Commit | Line | Data |
---|---|---|
2340bcd3 JVH |
1 | /* |
2 | * ALAC (Apple Lossless Audio Codec) decoder | |
3 | * Copyright (c) 2005 David Hammerton | |
4 | * All rights reserved. | |
5 | * | |
6 | * This is the actual decoder. | |
7 | * | |
8 | * http://crazney.net/programs/itunes/alac.html | |
9 | * | |
10 | * Permission is hereby granted, free of charge, to any person | |
11 | * obtaining a copy of this software and associated documentation | |
12 | * files (the "Software"), to deal in the Software without | |
13 | * restriction, including without limitation the rights to use, | |
14 | * copy, modify, merge, publish, distribute, sublicense, and/or | |
15 | * sell copies of the Software, and to permit persons to whom the | |
16 | * Software is furnished to do so, subject to the following conditions: | |
17 | * | |
18 | * The above copyright notice and this permission notice shall be | |
19 | * included in all copies or substantial portions of the Software. | |
20 | * | |
21 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
22 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES | |
23 | * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
24 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT | |
25 | * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, | |
26 | * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING | |
27 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR | |
28 | * OTHER DEALINGS IN THE SOFTWARE. | |
29 | * | |
30 | */ | |
31 | ||
32 | #ifdef __BIG_ENDIAN__ | |
33 | static const int host_bigendian = 1; | |
34 | #else | |
35 | static const int host_bigendian = 0; | |
36 | #endif | |
37 | ||
38 | #include <stdio.h> | |
39 | #include <stdlib.h> | |
40 | #include <string.h> | |
41 | #ifdef _WIN32 | |
42 | #include "stdint_win.h" | |
43 | #else | |
44 | #include <stdint.h> | |
45 | #endif | |
46 | ||
47 | #include "alac.h" | |
48 | ||
49 | #define _Swap32(v) do { \ | |
50 | v = (((v) & 0x000000FF) << 0x18) | \ | |
51 | (((v) & 0x0000FF00) << 0x08) | \ | |
52 | (((v) & 0x00FF0000) >> 0x08) | \ | |
53 | (((v) & 0xFF000000) >> 0x18); } while(0) | |
54 | ||
55 | #define _Swap16(v) do { \ | |
56 | v = (((v) & 0x00FF) << 0x08) | \ | |
57 | (((v) & 0xFF00) >> 0x08); } while (0) | |
58 | ||
59 | struct {signed int x:24;} se_struct_24; | |
60 | #define SignExtend24(val) (se_struct_24.x = val) | |
61 | ||
62 | struct alac_file | |
63 | { | |
64 | unsigned char *input_buffer; | |
65 | int input_buffer_bitaccumulator; /* used so we can do arbitary | |
66 | bit reads */ | |
67 | ||
68 | int samplesize; | |
69 | int numchannels; | |
70 | int bytespersample; | |
71 | ||
72 | ||
73 | /* buffers */ | |
74 | int32_t *predicterror_buffer_a; | |
75 | int32_t *predicterror_buffer_b; | |
76 | ||
77 | int32_t *outputsamples_buffer_a; | |
78 | int32_t *outputsamples_buffer_b; | |
79 | ||
80 | int32_t *uncompressed_bytes_buffer_a; | |
81 | int32_t *uncompressed_bytes_buffer_b; | |
82 | ||
83 | ||
84 | ||
85 | /* stuff from setinfo */ | |
86 | uint32_t setinfo_max_samples_per_frame; /* 0x1000 = 4096 */ /* max samples per frame? */ | |
87 | uint8_t setinfo_7a; /* 0x00 */ | |
88 | uint8_t setinfo_sample_size; /* 0x10 */ | |
89 | uint8_t setinfo_rice_historymult; /* 0x28 */ | |
90 | uint8_t setinfo_rice_initialhistory; /* 0x0a */ | |
91 | uint8_t setinfo_rice_kmodifier; /* 0x0e */ | |
92 | uint8_t setinfo_7f; /* 0x02 */ | |
93 | uint16_t setinfo_80; /* 0x00ff */ | |
94 | uint32_t setinfo_82; /* 0x000020e7 */ /* max sample size?? */ | |
95 | uint32_t setinfo_86; /* 0x00069fe4 */ /* bit rate (avarge)?? */ | |
96 | uint32_t setinfo_8a_rate; /* 0x0000ac44 */ | |
97 | /* end setinfo stuff */ | |
98 | ||
99 | }; | |
100 | ||
101 | ||
102 | static void allocate_buffers(alac_file *alac) | |
103 | { | |
104 | alac->predicterror_buffer_a = malloc(alac->setinfo_max_samples_per_frame * 4); | |
105 | alac->predicterror_buffer_b = malloc(alac->setinfo_max_samples_per_frame * 4); | |
106 | ||
107 | alac->outputsamples_buffer_a = malloc(alac->setinfo_max_samples_per_frame * 4); | |
108 | alac->outputsamples_buffer_b = malloc(alac->setinfo_max_samples_per_frame * 4); | |
109 | ||
110 | alac->uncompressed_bytes_buffer_a = malloc(alac->setinfo_max_samples_per_frame * 4); | |
111 | alac->uncompressed_bytes_buffer_b = malloc(alac->setinfo_max_samples_per_frame * 4); | |
112 | } | |
113 | ||
114 | void alac_set_info(alac_file *alac, char *inputbuffer) | |
115 | { | |
116 | char *ptr = inputbuffer; | |
117 | ptr += 4; /* size */ | |
118 | ptr += 4; /* frma */ | |
119 | ptr += 4; /* alac */ | |
120 | ptr += 4; /* size */ | |
121 | ptr += 4; /* alac */ | |
122 | ||
123 | ptr += 4; /* 0 ? */ | |
124 | ||
125 | alac->setinfo_max_samples_per_frame = *(uint32_t*)ptr; /* buffer size / 2 ? */ | |
126 | if (!host_bigendian) | |
127 | _Swap32(alac->setinfo_max_samples_per_frame); | |
128 | ptr += 4; | |
129 | alac->setinfo_7a = *(uint8_t*)ptr; | |
130 | ptr += 1; | |
131 | alac->setinfo_sample_size = *(uint8_t*)ptr; | |
132 | ptr += 1; | |
133 | alac->setinfo_rice_historymult = *(uint8_t*)ptr; | |
134 | ptr += 1; | |
135 | alac->setinfo_rice_initialhistory = *(uint8_t*)ptr; | |
136 | ptr += 1; | |
137 | alac->setinfo_rice_kmodifier = *(uint8_t*)ptr; | |
138 | ptr += 1; | |
139 | alac->setinfo_7f = *(uint8_t*)ptr; | |
140 | ptr += 1; | |
141 | alac->setinfo_80 = *(uint16_t*)ptr; | |
142 | if (!host_bigendian) | |
143 | _Swap16(alac->setinfo_80); | |
144 | ptr += 2; | |
145 | alac->setinfo_82 = *(uint32_t*)ptr; | |
146 | if (!host_bigendian) | |
147 | _Swap32(alac->setinfo_82); | |
148 | ptr += 4; | |
149 | alac->setinfo_86 = *(uint32_t*)ptr; | |
150 | if (!host_bigendian) | |
151 | _Swap32(alac->setinfo_86); | |
152 | ptr += 4; | |
153 | alac->setinfo_8a_rate = *(uint32_t*)ptr; | |
154 | if (!host_bigendian) | |
155 | _Swap32(alac->setinfo_8a_rate); | |
156 | ptr += 4; | |
157 | ||
158 | allocate_buffers(alac); | |
159 | ||
160 | } | |
161 | ||
162 | /* stream reading */ | |
163 | ||
164 | /* supports reading 1 to 16 bits, in big endian format */ | |
165 | static uint32_t readbits_16(alac_file *alac, int bits) | |
166 | { | |
167 | uint32_t result; | |
168 | int new_accumulator; | |
169 | ||
170 | result = (alac->input_buffer[0] << 16) | | |
171 | (alac->input_buffer[1] << 8) | | |
172 | (alac->input_buffer[2]); | |
173 | ||
174 | /* shift left by the number of bits we've already read, | |
175 | * so that the top 'n' bits of the 24 bits we read will | |
176 | * be the return bits */ | |
177 | result = result << alac->input_buffer_bitaccumulator; | |
178 | ||
179 | result = result & 0x00ffffff; | |
180 | ||
181 | /* and then only want the top 'n' bits from that, where | |
182 | * n is 'bits' */ | |
183 | result = result >> (24 - bits); | |
184 | ||
185 | new_accumulator = (alac->input_buffer_bitaccumulator + bits); | |
186 | ||
187 | /* increase the buffer pointer if we've read over n bytes. */ | |
188 | alac->input_buffer += (new_accumulator >> 3); | |
189 | ||
190 | /* and the remainder goes back into the bit accumulator */ | |
191 | alac->input_buffer_bitaccumulator = (new_accumulator & 7); | |
192 | ||
193 | return result; | |
194 | } | |
195 | ||
196 | /* supports reading 1 to 32 bits, in big endian format */ | |
197 | static uint32_t readbits(alac_file *alac, int bits) | |
198 | { | |
199 | int32_t result = 0; | |
200 | ||
201 | if (bits > 16) | |
202 | { | |
203 | bits -= 16; | |
204 | result = readbits_16(alac, 16) << bits; | |
205 | } | |
206 | ||
207 | result |= readbits_16(alac, bits); | |
208 | ||
209 | return result; | |
210 | } | |
211 | ||
212 | /* reads a single bit */ | |
213 | static int readbit(alac_file *alac) | |
214 | { | |
215 | int result; | |
216 | int new_accumulator; | |
217 | ||
218 | result = alac->input_buffer[0]; | |
219 | ||
220 | result = result << alac->input_buffer_bitaccumulator; | |
221 | ||
222 | result = result >> 7 & 1; | |
223 | ||
224 | new_accumulator = (alac->input_buffer_bitaccumulator + 1); | |
225 | ||
226 | alac->input_buffer += (new_accumulator / 8); | |
227 | ||
228 | alac->input_buffer_bitaccumulator = (new_accumulator % 8); | |
229 | ||
230 | return result; | |
231 | } | |
232 | ||
233 | static void unreadbits(alac_file *alac, int bits) | |
234 | { | |
235 | int new_accumulator = (alac->input_buffer_bitaccumulator - bits); | |
236 | ||
237 | alac->input_buffer += (new_accumulator >> 3); | |
238 | ||
239 | alac->input_buffer_bitaccumulator = (new_accumulator & 7); | |
240 | if (alac->input_buffer_bitaccumulator < 0) | |
241 | alac->input_buffer_bitaccumulator *= -1; | |
242 | } | |
243 | ||
244 | /* various implementations of count_leading_zero: | |
245 | * the first one is the original one, the simplest and most | |
246 | * obvious for what it's doing. never use this. | |
247 | * then there are the asm ones. fill in as necessary | |
248 | * and finally an unrolled and optimised c version | |
249 | * to fall back to | |
250 | */ | |
251 | #if 0 | |
252 | /* hideously inefficient. could use a bitmask search, | |
253 | * alternatively bsr on x86, | |
254 | */ | |
255 | static int count_leading_zeros(int32_t input) | |
256 | { | |
257 | int i = 0; | |
258 | while (!(0x80000000 & input) && i < 32) | |
259 | { | |
260 | i++; | |
261 | input = input << 1; | |
262 | } | |
263 | return i; | |
264 | } | |
265 | #elif defined(__GNUC__) && (defined(_X86) || defined(__i386) || defined(i386)) | |
266 | /* for some reason the unrolled version (below) is | |
267 | * actually faster than this. yay intel! | |
268 | */ | |
269 | static int count_leading_zeros(int input) | |
270 | { | |
271 | int output = 0; | |
272 | if (!input) return 32; | |
273 | __asm("bsr %1, %0\n" | |
274 | : "=r" (output) | |
275 | : "r" (input)); | |
276 | return (0x1f - output); | |
277 | } | |
278 | #elif defined(__GNUC__) | |
279 | static int count_leading_zeros(int input) | |
280 | { | |
281 | return __builtin_clz(input); | |
282 | } | |
283 | #elif defined(_MSC_VER) && defined(_M_IX86) | |
284 | static int count_leading_zeros(int input) | |
285 | { | |
286 | int output = 0; | |
287 | if (!input) return 32; | |
288 | __asm | |
289 | { | |
290 | mov eax, input; | |
291 | mov edx, 0x1f; | |
292 | bsr ecx, eax; | |
293 | sub edx, ecx; | |
294 | mov output, edx; | |
295 | } | |
296 | return output; | |
297 | } | |
298 | #else | |
299 | #warning using generic count leading zeroes. You may wish to write one for your CPU / compiler | |
300 | static int count_leading_zeros(int input) | |
301 | { | |
302 | int output = 0; | |
303 | int curbyte = 0; | |
304 | ||
305 | curbyte = input >> 24; | |
306 | if (curbyte) goto found; | |
307 | output += 8; | |
308 | ||
309 | curbyte = input >> 16; | |
310 | if (curbyte & 0xff) goto found; | |
311 | output += 8; | |
312 | ||
313 | curbyte = input >> 8; | |
314 | if (curbyte & 0xff) goto found; | |
315 | output += 8; | |
316 | ||
317 | curbyte = input; | |
318 | if (curbyte & 0xff) goto found; | |
319 | output += 8; | |
320 | ||
321 | return output; | |
322 | ||
323 | found: | |
324 | if (!(curbyte & 0xf0)) | |
325 | { | |
326 | output += 4; | |
327 | } | |
328 | else | |
329 | curbyte >>= 4; | |
330 | ||
331 | if (curbyte & 0x8) | |
332 | return output; | |
333 | if (curbyte & 0x4) | |
334 | return output + 1; | |
335 | if (curbyte & 0x2) | |
336 | return output + 2; | |
337 | if (curbyte & 0x1) | |
338 | return output + 3; | |
339 | ||
340 | /* shouldn't get here: */ | |
341 | return output + 4; | |
342 | } | |
343 | #endif | |
344 | ||
345 | #define RICE_THRESHOLD 8 // maximum number of bits for a rice prefix. | |
346 | ||
347 | int32_t entropy_decode_value(alac_file* alac, | |
348 | int readSampleSize, | |
349 | int k, | |
350 | int rice_kmodifier_mask) | |
351 | { | |
352 | int32_t x = 0; // decoded value | |
353 | ||
354 | // read x, number of 1s before 0 represent the rice value. | |
355 | while (x <= RICE_THRESHOLD && readbit(alac)) | |
356 | { | |
357 | x++; | |
358 | } | |
359 | ||
360 | if (x > RICE_THRESHOLD) | |
361 | { | |
362 | // read the number from the bit stream (raw value) | |
363 | int32_t value; | |
364 | ||
365 | value = readbits(alac, readSampleSize); | |
366 | ||
367 | // mask value | |
368 | value &= (((uint32_t)0xffffffff) >> (32 - readSampleSize)); | |
369 | ||
370 | x = value; | |
371 | } | |
372 | else | |
373 | { | |
374 | if (k != 1) | |
375 | { | |
376 | int extraBits = readbits(alac, k); | |
377 | ||
378 | // x = x * (2^k - 1) | |
379 | x *= (((1 << k) - 1) & rice_kmodifier_mask); | |
380 | ||
381 | if (extraBits > 1) | |
382 | x += extraBits - 1; | |
383 | else | |
384 | unreadbits(alac, 1); | |
385 | } | |
386 | } | |
387 | ||
388 | return x; | |
389 | } | |
390 | ||
391 | void entropy_rice_decode(alac_file* alac, | |
392 | int32_t* outputBuffer, | |
393 | int outputSize, | |
394 | int readSampleSize, | |
395 | int rice_initialhistory, | |
396 | int rice_kmodifier, | |
397 | int rice_historymult, | |
398 | int rice_kmodifier_mask) | |
399 | { | |
400 | int outputCount; | |
401 | int history = rice_initialhistory; | |
402 | int signModifier = 0; | |
403 | ||
404 | for (outputCount = 0; outputCount < outputSize; outputCount++) | |
405 | { | |
406 | int32_t decodedValue; | |
407 | int32_t finalValue; | |
408 | int32_t k; | |
409 | ||
410 | k = 31 - rice_kmodifier - count_leading_zeros((history >> 9) + 3); | |
411 | ||
412 | if (k < 0) k += rice_kmodifier; | |
413 | else k = rice_kmodifier; | |
414 | ||
415 | // note: don't use rice_kmodifier_mask here (set mask to 0xFFFFFFFF) | |
416 | decodedValue = entropy_decode_value(alac, readSampleSize, k, 0xFFFFFFFF); | |
417 | ||
418 | decodedValue += signModifier; | |
419 | finalValue = (decodedValue + 1) / 2; // inc by 1 and shift out sign bit | |
420 | if (decodedValue & 1) // the sign is stored in the low bit | |
421 | finalValue *= -1; | |
422 | ||
423 | outputBuffer[outputCount] = finalValue; | |
424 | ||
425 | signModifier = 0; | |
426 | ||
427 | // update history | |
428 | history += (decodedValue * rice_historymult) | |
429 | - ((history * rice_historymult) >> 9); | |
430 | ||
431 | if (decodedValue > 0xFFFF) | |
432 | history = 0xFFFF; | |
433 | ||
434 | // special case, for compressed blocks of 0 | |
435 | if ((history < 128) && (outputCount + 1 < outputSize)) | |
436 | { | |
437 | int32_t blockSize; | |
438 | ||
439 | signModifier = 1; | |
440 | ||
441 | k = count_leading_zeros(history) + ((history + 16) / 64) - 24; | |
442 | ||
443 | // note: blockSize is always 16bit | |
444 | blockSize = entropy_decode_value(alac, 16, k, rice_kmodifier_mask); | |
445 | ||
446 | // got blockSize 0s | |
447 | if (blockSize > 0) | |
448 | { | |
449 | memset(&outputBuffer[outputCount + 1], 0, blockSize * sizeof(*outputBuffer)); | |
450 | outputCount += blockSize; | |
451 | } | |
452 | ||
453 | if (blockSize > 0xFFFF) | |
454 | signModifier = 0; | |
455 | ||
456 | history = 0; | |
457 | } | |
458 | } | |
459 | } | |
460 | ||
461 | #define SIGN_EXTENDED32(val, bits) ((val << (32 - bits)) >> (32 - bits)) | |
462 | ||
463 | #define SIGN_ONLY(v) \ | |
464 | ((v < 0) ? (-1) : \ | |
465 | ((v > 0) ? (1) : \ | |
466 | (0))) | |
467 | ||
468 | static void predictor_decompress_fir_adapt(int32_t *error_buffer, | |
469 | int32_t *buffer_out, | |
470 | int output_size, | |
471 | int readsamplesize, | |
472 | int16_t *predictor_coef_table, | |
473 | int predictor_coef_num, | |
474 | int predictor_quantitization) | |
475 | { | |
476 | int i; | |
477 | ||
478 | /* first sample always copies */ | |
479 | *buffer_out = *error_buffer; | |
480 | ||
481 | if (!predictor_coef_num) | |
482 | { | |
483 | if (output_size <= 1) return; | |
484 | memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4); | |
485 | return; | |
486 | } | |
487 | ||
488 | if (predictor_coef_num == 0x1f) /* 11111 - max value of predictor_coef_num */ | |
489 | { /* second-best case scenario for fir decompression, | |
490 | * error describes a small difference from the previous sample only | |
491 | */ | |
492 | if (output_size <= 1) return; | |
493 | for (i = 0; i < output_size - 1; i++) | |
494 | { | |
495 | int32_t prev_value; | |
496 | int32_t error_value; | |
497 | ||
498 | prev_value = buffer_out[i]; | |
499 | error_value = error_buffer[i+1]; | |
500 | buffer_out[i+1] = SIGN_EXTENDED32((prev_value + error_value), readsamplesize); | |
501 | } | |
502 | return; | |
503 | } | |
504 | ||
505 | /* read warm-up samples */ | |
506 | if (predictor_coef_num > 0) | |
507 | { | |
508 | int i; | |
509 | for (i = 0; i < predictor_coef_num; i++) | |
510 | { | |
511 | int32_t val; | |
512 | ||
513 | val = buffer_out[i] + error_buffer[i+1]; | |
514 | ||
515 | val = SIGN_EXTENDED32(val, readsamplesize); | |
516 | ||
517 | buffer_out[i+1] = val; | |
518 | } | |
519 | } | |
520 | ||
521 | #if 0 | |
522 | /* 4 and 8 are very common cases (the only ones i've seen). these | |
523 | * should be unrolled and optimised | |
524 | */ | |
525 | if (predictor_coef_num == 4) | |
526 | { | |
527 | /* FIXME: optimised general case */ | |
528 | return; | |
529 | } | |
530 | ||
531 | if (predictor_coef_table == 8) | |
532 | { | |
533 | /* FIXME: optimised general case */ | |
534 | return; | |
535 | } | |
536 | #endif | |
537 | ||
538 | ||
539 | /* general case */ | |
540 | if (predictor_coef_num > 0) | |
541 | { | |
542 | for (i = predictor_coef_num + 1; | |
543 | i < output_size; | |
544 | i++) | |
545 | { | |
546 | int j; | |
547 | int sum = 0; | |
548 | int outval; | |
549 | int error_val = error_buffer[i]; | |
550 | ||
551 | for (j = 0; j < predictor_coef_num; j++) | |
552 | { | |
553 | sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) * | |
554 | predictor_coef_table[j]; | |
555 | } | |
556 | ||
557 | outval = (1 << (predictor_quantitization-1)) + sum; | |
558 | outval = outval >> predictor_quantitization; | |
559 | outval = outval + buffer_out[0] + error_val; | |
560 | outval = SIGN_EXTENDED32(outval, readsamplesize); | |
561 | ||
562 | buffer_out[predictor_coef_num+1] = outval; | |
563 | ||
564 | if (error_val > 0) | |
565 | { | |
566 | int predictor_num = predictor_coef_num - 1; | |
567 | ||
568 | while (predictor_num >= 0 && error_val > 0) | |
569 | { | |
570 | int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num]; | |
571 | int sign = SIGN_ONLY(val); | |
572 | ||
573 | predictor_coef_table[predictor_num] -= sign; | |
574 | ||
575 | val *= sign; /* absolute value */ | |
576 | ||
577 | error_val -= ((val >> predictor_quantitization) * | |
578 | (predictor_coef_num - predictor_num)); | |
579 | ||
580 | predictor_num--; | |
581 | } | |
582 | } | |
583 | else if (error_val < 0) | |
584 | { | |
585 | int predictor_num = predictor_coef_num - 1; | |
586 | ||
587 | while (predictor_num >= 0 && error_val < 0) | |
588 | { | |
589 | int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num]; | |
590 | int sign = - SIGN_ONLY(val); | |
591 | ||
592 | predictor_coef_table[predictor_num] -= sign; | |
593 | ||
594 | val *= sign; /* neg value */ | |
595 | ||
596 | error_val -= ((val >> predictor_quantitization) * | |
597 | (predictor_coef_num - predictor_num)); | |
598 | ||
599 | predictor_num--; | |
600 | } | |
601 | } | |
602 | ||
603 | buffer_out++; | |
604 | } | |
605 | } | |
606 | } | |
607 | ||
608 | void deinterlace_16(int32_t *buffer_a, int32_t *buffer_b, | |
609 | int16_t *buffer_out, | |
610 | int numchannels, int numsamples, | |
611 | uint8_t interlacing_shift, | |
612 | uint8_t interlacing_leftweight) | |
613 | { | |
614 | int i; | |
615 | if (numsamples <= 0) return; | |
616 | ||
617 | /* weighted interlacing */ | |
618 | if (interlacing_leftweight) | |
619 | { | |
620 | for (i = 0; i < numsamples; i++) | |
621 | { | |
622 | int32_t difference, midright; | |
623 | int16_t left; | |
624 | int16_t right; | |
625 | ||
626 | midright = buffer_a[i]; | |
627 | difference = buffer_b[i]; | |
628 | ||
629 | ||
630 | right = midright - ((difference * interlacing_leftweight) >> interlacing_shift); | |
631 | left = right + difference; | |
632 | ||
633 | /* output is always little endian */ | |
634 | if (host_bigendian) | |
635 | { | |
636 | _Swap16(left); | |
637 | _Swap16(right); | |
638 | } | |
639 | ||
640 | buffer_out[i*numchannels] = left; | |
641 | buffer_out[i*numchannels + 1] = right; | |
642 | } | |
643 | ||
644 | return; | |
645 | } | |
646 | ||
647 | /* otherwise basic interlacing took place */ | |
648 | for (i = 0; i < numsamples; i++) | |
649 | { | |
650 | int16_t left, right; | |
651 | ||
652 | left = buffer_a[i]; | |
653 | right = buffer_b[i]; | |
654 | ||
655 | /* output is always little endian */ | |
656 | if (host_bigendian) | |
657 | { | |
658 | _Swap16(left); | |
659 | _Swap16(right); | |
660 | } | |
661 | ||
662 | buffer_out[i*numchannels] = left; | |
663 | buffer_out[i*numchannels + 1] = right; | |
664 | } | |
665 | } | |
666 | ||
667 | void deinterlace_24(int32_t *buffer_a, int32_t *buffer_b, | |
668 | int uncompressed_bytes, | |
669 | int32_t *uncompressed_bytes_buffer_a, int32_t *uncompressed_bytes_buffer_b, | |
670 | void *buffer_out, | |
671 | int numchannels, int numsamples, | |
672 | uint8_t interlacing_shift, | |
673 | uint8_t interlacing_leftweight) | |
674 | { | |
675 | int i; | |
676 | if (numsamples <= 0) return; | |
677 | ||
678 | /* weighted interlacing */ | |
679 | if (interlacing_leftweight) | |
680 | { | |
681 | for (i = 0; i < numsamples; i++) | |
682 | { | |
683 | int32_t difference, midright; | |
684 | int32_t left; | |
685 | int32_t right; | |
686 | ||
687 | midright = buffer_a[i]; | |
688 | difference = buffer_b[i]; | |
689 | ||
690 | right = midright - ((difference * interlacing_leftweight) >> interlacing_shift); | |
691 | left = right + difference; | |
692 | ||
693 | if (uncompressed_bytes) | |
694 | { | |
695 | uint32_t mask = ~(0xFFFFFFFF << (uncompressed_bytes * 8)); | |
696 | left <<= (uncompressed_bytes * 8); | |
697 | right <<= (uncompressed_bytes * 8); | |
698 | ||
699 | left |= uncompressed_bytes_buffer_a[i] & mask; | |
700 | right |= uncompressed_bytes_buffer_b[i] & mask; | |
701 | } | |
702 | ||
703 | ((uint8_t*)buffer_out)[i * numchannels * 3] = (left) & 0xFF; | |
704 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 1] = (left >> 8) & 0xFF; | |
705 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 2] = (left >> 16) & 0xFF; | |
706 | ||
707 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 3] = (right) & 0xFF; | |
708 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 4] = (right >> 8) & 0xFF; | |
709 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 5] = (right >> 16) & 0xFF; | |
710 | } | |
711 | ||
712 | return; | |
713 | } | |
714 | ||
715 | /* otherwise basic interlacing took place */ | |
716 | for (i = 0; i < numsamples; i++) | |
717 | { | |
718 | int32_t left, right; | |
719 | ||
720 | left = buffer_a[i]; | |
721 | right = buffer_b[i]; | |
722 | ||
723 | if (uncompressed_bytes) | |
724 | { | |
725 | uint32_t mask = ~(0xFFFFFFFF << (uncompressed_bytes * 8)); | |
726 | left <<= (uncompressed_bytes * 8); | |
727 | right <<= (uncompressed_bytes * 8); | |
728 | ||
729 | left |= uncompressed_bytes_buffer_a[i] & mask; | |
730 | right |= uncompressed_bytes_buffer_b[i] & mask; | |
731 | } | |
732 | ||
733 | ((uint8_t*)buffer_out)[i * numchannels * 3] = (left) & 0xFF; | |
734 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 1] = (left >> 8) & 0xFF; | |
735 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 2] = (left >> 16) & 0xFF; | |
736 | ||
737 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 3] = (right) & 0xFF; | |
738 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 4] = (right >> 8) & 0xFF; | |
739 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 5] = (right >> 16) & 0xFF; | |
740 | ||
741 | } | |
742 | ||
743 | } | |
744 | ||
745 | void decode_frame(alac_file *alac, | |
746 | unsigned char *inbuffer, | |
747 | void *outbuffer, int *outputsize) | |
748 | { | |
749 | int channels; | |
750 | int32_t outputsamples = alac->setinfo_max_samples_per_frame; | |
751 | ||
752 | /* setup the stream */ | |
753 | alac->input_buffer = inbuffer; | |
754 | alac->input_buffer_bitaccumulator = 0; | |
755 | ||
756 | channels = readbits(alac, 3); | |
757 | ||
758 | *outputsize = outputsamples * alac->bytespersample; | |
759 | ||
760 | switch(channels) | |
761 | { | |
762 | case 0: /* 1 channel */ | |
763 | { | |
764 | int hassize; | |
765 | int isnotcompressed; | |
766 | int readsamplesize; | |
767 | ||
768 | int uncompressed_bytes; | |
769 | int ricemodifier; | |
770 | ||
771 | /* 2^result = something to do with output waiting. | |
772 | * perhaps matters if we read > 1 frame in a pass? | |
773 | */ | |
774 | readbits(alac, 4); | |
775 | ||
776 | readbits(alac, 12); /* unknown, skip 12 bits */ | |
777 | ||
778 | hassize = readbits(alac, 1); /* the output sample size is stored soon */ | |
779 | ||
780 | uncompressed_bytes = readbits(alac, 2); /* number of bytes in the (compressed) stream that are not compressed */ | |
781 | ||
782 | isnotcompressed = readbits(alac, 1); /* whether the frame is compressed */ | |
783 | ||
784 | if (hassize) | |
785 | { | |
786 | /* now read the number of samples, | |
787 | * as a 32bit integer */ | |
788 | outputsamples = readbits(alac, 32); | |
789 | *outputsize = outputsamples * alac->bytespersample; | |
790 | } | |
791 | ||
792 | readsamplesize = alac->setinfo_sample_size - (uncompressed_bytes * 8); | |
793 | ||
794 | if (!isnotcompressed) | |
795 | { /* so it is compressed */ | |
796 | int16_t predictor_coef_table[32]; | |
797 | int predictor_coef_num; | |
798 | int prediction_type; | |
799 | int prediction_quantitization; | |
800 | int i; | |
801 | ||
802 | /* skip 16 bits, not sure what they are. seem to be used in | |
803 | * two channel case */ | |
804 | readbits(alac, 8); | |
805 | readbits(alac, 8); | |
806 | ||
807 | prediction_type = readbits(alac, 4); | |
808 | prediction_quantitization = readbits(alac, 4); | |
809 | ||
810 | ricemodifier = readbits(alac, 3); | |
811 | predictor_coef_num = readbits(alac, 5); | |
812 | ||
813 | /* read the predictor table */ | |
814 | for (i = 0; i < predictor_coef_num; i++) | |
815 | { | |
816 | predictor_coef_table[i] = (int16_t)readbits(alac, 16); | |
817 | } | |
818 | ||
819 | if (uncompressed_bytes) | |
820 | { | |
821 | int i; | |
822 | for (i = 0; i < outputsamples; i++) | |
823 | { | |
824 | alac->uncompressed_bytes_buffer_a[i] = readbits(alac, uncompressed_bytes * 8); | |
825 | } | |
826 | } | |
827 | ||
828 | entropy_rice_decode(alac, | |
829 | alac->predicterror_buffer_a, | |
830 | outputsamples, | |
831 | readsamplesize, | |
832 | alac->setinfo_rice_initialhistory, | |
833 | alac->setinfo_rice_kmodifier, | |
834 | ricemodifier * alac->setinfo_rice_historymult / 4, | |
835 | (1 << alac->setinfo_rice_kmodifier) - 1); | |
836 | ||
837 | if (prediction_type == 0) | |
838 | { /* adaptive fir */ | |
839 | predictor_decompress_fir_adapt(alac->predicterror_buffer_a, | |
840 | alac->outputsamples_buffer_a, | |
841 | outputsamples, | |
842 | readsamplesize, | |
843 | predictor_coef_table, | |
844 | predictor_coef_num, | |
845 | prediction_quantitization); | |
846 | } | |
847 | else | |
848 | { | |
849 | fprintf(stderr, "FIXME: unhandled predicition type: %i\n", prediction_type); | |
850 | /* i think the only other prediction type (or perhaps this is just a | |
851 | * boolean?) runs adaptive fir twice.. like: | |
852 | * predictor_decompress_fir_adapt(predictor_error, tempout, ...) | |
853 | * predictor_decompress_fir_adapt(predictor_error, outputsamples ...) | |
854 | * little strange.. | |
855 | */ | |
856 | } | |
857 | ||
858 | } | |
859 | else | |
860 | { /* not compressed, easy case */ | |
861 | if (alac->setinfo_sample_size <= 16) | |
862 | { | |
863 | int i; | |
864 | for (i = 0; i < outputsamples; i++) | |
865 | { | |
866 | int32_t audiobits = readbits(alac, alac->setinfo_sample_size); | |
867 | ||
868 | audiobits = SIGN_EXTENDED32(audiobits, alac->setinfo_sample_size); | |
869 | ||
870 | alac->outputsamples_buffer_a[i] = audiobits; | |
871 | } | |
872 | } | |
873 | else | |
874 | { | |
875 | int i; | |
876 | for (i = 0; i < outputsamples; i++) | |
877 | { | |
878 | int32_t audiobits; | |
879 | ||
880 | audiobits = readbits(alac, 16); | |
881 | /* special case of sign extension.. | |
882 | * as we'll be ORing the low 16bits into this */ | |
883 | audiobits = audiobits << (alac->setinfo_sample_size - 16); | |
884 | audiobits |= readbits(alac, alac->setinfo_sample_size - 16); | |
885 | audiobits = SignExtend24(audiobits); | |
886 | ||
887 | alac->outputsamples_buffer_a[i] = audiobits; | |
888 | } | |
889 | } | |
890 | uncompressed_bytes = 0; // always 0 for uncompressed | |
891 | } | |
892 | ||
893 | switch(alac->setinfo_sample_size) | |
894 | { | |
895 | case 16: | |
896 | { | |
897 | int i; | |
898 | for (i = 0; i < outputsamples; i++) | |
899 | { | |
900 | int16_t sample = alac->outputsamples_buffer_a[i]; | |
901 | if (host_bigendian) | |
902 | _Swap16(sample); | |
903 | ((int16_t*)outbuffer)[i * alac->numchannels] = sample; | |
904 | } | |
905 | break; | |
906 | } | |
907 | case 24: | |
908 | { | |
909 | int i; | |
910 | for (i = 0; i < outputsamples; i++) | |
911 | { | |
912 | int32_t sample = alac->outputsamples_buffer_a[i]; | |
913 | ||
914 | if (uncompressed_bytes) | |
915 | { | |
916 | uint32_t mask; | |
917 | sample = sample << (uncompressed_bytes * 8); | |
918 | mask = ~(0xFFFFFFFF << (uncompressed_bytes * 8)); | |
919 | sample |= alac->uncompressed_bytes_buffer_a[i] & mask; | |
920 | } | |
921 | ||
922 | ((uint8_t*)outbuffer)[i * alac->numchannels * 3] = (sample) & 0xFF; | |
923 | ((uint8_t*)outbuffer)[i * alac->numchannels * 3 + 1] = (sample >> 8) & 0xFF; | |
924 | ((uint8_t*)outbuffer)[i * alac->numchannels * 3 + 2] = (sample >> 16) & 0xFF; | |
925 | } | |
926 | break; | |
927 | } | |
928 | case 20: | |
929 | case 32: | |
930 | fprintf(stderr, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size); | |
931 | break; | |
932 | default: | |
933 | break; | |
934 | } | |
935 | break; | |
936 | } | |
937 | case 1: /* 2 channels */ | |
938 | { | |
939 | int hassize; | |
940 | int isnotcompressed; | |
941 | int readsamplesize; | |
942 | ||
943 | int uncompressed_bytes; | |
944 | ||
945 | uint8_t interlacing_shift; | |
946 | uint8_t interlacing_leftweight; | |
947 | ||
948 | /* 2^result = something to do with output waiting. | |
949 | * perhaps matters if we read > 1 frame in a pass? | |
950 | */ | |
951 | readbits(alac, 4); | |
952 | ||
953 | readbits(alac, 12); /* unknown, skip 12 bits */ | |
954 | ||
955 | hassize = readbits(alac, 1); /* the output sample size is stored soon */ | |
956 | ||
957 | uncompressed_bytes = readbits(alac, 2); /* the number of bytes in the (compressed) stream that are not compressed */ | |
958 | ||
959 | isnotcompressed = readbits(alac, 1); /* whether the frame is compressed */ | |
960 | ||
961 | if (hassize) | |
962 | { | |
963 | /* now read the number of samples, | |
964 | * as a 32bit integer */ | |
965 | outputsamples = readbits(alac, 32); | |
966 | *outputsize = outputsamples * alac->bytespersample; | |
967 | } | |
968 | ||
969 | readsamplesize = alac->setinfo_sample_size - (uncompressed_bytes * 8) + 1; | |
970 | ||
971 | if (!isnotcompressed) | |
972 | { /* compressed */ | |
973 | int16_t predictor_coef_table_a[32]; | |
974 | int predictor_coef_num_a; | |
975 | int prediction_type_a; | |
976 | int prediction_quantitization_a; | |
977 | int ricemodifier_a; | |
978 | ||
979 | int16_t predictor_coef_table_b[32]; | |
980 | int predictor_coef_num_b; | |
981 | int prediction_type_b; | |
982 | int prediction_quantitization_b; | |
983 | int ricemodifier_b; | |
984 | ||
985 | int i; | |
986 | ||
987 | interlacing_shift = readbits(alac, 8); | |
988 | interlacing_leftweight = readbits(alac, 8); | |
989 | ||
990 | /******** channel 1 ***********/ | |
991 | prediction_type_a = readbits(alac, 4); | |
992 | prediction_quantitization_a = readbits(alac, 4); | |
993 | ||
994 | ricemodifier_a = readbits(alac, 3); | |
995 | predictor_coef_num_a = readbits(alac, 5); | |
996 | ||
997 | /* read the predictor table */ | |
998 | for (i = 0; i < predictor_coef_num_a; i++) | |
999 | { | |
1000 | predictor_coef_table_a[i] = (int16_t)readbits(alac, 16); | |
1001 | } | |
1002 | ||
1003 | /******** channel 2 *********/ | |
1004 | prediction_type_b = readbits(alac, 4); | |
1005 | prediction_quantitization_b = readbits(alac, 4); | |
1006 | ||
1007 | ricemodifier_b = readbits(alac, 3); | |
1008 | predictor_coef_num_b = readbits(alac, 5); | |
1009 | ||
1010 | /* read the predictor table */ | |
1011 | for (i = 0; i < predictor_coef_num_b; i++) | |
1012 | { | |
1013 | predictor_coef_table_b[i] = (int16_t)readbits(alac, 16); | |
1014 | } | |
1015 | ||
1016 | /*********************/ | |
1017 | if (uncompressed_bytes) | |
1018 | { /* see mono case */ | |
1019 | int i; | |
1020 | for (i = 0; i < outputsamples; i++) | |
1021 | { | |
1022 | alac->uncompressed_bytes_buffer_a[i] = readbits(alac, uncompressed_bytes * 8); | |
1023 | alac->uncompressed_bytes_buffer_b[i] = readbits(alac, uncompressed_bytes * 8); | |
1024 | } | |
1025 | } | |
1026 | ||
1027 | /* channel 1 */ | |
1028 | entropy_rice_decode(alac, | |
1029 | alac->predicterror_buffer_a, | |
1030 | outputsamples, | |
1031 | readsamplesize, | |
1032 | alac->setinfo_rice_initialhistory, | |
1033 | alac->setinfo_rice_kmodifier, | |
1034 | ricemodifier_a * alac->setinfo_rice_historymult / 4, | |
1035 | (1 << alac->setinfo_rice_kmodifier) - 1); | |
1036 | ||
1037 | if (prediction_type_a == 0) | |
1038 | { /* adaptive fir */ | |
1039 | predictor_decompress_fir_adapt(alac->predicterror_buffer_a, | |
1040 | alac->outputsamples_buffer_a, | |
1041 | outputsamples, | |
1042 | readsamplesize, | |
1043 | predictor_coef_table_a, | |
1044 | predictor_coef_num_a, | |
1045 | prediction_quantitization_a); | |
1046 | } | |
1047 | else | |
1048 | { /* see mono case */ | |
1049 | fprintf(stderr, "FIXME: unhandled predicition type: %i\n", prediction_type_a); | |
1050 | } | |
1051 | ||
1052 | /* channel 2 */ | |
1053 | entropy_rice_decode(alac, | |
1054 | alac->predicterror_buffer_b, | |
1055 | outputsamples, | |
1056 | readsamplesize, | |
1057 | alac->setinfo_rice_initialhistory, | |
1058 | alac->setinfo_rice_kmodifier, | |
1059 | ricemodifier_b * alac->setinfo_rice_historymult / 4, | |
1060 | (1 << alac->setinfo_rice_kmodifier) - 1); | |
1061 | ||
1062 | if (prediction_type_b == 0) | |
1063 | { /* adaptive fir */ | |
1064 | predictor_decompress_fir_adapt(alac->predicterror_buffer_b, | |
1065 | alac->outputsamples_buffer_b, | |
1066 | outputsamples, | |
1067 | readsamplesize, | |
1068 | predictor_coef_table_b, | |
1069 | predictor_coef_num_b, | |
1070 | prediction_quantitization_b); | |
1071 | } | |
1072 | else | |
1073 | { | |
1074 | fprintf(stderr, "FIXME: unhandled predicition type: %i\n", prediction_type_b); | |
1075 | } | |
1076 | } | |
1077 | else | |
1078 | { /* not compressed, easy case */ | |
1079 | if (alac->setinfo_sample_size <= 16) | |
1080 | { | |
1081 | int i; | |
1082 | for (i = 0; i < outputsamples; i++) | |
1083 | { | |
1084 | int32_t audiobits_a, audiobits_b; | |
1085 | ||
1086 | audiobits_a = readbits(alac, alac->setinfo_sample_size); | |
1087 | audiobits_b = readbits(alac, alac->setinfo_sample_size); | |
1088 | ||
1089 | audiobits_a = SIGN_EXTENDED32(audiobits_a, alac->setinfo_sample_size); | |
1090 | audiobits_b = SIGN_EXTENDED32(audiobits_b, alac->setinfo_sample_size); | |
1091 | ||
1092 | alac->outputsamples_buffer_a[i] = audiobits_a; | |
1093 | alac->outputsamples_buffer_b[i] = audiobits_b; | |
1094 | } | |
1095 | } | |
1096 | else | |
1097 | { | |
1098 | int i; | |
1099 | for (i = 0; i < outputsamples; i++) | |
1100 | { | |
1101 | int32_t audiobits_a, audiobits_b; | |
1102 | ||
1103 | audiobits_a = readbits(alac, 16); | |
1104 | audiobits_a = audiobits_a << (alac->setinfo_sample_size - 16); | |
1105 | audiobits_a |= readbits(alac, alac->setinfo_sample_size - 16); | |
1106 | audiobits_a = SignExtend24(audiobits_a); | |
1107 | ||
1108 | audiobits_b = readbits(alac, 16); | |
1109 | audiobits_b = audiobits_b << (alac->setinfo_sample_size - 16); | |
1110 | audiobits_b |= readbits(alac, alac->setinfo_sample_size - 16); | |
1111 | audiobits_b = SignExtend24(audiobits_b); | |
1112 | ||
1113 | alac->outputsamples_buffer_a[i] = audiobits_a; | |
1114 | alac->outputsamples_buffer_b[i] = audiobits_b; | |
1115 | } | |
1116 | } | |
1117 | uncompressed_bytes = 0; // always 0 for uncompressed | |
1118 | interlacing_shift = 0; | |
1119 | interlacing_leftweight = 0; | |
1120 | } | |
1121 | ||
1122 | switch(alac->setinfo_sample_size) | |
1123 | { | |
1124 | case 16: | |
1125 | { | |
1126 | deinterlace_16(alac->outputsamples_buffer_a, | |
1127 | alac->outputsamples_buffer_b, | |
1128 | (int16_t*)outbuffer, | |
1129 | alac->numchannels, | |
1130 | outputsamples, | |
1131 | interlacing_shift, | |
1132 | interlacing_leftweight); | |
1133 | break; | |
1134 | } | |
1135 | case 24: | |
1136 | { | |
1137 | deinterlace_24(alac->outputsamples_buffer_a, | |
1138 | alac->outputsamples_buffer_b, | |
1139 | uncompressed_bytes, | |
1140 | alac->uncompressed_bytes_buffer_a, | |
1141 | alac->uncompressed_bytes_buffer_b, | |
1142 | (int16_t*)outbuffer, | |
1143 | alac->numchannels, | |
1144 | outputsamples, | |
1145 | interlacing_shift, | |
1146 | interlacing_leftweight); | |
1147 | break; | |
1148 | } | |
1149 | case 20: | |
1150 | case 32: | |
1151 | fprintf(stderr, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size); | |
1152 | break; | |
1153 | default: | |
1154 | break; | |
1155 | } | |
1156 | ||
1157 | break; | |
1158 | } | |
1159 | } | |
1160 | } | |
1161 | ||
1162 | alac_file *create_alac(int samplesize, int numchannels) | |
1163 | { | |
1164 | alac_file *newfile = malloc(sizeof(alac_file)); | |
1165 | ||
1166 | newfile->samplesize = samplesize; | |
1167 | newfile->numchannels = numchannels; | |
1168 | newfile->bytespersample = (samplesize / 8) * numchannels; | |
1169 | ||
1170 | return newfile; | |
1171 | } | |
1172 |