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];
75 /* next header (used in free format parsing) */
76 uint32_t free_format_next_header;
79 DECLARE_ALIGNED(32, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
80 int synth_buf_offset[MPA_MAX_CHANNELS];
81 DECLARE_ALIGNED(32, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
82 INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
83 GranuleDef granules[2][2]; /* Used in Layer 3 */
84 int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
87 AVCodecContext* avctx;
89 AVFloatDSPContext *fdsp;
95 #include "mpegaudiodata.h"
96 #include "mpegaudiodectab.h"
98 /* vlc structure for decoding layer 3 huffman tables */
99 static VLC huff_vlc[16];
100 static VLC_TYPE huff_vlc_tables[
101 0 + 128 + 128 + 128 + 130 + 128 + 154 + 166 +
102 142 + 204 + 190 + 170 + 542 + 460 + 662 + 414
104 static const int huff_vlc_tables_sizes[16] = {
105 0, 128, 128, 128, 130, 128, 154, 166,
106 142, 204, 190, 170, 542, 460, 662, 414
108 static VLC huff_quad_vlc[2];
109 static VLC_TYPE huff_quad_vlc_tables[128+16][2];
110 static const int huff_quad_vlc_tables_sizes[2] = { 128, 16 };
111 /* computed from band_size_long */
112 static uint16_t band_index_long[9][23];
113 #include "mpegaudio_tablegen.h"
114 /* intensity stereo coef table */
115 static INTFLOAT is_table[2][16];
116 static INTFLOAT is_table_lsf[2][2][16];
117 static INTFLOAT csa_table[8][4];
119 static int16_t division_tab3[1<<6 ];
120 static int16_t division_tab5[1<<8 ];
121 static int16_t division_tab9[1<<11];
123 static int16_t * const division_tabs[4] = {
124 division_tab3, division_tab5, NULL, division_tab9
127 /* lower 2 bits: modulo 3, higher bits: shift */
128 static uint16_t scale_factor_modshift[64];
129 /* [i][j]: 2^(-j/3) * FRAC_ONE * 2^(i+2) / (2^(i+2) - 1) */
130 static int32_t scale_factor_mult[15][3];
131 /* mult table for layer 2 group quantization */
133 #define SCALE_GEN(v) \
134 { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) }
136 static const int32_t scale_factor_mult2[3][3] = {
137 SCALE_GEN(4.0 / 3.0), /* 3 steps */
138 SCALE_GEN(4.0 / 5.0), /* 5 steps */
139 SCALE_GEN(4.0 / 9.0), /* 9 steps */
143 * Convert region offsets to region sizes and truncate
144 * size to big_values.
146 static void region_offset2size(GranuleDef *g)
149 g->region_size[2] = 576 / 2;
150 for (i = 0; i < 3; i++) {
151 k = FFMIN(g->region_size[i], g->big_values);
152 g->region_size[i] = k - j;
157 static void init_short_region(MPADecodeContext *s, GranuleDef *g)
159 if (g->block_type == 2) {
160 if (s->sample_rate_index != 8)
161 g->region_size[0] = (36 / 2);
163 g->region_size[0] = (72 / 2);
165 if (s->sample_rate_index <= 2)
166 g->region_size[0] = (36 / 2);
167 else if (s->sample_rate_index != 8)
168 g->region_size[0] = (54 / 2);
170 g->region_size[0] = (108 / 2);
172 g->region_size[1] = (576 / 2);
175 static void init_long_region(MPADecodeContext *s, GranuleDef *g,
179 g->region_size[0] = band_index_long[s->sample_rate_index][ra1 + 1] >> 1;
180 /* should not overflow */
181 l = FFMIN(ra1 + ra2 + 2, 22);
182 g->region_size[1] = band_index_long[s->sample_rate_index][ l] >> 1;
185 static void compute_band_indexes(MPADecodeContext *s, GranuleDef *g)
187 if (g->block_type == 2) {
188 if (g->switch_point) {
189 if(s->sample_rate_index == 8)
190 avpriv_request_sample(s->avctx, "switch point in 8khz");
191 /* if switched mode, we handle the 36 first samples as
192 long blocks. For 8000Hz, we handle the 72 first
193 exponents as long blocks */
194 if (s->sample_rate_index <= 2)
210 /* layer 1 unscaling */
211 /* n = number of bits of the mantissa minus 1 */
212 static inline int l1_unscale(int n, int mant, int scale_factor)
217 shift = scale_factor_modshift[scale_factor];
220 val = MUL64((int)(mant + (-1U << n) + 1), scale_factor_mult[n-1][mod]);
222 /* NOTE: at this point, 1 <= shift >= 21 + 15 */
223 return (int)((val + (1LL << (shift - 1))) >> shift);
226 static inline int l2_unscale_group(int steps, int mant, int scale_factor)
230 shift = scale_factor_modshift[scale_factor];
234 val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod];
235 /* NOTE: at this point, 0 <= shift <= 21 */
237 val = (val + (1 << (shift - 1))) >> shift;
241 /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */
242 static inline int l3_unscale(int value, int exponent)
247 e = table_4_3_exp [4 * value + (exponent & 3)];
248 m = table_4_3_value[4 * value + (exponent & 3)];
252 av_log(NULL, AV_LOG_WARNING, "l3_unscale: e is %d\n", e);
256 m = (m + ((1U << e)>>1)) >> e;
261 static av_cold void decode_init_static(void)
266 /* scale factors table for layer 1/2 */
267 for (i = 0; i < 64; i++) {
269 /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */
272 scale_factor_modshift[i] = mod | (shift << 2);
275 /* scale factor multiply for layer 1 */
276 for (i = 0; i < 15; i++) {
279 norm = ((INT64_C(1) << n) * FRAC_ONE) / ((1 << n) - 1);
280 scale_factor_mult[i][0] = MULLx(norm, FIXR(1.0 * 2.0), FRAC_BITS);
281 scale_factor_mult[i][1] = MULLx(norm, FIXR(0.7937005259 * 2.0), FRAC_BITS);
282 scale_factor_mult[i][2] = MULLx(norm, FIXR(0.6299605249 * 2.0), FRAC_BITS);
283 ff_dlog(NULL, "%d: norm=%x s=%x %x %x\n", i, norm,
284 scale_factor_mult[i][0],
285 scale_factor_mult[i][1],
286 scale_factor_mult[i][2]);
289 RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window));
291 /* huffman decode tables */
293 for (i = 1; i < 16; i++) {
294 const HuffTable *h = &mpa_huff_tables[i];
296 uint8_t tmp_bits [512] = { 0 };
297 uint16_t tmp_codes[512] = { 0 };
302 for (x = 0; x < xsize; x++) {
303 for (y = 0; y < xsize; y++) {
304 tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ];
305 tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++];
310 huff_vlc[i].table = huff_vlc_tables+offset;
311 huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i];
312 init_vlc(&huff_vlc[i], 7, 512,
313 tmp_bits, 1, 1, tmp_codes, 2, 2,
314 INIT_VLC_USE_NEW_STATIC);
315 offset += huff_vlc_tables_sizes[i];
317 av_assert0(offset == FF_ARRAY_ELEMS(huff_vlc_tables));
320 for (i = 0; i < 2; i++) {
321 huff_quad_vlc[i].table = huff_quad_vlc_tables+offset;
322 huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i];
323 init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16,
324 mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1,
325 INIT_VLC_USE_NEW_STATIC);
326 offset += huff_quad_vlc_tables_sizes[i];
328 av_assert0(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables));
330 for (i = 0; i < 9; i++) {
332 for (j = 0; j < 22; j++) {
333 band_index_long[i][j] = k;
334 k += band_size_long[i][j];
336 band_index_long[i][22] = k;
339 /* compute n ^ (4/3) and store it in mantissa/exp format */
341 mpegaudio_tableinit();
343 for (i = 0; i < 4; i++) {
344 if (ff_mpa_quant_bits[i] < 0) {
345 for (j = 0; j < (1 << (-ff_mpa_quant_bits[i]+1)); j++) {
346 int val1, val2, val3, steps;
348 steps = ff_mpa_quant_steps[i];
353 division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8);
359 for (i = 0; i < 7; i++) {
363 f = tan((double)i * M_PI / 12.0);
364 v = FIXR(f / (1.0 + f));
369 is_table[1][6 - i] = v;
372 for (i = 7; i < 16; i++)
373 is_table[0][i] = is_table[1][i] = 0.0;
375 for (i = 0; i < 16; i++) {
379 for (j = 0; j < 2; j++) {
380 e = -(j + 1) * ((i + 1) >> 1);
383 is_table_lsf[j][k ^ 1][i] = FIXR(f);
384 is_table_lsf[j][k ][i] = FIXR(1.0);
385 ff_dlog(NULL, "is_table_lsf %d %d: %f %f\n",
386 i, j, (float) is_table_lsf[j][0][i],
387 (float) is_table_lsf[j][1][i]);
391 for (i = 0; i < 8; i++) {
394 cs = 1.0 / sqrt(1.0 + ci * ci);
397 csa_table[i][0] = FIXHR(cs/4);
398 csa_table[i][1] = FIXHR(ca/4);
399 csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4);
400 csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4);
402 csa_table[i][0] = cs;
403 csa_table[i][1] = ca;
404 csa_table[i][2] = ca + cs;
405 csa_table[i][3] = ca - cs;
411 static av_cold int decode_close(AVCodecContext * avctx)
413 MPADecodeContext *s = avctx->priv_data;
420 static av_cold int decode_init(AVCodecContext * avctx)
422 static int initialized_tables = 0;
423 MPADecodeContext *s = avctx->priv_data;
425 if (!initialized_tables) {
426 decode_init_static();
427 initialized_tables = 1;
433 s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
435 return AVERROR(ENOMEM);
438 ff_mpadsp_init(&s->mpadsp);
440 if (avctx->request_sample_fmt == OUT_FMT &&
441 avctx->codec_id != AV_CODEC_ID_MP3ON4)
442 avctx->sample_fmt = OUT_FMT;
444 avctx->sample_fmt = OUT_FMT_P;
445 s->err_recognition = avctx->err_recognition;
447 if (avctx->codec_id == AV_CODEC_ID_MP3ADU)
453 #define C3 FIXHR(0.86602540378443864676/2)
454 #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36)
455 #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36)
456 #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36)
458 /* 12 points IMDCT. We compute it "by hand" by factorizing obvious
460 static void imdct12(INTFLOAT *out, SUINTFLOAT *in)
462 SUINTFLOAT in0, in1, in2, in3, in4, in5, t1, t2;
465 in1 = in[1*3] + in[0*3];
466 in2 = in[2*3] + in[1*3];
467 in3 = in[3*3] + in[2*3];
468 in4 = in[4*3] + in[3*3];
469 in5 = in[5*3] + in[4*3];
473 in2 = MULH3(in2, C3, 2);
474 in3 = MULH3(in3, C3, 4);
477 t2 = MULH3(in1 - in5, C4, 2);
487 in1 = MULH3(in5 + in3, C5, 1);
494 in5 = MULH3(in5 - in3, C6, 2);
501 /* return the number of decoded frames */
502 static int mp_decode_layer1(MPADecodeContext *s)
504 int bound, i, v, n, ch, j, mant;
505 uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
506 uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];
508 if (s->mode == MPA_JSTEREO)
509 bound = (s->mode_ext + 1) * 4;
513 /* allocation bits */
514 for (i = 0; i < bound; i++) {
515 for (ch = 0; ch < s->nb_channels; ch++) {
516 allocation[ch][i] = get_bits(&s->gb, 4);
519 for (i = bound; i < SBLIMIT; i++)
520 allocation[0][i] = get_bits(&s->gb, 4);
523 for (i = 0; i < bound; i++) {
524 for (ch = 0; ch < s->nb_channels; ch++) {
525 if (allocation[ch][i])
526 scale_factors[ch][i] = get_bits(&s->gb, 6);
529 for (i = bound; i < SBLIMIT; i++) {
530 if (allocation[0][i]) {
531 scale_factors[0][i] = get_bits(&s->gb, 6);
532 scale_factors[1][i] = get_bits(&s->gb, 6);
536 /* compute samples */
537 for (j = 0; j < 12; j++) {
538 for (i = 0; i < bound; i++) {
539 for (ch = 0; ch < s->nb_channels; ch++) {
540 n = allocation[ch][i];
542 mant = get_bits(&s->gb, n + 1);
543 v = l1_unscale(n, mant, scale_factors[ch][i]);
547 s->sb_samples[ch][j][i] = v;
550 for (i = bound; i < SBLIMIT; i++) {
551 n = allocation[0][i];
553 mant = get_bits(&s->gb, n + 1);
554 v = l1_unscale(n, mant, scale_factors[0][i]);
555 s->sb_samples[0][j][i] = v;
556 v = l1_unscale(n, mant, scale_factors[1][i]);
557 s->sb_samples[1][j][i] = v;
559 s->sb_samples[0][j][i] = 0;
560 s->sb_samples[1][j][i] = 0;
567 static int mp_decode_layer2(MPADecodeContext *s)
569 int sblimit; /* number of used subbands */
570 const unsigned char *alloc_table;
571 int table, bit_alloc_bits, i, j, ch, bound, v;
572 unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
573 unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
574 unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
575 int scale, qindex, bits, steps, k, l, m, b;
577 /* select decoding table */
578 table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
579 s->sample_rate, s->lsf);
580 sblimit = ff_mpa_sblimit_table[table];
581 alloc_table = ff_mpa_alloc_tables[table];
583 if (s->mode == MPA_JSTEREO)
584 bound = (s->mode_ext + 1) * 4;
588 ff_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);
594 /* parse bit allocation */
596 for (i = 0; i < bound; i++) {
597 bit_alloc_bits = alloc_table[j];
598 for (ch = 0; ch < s->nb_channels; ch++)
599 bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
600 j += 1 << bit_alloc_bits;
602 for (i = bound; i < sblimit; i++) {
603 bit_alloc_bits = alloc_table[j];
604 v = get_bits(&s->gb, bit_alloc_bits);
607 j += 1 << bit_alloc_bits;
611 for (i = 0; i < sblimit; i++) {
612 for (ch = 0; ch < s->nb_channels; ch++) {
613 if (bit_alloc[ch][i])
614 scale_code[ch][i] = get_bits(&s->gb, 2);
619 for (i = 0; i < sblimit; i++) {
620 for (ch = 0; ch < s->nb_channels; ch++) {
621 if (bit_alloc[ch][i]) {
622 sf = scale_factors[ch][i];
623 switch (scale_code[ch][i]) {
626 sf[0] = get_bits(&s->gb, 6);
627 sf[1] = get_bits(&s->gb, 6);
628 sf[2] = get_bits(&s->gb, 6);
631 sf[0] = get_bits(&s->gb, 6);
636 sf[0] = get_bits(&s->gb, 6);
637 sf[2] = get_bits(&s->gb, 6);
641 sf[0] = get_bits(&s->gb, 6);
642 sf[2] = get_bits(&s->gb, 6);
651 for (k = 0; k < 3; k++) {
652 for (l = 0; l < 12; l += 3) {
654 for (i = 0; i < bound; i++) {
655 bit_alloc_bits = alloc_table[j];
656 for (ch = 0; ch < s->nb_channels; ch++) {
657 b = bit_alloc[ch][i];
659 scale = scale_factors[ch][i][k];
660 qindex = alloc_table[j+b];
661 bits = ff_mpa_quant_bits[qindex];
664 /* 3 values at the same time */
665 v = get_bits(&s->gb, -bits);
666 v2 = division_tabs[qindex][v];
667 steps = ff_mpa_quant_steps[qindex];
669 s->sb_samples[ch][k * 12 + l + 0][i] =
670 l2_unscale_group(steps, v2 & 15, scale);
671 s->sb_samples[ch][k * 12 + l + 1][i] =
672 l2_unscale_group(steps, (v2 >> 4) & 15, scale);
673 s->sb_samples[ch][k * 12 + l + 2][i] =
674 l2_unscale_group(steps, v2 >> 8 , scale);
676 for (m = 0; m < 3; m++) {
677 v = get_bits(&s->gb, bits);
678 v = l1_unscale(bits - 1, v, scale);
679 s->sb_samples[ch][k * 12 + l + m][i] = v;
683 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
684 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
685 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
688 /* next subband in alloc table */
689 j += 1 << bit_alloc_bits;
691 /* XXX: find a way to avoid this duplication of code */
692 for (i = bound; i < sblimit; i++) {
693 bit_alloc_bits = alloc_table[j];
696 int mant, scale0, scale1;
697 scale0 = scale_factors[0][i][k];
698 scale1 = scale_factors[1][i][k];
699 qindex = alloc_table[j+b];
700 bits = ff_mpa_quant_bits[qindex];
702 /* 3 values at the same time */
703 v = get_bits(&s->gb, -bits);
704 steps = ff_mpa_quant_steps[qindex];
707 s->sb_samples[0][k * 12 + l + 0][i] =
708 l2_unscale_group(steps, mant, scale0);
709 s->sb_samples[1][k * 12 + l + 0][i] =
710 l2_unscale_group(steps, mant, scale1);
713 s->sb_samples[0][k * 12 + l + 1][i] =
714 l2_unscale_group(steps, mant, scale0);
715 s->sb_samples[1][k * 12 + l + 1][i] =
716 l2_unscale_group(steps, mant, scale1);
717 s->sb_samples[0][k * 12 + l + 2][i] =
718 l2_unscale_group(steps, v, scale0);
719 s->sb_samples[1][k * 12 + l + 2][i] =
720 l2_unscale_group(steps, v, scale1);
722 for (m = 0; m < 3; m++) {
723 mant = get_bits(&s->gb, bits);
724 s->sb_samples[0][k * 12 + l + m][i] =
725 l1_unscale(bits - 1, mant, scale0);
726 s->sb_samples[1][k * 12 + l + m][i] =
727 l1_unscale(bits - 1, mant, scale1);
731 s->sb_samples[0][k * 12 + l + 0][i] = 0;
732 s->sb_samples[0][k * 12 + l + 1][i] = 0;
733 s->sb_samples[0][k * 12 + l + 2][i] = 0;
734 s->sb_samples[1][k * 12 + l + 0][i] = 0;
735 s->sb_samples[1][k * 12 + l + 1][i] = 0;
736 s->sb_samples[1][k * 12 + l + 2][i] = 0;
738 /* next subband in alloc table */
739 j += 1 << bit_alloc_bits;
741 /* fill remaining samples to zero */
742 for (i = sblimit; i < SBLIMIT; i++) {
743 for (ch = 0; ch < s->nb_channels; ch++) {
744 s->sb_samples[ch][k * 12 + l + 0][i] = 0;
745 s->sb_samples[ch][k * 12 + l + 1][i] = 0;
746 s->sb_samples[ch][k * 12 + l + 2][i] = 0;
754 #define SPLIT(dst,sf,n) \
756 int m = (sf * 171) >> 9; \
759 } else if (n == 4) { \
762 } else if (n == 5) { \
763 int m = (sf * 205) >> 10; \
766 } else if (n == 6) { \
767 int m = (sf * 171) >> 10; \
774 static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2,
777 SPLIT(slen[3], sf, n3)
778 SPLIT(slen[2], sf, n2)
779 SPLIT(slen[1], sf, n1)
783 static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g,
786 const uint8_t *bstab, *pretab;
787 int len, i, j, k, l, v0, shift, gain, gains[3];
791 gain = g->global_gain - 210;
792 shift = g->scalefac_scale + 1;
794 bstab = band_size_long[s->sample_rate_index];
795 pretab = mpa_pretab[g->preflag];
796 for (i = 0; i < g->long_end; i++) {
797 v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400;
799 for (j = len; j > 0; j--)
803 if (g->short_start < 13) {
804 bstab = band_size_short[s->sample_rate_index];
805 gains[0] = gain - (g->subblock_gain[0] << 3);
806 gains[1] = gain - (g->subblock_gain[1] << 3);
807 gains[2] = gain - (g->subblock_gain[2] << 3);
809 for (i = g->short_start; i < 13; i++) {
811 for (l = 0; l < 3; l++) {
812 v0 = gains[l] - (g->scale_factors[k++] << shift) + 400;
813 for (j = len; j > 0; j--)
820 static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos,
823 if (s->in_gb.buffer && *pos >= s->gb.size_in_bits - s->extrasize * 8) {
825 s->in_gb.buffer = NULL;
827 av_assert2((get_bits_count(&s->gb) & 7) == 0);
828 skip_bits_long(&s->gb, *pos - *end_pos);
830 *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos;
831 *pos = get_bits_count(&s->gb);
835 /* Following is an optimized code for
837 if(get_bits1(&s->gb))
842 #define READ_FLIP_SIGN(dst,src) \
843 v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \
846 #define READ_FLIP_SIGN(dst,src) \
847 v = -get_bits1(&s->gb); \
848 *(dst) = (*(src) ^ v) - v;
851 static int huffman_decode(MPADecodeContext *s, GranuleDef *g,
852 int16_t *exponents, int end_pos2)
856 int last_pos, bits_left;
858 int end_pos = FFMIN(end_pos2, s->gb.size_in_bits - s->extrasize * 8);
860 /* low frequencies (called big values) */
862 for (i = 0; i < 3; i++) {
863 int j, k, l, linbits;
864 j = g->region_size[i];
867 /* select vlc table */
868 k = g->table_select[i];
869 l = mpa_huff_data[k][0];
870 linbits = mpa_huff_data[k][1];
874 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j);
879 /* read huffcode and compute each couple */
883 int pos = get_bits_count(&s->gb);
886 switch_buffer(s, &pos, &end_pos, &end_pos2);
890 y = get_vlc2(&s->gb, vlc->table, 7, 3);
893 g->sb_hybrid[s_index ] =
894 g->sb_hybrid[s_index+1] = 0;
899 exponent= exponents[s_index];
901 ff_dlog(s->avctx, "region=%d n=%d y=%d exp=%d\n",
902 i, g->region_size[i] - j, y, exponent);
907 READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x)
909 x += get_bitsz(&s->gb, linbits);
910 v = l3_unscale(x, exponent);
911 if (get_bits1(&s->gb))
913 g->sb_hybrid[s_index] = v;
916 READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y)
918 y += get_bitsz(&s->gb, linbits);
919 v = l3_unscale(y, exponent);
920 if (get_bits1(&s->gb))
922 g->sb_hybrid[s_index+1] = v;
929 READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x)
931 x += get_bitsz(&s->gb, linbits);
932 v = l3_unscale(x, exponent);
933 if (get_bits1(&s->gb))
935 g->sb_hybrid[s_index+!!y] = v;
937 g->sb_hybrid[s_index + !y] = 0;
943 /* high frequencies */
944 vlc = &huff_quad_vlc[g->count1table_select];
946 while (s_index <= 572) {
948 pos = get_bits_count(&s->gb);
949 if (pos >= end_pos) {
950 if (pos > end_pos2 && last_pos) {
951 /* some encoders generate an incorrect size for this
952 part. We must go back into the data */
954 skip_bits_long(&s->gb, last_pos - pos);
955 av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos);
956 if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT))
960 switch_buffer(s, &pos, &end_pos, &end_pos2);
966 code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1);
967 ff_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code);
968 g->sb_hybrid[s_index+0] =
969 g->sb_hybrid[s_index+1] =
970 g->sb_hybrid[s_index+2] =
971 g->sb_hybrid[s_index+3] = 0;
973 static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 };
975 int pos = s_index + idxtab[code];
976 code ^= 8 >> idxtab[code];
977 READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos])
981 /* skip extension bits */
982 bits_left = end_pos2 - get_bits_count(&s->gb);
983 if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) {
984 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
986 } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) {
987 av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left);
990 memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index));
991 skip_bits_long(&s->gb, bits_left);
993 i = get_bits_count(&s->gb);
994 switch_buffer(s, &i, &end_pos, &end_pos2);
999 /* Reorder short blocks from bitstream order to interleaved order. It
1000 would be faster to do it in parsing, but the code would be far more
1002 static void reorder_block(MPADecodeContext *s, GranuleDef *g)
1005 INTFLOAT *ptr, *dst, *ptr1;
1008 if (g->block_type != 2)
1011 if (g->switch_point) {
1012 if (s->sample_rate_index != 8)
1013 ptr = g->sb_hybrid + 36;
1015 ptr = g->sb_hybrid + 72;
1020 for (i = g->short_start; i < 13; i++) {
1021 len = band_size_short[s->sample_rate_index][i];
1024 for (j = len; j > 0; j--) {
1025 *dst++ = ptr[0*len];
1026 *dst++ = ptr[1*len];
1027 *dst++ = ptr[2*len];
1031 memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1));
1035 #define ISQRT2 FIXR(0.70710678118654752440)
1037 static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1)
1040 int sf_max, sf, len, non_zero_found;
1041 INTFLOAT (*is_tab)[16], *tab0, *tab1, v1, v2;
1042 SUINTFLOAT tmp0, tmp1;
1043 int non_zero_found_short[3];
1045 /* intensity stereo */
1046 if (s->mode_ext & MODE_EXT_I_STEREO) {
1051 is_tab = is_table_lsf[g1->scalefac_compress & 1];
1055 tab0 = g0->sb_hybrid + 576;
1056 tab1 = g1->sb_hybrid + 576;
1058 non_zero_found_short[0] = 0;
1059 non_zero_found_short[1] = 0;
1060 non_zero_found_short[2] = 0;
1061 k = (13 - g1->short_start) * 3 + g1->long_end - 3;
1062 for (i = 12; i >= g1->short_start; i--) {
1063 /* for last band, use previous scale factor */
1066 len = band_size_short[s->sample_rate_index][i];
1067 for (l = 2; l >= 0; l--) {
1070 if (!non_zero_found_short[l]) {
1071 /* test if non zero band. if so, stop doing i-stereo */
1072 for (j = 0; j < len; j++) {
1074 non_zero_found_short[l] = 1;
1078 sf = g1->scale_factors[k + l];
1084 for (j = 0; j < len; j++) {
1086 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1087 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1091 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1092 /* lower part of the spectrum : do ms stereo
1094 for (j = 0; j < len; j++) {
1097 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1098 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1105 non_zero_found = non_zero_found_short[0] |
1106 non_zero_found_short[1] |
1107 non_zero_found_short[2];
1109 for (i = g1->long_end - 1;i >= 0;i--) {
1110 len = band_size_long[s->sample_rate_index][i];
1113 /* test if non zero band. if so, stop doing i-stereo */
1114 if (!non_zero_found) {
1115 for (j = 0; j < len; j++) {
1121 /* for last band, use previous scale factor */
1122 k = (i == 21) ? 20 : i;
1123 sf = g1->scale_factors[k];
1128 for (j = 0; j < len; j++) {
1130 tab0[j] = MULLx(tmp0, v1, FRAC_BITS);
1131 tab1[j] = MULLx(tmp0, v2, FRAC_BITS);
1135 if (s->mode_ext & MODE_EXT_MS_STEREO) {
1136 /* lower part of the spectrum : do ms stereo
1138 for (j = 0; j < len; j++) {
1141 tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS);
1142 tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS);
1147 } else if (s->mode_ext & MODE_EXT_MS_STEREO) {
1148 /* ms stereo ONLY */
1149 /* NOTE: the 1/sqrt(2) normalization factor is included in the
1152 s->fdsp->butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
1154 tab0 = g0->sb_hybrid;
1155 tab1 = g1->sb_hybrid;
1156 for (i = 0; i < 576; i++) {
1159 tab0[i] = tmp0 + tmp1;
1160 tab1[i] = tmp0 - tmp1;
1168 # include "mips/compute_antialias_float.h"
1169 #endif /* HAVE_MIPSFPU */
1172 # include "mips/compute_antialias_fixed.h"
1173 #endif /* HAVE_MIPSDSP */
1174 #endif /* USE_FLOATS */
1176 #ifndef compute_antialias
1178 #define AA(j) do { \
1179 float tmp0 = ptr[-1-j]; \
1180 float tmp1 = ptr[ j]; \
1181 ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \
1182 ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \
1185 #define AA(j) do { \
1186 SUINT tmp0 = ptr[-1-j]; \
1187 SUINT tmp1 = ptr[ j]; \
1188 SUINT tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \
1189 ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \
1190 ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \
1194 static void compute_antialias(MPADecodeContext *s, GranuleDef *g)
1199 /* we antialias only "long" bands */
1200 if (g->block_type == 2) {
1201 if (!g->switch_point)
1203 /* XXX: check this for 8000Hz case */
1209 ptr = g->sb_hybrid + 18;
1210 for (i = n; i > 0; i--) {
1223 #endif /* compute_antialias */
1225 static void compute_imdct(MPADecodeContext *s, GranuleDef *g,
1226 INTFLOAT *sb_samples, INTFLOAT *mdct_buf)
1228 INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1;
1230 int i, j, mdct_long_end, sblimit;
1232 /* find last non zero block */
1233 ptr = g->sb_hybrid + 576;
1234 ptr1 = g->sb_hybrid + 2 * 18;
1235 while (ptr >= ptr1) {
1239 if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5])
1242 sblimit = ((ptr - g->sb_hybrid) / 18) + 1;
1244 if (g->block_type == 2) {
1245 /* XXX: check for 8000 Hz */
1246 if (g->switch_point)
1251 mdct_long_end = sblimit;
1254 s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid,
1255 mdct_long_end, g->switch_point,
1258 buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3);
1259 ptr = g->sb_hybrid + 18 * mdct_long_end;
1261 for (j = mdct_long_end; j < sblimit; j++) {
1262 /* select frequency inversion */
1263 win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))];
1264 out_ptr = sb_samples + j;
1266 for (i = 0; i < 6; i++) {
1267 *out_ptr = buf[4*i];
1270 imdct12(out2, ptr + 0);
1271 for (i = 0; i < 6; i++) {
1272 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)];
1273 buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1);
1276 imdct12(out2, ptr + 1);
1277 for (i = 0; i < 6; i++) {
1278 *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)];
1279 buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1);
1282 imdct12(out2, ptr + 2);
1283 for (i = 0; i < 6; i++) {
1284 buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)];
1285 buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1);
1286 buf[4*(i + 6*2)] = 0;
1289 buf += (j&3) != 3 ? 1 : (4*18-3);
1292 for (j = sblimit; j < SBLIMIT; j++) {
1294 out_ptr = sb_samples + j;
1295 for (i = 0; i < 18; i++) {
1296 *out_ptr = buf[4*i];
1300 buf += (j&3) != 3 ? 1 : (4*18-3);
1304 /* main layer3 decoding function */
1305 static int mp_decode_layer3(MPADecodeContext *s)
1307 int nb_granules, main_data_begin;
1308 int gr, ch, blocksplit_flag, i, j, k, n, bits_pos;
1310 int16_t exponents[576]; //FIXME try INTFLOAT
1312 /* read side info */
1314 main_data_begin = get_bits(&s->gb, 8);
1315 skip_bits(&s->gb, s->nb_channels);
1318 main_data_begin = get_bits(&s->gb, 9);
1319 if (s->nb_channels == 2)
1320 skip_bits(&s->gb, 3);
1322 skip_bits(&s->gb, 5);
1324 for (ch = 0; ch < s->nb_channels; ch++) {
1325 s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */
1326 s->granules[ch][1].scfsi = get_bits(&s->gb, 4);
1330 for (gr = 0; gr < nb_granules; gr++) {
1331 for (ch = 0; ch < s->nb_channels; ch++) {
1332 ff_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch);
1333 g = &s->granules[ch][gr];
1334 g->part2_3_length = get_bits(&s->gb, 12);
1335 g->big_values = get_bits(&s->gb, 9);
1336 if (g->big_values > 288) {
1337 av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n");
1338 return AVERROR_INVALIDDATA;
1341 g->global_gain = get_bits(&s->gb, 8);
1342 /* if MS stereo only is selected, we precompute the
1343 1/sqrt(2) renormalization factor */
1344 if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) ==
1346 g->global_gain -= 2;
1348 g->scalefac_compress = get_bits(&s->gb, 9);
1350 g->scalefac_compress = get_bits(&s->gb, 4);
1351 blocksplit_flag = get_bits1(&s->gb);
1352 if (blocksplit_flag) {
1353 g->block_type = get_bits(&s->gb, 2);
1354 if (g->block_type == 0) {
1355 av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n");
1356 return AVERROR_INVALIDDATA;
1358 g->switch_point = get_bits1(&s->gb);
1359 for (i = 0; i < 2; i++)
1360 g->table_select[i] = get_bits(&s->gb, 5);
1361 for (i = 0; i < 3; i++)
1362 g->subblock_gain[i] = get_bits(&s->gb, 3);
1363 init_short_region(s, g);
1365 int region_address1, region_address2;
1367 g->switch_point = 0;
1368 for (i = 0; i < 3; i++)
1369 g->table_select[i] = get_bits(&s->gb, 5);
1370 /* compute huffman coded region sizes */
1371 region_address1 = get_bits(&s->gb, 4);
1372 region_address2 = get_bits(&s->gb, 3);
1373 ff_dlog(s->avctx, "region1=%d region2=%d\n",
1374 region_address1, region_address2);
1375 init_long_region(s, g, region_address1, region_address2);
1377 region_offset2size(g);
1378 compute_band_indexes(s, g);
1382 g->preflag = get_bits1(&s->gb);
1383 g->scalefac_scale = get_bits1(&s->gb);
1384 g->count1table_select = get_bits1(&s->gb);
1385 ff_dlog(s->avctx, "block_type=%d switch_point=%d\n",
1386 g->block_type, g->switch_point);
1392 const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3);
1393 s->extrasize = av_clip((get_bits_left(&s->gb) >> 3) - s->extrasize, 0,
1394 FFMAX(0, LAST_BUF_SIZE - s->last_buf_size));
1395 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1396 /* now we get bits from the main_data_begin offset */
1397 ff_dlog(s->avctx, "seekback:%d, lastbuf:%d\n",
1398 main_data_begin, s->last_buf_size);
1400 memcpy(s->last_buf + s->last_buf_size, ptr, s->extrasize);
1402 init_get_bits(&s->gb, s->last_buf, (s->last_buf_size + s->extrasize) * 8);
1403 s->last_buf_size <<= 3;
1404 for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) {
1405 for (ch = 0; ch < s->nb_channels; ch++) {
1406 g = &s->granules[ch][gr];
1407 s->last_buf_size += g->part2_3_length;
1408 memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid));
1409 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1412 skip = s->last_buf_size - 8 * main_data_begin;
1413 if (skip >= s->gb.size_in_bits - s->extrasize * 8 && s->in_gb.buffer) {
1414 skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits + s->extrasize * 8);
1416 s->in_gb.buffer = NULL;
1419 skip_bits_long(&s->gb, skip);
1426 for (; gr < nb_granules; gr++) {
1427 for (ch = 0; ch < s->nb_channels; ch++) {
1428 g = &s->granules[ch][gr];
1429 bits_pos = get_bits_count(&s->gb);
1433 int slen, slen1, slen2;
1435 /* MPEG-1 scale factors */
1436 slen1 = slen_table[0][g->scalefac_compress];
1437 slen2 = slen_table[1][g->scalefac_compress];
1438 ff_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2);
1439 if (g->block_type == 2) {
1440 n = g->switch_point ? 17 : 18;
1443 for (i = 0; i < n; i++)
1444 g->scale_factors[j++] = get_bits(&s->gb, slen1);
1446 for (i = 0; i < n; i++)
1447 g->scale_factors[j++] = 0;
1450 for (i = 0; i < 18; i++)
1451 g->scale_factors[j++] = get_bits(&s->gb, slen2);
1452 for (i = 0; i < 3; i++)
1453 g->scale_factors[j++] = 0;
1455 for (i = 0; i < 21; i++)
1456 g->scale_factors[j++] = 0;
1459 sc = s->granules[ch][0].scale_factors;
1461 for (k = 0; k < 4; k++) {
1463 if ((g->scfsi & (0x8 >> k)) == 0) {
1464 slen = (k < 2) ? slen1 : slen2;
1466 for (i = 0; i < n; i++)
1467 g->scale_factors[j++] = get_bits(&s->gb, slen);
1469 for (i = 0; i < n; i++)
1470 g->scale_factors[j++] = 0;
1473 /* simply copy from last granule */
1474 for (i = 0; i < n; i++) {
1475 g->scale_factors[j] = sc[j];
1480 g->scale_factors[j++] = 0;
1483 int tindex, tindex2, slen[4], sl, sf;
1485 /* LSF scale factors */
1486 if (g->block_type == 2)
1487 tindex = g->switch_point ? 2 : 1;
1491 sf = g->scalefac_compress;
1492 if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) {
1493 /* intensity stereo case */
1496 lsf_sf_expand(slen, sf, 6, 6, 0);
1498 } else if (sf < 244) {
1499 lsf_sf_expand(slen, sf - 180, 4, 4, 0);
1502 lsf_sf_expand(slen, sf - 244, 3, 0, 0);
1508 lsf_sf_expand(slen, sf, 5, 4, 4);
1510 } else if (sf < 500) {
1511 lsf_sf_expand(slen, sf - 400, 5, 4, 0);
1514 lsf_sf_expand(slen, sf - 500, 3, 0, 0);
1521 for (k = 0; k < 4; k++) {
1522 n = lsf_nsf_table[tindex2][tindex][k];
1525 for (i = 0; i < n; i++)
1526 g->scale_factors[j++] = get_bits(&s->gb, sl);
1528 for (i = 0; i < n; i++)
1529 g->scale_factors[j++] = 0;
1532 /* XXX: should compute exact size */
1534 g->scale_factors[j] = 0;
1537 exponents_from_scale_factors(s, g, exponents);
1539 /* read Huffman coded residue */
1540 huffman_decode(s, g, exponents, bits_pos + g->part2_3_length);
1543 if (s->mode == MPA_JSTEREO)
1544 compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]);
1546 for (ch = 0; ch < s->nb_channels; ch++) {
1547 g = &s->granules[ch][gr];
1549 reorder_block(s, g);
1550 compute_antialias(s, g);
1551 compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]);
1554 if (get_bits_count(&s->gb) < 0)
1555 skip_bits_long(&s->gb, -get_bits_count(&s->gb));
1556 return nb_granules * 18;
1559 static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples,
1560 const uint8_t *buf, int buf_size)
1562 int i, nb_frames, ch, ret;
1563 OUT_INT *samples_ptr;
1565 init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8);
1567 /* skip error protection field */
1568 if (s->error_protection)
1569 skip_bits(&s->gb, 16);
1573 s->avctx->frame_size = 384;
1574 nb_frames = mp_decode_layer1(s);
1577 s->avctx->frame_size = 1152;
1578 nb_frames = mp_decode_layer2(s);
1581 s->avctx->frame_size = s->lsf ? 576 : 1152;
1583 nb_frames = mp_decode_layer3(s);
1586 if (s->in_gb.buffer) {
1587 align_get_bits(&s->gb);
1588 i = (get_bits_left(&s->gb) >> 3) - s->extrasize;
1589 if (i >= 0 && i <= BACKSTEP_SIZE) {
1590 memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i);
1593 av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i);
1595 s->in_gb.buffer = NULL;
1599 align_get_bits(&s->gb);
1600 av_assert1((get_bits_count(&s->gb) & 7) == 0);
1601 i = (get_bits_left(&s->gb) >> 3) - s->extrasize;
1602 if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) {
1604 av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i);
1605 i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE);
1607 av_assert1(i <= buf_size - HEADER_SIZE && i >= 0);
1608 memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i);
1609 s->last_buf_size += i;
1615 /* get output buffer */
1617 av_assert0(s->frame);
1618 s->frame->nb_samples = s->avctx->frame_size;
1619 if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0)
1621 samples = (OUT_INT **)s->frame->extended_data;
1624 /* apply the synthesis filter */
1625 for (ch = 0; ch < s->nb_channels; ch++) {
1627 if (s->avctx->sample_fmt == OUT_FMT_P) {
1628 samples_ptr = samples[ch];
1631 samples_ptr = samples[0] + ch;
1632 sample_stride = s->nb_channels;
1634 for (i = 0; i < nb_frames; i++) {
1635 RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch],
1636 &(s->synth_buf_offset[ch]),
1637 RENAME(ff_mpa_synth_window),
1638 &s->dither_state, samples_ptr,
1639 sample_stride, s->sb_samples[ch][i]);
1640 samples_ptr += 32 * sample_stride;
1644 return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels;
1647 static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr,
1650 const uint8_t *buf = avpkt->data;
1651 int buf_size = avpkt->size;
1652 MPADecodeContext *s = avctx->priv_data;
1657 while(buf_size && !*buf){
1663 if (buf_size < HEADER_SIZE)
1664 return AVERROR_INVALIDDATA;
1666 header = AV_RB32(buf);
1667 if (header>>8 == AV_RB32("TAG")>>8) {
1668 av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n");
1669 return buf_size + skipped;
1671 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1673 av_log(avctx, AV_LOG_ERROR, "Header missing\n");
1674 return AVERROR_INVALIDDATA;
1675 } else if (ret == 1) {
1676 /* free format: prepare to compute frame size */
1678 return AVERROR_INVALIDDATA;
1680 /* update codec info */
1681 avctx->channels = s->nb_channels;
1682 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1683 if (!avctx->bit_rate)
1684 avctx->bit_rate = s->bit_rate;
1686 if (s->frame_size <= 0) {
1687 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1688 return AVERROR_INVALIDDATA;
1689 } else if (s->frame_size < buf_size) {
1690 av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n");
1691 buf_size= s->frame_size;
1696 ret = mp_decode_frame(s, NULL, buf, buf_size);
1698 s->frame->nb_samples = avctx->frame_size;
1700 avctx->sample_rate = s->sample_rate;
1701 //FIXME maybe move the other codec info stuff from above here too
1703 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1704 /* Only return an error if the bad frame makes up the whole packet or
1705 * the error is related to buffer management.
1706 * If there is more data in the packet, just consume the bad frame
1707 * instead of returning an error, which would discard the whole
1710 if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA)
1714 return buf_size + skipped;
1717 static void mp_flush(MPADecodeContext *ctx)
1719 memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf));
1720 memset(ctx->mdct_buf, 0, sizeof(ctx->mdct_buf));
1721 ctx->last_buf_size = 0;
1722 ctx->dither_state = 0;
1725 static void flush(AVCodecContext *avctx)
1727 mp_flush(avctx->priv_data);
1730 #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER
1731 static int decode_frame_adu(AVCodecContext *avctx, void *data,
1732 int *got_frame_ptr, AVPacket *avpkt)
1734 const uint8_t *buf = avpkt->data;
1735 int buf_size = avpkt->size;
1736 MPADecodeContext *s = avctx->priv_data;
1739 int av_unused out_size;
1743 // Discard too short frames
1744 if (buf_size < HEADER_SIZE) {
1745 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1746 return AVERROR_INVALIDDATA;
1750 if (len > MPA_MAX_CODED_FRAME_SIZE)
1751 len = MPA_MAX_CODED_FRAME_SIZE;
1753 // Get header and restore sync word
1754 header = AV_RB32(buf) | 0xffe00000;
1756 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header);
1758 av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n");
1761 /* update codec info */
1762 avctx->sample_rate = s->sample_rate;
1763 avctx->channels = s->nb_channels;
1764 avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
1765 if (!avctx->bit_rate)
1766 avctx->bit_rate = s->bit_rate;
1768 s->frame_size = len;
1772 ret = mp_decode_frame(s, NULL, buf, buf_size);
1774 av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n");
1782 #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */
1784 #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER
1787 * Context for MP3On4 decoder
1789 typedef struct MP3On4DecodeContext {
1790 int frames; ///< number of mp3 frames per block (number of mp3 decoder instances)
1791 int syncword; ///< syncword patch
1792 const uint8_t *coff; ///< channel offsets in output buffer
1793 MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance
1794 } MP3On4DecodeContext;
1796 #include "mpeg4audio.h"
1798 /* Next 3 arrays are indexed by channel config number (passed via codecdata) */
1800 /* number of mp3 decoder instances */
1801 static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 };
1803 /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */
1804 static const uint8_t chan_offset[8][5] = {
1809 { 2, 0, 3 }, // C FLR BS
1810 { 2, 0, 3 }, // C FLR BLRS
1811 { 2, 0, 4, 3 }, // C FLR BLRS LFE
1812 { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE
1815 /* mp3on4 channel layouts */
1816 static const int16_t chan_layout[8] = {
1819 AV_CH_LAYOUT_STEREO,
1820 AV_CH_LAYOUT_SURROUND,
1821 AV_CH_LAYOUT_4POINT0,
1822 AV_CH_LAYOUT_5POINT0,
1823 AV_CH_LAYOUT_5POINT1,
1824 AV_CH_LAYOUT_7POINT1
1827 static av_cold int decode_close_mp3on4(AVCodecContext * avctx)
1829 MP3On4DecodeContext *s = avctx->priv_data;
1832 if (s->mp3decctx[0])
1833 av_freep(&s->mp3decctx[0]->fdsp);
1835 for (i = 0; i < s->frames; i++)
1836 av_freep(&s->mp3decctx[i]);
1842 static av_cold int decode_init_mp3on4(AVCodecContext * avctx)
1844 MP3On4DecodeContext *s = avctx->priv_data;
1845 MPEG4AudioConfig cfg;
1848 if ((avctx->extradata_size < 2) || !avctx->extradata) {
1849 av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n");
1850 return AVERROR_INVALIDDATA;
1853 avpriv_mpeg4audio_get_config(&cfg, avctx->extradata,
1854 avctx->extradata_size * 8, 1);
1855 if (!cfg.chan_config || cfg.chan_config > 7) {
1856 av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n");
1857 return AVERROR_INVALIDDATA;
1859 s->frames = mp3Frames[cfg.chan_config];
1860 s->coff = chan_offset[cfg.chan_config];
1861 avctx->channels = ff_mpeg4audio_channels[cfg.chan_config];
1862 avctx->channel_layout = chan_layout[cfg.chan_config];
1864 if (cfg.sample_rate < 16000)
1865 s->syncword = 0xffe00000;
1867 s->syncword = 0xfff00000;
1869 /* Init the first mp3 decoder in standard way, so that all tables get builded
1870 * We replace avctx->priv_data with the context of the first decoder so that
1871 * decode_init() does not have to be changed.
1872 * Other decoders will be initialized here copying data from the first context
1874 // Allocate zeroed memory for the first decoder context
1875 s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext));
1876 if (!s->mp3decctx[0])
1878 // Put decoder context in place to make init_decode() happy
1879 avctx->priv_data = s->mp3decctx[0];
1881 // Restore mp3on4 context pointer
1882 avctx->priv_data = s;
1883 s->mp3decctx[0]->adu_mode = 1; // Set adu mode
1885 /* Create a separate codec/context for each frame (first is already ok).
1886 * Each frame is 1 or 2 channels - up to 5 frames allowed
1888 for (i = 1; i < s->frames; i++) {
1889 s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext));
1890 if (!s->mp3decctx[i])
1892 s->mp3decctx[i]->adu_mode = 1;
1893 s->mp3decctx[i]->avctx = avctx;
1894 s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp;
1895 s->mp3decctx[i]->fdsp = s->mp3decctx[0]->fdsp;
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 ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header);
1956 av_log(avctx, AV_LOG_ERROR, "Bad header, discard block\n");
1957 return AVERROR_INVALIDDATA;
1960 if (ch + m->nb_channels > avctx->channels ||
1961 s->coff[fr] + m->nb_channels > avctx->channels) {
1962 av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec "
1964 return AVERROR_INVALIDDATA;
1966 ch += m->nb_channels;
1968 outptr[0] = out_samples[s->coff[fr]];
1969 if (m->nb_channels > 1)
1970 outptr[1] = out_samples[s->coff[fr] + 1];
1972 if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0) {
1973 av_log(avctx, AV_LOG_ERROR, "failed to decode channel %d\n", ch);
1974 memset(outptr[0], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1975 if (m->nb_channels > 1)
1976 memset(outptr[1], 0, MPA_FRAME_SIZE*sizeof(OUT_INT));
1977 ret = m->nb_channels * MPA_FRAME_SIZE*sizeof(OUT_INT);
1984 avctx->bit_rate += m->bit_rate;
1986 if (ch != avctx->channels) {
1987 av_log(avctx, AV_LOG_ERROR, "failed to decode all channels\n");
1988 return AVERROR_INVALIDDATA;
1991 /* update codec info */
1992 avctx->sample_rate = s->mp3decctx[0]->sample_rate;
1994 frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT));
1999 #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */