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((int)(mant + (-1U << 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 ff_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 ff_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;
410 static av_cold int decode_close(AVCodecContext * avctx)
412 MPADecodeContext *s = avctx->priv_data;
419 static av_cold int decode_init(AVCodecContext * avctx)
421 static int initialized_tables = 0;
422 MPADecodeContext *s = avctx->priv_data;
424 if (!initialized_tables) {
425 decode_init_static();
426 initialized_tables = 1;
432 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
434 return AVERROR(ENOMEM);
437 ff_mpadsp_init(&s->mpadsp);
439 if (avctx->request_sample_fmt == OUT_FMT &&
440 avctx->codec_id != AV_CODEC_ID_MP3ON4)
441 avctx->sample_fmt = OUT_FMT;
443 avctx->sample_fmt = OUT_FMT_P;
444 s->err_recognition = avctx->err_recognition;
446 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
452 #define C3 FIXHR(0.86602540378443864676/2)
453 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
454 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
455 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
457 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
459 static void imdct12(INTFLOAT *out, INTFLOAT *in)
461 INTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
464 in1 = in[1*3] + in[0*3];
465 in2 = in[2*3] + in[1*3];
466 in3 = in[3*3] + in[2*3];
467 in4 = in[4*3] + in[3*3];
468 in5 = in[5*3] + in[4*3];
472 in2 = MULH3(in2, C3, 2);
473 in3 = MULH3(in3, C3, 4);
476 t2 = MULH3(in1 - in5, C4, 2);
486 in1 = MULH3(in5 + in3, C5, 1);
493 in5 = MULH3(in5 - in3, C6, 2);
500 /* return the number of decoded frames */
501 static int mp_decode_layer1(MPADecodeContext *s)
503 int bound, i, v, n, ch, j, mant;
504 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
505 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
507 if (s->mode == MPA_JSTEREO)
508 bound = (s->mode_ext + 1) * 4;
512 /* allocation bits */
513 for (i = 0; i < bound; i++) {
514 for (ch = 0; ch < s->nb_channels; ch++) {
515 allocation[ch][i] = get_bits(&s->gb, 4);
518 for (i = bound; i < SBLIMIT; i++)
519 allocation[0][i] = get_bits(&s->gb, 4);
522 for (i = 0; i < bound; i++) {
523 for (ch = 0; ch < s->nb_channels; ch++) {
524 if (allocation[ch][i])
525 scale_factors[ch][i] = get_bits(&s->gb, 6);
528 for (i = bound; i < SBLIMIT; i++) {
529 if (allocation[0][i]) {
530 scale_factors[0][i] = get_bits(&s->gb, 6);
531 scale_factors[1][i] = get_bits(&s->gb, 6);
535 /* compute samples */
536 for (j = 0; j < 12; j++) {
537 for (i = 0; i < bound; i++) {
538 for (ch = 0; ch < s->nb_channels; ch++) {
539 n = allocation[ch][i];
541 mant = get_bits(&s->gb, n + 1);
542 v = l1_unscale(n, mant, scale_factors[ch][i]);
546 s->sb_samples[ch][j][i] = v;
549 for (i = bound; i < SBLIMIT; i++) {
550 n = allocation[0][i];
552 mant = get_bits(&s->gb, n + 1);
553 v = l1_unscale(n, mant, scale_factors[0][i]);
554 s->sb_samples[0][j][i] = v;
555 v = l1_unscale(n, mant, scale_factors[1][i]);
556 s->sb_samples[1][j][i] = v;
558 s->sb_samples[0][j][i] = 0;
559 s->sb_samples[1][j][i] = 0;
566 static int mp_decode_layer2(MPADecodeContext *s)
568 int sblimit; /* number of used subbands */
569 const unsigned char *alloc_table;
570 int table, bit_alloc_bits, i, j, ch, bound, v;
571 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
572 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
573 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
574 int scale, qindex, bits, steps, k, l, m, b;
576 /* select decoding table */
577 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
578 s->sample_rate, s->lsf);
579 sblimit = ff_mpa_sblimit_table[table];
580 alloc_table = ff_mpa_alloc_tables[table];
582 if (s->mode == MPA_JSTEREO)
583 bound = (s->mode_ext + 1) * 4;
587 ff_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
593 /* parse bit allocation */
595 for (i = 0; i < bound; i++) {
596 bit_alloc_bits = alloc_table[j];
597 for (ch = 0; ch < s->nb_channels; ch++)
598 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
599 j += 1 << bit_alloc_bits;
601 for (i = bound; i < sblimit; i++) {
602 bit_alloc_bits = alloc_table[j];
603 v = get_bits(&s->gb, bit_alloc_bits);
606 j += 1 << bit_alloc_bits;
610 for (i = 0; i < sblimit; i++) {
611 for (ch = 0; ch < s->nb_channels; ch++) {
612 if (bit_alloc[ch][i])
613 scale_code[ch][i] = get_bits(&s->gb, 2);
618 for (i = 0; i < sblimit; i++) {
619 for (ch = 0; ch < s->nb_channels; ch++) {
620 if (bit_alloc[ch][i]) {
621 sf = scale_factors[ch][i];
622 switch (scale_code[ch][i]) {
625 sf[0] = get_bits(&s->gb, 6);
626 sf[1] = get_bits(&s->gb, 6);
627 sf[2] = get_bits(&s->gb, 6);
630 sf[0] = get_bits(&s->gb, 6);
635 sf[0] = get_bits(&s->gb, 6);
636 sf[2] = get_bits(&s->gb, 6);
640 sf[0] = get_bits(&s->gb, 6);
641 sf[2] = get_bits(&s->gb, 6);
650 for (k = 0; k < 3; k++) {
651 for (l = 0; l < 12; l += 3) {
653 for (i = 0; i < bound; i++) {
654 bit_alloc_bits = alloc_table[j];
655 for (ch = 0; ch < s->nb_channels; ch++) {
656 b = bit_alloc[ch][i];
658 scale = scale_factors[ch][i][k];
659 qindex = alloc_table[j+b];
660 bits = ff_mpa_quant_bits[qindex];
663 /* 3 values at the same time */
664 v = get_bits(&s->gb, -bits);
665 v2 = division_tabs[qindex][v];
666 steps = ff_mpa_quant_steps[qindex];
668 s->sb_samples[ch][k * 12 + l + 0][i] =
669 l2_unscale_group(steps, v2 & 15, scale);
670 s->sb_samples[ch][k * 12 + l + 1][i] =
671 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
672 s->sb_samples[ch][k * 12 + l + 2][i] =
673 l2_unscale_group(steps, v2 >> 8 , scale);
675 for (m = 0; m < 3; m++) {
676 v = get_bits(&s->gb, bits);
677 v = l1_unscale(bits - 1, v, scale);
678 s->sb_samples[ch][k * 12 + l + m][i] = v;
682 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
683 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
684 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
687 /* next subband in alloc table */
688 j += 1 << bit_alloc_bits;
690 /* XXX: find a way to avoid this duplication of code */
691 for (i = bound; i < sblimit; i++) {
692 bit_alloc_bits = alloc_table[j];
695 int mant, scale0, scale1;
696 scale0 = scale_factors[0][i][k];
697 scale1 = scale_factors[1][i][k];
698 qindex = alloc_table[j+b];
699 bits = ff_mpa_quant_bits[qindex];
701 /* 3 values at the same time */
702 v = get_bits(&s->gb, -bits);
703 steps = ff_mpa_quant_steps[qindex];
706 s->sb_samples[0][k * 12 + l + 0][i] =
707 l2_unscale_group(steps, mant, scale0);
708 s->sb_samples[1][k * 12 + l + 0][i] =
709 l2_unscale_group(steps, mant, scale1);
712 s->sb_samples[0][k * 12 + l + 1][i] =
713 l2_unscale_group(steps, mant, scale0);
714 s->sb_samples[1][k * 12 + l + 1][i] =
715 l2_unscale_group(steps, mant, scale1);
716 s->sb_samples[0][k * 12 + l + 2][i] =
717 l2_unscale_group(steps, v, scale0);
718 s->sb_samples[1][k * 12 + l + 2][i] =
719 l2_unscale_group(steps, v, scale1);
721 for (m = 0; m < 3; m++) {
722 mant = get_bits(&s->gb, bits);
723 s->sb_samples[0][k * 12 + l + m][i] =
724 l1_unscale(bits - 1, mant, scale0);
725 s->sb_samples[1][k * 12 + l + m][i] =
726 l1_unscale(bits - 1, mant, scale1);
730 s->sb_samples[0][k * 12 + l + 0][i] = 0;
731 s->sb_samples[0][k * 12 + l + 1][i] = 0;
732 s->sb_samples[0][k * 12 + l + 2][i] = 0;
733 s->sb_samples[1][k * 12 + l + 0][i] = 0;
734 s->sb_samples[1][k * 12 + l + 1][i] = 0;
735 s->sb_samples[1][k * 12 + l + 2][i] = 0;
737 /* next subband in alloc table */
738 j += 1 << bit_alloc_bits;
740 /* fill remaining samples to zero */
741 for (i = sblimit; i < SBLIMIT; i++) {
742 for (ch = 0; ch < s->nb_channels; ch++) {
743 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
744 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
745 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
753 #define SPLIT(dst,sf,n) \
755 int m = (sf * 171) >> 9; \
758 } else if (n == 4) { \
761 } else if (n == 5) { \
762 int m = (sf * 205) >> 10; \
765 } else if (n == 6) { \
766 int m = (sf * 171) >> 10; \
773 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
776 SPLIT(slen[3], sf, n3)
777 SPLIT(slen[2], sf, n2)
778 SPLIT(slen[1], sf, n1)
782 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
785 const uint8_t *bstab, *pretab;
786 int len, i, j, k, l, v0, shift, gain, gains[3];
790 gain = g->global_gain - 210;
791 shift = g->scalefac_scale + 1;
793 bstab = band_size_long[s->sample_rate_index];
794 pretab = mpa_pretab[g->preflag];
795 for (i = 0; i < g->long_end; i++) {
796 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
798 for (j = len; j > 0; j--)
802 if (g->short_start < 13) {
803 bstab = band_size_short[s->sample_rate_index];
804 gains[0] = gain - (g->subblock_gain[0] << 3);
805 gains[1] = gain - (g->subblock_gain[1] << 3);
806 gains[2] = gain - (g->subblock_gain[2] << 3);
808 for (i = g->short_start; i < 13; i++) {
810 for (l = 0; l < 3; l++) {
811 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
812 for (j = len; j > 0; j--)
819 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
822 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits) {
824 s->in_gb.buffer = NULL;
825 av_assert2((get_bits_count(&s->gb) & 7) == 0);
826 skip_bits_long(&s->gb, *pos - *end_pos);
828 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
829 *pos = get_bits_count(&s->gb);
833 /* Following is an optimized code for
835 if(get_bits1(&s->gb))
840 #define READ_FLIP_SIGN(dst,src) \
841 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
844 #define READ_FLIP_SIGN(dst,src) \
845 v = -get_bits1(&s->gb); \
846 *(dst) = (*(src) ^ v) - v;
849 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
850 int16_t *exponents, int end_pos2)
854 int last_pos, bits_left;
856 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits);
858 /* low frequencies (called big values) */
860 for (i = 0; i < 3; i++) {
861 int j, k, l, linbits;
862 j = g->region_size[i];
865 /* select vlc table */
866 k = g->table_select[i];
867 l = mpa_huff_data[k][0];
868 linbits = mpa_huff_data[k][1];
872 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
877 /* read huffcode and compute each couple */
881 int pos = get_bits_count(&s->gb);
884 switch_buffer(s, &pos, &end_pos, &end_pos2);
888 y = get_vlc2(&s->gb, vlc->table, 7, 3);
891 g->sb_hybrid[s_index ] =
892 g->sb_hybrid[s_index+1] = 0;
897 exponent= exponents[s_index];
899 ff_dlog(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
900 i, g->region_size[i] - j, x, y, exponent);
905 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
907 x += get_bitsz(&s->gb, linbits);
908 v = l3_unscale(x, exponent);
909 if (get_bits1(&s->gb))
911 g->sb_hybrid[s_index] = v;
914 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
916 y += get_bitsz(&s->gb, linbits);
917 v = l3_unscale(y, exponent);
918 if (get_bits1(&s->gb))
920 g->sb_hybrid[s_index+1] = v;
927 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
929 x += get_bitsz(&s->gb, linbits);
930 v = l3_unscale(x, exponent);
931 if (get_bits1(&s->gb))
933 g->sb_hybrid[s_index+!!y] = v;
935 g->sb_hybrid[s_index + !y] = 0;
941 /* high frequencies */
942 vlc = &huff_quad_vlc[g->count1table_select];
944 while (s_index <= 572) {
946 pos = get_bits_count(&s->gb);
947 if (pos >= end_pos) {
948 if (pos > end_pos2 && last_pos) {
949 /* some encoders generate an incorrect size for this
950 part. We must go back into the data */
952 skip_bits_long(&s->gb, last_pos - pos);
953 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
954 if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT))
958 switch_buffer(s, &pos, &end_pos, &end_pos2);
964 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
965 ff_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
966 g->sb_hybrid[s_index+0] =
967 g->sb_hybrid[s_index+1] =
968 g->sb_hybrid[s_index+2] =
969 g->sb_hybrid[s_index+3] = 0;
971 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
973 int pos = s_index + idxtab[code];
974 code ^= 8 >> idxtab[code];
975 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
979 /* skip extension bits */
980 bits_left = end_pos2 - get_bits_count(&s->gb);
981 if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) {
982 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
984 } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) {
985 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
988 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
989 skip_bits_long(&s->gb, bits_left);
991 i = get_bits_count(&s->gb);
992 switch_buffer(s, &i, &end_pos, &end_pos2);
997 /* Reorder short blocks from bitstream order to interleaved order. It
998 would be faster to do it in parsing, but the code would be far more
1000 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1003 INTFLOAT *ptr, *dst, *ptr1;
1006 if (g->block_type != 2)
1009 if (g->switch_point) {
1010 if (s->sample_rate_index != 8)
1011 ptr = g->sb_hybrid + 36;
1013 ptr = g->sb_hybrid + 72;
1018 for (i = g->short_start; i < 13; i++) {
1019 len = band_size_short[s->sample_rate_index][i];
1022 for (j = len; j > 0; j--) {
1023 *dst++ = ptr[0*len];
1024 *dst++ = ptr[1*len];
1025 *dst++ = ptr[2*len];
1029 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1033 #define ISQRT2 FIXR(0.70710678118654752440)
1035 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1038 int sf_max, sf, len, non_zero_found;
1039 INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1040 int non_zero_found_short[3];
1042 /* intensity stereo */
1043 if (s->mode_ext & MODE_EXT_I_STEREO) {
1048 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1052 tab0 = g0->sb_hybrid + 576;
1053 tab1 = g1->sb_hybrid + 576;
1055 non_zero_found_short[0] = 0;
1056 non_zero_found_short[1] = 0;
1057 non_zero_found_short[2] = 0;
1058 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1059 for (i = 12; i >= g1->short_start; i--) {
1060 /* for last band, use previous scale factor */
1063 len = band_size_short[s->sample_rate_index][i];
1064 for (l = 2; l >= 0; l--) {
1067 if (!non_zero_found_short[l]) {
1068 /* test if non zero band. if so, stop doing i-stereo */
1069 for (j = 0; j < len; j++) {
1071 non_zero_found_short[l] = 1;
1075 sf = g1->scale_factors[k + l];
1081 for (j = 0; j < len; j++) {
1083 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1084 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1088 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1089 /* lower part of the spectrum : do ms stereo
1091 for (j = 0; j < len; j++) {
1094 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1095 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1102 non_zero_found = non_zero_found_short[0] |
1103 non_zero_found_short[1] |
1104 non_zero_found_short[2];
1106 for (i = g1->long_end - 1;i >= 0;i--) {
1107 len = band_size_long[s->sample_rate_index][i];
1110 /* test if non zero band. if so, stop doing i-stereo */
1111 if (!non_zero_found) {
1112 for (j = 0; j < len; j++) {
1118 /* for last band, use previous scale factor */
1119 k = (i == 21) ? 20 : i;
1120 sf = g1->scale_factors[k];
1125 for (j = 0; j < len; j++) {
1127 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1128 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1132 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1133 /* lower part of the spectrum : do ms stereo
1135 for (j = 0; j < len; j++) {
1138 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1139 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1144 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1145 /* ms stereo ONLY */
1146 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1149 s->fdsp->butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1151 tab0 = g0->sb_hybrid;
1152 tab1 = g1->sb_hybrid;
1153 for (i = 0; i < 576; i++) {
1156 tab0[i] = tmp0 + tmp1;
1157 tab1[i] = tmp0 - tmp1;
1165 # include "mips/compute_antialias_float.h"
1166 #endif /* HAVE_MIPSFPU */
1169 # include "mips/compute_antialias_fixed.h"
1170 #endif /* HAVE_MIPSDSP */
1171 #endif /* USE_FLOATS */
1173 #ifndef compute_antialias
1175 #define AA(j) do { \
1176 float tmp0 = ptr[-1-j]; \
1177 float tmp1 = ptr[ j]; \
1178 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1179 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1182 #define AA(j) do { \
1183 int tmp0 = ptr[-1-j]; \
1184 int tmp1 = ptr[ j]; \
1185 int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1186 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1187 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1191 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1196 /* we antialias only "long" bands */
1197 if (g->block_type == 2) {
1198 if (!g->switch_point)
1200 /* XXX: check this for 8000Hz case */
1206 ptr = g->sb_hybrid + 18;
1207 for (i = n; i > 0; i--) {
1220 #endif /* compute_antialias */
1222 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1223 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1225 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1227 int i, j, mdct_long_end, sblimit;
1229 /* find last non zero block */
1230 ptr = g->sb_hybrid + 576;
1231 ptr1 = g->sb_hybrid + 2 * 18;
1232 while (ptr >= ptr1) {
1236 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1239 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1241 if (g->block_type == 2) {
1242 /* XXX: check for 8000 Hz */
1243 if (g->switch_point)
1248 mdct_long_end = sblimit;
1251 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1252 mdct_long_end, g->switch_point,
1255 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1256 ptr = g->sb_hybrid + 18 * mdct_long_end;
1258 for (j = mdct_long_end; j < sblimit; j++) {
1259 /* select frequency inversion */
1260 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1261 out_ptr = sb_samples + j;
1263 for (i = 0; i < 6; i++) {
1264 *out_ptr = buf[4*i];
1267 imdct12(out2, ptr + 0);
1268 for (i = 0; i < 6; i++) {
1269 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1270 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1273 imdct12(out2, ptr + 1);
1274 for (i = 0; i < 6; i++) {
1275 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1276 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1279 imdct12(out2, ptr + 2);
1280 for (i = 0; i < 6; i++) {
1281 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1282 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1283 buf[4*(i + 6*2)] = 0;
1286 buf += (j&3) != 3 ? 1 : (4*18-3);
1289 for (j = sblimit; j < SBLIMIT; j++) {
1291 out_ptr = sb_samples + j;
1292 for (i = 0; i < 18; i++) {
1293 *out_ptr = buf[4*i];
1297 buf += (j&3) != 3 ? 1 : (4*18-3);
1301 /* main layer3 decoding function */
1302 static int mp_decode_layer3(MPADecodeContext *s)
1304 int nb_granules, main_data_begin;
1305 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1307 int16_t exponents[576]; //FIXME try INTFLOAT
1309 /* read side info */
1311 main_data_begin = get_bits(&s->gb, 8);
1312 skip_bits(&s->gb, s->nb_channels);
1315 main_data_begin = get_bits(&s->gb, 9);
1316 if (s->nb_channels == 2)
1317 skip_bits(&s->gb, 3);
1319 skip_bits(&s->gb, 5);
1321 for (ch = 0; ch < s->nb_channels; ch++) {
1322 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1323 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1327 for (gr = 0; gr < nb_granules; gr++) {
1328 for (ch = 0; ch < s->nb_channels; ch++) {
1329 ff_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1330 g = &s->granules[ch][gr];
1331 g->part2_3_length = get_bits(&s->gb, 12);
1332 g->big_values = get_bits(&s->gb, 9);
1333 if (g->big_values > 288) {
1334 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1335 return AVERROR_INVALIDDATA;
1338 g->global_gain = get_bits(&s->gb, 8);
1339 /* if MS stereo only is selected, we precompute the
1340 1/sqrt(2) renormalization factor */
1341 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1343 g->global_gain -= 2;
1345 g->scalefac_compress = get_bits(&s->gb, 9);
1347 g->scalefac_compress = get_bits(&s->gb, 4);
1348 blocksplit_flag = get_bits1(&s->gb);
1349 if (blocksplit_flag) {
1350 g->block_type = get_bits(&s->gb, 2);
1351 if (g->block_type == 0) {
1352 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1353 return AVERROR_INVALIDDATA;
1355 g->switch_point = get_bits1(&s->gb);
1356 for (i = 0; i < 2; i++)
1357 g->table_select[i] = get_bits(&s->gb, 5);
1358 for (i = 0; i < 3; i++)
1359 g->subblock_gain[i] = get_bits(&s->gb, 3);
1360 init_short_region(s, g);
1362 int region_address1, region_address2;
1364 g->switch_point = 0;
1365 for (i = 0; i < 3; i++)
1366 g->table_select[i] = get_bits(&s->gb, 5);
1367 /* compute huffman coded region sizes */
1368 region_address1 = get_bits(&s->gb, 4);
1369 region_address2 = get_bits(&s->gb, 3);
1370 ff_dlog(s->avctx, "region1=%d region2=%d\n",
1371 region_address1, region_address2);
1372 init_long_region(s, g, region_address1, region_address2);
1374 region_offset2size(g);
1375 compute_band_indexes(s, g);
1379 g->preflag = get_bits1(&s->gb);
1380 g->scalefac_scale = get_bits1(&s->gb);
1381 g->count1table_select = get_bits1(&s->gb);
1382 ff_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1383 g->block_type, g->switch_point);
1389 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1390 int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0, EXTRABYTES);
1391 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1392 /* now we get bits from the main_data_begin offset */
1393 ff_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1394 main_data_begin, s->last_buf_size);
1396 memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);
1398 init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
1399 #if !UNCHECKED_BITSTREAM_READER
1400 s->gb.size_in_bits_plus8 += FFMAX(extrasize, LAST_BUF_SIZE - s->last_buf_size) * 8;
1402 s->last_buf_size <<= 3;
1403 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1404 for (ch = 0; ch < s->nb_channels; ch++) {
1405 g = &s->granules[ch][gr];
1406 s->last_buf_size += g->part2_3_length;
1407 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1408 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1411 skip = s->last_buf_size - 8 * main_data_begin;
1412 if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {
1413 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);
1415 s->in_gb.buffer = NULL;
1417 skip_bits_long(&s->gb, skip);
1423 for (; gr < nb_granules; gr++) {
1424 for (ch = 0; ch < s->nb_channels; ch++) {
1425 g = &s->granules[ch][gr];
1426 bits_pos = get_bits_count(&s->gb);
1430 int slen, slen1, slen2;
1432 /* MPEG1 scale factors */
1433 slen1 = slen_table[0][g->scalefac_compress];
1434 slen2 = slen_table[1][g->scalefac_compress];
1435 ff_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1436 if (g->block_type == 2) {
1437 n = g->switch_point ? 17 : 18;
1440 for (i = 0; i < n; i++)
1441 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1443 for (i = 0; i < n; i++)
1444 g->scale_factors[j++] = 0;
1447 for (i = 0; i < 18; i++)
1448 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1449 for (i = 0; i < 3; i++)
1450 g->scale_factors[j++] = 0;
1452 for (i = 0; i < 21; i++)
1453 g->scale_factors[j++] = 0;
1456 sc = s->granules[ch][0].scale_factors;
1458 for (k = 0; k < 4; k++) {
1460 if ((g->scfsi & (0x8 >> k)) == 0) {
1461 slen = (k < 2) ? slen1 : slen2;
1463 for (i = 0; i < n; i++)
1464 g->scale_factors[j++] = get_bits(&s->gb, slen);
1466 for (i = 0; i < n; i++)
1467 g->scale_factors[j++] = 0;
1470 /* simply copy from last granule */
1471 for (i = 0; i < n; i++) {
1472 g->scale_factors[j] = sc[j];
1477 g->scale_factors[j++] = 0;
1480 int tindex, tindex2, slen[4], sl, sf;
1482 /* LSF scale factors */
1483 if (g->block_type == 2)
1484 tindex = g->switch_point ? 2 : 1;
1488 sf = g->scalefac_compress;
1489 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1490 /* intensity stereo case */
1493 lsf_sf_expand(slen, sf, 6, 6, 0);
1495 } else if (sf < 244) {
1496 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1499 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1505 lsf_sf_expand(slen, sf, 5, 4, 4);
1507 } else if (sf < 500) {
1508 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1511 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1518 for (k = 0; k < 4; k++) {
1519 n = lsf_nsf_table[tindex2][tindex][k];
1522 for (i = 0; i < n; i++)
1523 g->scale_factors[j++] = get_bits(&s->gb, sl);
1525 for (i = 0; i < n; i++)
1526 g->scale_factors[j++] = 0;
1529 /* XXX: should compute exact size */
1531 g->scale_factors[j] = 0;
1534 exponents_from_scale_factors(s, g, exponents);
1536 /* read Huffman coded residue */
1537 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1540 if (s->mode == MPA_JSTEREO)
1541 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1543 for (ch = 0; ch < s->nb_channels; ch++) {
1544 g = &s->granules[ch][gr];
1546 reorder_block(s, g);
1547 compute_antialias(s, g);
1548 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1551 if (get_bits_count(&s->gb) < 0)
1552 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1553 return nb_granules * 18;
1556 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1557 const uint8_t *buf, int buf_size)
1559 int i, nb_frames, ch, ret;
1560 OUT_INT *samples_ptr;
1562 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1564 /* skip error protection field */
1565 if (s->error_protection)
1566 skip_bits(&s->gb, 16);
1570 s->avctx->frame_size = 384;
1571 nb_frames = mp_decode_layer1(s);
1574 s->avctx->frame_size = 1152;
1575 nb_frames = mp_decode_layer2(s);
1578 s->avctx->frame_size = s->lsf ? 576 : 1152;
1580 nb_frames = mp_decode_layer3(s);
1583 if (s->in_gb.buffer) {
1584 align_get_bits(&s->gb);
1585 i = get_bits_left(&s->gb)>>3;
1586 if (i >= 0 && i <= BACKSTEP_SIZE) {
1587 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1590 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1592 s->in_gb.buffer = NULL;
1595 align_get_bits(&s->gb);
1596 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1597 i = get_bits_left(&s->gb) >> 3;
1599 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1601 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1602 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1604 av_assert1(i <= buf_size - HEADER_SIZE && i >= 0);
1605 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1606 s->last_buf_size += i;
1612 /* get output buffer */
1614 av_assert0(s->frame);
1615 s->frame->nb_samples = s->avctx->frame_size;
1616 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0)
1618 samples = (OUT_INT **)s->frame->extended_data;
1621 /* apply the synthesis filter */
1622 for (ch = 0; ch < s->nb_channels; ch++) {
1624 if (s->avctx->sample_fmt == OUT_FMT_P) {
1625 samples_ptr = samples[ch];
1628 samples_ptr = samples[0] + ch;
1629 sample_stride = s->nb_channels;
1631 for (i = 0; i < nb_frames; i++) {
1632 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1633 &(s->synth_buf_offset[ch]),
1634 RENAME(ff_mpa_synth_window),
1635 &s->dither_state, samples_ptr,
1636 sample_stride, s->sb_samples[ch][i]);
1637 samples_ptr += 32 * sample_stride;
1641 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1644 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1647 const uint8_t *buf = avpkt->data;
1648 int buf_size = avpkt->size;
1649 MPADecodeContext *s = avctx->priv_data;
1654 while(buf_size && !*buf){
1660 if (buf_size < HEADER_SIZE)
1661 return AVERROR_INVALIDDATA;
1663 header = AV_RB32(buf);
1664 if (header>>8 == AV_RB32("TAG")>>8) {
1665 av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n");
1668 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1670 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1671 return AVERROR_INVALIDDATA;
1672 } else if (ret == 1) {
1673 /* free format: prepare to compute frame size */
1675 return AVERROR_INVALIDDATA;
1677 /* update codec info */
1678 avctx->channels = s->nb_channels;
1679 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1680 if (!avctx->bit_rate)
1681 avctx->bit_rate = s->bit_rate;
1683 if (s->frame_size <= 0) {
1684 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1685 return AVERROR_INVALIDDATA;
1686 } else if (s->frame_size < buf_size) {
1687 av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");
1688 buf_size= s->frame_size;
1693 ret = mp_decode_frame(s, NULL, buf, buf_size);
1695 s->frame->nb_samples = avctx->frame_size;
1697 avctx->sample_rate = s->sample_rate;
1698 //FIXME maybe move the other codec info stuff from above here too
1700 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1701 /* Only return an error if the bad frame makes up the whole packet or
1702 * the error is related to buffer management.
1703 * If there is more data in the packet, just consume the bad frame
1704 * instead of returning an error, which would discard the whole
1707 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1711 return buf_size + skipped;
1714 static void mp_flush(MPADecodeContext *ctx)
1716 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1717 memset(ctx->mdct_buf, 0, sizeof(ctx->mdct_buf));
1718 ctx->last_buf_size = 0;
1719 ctx->dither_state = 0;
1722 static void flush(AVCodecContext *avctx)
1724 mp_flush(avctx->priv_data);
1727 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1728 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1729 int *got_frame_ptr, AVPacket *avpkt)
1731 const uint8_t *buf = avpkt->data;
1732 int buf_size = avpkt->size;
1733 MPADecodeContext *s = avctx->priv_data;
1736 int av_unused out_size;
1740 // Discard too short frames
1741 if (buf_size < HEADER_SIZE) {
1742 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1743 return AVERROR_INVALIDDATA;
1747 if (len > MPA_MAX_CODED_FRAME_SIZE)
1748 len = MPA_MAX_CODED_FRAME_SIZE;
1750 // Get header and restore sync word
1751 header = AV_RB32(buf) | 0xffe00000;
1753 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1755 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1758 /* update codec info */
1759 avctx->sample_rate = s->sample_rate;
1760 avctx->channels = s->nb_channels;
1761 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1762 if (!avctx->bit_rate)
1763 avctx->bit_rate = s->bit_rate;
1765 s->frame_size = len;
1769 ret = mp_decode_frame(s, NULL, buf, buf_size);
1771 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1779 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1781 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1784 * Context for MP3On4 decoder
1786 typedef struct MP3On4DecodeContext {
1787 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1788 int syncword; ///< syncword patch
1789 const uint8_t *coff; ///< channel offsets in output buffer
1790 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1791 } MP3On4DecodeContext;
1793 #include "mpeg4audio.h"
1795 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1797 /* number of mp3 decoder instances */
1798 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1800 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1801 static const uint8_t chan_offset[8][5] = {
1806 { 2, 0, 3 }, // C FLR BS
1807 { 2, 0, 3 }, // C FLR BLRS
1808 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1809 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1812 /* mp3on4 channel layouts */
1813 static const int16_t chan_layout[8] = {
1816 AV_CH_LAYOUT_STEREO,
1817 AV_CH_LAYOUT_SURROUND,
1818 AV_CH_LAYOUT_4POINT0,
1819 AV_CH_LAYOUT_5POINT0,
1820 AV_CH_LAYOUT_5POINT1,
1821 AV_CH_LAYOUT_7POINT1
1824 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1826 MP3On4DecodeContext *s = avctx->priv_data;
1829 for (i = 0; i < s->frames; i++)
1830 av_freep(&s->mp3decctx[i]);
1836 static av_cold int decode_init_mp3on4(AVCodecContext * avctx)
1838 MP3On4DecodeContext *s = avctx->priv_data;
1839 MPEG4AudioConfig cfg;
1842 if ((avctx->extradata_size < 2) || !avctx->extradata) {
1843 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1844 return AVERROR_INVALIDDATA;
1847 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
1848 avctx->extradata_size * 8, 1);
1849 if (!cfg.chan_config || cfg.chan_config > 7) {
1850 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1851 return AVERROR_INVALIDDATA;
1853 s->frames = mp3Frames[cfg.chan_config];
1854 s->coff = chan_offset[cfg.chan_config];
1855 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1856 avctx->channel_layout = chan_layout[cfg.chan_config];
1858 if (cfg.sample_rate < 16000)
1859 s->syncword = 0xffe00000;
1861 s->syncword = 0xfff00000;
1863 /* Init the first mp3 decoder in standard way, so that all tables get builded
1864 * We replace avctx->priv_data with the context of the first decoder so that
1865 * decode_init() does not have to be changed.
1866 * Other decoders will be initialized here copying data from the first context
1868 // Allocate zeroed memory for the first decoder context
1869 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1870 if (!s->mp3decctx[0])
1872 // Put decoder context in place to make init_decode() happy
1873 avctx->priv_data = s->mp3decctx[0];
1875 // Restore mp3on4 context pointer
1876 avctx->priv_data = s;
1877 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1879 /* Create a separate codec/context for each frame (first is already ok).
1880 * Each frame is 1 or 2 channels - up to 5 frames allowed
1882 for (i = 1; i < s->frames; i++) {
1883 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1884 if (!s->mp3decctx[i])
1886 s->mp3decctx[i]->adu_mode = 1;
1887 s->mp3decctx[i]->avctx = avctx;
1888 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1889 s->mp3decctx[i]->fdsp = s->mp3decctx[0]->fdsp;
1894 decode_close_mp3on4(avctx);
1895 return AVERROR(ENOMEM);
1899 static void flush_mp3on4(AVCodecContext *avctx)
1902 MP3On4DecodeContext *s = avctx->priv_data;
1904 for (i = 0; i < s->frames; i++)
1905 mp_flush(s->mp3decctx[i]);
1909 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1910 int *got_frame_ptr, AVPacket *avpkt)
1912 AVFrame *frame = data;
1913 const uint8_t *buf = avpkt->data;
1914 int buf_size = avpkt->size;
1915 MP3On4DecodeContext *s = avctx->priv_data;
1916 MPADecodeContext *m;
1917 int fsize, len = buf_size, out_size = 0;
1919 OUT_INT **out_samples;
1923 /* get output buffer */
1924 frame->nb_samples = MPA_FRAME_SIZE;
1925 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1927 out_samples = (OUT_INT **)frame->extended_data;
1929 // Discard too short frames
1930 if (buf_size < HEADER_SIZE)
1931 return AVERROR_INVALIDDATA;
1933 avctx->bit_rate = 0;
1936 for (fr = 0; fr < s->frames; fr++) {
1937 fsize = AV_RB16(buf) >> 4;
1938 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1939 m = s->mp3decctx[fr];
1942 if (fsize < HEADER_SIZE) {
1943 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1944 return AVERROR_INVALIDDATA;
1946 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1948 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1950 av_log(avctx, AV_LOG_ERROR, "Bad header, discard block\n");
1951 return AVERROR_INVALIDDATA;
1954 if (ch + m->nb_channels > avctx->channels ||
1955 s->coff[fr] + m->nb_channels > avctx->channels) {
1956 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1958 return AVERROR_INVALIDDATA;
1960 ch += m->nb_channels;
1962 outptr[0] = out_samples[s->coff[fr]];
1963 if (m->nb_channels > 1)
1964 outptr[1] = out_samples[s->coff[fr] + 1];
1966 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0) {
1967 av_log(avctx, AV_LOG_ERROR, "failed to decode channel %d\n", ch);
1968 memset(outptr[0], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1969 if (m->nb_channels > 1)
1970 memset(outptr[1], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1971 ret = m->nb_channels * MPA_FRAME_SIZE*sizeof(OUT_INT);
1978 avctx->bit_rate += m->bit_rate;
1980 if (ch != avctx->channels) {
1981 av_log(avctx, AV_LOG_ERROR, "failed to decode all channels\n");
1982 return AVERROR_INVALIDDATA;
1985 /* update codec info */
1986 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1988 frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1993 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */