blob: 9cce88e263b730981feb49506b94ba32a121b732
1 | /* |
2 | * MPEG Audio decoder |
3 | * Copyright (c) 2001, 2002 Fabrice Bellard |
4 | * |
5 | * This file is part of FFmpeg. |
6 | * |
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. |
11 | * |
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. |
16 | * |
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 |
20 | */ |
21 | |
22 | /** |
23 | * @file |
24 | * MPEG Audio decoder |
25 | */ |
26 | |
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" |
32 | #include "avcodec.h" |
33 | #include "get_bits.h" |
34 | #include "internal.h" |
35 | #include "mathops.h" |
36 | #include "mpegaudiodsp.h" |
37 | |
38 | /* |
39 | * TODO: |
40 | * - test lsf / mpeg25 extensively. |
41 | */ |
42 | |
43 | #include "mpegaudio.h" |
44 | #include "mpegaudiodecheader.h" |
45 | |
46 | #define BACKSTEP_SIZE 512 |
47 | #define EXTRABYTES 24 |
48 | #define LAST_BUF_SIZE 2 * BACKSTEP_SIZE + EXTRABYTES |
49 | |
50 | /* layer 3 "granule" */ |
51 | typedef struct GranuleDef { |
52 | uint8_t scfsi; |
53 | int part2_3_length; |
54 | int big_values; |
55 | int global_gain; |
56 | int scalefac_compress; |
57 | uint8_t block_type; |
58 | uint8_t switch_point; |
59 | int table_select[3]; |
60 | int subblock_gain[3]; |
61 | uint8_t scalefac_scale; |
62 | uint8_t count1table_select; |
63 | int region_size[3]; /* number of huffman codes in each region */ |
64 | int preflag; |
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 */ |
68 | } GranuleDef; |
69 | |
70 | typedef struct MPADecodeContext { |
71 | MPA_DECODE_HEADER |
72 | uint8_t last_buf[LAST_BUF_SIZE]; |
73 | int last_buf_size; |
74 | int extrasize; |
75 | /* next header (used in free format parsing) */ |
76 | uint32_t free_format_next_header; |
77 | GetBitContext gb; |
78 | GetBitContext in_gb; |
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 |
85 | int dither_state; |
86 | int err_recognition; |
87 | AVCodecContext* avctx; |
88 | MPADSPContext mpadsp; |
89 | AVFloatDSPContext *fdsp; |
90 | AVFrame *frame; |
91 | } MPADecodeContext; |
92 | |
93 | #define HEADER_SIZE 4 |
94 | |
95 | #include "mpegaudiodata.h" |
96 | #include "mpegaudiodectab.h" |
97 | |
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 |
103 | ][2]; |
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 |
107 | }; |
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]; |
118 | |
119 | static int16_t division_tab3[1<<6 ]; |
120 | static int16_t division_tab5[1<<8 ]; |
121 | static int16_t division_tab9[1<<11]; |
122 | |
123 | static int16_t * const division_tabs[4] = { |
124 | division_tab3, division_tab5, NULL, division_tab9 |
125 | }; |
126 | |
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 */ |
132 | |
133 | #define SCALE_GEN(v) \ |
134 | { FIXR_OLD(1.0 * (v)), FIXR_OLD(0.7937005259 * (v)), FIXR_OLD(0.6299605249 * (v)) } |
135 | |
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 */ |
140 | }; |
141 | |
142 | /** |
143 | * Convert region offsets to region sizes and truncate |
144 | * size to big_values. |
145 | */ |
146 | static void region_offset2size(GranuleDef *g) |
147 | { |
148 | int i, k, j = 0; |
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; |
153 | j = k; |
154 | } |
155 | } |
156 | |
157 | static void init_short_region(MPADecodeContext *s, GranuleDef *g) |
158 | { |
159 | if (g->block_type == 2) { |
160 | if (s->sample_rate_index != 8) |
161 | g->region_size[0] = (36 / 2); |
162 | else |
163 | g->region_size[0] = (72 / 2); |
164 | } else { |
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); |
169 | else |
170 | g->region_size[0] = (108 / 2); |
171 | } |
172 | g->region_size[1] = (576 / 2); |
173 | } |
174 | |
175 | static void init_long_region(MPADecodeContext *s, GranuleDef *g, |
176 | int ra1, int ra2) |
177 | { |
178 | int l; |
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; |
183 | } |
184 | |
185 | static void compute_band_indexes(MPADecodeContext *s, GranuleDef *g) |
186 | { |
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) |
195 | g->long_end = 8; |
196 | else |
197 | g->long_end = 6; |
198 | |
199 | g->short_start = 3; |
200 | } else { |
201 | g->long_end = 0; |
202 | g->short_start = 0; |
203 | } |
204 | } else { |
205 | g->short_start = 13; |
206 | g->long_end = 22; |
207 | } |
208 | } |
209 | |
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) |
213 | { |
214 | int shift, mod; |
215 | int64_t val; |
216 | |
217 | shift = scale_factor_modshift[scale_factor]; |
218 | mod = shift & 3; |
219 | shift >>= 2; |
220 | val = MUL64((int)(mant + (-1U << n) + 1), scale_factor_mult[n-1][mod]); |
221 | shift += n; |
222 | /* NOTE: at this point, 1 <= shift >= 21 + 15 */ |
223 | return (int)((val + (1LL << (shift - 1))) >> shift); |
224 | } |
225 | |
226 | static inline int l2_unscale_group(int steps, int mant, int scale_factor) |
227 | { |
228 | int shift, mod, val; |
229 | |
230 | shift = scale_factor_modshift[scale_factor]; |
231 | mod = shift & 3; |
232 | shift >>= 2; |
233 | |
234 | val = (mant - (steps >> 1)) * scale_factor_mult2[steps >> 2][mod]; |
235 | /* NOTE: at this point, 0 <= shift <= 21 */ |
236 | if (shift > 0) |
237 | val = (val + (1 << (shift - 1))) >> shift; |
238 | return val; |
239 | } |
240 | |
241 | /* compute value^(4/3) * 2^(exponent/4). It normalized to FRAC_BITS */ |
242 | static inline int l3_unscale(int value, int exponent) |
243 | { |
244 | unsigned int m; |
245 | int e; |
246 | |
247 | e = table_4_3_exp [4 * value + (exponent & 3)]; |
248 | m = table_4_3_value[4 * value + (exponent & 3)]; |
249 | e -= exponent >> 2; |
250 | #ifdef DEBUG |
251 | if(e < 1) |
252 | av_log(NULL, AV_LOG_WARNING, "l3_unscale: e is %d\n", e); |
253 | #endif |
254 | if (e > (SUINT)31) |
255 | return 0; |
256 | m = (m + ((1U << e)>>1)) >> e; |
257 | |
258 | return m; |
259 | } |
260 | |
261 | static av_cold void decode_init_static(void) |
262 | { |
263 | int i, j, k; |
264 | int offset; |
265 | |
266 | /* scale factors table for layer 1/2 */ |
267 | for (i = 0; i < 64; i++) { |
268 | int shift, mod; |
269 | /* 1.0 (i = 3) is normalized to 2 ^ FRAC_BITS */ |
270 | shift = i / 3; |
271 | mod = i % 3; |
272 | scale_factor_modshift[i] = mod | (shift << 2); |
273 | } |
274 | |
275 | /* scale factor multiply for layer 1 */ |
276 | for (i = 0; i < 15; i++) { |
277 | int n, norm; |
278 | n = i + 2; |
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=%"PRIx32" %"PRIx32" %"PRIx32"\n", i, |
284 | (unsigned)norm, |
285 | scale_factor_mult[i][0], |
286 | scale_factor_mult[i][1], |
287 | scale_factor_mult[i][2]); |
288 | } |
289 | |
290 | RENAME(ff_mpa_synth_init)(RENAME(ff_mpa_synth_window)); |
291 | |
292 | /* huffman decode tables */ |
293 | offset = 0; |
294 | for (i = 1; i < 16; i++) { |
295 | const HuffTable *h = &mpa_huff_tables[i]; |
296 | int xsize, x, y; |
297 | uint8_t tmp_bits [512] = { 0 }; |
298 | uint16_t tmp_codes[512] = { 0 }; |
299 | |
300 | xsize = h->xsize; |
301 | |
302 | j = 0; |
303 | for (x = 0; x < xsize; x++) { |
304 | for (y = 0; y < xsize; y++) { |
305 | tmp_bits [(x << 5) | y | ((x&&y)<<4)]= h->bits [j ]; |
306 | tmp_codes[(x << 5) | y | ((x&&y)<<4)]= h->codes[j++]; |
307 | } |
308 | } |
309 | |
310 | /* XXX: fail test */ |
311 | huff_vlc[i].table = huff_vlc_tables+offset; |
312 | huff_vlc[i].table_allocated = huff_vlc_tables_sizes[i]; |
313 | init_vlc(&huff_vlc[i], 7, 512, |
314 | tmp_bits, 1, 1, tmp_codes, 2, 2, |
315 | INIT_VLC_USE_NEW_STATIC); |
316 | offset += huff_vlc_tables_sizes[i]; |
317 | } |
318 | av_assert0(offset == FF_ARRAY_ELEMS(huff_vlc_tables)); |
319 | |
320 | offset = 0; |
321 | for (i = 0; i < 2; i++) { |
322 | huff_quad_vlc[i].table = huff_quad_vlc_tables+offset; |
323 | huff_quad_vlc[i].table_allocated = huff_quad_vlc_tables_sizes[i]; |
324 | init_vlc(&huff_quad_vlc[i], i == 0 ? 7 : 4, 16, |
325 | mpa_quad_bits[i], 1, 1, mpa_quad_codes[i], 1, 1, |
326 | INIT_VLC_USE_NEW_STATIC); |
327 | offset += huff_quad_vlc_tables_sizes[i]; |
328 | } |
329 | av_assert0(offset == FF_ARRAY_ELEMS(huff_quad_vlc_tables)); |
330 | |
331 | for (i = 0; i < 9; i++) { |
332 | k = 0; |
333 | for (j = 0; j < 22; j++) { |
334 | band_index_long[i][j] = k; |
335 | k += band_size_long[i][j]; |
336 | } |
337 | band_index_long[i][22] = k; |
338 | } |
339 | |
340 | /* compute n ^ (4/3) and store it in mantissa/exp format */ |
341 | |
342 | mpegaudio_tableinit(); |
343 | |
344 | for (i = 0; i < 4; i++) { |
345 | if (ff_mpa_quant_bits[i] < 0) { |
346 | for (j = 0; j < (1 << (-ff_mpa_quant_bits[i]+1)); j++) { |
347 | int val1, val2, val3, steps; |
348 | int val = j; |
349 | steps = ff_mpa_quant_steps[i]; |
350 | val1 = val % steps; |
351 | val /= steps; |
352 | val2 = val % steps; |
353 | val3 = val / steps; |
354 | division_tabs[i][j] = val1 + (val2 << 4) + (val3 << 8); |
355 | } |
356 | } |
357 | } |
358 | |
359 | |
360 | for (i = 0; i < 7; i++) { |
361 | float f; |
362 | INTFLOAT v; |
363 | if (i != 6) { |
364 | f = tan((double)i * M_PI / 12.0); |
365 | v = FIXR(f / (1.0 + f)); |
366 | } else { |
367 | v = FIXR(1.0); |
368 | } |
369 | is_table[0][ i] = v; |
370 | is_table[1][6 - i] = v; |
371 | } |
372 | /* invalid values */ |
373 | for (i = 7; i < 16; i++) |
374 | is_table[0][i] = is_table[1][i] = 0.0; |
375 | |
376 | for (i = 0; i < 16; i++) { |
377 | double f; |
378 | int e, k; |
379 | |
380 | for (j = 0; j < 2; j++) { |
381 | e = -(j + 1) * ((i + 1) >> 1); |
382 | f = exp2(e / 4.0); |
383 | k = i & 1; |
384 | is_table_lsf[j][k ^ 1][i] = FIXR(f); |
385 | is_table_lsf[j][k ][i] = FIXR(1.0); |
386 | ff_dlog(NULL, "is_table_lsf %d %d: %f %f\n", |
387 | i, j, (float) is_table_lsf[j][0][i], |
388 | (float) is_table_lsf[j][1][i]); |
389 | } |
390 | } |
391 | |
392 | for (i = 0; i < 8; i++) { |
393 | double ci, cs, ca; |
394 | ci = ci_table[i]; |
395 | cs = 1.0 / sqrt(1.0 + ci * ci); |
396 | ca = cs * ci; |
397 | #if !USE_FLOATS |
398 | csa_table[i][0] = FIXHR(cs/4); |
399 | csa_table[i][1] = FIXHR(ca/4); |
400 | csa_table[i][2] = FIXHR(ca/4) + FIXHR(cs/4); |
401 | csa_table[i][3] = FIXHR(ca/4) - FIXHR(cs/4); |
402 | #else |
403 | csa_table[i][0] = cs; |
404 | csa_table[i][1] = ca; |
405 | csa_table[i][2] = ca + cs; |
406 | csa_table[i][3] = ca - cs; |
407 | #endif |
408 | } |
409 | } |
410 | |
411 | #if USE_FLOATS |
412 | static av_cold int decode_close(AVCodecContext * avctx) |
413 | { |
414 | MPADecodeContext *s = avctx->priv_data; |
415 | av_freep(&s->fdsp); |
416 | |
417 | return 0; |
418 | } |
419 | #endif |
420 | |
421 | static av_cold int decode_init(AVCodecContext * avctx) |
422 | { |
423 | static int initialized_tables = 0; |
424 | MPADecodeContext *s = avctx->priv_data; |
425 | |
426 | if (!initialized_tables) { |
427 | decode_init_static(); |
428 | initialized_tables = 1; |
429 | } |
430 | |
431 | s->avctx = avctx; |
432 | |
433 | #if USE_FLOATS |
434 | s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT); |
435 | if (!s->fdsp) |
436 | return AVERROR(ENOMEM); |
437 | #endif |
438 | |
439 | ff_mpadsp_init(&s->mpadsp); |
440 | |
441 | if (avctx->request_sample_fmt == OUT_FMT && |
442 | avctx->codec_id != AV_CODEC_ID_MP3ON4) |
443 | avctx->sample_fmt = OUT_FMT; |
444 | else |
445 | avctx->sample_fmt = OUT_FMT_P; |
446 | s->err_recognition = avctx->err_recognition; |
447 | |
448 | if (avctx->codec_id == AV_CODEC_ID_MP3ADU) |
449 | s->adu_mode = 1; |
450 | |
451 | return 0; |
452 | } |
453 | |
454 | #define C3 FIXHR(0.86602540378443864676/2) |
455 | #define C4 FIXHR(0.70710678118654752439/2) //0.5 / cos(pi*(9)/36) |
456 | #define C5 FIXHR(0.51763809020504152469/2) //0.5 / cos(pi*(5)/36) |
457 | #define C6 FIXHR(1.93185165257813657349/4) //0.5 / cos(pi*(15)/36) |
458 | |
459 | /* 12 points IMDCT. We compute it "by hand" by factorizing obvious |
460 | cases. */ |
461 | static void imdct12(INTFLOAT *out, SUINTFLOAT *in) |
462 | { |
463 | SUINTFLOAT in0, in1, in2, in3, in4, in5, t1, t2; |
464 | |
465 | in0 = in[0*3]; |
466 | in1 = in[1*3] + in[0*3]; |
467 | in2 = in[2*3] + in[1*3]; |
468 | in3 = in[3*3] + in[2*3]; |
469 | in4 = in[4*3] + in[3*3]; |
470 | in5 = in[5*3] + in[4*3]; |
471 | in5 += in3; |
472 | in3 += in1; |
473 | |
474 | in2 = MULH3(in2, C3, 2); |
475 | in3 = MULH3(in3, C3, 4); |
476 | |
477 | t1 = in0 - in4; |
478 | t2 = MULH3(in1 - in5, C4, 2); |
479 | |
480 | out[ 7] = |
481 | out[10] = t1 + t2; |
482 | out[ 1] = |
483 | out[ 4] = t1 - t2; |
484 | |
485 | in0 += SHR(in4, 1); |
486 | in4 = in0 + in2; |
487 | in5 += 2*in1; |
488 | in1 = MULH3(in5 + in3, C5, 1); |
489 | out[ 8] = |
490 | out[ 9] = in4 + in1; |
491 | out[ 2] = |
492 | out[ 3] = in4 - in1; |
493 | |
494 | in0 -= in2; |
495 | in5 = MULH3(in5 - in3, C6, 2); |
496 | out[ 0] = |
497 | out[ 5] = in0 - in5; |
498 | out[ 6] = |
499 | out[11] = in0 + in5; |
500 | } |
501 | |
502 | /* return the number of decoded frames */ |
503 | static int mp_decode_layer1(MPADecodeContext *s) |
504 | { |
505 | int bound, i, v, n, ch, j, mant; |
506 | uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT]; |
507 | uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT]; |
508 | |
509 | if (s->mode == MPA_JSTEREO) |
510 | bound = (s->mode_ext + 1) * 4; |
511 | else |
512 | bound = SBLIMIT; |
513 | |
514 | /* allocation bits */ |
515 | for (i = 0; i < bound; i++) { |
516 | for (ch = 0; ch < s->nb_channels; ch++) { |
517 | allocation[ch][i] = get_bits(&s->gb, 4); |
518 | } |
519 | } |
520 | for (i = bound; i < SBLIMIT; i++) |
521 | allocation[0][i] = get_bits(&s->gb, 4); |
522 | |
523 | /* scale factors */ |
524 | for (i = 0; i < bound; i++) { |
525 | for (ch = 0; ch < s->nb_channels; ch++) { |
526 | if (allocation[ch][i]) |
527 | scale_factors[ch][i] = get_bits(&s->gb, 6); |
528 | } |
529 | } |
530 | for (i = bound; i < SBLIMIT; i++) { |
531 | if (allocation[0][i]) { |
532 | scale_factors[0][i] = get_bits(&s->gb, 6); |
533 | scale_factors[1][i] = get_bits(&s->gb, 6); |
534 | } |
535 | } |
536 | |
537 | /* compute samples */ |
538 | for (j = 0; j < 12; j++) { |
539 | for (i = 0; i < bound; i++) { |
540 | for (ch = 0; ch < s->nb_channels; ch++) { |
541 | n = allocation[ch][i]; |
542 | if (n) { |
543 | mant = get_bits(&s->gb, n + 1); |
544 | v = l1_unscale(n, mant, scale_factors[ch][i]); |
545 | } else { |
546 | v = 0; |
547 | } |
548 | s->sb_samples[ch][j][i] = v; |
549 | } |
550 | } |
551 | for (i = bound; i < SBLIMIT; i++) { |
552 | n = allocation[0][i]; |
553 | if (n) { |
554 | mant = get_bits(&s->gb, n + 1); |
555 | v = l1_unscale(n, mant, scale_factors[0][i]); |
556 | s->sb_samples[0][j][i] = v; |
557 | v = l1_unscale(n, mant, scale_factors[1][i]); |
558 | s->sb_samples[1][j][i] = v; |
559 | } else { |
560 | s->sb_samples[0][j][i] = 0; |
561 | s->sb_samples[1][j][i] = 0; |
562 | } |
563 | } |
564 | } |
565 | return 12; |
566 | } |
567 | |
568 | static int mp_decode_layer2(MPADecodeContext *s) |
569 | { |
570 | int sblimit; /* number of used subbands */ |
571 | const unsigned char *alloc_table; |
572 | int table, bit_alloc_bits, i, j, ch, bound, v; |
573 | unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT]; |
574 | unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT]; |
575 | unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf; |
576 | int scale, qindex, bits, steps, k, l, m, b; |
577 | |
578 | /* select decoding table */ |
579 | table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels, |
580 | s->sample_rate, s->lsf); |
581 | sblimit = ff_mpa_sblimit_table[table]; |
582 | alloc_table = ff_mpa_alloc_tables[table]; |
583 | |
584 | if (s->mode == MPA_JSTEREO) |
585 | bound = (s->mode_ext + 1) * 4; |
586 | else |
587 | bound = sblimit; |
588 | |
589 | ff_dlog(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit); |
590 | |
591 | /* sanity check */ |
592 | if (bound > sblimit) |
593 | bound = sblimit; |
594 | |
595 | /* parse bit allocation */ |
596 | j = 0; |
597 | for (i = 0; i < bound; i++) { |
598 | bit_alloc_bits = alloc_table[j]; |
599 | for (ch = 0; ch < s->nb_channels; ch++) |
600 | bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits); |
601 | j += 1 << bit_alloc_bits; |
602 | } |
603 | for (i = bound; i < sblimit; i++) { |
604 | bit_alloc_bits = alloc_table[j]; |
605 | v = get_bits(&s->gb, bit_alloc_bits); |
606 | bit_alloc[0][i] = v; |
607 | bit_alloc[1][i] = v; |
608 | j += 1 << bit_alloc_bits; |
609 | } |
610 | |
611 | /* scale codes */ |
612 | for (i = 0; i < sblimit; i++) { |
613 | for (ch = 0; ch < s->nb_channels; ch++) { |
614 | if (bit_alloc[ch][i]) |
615 | scale_code[ch][i] = get_bits(&s->gb, 2); |
616 | } |
617 | } |
618 | |
619 | /* scale factors */ |
620 | for (i = 0; i < sblimit; i++) { |
621 | for (ch = 0; ch < s->nb_channels; ch++) { |
622 | if (bit_alloc[ch][i]) { |
623 | sf = scale_factors[ch][i]; |
624 | switch (scale_code[ch][i]) { |
625 | default: |
626 | case 0: |
627 | sf[0] = get_bits(&s->gb, 6); |
628 | sf[1] = get_bits(&s->gb, 6); |
629 | sf[2] = get_bits(&s->gb, 6); |
630 | break; |
631 | case 2: |
632 | sf[0] = get_bits(&s->gb, 6); |
633 | sf[1] = sf[0]; |
634 | sf[2] = sf[0]; |
635 | break; |
636 | case 1: |
637 | sf[0] = get_bits(&s->gb, 6); |
638 | sf[2] = get_bits(&s->gb, 6); |
639 | sf[1] = sf[0]; |
640 | break; |
641 | case 3: |
642 | sf[0] = get_bits(&s->gb, 6); |
643 | sf[2] = get_bits(&s->gb, 6); |
644 | sf[1] = sf[2]; |
645 | break; |
646 | } |
647 | } |
648 | } |
649 | } |
650 | |
651 | /* samples */ |
652 | for (k = 0; k < 3; k++) { |
653 | for (l = 0; l < 12; l += 3) { |
654 | j = 0; |
655 | for (i = 0; i < bound; i++) { |
656 | bit_alloc_bits = alloc_table[j]; |
657 | for (ch = 0; ch < s->nb_channels; ch++) { |
658 | b = bit_alloc[ch][i]; |
659 | if (b) { |
660 | scale = scale_factors[ch][i][k]; |
661 | qindex = alloc_table[j+b]; |
662 | bits = ff_mpa_quant_bits[qindex]; |
663 | if (bits < 0) { |
664 | int v2; |
665 | /* 3 values at the same time */ |
666 | v = get_bits(&s->gb, -bits); |
667 | v2 = division_tabs[qindex][v]; |
668 | steps = ff_mpa_quant_steps[qindex]; |
669 | |
670 | s->sb_samples[ch][k * 12 + l + 0][i] = |
671 | l2_unscale_group(steps, v2 & 15, scale); |
672 | s->sb_samples[ch][k * 12 + l + 1][i] = |
673 | l2_unscale_group(steps, (v2 >> 4) & 15, scale); |
674 | s->sb_samples[ch][k * 12 + l + 2][i] = |
675 | l2_unscale_group(steps, v2 >> 8 , scale); |
676 | } else { |
677 | for (m = 0; m < 3; m++) { |
678 | v = get_bits(&s->gb, bits); |
679 | v = l1_unscale(bits - 1, v, scale); |
680 | s->sb_samples[ch][k * 12 + l + m][i] = v; |
681 | } |
682 | } |
683 | } else { |
684 | s->sb_samples[ch][k * 12 + l + 0][i] = 0; |
685 | s->sb_samples[ch][k * 12 + l + 1][i] = 0; |
686 | s->sb_samples[ch][k * 12 + l + 2][i] = 0; |
687 | } |
688 | } |
689 | /* next subband in alloc table */ |
690 | j += 1 << bit_alloc_bits; |
691 | } |
692 | /* XXX: find a way to avoid this duplication of code */ |
693 | for (i = bound; i < sblimit; i++) { |
694 | bit_alloc_bits = alloc_table[j]; |
695 | b = bit_alloc[0][i]; |
696 | if (b) { |
697 | int mant, scale0, scale1; |
698 | scale0 = scale_factors[0][i][k]; |
699 | scale1 = scale_factors[1][i][k]; |
700 | qindex = alloc_table[j+b]; |
701 | bits = ff_mpa_quant_bits[qindex]; |
702 | if (bits < 0) { |
703 | /* 3 values at the same time */ |
704 | v = get_bits(&s->gb, -bits); |
705 | steps = ff_mpa_quant_steps[qindex]; |
706 | mant = v % steps; |
707 | v = v / steps; |
708 | s->sb_samples[0][k * 12 + l + 0][i] = |
709 | l2_unscale_group(steps, mant, scale0); |
710 | s->sb_samples[1][k * 12 + l + 0][i] = |
711 | l2_unscale_group(steps, mant, scale1); |
712 | mant = v % steps; |
713 | v = v / steps; |
714 | s->sb_samples[0][k * 12 + l + 1][i] = |
715 | l2_unscale_group(steps, mant, scale0); |
716 | s->sb_samples[1][k * 12 + l + 1][i] = |
717 | l2_unscale_group(steps, mant, scale1); |
718 | s->sb_samples[0][k * 12 + l + 2][i] = |
719 | l2_unscale_group(steps, v, scale0); |
720 | s->sb_samples[1][k * 12 + l + 2][i] = |
721 | l2_unscale_group(steps, v, scale1); |
722 | } else { |
723 | for (m = 0; m < 3; m++) { |
724 | mant = get_bits(&s->gb, bits); |
725 | s->sb_samples[0][k * 12 + l + m][i] = |
726 | l1_unscale(bits - 1, mant, scale0); |
727 | s->sb_samples[1][k * 12 + l + m][i] = |
728 | l1_unscale(bits - 1, mant, scale1); |
729 | } |
730 | } |
731 | } else { |
732 | s->sb_samples[0][k * 12 + l + 0][i] = 0; |
733 | s->sb_samples[0][k * 12 + l + 1][i] = 0; |
734 | s->sb_samples[0][k * 12 + l + 2][i] = 0; |
735 | s->sb_samples[1][k * 12 + l + 0][i] = 0; |
736 | s->sb_samples[1][k * 12 + l + 1][i] = 0; |
737 | s->sb_samples[1][k * 12 + l + 2][i] = 0; |
738 | } |
739 | /* next subband in alloc table */ |
740 | j += 1 << bit_alloc_bits; |
741 | } |
742 | /* fill remaining samples to zero */ |
743 | for (i = sblimit; i < SBLIMIT; i++) { |
744 | for (ch = 0; ch < s->nb_channels; ch++) { |
745 | s->sb_samples[ch][k * 12 + l + 0][i] = 0; |
746 | s->sb_samples[ch][k * 12 + l + 1][i] = 0; |
747 | s->sb_samples[ch][k * 12 + l + 2][i] = 0; |
748 | } |
749 | } |
750 | } |
751 | } |
752 | return 3 * 12; |
753 | } |
754 | |
755 | #define SPLIT(dst,sf,n) \ |
756 | if (n == 3) { \ |
757 | int m = (sf * 171) >> 9; \ |
758 | dst = sf - 3 * m; \ |
759 | sf = m; \ |
760 | } else if (n == 4) { \ |
761 | dst = sf & 3; \ |
762 | sf >>= 2; \ |
763 | } else if (n == 5) { \ |
764 | int m = (sf * 205) >> 10; \ |
765 | dst = sf - 5 * m; \ |
766 | sf = m; \ |
767 | } else if (n == 6) { \ |
768 | int m = (sf * 171) >> 10; \ |
769 | dst = sf - 6 * m; \ |
770 | sf = m; \ |
771 | } else { \ |
772 | dst = 0; \ |
773 | } |
774 | |
775 | static av_always_inline void lsf_sf_expand(int *slen, int sf, int n1, int n2, |
776 | int n3) |
777 | { |
778 | SPLIT(slen[3], sf, n3) |
779 | SPLIT(slen[2], sf, n2) |
780 | SPLIT(slen[1], sf, n1) |
781 | slen[0] = sf; |
782 | } |
783 | |
784 | static void exponents_from_scale_factors(MPADecodeContext *s, GranuleDef *g, |
785 | int16_t *exponents) |
786 | { |
787 | const uint8_t *bstab, *pretab; |
788 | int len, i, j, k, l, v0, shift, gain, gains[3]; |
789 | int16_t *exp_ptr; |
790 | |
791 | exp_ptr = exponents; |
792 | gain = g->global_gain - 210; |
793 | shift = g->scalefac_scale + 1; |
794 | |
795 | bstab = band_size_long[s->sample_rate_index]; |
796 | pretab = mpa_pretab[g->preflag]; |
797 | for (i = 0; i < g->long_end; i++) { |
798 | v0 = gain - ((g->scale_factors[i] + pretab[i]) << shift) + 400; |
799 | len = bstab[i]; |
800 | for (j = len; j > 0; j--) |
801 | *exp_ptr++ = v0; |
802 | } |
803 | |
804 | if (g->short_start < 13) { |
805 | bstab = band_size_short[s->sample_rate_index]; |
806 | gains[0] = gain - (g->subblock_gain[0] << 3); |
807 | gains[1] = gain - (g->subblock_gain[1] << 3); |
808 | gains[2] = gain - (g->subblock_gain[2] << 3); |
809 | k = g->long_end; |
810 | for (i = g->short_start; i < 13; i++) { |
811 | len = bstab[i]; |
812 | for (l = 0; l < 3; l++) { |
813 | v0 = gains[l] - (g->scale_factors[k++] << shift) + 400; |
814 | for (j = len; j > 0; j--) |
815 | *exp_ptr++ = v0; |
816 | } |
817 | } |
818 | } |
819 | } |
820 | |
821 | static void switch_buffer(MPADecodeContext *s, int *pos, int *end_pos, |
822 | int *end_pos2) |
823 | { |
824 | if (s->in_gb.buffer && *pos >= s->gb.size_in_bits - s->extrasize * 8) { |
825 | s->gb = s->in_gb; |
826 | s->in_gb.buffer = NULL; |
827 | s->extrasize = 0; |
828 | av_assert2((get_bits_count(&s->gb) & 7) == 0); |
829 | skip_bits_long(&s->gb, *pos - *end_pos); |
830 | *end_pos2 = |
831 | *end_pos = *end_pos2 + get_bits_count(&s->gb) - *pos; |
832 | *pos = get_bits_count(&s->gb); |
833 | } |
834 | } |
835 | |
836 | /* Following is an optimized code for |
837 | INTFLOAT v = *src |
838 | if(get_bits1(&s->gb)) |
839 | v = -v; |
840 | *dst = v; |
841 | */ |
842 | #if USE_FLOATS |
843 | #define READ_FLIP_SIGN(dst,src) \ |
844 | v = AV_RN32A(src) ^ (get_bits1(&s->gb) << 31); \ |
845 | AV_WN32A(dst, v); |
846 | #else |
847 | #define READ_FLIP_SIGN(dst,src) \ |
848 | v = -get_bits1(&s->gb); \ |
849 | *(dst) = (*(src) ^ v) - v; |
850 | #endif |
851 | |
852 | static int huffman_decode(MPADecodeContext *s, GranuleDef *g, |
853 | int16_t *exponents, int end_pos2) |
854 | { |
855 | int s_index; |
856 | int i; |
857 | int last_pos, bits_left; |
858 | VLC *vlc; |
859 | int end_pos = FFMIN(end_pos2, s->gb.size_in_bits - s->extrasize * 8); |
860 | |
861 | /* low frequencies (called big values) */ |
862 | s_index = 0; |
863 | for (i = 0; i < 3; i++) { |
864 | int j, k, l, linbits; |
865 | j = g->region_size[i]; |
866 | if (j == 0) |
867 | continue; |
868 | /* select vlc table */ |
869 | k = g->table_select[i]; |
870 | l = mpa_huff_data[k][0]; |
871 | linbits = mpa_huff_data[k][1]; |
872 | vlc = &huff_vlc[l]; |
873 | |
874 | if (!l) { |
875 | memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * 2 * j); |
876 | s_index += 2 * j; |
877 | continue; |
878 | } |
879 | |
880 | /* read huffcode and compute each couple */ |
881 | for (; j > 0; j--) { |
882 | int exponent, x, y; |
883 | int v; |
884 | int pos = get_bits_count(&s->gb); |
885 | |
886 | if (pos >= end_pos){ |
887 | switch_buffer(s, &pos, &end_pos, &end_pos2); |
888 | if (pos >= end_pos) |
889 | break; |
890 | } |
891 | y = get_vlc2(&s->gb, vlc->table, 7, 3); |
892 | |
893 | if (!y) { |
894 | g->sb_hybrid[s_index ] = |
895 | g->sb_hybrid[s_index+1] = 0; |
896 | s_index += 2; |
897 | continue; |
898 | } |
899 | |
900 | exponent= exponents[s_index]; |
901 | |
902 | ff_dlog(s->avctx, "region=%d n=%d y=%d exp=%d\n", |
903 | i, g->region_size[i] - j, y, exponent); |
904 | if (y & 16) { |
905 | x = y >> 5; |
906 | y = y & 0x0f; |
907 | if (x < 15) { |
908 | READ_FLIP_SIGN(g->sb_hybrid + s_index, RENAME(expval_table)[exponent] + x) |
909 | } else { |
910 | x += get_bitsz(&s->gb, linbits); |
911 | v = l3_unscale(x, exponent); |
912 | if (get_bits1(&s->gb)) |
913 | v = -v; |
914 | g->sb_hybrid[s_index] = v; |
915 | } |
916 | if (y < 15) { |
917 | READ_FLIP_SIGN(g->sb_hybrid + s_index + 1, RENAME(expval_table)[exponent] + y) |
918 | } else { |
919 | y += get_bitsz(&s->gb, linbits); |
920 | v = l3_unscale(y, exponent); |
921 | if (get_bits1(&s->gb)) |
922 | v = -v; |
923 | g->sb_hybrid[s_index+1] = v; |
924 | } |
925 | } else { |
926 | x = y >> 5; |
927 | y = y & 0x0f; |
928 | x += y; |
929 | if (x < 15) { |
930 | READ_FLIP_SIGN(g->sb_hybrid + s_index + !!y, RENAME(expval_table)[exponent] + x) |
931 | } else { |
932 | x += get_bitsz(&s->gb, linbits); |
933 | v = l3_unscale(x, exponent); |
934 | if (get_bits1(&s->gb)) |
935 | v = -v; |
936 | g->sb_hybrid[s_index+!!y] = v; |
937 | } |
938 | g->sb_hybrid[s_index + !y] = 0; |
939 | } |
940 | s_index += 2; |
941 | } |
942 | } |
943 | |
944 | /* high frequencies */ |
945 | vlc = &huff_quad_vlc[g->count1table_select]; |
946 | last_pos = 0; |
947 | while (s_index <= 572) { |
948 | int pos, code; |
949 | pos = get_bits_count(&s->gb); |
950 | if (pos >= end_pos) { |
951 | if (pos > end_pos2 && last_pos) { |
952 | /* some encoders generate an incorrect size for this |
953 | part. We must go back into the data */ |
954 | s_index -= 4; |
955 | skip_bits_long(&s->gb, last_pos - pos); |
956 | av_log(s->avctx, AV_LOG_INFO, "overread, skip %d enddists: %d %d\n", last_pos - pos, end_pos-pos, end_pos2-pos); |
957 | if(s->err_recognition & (AV_EF_BITSTREAM|AV_EF_COMPLIANT)) |
958 | s_index=0; |
959 | break; |
960 | } |
961 | switch_buffer(s, &pos, &end_pos, &end_pos2); |
962 | if (pos >= end_pos) |
963 | break; |
964 | } |
965 | last_pos = pos; |
966 | |
967 | code = get_vlc2(&s->gb, vlc->table, vlc->bits, 1); |
968 | ff_dlog(s->avctx, "t=%d code=%d\n", g->count1table_select, code); |
969 | g->sb_hybrid[s_index+0] = |
970 | g->sb_hybrid[s_index+1] = |
971 | g->sb_hybrid[s_index+2] = |
972 | g->sb_hybrid[s_index+3] = 0; |
973 | while (code) { |
974 | static const int idxtab[16] = { 3,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0 }; |
975 | int v; |
976 | int pos = s_index + idxtab[code]; |
977 | code ^= 8 >> idxtab[code]; |
978 | READ_FLIP_SIGN(g->sb_hybrid + pos, RENAME(exp_table)+exponents[pos]) |
979 | } |
980 | s_index += 4; |
981 | } |
982 | /* skip extension bits */ |
983 | bits_left = end_pos2 - get_bits_count(&s->gb); |
984 | if (bits_left < 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_COMPLIANT))) { |
985 | av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left); |
986 | s_index=0; |
987 | } else if (bits_left > 0 && (s->err_recognition & (AV_EF_BUFFER|AV_EF_AGGRESSIVE))) { |
988 | av_log(s->avctx, AV_LOG_ERROR, "bits_left=%d\n", bits_left); |
989 | s_index = 0; |
990 | } |
991 | memset(&g->sb_hybrid[s_index], 0, sizeof(*g->sb_hybrid) * (576 - s_index)); |
992 | skip_bits_long(&s->gb, bits_left); |
993 | |
994 | i = get_bits_count(&s->gb); |
995 | switch_buffer(s, &i, &end_pos, &end_pos2); |
996 | |
997 | return 0; |
998 | } |
999 | |
1000 | /* Reorder short blocks from bitstream order to interleaved order. It |
1001 | would be faster to do it in parsing, but the code would be far more |
1002 | complicated */ |
1003 | static void reorder_block(MPADecodeContext *s, GranuleDef *g) |
1004 | { |
1005 | int i, j, len; |
1006 | INTFLOAT *ptr, *dst, *ptr1; |
1007 | INTFLOAT tmp[576]; |
1008 | |
1009 | if (g->block_type != 2) |
1010 | return; |
1011 | |
1012 | if (g->switch_point) { |
1013 | if (s->sample_rate_index != 8) |
1014 | ptr = g->sb_hybrid + 36; |
1015 | else |
1016 | ptr = g->sb_hybrid + 72; |
1017 | } else { |
1018 | ptr = g->sb_hybrid; |
1019 | } |
1020 | |
1021 | for (i = g->short_start; i < 13; i++) { |
1022 | len = band_size_short[s->sample_rate_index][i]; |
1023 | ptr1 = ptr; |
1024 | dst = tmp; |
1025 | for (j = len; j > 0; j--) { |
1026 | *dst++ = ptr[0*len]; |
1027 | *dst++ = ptr[1*len]; |
1028 | *dst++ = ptr[2*len]; |
1029 | ptr++; |
1030 | } |
1031 | ptr += 2 * len; |
1032 | memcpy(ptr1, tmp, len * 3 * sizeof(*ptr1)); |
1033 | } |
1034 | } |
1035 | |
1036 | #define ISQRT2 FIXR(0.70710678118654752440) |
1037 | |
1038 | static void compute_stereo(MPADecodeContext *s, GranuleDef *g0, GranuleDef *g1) |
1039 | { |
1040 | int i, j, k, l; |
1041 | int sf_max, sf, len, non_zero_found; |
1042 | INTFLOAT (*is_tab)[16], *tab0, *tab1, v1, v2; |
1043 | SUINTFLOAT tmp0, tmp1; |
1044 | int non_zero_found_short[3]; |
1045 | |
1046 | /* intensity stereo */ |
1047 | if (s->mode_ext & MODE_EXT_I_STEREO) { |
1048 | if (!s->lsf) { |
1049 | is_tab = is_table; |
1050 | sf_max = 7; |
1051 | } else { |
1052 | is_tab = is_table_lsf[g1->scalefac_compress & 1]; |
1053 | sf_max = 16; |
1054 | } |
1055 | |
1056 | tab0 = g0->sb_hybrid + 576; |
1057 | tab1 = g1->sb_hybrid + 576; |
1058 | |
1059 | non_zero_found_short[0] = 0; |
1060 | non_zero_found_short[1] = 0; |
1061 | non_zero_found_short[2] = 0; |
1062 | k = (13 - g1->short_start) * 3 + g1->long_end - 3; |
1063 | for (i = 12; i >= g1->short_start; i--) { |
1064 | /* for last band, use previous scale factor */ |
1065 | if (i != 11) |
1066 | k -= 3; |
1067 | len = band_size_short[s->sample_rate_index][i]; |
1068 | for (l = 2; l >= 0; l--) { |
1069 | tab0 -= len; |
1070 | tab1 -= len; |
1071 | if (!non_zero_found_short[l]) { |
1072 | /* test if non zero band. if so, stop doing i-stereo */ |
1073 | for (j = 0; j < len; j++) { |
1074 | if (tab1[j] != 0) { |
1075 | non_zero_found_short[l] = 1; |
1076 | goto found1; |
1077 | } |
1078 | } |
1079 | sf = g1->scale_factors[k + l]; |
1080 | if (sf >= sf_max) |
1081 | goto found1; |
1082 | |
1083 | v1 = is_tab[0][sf]; |
1084 | v2 = is_tab[1][sf]; |
1085 | for (j = 0; j < len; j++) { |
1086 | tmp0 = tab0[j]; |
1087 | tab0[j] = MULLx(tmp0, v1, FRAC_BITS); |
1088 | tab1[j] = MULLx(tmp0, v2, FRAC_BITS); |
1089 | } |
1090 | } else { |
1091 | found1: |
1092 | if (s->mode_ext & MODE_EXT_MS_STEREO) { |
1093 | /* lower part of the spectrum : do ms stereo |
1094 | if enabled */ |
1095 | for (j = 0; j < len; j++) { |
1096 | tmp0 = tab0[j]; |
1097 | tmp1 = tab1[j]; |
1098 | tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS); |
1099 | tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS); |
1100 | } |
1101 | } |
1102 | } |
1103 | } |
1104 | } |
1105 | |
1106 | non_zero_found = non_zero_found_short[0] | |
1107 | non_zero_found_short[1] | |
1108 | non_zero_found_short[2]; |
1109 | |
1110 | for (i = g1->long_end - 1;i >= 0;i--) { |
1111 | len = band_size_long[s->sample_rate_index][i]; |
1112 | tab0 -= len; |
1113 | tab1 -= len; |
1114 | /* test if non zero band. if so, stop doing i-stereo */ |
1115 | if (!non_zero_found) { |
1116 | for (j = 0; j < len; j++) { |
1117 | if (tab1[j] != 0) { |
1118 | non_zero_found = 1; |
1119 | goto found2; |
1120 | } |
1121 | } |
1122 | /* for last band, use previous scale factor */ |
1123 | k = (i == 21) ? 20 : i; |
1124 | sf = g1->scale_factors[k]; |
1125 | if (sf >= sf_max) |
1126 | goto found2; |
1127 | v1 = is_tab[0][sf]; |
1128 | v2 = is_tab[1][sf]; |
1129 | for (j = 0; j < len; j++) { |
1130 | tmp0 = tab0[j]; |
1131 | tab0[j] = MULLx(tmp0, v1, FRAC_BITS); |
1132 | tab1[j] = MULLx(tmp0, v2, FRAC_BITS); |
1133 | } |
1134 | } else { |
1135 | found2: |
1136 | if (s->mode_ext & MODE_EXT_MS_STEREO) { |
1137 | /* lower part of the spectrum : do ms stereo |
1138 | if enabled */ |
1139 | for (j = 0; j < len; j++) { |
1140 | tmp0 = tab0[j]; |
1141 | tmp1 = tab1[j]; |
1142 | tab0[j] = MULLx(tmp0 + tmp1, ISQRT2, FRAC_BITS); |
1143 | tab1[j] = MULLx(tmp0 - tmp1, ISQRT2, FRAC_BITS); |
1144 | } |
1145 | } |
1146 | } |
1147 | } |
1148 | } else if (s->mode_ext & MODE_EXT_MS_STEREO) { |
1149 | /* ms stereo ONLY */ |
1150 | /* NOTE: the 1/sqrt(2) normalization factor is included in the |
1151 | global gain */ |
1152 | #if USE_FLOATS |
1153 | s->fdsp->butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576); |
1154 | #else |
1155 | tab0 = g0->sb_hybrid; |
1156 | tab1 = g1->sb_hybrid; |
1157 | for (i = 0; i < 576; i++) { |
1158 | tmp0 = tab0[i]; |
1159 | tmp1 = tab1[i]; |
1160 | tab0[i] = tmp0 + tmp1; |
1161 | tab1[i] = tmp0 - tmp1; |
1162 | } |
1163 | #endif |
1164 | } |
1165 | } |
1166 | |
1167 | #if USE_FLOATS |
1168 | #if HAVE_MIPSFPU |
1169 | # include "mips/compute_antialias_float.h" |
1170 | #endif /* HAVE_MIPSFPU */ |
1171 | #else |
1172 | #if HAVE_MIPSDSP |
1173 | # include "mips/compute_antialias_fixed.h" |
1174 | #endif /* HAVE_MIPSDSP */ |
1175 | #endif /* USE_FLOATS */ |
1176 | |
1177 | #ifndef compute_antialias |
1178 | #if USE_FLOATS |
1179 | #define AA(j) do { \ |
1180 | float tmp0 = ptr[-1-j]; \ |
1181 | float tmp1 = ptr[ j]; \ |
1182 | ptr[-1-j] = tmp0 * csa_table[j][0] - tmp1 * csa_table[j][1]; \ |
1183 | ptr[ j] = tmp0 * csa_table[j][1] + tmp1 * csa_table[j][0]; \ |
1184 | } while (0) |
1185 | #else |
1186 | #define AA(j) do { \ |
1187 | SUINT tmp0 = ptr[-1-j]; \ |
1188 | SUINT tmp1 = ptr[ j]; \ |
1189 | SUINT tmp2 = MULH(tmp0 + tmp1, csa_table[j][0]); \ |
1190 | ptr[-1-j] = 4 * (tmp2 - MULH(tmp1, csa_table[j][2])); \ |
1191 | ptr[ j] = 4 * (tmp2 + MULH(tmp0, csa_table[j][3])); \ |
1192 | } while (0) |
1193 | #endif |
1194 | |
1195 | static void compute_antialias(MPADecodeContext *s, GranuleDef *g) |
1196 | { |
1197 | INTFLOAT *ptr; |
1198 | int n, i; |
1199 | |
1200 | /* we antialias only "long" bands */ |
1201 | if (g->block_type == 2) { |
1202 | if (!g->switch_point) |
1203 | return; |
1204 | /* XXX: check this for 8000Hz case */ |
1205 | n = 1; |
1206 | } else { |
1207 | n = SBLIMIT - 1; |
1208 | } |
1209 | |
1210 | ptr = g->sb_hybrid + 18; |
1211 | for (i = n; i > 0; i--) { |
1212 | AA(0); |
1213 | AA(1); |
1214 | AA(2); |
1215 | AA(3); |
1216 | AA(4); |
1217 | AA(5); |
1218 | AA(6); |
1219 | AA(7); |
1220 | |
1221 | ptr += 18; |
1222 | } |
1223 | } |
1224 | #endif /* compute_antialias */ |
1225 | |
1226 | static void compute_imdct(MPADecodeContext *s, GranuleDef *g, |
1227 | INTFLOAT *sb_samples, INTFLOAT *mdct_buf) |
1228 | { |
1229 | INTFLOAT *win, *out_ptr, *ptr, *buf, *ptr1; |
1230 | INTFLOAT out2[12]; |
1231 | int i, j, mdct_long_end, sblimit; |
1232 | |
1233 | /* find last non zero block */ |
1234 | ptr = g->sb_hybrid + 576; |
1235 | ptr1 = g->sb_hybrid + 2 * 18; |
1236 | while (ptr >= ptr1) { |
1237 | int32_t *p; |
1238 | ptr -= 6; |
1239 | p = (int32_t*)ptr; |
1240 | if (p[0] | p[1] | p[2] | p[3] | p[4] | p[5]) |
1241 | break; |
1242 | } |
1243 | sblimit = ((ptr - g->sb_hybrid) / 18) + 1; |
1244 | |
1245 | if (g->block_type == 2) { |
1246 | /* XXX: check for 8000 Hz */ |
1247 | if (g->switch_point) |
1248 | mdct_long_end = 2; |
1249 | else |
1250 | mdct_long_end = 0; |
1251 | } else { |
1252 | mdct_long_end = sblimit; |
1253 | } |
1254 | |
1255 | s->mpadsp.RENAME(imdct36_blocks)(sb_samples, mdct_buf, g->sb_hybrid, |
1256 | mdct_long_end, g->switch_point, |
1257 | g->block_type); |
1258 | |
1259 | buf = mdct_buf + 4*18*(mdct_long_end >> 2) + (mdct_long_end & 3); |
1260 | ptr = g->sb_hybrid + 18 * mdct_long_end; |
1261 | |
1262 | for (j = mdct_long_end; j < sblimit; j++) { |
1263 | /* select frequency inversion */ |
1264 | win = RENAME(ff_mdct_win)[2 + (4 & -(j & 1))]; |
1265 | out_ptr = sb_samples + j; |
1266 | |
1267 | for (i = 0; i < 6; i++) { |
1268 | *out_ptr = buf[4*i]; |
1269 | out_ptr += SBLIMIT; |
1270 | } |
1271 | imdct12(out2, ptr + 0); |
1272 | for (i = 0; i < 6; i++) { |
1273 | *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*1)]; |
1274 | buf[4*(i + 6*2)] = MULH3(out2[i + 6], win[i + 6], 1); |
1275 | out_ptr += SBLIMIT; |
1276 | } |
1277 | imdct12(out2, ptr + 1); |
1278 | for (i = 0; i < 6; i++) { |
1279 | *out_ptr = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*2)]; |
1280 | buf[4*(i + 6*0)] = MULH3(out2[i + 6], win[i + 6], 1); |
1281 | out_ptr += SBLIMIT; |
1282 | } |
1283 | imdct12(out2, ptr + 2); |
1284 | for (i = 0; i < 6; i++) { |
1285 | buf[4*(i + 6*0)] = MULH3(out2[i ], win[i ], 1) + buf[4*(i + 6*0)]; |
1286 | buf[4*(i + 6*1)] = MULH3(out2[i + 6], win[i + 6], 1); |
1287 | buf[4*(i + 6*2)] = 0; |
1288 | } |
1289 | ptr += 18; |
1290 | buf += (j&3) != 3 ? 1 : (4*18-3); |
1291 | } |
1292 | /* zero bands */ |
1293 | for (j = sblimit; j < SBLIMIT; j++) { |
1294 | /* overlap */ |
1295 | out_ptr = sb_samples + j; |
1296 | for (i = 0; i < 18; i++) { |
1297 | *out_ptr = buf[4*i]; |
1298 | buf[4*i] = 0; |
1299 | out_ptr += SBLIMIT; |
1300 | } |
1301 | buf += (j&3) != 3 ? 1 : (4*18-3); |
1302 | } |
1303 | } |
1304 | |
1305 | /* main layer3 decoding function */ |
1306 | static int mp_decode_layer3(MPADecodeContext *s) |
1307 | { |
1308 | int nb_granules, main_data_begin; |
1309 | int gr, ch, blocksplit_flag, i, j, k, n, bits_pos; |
1310 | GranuleDef *g; |
1311 | int16_t exponents[576]; //FIXME try INTFLOAT |
1312 | |
1313 | /* read side info */ |
1314 | if (s->lsf) { |
1315 | main_data_begin = get_bits(&s->gb, 8); |
1316 | skip_bits(&s->gb, s->nb_channels); |
1317 | nb_granules = 1; |
1318 | } else { |
1319 | main_data_begin = get_bits(&s->gb, 9); |
1320 | if (s->nb_channels == 2) |
1321 | skip_bits(&s->gb, 3); |
1322 | else |
1323 | skip_bits(&s->gb, 5); |
1324 | nb_granules = 2; |
1325 | for (ch = 0; ch < s->nb_channels; ch++) { |
1326 | s->granules[ch][0].scfsi = 0;/* all scale factors are transmitted */ |
1327 | s->granules[ch][1].scfsi = get_bits(&s->gb, 4); |
1328 | } |
1329 | } |
1330 | |
1331 | for (gr = 0; gr < nb_granules; gr++) { |
1332 | for (ch = 0; ch < s->nb_channels; ch++) { |
1333 | ff_dlog(s->avctx, "gr=%d ch=%d: side_info\n", gr, ch); |
1334 | g = &s->granules[ch][gr]; |
1335 | g->part2_3_length = get_bits(&s->gb, 12); |
1336 | g->big_values = get_bits(&s->gb, 9); |
1337 | if (g->big_values > 288) { |
1338 | av_log(s->avctx, AV_LOG_ERROR, "big_values too big\n"); |
1339 | return AVERROR_INVALIDDATA; |
1340 | } |
1341 | |
1342 | g->global_gain = get_bits(&s->gb, 8); |
1343 | /* if MS stereo only is selected, we precompute the |
1344 | 1/sqrt(2) renormalization factor */ |
1345 | if ((s->mode_ext & (MODE_EXT_MS_STEREO | MODE_EXT_I_STEREO)) == |
1346 | MODE_EXT_MS_STEREO) |
1347 | g->global_gain -= 2; |
1348 | if (s->lsf) |
1349 | g->scalefac_compress = get_bits(&s->gb, 9); |
1350 | else |
1351 | g->scalefac_compress = get_bits(&s->gb, 4); |
1352 | blocksplit_flag = get_bits1(&s->gb); |
1353 | if (blocksplit_flag) { |
1354 | g->block_type = get_bits(&s->gb, 2); |
1355 | if (g->block_type == 0) { |
1356 | av_log(s->avctx, AV_LOG_ERROR, "invalid block type\n"); |
1357 | return AVERROR_INVALIDDATA; |
1358 | } |
1359 | g->switch_point = get_bits1(&s->gb); |
1360 | for (i = 0; i < 2; i++) |
1361 | g->table_select[i] = get_bits(&s->gb, 5); |
1362 | for (i = 0; i < 3; i++) |
1363 | g->subblock_gain[i] = get_bits(&s->gb, 3); |
1364 | init_short_region(s, g); |
1365 | } else { |
1366 | int region_address1, region_address2; |
1367 | g->block_type = 0; |
1368 | g->switch_point = 0; |
1369 | for (i = 0; i < 3; i++) |
1370 | g->table_select[i] = get_bits(&s->gb, 5); |
1371 | /* compute huffman coded region sizes */ |
1372 | region_address1 = get_bits(&s->gb, 4); |
1373 | region_address2 = get_bits(&s->gb, 3); |
1374 | ff_dlog(s->avctx, "region1=%d region2=%d\n", |
1375 | region_address1, region_address2); |
1376 | init_long_region(s, g, region_address1, region_address2); |
1377 | } |
1378 | region_offset2size(g); |
1379 | compute_band_indexes(s, g); |
1380 | |
1381 | g->preflag = 0; |
1382 | if (!s->lsf) |
1383 | g->preflag = get_bits1(&s->gb); |
1384 | g->scalefac_scale = get_bits1(&s->gb); |
1385 | g->count1table_select = get_bits1(&s->gb); |
1386 | ff_dlog(s->avctx, "block_type=%d switch_point=%d\n", |
1387 | g->block_type, g->switch_point); |
1388 | } |
1389 | } |
1390 | |
1391 | if (!s->adu_mode) { |
1392 | int skip; |
1393 | const uint8_t *ptr = s->gb.buffer + (get_bits_count(&s->gb)>>3); |
1394 | s->extrasize = av_clip((get_bits_left(&s->gb) >> 3) - s->extrasize, 0, |
1395 | FFMAX(0, LAST_BUF_SIZE - s->last_buf_size)); |
1396 | av_assert1((get_bits_count(&s->gb) & 7) == 0); |
1397 | /* now we get bits from the main_data_begin offset */ |
1398 | ff_dlog(s->avctx, "seekback:%d, lastbuf:%d\n", |
1399 | main_data_begin, s->last_buf_size); |
1400 | |
1401 | memcpy(s->last_buf + s->last_buf_size, ptr, s->extrasize); |
1402 | s->in_gb = s->gb; |
1403 | init_get_bits(&s->gb, s->last_buf, (s->last_buf_size + s->extrasize) * 8); |
1404 | s->last_buf_size <<= 3; |
1405 | for (gr = 0; gr < nb_granules && (s->last_buf_size >> 3) < main_data_begin; gr++) { |
1406 | for (ch = 0; ch < s->nb_channels; ch++) { |
1407 | g = &s->granules[ch][gr]; |
1408 | s->last_buf_size += g->part2_3_length; |
1409 | memset(g->sb_hybrid, 0, sizeof(g->sb_hybrid)); |
1410 | compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]); |
1411 | } |
1412 | } |
1413 | skip = s->last_buf_size - 8 * main_data_begin; |
1414 | if (skip >= s->gb.size_in_bits - s->extrasize * 8 && s->in_gb.buffer) { |
1415 | skip_bits_long(&s->in_gb, skip - s->gb.size_in_bits + s->extrasize * 8); |
1416 | s->gb = s->in_gb; |
1417 | s->in_gb.buffer = NULL; |
1418 | s->extrasize = 0; |
1419 | } else { |
1420 | skip_bits_long(&s->gb, skip); |
1421 | } |
1422 | } else { |
1423 | gr = 0; |
1424 | s->extrasize = 0; |
1425 | } |
1426 | |
1427 | for (; gr < nb_granules; gr++) { |
1428 | for (ch = 0; ch < s->nb_channels; ch++) { |
1429 | g = &s->granules[ch][gr]; |
1430 | bits_pos = get_bits_count(&s->gb); |
1431 | |
1432 | if (!s->lsf) { |
1433 | uint8_t *sc; |
1434 | int slen, slen1, slen2; |
1435 | |
1436 | /* MPEG-1 scale factors */ |
1437 | slen1 = slen_table[0][g->scalefac_compress]; |
1438 | slen2 = slen_table[1][g->scalefac_compress]; |
1439 | ff_dlog(s->avctx, "slen1=%d slen2=%d\n", slen1, slen2); |
1440 | if (g->block_type == 2) { |
1441 | n = g->switch_point ? 17 : 18; |
1442 | j = 0; |
1443 | if (slen1) { |
1444 | for (i = 0; i < n; i++) |
1445 | g->scale_factors[j++] = get_bits(&s->gb, slen1); |
1446 | } else { |
1447 | for (i = 0; i < n; i++) |
1448 | g->scale_factors[j++] = 0; |
1449 | } |
1450 | if (slen2) { |
1451 | for (i = 0; i < 18; i++) |
1452 | g->scale_factors[j++] = get_bits(&s->gb, slen2); |
1453 | for (i = 0; i < 3; i++) |
1454 | g->scale_factors[j++] = 0; |
1455 | } else { |
1456 | for (i = 0; i < 21; i++) |
1457 | g->scale_factors[j++] = 0; |
1458 | } |
1459 | } else { |
1460 | sc = s->granules[ch][0].scale_factors; |
1461 | j = 0; |
1462 | for (k = 0; k < 4; k++) { |
1463 | n = k == 0 ? 6 : 5; |
1464 | if ((g->scfsi & (0x8 >> k)) == 0) { |
1465 | slen = (k < 2) ? slen1 : slen2; |
1466 | if (slen) { |
1467 | for (i = 0; i < n; i++) |
1468 | g->scale_factors[j++] = get_bits(&s->gb, slen); |
1469 | } else { |
1470 | for (i = 0; i < n; i++) |
1471 | g->scale_factors[j++] = 0; |
1472 | } |
1473 | } else { |
1474 | /* simply copy from last granule */ |
1475 | for (i = 0; i < n; i++) { |
1476 | g->scale_factors[j] = sc[j]; |
1477 | j++; |
1478 | } |
1479 | } |
1480 | } |
1481 | g->scale_factors[j++] = 0; |
1482 | } |
1483 | } else { |
1484 | int tindex, tindex2, slen[4], sl, sf; |
1485 | |
1486 | /* LSF scale factors */ |
1487 | if (g->block_type == 2) |
1488 | tindex = g->switch_point ? 2 : 1; |
1489 | else |
1490 | tindex = 0; |
1491 | |
1492 | sf = g->scalefac_compress; |
1493 | if ((s->mode_ext & MODE_EXT_I_STEREO) && ch == 1) { |
1494 | /* intensity stereo case */ |
1495 | sf >>= 1; |
1496 | if (sf < 180) { |
1497 | lsf_sf_expand(slen, sf, 6, 6, 0); |
1498 | tindex2 = 3; |
1499 | } else if (sf < 244) { |
1500 | lsf_sf_expand(slen, sf - 180, 4, 4, 0); |
1501 | tindex2 = 4; |
1502 | } else { |
1503 | lsf_sf_expand(slen, sf - 244, 3, 0, 0); |
1504 | tindex2 = 5; |
1505 | } |
1506 | } else { |
1507 | /* normal case */ |
1508 | if (sf < 400) { |
1509 | lsf_sf_expand(slen, sf, 5, 4, 4); |
1510 | tindex2 = 0; |
1511 | } else if (sf < 500) { |
1512 | lsf_sf_expand(slen, sf - 400, 5, 4, 0); |
1513 | tindex2 = 1; |
1514 | } else { |
1515 | lsf_sf_expand(slen, sf - 500, 3, 0, 0); |
1516 | tindex2 = 2; |
1517 | g->preflag = 1; |
1518 | } |
1519 | } |
1520 | |
1521 | j = 0; |
1522 | for (k = 0; k < 4; k++) { |
1523 | n = lsf_nsf_table[tindex2][tindex][k]; |
1524 | sl = slen[k]; |
1525 | if (sl) { |
1526 | for (i = 0; i < n; i++) |
1527 | g->scale_factors[j++] = get_bits(&s->gb, sl); |
1528 | } else { |
1529 | for (i = 0; i < n; i++) |
1530 | g->scale_factors[j++] = 0; |
1531 | } |
1532 | } |
1533 | /* XXX: should compute exact size */ |
1534 | for (; j < 40; j++) |
1535 | g->scale_factors[j] = 0; |
1536 | } |
1537 | |
1538 | exponents_from_scale_factors(s, g, exponents); |
1539 | |
1540 | /* read Huffman coded residue */ |
1541 | huffman_decode(s, g, exponents, bits_pos + g->part2_3_length); |
1542 | } /* ch */ |
1543 | |
1544 | if (s->mode == MPA_JSTEREO) |
1545 | compute_stereo(s, &s->granules[0][gr], &s->granules[1][gr]); |
1546 | |
1547 | for (ch = 0; ch < s->nb_channels; ch++) { |
1548 | g = &s->granules[ch][gr]; |
1549 | |
1550 | reorder_block(s, g); |
1551 | compute_antialias(s, g); |
1552 | compute_imdct(s, g, &s->sb_samples[ch][18 * gr][0], s->mdct_buf[ch]); |
1553 | } |
1554 | } /* gr */ |
1555 | if (get_bits_count(&s->gb) < 0) |
1556 | skip_bits_long(&s->gb, -get_bits_count(&s->gb)); |
1557 | return nb_granules * 18; |
1558 | } |
1559 | |
1560 | static int mp_decode_frame(MPADecodeContext *s, OUT_INT **samples, |
1561 | const uint8_t *buf, int buf_size) |
1562 | { |
1563 | int i, nb_frames, ch, ret; |
1564 | OUT_INT *samples_ptr; |
1565 | |
1566 | init_get_bits(&s->gb, buf + HEADER_SIZE, (buf_size - HEADER_SIZE) * 8); |
1567 | |
1568 | /* skip error protection field */ |
1569 | if (s->error_protection) |
1570 | skip_bits(&s->gb, 16); |
1571 | |
1572 | switch(s->layer) { |
1573 | case 1: |
1574 | s->avctx->frame_size = 384; |
1575 | nb_frames = mp_decode_layer1(s); |
1576 | break; |
1577 | case 2: |
1578 | s->avctx->frame_size = 1152; |
1579 | nb_frames = mp_decode_layer2(s); |
1580 | break; |
1581 | case 3: |
1582 | s->avctx->frame_size = s->lsf ? 576 : 1152; |
1583 | default: |
1584 | nb_frames = mp_decode_layer3(s); |
1585 | |
1586 | s->last_buf_size=0; |
1587 | if (s->in_gb.buffer) { |
1588 | align_get_bits(&s->gb); |
1589 | i = (get_bits_left(&s->gb) >> 3) - s->extrasize; |
1590 | if (i >= 0 && i <= BACKSTEP_SIZE) { |
1591 | memmove(s->last_buf, s->gb.buffer + (get_bits_count(&s->gb)>>3), i); |
1592 | s->last_buf_size=i; |
1593 | } else |
1594 | av_log(s->avctx, AV_LOG_ERROR, "invalid old backstep %d\n", i); |
1595 | s->gb = s->in_gb; |
1596 | s->in_gb.buffer = NULL; |
1597 | s->extrasize = 0; |
1598 | } |
1599 | |
1600 | align_get_bits(&s->gb); |
1601 | av_assert1((get_bits_count(&s->gb) & 7) == 0); |
1602 | i = (get_bits_left(&s->gb) >> 3) - s->extrasize; |
1603 | if (i < 0 || i > BACKSTEP_SIZE || nb_frames < 0) { |
1604 | if (i < 0) |
1605 | av_log(s->avctx, AV_LOG_ERROR, "invalid new backstep %d\n", i); |
1606 | i = FFMIN(BACKSTEP_SIZE, buf_size - HEADER_SIZE); |
1607 | } |
1608 | av_assert1(i <= buf_size - HEADER_SIZE && i >= 0); |
1609 | memcpy(s->last_buf + s->last_buf_size, s->gb.buffer + buf_size - HEADER_SIZE - i, i); |
1610 | s->last_buf_size += i; |
1611 | } |
1612 | |
1613 | if(nb_frames < 0) |
1614 | return nb_frames; |
1615 | |
1616 | /* get output buffer */ |
1617 | if (!samples) { |
1618 | av_assert0(s->frame); |
1619 | s->frame->nb_samples = s->avctx->frame_size; |
1620 | if ((ret = ff_get_buffer(s->avctx, s->frame, 0)) < 0) |
1621 | return ret; |
1622 | samples = (OUT_INT **)s->frame->extended_data; |
1623 | } |
1624 | |
1625 | /* apply the synthesis filter */ |
1626 | for (ch = 0; ch < s->nb_channels; ch++) { |
1627 | int sample_stride; |
1628 | if (s->avctx->sample_fmt == OUT_FMT_P) { |
1629 | samples_ptr = samples[ch]; |
1630 | sample_stride = 1; |
1631 | } else { |
1632 | samples_ptr = samples[0] + ch; |
1633 | sample_stride = s->nb_channels; |
1634 | } |
1635 | for (i = 0; i < nb_frames; i++) { |
1636 | RENAME(ff_mpa_synth_filter)(&s->mpadsp, s->synth_buf[ch], |
1637 | &(s->synth_buf_offset[ch]), |
1638 | RENAME(ff_mpa_synth_window), |
1639 | &s->dither_state, samples_ptr, |
1640 | sample_stride, s->sb_samples[ch][i]); |
1641 | samples_ptr += 32 * sample_stride; |
1642 | } |
1643 | } |
1644 | |
1645 | return nb_frames * 32 * sizeof(OUT_INT) * s->nb_channels; |
1646 | } |
1647 | |
1648 | static int decode_frame(AVCodecContext * avctx, void *data, int *got_frame_ptr, |
1649 | AVPacket *avpkt) |
1650 | { |
1651 | const uint8_t *buf = avpkt->data; |
1652 | int buf_size = avpkt->size; |
1653 | MPADecodeContext *s = avctx->priv_data; |
1654 | uint32_t header; |
1655 | int ret; |
1656 | |
1657 | int skipped = 0; |
1658 | while(buf_size && !*buf){ |
1659 | buf++; |
1660 | buf_size--; |
1661 | skipped++; |
1662 | } |
1663 | |
1664 | if (buf_size < HEADER_SIZE) |
1665 | return AVERROR_INVALIDDATA; |
1666 | |
1667 | header = AV_RB32(buf); |
1668 | if (header>>8 == AV_RB32("TAG")>>8) { |
1669 | av_log(avctx, AV_LOG_DEBUG, "discarding ID3 tag\n"); |
1670 | return buf_size + skipped; |
1671 | } |
1672 | ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header); |
1673 | if (ret < 0) { |
1674 | av_log(avctx, AV_LOG_ERROR, "Header missing\n"); |
1675 | return AVERROR_INVALIDDATA; |
1676 | } else if (ret == 1) { |
1677 | /* free format: prepare to compute frame size */ |
1678 | s->frame_size = -1; |
1679 | return AVERROR_INVALIDDATA; |
1680 | } |
1681 | /* update codec info */ |
1682 | avctx->channels = s->nb_channels; |
1683 | avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO; |
1684 | if (!avctx->bit_rate) |
1685 | avctx->bit_rate = s->bit_rate; |
1686 | |
1687 | if (s->frame_size <= 0) { |
1688 | av_log(avctx, AV_LOG_ERROR, "incomplete frame\n"); |
1689 | return AVERROR_INVALIDDATA; |
1690 | } else if (s->frame_size < buf_size) { |
1691 | av_log(avctx, AV_LOG_DEBUG, "incorrect frame size - multiple frames in buffer?\n"); |
1692 | buf_size= s->frame_size; |
1693 | } |
1694 | |
1695 | s->frame = data; |
1696 | |
1697 | ret = mp_decode_frame(s, NULL, buf, buf_size); |
1698 | if (ret >= 0) { |
1699 | s->frame->nb_samples = avctx->frame_size; |
1700 | *got_frame_ptr = 1; |
1701 | avctx->sample_rate = s->sample_rate; |
1702 | //FIXME maybe move the other codec info stuff from above here too |
1703 | } else { |
1704 | av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n"); |
1705 | /* Only return an error if the bad frame makes up the whole packet or |
1706 | * the error is related to buffer management. |
1707 | * If there is more data in the packet, just consume the bad frame |
1708 | * instead of returning an error, which would discard the whole |
1709 | * packet. */ |
1710 | *got_frame_ptr = 0; |
1711 | if (buf_size == avpkt->size || ret != AVERROR_INVALIDDATA) |
1712 | return ret; |
1713 | } |
1714 | s->frame_size = 0; |
1715 | return buf_size + skipped; |
1716 | } |
1717 | |
1718 | static void mp_flush(MPADecodeContext *ctx) |
1719 | { |
1720 | memset(ctx->synth_buf, 0, sizeof(ctx->synth_buf)); |
1721 | memset(ctx->mdct_buf, 0, sizeof(ctx->mdct_buf)); |
1722 | ctx->last_buf_size = 0; |
1723 | ctx->dither_state = 0; |
1724 | } |
1725 | |
1726 | static void flush(AVCodecContext *avctx) |
1727 | { |
1728 | mp_flush(avctx->priv_data); |
1729 | } |
1730 | |
1731 | #if CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER |
1732 | static int decode_frame_adu(AVCodecContext *avctx, void *data, |
1733 | int *got_frame_ptr, AVPacket *avpkt) |
1734 | { |
1735 | const uint8_t *buf = avpkt->data; |
1736 | int buf_size = avpkt->size; |
1737 | MPADecodeContext *s = avctx->priv_data; |
1738 | uint32_t header; |
1739 | int len, ret; |
1740 | int av_unused out_size; |
1741 | |
1742 | len = buf_size; |
1743 | |
1744 | // Discard too short frames |
1745 | if (buf_size < HEADER_SIZE) { |
1746 | av_log(avctx, AV_LOG_ERROR, "Packet is too small\n"); |
1747 | return AVERROR_INVALIDDATA; |
1748 | } |
1749 | |
1750 | |
1751 | if (len > MPA_MAX_CODED_FRAME_SIZE) |
1752 | len = MPA_MAX_CODED_FRAME_SIZE; |
1753 | |
1754 | // Get header and restore sync word |
1755 | header = AV_RB32(buf) | 0xffe00000; |
1756 | |
1757 | ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)s, header); |
1758 | if (ret < 0) { |
1759 | av_log(avctx, AV_LOG_ERROR, "Invalid frame header\n"); |
1760 | return ret; |
1761 | } |
1762 | /* update codec info */ |
1763 | avctx->sample_rate = s->sample_rate; |
1764 | avctx->channels = s->nb_channels; |
1765 | avctx->channel_layout = s->nb_channels == 1 ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO; |
1766 | if (!avctx->bit_rate) |
1767 | avctx->bit_rate = s->bit_rate; |
1768 | |
1769 | s->frame_size = len; |
1770 | |
1771 | s->frame = data; |
1772 | |
1773 | ret = mp_decode_frame(s, NULL, buf, buf_size); |
1774 | if (ret < 0) { |
1775 | av_log(avctx, AV_LOG_ERROR, "Error while decoding MPEG audio frame.\n"); |
1776 | return ret; |
1777 | } |
1778 | |
1779 | *got_frame_ptr = 1; |
1780 | |
1781 | return buf_size; |
1782 | } |
1783 | #endif /* CONFIG_MP3ADU_DECODER || CONFIG_MP3ADUFLOAT_DECODER */ |
1784 | |
1785 | #if CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER |
1786 | |
1787 | /** |
1788 | * Context for MP3On4 decoder |
1789 | */ |
1790 | typedef struct MP3On4DecodeContext { |
1791 | int frames; ///< number of mp3 frames per block (number of mp3 decoder instances) |
1792 | int syncword; ///< syncword patch |
1793 | const uint8_t *coff; ///< channel offsets in output buffer |
1794 | MPADecodeContext *mp3decctx[5]; ///< MPADecodeContext for every decoder instance |
1795 | } MP3On4DecodeContext; |
1796 | |
1797 | #include "mpeg4audio.h" |
1798 | |
1799 | /* Next 3 arrays are indexed by channel config number (passed via codecdata) */ |
1800 | |
1801 | /* number of mp3 decoder instances */ |
1802 | static const uint8_t mp3Frames[8] = { 0, 1, 1, 2, 3, 3, 4, 5 }; |
1803 | |
1804 | /* offsets into output buffer, assume output order is FL FR C LFE BL BR SL SR */ |
1805 | static const uint8_t chan_offset[8][5] = { |
1806 | { 0 }, |
1807 | { 0 }, // C |
1808 | { 0 }, // FLR |
1809 | { 2, 0 }, // C FLR |
1810 | { 2, 0, 3 }, // C FLR BS |
1811 | { 2, 0, 3 }, // C FLR BLRS |
1812 | { 2, 0, 4, 3 }, // C FLR BLRS LFE |
1813 | { 2, 0, 6, 4, 3 }, // C FLR BLRS BLR LFE |
1814 | }; |
1815 | |
1816 | /* mp3on4 channel layouts */ |
1817 | static const int16_t chan_layout[8] = { |
1818 | 0, |
1819 | AV_CH_LAYOUT_MONO, |
1820 | AV_CH_LAYOUT_STEREO, |
1821 | AV_CH_LAYOUT_SURROUND, |
1822 | AV_CH_LAYOUT_4POINT0, |
1823 | AV_CH_LAYOUT_5POINT0, |
1824 | AV_CH_LAYOUT_5POINT1, |
1825 | AV_CH_LAYOUT_7POINT1 |
1826 | }; |
1827 | |
1828 | static av_cold int decode_close_mp3on4(AVCodecContext * avctx) |
1829 | { |
1830 | MP3On4DecodeContext *s = avctx->priv_data; |
1831 | int i; |
1832 | |
1833 | if (s->mp3decctx[0]) |
1834 | av_freep(&s->mp3decctx[0]->fdsp); |
1835 | |
1836 | for (i = 0; i < s->frames; i++) |
1837 | av_freep(&s->mp3decctx[i]); |
1838 | |
1839 | return 0; |
1840 | } |
1841 | |
1842 | |
1843 | static av_cold int decode_init_mp3on4(AVCodecContext * avctx) |
1844 | { |
1845 | MP3On4DecodeContext *s = avctx->priv_data; |
1846 | MPEG4AudioConfig cfg; |
1847 | int i; |
1848 | |
1849 | if ((avctx->extradata_size < 2) || !avctx->extradata) { |
1850 | av_log(avctx, AV_LOG_ERROR, "Codec extradata missing or too short.\n"); |
1851 | return AVERROR_INVALIDDATA; |
1852 | } |
1853 | |
1854 | avpriv_mpeg4audio_get_config(&cfg, avctx->extradata, |
1855 | avctx->extradata_size * 8, 1); |
1856 | if (!cfg.chan_config || cfg.chan_config > 7) { |
1857 | av_log(avctx, AV_LOG_ERROR, "Invalid channel config number.\n"); |
1858 | return AVERROR_INVALIDDATA; |
1859 | } |
1860 | s->frames = mp3Frames[cfg.chan_config]; |
1861 | s->coff = chan_offset[cfg.chan_config]; |
1862 | avctx->channels = ff_mpeg4audio_channels[cfg.chan_config]; |
1863 | avctx->channel_layout = chan_layout[cfg.chan_config]; |
1864 | |
1865 | if (cfg.sample_rate < 16000) |
1866 | s->syncword = 0xffe00000; |
1867 | else |
1868 | s->syncword = 0xfff00000; |
1869 | |
1870 | /* Init the first mp3 decoder in standard way, so that all tables get builded |
1871 | * We replace avctx->priv_data with the context of the first decoder so that |
1872 | * decode_init() does not have to be changed. |
1873 | * Other decoders will be initialized here copying data from the first context |
1874 | */ |
1875 | // Allocate zeroed memory for the first decoder context |
1876 | s->mp3decctx[0] = av_mallocz(sizeof(MPADecodeContext)); |
1877 | if (!s->mp3decctx[0]) |
1878 | goto alloc_fail; |
1879 | // Put decoder context in place to make init_decode() happy |
1880 | avctx->priv_data = s->mp3decctx[0]; |
1881 | decode_init(avctx); |
1882 | // Restore mp3on4 context pointer |
1883 | avctx->priv_data = s; |
1884 | s->mp3decctx[0]->adu_mode = 1; // Set adu mode |
1885 | |
1886 | /* Create a separate codec/context for each frame (first is already ok). |
1887 | * Each frame is 1 or 2 channels - up to 5 frames allowed |
1888 | */ |
1889 | for (i = 1; i < s->frames; i++) { |
1890 | s->mp3decctx[i] = av_mallocz(sizeof(MPADecodeContext)); |
1891 | if (!s->mp3decctx[i]) |
1892 | goto alloc_fail; |
1893 | s->mp3decctx[i]->adu_mode = 1; |
1894 | s->mp3decctx[i]->avctx = avctx; |
1895 | s->mp3decctx[i]->mpadsp = s->mp3decctx[0]->mpadsp; |
1896 | s->mp3decctx[i]->fdsp = s->mp3decctx[0]->fdsp; |
1897 | } |
1898 | |
1899 | return 0; |
1900 | alloc_fail: |
1901 | decode_close_mp3on4(avctx); |
1902 | return AVERROR(ENOMEM); |
1903 | } |
1904 | |
1905 | |
1906 | static void flush_mp3on4(AVCodecContext *avctx) |
1907 | { |
1908 | int i; |
1909 | MP3On4DecodeContext *s = avctx->priv_data; |
1910 | |
1911 | for (i = 0; i < s->frames; i++) |
1912 | mp_flush(s->mp3decctx[i]); |
1913 | } |
1914 | |
1915 | |
1916 | static int decode_frame_mp3on4(AVCodecContext *avctx, void *data, |
1917 | int *got_frame_ptr, AVPacket *avpkt) |
1918 | { |
1919 | AVFrame *frame = data; |
1920 | const uint8_t *buf = avpkt->data; |
1921 | int buf_size = avpkt->size; |
1922 | MP3On4DecodeContext *s = avctx->priv_data; |
1923 | MPADecodeContext *m; |
1924 | int fsize, len = buf_size, out_size = 0; |
1925 | uint32_t header; |
1926 | OUT_INT **out_samples; |
1927 | OUT_INT *outptr[2]; |
1928 | int fr, ch, ret; |
1929 | |
1930 | /* get output buffer */ |
1931 | frame->nb_samples = MPA_FRAME_SIZE; |
1932 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
1933 | return ret; |
1934 | out_samples = (OUT_INT **)frame->extended_data; |
1935 | |
1936 | // Discard too short frames |
1937 | if (buf_size < HEADER_SIZE) |
1938 | return AVERROR_INVALIDDATA; |
1939 | |
1940 | avctx->bit_rate = 0; |
1941 | |
1942 | ch = 0; |
1943 | for (fr = 0; fr < s->frames; fr++) { |
1944 | fsize = AV_RB16(buf) >> 4; |
1945 | fsize = FFMIN3(fsize, len, MPA_MAX_CODED_FRAME_SIZE); |
1946 | m = s->mp3decctx[fr]; |
1947 | av_assert1(m); |
1948 | |
1949 | if (fsize < HEADER_SIZE) { |
1950 | av_log(avctx, AV_LOG_ERROR, "Frame size smaller than header size\n"); |
1951 | return AVERROR_INVALIDDATA; |
1952 | } |
1953 | header = (AV_RB32(buf) & 0x000fffff) | s->syncword; // patch header |
1954 | |
1955 | ret = avpriv_mpegaudio_decode_header((MPADecodeHeader *)m, header); |
1956 | if (ret < 0) { |
1957 | av_log(avctx, AV_LOG_ERROR, "Bad header, discard block\n"); |
1958 | return AVERROR_INVALIDDATA; |
1959 | } |
1960 | |
1961 | if (ch + m->nb_channels > avctx->channels || |
1962 | s->coff[fr] + m->nb_channels > avctx->channels) { |
1963 | av_log(avctx, AV_LOG_ERROR, "frame channel count exceeds codec " |
1964 | "channel count\n"); |
1965 | return AVERROR_INVALIDDATA; |
1966 | } |
1967 | ch += m->nb_channels; |
1968 | |
1969 | outptr[0] = out_samples[s->coff[fr]]; |
1970 | if (m->nb_channels > 1) |
1971 | outptr[1] = out_samples[s->coff[fr] + 1]; |
1972 | |
1973 | if ((ret = mp_decode_frame(m, outptr, buf, fsize)) < 0) { |
1974 | av_log(avctx, AV_LOG_ERROR, "failed to decode channel %d\n", ch); |
1975 | memset(outptr[0], 0, MPA_FRAME_SIZE*sizeof(OUT_INT)); |
1976 | if (m->nb_channels > 1) |
1977 | memset(outptr[1], 0, MPA_FRAME_SIZE*sizeof(OUT_INT)); |
1978 | ret = m->nb_channels * MPA_FRAME_SIZE*sizeof(OUT_INT); |
1979 | } |
1980 | |
1981 | out_size += ret; |
1982 | buf += fsize; |
1983 | len -= fsize; |
1984 | |
1985 | avctx->bit_rate += m->bit_rate; |
1986 | } |
1987 | if (ch != avctx->channels) { |
1988 | av_log(avctx, AV_LOG_ERROR, "failed to decode all channels\n"); |
1989 | return AVERROR_INVALIDDATA; |
1990 | } |
1991 | |
1992 | /* update codec info */ |
1993 | avctx->sample_rate = s->mp3decctx[0]->sample_rate; |
1994 | |
1995 | frame->nb_samples = out_size / (avctx->channels * sizeof(OUT_INT)); |
1996 | *got_frame_ptr = 1; |
1997 | |
1998 | return buf_size; |
1999 | } |
2000 | #endif /* CONFIG_MP3ON4_DECODER || CONFIG_MP3ON4FLOAT_DECODER */ |
2001 |