16_t tmp_codes[512] = { 0 };
301 for (x = 0; x < xsize; x++) {
302 for (y = 0; y < xsize; y++) {
303 tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ];
304 tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
309 huff_vlc[i].table = huff_vlc_tables+offset;
310 huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
311 init_vlc(&huff_vlc[i], 7, 512,
312 tmp_bits, 1, 1, tmp_codes, 2, 2,
313 INIT_VLC_USE_NEW_STATIC);
314 offset += huff_vlc_tables_sizes[i];
316 av_assert0(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
319 for (i = 0; i < 2; i++) {
320 huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
321 huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
322 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
323 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
324 INIT_VLC_USE_NEW_STATIC);
325 offset += huff_quad_vlc_tables_sizes[i];
327 av_assert0(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
329 for (i = 0; i < 9; i++) {
331 for (j = 0; j < 22; j++) {
332 band_index_long[i][j] = k;
333 k += band_size_long[i][j];
335 band_index_long[i][22] = k;
338 /* compute n ^ (4/3) and store it in mantissa/exp format */
340 mpegaudio_tableinit();
342 for (i = 0; i < 4; i++) {
343 if (ff_mpa_quant_bits[i] < 0) {
344 for (j = 0; j < (1 << (-ff_mpa_quant_bits[i]+1)); j++) {
345 int val1, val2, val3, steps;
347 steps = ff_mpa_quant_steps[i];
352 division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
358 for (i = 0; i < 7; i++) {
362 f = tan((double)i * M_PI / 12.0);
363 v = FIXR(f / (1.0 + f));
368 is_table[1][6 - i] = v;
371 for (i = 7; i < 16; i++)
372 is_table[0][i] = is_table[1][i] = 0.0;
374 for (i = 0; i < 16; i++) {
378 for (j = 0; j < 2; j++) {
379 e = -(j + 1) * ((i + 1) >> 1);
382 is_table_lsf[j][k ^ 1][i] = FIXR(f);
383 is_table_lsf[j][k ][i] = FIXR(1.0);
384 av_dlog(NULL, "is_table_lsf %d %d: %f %f\n",
385 i, j, (float) is_table_lsf[j][0][i],
386 (float) is_table_lsf[j][1][i]);
390 for (i = 0; i < 8; i++) {
393 cs = 1.0 / sqrt(1.0 + ci * ci);
396 csa_table[i][0] = FIXHR(cs/4);
397 csa_table[i][1] = FIXHR(ca/4);
398 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
399 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
401 csa_table[i][0] = cs;
402 csa_table[i][1] = ca;
403 csa_table[i][2] = ca + cs;
404 csa_table[i][3] = ca - cs;
409 static av_cold int decode_init(AVCodecContext * avctx)
411 static int initialized_tables = 0;
412 MPADecodeContext *s = avctx->priv_data;
414 if (!initialized_tables) {
415 decode_init_static();
416 initialized_tables = 1;
421 avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
422 ff_mpadsp_init(&s->mpadsp);
424 if (avctx->request_sample_fmt == OUT_FMT &&
425 avctx->codec_id != AV_CODEC_ID_MP3ON4)
426 avctx->sample_fmt = OUT_FMT;
428 avctx->sample_fmt = OUT_FMT_P;
429 s->err_recognition = avctx->err_recognition;
431 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
437 #define C3 FIXHR(0.86602540378443864676/2)
438 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
439 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
440 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
442 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
444 static void imdct12(INTFLOAT *out, INTFLOAT *in)
446 INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
449 in1 = in[1*3] + in[0*3];
450 in2 = in[2*3] + in[1*3];
451 in3 = in[3*3] + in[2*3];
452 in4 = in[4*3] + in[3*3];
453 in5 = in[5*3] + in[4*3];
457 in2 = MULH3(in2, C3, 2);
458 in3 = MULH3(in3, C3, 4);
461 t2 = MULH3(in1 - in5, C4, 2);
471 in1 = MULH3(in5 + in3, C5, 1);
478 in5 = MULH3(in5 - in3, C6, 2);
485 /* return the number of decoded frames */
486 static int mp_decode_layer1(MPADecodeContext *s)
488 int bound, i, v, n, ch, j, mant;
489 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
490 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
492 if (s->mode == MPA_JSTEREO)
493 bound = (s->mode_ext + 1) * 4;
497 /* allocation bits */
498 for (i = 0; i < bound; i++) {
499 for (ch = 0; ch < s->nb_channels; ch++) {
500 allocation[ch][i] = get_bits(&s->gb, 4);
503 for (i = bound; i < SBLIMIT; i++)
504 allocation[0][i] = get_bits(&s->gb, 4);
507 for (i = 0; i < bound; i++) {
508 for (ch = 0; ch < s->nb_channels; ch++) {
509 if (allocation[ch][i])
510 scale_factors[ch][i] = get_bits(&s->gb, 6);
513 for (i = bound; i < SBLIMIT; i++) {
514 if (allocation[0][i]) {
515 scale_factors[0][i] = get_bits(&s->gb, 6);
516 scale_factors[1][i] = get_bits(&s->gb, 6);
520 /* compute samples */
521 for (j = 0; j < 12; j++) {
522 for (i = 0; i < bound; i++) {
523 for (ch = 0; ch < s->nb_channels; ch++) {
524 n = allocation[ch][i];
526 mant = get_bits(&s->gb, n + 1);
527 v = l1_unscale(n, mant, scale_factors[ch][i]);
531 s->sb_samples[ch][j][i] = v;
534 for (i = bound; i < SBLIMIT; i++) {
535 n = allocation[0][i];
537 mant = get_bits(&s->gb, n + 1);
538 v = l1_unscale(n, mant, scale_factors[0][i]);
539 s->sb_samples[0][j][i] = v;
540 v = l1_unscale(n, mant, scale_factors[1][i]);
541 s->sb_samples[1][j][i] = v;
543 s->sb_samples[0][j][i] = 0;
544 s->sb_samples[1][j][i] = 0;
551 static int mp_decode_layer2(MPADecodeContext *s)
553 int sblimit; /* number of used subbands */
554 const unsigned char *alloc_table;
555 int table, bit_alloc_bits, i, j, ch, bound, v;
556 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
557 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
558 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
559 int scale, qindex, bits, steps, k, l, m, b;
561 /* select decoding table */
562 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
563 s->sample_rate, s->lsf);
564 sblimit = ff_mpa_sblimit_table[table];
565 alloc_table = ff_mpa_alloc_tables[table];
567 if (s->mode == MPA_JSTEREO)
568 bound = (s->mode_ext + 1) * 4;
572 av_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
578 /* parse bit allocation */
580 for (i = 0; i < bound; i++) {
581 bit_alloc_bits = alloc_table[j];
582 for (ch = 0; ch < s->nb_channels; ch++)
583 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
584 j += 1 << bit_alloc_bits;
586 for (i = bound; i < sblimit; i++) {
587 bit_alloc_bits = alloc_table[j];
588 v = get_bits(&s->gb, bit_alloc_bits);
591 j += 1 << bit_alloc_bits;
595 for (i = 0; i < sblimit; i++) {
596 for (ch = 0; ch < s->nb_channels; ch++) {
597 if (bit_alloc[ch][i])
598 scale_code[ch][i] = get_bits(&s->gb, 2);
603 for (i = 0; i < sblimit; i++) {
604 for (ch = 0; ch < s->nb_channels; ch++) {
605 if (bit_alloc[ch][i]) {
606 sf = scale_factors[ch][i];
607 switch (scale_code[ch][i]) {
610 sf[0] = get_bits(&s->gb, 6);
611 sf[1] = get_bits(&s->gb, 6);
612 sf[2] = get_bits(&s->gb, 6);
615 sf[0] = get_bits(&s->gb, 6);
620 sf[0] = get_bits(&s->gb, 6);
621 sf[2] = get_bits(&s->gb, 6);
625 sf[0] = get_bits(&s->gb, 6);
626 sf[2] = get_bits(&s->gb, 6);
635 for (k = 0; k < 3; k++) {
636 for (l = 0; l < 12; l += 3) {
638 for (i = 0; i < bound; i++) {
639 bit_alloc_bits = alloc_table[j];
640 for (ch = 0; ch < s->nb_channels; ch++) {
641 b = bit_alloc[ch][i];
643 scale = scale_factors[ch][i][k];
644 qindex = alloc_table[j+b];
645 bits = ff_mpa_quant_bits[qindex];
648 /* 3 values at the same time */
649 v = get_bits(&s->gb, -bits);
650 v2 = division_tabs[qindex][v];
651 steps = ff_mpa_quant_steps[qindex];
653 s->sb_samples[ch][k * 12 + l + 0][i] =
654 l2_unscale_group(steps, v2 & 15, scale);
655 s->sb_samples[ch][k * 12 + l + 1][i] =
656 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
657 s->sb_samples[ch][k * 12 + l + 2][i] =
658 l2_unscale_group(steps, v2 >> 8 , scale);
660 for (m = 0; m < 3; m++) {
661 v = get_bits(&s->gb, bits);
662 v = l1_unscale(bits - 1, v, scale);
663 s->sb_samples[ch][k * 12 + l + m][i] = v;
667 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
668 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
669 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
672 /* next subband in alloc table */
673 j += 1 << bit_alloc_bits;
675 /* XXX: find a way to avoid this duplication of code */
676 for (i = bound; i < sblimit; i++) {
677 bit_alloc_bits = alloc_table[j];
680 int mant, scale0, scale1;
681 scale0 = scale_factors[0][i][k];
682 scale1 = scale_factors[1][i][k];
683 qindex = alloc_table[j+b];
684 bits = ff_mpa_quant_bits[qindex];
686 /* 3 values at the same time */
687 v = get_bits(&s->gb, -bits);
688 steps = ff_mpa_quant_steps[qindex];
691 s->sb_samples[0][k * 12 + l + 0][i] =
692 l2_unscale_group(steps, mant, scale0);
693 s->sb_samples[1][k * 12 + l + 0][i] =
694 l2_unscale_group(steps, mant, scale1);
697 s->sb_samples[0][k * 12 + l + 1][i] =
698 l2_unscale_group(steps, mant, scale0);
699 s->sb_samples[1][k * 12 + l + 1][i] =
700 l2_unscale_group(steps, mant, scale1);
701 s->sb_samples[0][k * 12 + l + 2][i] =
702 l2_unscale_group(steps, v, scale0);
703 s->sb_samples[1][k * 12 + l + 2][i] =
704 l2_unscale_group(steps, v, scale1);
706 for (m = 0; m < 3; m++) {
707 mant = get_bits(&s->gb, bits);
708 s->sb_samples[0][k * 12 + l + m][i] =
709 l1_unscale(bits - 1, mant, scale0);
710 s->sb_samples[1][k * 12 + l + m][i] =
711 l1_unscale(bits - 1, mant, scale1);
715 s->sb_samples[0][k * 12 + l + 0][i] = 0;
716 s->sb_samples[0][k * 12 + l + 1][i] = 0;
717 s->sb_samples[0][k * 12 + l + 2][i] = 0;
718 s->sb_samples[1][k * 12 + l + 0][i] = 0;
719 s->sb_samples[1][k * 12 + l + 1][i] = 0;
720 s->sb_samples[1][k * 12 + l + 2][i] = 0;
722 /* next subband in alloc table */
723 j += 1 << bit_alloc_bits;
725 /* fill remaining samples to zero */
726 for (i = sblimit; i < SBLIMIT; i++) {
727 for (ch = 0; ch < s->nb_channels; ch++) {
728 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
729 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
730 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
738 #define SPLIT(dst,sf,n) \
740 int m = (sf * 171) >> 9; \
743 } else if (n == 4) { \
746 } else if (n == 5) { \
747 int m = (sf * 205) >> 10; \
750 } else if (n == 6) { \
751 int m = (sf * 171) >> 10; \
758 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
761 SPLIT(slen[3], sf, n3)
762 SPLIT(slen[2], sf, n2)
763 SPLIT(slen[1], sf, n1)
767 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
770 const uint8_t *bstab, *pretab;
771 int len, i, j, k, l, v0, shift, gain, gains[3];
775 gain = g->global_gain - 210;
776 shift = g->scalefac_scale + 1;
778 bstab = band_size_long[s->sample_rate_index];
779 pretab = mpa_pretab[g->preflag];
780 for (i = 0; i < g->long_end; i++) {
781 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
783 for (j = len; j > 0; j--)
787 if (g->short_start < 13) {
788 bstab = band_size_short[s->sample_rate_index];
789 gains[0] = gain - (g->subblock_gain[0] << 3);
790 gains[1] = gain - (g->subblock_gain[1] << 3);
791 gains[2] = gain - (g->subblock_gain[2] << 3);
793 for (i = g->short_start; i < 13; i++) {
795 for (l = 0; l < 3; l++) {
796 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
797 for (j = len; j > 0; j--)
804 /* handle n = 0 too */
805 static inline int get_bitsz(GetBitContext *s, int n)
807 return n ? get_bits(s, n) : 0;
811 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
814 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits) {
816 s->in_gb.buffer = NULL;
817 av_assert2((get_bits_count(&s->gb) & 7) == 0);
818 skip_bits_long(&s->gb, *pos - *end_pos);
820 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
821 *pos = get_bits_count(&s->gb);
825 /* Following is a optimized code for
827 if(get_bits1(&s->gb))
832 #define READ_FLIP_SIGN(dst,src) \
833 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
836 #define READ_FLIP_SIGN(dst,src) \
837 v = -get_bits1(&s->gb); \
838 *(dst) = (*(src) ^ v) - v;
841 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
842 int16_t *exponents, int end_pos2)
846 int last_pos, bits_left;
848 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits);
850 /* low frequencies (called big values) */
852 for (i = 0; i < 3; i++) {
853 int j, k, l, linbits;
854 j = g->region_size[i];
857 /* select vlc table */
858 k = g->table_select[i];
859 l = mpa_huff_data[k][0];
860 linbits = mpa_huff_data[k][1];
864 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
869 /* read huffcode and compute each couple */
873 int pos = get_bits_count(&s->gb);
876 switch_buffer(s, &pos, &end_pos, &end_pos2);
880 y = get_vlc2(&s->gb, vlc->table, 7, 3);
883 g->sb_hybrid[s_index ] =
884 g->sb_hybrid[s_index+1] = 0;
889 exponent= exponents[s_index];
891 av_dlog(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
892 i, g->region_size[i] - j, x, y, exponent);
897 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
899 x += get_bitsz(&s->gb, linbits);
900 v = l3_unscale(x, exponent);
901 if (get_bits1(&s->gb))
903 g->sb_hybrid[s_index] = v;
906 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
908 y += get_bitsz(&s->gb, linbits);
909 v = l3_unscale(y, exponent);
910 if (get_bits1(&s->gb))
912 g->sb_hybrid[s_index+1] = v;
919 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
921 x += get_bitsz(&s->gb, linbits);
922 v = l3_unscale(x, exponent);
923 if (get_bits1(&s->gb))
925 g->sb_hybrid[s_index+!!y] = v;
927 g->sb_hybrid[s_index + !y] = 0;
933 /* high frequencies */
934 vlc = &huff_quad_vlc[g->count1table_select];
936 while (s_index <= 572) {
938 pos = get_bits_count(&s->gb);
939 if (pos >= end_pos) {
940 if (pos > end_pos2 && last_pos) {
941 /* some encoders generate an incorrect size for this
942 part. We must go back into the data */
944 skip_bits_long(&s->gb, last_pos - pos);
945 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
946 if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT))
950 switch_buffer(s, &pos, &end_pos, &end_pos2);
956 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
957 av_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
958 g->sb_hybrid[s_index+0] =
959 g->sb_hybrid[s_index+1] =
960 g->sb_hybrid[s_index+2] =
961 g->sb_hybrid[s_index+3] = 0;
963 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
965 int pos = s_index + idxtab[code];
966 code ^= 8 >> idxtab[code];
967 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
971 /* skip extension bits */
972 bits_left = end_pos2 - get_bits_count(&s->gb);
973 if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) {
974 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
976 } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) {
977 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
980 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
981 skip_bits_long(&s->gb, bits_left);
983 i = get_bits_count(&s->gb);
984 switch_buffer(s, &i, &end_pos, &end_pos2);
989 /* Reorder short blocks from bitstream order to interleaved order. It
990 would be faster to do it in parsing, but the code would be far more
992 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
995 INTFLOAT *ptr, *dst, *ptr1;
998 if (g->block_type != 2)
1001 if (g->switch_point) {
1002 if (s->sample_rate_index != 8)
1003 ptr = g->sb_hybrid + 36;
1005 ptr = g->sb_hybrid + 72;
1010 for (i = g->short_start; i < 13; i++) {
1011 len = band_size_short[s->sample_rate_index][i];
1014 for (j = len; j > 0; j--) {
1015 *dst++ = ptr[0*len];
1016 *dst++ = ptr[1*len];
1017 *dst++ = ptr[2*len];
1021 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1025 #define ISQRT2 FIXR(0.70710678118654752440)
1027 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1030 int sf_max, sf, len, non_zero_found;
1031 INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1032 int non_zero_found_short[3];
1034 /* intensity stereo */
1035 if (s->mode_ext & MODE_EXT_I_STEREO) {
1040 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1044 tab0 = g0->sb_hybrid + 576;
1045 tab1 = g1->sb_hybrid + 576;
1047 non_zero_found_short[0] = 0;
1048 non_zero_found_short[1] = 0;
1049 non_zero_found_short[2] = 0;
1050 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1051 for (i = 12; i >= g1->short_start; i--) {
1052 /* for last band, use previous scale factor */
1055 len = band_size_short[s->sample_rate_index][i];
1056 for (l = 2; l >= 0; l--) {
1059 if (!non_zero_found_short[l]) {
1060 /* test if non zero band. if so, stop doing i-stereo */
1061 for (j = 0; j < len; j++) {
1063 non_zero_found_short[l] = 1;
1067 sf = g1->scale_factors[k + l];
1073 for (j = 0; j < len; j++) {
1075 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1076 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1080 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1081 /* lower part of the spectrum : do ms stereo
1083 for (j = 0; j < len; j++) {
1086 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1087 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1094 non_zero_found = non_zero_found_short[0] |
1095 non_zero_found_short[1] |
1096 non_zero_found_short[2];
1098 for (i = g1->long_end - 1;i >= 0;i--) {
1099 len = band_size_long[s->sample_rate_index][i];
1102 /* test if non zero band. if so, stop doing i-stereo */
1103 if (!non_zero_found) {
1104 for (j = 0; j < len; j++) {
1110 /* for last band, use previous scale factor */
1111 k = (i == 21) ? 20 : i;
1112 sf = g1->scale_factors[k];
1117 for (j = 0; j < len; j++) {
1119 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1120 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1124 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1125 /* lower part of the spectrum : do ms stereo
1127 for (j = 0; j < len; j++) {
1130 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1131 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1136 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1137 /* ms stereo ONLY */
1138 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1141 s->fdsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1143 tab0 = g0->sb_hybrid;
1144 tab1 = g1->sb_hybrid;
1145 for (i = 0; i < 576; i++) {
1148 tab0[i] = tmp0 + tmp1;
1149 tab1[i] = tmp0 - tmp1;
1157 # include "mips/compute_antialias_float.h"
1158 #endif /* HAVE_MIPSFPU */
1161 # include "mips/compute_antialias_fixed.h"
1162 #endif /* HAVE_MIPSDSPR1 */
1163 #endif /* USE_FLOATS */
1165 #ifndef compute_antialias
1167 #define AA(j) do { \
1168 float tmp0 = ptr[-1-j]; \
1169 float tmp1 = ptr[ j]; \
1170 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1171 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1174 #define AA(j) do { \
1175 int tmp0 = ptr[-1-j]; \
1176 int tmp1 = ptr[ j]; \
1177 int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1178 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1179 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1183 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1188 /* we antialias only "long" bands */
1189 if (g->block_type == 2) {
1190 if (!g->switch_point)
1192 /* XXX: check this for 8000Hz case */
1198 ptr = g->sb_hybrid + 18;
1199 for (i = n; i > 0; i--) {
1212 #endif /* compute_antialias */
1214 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1215 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1217 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1219 int i, j, mdct_long_end, sblimit;
1221 /* find last non zero block */
1222 ptr = g->sb_hybrid + 576;
1223 ptr1 = g->sb_hybrid + 2 * 18;
1224 while (ptr >= ptr1) {
1228 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1231 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1233 if (g->block_type == 2) {
1234 /* XXX: check for 8000 Hz */
1235 if (g->switch_point)
1240 mdct_long_end = sblimit;
1243 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1244 mdct_long_end, g->switch_point,
1247 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1248 ptr = g->sb_hybrid + 18 * mdct_long_end;
1250 for (j = mdct_long_end; j < sblimit; j++) {
1251 /* select frequency inversion */
1252 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1253 out_ptr = sb_samples + j;
1255 for (i = 0; i < 6; i++) {
1256 *out_ptr = buf[4*i];
1259 imdct12(out2, ptr + 0);
1260 for (i = 0; i < 6; i++) {
1261 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1262 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1265 imdct12(out2, ptr + 1);
1266 for (i = 0; i < 6; i++) {
1267 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1268 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1271 imdct12(out2, ptr + 2);
1272 for (i = 0; i < 6; i++) {
1273 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1274 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1275 buf[4*(i + 6*2)] = 0;
1278 buf += (j&3) != 3 ? 1 : (4*18-3);
1281 for (j = sblimit; j < SBLIMIT; j++) {
1283 out_ptr = sb_samples + j;
1284 for (i = 0; i < 18; i++) {
1285 *out_ptr = buf[4*i];
1289 buf += (j&3) != 3 ? 1 : (4*18-3);
1293 /* main layer3 decoding function */
1294 static int mp_decode_layer3(MPADecodeContext *s)
1296 int nb_granules, main_data_begin;
1297 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1299 int16_t exponents[576]; //FIXME try INTFLOAT
1301 /* read side info */
1303 main_data_begin = get_bits(&s->gb, 8);
1304 skip_bits(&s->gb, s->nb_channels);
1307 main_data_begin = get_bits(&s->gb, 9);
1308 if (s->nb_channels == 2)
1309 skip_bits(&s->gb, 3);
1311 skip_bits(&s->gb, 5);
1313 for (ch = 0; ch < s->nb_channels; ch++) {
1314 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1315 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1319 for (gr = 0; gr < nb_granules; gr++) {
1320 for (ch = 0; ch < s->nb_channels; ch++) {
1321 av_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1322 g = &s->granules[ch][gr];
1323 g->part2_3_length = get_bits(&s->gb, 12);
1324 g->big_values = get_bits(&s->gb, 9);
1325 if (g->big_values > 288) {
1326 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1327 return AVERROR_INVALIDDATA;
1330 g->global_gain = get_bits(&s->gb, 8);
1331 /* if MS stereo only is selected, we precompute the
1332 1/sqrt(2) renormalization factor */
1333 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1335 g->global_gain -= 2;
1337 g->scalefac_compress = get_bits(&s->gb, 9);
1339 g->scalefac_compress = get_bits(&s->gb, 4);
1340 blocksplit_flag = get_bits1(&s->gb);
1341 if (blocksplit_flag) {
1342 g->block_type = get_bits(&s->gb, 2);
1343 if (g->block_type == 0) {
1344 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1345 return AVERROR_INVALIDDATA;
1347 g->switch_point = get_bits1(&s->gb);
1348 for (i = 0; i < 2; i++)
1349 g->table_select[i] = get_bits(&s->gb, 5);
1350 for (i = 0; i < 3; i++)
1351 g->subblock_gain[i] = get_bits(&s->gb, 3);
1352 init_short_region(s, g);
1354 int region_address1, region_address2;
1356 g->switch_point = 0;
1357 for (i = 0; i < 3; i++)
1358 g->table_select[i] = get_bits(&s->gb, 5);
1359 /* compute huffman coded region sizes */
1360 region_address1 = get_bits(&s->gb, 4);
1361 region_address2 = get_bits(&s->gb, 3);
1362 av_dlog(s->avctx, "region1=%d region2=%d\n",
1363 region_address1, region_address2);
1364 init_long_region(s, g, region_address1, region_address2);
1366 region_offset2size(g);
1367 compute_band_indexes(s, g);
1371 g->preflag = get_bits1(&s->gb);
1372 g->scalefac_scale = get_bits1(&s->gb);
1373 g->count1table_select = get_bits1(&s->gb);
1374 av_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1375 g->block_type, g->switch_point);
1381 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1382 int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0, EXTRABYTES);
1383 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1384 /* now we get bits from the main_data_begin offset */
1385 av_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1386 main_data_begin, s->last_buf_size);
1388 memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);
1390 init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
1391 #if !UNCHECKED_BITSTREAM_READER
1392 s->gb.size_in_bits_plus8 += FFMAX(extrasize, LAST_BUF_SIZE - s->last_buf_size) * 8;
1394 s->last_buf_size <<= 3;
1395 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1396 for (ch = 0; ch < s->nb_channels; ch++) {
1397 g = &s->granules[ch][gr];
1398 s->last_buf_size += g->part2_3_length;
1399 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1400 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1403 skip = s->last_buf_size - 8 * main_data_begin;
1404 if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {
1405 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);
1407 s->in_gb.buffer = NULL;
1409 skip_bits_long(&s->gb, skip);
1415 for (; gr < nb_granules; gr++) {
1416 for (ch = 0; ch < s->nb_channels; ch++) {
1417 g = &s->granules[ch][gr];
1418 bits_pos = get_bits_count(&s->gb);
1422 int slen, slen1, slen2;
1424 /* MPEG1 scale factors */
1425 slen1 = slen_table[0][g->scalefac_compress];
1426 slen2 = slen_table[1][g->scalefac_compress];
1427 av_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1428 if (g->block_type == 2) {
1429 n = g->switch_point ? 17 : 18;
1432 for (i = 0; i < n; i++)
1433 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1435 for (i = 0; i < n; i++)
1436 g->scale_factors[j++] = 0;
1439 for (i = 0; i < 18; i++)
1440 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1441 for (i = 0; i < 3; i++)
1442 g->scale_factors[j++] = 0;
1444 for (i = 0; i < 21; i++)
1445 g->scale_factors[j++] = 0;
1448 sc = s->granules[ch][0].scale_factors;
1450 for (k = 0; k < 4; k++) {
1452 if ((g->scfsi & (0x8 >> k)) == 0) {
1453 slen = (k < 2) ? slen1 : slen2;
1455 for (i = 0; i < n; i++)
1456 g->scale_factors[j++] = get_bits(&s->gb, slen);
1458 for (i = 0; i < n; i++)
1459 g->scale_factors[j++] = 0;
1462 /* simply copy from last granule */
1463 for (i = 0; i < n; i++) {
1464 g->scale_factors[j] = sc[j];
1469 g->scale_factors[j++] = 0;
1472 int tindex, tindex2, slen[4], sl, sf;
1474 /* LSF scale factors */
1475 if (g->block_type == 2)
1476 tindex = g->switch_point ? 2 : 1;
1480 sf = g->scalefac_compress;
1481 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1482 /* intensity stereo case */
1485 lsf_sf_expand(slen, sf, 6, 6, 0);
1487 } else if (sf < 244) {
1488 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1491 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1497 lsf_sf_expand(slen, sf, 5, 4, 4);
1499 } else if (sf < 500) {
1500 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1503 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1510 for (k = 0; k < 4; k++) {
1511 n = lsf_nsf_table[tindex2][tindex][k];
1514 for (i = 0; i < n; i++)
1515 g->scale_factors[j++] = get_bits(&s->gb, sl);
1517 for (i = 0; i < n; i++)
1518 g->scale_factors[j++] = 0;
1521 /* XXX: should compute exact size */
1523 g->scale_factors[j] = 0;
1526 exponents_from_scale_factors(s, g, exponents);
1528 /* read Huffman coded residue */
1529 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1532 if (s->mode == MPA_JSTEREO)
1533 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1535 for (ch = 0; ch < s->nb_channels; ch++) {
1536 g = &s->granules[ch][gr];
1538 reorder_block(s, g);
1539 compute_antialias(s, g);
1540 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1543 if (get_bits_count(&s->gb) < 0)
1544 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1545 return nb_granules * 18;
1548 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1549 const uint8_t *buf, int buf_size)
1551 int i, nb_frames, ch, ret;
1552 OUT_INT *samples_ptr;
1554 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1556 /* skip error protection field */
1557 if (s->error_protection)
1558 skip_bits(&s->gb, 16);
1562 s->avctx->frame_size = 384;
1563 nb_frames = mp_decode_layer1(s);
1566 s->avctx->frame_size = 1152;
1567 nb_frames = mp_decode_layer2(s);
1570 s->avctx->frame_size = s->lsf ? 576 : 1152;
1572 nb_frames = mp_decode_layer3(s);
1575 if (s->in_gb.buffer) {
1576 align_get_bits(&s->gb);
1577 i = get_bits_left(&s->gb)>>3;
1578 if (i >= 0 && i <= BACKSTEP_SIZE) {
1579 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1582 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1584 s->in_gb.buffer = NULL;
1587 align_get_bits(&s->gb);
1588 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1589 i = get_bits_left(&s->gb) >> 3;
1591 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1593 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1594 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1596 av_assert1(i <= buf_size - HEADER_SIZE && i >= 0);
1597 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1598 s->last_buf_size += i;
1604 /* get output buffer */
1606 av_assert0(s->frame != NULL);
1607 s->frame->nb_samples = s->avctx->frame_size;
1608 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0)
1610 samples = (OUT_INT **)s->frame->extended_data;
1613 /* apply the synthesis filter */
1614 for (ch = 0; ch < s->nb_channels; ch++) {
1616 if (s->avctx->sample_fmt == OUT_FMT_P) {
1617 samples_ptr = samples[ch];
1620 samples_ptr = samples[0] + ch;
1621 sample_stride = s->nb_channels;
1623 for (i = 0; i < nb_frames; i++) {
1624 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1625 &(s->synth_buf_offset[ch]),
1626 RENAME(ff_mpa_synth_window),
1627 &s->dither_state, samples_ptr,
1628 sample_stride, s->sb_samples[ch][i]);
1629 samples_ptr += 32 * sample_stride;
1633 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1636 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1639 const uint8_t *buf = avpkt->data;
1640 int buf_size = avpkt->size;
1641 MPADecodeContext *s = avctx->priv_data;
1645 while(buf_size && !*buf){
1650 if (buf_size < HEADER_SIZE)
1651 return AVERROR_INVALIDDATA;
1653 header = AV_RB32(buf);
1654 if (header>>8 == AV_RB32("TAG")>>8) {
1655 av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n");
1658 if (ff_mpa_check_header(header) < 0) {
1659 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1660 return AVERROR_INVALIDDATA;
1663 if (avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
1664 /* free format: prepare to compute frame size */
1666 return AVERROR_INVALIDDATA;
1668 /* update codec info */
1669 avctx->channels = s->nb_channels;
1670 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1671 if (!avctx->bit_rate)
1672 avctx->bit_rate = s->bit_rate;
1674 if (s->frame_size <= 0 || s->frame_size > buf_size) {
1675 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1676 return AVERROR_INVALIDDATA;
1677 } else if (s->frame_size < buf_size) {
1678 av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");
1679 buf_size= s->frame_size;
1684 ret = mp_decode_frame(s, NULL, buf, buf_size);
1686 s->frame->nb_samples = avctx->frame_size;
1688 avctx->sample_rate = s->sample_rate;
1689 //FIXME maybe move the other codec info stuff from above here too
1691 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1692 /* Only return an error if the bad frame makes up the whole packet or
1693 * the error is related to buffer management.
1694 * If there is more data in the packet, just consume the bad frame
1695 * instead of returning an error, which would discard the whole
1698 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1705 static void mp_flush(MPADecodeContext *ctx)
1707 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1708 ctx->last_buf_size = 0;
1711 static void flush(AVCodecContext *avctx)
1713 mp_flush(avctx->priv_data);
1716 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1717 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1718 int *got_frame_ptr, AVPacket *avpkt)
1720 const uint8_t *buf = avpkt->data;
1721 int buf_size = avpkt->size;
1722 MPADecodeContext *s = avctx->priv_data;
1725 int av_unused out_size;
1729 // Discard too short frames
1730 if (buf_size < HEADER_SIZE) {
1731 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1732 return AVERROR_INVALIDDATA;
1736 if (len > MPA_MAX_CODED_FRAME_SIZE)
1737 len = MPA_MAX_CODED_FRAME_SIZE;
1739 // Get header and restore sync word
1740 header = AV_RB32(buf) | 0xffe00000;
1742 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
1743 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1744 return AVERROR_INVALIDDATA;
1747 avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1748 /* update codec info */
1749 avctx->sample_rate = s->sample_rate;
1750 avctx->channels = s->nb_channels;
1751 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1752 if (!avctx->bit_rate)
1753 avctx->bit_rate = s->bit_rate;
1755 s->frame_size = len;
1759 ret = mp_decode_frame(s, NULL, buf, buf_size);
1761 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1769 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1771 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1774 * Context for MP3On4 decoder
1776 typedef struct MP3On4DecodeContext {
1777 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1778 int syncword; ///< syncword patch
1779 const uint8_t *coff; ///< channel offsets in output buffer
1780 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1781 } MP3On4DecodeContext;
1783 #include "mpeg4audio.h"
1785 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1787 /* number of mp3 decoder instances */
1788 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1790 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1791 static const uint8_t chan_offset[8][5] = {
1796 { 2, 0, 3 }, // C FLR BS
1797 { 2, 0, 3 }, // C FLR BLRS
1798 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1799 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1802 /* mp3on4 channel layouts */
1803 static const int16_t chan_layout[8] = {
1806 AV_CH_LAYOUT_STEREO,
1807 AV_CH_LAYOUT_SURROUND,
1808 AV_CH_LAYOUT_4POINT0,
1809 AV_CH_LAYOUT_5POINT0,
1810 AV_CH_LAYOUT_5POINT1,
1811 AV_CH_LAYOUT_7POINT1
1814 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1816 MP3On4DecodeContext *s = avctx->priv_data;
1819 for (i = 0; i < s->frames; i++)
1820 av_free(s->mp3decctx[i]);
1826 static av_cold int decode_init_mp3on4(AVCodecContext * avctx)
1828 MP3On4DecodeContext *s = avctx->priv_data;
1829 MPEG4AudioConfig cfg;
1832 if ((avctx->extradata_size < 2) || (avctx->extradata == NULL)) {
1833 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1834 return AVERROR_INVALIDDATA;
1837 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
1838 avctx->extradata_size * 8, 1);
1839 if (!cfg.chan_config || cfg.chan_config > 7) {
1840 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1841 return AVERROR_INVALIDDATA;
1843 s->frames = mp3Frames[cfg.chan_config];
1844 s->coff = chan_offset[cfg.chan_config];
1845 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1846 avctx->channel_layout = chan_layout[cfg.chan_config];
1848 if (cfg.sample_rate < 16000)
1849 s->syncword = 0xffe00000;
1851 s->syncword = 0xfff00000;
1853 /* Init the first mp3 decoder in standard way, so that all tables get builded
1854 * We replace avctx->priv_data with the context of the first decoder so that
1855 * decode_init() does not have to be changed.
1856 * Other decoders will be initialized here copying data from the first context
1858 // Allocate zeroed memory for the first decoder context
1859 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1860 if (!s->mp3decctx[0])
1862 // Put decoder context in place to make init_decode() happy
1863 avctx->priv_data = s->mp3decctx[0];
1865 // Restore mp3on4 context pointer
1866 avctx->priv_data = s;
1867 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1869 /* Create a separate codec/context for each frame (first is already ok).
1870 * Each frame is 1 or 2 channels - up to 5 frames allowed
1872 for (i = 1; i < s->frames; i++) {
1873 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1874 if (!s->mp3decctx[i])
1876 s->mp3decctx[i]->adu_mode = 1;
1877 s->mp3decctx[i]->avctx = avctx;
1878 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1883 decode_close_mp3on4(avctx);
1884 return AVERROR(ENOMEM);
1888 static void flush_mp3on4(AVCodecContext *avctx)
1891 MP3On4DecodeContext *s = avctx->priv_data;
1893 for (i = 0; i < s->frames; i++)
1894 mp_flush(s->mp3decctx[i]);
1898 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1899 int *got_frame_ptr, AVPacket *avpkt)
1901 AVFrame *frame = data;
1902 const uint8_t *buf = avpkt->data;
1903 int buf_size = avpkt->size;
1904 MP3On4DecodeContext *s = avctx->priv_data;
1905 MPADecodeContext *m;
1906 int fsize, len = buf_size, out_size = 0;
1908 OUT_INT **out_samples;
1912 /* get output buffer */
1913 frame->nb_samples = MPA_FRAME_SIZE;
1914 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1916 out_samples = (OUT_INT **)frame->extended_data;
1918 // Discard too short frames
1919 if (buf_size < HEADER_SIZE)
1920 return AVERROR_INVALIDDATA;
1922 avctx->bit_rate = 0;
1925 for (fr = 0; fr < s->frames; fr++) {
1926 fsize = AV_RB16(buf) >> 4;
1927 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1928 m = s->mp3decctx[fr];
1931 if (fsize < HEADER_SIZE) {
1932 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1933 return AVERROR_INVALIDDATA;
1935 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1937 if (ff_mpa_check_header(header) < 0) // Bad header, discard block
1940 avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1942 if (ch + m->nb_channels > avctx->channels ||
1943 s->coff[fr] + m->nb_channels > avctx->channels) {
1944 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1946 return AVERROR_INVALIDDATA;
1948 ch += m->nb_channels;
1950 outptr[0] = out_samples[s->coff[fr]];
1951 if (m->nb_channels > 1)
1952 outptr[1] = out_samples[s->coff[fr] + 1];
1954 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0)
1961 avctx->bit_rate += m->bit_rate;
1964 /* update codec info */
1965 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1967 frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1972 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */