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 & 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 /* handle n = 0 too */
820 static inline int get_bitsz(GetBitContext *s, int n)
822 return n ? get_bits(s, n) : 0;
826 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
829 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits) {
831 s->in_gb.buffer = NULL;
832 av_assert2((get_bits_count(&s->gb) & 7) == 0);
833 skip_bits_long(&s->gb, *pos - *end_pos);
835 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
836 *pos = get_bits_count(&s->gb);
840 /* Following is a optimized code for
842 if(get_bits1(&s->gb))
847 #define READ_FLIP_SIGN(dst,src) \
848 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
851 #define READ_FLIP_SIGN(dst,src) \
852 v = -get_bits1(&s->gb); \
853 *(dst) = (*(src) ^ v) - v;
856 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
857 int16_t *exponents, int end_pos2)
861 int last_pos, bits_left;
863 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits);
865 /* low frequencies (called big values) */
867 for (i = 0; i < 3; i++) {
868 int j, k, l, linbits;
869 j = g->region_size[i];
872 /* select vlc table */
873 k = g->table_select[i];
874 l = mpa_huff_data[k][0];
875 linbits = mpa_huff_data[k][1];
879 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
884 /* read huffcode and compute each couple */
888 int pos = get_bits_count(&s->gb);
891 switch_buffer(s, &pos, &end_pos, &end_pos2);
895 y = get_vlc2(&s->gb, vlc->table, 7, 3);
898 g->sb_hybrid[s_index ] =
899 g->sb_hybrid[s_index+1] = 0;
904 exponent= exponents[s_index];
906 ff_dlog(s->avctx, "region=%d n=%d x=%d y=%d exp=%d\n",
907 i, g->region_size[i] - j, x, y, exponent);
912 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
914 x += get_bitsz(&s->gb, linbits);
915 v = l3_unscale(x, exponent);
916 if (get_bits1(&s->gb))
918 g->sb_hybrid[s_index] = v;
921 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
923 y += get_bitsz(&s->gb, linbits);
924 v = l3_unscale(y, exponent);
925 if (get_bits1(&s->gb))
927 g->sb_hybrid[s_index+1] = v;
934 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
936 x += get_bitsz(&s->gb, linbits);
937 v = l3_unscale(x, exponent);
938 if (get_bits1(&s->gb))
940 g->sb_hybrid[s_index+!!y] = v;
942 g->sb_hybrid[s_index + !y] = 0;
948 /* high frequencies */
949 vlc = &huff_quad_vlc[g->count1table_select];
951 while (s_index <= 572) {
953 pos = get_bits_count(&s->gb);
954 if (pos >= end_pos) {
955 if (pos > end_pos2 && last_pos) {
956 /* some encoders generate an incorrect size for this
957 part. We must go back into the data */
959 skip_bits_long(&s->gb, last_pos - pos);
960 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
961 if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT))
965 switch_buffer(s, &pos, &end_pos, &end_pos2);
971 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
972 ff_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
973 g->sb_hybrid[s_index+0] =
974 g->sb_hybrid[s_index+1] =
975 g->sb_hybrid[s_index+2] =
976 g->sb_hybrid[s_index+3] = 0;
978 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
980 int pos = s_index + idxtab[code];
981 code ^= 8 >> idxtab[code];
982 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
986 /* skip extension bits */
987 bits_left = end_pos2 - get_bits_count(&s->gb);
988 if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) {
989 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
991 } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) {
992 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
995 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
996 skip_bits_long(&s->gb, bits_left);
998 i = get_bits_count(&s->gb);
999 switch_buffer(s, &i, &end_pos, &end_pos2);
1004 /* Reorder short blocks from bitstream order to interleaved order. It
1005 would be faster to do it in parsing, but the code would be far more
1007 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1010 INTFLOAT *ptr, *dst, *ptr1;
1013 if (g->block_type != 2)
1016 if (g->switch_point) {
1017 if (s->sample_rate_index != 8)
1018 ptr = g->sb_hybrid + 36;
1020 ptr = g->sb_hybrid + 72;
1025 for (i = g->short_start; i < 13; i++) {
1026 len = band_size_short[s->sample_rate_index][i];
1029 for (j = len; j > 0; j--) {
1030 *dst++ = ptr[0*len];
1031 *dst++ = ptr[1*len];
1032 *dst++ = ptr[2*len];
1036 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1040 #define ISQRT2 FIXR(0.70710678118654752440)
1042 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1045 int sf_max, sf, len, non_zero_found;
1046 INTFLOAT (*is_tab)[16], *tab0, *tab1, tmp0, tmp1, v1, v2;
1047 int non_zero_found_short[3];
1049 /* intensity stereo */
1050 if (s->mode_ext & MODE_EXT_I_STEREO) {
1055 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1059 tab0 = g0->sb_hybrid + 576;
1060 tab1 = g1->sb_hybrid + 576;
1062 non_zero_found_short[0] = 0;
1063 non_zero_found_short[1] = 0;
1064 non_zero_found_short[2] = 0;
1065 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1066 for (i = 12; i >= g1->short_start; i--) {
1067 /* for last band, use previous scale factor */
1070 len = band_size_short[s->sample_rate_index][i];
1071 for (l = 2; l >= 0; l--) {
1074 if (!non_zero_found_short[l]) {
1075 /* test if non zero band. if so, stop doing i-stereo */
1076 for (j = 0; j < len; j++) {
1078 non_zero_found_short[l] = 1;
1082 sf = g1->scale_factors[k + l];
1088 for (j = 0; j < len; j++) {
1090 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1091 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1095 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1096 /* lower part of the spectrum : do ms stereo
1098 for (j = 0; j < len; j++) {
1101 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1102 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1109 non_zero_found = non_zero_found_short[0] |
1110 non_zero_found_short[1] |
1111 non_zero_found_short[2];
1113 for (i = g1->long_end - 1;i >= 0;i--) {
1114 len = band_size_long[s->sample_rate_index][i];
1117 /* test if non zero band. if so, stop doing i-stereo */
1118 if (!non_zero_found) {
1119 for (j = 0; j < len; j++) {
1125 /* for last band, use previous scale factor */
1126 k = (i == 21) ? 20 : i;
1127 sf = g1->scale_factors[k];
1132 for (j = 0; j < len; j++) {
1134 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1135 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1139 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1140 /* lower part of the spectrum : do ms stereo
1142 for (j = 0; j < len; j++) {
1145 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1146 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1151 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1152 /* ms stereo ONLY */
1153 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1156 s->fdsp->butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1158 tab0 = g0->sb_hybrid;
1159 tab1 = g1->sb_hybrid;
1160 for (i = 0; i < 576; i++) {
1163 tab0[i] = tmp0 + tmp1;
1164 tab1[i] = tmp0 - tmp1;
1172 # include "mips/compute_antialias_float.h"
1173 #endif /* HAVE_MIPSFPU */
1176 # include "mips/compute_antialias_fixed.h"
1177 #endif /* HAVE_MIPSDSPR1 */
1178 #endif /* USE_FLOATS */
1180 #ifndef compute_antialias
1182 #define AA(j) do { \
1183 float tmp0 = ptr[-1-j]; \
1184 float tmp1 = ptr[ j]; \
1185 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1186 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1189 #define AA(j) do { \
1190 int tmp0 = ptr[-1-j]; \
1191 int tmp1 = ptr[ j]; \
1192 int tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1193 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1194 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1198 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1203 /* we antialias only "long" bands */
1204 if (g->block_type == 2) {
1205 if (!g->switch_point)
1207 /* XXX: check this for 8000Hz case */
1213 ptr = g->sb_hybrid + 18;
1214 for (i = n; i > 0; i--) {
1227 #endif /* compute_antialias */
1229 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1230 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1232 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1234 int i, j, mdct_long_end, sblimit;
1236 /* find last non zero block */
1237 ptr = g->sb_hybrid + 576;
1238 ptr1 = g->sb_hybrid + 2 * 18;
1239 while (ptr >= ptr1) {
1243 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1246 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1248 if (g->block_type == 2) {
1249 /* XXX: check for 8000 Hz */
1250 if (g->switch_point)
1255 mdct_long_end = sblimit;
1258 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1259 mdct_long_end, g->switch_point,
1262 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1263 ptr = g->sb_hybrid + 18 * mdct_long_end;
1265 for (j = mdct_long_end; j < sblimit; j++) {
1266 /* select frequency inversion */
1267 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1268 out_ptr = sb_samples + j;
1270 for (i = 0; i < 6; i++) {
1271 *out_ptr = buf[4*i];
1274 imdct12(out2, ptr + 0);
1275 for (i = 0; i < 6; i++) {
1276 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1277 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1280 imdct12(out2, ptr + 1);
1281 for (i = 0; i < 6; i++) {
1282 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1283 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1286 imdct12(out2, ptr + 2);
1287 for (i = 0; i < 6; i++) {
1288 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1289 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1290 buf[4*(i + 6*2)] = 0;
1293 buf += (j&3) != 3 ? 1 : (4*18-3);
1296 for (j = sblimit; j < SBLIMIT; j++) {
1298 out_ptr = sb_samples + j;
1299 for (i = 0; i < 18; i++) {
1300 *out_ptr = buf[4*i];
1304 buf += (j&3) != 3 ? 1 : (4*18-3);
1308 /* main layer3 decoding function */
1309 static int mp_decode_layer3(MPADecodeContext *s)
1311 int nb_granules, main_data_begin;
1312 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1314 int16_t exponents[576]; //FIXME try INTFLOAT
1316 /* read side info */
1318 main_data_begin = get_bits(&s->gb, 8);
1319 skip_bits(&s->gb, s->nb_channels);
1322 main_data_begin = get_bits(&s->gb, 9);
1323 if (s->nb_channels == 2)
1324 skip_bits(&s->gb, 3);
1326 skip_bits(&s->gb, 5);
1328 for (ch = 0; ch < s->nb_channels; ch++) {
1329 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1330 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1334 for (gr = 0; gr < nb_granules; gr++) {
1335 for (ch = 0; ch < s->nb_channels; ch++) {
1336 ff_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1337 g = &s->granules[ch][gr];
1338 g->part2_3_length = get_bits(&s->gb, 12);
1339 g->big_values = get_bits(&s->gb, 9);
1340 if (g->big_values > 288) {
1341 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1342 return AVERROR_INVALIDDATA;
1345 g->global_gain = get_bits(&s->gb, 8);
1346 /* if MS stereo only is selected, we precompute the
1347 1/sqrt(2) renormalization factor */
1348 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1350 g->global_gain -= 2;
1352 g->scalefac_compress = get_bits(&s->gb, 9);
1354 g->scalefac_compress = get_bits(&s->gb, 4);
1355 blocksplit_flag = get_bits1(&s->gb);
1356 if (blocksplit_flag) {
1357 g->block_type = get_bits(&s->gb, 2);
1358 if (g->block_type == 0) {
1359 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1360 return AVERROR_INVALIDDATA;
1362 g->switch_point = get_bits1(&s->gb);
1363 for (i = 0; i < 2; i++)
1364 g->table_select[i] = get_bits(&s->gb, 5);
1365 for (i = 0; i < 3; i++)
1366 g->subblock_gain[i] = get_bits(&s->gb, 3);
1367 init_short_region(s, g);
1369 int region_address1, region_address2;
1371 g->switch_point = 0;
1372 for (i = 0; i < 3; i++)
1373 g->table_select[i] = get_bits(&s->gb, 5);
1374 /* compute huffman coded region sizes */
1375 region_address1 = get_bits(&s->gb, 4);
1376 region_address2 = get_bits(&s->gb, 3);
1377 ff_dlog(s->avctx, "region1=%d region2=%d\n",
1378 region_address1, region_address2);
1379 init_long_region(s, g, region_address1, region_address2);
1381 region_offset2size(g);
1382 compute_band_indexes(s, g);
1386 g->preflag = get_bits1(&s->gb);
1387 g->scalefac_scale = get_bits1(&s->gb);
1388 g->count1table_select = get_bits1(&s->gb);
1389 ff_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1390 g->block_type, g->switch_point);
1396 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1397 int extrasize = av_clip(get_bits_left(&s->gb) >> 3, 0, EXTRABYTES);
1398 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1399 /* now we get bits from the main_data_begin offset */
1400 ff_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1401 main_data_begin, s->last_buf_size);
1403 memcpy(s->last_buf + s->last_buf_size, ptr, extrasize);
1405 init_get_bits(&s->gb, s->last_buf, s->last_buf_size*8);
1406 #if !UNCHECKED_BITSTREAM_READER
1407 s->gb.size_in_bits_plus8 += FFMAX(extrasize, LAST_BUF_SIZE - s->last_buf_size) * 8;
1409 s->last_buf_size <<= 3;
1410 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1411 for (ch = 0; ch < s->nb_channels; ch++) {
1412 g = &s->granules[ch][gr];
1413 s->last_buf_size += g->part2_3_length;
1414 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1415 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1418 skip = s->last_buf_size - 8 * main_data_begin;
1419 if (skip >= s->gb.size_in_bits && s->in_gb.buffer) {
1420 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits);
1422 s->in_gb.buffer = NULL;
1424 skip_bits_long(&s->gb, skip);
1430 for (; gr < nb_granules; gr++) {
1431 for (ch = 0; ch < s->nb_channels; ch++) {
1432 g = &s->granules[ch][gr];
1433 bits_pos = get_bits_count(&s->gb);
1437 int slen, slen1, slen2;
1439 /* MPEG1 scale factors */
1440 slen1 = slen_table[0][g->scalefac_compress];
1441 slen2 = slen_table[1][g->scalefac_compress];
1442 ff_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1443 if (g->block_type == 2) {
1444 n = g->switch_point ? 17 : 18;
1447 for (i = 0; i < n; i++)
1448 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1450 for (i = 0; i < n; i++)
1451 g->scale_factors[j++] = 0;
1454 for (i = 0; i < 18; i++)
1455 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1456 for (i = 0; i < 3; i++)
1457 g->scale_factors[j++] = 0;
1459 for (i = 0; i < 21; i++)
1460 g->scale_factors[j++] = 0;
1463 sc = s->granules[ch][0].scale_factors;
1465 for (k = 0; k < 4; k++) {
1467 if ((g->scfsi & (0x8 >> k)) == 0) {
1468 slen = (k < 2) ? slen1 : slen2;
1470 for (i = 0; i < n; i++)
1471 g->scale_factors[j++] = get_bits(&s->gb, slen);
1473 for (i = 0; i < n; i++)
1474 g->scale_factors[j++] = 0;
1477 /* simply copy from last granule */
1478 for (i = 0; i < n; i++) {
1479 g->scale_factors[j] = sc[j];
1484 g->scale_factors[j++] = 0;
1487 int tindex, tindex2, slen[4], sl, sf;
1489 /* LSF scale factors */
1490 if (g->block_type == 2)
1491 tindex = g->switch_point ? 2 : 1;
1495 sf = g->scalefac_compress;
1496 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1497 /* intensity stereo case */
1500 lsf_sf_expand(slen, sf, 6, 6, 0);
1502 } else if (sf < 244) {
1503 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1506 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1512 lsf_sf_expand(slen, sf, 5, 4, 4);
1514 } else if (sf < 500) {
1515 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1518 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1525 for (k = 0; k < 4; k++) {
1526 n = lsf_nsf_table[tindex2][tindex][k];
1529 for (i = 0; i < n; i++)
1530 g->scale_factors[j++] = get_bits(&s->gb, sl);
1532 for (i = 0; i < n; i++)
1533 g->scale_factors[j++] = 0;
1536 /* XXX: should compute exact size */
1538 g->scale_factors[j] = 0;
1541 exponents_from_scale_factors(s, g, exponents);
1543 /* read Huffman coded residue */
1544 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1547 if (s->mode == MPA_JSTEREO)
1548 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1550 for (ch = 0; ch < s->nb_channels; ch++) {
1551 g = &s->granules[ch][gr];
1553 reorder_block(s, g);
1554 compute_antialias(s, g);
1555 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1558 if (get_bits_count(&s->gb) < 0)
1559 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1560 return nb_granules * 18;
1563 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1564 const uint8_t *buf, int buf_size)
1566 int i, nb_frames, ch, ret;
1567 OUT_INT *samples_ptr;
1569 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1571 /* skip error protection field */
1572 if (s->error_protection)
1573 skip_bits(&s->gb, 16);
1577 s->avctx->frame_size = 384;
1578 nb_frames = mp_decode_layer1(s);
1581 s->avctx->frame_size = 1152;
1582 nb_frames = mp_decode_layer2(s);
1585 s->avctx->frame_size = s->lsf ? 576 : 1152;
1587 nb_frames = mp_decode_layer3(s);
1590 if (s->in_gb.buffer) {
1591 align_get_bits(&s->gb);
1592 i = get_bits_left(&s->gb)>>3;
1593 if (i >= 0 && i <= BACKSTEP_SIZE) {
1594 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1597 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1599 s->in_gb.buffer = NULL;
1602 align_get_bits(&s->gb);
1603 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1604 i = get_bits_left(&s->gb) >> 3;
1606 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1608 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1609 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1611 av_assert1(i <= buf_size - HEADER_SIZE && i >= 0);
1612 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1613 s->last_buf_size += i;
1619 /* get output buffer */
1621 av_assert0(s->frame);
1622 s->frame->nb_samples = s->avctx->frame_size;
1623 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0)
1625 samples = (OUT_INT **)s->frame->extended_data;
1628 /* apply the synthesis filter */
1629 for (ch = 0; ch < s->nb_channels; ch++) {
1631 if (s->avctx->sample_fmt == OUT_FMT_P) {
1632 samples_ptr = samples[ch];
1635 samples_ptr = samples[0] + ch;
1636 sample_stride = s->nb_channels;
1638 for (i = 0; i < nb_frames; i++) {
1639 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1640 &(s->synth_buf_offset[ch]),
1641 RENAME(ff_mpa_synth_window),
1642 &s->dither_state, samples_ptr,
1643 sample_stride, s->sb_samples[ch][i]);
1644 samples_ptr += 32 * sample_stride;
1648 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1651 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1654 const uint8_t *buf = avpkt->data;
1655 int buf_size = avpkt->size;
1656 MPADecodeContext *s = avctx->priv_data;
1660 while(buf_size && !*buf){
1665 if (buf_size < HEADER_SIZE)
1666 return AVERROR_INVALIDDATA;
1668 header = AV_RB32(buf);
1669 if (header>>8 == AV_RB32("TAG")>>8) {
1670 av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n");
1673 if (ff_mpa_check_header(header) < 0) {
1674 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1675 return AVERROR_INVALIDDATA;
1678 if (avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header) == 1) {
1679 /* free format: prepare to compute frame size */
1681 return AVERROR_INVALIDDATA;
1683 /* update codec info */
1684 avctx->channels = s->nb_channels;
1685 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1686 if (!avctx->bit_rate)
1687 avctx->bit_rate = s->bit_rate;
1689 if (s->frame_size <= 0 || s->frame_size > buf_size) {
1690 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1691 return AVERROR_INVALIDDATA;
1692 } else if (s->frame_size < buf_size) {
1693 av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");
1694 buf_size= s->frame_size;
1699 ret = mp_decode_frame(s, NULL, buf, buf_size);
1701 s->frame->nb_samples = avctx->frame_size;
1703 avctx->sample_rate = s->sample_rate;
1704 //FIXME maybe move the other codec info stuff from above here too
1706 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1707 /* Only return an error if the bad frame makes up the whole packet or
1708 * the error is related to buffer management.
1709 * If there is more data in the packet, just consume the bad frame
1710 * instead of returning an error, which would discard the whole
1713 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1720 static void mp_flush(MPADecodeContext *ctx)
1722 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1723 memset(ctx->mdct_buf, 0, sizeof(ctx->mdct_buf));
1724 ctx->last_buf_size = 0;
1725 ctx->dither_state = 0;
1728 static void flush(AVCodecContext *avctx)
1730 mp_flush(avctx->priv_data);
1733 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1734 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1735 int *got_frame_ptr, AVPacket *avpkt)
1737 const uint8_t *buf = avpkt->data;
1738 int buf_size = avpkt->size;
1739 MPADecodeContext *s = avctx->priv_data;
1742 int av_unused out_size;
1746 // Discard too short frames
1747 if (buf_size < HEADER_SIZE) {
1748 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1749 return AVERROR_INVALIDDATA;
1753 if (len > MPA_MAX_CODED_FRAME_SIZE)
1754 len = MPA_MAX_CODED_FRAME_SIZE;
1756 // Get header and restore sync word
1757 header = AV_RB32(buf) | 0xffe00000;
1759 if (ff_mpa_check_header(header) < 0) { // Bad header, discard frame
1760 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1761 return AVERROR_INVALIDDATA;
1764 avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1765 /* update codec info */
1766 avctx->sample_rate = s->sample_rate;
1767 avctx->channels = s->nb_channels;
1768 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1769 if (!avctx->bit_rate)
1770 avctx->bit_rate = s->bit_rate;
1772 s->frame_size = len;
1776 ret = mp_decode_frame(s, NULL, buf, buf_size);
1778 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1786 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1788 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1791 * Context for MP3On4 decoder
1793 typedef struct MP3On4DecodeContext {
1794 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1795 int syncword; ///< syncword patch
1796 const uint8_t *coff; ///< channel offsets in output buffer
1797 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1798 } MP3On4DecodeContext;
1800 #include "mpeg4audio.h"
1802 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1804 /* number of mp3 decoder instances */
1805 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1807 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1808 static const uint8_t chan_offset[8][5] = {
1813 { 2, 0, 3 }, // C FLR BS
1814 { 2, 0, 3 }, // C FLR BLRS
1815 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1816 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1819 /* mp3on4 channel layouts */
1820 static const int16_t chan_layout[8] = {
1823 AV_CH_LAYOUT_STEREO,
1824 AV_CH_LAYOUT_SURROUND,
1825 AV_CH_LAYOUT_4POINT0,
1826 AV_CH_LAYOUT_5POINT0,
1827 AV_CH_LAYOUT_5POINT1,
1828 AV_CH_LAYOUT_7POINT1
1831 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1833 MP3On4DecodeContext *s = avctx->priv_data;
1836 for (i = 0; i < s->frames; i++)
1837 av_freep(&s->mp3decctx[i]);
1843 static av_cold int decode_init_mp3on4(AVCodecContext * avctx)
1845 MP3On4DecodeContext *s = avctx->priv_data;
1846 MPEG4AudioConfig cfg;
1849 if ((avctx->extradata_size < 2) || !avctx->extradata) {
1850 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1851 return AVERROR_INVALIDDATA;
1854 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
1855 avctx->extradata_size * 8, 1);
1856 if (!cfg.chan_config || cfg.chan_config > 7) {
1857 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1858 return AVERROR_INVALIDDATA;
1860 s->frames = mp3Frames[cfg.chan_config];
1861 s->coff = chan_offset[cfg.chan_config];
1862 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1863 avctx->channel_layout = chan_layout[cfg.chan_config];
1865 if (cfg.sample_rate < 16000)
1866 s->syncword = 0xffe00000;
1868 s->syncword = 0xfff00000;
1870 /* Init the first mp3 decoder in standard way, so that all tables get builded
1871 * We replace avctx->priv_data with the context of the first decoder so that
1872 * decode_init() does not have to be changed.
1873 * Other decoders will be initialized here copying data from the first context
1875 // Allocate zeroed memory for the first decoder context
1876 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1877 if (!s->mp3decctx[0])
1879 // Put decoder context in place to make init_decode() happy
1880 avctx->priv_data = s->mp3decctx[0];
1882 // Restore mp3on4 context pointer
1883 avctx->priv_data = s;
1884 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1886 /* Create a separate codec/context for each frame (first is already ok).
1887 * Each frame is 1 or 2 channels - up to 5 frames allowed
1889 for (i = 1; i < s->frames; i++) {
1890 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1891 if (!s->mp3decctx[i])
1893 s->mp3decctx[i]->adu_mode = 1;
1894 s->mp3decctx[i]->avctx = avctx;
1895 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1900 decode_close_mp3on4(avctx);
1901 return AVERROR(ENOMEM);
1905 static void flush_mp3on4(AVCodecContext *avctx)
1908 MP3On4DecodeContext *s = avctx->priv_data;
1910 for (i = 0; i < s->frames; i++)
1911 mp_flush(s->mp3decctx[i]);
1915 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1916 int *got_frame_ptr, AVPacket *avpkt)
1918 AVFrame *frame = data;
1919 const uint8_t *buf = avpkt->data;
1920 int buf_size = avpkt->size;
1921 MP3On4DecodeContext *s = avctx->priv_data;
1922 MPADecodeContext *m;
1923 int fsize, len = buf_size, out_size = 0;
1925 OUT_INT **out_samples;
1929 /* get output buffer */
1930 frame->nb_samples = MPA_FRAME_SIZE;
1931 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1933 out_samples = (OUT_INT **)frame->extended_data;
1935 // Discard too short frames
1936 if (buf_size < HEADER_SIZE)
1937 return AVERROR_INVALIDDATA;
1939 avctx->bit_rate = 0;
1942 for (fr = 0; fr < s->frames; fr++) {
1943 fsize = AV_RB16(buf) >> 4;
1944 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1945 m = s->mp3decctx[fr];
1948 if (fsize < HEADER_SIZE) {
1949 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1950 return AVERROR_INVALIDDATA;
1952 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1954 if (ff_mpa_check_header(header) < 0) {
1955 av_log(avctx, AV_LOG_ERROR, "Bad header, discard block\n");
1956 return AVERROR_INVALIDDATA;
1959 avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1961 if (ch + m->nb_channels > avctx->channels ||
1962 s->coff[fr] + m->nb_channels > avctx->channels) {
1963 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1965 return AVERROR_INVALIDDATA;
1967 ch += m->nb_channels;
1969 outptr[0] = out_samples[s->coff[fr]];
1970 if (m->nb_channels > 1)
1971 outptr[1] = out_samples[s->coff[fr] + 1];
1973 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0) {
1974 av_log(avctx, AV_LOG_ERROR, "failed to decode channel %d\n", ch);
1975 memset(outptr[0], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1976 if (m->nb_channels > 1)
1977 memset(outptr[1], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1978 ret = m->nb_channels * MPA_FRAME_SIZE*sizeof(OUT_INT);
1985 avctx->bit_rate += m->bit_rate;
1987 if (ch != avctx->channels) {
1988 av_log(avctx, AV_LOG_ERROR, "failed to decode all channels\n");
1989 return AVERROR_INVALIDDATA;
1992 /* update codec info */
1993 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1995 frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
2000 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */