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/crc.h"
31 #include "libavutil/float_dsp.h"
32 #include "libavutil/libm.h"
37 #include "mpegaudiodsp.h"
41 * - test lsf / mpeg25 extensively.
44 #include "mpegaudio.h"
45 #include "mpegaudiodecheader.h"
47 #define BACKSTEP_SIZE 512
49 #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES
51 /* layer 3 "granule" */
52 typedef struct GranuleDef {
57 int scalefac_compress;
62 uint8_t scalefac_scale;
63 uint8_t count1table_select;
64 int region_size[3]; /* number of huffman codes in each region */
66 int short_start, long_end; /* long/short band indexes */
67 uint8_t scale_factors[40];
68 DECLARE_ALIGNED(16, INTFLOAT, sb_hybrid)[SBLIMIT * 18]; /* 576 samples */
71 typedef struct MPADecodeContext {
73 uint8_t last_buf[LAST_BUF_SIZE];
76 /* next header (used in free format parsing) */
77 uint32_t free_format_next_header;
80 DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
81 int synth_buf_offset[MPA_MAX_CHANNELS];
82 DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
83 INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
84 GranuleDef granules[2][2]; /* Used in Layer 3 */
85 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
88 AVCodecContext* avctx;
90 AVFloatDSPContext *fdsp;
96 #include "mpegaudiodata.h"
97 #include "mpegaudiodectab.h"
99 /* vlc structure for decoding layer 3 huffman tables */
100 static VLC huff_vlc[16];
101 static VLC_TYPE huff_vlc_tables[
102 0 + 128 + 128 + 128 + 130 + 128 + 154 + 166 +
103 142 + 204 + 190 + 170 + 542 + 460 + 662 + 414
105 static const int huff_vlc_tables_sizes[16] = {
106 0, 128, 128, 128, 130, 128, 154, 166,
107 142, 204, 190, 170, 542, 460, 662, 414
109 static VLC huff_quad_vlc[2];
110 static VLC_TYPE huff_quad_vlc_tables[128+16][2];
111 static const int huff_quad_vlc_tables_sizes[2] = { 128, 16 };
112 /* computed from band_size_long */
113 static uint16_t band_index_long[9][23];
114 #include "mpegaudio_tablegen.h"
115 /* intensity stereo coef table */
116 static INTFLOAT is_table[2][16];
117 static INTFLOAT is_table_lsf[2][2][16];
118 static INTFLOAT csa_table[8][4];
120 static int16_t division_tab3[1<<6 ];
121 static int16_t division_tab5[1<<8 ];
122 static int16_t division_tab9[1<<11];
124 static int16_t * const division_tabs[4] = {
125 division_tab3, division_tab5, NULL, division_tab9
128 /* lower 2 bits: modulo 3, higher bits: shift */
129 static uint16_t scale_factor_modshift[64];
130 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
131 static int32_t scale_factor_mult[15][3];
132 /* mult table for layer 2 group quantization */
134 #define SCALE_GEN(v) \
135 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
137 static const int32_t scale_factor_mult2[3][3] = {
138 SCALE_GEN(4.0 / 3.0), /* 3 steps */
139 SCALE_GEN(4.0 / 5.0), /* 5 steps */
140 SCALE_GEN(4.0 / 9.0), /* 9 steps */
144 * Convert region offsets to region sizes and truncate
145 * size to big_values.
147 static void region_offset2size(GranuleDef *g)
150 g->region_size[2] = 576 / 2;
151 for (i = 0; i < 3; i++) {
152 k = FFMIN(g->region_size[i], g->big_values);
153 g->region_size[i] = k - j;
158 static void init_short_region(MPADecodeContext *s, GranuleDef *g)
160 if (g->block_type == 2) {
161 if (s->sample_rate_index != 8)
162 g->region_size[0] = (36 / 2);
164 g->region_size[0] = (72 / 2);
166 if (s->sample_rate_index <= 2)
167 g->region_size[0] = (36 / 2);
168 else if (s->sample_rate_index != 8)
169 g->region_size[0] = (54 / 2);
171 g->region_size[0] = (108 / 2);
173 g->region_size[1] = (576 / 2);
176 static void init_long_region(MPADecodeContext *s, GranuleDef *g,
180 g->region_size[0] = band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
181 /* should not overflow */
182 l = FFMIN(ra1 + ra2 + 2, 22);
183 g->region_size[1] = band_index_long[s->sample_rate_index][ l] >> 1;
186 static void compute_band_indexes(MPADecodeContext *s, GranuleDef *g)
188 if (g->block_type == 2) {
189 if (g->switch_point) {
190 if(s->sample_rate_index == 8)
191 avpriv_request_sample(s->avctx, "switch point in 8khz");
192 /* if switched mode, we handle the 36 first samples as
193 long blocks. For 8000Hz, we handle the 72 first
194 exponents as long blocks */
195 if (s->sample_rate_index <= 2)
211 /* layer 1 unscaling */
212 /* n = number of bits of the mantissa minus 1 */
213 static inline int l1_unscale(int n, int mant, int scale_factor)
218 shift = scale_factor_modshift[scale_factor];
221 val = MUL64((int)(mant + (-1U << n) + 1), scale_factor_mult[n-1][mod]);
223 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
224 return (int)((val + (1LL << (shift - 1))) >> shift);
227 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
231 shift = scale_factor_modshift[scale_factor];
235 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
236 /* NOTE: at this point, 0 <= shift <= 21 */
238 val = (val + (1 << (shift - 1))) >> shift;
242 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
243 static inline int l3_unscale(int value, int exponent)
248 e = table_4_3_exp [4 * value + (exponent & 3)];
249 m = table_4_3_value[4 * value + (exponent & 3)];
253 av_log(NULL, AV_LOG_WARNING, "l3_unscale: e is %d\n", e);
257 m = (m + ((1U << e)>>1)) >> e;
262 static av_cold void decode_init_static(void)
267 /* scale factors table for layer 1/2 */
268 for (i = 0; i < 64; i++) {
270 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
273 scale_factor_modshift[i] = mod | (shift << 2);
276 /* scale factor multiply for layer 1 */
277 for (i = 0; i < 15; i++) {
280 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
281 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
282 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
283 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
284 ff_dlog(NULL, "%d: norm=%x s=%"PRIx32" %"PRIx32" %"PRIx32"\n", i,
286 scale_factor_mult[i][0],
287 scale_factor_mult[i][1],
288 scale_factor_mult[i][2]);
291 RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
293 /* huffman decode tables */
295 for (i = 1; i < 16; i++) {
296 const HuffTable *h = &mpa_huff_tables[i];
298 uint8_t tmp_bits [512] = { 0 };
299 uint16_t tmp_codes[512] = { 0 };
304 for (x = 0; x < xsize; x++) {
305 for (y = 0; y < xsize; y++) {
306 tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ];
307 tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
312 huff_vlc[i].table = huff_vlc_tables+offset;
313 huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
314 init_vlc(&huff_vlc[i], 7, 512,
315 tmp_bits, 1, 1, tmp_codes, 2, 2,
316 INIT_VLC_USE_NEW_STATIC);
317 offset += huff_vlc_tables_sizes[i];
319 av_assert0(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
322 for (i = 0; i < 2; i++) {
323 huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
324 huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
325 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
326 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
327 INIT_VLC_USE_NEW_STATIC);
328 offset += huff_quad_vlc_tables_sizes[i];
330 av_assert0(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
332 for (i = 0; i < 9; i++) {
334 for (j = 0; j < 22; j++) {
335 band_index_long[i][j] = k;
336 k += band_size_long[i][j];
338 band_index_long[i][22] = k;
341 /* compute n ^ (4/3) and store it in mantissa/exp format */
343 mpegaudio_tableinit();
345 for (i = 0; i < 4; i++) {
346 if (ff_mpa_quant_bits[i] < 0) {
347 for (j = 0; j < (1 << (-ff_mpa_quant_bits[i]+1)); j++) {
348 int val1, val2, val3, steps;
350 steps = ff_mpa_quant_steps[i];
355 division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
361 for (i = 0; i < 7; i++) {
365 f = tan((double)i * M_PI / 12.0);
366 v = FIXR(f / (1.0 + f));
371 is_table[1][6 - i] = v;
374 for (i = 7; i < 16; i++)
375 is_table[0][i] = is_table[1][i] = 0.0;
377 for (i = 0; i < 16; i++) {
381 for (j = 0; j < 2; j++) {
382 e = -(j + 1) * ((i + 1) >> 1);
385 is_table_lsf[j][k ^ 1][i] = FIXR(f);
386 is_table_lsf[j][k ][i] = FIXR(1.0);
387 ff_dlog(NULL, "is_table_lsf %d %d: %f %f\n",
388 i, j, (float) is_table_lsf[j][0][i],
389 (float) is_table_lsf[j][1][i]);
393 for (i = 0; i < 8; i++) {
396 cs = 1.0 / sqrt(1.0 + ci * ci);
399 csa_table[i][0] = FIXHR(cs/4);
400 csa_table[i][1] = FIXHR(ca/4);
401 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
402 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
404 csa_table[i][0] = cs;
405 csa_table[i][1] = ca;
406 csa_table[i][2] = ca + cs;
407 csa_table[i][3] = ca - cs;
413 static av_cold int decode_close(AVCodecContext * avctx)
415 MPADecodeContext *s = avctx->priv_data;
422 static av_cold int decode_init(AVCodecContext * avctx)
424 static int initialized_tables = 0;
425 MPADecodeContext *s = avctx->priv_data;
427 if (!initialized_tables) {
428 decode_init_static();
429 initialized_tables = 1;
435 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
437 return AVERROR(ENOMEM);
440 ff_mpadsp_init(&s->mpadsp);
442 if (avctx->request_sample_fmt == OUT_FMT &&
443 avctx->codec_id != AV_CODEC_ID_MP3ON4)
444 avctx->sample_fmt = OUT_FMT;
446 avctx->sample_fmt = OUT_FMT_P;
447 s->err_recognition = avctx->err_recognition;
449 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
455 #define C3 FIXHR(0.86602540378443864676/2)
456 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
457 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
458 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
460 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
462 static void imdct12(INTFLOAT *out, SUINTFLOAT *in)
464 SUINTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
467 in1 = in[1*3] + in[0*3];
468 in2 = in[2*3] + in[1*3];
469 in3 = in[3*3] + in[2*3];
470 in4 = in[4*3] + in[3*3];
471 in5 = in[5*3] + in[4*3];
475 in2 = MULH3(in2, C3, 2);
476 in3 = MULH3(in3, C3, 4);
479 t2 = MULH3(in1 - in5, C4, 2);
489 in1 = MULH3(in5 + in3, C5, 1);
496 in5 = MULH3(in5 - in3, C6, 2);
503 /* return the number of decoded frames */
504 static int mp_decode_layer1(MPADecodeContext *s)
506 int bound, i, v, n, ch, j, mant;
507 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
508 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
510 if (s->mode == MPA_JSTEREO)
511 bound = (s->mode_ext + 1) * 4;
515 /* allocation bits */
516 for (i = 0; i < bound; i++) {
517 for (ch = 0; ch < s->nb_channels; ch++) {
518 allocation[ch][i] = get_bits(&s->gb, 4);
521 for (i = bound; i < SBLIMIT; i++)
522 allocation[0][i] = get_bits(&s->gb, 4);
525 for (i = 0; i < bound; i++) {
526 for (ch = 0; ch < s->nb_channels; ch++) {
527 if (allocation[ch][i])
528 scale_factors[ch][i] = get_bits(&s->gb, 6);
531 for (i = bound; i < SBLIMIT; i++) {
532 if (allocation[0][i]) {
533 scale_factors[0][i] = get_bits(&s->gb, 6);
534 scale_factors[1][i] = get_bits(&s->gb, 6);
538 /* compute samples */
539 for (j = 0; j < 12; j++) {
540 for (i = 0; i < bound; i++) {
541 for (ch = 0; ch < s->nb_channels; ch++) {
542 n = allocation[ch][i];
544 mant = get_bits(&s->gb, n + 1);
545 v = l1_unscale(n, mant, scale_factors[ch][i]);
549 s->sb_samples[ch][j][i] = v;
552 for (i = bound; i < SBLIMIT; i++) {
553 n = allocation[0][i];
555 mant = get_bits(&s->gb, n + 1);
556 v = l1_unscale(n, mant, scale_factors[0][i]);
557 s->sb_samples[0][j][i] = v;
558 v = l1_unscale(n, mant, scale_factors[1][i]);
559 s->sb_samples[1][j][i] = v;
561 s->sb_samples[0][j][i] = 0;
562 s->sb_samples[1][j][i] = 0;
569 static int mp_decode_layer2(MPADecodeContext *s)
571 int sblimit; /* number of used subbands */
572 const unsigned char *alloc_table;
573 int table, bit_alloc_bits, i, j, ch, bound, v;
574 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
575 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
576 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
577 int scale, qindex, bits, steps, k, l, m, b;
579 /* select decoding table */
580 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
581 s->sample_rate, s->lsf);
582 sblimit = ff_mpa_sblimit_table[table];
583 alloc_table = ff_mpa_alloc_tables[table];
585 if (s->mode == MPA_JSTEREO)
586 bound = (s->mode_ext + 1) * 4;
590 ff_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
596 /* parse bit allocation */
598 for (i = 0; i < bound; i++) {
599 bit_alloc_bits = alloc_table[j];
600 for (ch = 0; ch < s->nb_channels; ch++)
601 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
602 j += 1 << bit_alloc_bits;
604 for (i = bound; i < sblimit; i++) {
605 bit_alloc_bits = alloc_table[j];
606 v = get_bits(&s->gb, bit_alloc_bits);
609 j += 1 << bit_alloc_bits;
613 for (i = 0; i < sblimit; i++) {
614 for (ch = 0; ch < s->nb_channels; ch++) {
615 if (bit_alloc[ch][i])
616 scale_code[ch][i] = get_bits(&s->gb, 2);
621 for (i = 0; i < sblimit; i++) {
622 for (ch = 0; ch < s->nb_channels; ch++) {
623 if (bit_alloc[ch][i]) {
624 sf = scale_factors[ch][i];
625 switch (scale_code[ch][i]) {
628 sf[0] = get_bits(&s->gb, 6);
629 sf[1] = get_bits(&s->gb, 6);
630 sf[2] = get_bits(&s->gb, 6);
633 sf[0] = get_bits(&s->gb, 6);
638 sf[0] = get_bits(&s->gb, 6);
639 sf[2] = get_bits(&s->gb, 6);
643 sf[0] = get_bits(&s->gb, 6);
644 sf[2] = get_bits(&s->gb, 6);
653 for (k = 0; k < 3; k++) {
654 for (l = 0; l < 12; l += 3) {
656 for (i = 0; i < bound; i++) {
657 bit_alloc_bits = alloc_table[j];
658 for (ch = 0; ch < s->nb_channels; ch++) {
659 b = bit_alloc[ch][i];
661 scale = scale_factors[ch][i][k];
662 qindex = alloc_table[j+b];
663 bits = ff_mpa_quant_bits[qindex];
666 /* 3 values at the same time */
667 v = get_bits(&s->gb, -bits);
668 v2 = division_tabs[qindex][v];
669 steps = ff_mpa_quant_steps[qindex];
671 s->sb_samples[ch][k * 12 + l + 0][i] =
672 l2_unscale_group(steps, v2 & 15, scale);
673 s->sb_samples[ch][k * 12 + l + 1][i] =
674 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
675 s->sb_samples[ch][k * 12 + l + 2][i] =
676 l2_unscale_group(steps, v2 >> 8 , scale);
678 for (m = 0; m < 3; m++) {
679 v = get_bits(&s->gb, bits);
680 v = l1_unscale(bits - 1, v, scale);
681 s->sb_samples[ch][k * 12 + l + m][i] = v;
685 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
686 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
687 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
690 /* next subband in alloc table */
691 j += 1 << bit_alloc_bits;
693 /* XXX: find a way to avoid this duplication of code */
694 for (i = bound; i < sblimit; i++) {
695 bit_alloc_bits = alloc_table[j];
698 int mant, scale0, scale1;
699 scale0 = scale_factors[0][i][k];
700 scale1 = scale_factors[1][i][k];
701 qindex = alloc_table[j+b];
702 bits = ff_mpa_quant_bits[qindex];
704 /* 3 values at the same time */
705 v = get_bits(&s->gb, -bits);
706 steps = ff_mpa_quant_steps[qindex];
709 s->sb_samples[0][k * 12 + l + 0][i] =
710 l2_unscale_group(steps, mant, scale0);
711 s->sb_samples[1][k * 12 + l + 0][i] =
712 l2_unscale_group(steps, mant, scale1);
715 s->sb_samples[0][k * 12 + l + 1][i] =
716 l2_unscale_group(steps, mant, scale0);
717 s->sb_samples[1][k * 12 + l + 1][i] =
718 l2_unscale_group(steps, mant, scale1);
719 s->sb_samples[0][k * 12 + l + 2][i] =
720 l2_unscale_group(steps, v, scale0);
721 s->sb_samples[1][k * 12 + l + 2][i] =
722 l2_unscale_group(steps, v, scale1);
724 for (m = 0; m < 3; m++) {
725 mant = get_bits(&s->gb, bits);
726 s->sb_samples[0][k * 12 + l + m][i] =
727 l1_unscale(bits - 1, mant, scale0);
728 s->sb_samples[1][k * 12 + l + m][i] =
729 l1_unscale(bits - 1, mant, scale1);
733 s->sb_samples[0][k * 12 + l + 0][i] = 0;
734 s->sb_samples[0][k * 12 + l + 1][i] = 0;
735 s->sb_samples[0][k * 12 + l + 2][i] = 0;
736 s->sb_samples[1][k * 12 + l + 0][i] = 0;
737 s->sb_samples[1][k * 12 + l + 1][i] = 0;
738 s->sb_samples[1][k * 12 + l + 2][i] = 0;
740 /* next subband in alloc table */
741 j += 1 << bit_alloc_bits;
743 /* fill remaining samples to zero */
744 for (i = sblimit; i < SBLIMIT; i++) {
745 for (ch = 0; ch < s->nb_channels; ch++) {
746 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
747 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
748 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
756 #define SPLIT(dst,sf,n) \
758 int m = (sf * 171) >> 9; \
761 } else if (n == 4) { \
764 } else if (n == 5) { \
765 int m = (sf * 205) >> 10; \
768 } else if (n == 6) { \
769 int m = (sf * 171) >> 10; \
776 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
779 SPLIT(slen[3], sf, n3)
780 SPLIT(slen[2], sf, n2)
781 SPLIT(slen[1], sf, n1)
785 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
788 const uint8_t *bstab, *pretab;
789 int len, i, j, k, l, v0, shift, gain, gains[3];
793 gain = g->global_gain - 210;
794 shift = g->scalefac_scale + 1;
796 bstab = band_size_long[s->sample_rate_index];
797 pretab = mpa_pretab[g->preflag];
798 for (i = 0; i < g->long_end; i++) {
799 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
801 for (j = len; j > 0; j--)
805 if (g->short_start < 13) {
806 bstab = band_size_short[s->sample_rate_index];
807 gains[0] = gain - (g->subblock_gain[0] << 3);
808 gains[1] = gain - (g->subblock_gain[1] << 3);
809 gains[2] = gain - (g->subblock_gain[2] << 3);
811 for (i = g->short_start; i < 13; i++) {
813 for (l = 0; l < 3; l++) {
814 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
815 for (j = len; j > 0; j--)
822 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
825 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits - s->extrasize * 8) {
827 s->in_gb.buffer = NULL;
829 av_assert2((get_bits_count(&s->gb) & 7) == 0);
830 skip_bits_long(&s->gb, *pos - *end_pos);
832 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
833 *pos = get_bits_count(&s->gb);
837 /* Following is an optimized code for
839 if(get_bits1(&s->gb))
844 #define READ_FLIP_SIGN(dst,src) \
845 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
848 #define READ_FLIP_SIGN(dst,src) \
849 v = -get_bits1(&s->gb); \
850 *(dst) = (*(src) ^ v) - v;
853 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
854 int16_t *exponents, int end_pos2)
858 int last_pos, bits_left;
860 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits - s->extrasize * 8);
862 /* low frequencies (called big values) */
864 for (i = 0; i < 3; i++) {
865 int j, k, l, linbits;
866 j = g->region_size[i];
869 /* select vlc table */
870 k = g->table_select[i];
871 l = mpa_huff_data[k][0];
872 linbits = mpa_huff_data[k][1];
876 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
881 /* read huffcode and compute each couple */
885 int pos = get_bits_count(&s->gb);
888 switch_buffer(s, &pos, &end_pos, &end_pos2);
892 y = get_vlc2(&s->gb, vlc->table, 7, 3);
895 g->sb_hybrid[s_index ] =
896 g->sb_hybrid[s_index+1] = 0;
901 exponent= exponents[s_index];
903 ff_dlog(s->avctx, "region=%d n=%d y=%d exp=%d\n",
904 i, g->region_size[i] - j, y, exponent);
909 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
911 x += get_bitsz(&s->gb, linbits);
912 v = l3_unscale(x, exponent);
913 if (get_bits1(&s->gb))
915 g->sb_hybrid[s_index] = v;
918 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
920 y += get_bitsz(&s->gb, linbits);
921 v = l3_unscale(y, exponent);
922 if (get_bits1(&s->gb))
924 g->sb_hybrid[s_index+1] = v;
931 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
933 x += get_bitsz(&s->gb, linbits);
934 v = l3_unscale(x, exponent);
935 if (get_bits1(&s->gb))
937 g->sb_hybrid[s_index+!!y] = v;
939 g->sb_hybrid[s_index + !y] = 0;
945 /* high frequencies */
946 vlc = &huff_quad_vlc[g->count1table_select];
948 while (s_index <= 572) {
950 pos = get_bits_count(&s->gb);
951 if (pos >= end_pos) {
952 if (pos > end_pos2 && last_pos) {
953 /* some encoders generate an incorrect size for this
954 part. We must go back into the data */
956 skip_bits_long(&s->gb, last_pos - pos);
957 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
958 if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT))
962 switch_buffer(s, &pos, &end_pos, &end_pos2);
968 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
969 ff_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
970 g->sb_hybrid[s_index+0] =
971 g->sb_hybrid[s_index+1] =
972 g->sb_hybrid[s_index+2] =
973 g->sb_hybrid[s_index+3] = 0;
975 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
977 int pos = s_index + idxtab[code];
978 code ^= 8 >> idxtab[code];
979 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
983 /* skip extension bits */
984 bits_left = end_pos2 - get_bits_count(&s->gb);
985 if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) {
986 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
988 } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) {
989 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
992 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
993 skip_bits_long(&s->gb, bits_left);
995 i = get_bits_count(&s->gb);
996 switch_buffer(s, &i, &end_pos, &end_pos2);
1001 /* Reorder short blocks from bitstream order to interleaved order. It
1002 would be faster to do it in parsing, but the code would be far more
1004 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1007 INTFLOAT *ptr, *dst, *ptr1;
1010 if (g->block_type != 2)
1013 if (g->switch_point) {
1014 if (s->sample_rate_index != 8)
1015 ptr = g->sb_hybrid + 36;
1017 ptr = g->sb_hybrid + 72;
1022 for (i = g->short_start; i < 13; i++) {
1023 len = band_size_short[s->sample_rate_index][i];
1026 for (j = len; j > 0; j--) {
1027 *dst++ = ptr[0*len];
1028 *dst++ = ptr[1*len];
1029 *dst++ = ptr[2*len];
1033 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1037 #define ISQRT2 FIXR(0.70710678118654752440)
1039 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1042 int sf_max, sf, len, non_zero_found;
1043 INTFLOAT (*is_tab)[16], *tab0, *tab1, v1, v2;
1044 SUINTFLOAT tmp0, tmp1;
1045 int non_zero_found_short[3];
1047 /* intensity stereo */
1048 if (s->mode_ext & MODE_EXT_I_STEREO) {
1053 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1057 tab0 = g0->sb_hybrid + 576;
1058 tab1 = g1->sb_hybrid + 576;
1060 non_zero_found_short[0] = 0;
1061 non_zero_found_short[1] = 0;
1062 non_zero_found_short[2] = 0;
1063 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1064 for (i = 12; i >= g1->short_start; i--) {
1065 /* for last band, use previous scale factor */
1068 len = band_size_short[s->sample_rate_index][i];
1069 for (l = 2; l >= 0; l--) {
1072 if (!non_zero_found_short[l]) {
1073 /* test if non zero band. if so, stop doing i-stereo */
1074 for (j = 0; j < len; j++) {
1076 non_zero_found_short[l] = 1;
1080 sf = g1->scale_factors[k + l];
1086 for (j = 0; j < len; j++) {
1088 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1089 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1093 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1094 /* lower part of the spectrum : do ms stereo
1096 for (j = 0; j < len; j++) {
1099 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1100 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1107 non_zero_found = non_zero_found_short[0] |
1108 non_zero_found_short[1] |
1109 non_zero_found_short[2];
1111 for (i = g1->long_end - 1;i >= 0;i--) {
1112 len = band_size_long[s->sample_rate_index][i];
1115 /* test if non zero band. if so, stop doing i-stereo */
1116 if (!non_zero_found) {
1117 for (j = 0; j < len; j++) {
1123 /* for last band, use previous scale factor */
1124 k = (i == 21) ? 20 : i;
1125 sf = g1->scale_factors[k];
1130 for (j = 0; j < len; j++) {
1132 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1133 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1137 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1138 /* lower part of the spectrum : do ms stereo
1140 for (j = 0; j < len; j++) {
1143 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1144 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1149 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1150 /* ms stereo ONLY */
1151 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1154 s->fdsp->butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1156 tab0 = g0->sb_hybrid;
1157 tab1 = g1->sb_hybrid;
1158 for (i = 0; i < 576; i++) {
1161 tab0[i] = tmp0 + tmp1;
1162 tab1[i] = tmp0 - tmp1;
1170 # include "mips/compute_antialias_float.h"
1171 #endif /* HAVE_MIPSFPU */
1174 # include "mips/compute_antialias_fixed.h"
1175 #endif /* HAVE_MIPSDSP */
1176 #endif /* USE_FLOATS */
1178 #ifndef compute_antialias
1180 #define AA(j) do { \
1181 float tmp0 = ptr[-1-j]; \
1182 float tmp1 = ptr[ j]; \
1183 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1184 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1187 #define AA(j) do { \
1188 SUINT tmp0 = ptr[-1-j]; \
1189 SUINT tmp1 = ptr[ j]; \
1190 SUINT tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1191 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1192 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1196 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1201 /* we antialias only "long" bands */
1202 if (g->block_type == 2) {
1203 if (!g->switch_point)
1205 /* XXX: check this for 8000Hz case */
1211 ptr = g->sb_hybrid + 18;
1212 for (i = n; i > 0; i--) {
1225 #endif /* compute_antialias */
1227 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1228 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1230 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1232 int i, j, mdct_long_end, sblimit;
1234 /* find last non zero block */
1235 ptr = g->sb_hybrid + 576;
1236 ptr1 = g->sb_hybrid + 2 * 18;
1237 while (ptr >= ptr1) {
1241 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1244 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1246 if (g->block_type == 2) {
1247 /* XXX: check for 8000 Hz */
1248 if (g->switch_point)
1253 mdct_long_end = sblimit;
1256 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1257 mdct_long_end, g->switch_point,
1260 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1261 ptr = g->sb_hybrid + 18 * mdct_long_end;
1263 for (j = mdct_long_end; j < sblimit; j++) {
1264 /* select frequency inversion */
1265 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1266 out_ptr = sb_samples + j;
1268 for (i = 0; i < 6; i++) {
1269 *out_ptr = buf[4*i];
1272 imdct12(out2, ptr + 0);
1273 for (i = 0; i < 6; i++) {
1274 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1275 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1278 imdct12(out2, ptr + 1);
1279 for (i = 0; i < 6; i++) {
1280 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1281 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1284 imdct12(out2, ptr + 2);
1285 for (i = 0; i < 6; i++) {
1286 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1287 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1288 buf[4*(i + 6*2)] = 0;
1291 buf += (j&3) != 3 ? 1 : (4*18-3);
1294 for (j = sblimit; j < SBLIMIT; j++) {
1296 out_ptr = sb_samples + j;
1297 for (i = 0; i < 18; i++) {
1298 *out_ptr = buf[4*i];
1302 buf += (j&3) != 3 ? 1 : (4*18-3);
1306 /* main layer3 decoding function */
1307 static int mp_decode_layer3(MPADecodeContext *s)
1309 int nb_granules, main_data_begin;
1310 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1312 int16_t exponents[576]; //FIXME try INTFLOAT
1314 /* read side info */
1316 main_data_begin = get_bits(&s->gb, 8);
1317 skip_bits(&s->gb, s->nb_channels);
1320 main_data_begin = get_bits(&s->gb, 9);
1321 if (s->nb_channels == 2)
1322 skip_bits(&s->gb, 3);
1324 skip_bits(&s->gb, 5);
1326 for (ch = 0; ch < s->nb_channels; ch++) {
1327 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1328 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1332 for (gr = 0; gr < nb_granules; gr++) {
1333 for (ch = 0; ch < s->nb_channels; ch++) {
1334 ff_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1335 g = &s->granules[ch][gr];
1336 g->part2_3_length = get_bits(&s->gb, 12);
1337 g->big_values = get_bits(&s->gb, 9);
1338 if (g->big_values > 288) {
1339 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1340 return AVERROR_INVALIDDATA;
1343 g->global_gain = get_bits(&s->gb, 8);
1344 /* if MS stereo only is selected, we precompute the
1345 1/sqrt(2) renormalization factor */
1346 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1348 g->global_gain -= 2;
1350 g->scalefac_compress = get_bits(&s->gb, 9);
1352 g->scalefac_compress = get_bits(&s->gb, 4);
1353 blocksplit_flag = get_bits1(&s->gb);
1354 if (blocksplit_flag) {
1355 g->block_type = get_bits(&s->gb, 2);
1356 if (g->block_type == 0) {
1357 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1358 return AVERROR_INVALIDDATA;
1360 g->switch_point = get_bits1(&s->gb);
1361 for (i = 0; i < 2; i++)
1362 g->table_select[i] = get_bits(&s->gb, 5);
1363 for (i = 0; i < 3; i++)
1364 g->subblock_gain[i] = get_bits(&s->gb, 3);
1365 init_short_region(s, g);
1367 int region_address1, region_address2;
1369 g->switch_point = 0;
1370 for (i = 0; i < 3; i++)
1371 g->table_select[i] = get_bits(&s->gb, 5);
1372 /* compute huffman coded region sizes */
1373 region_address1 = get_bits(&s->gb, 4);
1374 region_address2 = get_bits(&s->gb, 3);
1375 ff_dlog(s->avctx, "region1=%d region2=%d\n",
1376 region_address1, region_address2);
1377 init_long_region(s, g, region_address1, region_address2);
1379 region_offset2size(g);
1380 compute_band_indexes(s, g);
1384 g->preflag = get_bits1(&s->gb);
1385 g->scalefac_scale = get_bits1(&s->gb);
1386 g->count1table_select = get_bits1(&s->gb);
1387 ff_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1388 g->block_type, g->switch_point);
1394 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1395 s->extrasize = av_clip((get_bits_left(&s->gb) >> 3) - s->extrasize, 0,
1396 FFMAX(0, LAST_BUF_SIZE - s->last_buf_size));
1397 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1398 /* now we get bits from the main_data_begin offset */
1399 ff_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1400 main_data_begin, s->last_buf_size);
1402 memcpy(s->last_buf + s->last_buf_size, ptr, s->extrasize);
1404 init_get_bits(&s->gb, s->last_buf, (s->last_buf_size + s->extrasize) * 8);
1405 s->last_buf_size <<= 3;
1406 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1407 for (ch = 0; ch < s->nb_channels; ch++) {
1408 g = &s->granules[ch][gr];
1409 s->last_buf_size += g->part2_3_length;
1410 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1411 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1414 skip = s->last_buf_size - 8 * main_data_begin;
1415 if (skip >= s->gb.size_in_bits - s->extrasize * 8 && s->in_gb.buffer) {
1416 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits + s->extrasize * 8);
1418 s->in_gb.buffer = NULL;
1421 skip_bits_long(&s->gb, skip);
1428 for (; gr < nb_granules; gr++) {
1429 for (ch = 0; ch < s->nb_channels; ch++) {
1430 g = &s->granules[ch][gr];
1431 bits_pos = get_bits_count(&s->gb);
1435 int slen, slen1, slen2;
1437 /* MPEG-1 scale factors */
1438 slen1 = slen_table[0][g->scalefac_compress];
1439 slen2 = slen_table[1][g->scalefac_compress];
1440 ff_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1441 if (g->block_type == 2) {
1442 n = g->switch_point ? 17 : 18;
1445 for (i = 0; i < n; i++)
1446 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1448 for (i = 0; i < n; i++)
1449 g->scale_factors[j++] = 0;
1452 for (i = 0; i < 18; i++)
1453 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1454 for (i = 0; i < 3; i++)
1455 g->scale_factors[j++] = 0;
1457 for (i = 0; i < 21; i++)
1458 g->scale_factors[j++] = 0;
1461 sc = s->granules[ch][0].scale_factors;
1463 for (k = 0; k < 4; k++) {
1465 if ((g->scfsi & (0x8 >> k)) == 0) {
1466 slen = (k < 2) ? slen1 : slen2;
1468 for (i = 0; i < n; i++)
1469 g->scale_factors[j++] = get_bits(&s->gb, slen);
1471 for (i = 0; i < n; i++)
1472 g->scale_factors[j++] = 0;
1475 /* simply copy from last granule */
1476 for (i = 0; i < n; i++) {
1477 g->scale_factors[j] = sc[j];
1482 g->scale_factors[j++] = 0;
1485 int tindex, tindex2, slen[4], sl, sf;
1487 /* LSF scale factors */
1488 if (g->block_type == 2)
1489 tindex = g->switch_point ? 2 : 1;
1493 sf = g->scalefac_compress;
1494 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1495 /* intensity stereo case */
1498 lsf_sf_expand(slen, sf, 6, 6, 0);
1500 } else if (sf < 244) {
1501 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1504 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1510 lsf_sf_expand(slen, sf, 5, 4, 4);
1512 } else if (sf < 500) {
1513 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1516 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1523 for (k = 0; k < 4; k++) {
1524 n = lsf_nsf_table[tindex2][tindex][k];
1527 for (i = 0; i < n; i++)
1528 g->scale_factors[j++] = get_bits(&s->gb, sl);
1530 for (i = 0; i < n; i++)
1531 g->scale_factors[j++] = 0;
1534 /* XXX: should compute exact size */
1536 g->scale_factors[j] = 0;
1539 exponents_from_scale_factors(s, g, exponents);
1541 /* read Huffman coded residue */
1542 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1545 if (s->mode == MPA_JSTEREO)
1546 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1548 for (ch = 0; ch < s->nb_channels; ch++) {
1549 g = &s->granules[ch][gr];
1551 reorder_block(s, g);
1552 compute_antialias(s, g);
1553 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1556 if (get_bits_count(&s->gb) < 0)
1557 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1558 return nb_granules * 18;
1561 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1562 const uint8_t *buf, int buf_size)
1564 int i, nb_frames, ch, ret;
1565 OUT_INT *samples_ptr;
1567 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1569 if (s->error_protection) {
1570 uint16_t crc = get_bits(&s->gb, 16);
1571 if (s->err_recognition & AV_EF_CRCCHECK) {
1572 const int sec_len = s->lsf ? ((s->nb_channels == 1) ? 9 : 17) :
1573 ((s->nb_channels == 1) ? 17 : 32);
1574 const AVCRC *crc_tab = av_crc_get_table(AV_CRC_16_ANSI);
1575 uint32_t crc_cal = av_crc(crc_tab, UINT16_MAX, &buf[2], 2);
1576 crc_cal = av_crc(crc_tab, crc_cal, &buf[6], sec_len);
1578 if (av_bswap16(crc) ^ crc_cal) {
1579 av_log(s->avctx, AV_LOG_ERROR, "CRC mismatch!\n");
1580 if (s->err_recognition & AV_EF_EXPLODE)
1581 return AVERROR_INVALIDDATA;
1588 s->avctx->frame_size = 384;
1589 nb_frames = mp_decode_layer1(s);
1592 s->avctx->frame_size = 1152;
1593 nb_frames = mp_decode_layer2(s);
1596 s->avctx->frame_size = s->lsf ? 576 : 1152;
1598 nb_frames = mp_decode_layer3(s);
1601 if (s->in_gb.buffer) {
1602 align_get_bits(&s->gb);
1603 i = (get_bits_left(&s->gb) >> 3) - s->extrasize;
1604 if (i >= 0 && i <= BACKSTEP_SIZE) {
1605 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1608 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1610 s->in_gb.buffer = NULL;
1614 align_get_bits(&s->gb);
1615 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1616 i = (get_bits_left(&s->gb) >> 3) - s->extrasize;
1617 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1619 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1620 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1622 av_assert1(i <= buf_size - HEADER_SIZE && i >= 0);
1623 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1624 s->last_buf_size += i;
1630 /* get output buffer */
1632 av_assert0(s->frame);
1633 s->frame->nb_samples = s->avctx->frame_size;
1634 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0)
1636 samples = (OUT_INT **)s->frame->extended_data;
1639 /* apply the synthesis filter */
1640 for (ch = 0; ch < s->nb_channels; ch++) {
1642 if (s->avctx->sample_fmt == OUT_FMT_P) {
1643 samples_ptr = samples[ch];
1646 samples_ptr = samples[0] + ch;
1647 sample_stride = s->nb_channels;
1649 for (i = 0; i < nb_frames; i++) {
1650 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1651 &(s->synth_buf_offset[ch]),
1652 RENAME(ff_mpa_synth_window),
1653 &s->dither_state, samples_ptr,
1654 sample_stride, s->sb_samples[ch][i]);
1655 samples_ptr += 32 * sample_stride;
1659 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1662 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1665 const uint8_t *buf = avpkt->data;
1666 int buf_size = avpkt->size;
1667 MPADecodeContext *s = avctx->priv_data;
1672 while(buf_size && !*buf){
1678 if (buf_size < HEADER_SIZE)
1679 return AVERROR_INVALIDDATA;
1681 header = AV_RB32(buf);
1682 if (header>>8 == AV_RB32("TAG")>>8) {
1683 av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n");
1684 return buf_size + skipped;
1686 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1688 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1689 return AVERROR_INVALIDDATA;
1690 } else if (ret == 1) {
1691 /* free format: prepare to compute frame size */
1693 return AVERROR_INVALIDDATA;
1695 /* update codec info */
1696 avctx->channels = s->nb_channels;
1697 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1698 if (!avctx->bit_rate)
1699 avctx->bit_rate = s->bit_rate;
1701 if (s->frame_size <= 0) {
1702 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1703 return AVERROR_INVALIDDATA;
1704 } else if (s->frame_size < buf_size) {
1705 av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");
1706 buf_size= s->frame_size;
1711 ret = mp_decode_frame(s, NULL, buf, buf_size);
1713 s->frame->nb_samples = avctx->frame_size;
1715 avctx->sample_rate = s->sample_rate;
1716 //FIXME maybe move the other codec info stuff from above here too
1718 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1719 /* Only return an error if the bad frame makes up the whole packet or
1720 * the error is related to buffer management.
1721 * If there is more data in the packet, just consume the bad frame
1722 * instead of returning an error, which would discard the whole
1725 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1729 return buf_size + skipped;
1732 static void mp_flush(MPADecodeContext *ctx)
1734 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1735 memset(ctx->mdct_buf, 0, sizeof(ctx->mdct_buf));
1736 ctx->last_buf_size = 0;
1737 ctx->dither_state = 0;
1740 static void flush(AVCodecContext *avctx)
1742 mp_flush(avctx->priv_data);
1745 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1746 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1747 int *got_frame_ptr, AVPacket *avpkt)
1749 const uint8_t *buf = avpkt->data;
1750 int buf_size = avpkt->size;
1751 MPADecodeContext *s = avctx->priv_data;
1754 int av_unused out_size;
1758 // Discard too short frames
1759 if (buf_size < HEADER_SIZE) {
1760 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1761 return AVERROR_INVALIDDATA;
1765 if (len > MPA_MAX_CODED_FRAME_SIZE)
1766 len = MPA_MAX_CODED_FRAME_SIZE;
1768 // Get header and restore sync word
1769 header = AV_RB32(buf) | 0xffe00000;
1771 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1773 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1776 /* update codec info */
1777 avctx->sample_rate = s->sample_rate;
1778 avctx->channels = s->nb_channels;
1779 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1780 if (!avctx->bit_rate)
1781 avctx->bit_rate = s->bit_rate;
1783 s->frame_size = len;
1787 ret = mp_decode_frame(s, NULL, buf, buf_size);
1789 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1797 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1799 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1802 * Context for MP3On4 decoder
1804 typedef struct MP3On4DecodeContext {
1805 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1806 int syncword; ///< syncword patch
1807 const uint8_t *coff; ///< channel offsets in output buffer
1808 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1809 } MP3On4DecodeContext;
1811 #include "mpeg4audio.h"
1813 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1815 /* number of mp3 decoder instances */
1816 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1818 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1819 static const uint8_t chan_offset[8][5] = {
1824 { 2, 0, 3 }, // C FLR BS
1825 { 2, 0, 3 }, // C FLR BLRS
1826 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1827 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1830 /* mp3on4 channel layouts */
1831 static const int16_t chan_layout[8] = {
1834 AV_CH_LAYOUT_STEREO,
1835 AV_CH_LAYOUT_SURROUND,
1836 AV_CH_LAYOUT_4POINT0,
1837 AV_CH_LAYOUT_5POINT0,
1838 AV_CH_LAYOUT_5POINT1,
1839 AV_CH_LAYOUT_7POINT1
1842 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1844 MP3On4DecodeContext *s = avctx->priv_data;
1847 if (s->mp3decctx[0])
1848 av_freep(&s->mp3decctx[0]->fdsp);
1850 for (i = 0; i < s->frames; i++)
1851 av_freep(&s->mp3decctx[i]);
1857 static av_cold int decode_init_mp3on4(AVCodecContext * avctx)
1859 MP3On4DecodeContext *s = avctx->priv_data;
1860 MPEG4AudioConfig cfg;
1863 if ((avctx->extradata_size < 2) || !avctx->extradata) {
1864 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1865 return AVERROR_INVALIDDATA;
1868 avpriv_mpeg4audio_get_config2(&cfg, avctx->extradata,
1869 avctx->extradata_size, 1, avctx);
1870 if (!cfg.chan_config || cfg.chan_config > 7) {
1871 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1872 return AVERROR_INVALIDDATA;
1874 s->frames = mp3Frames[cfg.chan_config];
1875 s->coff = chan_offset[cfg.chan_config];
1876 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1877 avctx->channel_layout = chan_layout[cfg.chan_config];
1879 if (cfg.sample_rate < 16000)
1880 s->syncword = 0xffe00000;
1882 s->syncword = 0xfff00000;
1884 /* Init the first mp3 decoder in standard way, so that all tables get builded
1885 * We replace avctx->priv_data with the context of the first decoder so that
1886 * decode_init() does not have to be changed.
1887 * Other decoders will be initialized here copying data from the first context
1889 // Allocate zeroed memory for the first decoder context
1890 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1891 if (!s->mp3decctx[0])
1893 // Put decoder context in place to make init_decode() happy
1894 avctx->priv_data = s->mp3decctx[0];
1896 // Restore mp3on4 context pointer
1897 avctx->priv_data = s;
1898 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1900 /* Create a separate codec/context for each frame (first is already ok).
1901 * Each frame is 1 or 2 channels - up to 5 frames allowed
1903 for (i = 1; i < s->frames; i++) {
1904 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1905 if (!s->mp3decctx[i])
1907 s->mp3decctx[i]->adu_mode = 1;
1908 s->mp3decctx[i]->avctx = avctx;
1909 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1910 s->mp3decctx[i]->fdsp = s->mp3decctx[0]->fdsp;
1915 decode_close_mp3on4(avctx);
1916 return AVERROR(ENOMEM);
1920 static void flush_mp3on4(AVCodecContext *avctx)
1923 MP3On4DecodeContext *s = avctx->priv_data;
1925 for (i = 0; i < s->frames; i++)
1926 mp_flush(s->mp3decctx[i]);
1930 static int decode_frame_mp3on4(AVCodecContext *avctx, void *data,
1931 int *got_frame_ptr, AVPacket *avpkt)
1933 AVFrame *frame = data;
1934 const uint8_t *buf = avpkt->data;
1935 int buf_size = avpkt->size;
1936 MP3On4DecodeContext *s = avctx->priv_data;
1937 MPADecodeContext *m;
1938 int fsize, len = buf_size, out_size = 0;
1940 OUT_INT **out_samples;
1944 /* get output buffer */
1945 frame->nb_samples = MPA_FRAME_SIZE;
1946 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1948 out_samples = (OUT_INT **)frame->extended_data;
1950 // Discard too short frames
1951 if (buf_size < HEADER_SIZE)
1952 return AVERROR_INVALIDDATA;
1954 avctx->bit_rate = 0;
1957 for (fr = 0; fr < s->frames; fr++) {
1958 fsize = AV_RB16(buf) >> 4;
1959 fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE);
1960 m = s->mp3decctx[fr];
1963 if (fsize < HEADER_SIZE) {
1964 av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n");
1965 return AVERROR_INVALIDDATA;
1967 header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header
1969 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1971 av_log(avctx, AV_LOG_ERROR, "Bad header, discard block\n");
1972 return AVERROR_INVALIDDATA;
1975 if (ch + m->nb_channels > avctx->channels ||
1976 s->coff[fr] + m->nb_channels > avctx->channels) {
1977 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1979 return AVERROR_INVALIDDATA;
1981 ch += m->nb_channels;
1983 outptr[0] = out_samples[s->coff[fr]];
1984 if (m->nb_channels > 1)
1985 outptr[1] = out_samples[s->coff[fr] + 1];
1987 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0) {
1988 av_log(avctx, AV_LOG_ERROR, "failed to decode channel %d\n", ch);
1989 memset(outptr[0], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1990 if (m->nb_channels > 1)
1991 memset(outptr[1], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1992 ret = m->nb_channels * MPA_FRAME_SIZE*sizeof(OUT_INT);
1999 avctx->bit_rate += m->bit_rate;
2001 if (ch != avctx->channels) {
2002 av_log(avctx, AV_LOG_ERROR, "failed to decode all channels\n");
2003 return AVERROR_INVALIDDATA;
2006 /* update codec info */
2007 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
2009 frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
2014 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */