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 | ||
982a0c26 JVH |
114 | static void deallocate_buffers(alac_file *alac) |
115 | { | |
116 | free(alac->predicterror_buffer_a); | |
117 | free(alac->predicterror_buffer_b); | |
118 | alac->predicterror_buffer_a = NULL; | |
119 | alac->predicterror_buffer_b = NULL; | |
120 | ||
121 | free(alac->outputsamples_buffer_a); | |
122 | free(alac->outputsamples_buffer_b); | |
123 | alac->outputsamples_buffer_a = NULL; | |
124 | alac->outputsamples_buffer_b = NULL; | |
125 | ||
126 | free(alac->uncompressed_bytes_buffer_a); | |
127 | free(alac->uncompressed_bytes_buffer_b); | |
128 | alac->uncompressed_bytes_buffer_a = NULL; | |
129 | alac->uncompressed_bytes_buffer_b = NULL; | |
130 | } | |
131 | ||
2340bcd3 JVH |
132 | void alac_set_info(alac_file *alac, char *inputbuffer) |
133 | { | |
134 | char *ptr = inputbuffer; | |
135 | ptr += 4; /* size */ | |
136 | ptr += 4; /* frma */ | |
137 | ptr += 4; /* alac */ | |
138 | ptr += 4; /* size */ | |
139 | ptr += 4; /* alac */ | |
140 | ||
141 | ptr += 4; /* 0 ? */ | |
142 | ||
143 | alac->setinfo_max_samples_per_frame = *(uint32_t*)ptr; /* buffer size / 2 ? */ | |
144 | if (!host_bigendian) | |
145 | _Swap32(alac->setinfo_max_samples_per_frame); | |
146 | ptr += 4; | |
147 | alac->setinfo_7a = *(uint8_t*)ptr; | |
148 | ptr += 1; | |
149 | alac->setinfo_sample_size = *(uint8_t*)ptr; | |
150 | ptr += 1; | |
151 | alac->setinfo_rice_historymult = *(uint8_t*)ptr; | |
152 | ptr += 1; | |
153 | alac->setinfo_rice_initialhistory = *(uint8_t*)ptr; | |
154 | ptr += 1; | |
155 | alac->setinfo_rice_kmodifier = *(uint8_t*)ptr; | |
156 | ptr += 1; | |
157 | alac->setinfo_7f = *(uint8_t*)ptr; | |
158 | ptr += 1; | |
159 | alac->setinfo_80 = *(uint16_t*)ptr; | |
160 | if (!host_bigendian) | |
161 | _Swap16(alac->setinfo_80); | |
162 | ptr += 2; | |
163 | alac->setinfo_82 = *(uint32_t*)ptr; | |
164 | if (!host_bigendian) | |
165 | _Swap32(alac->setinfo_82); | |
166 | ptr += 4; | |
167 | alac->setinfo_86 = *(uint32_t*)ptr; | |
168 | if (!host_bigendian) | |
169 | _Swap32(alac->setinfo_86); | |
170 | ptr += 4; | |
171 | alac->setinfo_8a_rate = *(uint32_t*)ptr; | |
172 | if (!host_bigendian) | |
173 | _Swap32(alac->setinfo_8a_rate); | |
174 | ptr += 4; | |
175 | ||
176 | allocate_buffers(alac); | |
177 | ||
178 | } | |
179 | ||
180 | /* stream reading */ | |
181 | ||
182 | /* supports reading 1 to 16 bits, in big endian format */ | |
183 | static uint32_t readbits_16(alac_file *alac, int bits) | |
184 | { | |
185 | uint32_t result; | |
186 | int new_accumulator; | |
187 | ||
188 | result = (alac->input_buffer[0] << 16) | | |
189 | (alac->input_buffer[1] << 8) | | |
190 | (alac->input_buffer[2]); | |
191 | ||
192 | /* shift left by the number of bits we've already read, | |
193 | * so that the top 'n' bits of the 24 bits we read will | |
194 | * be the return bits */ | |
195 | result = result << alac->input_buffer_bitaccumulator; | |
196 | ||
197 | result = result & 0x00ffffff; | |
198 | ||
199 | /* and then only want the top 'n' bits from that, where | |
200 | * n is 'bits' */ | |
201 | result = result >> (24 - bits); | |
202 | ||
203 | new_accumulator = (alac->input_buffer_bitaccumulator + bits); | |
204 | ||
205 | /* increase the buffer pointer if we've read over n bytes. */ | |
206 | alac->input_buffer += (new_accumulator >> 3); | |
207 | ||
208 | /* and the remainder goes back into the bit accumulator */ | |
209 | alac->input_buffer_bitaccumulator = (new_accumulator & 7); | |
210 | ||
211 | return result; | |
212 | } | |
213 | ||
214 | /* supports reading 1 to 32 bits, in big endian format */ | |
215 | static uint32_t readbits(alac_file *alac, int bits) | |
216 | { | |
217 | int32_t result = 0; | |
218 | ||
219 | if (bits > 16) | |
220 | { | |
221 | bits -= 16; | |
222 | result = readbits_16(alac, 16) << bits; | |
223 | } | |
224 | ||
225 | result |= readbits_16(alac, bits); | |
226 | ||
227 | return result; | |
228 | } | |
229 | ||
230 | /* reads a single bit */ | |
231 | static int readbit(alac_file *alac) | |
232 | { | |
233 | int result; | |
234 | int new_accumulator; | |
235 | ||
236 | result = alac->input_buffer[0]; | |
237 | ||
238 | result = result << alac->input_buffer_bitaccumulator; | |
239 | ||
240 | result = result >> 7 & 1; | |
241 | ||
242 | new_accumulator = (alac->input_buffer_bitaccumulator + 1); | |
243 | ||
244 | alac->input_buffer += (new_accumulator / 8); | |
245 | ||
246 | alac->input_buffer_bitaccumulator = (new_accumulator % 8); | |
247 | ||
248 | return result; | |
249 | } | |
250 | ||
251 | static void unreadbits(alac_file *alac, int bits) | |
252 | { | |
253 | int new_accumulator = (alac->input_buffer_bitaccumulator - bits); | |
254 | ||
255 | alac->input_buffer += (new_accumulator >> 3); | |
256 | ||
257 | alac->input_buffer_bitaccumulator = (new_accumulator & 7); | |
258 | if (alac->input_buffer_bitaccumulator < 0) | |
259 | alac->input_buffer_bitaccumulator *= -1; | |
260 | } | |
261 | ||
262 | /* various implementations of count_leading_zero: | |
263 | * the first one is the original one, the simplest and most | |
264 | * obvious for what it's doing. never use this. | |
265 | * then there are the asm ones. fill in as necessary | |
266 | * and finally an unrolled and optimised c version | |
267 | * to fall back to | |
268 | */ | |
269 | #if 0 | |
270 | /* hideously inefficient. could use a bitmask search, | |
271 | * alternatively bsr on x86, | |
272 | */ | |
273 | static int count_leading_zeros(int32_t input) | |
274 | { | |
275 | int i = 0; | |
276 | while (!(0x80000000 & input) && i < 32) | |
277 | { | |
278 | i++; | |
279 | input = input << 1; | |
280 | } | |
281 | return i; | |
282 | } | |
283 | #elif defined(__GNUC__) && (defined(_X86) || defined(__i386) || defined(i386)) | |
284 | /* for some reason the unrolled version (below) is | |
285 | * actually faster than this. yay intel! | |
286 | */ | |
287 | static int count_leading_zeros(int input) | |
288 | { | |
289 | int output = 0; | |
290 | if (!input) return 32; | |
291 | __asm("bsr %1, %0\n" | |
292 | : "=r" (output) | |
293 | : "r" (input)); | |
294 | return (0x1f - output); | |
295 | } | |
296 | #elif defined(__GNUC__) | |
297 | static int count_leading_zeros(int input) | |
298 | { | |
299 | return __builtin_clz(input); | |
300 | } | |
301 | #elif defined(_MSC_VER) && defined(_M_IX86) | |
302 | static int count_leading_zeros(int input) | |
303 | { | |
304 | int output = 0; | |
305 | if (!input) return 32; | |
306 | __asm | |
307 | { | |
308 | mov eax, input; | |
309 | mov edx, 0x1f; | |
310 | bsr ecx, eax; | |
311 | sub edx, ecx; | |
312 | mov output, edx; | |
313 | } | |
314 | return output; | |
315 | } | |
316 | #else | |
317 | #warning using generic count leading zeroes. You may wish to write one for your CPU / compiler | |
318 | static int count_leading_zeros(int input) | |
319 | { | |
320 | int output = 0; | |
321 | int curbyte = 0; | |
322 | ||
323 | curbyte = input >> 24; | |
324 | if (curbyte) goto found; | |
325 | output += 8; | |
326 | ||
327 | curbyte = input >> 16; | |
328 | if (curbyte & 0xff) goto found; | |
329 | output += 8; | |
330 | ||
331 | curbyte = input >> 8; | |
332 | if (curbyte & 0xff) goto found; | |
333 | output += 8; | |
334 | ||
335 | curbyte = input; | |
336 | if (curbyte & 0xff) goto found; | |
337 | output += 8; | |
338 | ||
339 | return output; | |
340 | ||
341 | found: | |
342 | if (!(curbyte & 0xf0)) | |
343 | { | |
344 | output += 4; | |
345 | } | |
346 | else | |
347 | curbyte >>= 4; | |
348 | ||
349 | if (curbyte & 0x8) | |
350 | return output; | |
351 | if (curbyte & 0x4) | |
352 | return output + 1; | |
353 | if (curbyte & 0x2) | |
354 | return output + 2; | |
355 | if (curbyte & 0x1) | |
356 | return output + 3; | |
357 | ||
358 | /* shouldn't get here: */ | |
359 | return output + 4; | |
360 | } | |
361 | #endif | |
362 | ||
363 | #define RICE_THRESHOLD 8 // maximum number of bits for a rice prefix. | |
364 | ||
365 | int32_t entropy_decode_value(alac_file* alac, | |
366 | int readSampleSize, | |
367 | int k, | |
368 | int rice_kmodifier_mask) | |
369 | { | |
370 | int32_t x = 0; // decoded value | |
371 | ||
372 | // read x, number of 1s before 0 represent the rice value. | |
373 | while (x <= RICE_THRESHOLD && readbit(alac)) | |
374 | { | |
375 | x++; | |
376 | } | |
377 | ||
378 | if (x > RICE_THRESHOLD) | |
379 | { | |
380 | // read the number from the bit stream (raw value) | |
381 | int32_t value; | |
382 | ||
383 | value = readbits(alac, readSampleSize); | |
384 | ||
385 | // mask value | |
386 | value &= (((uint32_t)0xffffffff) >> (32 - readSampleSize)); | |
387 | ||
388 | x = value; | |
389 | } | |
390 | else | |
391 | { | |
392 | if (k != 1) | |
393 | { | |
394 | int extraBits = readbits(alac, k); | |
395 | ||
396 | // x = x * (2^k - 1) | |
397 | x *= (((1 << k) - 1) & rice_kmodifier_mask); | |
398 | ||
399 | if (extraBits > 1) | |
400 | x += extraBits - 1; | |
401 | else | |
402 | unreadbits(alac, 1); | |
403 | } | |
404 | } | |
405 | ||
406 | return x; | |
407 | } | |
408 | ||
409 | void entropy_rice_decode(alac_file* alac, | |
410 | int32_t* outputBuffer, | |
411 | int outputSize, | |
412 | int readSampleSize, | |
413 | int rice_initialhistory, | |
414 | int rice_kmodifier, | |
415 | int rice_historymult, | |
416 | int rice_kmodifier_mask) | |
417 | { | |
418 | int outputCount; | |
419 | int history = rice_initialhistory; | |
420 | int signModifier = 0; | |
421 | ||
422 | for (outputCount = 0; outputCount < outputSize; outputCount++) | |
423 | { | |
424 | int32_t decodedValue; | |
425 | int32_t finalValue; | |
426 | int32_t k; | |
427 | ||
428 | k = 31 - rice_kmodifier - count_leading_zeros((history >> 9) + 3); | |
429 | ||
430 | if (k < 0) k += rice_kmodifier; | |
431 | else k = rice_kmodifier; | |
432 | ||
433 | // note: don't use rice_kmodifier_mask here (set mask to 0xFFFFFFFF) | |
434 | decodedValue = entropy_decode_value(alac, readSampleSize, k, 0xFFFFFFFF); | |
435 | ||
436 | decodedValue += signModifier; | |
437 | finalValue = (decodedValue + 1) / 2; // inc by 1 and shift out sign bit | |
438 | if (decodedValue & 1) // the sign is stored in the low bit | |
439 | finalValue *= -1; | |
440 | ||
441 | outputBuffer[outputCount] = finalValue; | |
442 | ||
443 | signModifier = 0; | |
444 | ||
445 | // update history | |
446 | history += (decodedValue * rice_historymult) | |
447 | - ((history * rice_historymult) >> 9); | |
448 | ||
449 | if (decodedValue > 0xFFFF) | |
450 | history = 0xFFFF; | |
451 | ||
452 | // special case, for compressed blocks of 0 | |
453 | if ((history < 128) && (outputCount + 1 < outputSize)) | |
454 | { | |
455 | int32_t blockSize; | |
456 | ||
457 | signModifier = 1; | |
458 | ||
459 | k = count_leading_zeros(history) + ((history + 16) / 64) - 24; | |
460 | ||
461 | // note: blockSize is always 16bit | |
462 | blockSize = entropy_decode_value(alac, 16, k, rice_kmodifier_mask); | |
463 | ||
464 | // got blockSize 0s | |
465 | if (blockSize > 0) | |
466 | { | |
467 | memset(&outputBuffer[outputCount + 1], 0, blockSize * sizeof(*outputBuffer)); | |
468 | outputCount += blockSize; | |
469 | } | |
470 | ||
471 | if (blockSize > 0xFFFF) | |
472 | signModifier = 0; | |
473 | ||
474 | history = 0; | |
475 | } | |
476 | } | |
477 | } | |
478 | ||
479 | #define SIGN_EXTENDED32(val, bits) ((val << (32 - bits)) >> (32 - bits)) | |
480 | ||
481 | #define SIGN_ONLY(v) \ | |
482 | ((v < 0) ? (-1) : \ | |
483 | ((v > 0) ? (1) : \ | |
484 | (0))) | |
485 | ||
486 | static void predictor_decompress_fir_adapt(int32_t *error_buffer, | |
487 | int32_t *buffer_out, | |
488 | int output_size, | |
489 | int readsamplesize, | |
490 | int16_t *predictor_coef_table, | |
491 | int predictor_coef_num, | |
492 | int predictor_quantitization) | |
493 | { | |
494 | int i; | |
495 | ||
496 | /* first sample always copies */ | |
497 | *buffer_out = *error_buffer; | |
498 | ||
499 | if (!predictor_coef_num) | |
500 | { | |
501 | if (output_size <= 1) return; | |
502 | memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4); | |
503 | return; | |
504 | } | |
505 | ||
506 | if (predictor_coef_num == 0x1f) /* 11111 - max value of predictor_coef_num */ | |
507 | { /* second-best case scenario for fir decompression, | |
508 | * error describes a small difference from the previous sample only | |
509 | */ | |
510 | if (output_size <= 1) return; | |
511 | for (i = 0; i < output_size - 1; i++) | |
512 | { | |
513 | int32_t prev_value; | |
514 | int32_t error_value; | |
515 | ||
516 | prev_value = buffer_out[i]; | |
517 | error_value = error_buffer[i+1]; | |
518 | buffer_out[i+1] = SIGN_EXTENDED32((prev_value + error_value), readsamplesize); | |
519 | } | |
520 | return; | |
521 | } | |
522 | ||
523 | /* read warm-up samples */ | |
524 | if (predictor_coef_num > 0) | |
525 | { | |
526 | int i; | |
527 | for (i = 0; i < predictor_coef_num; i++) | |
528 | { | |
529 | int32_t val; | |
530 | ||
531 | val = buffer_out[i] + error_buffer[i+1]; | |
532 | ||
533 | val = SIGN_EXTENDED32(val, readsamplesize); | |
534 | ||
535 | buffer_out[i+1] = val; | |
536 | } | |
537 | } | |
538 | ||
539 | #if 0 | |
540 | /* 4 and 8 are very common cases (the only ones i've seen). these | |
541 | * should be unrolled and optimised | |
542 | */ | |
543 | if (predictor_coef_num == 4) | |
544 | { | |
545 | /* FIXME: optimised general case */ | |
546 | return; | |
547 | } | |
548 | ||
549 | if (predictor_coef_table == 8) | |
550 | { | |
551 | /* FIXME: optimised general case */ | |
552 | return; | |
553 | } | |
554 | #endif | |
555 | ||
556 | ||
557 | /* general case */ | |
558 | if (predictor_coef_num > 0) | |
559 | { | |
560 | for (i = predictor_coef_num + 1; | |
561 | i < output_size; | |
562 | i++) | |
563 | { | |
564 | int j; | |
565 | int sum = 0; | |
566 | int outval; | |
567 | int error_val = error_buffer[i]; | |
568 | ||
569 | for (j = 0; j < predictor_coef_num; j++) | |
570 | { | |
571 | sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) * | |
572 | predictor_coef_table[j]; | |
573 | } | |
574 | ||
575 | outval = (1 << (predictor_quantitization-1)) + sum; | |
576 | outval = outval >> predictor_quantitization; | |
577 | outval = outval + buffer_out[0] + error_val; | |
578 | outval = SIGN_EXTENDED32(outval, readsamplesize); | |
579 | ||
580 | buffer_out[predictor_coef_num+1] = outval; | |
581 | ||
582 | if (error_val > 0) | |
583 | { | |
584 | int predictor_num = predictor_coef_num - 1; | |
585 | ||
586 | while (predictor_num >= 0 && error_val > 0) | |
587 | { | |
588 | int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num]; | |
589 | int sign = SIGN_ONLY(val); | |
590 | ||
591 | predictor_coef_table[predictor_num] -= sign; | |
592 | ||
593 | val *= sign; /* absolute value */ | |
594 | ||
595 | error_val -= ((val >> predictor_quantitization) * | |
596 | (predictor_coef_num - predictor_num)); | |
597 | ||
598 | predictor_num--; | |
599 | } | |
600 | } | |
601 | else if (error_val < 0) | |
602 | { | |
603 | int predictor_num = predictor_coef_num - 1; | |
604 | ||
605 | while (predictor_num >= 0 && error_val < 0) | |
606 | { | |
607 | int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num]; | |
608 | int sign = - SIGN_ONLY(val); | |
609 | ||
610 | predictor_coef_table[predictor_num] -= sign; | |
611 | ||
612 | val *= sign; /* neg value */ | |
613 | ||
614 | error_val -= ((val >> predictor_quantitization) * | |
615 | (predictor_coef_num - predictor_num)); | |
616 | ||
617 | predictor_num--; | |
618 | } | |
619 | } | |
620 | ||
621 | buffer_out++; | |
622 | } | |
623 | } | |
624 | } | |
625 | ||
626 | void deinterlace_16(int32_t *buffer_a, int32_t *buffer_b, | |
627 | int16_t *buffer_out, | |
628 | int numchannels, int numsamples, | |
629 | uint8_t interlacing_shift, | |
630 | uint8_t interlacing_leftweight) | |
631 | { | |
632 | int i; | |
633 | if (numsamples <= 0) return; | |
634 | ||
635 | /* weighted interlacing */ | |
636 | if (interlacing_leftweight) | |
637 | { | |
638 | for (i = 0; i < numsamples; i++) | |
639 | { | |
640 | int32_t difference, midright; | |
641 | int16_t left; | |
642 | int16_t right; | |
643 | ||
644 | midright = buffer_a[i]; | |
645 | difference = buffer_b[i]; | |
646 | ||
647 | ||
648 | right = midright - ((difference * interlacing_leftweight) >> interlacing_shift); | |
649 | left = right + difference; | |
650 | ||
651 | /* output is always little endian */ | |
652 | if (host_bigendian) | |
653 | { | |
654 | _Swap16(left); | |
655 | _Swap16(right); | |
656 | } | |
657 | ||
658 | buffer_out[i*numchannels] = left; | |
659 | buffer_out[i*numchannels + 1] = right; | |
660 | } | |
661 | ||
662 | return; | |
663 | } | |
664 | ||
665 | /* otherwise basic interlacing took place */ | |
666 | for (i = 0; i < numsamples; i++) | |
667 | { | |
668 | int16_t left, right; | |
669 | ||
670 | left = buffer_a[i]; | |
671 | right = buffer_b[i]; | |
672 | ||
673 | /* output is always little endian */ | |
674 | if (host_bigendian) | |
675 | { | |
676 | _Swap16(left); | |
677 | _Swap16(right); | |
678 | } | |
679 | ||
680 | buffer_out[i*numchannels] = left; | |
681 | buffer_out[i*numchannels + 1] = right; | |
682 | } | |
683 | } | |
684 | ||
685 | void deinterlace_24(int32_t *buffer_a, int32_t *buffer_b, | |
686 | int uncompressed_bytes, | |
687 | int32_t *uncompressed_bytes_buffer_a, int32_t *uncompressed_bytes_buffer_b, | |
688 | void *buffer_out, | |
689 | int numchannels, int numsamples, | |
690 | uint8_t interlacing_shift, | |
691 | uint8_t interlacing_leftweight) | |
692 | { | |
693 | int i; | |
694 | if (numsamples <= 0) return; | |
695 | ||
696 | /* weighted interlacing */ | |
697 | if (interlacing_leftweight) | |
698 | { | |
699 | for (i = 0; i < numsamples; i++) | |
700 | { | |
701 | int32_t difference, midright; | |
702 | int32_t left; | |
703 | int32_t right; | |
704 | ||
705 | midright = buffer_a[i]; | |
706 | difference = buffer_b[i]; | |
707 | ||
708 | right = midright - ((difference * interlacing_leftweight) >> interlacing_shift); | |
709 | left = right + difference; | |
710 | ||
711 | if (uncompressed_bytes) | |
712 | { | |
713 | uint32_t mask = ~(0xFFFFFFFF << (uncompressed_bytes * 8)); | |
714 | left <<= (uncompressed_bytes * 8); | |
715 | right <<= (uncompressed_bytes * 8); | |
716 | ||
717 | left |= uncompressed_bytes_buffer_a[i] & mask; | |
718 | right |= uncompressed_bytes_buffer_b[i] & mask; | |
719 | } | |
720 | ||
721 | ((uint8_t*)buffer_out)[i * numchannels * 3] = (left) & 0xFF; | |
722 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 1] = (left >> 8) & 0xFF; | |
723 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 2] = (left >> 16) & 0xFF; | |
724 | ||
725 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 3] = (right) & 0xFF; | |
726 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 4] = (right >> 8) & 0xFF; | |
727 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 5] = (right >> 16) & 0xFF; | |
728 | } | |
729 | ||
730 | return; | |
731 | } | |
732 | ||
733 | /* otherwise basic interlacing took place */ | |
734 | for (i = 0; i < numsamples; i++) | |
735 | { | |
736 | int32_t left, right; | |
737 | ||
738 | left = buffer_a[i]; | |
739 | right = buffer_b[i]; | |
740 | ||
741 | if (uncompressed_bytes) | |
742 | { | |
743 | uint32_t mask = ~(0xFFFFFFFF << (uncompressed_bytes * 8)); | |
744 | left <<= (uncompressed_bytes * 8); | |
745 | right <<= (uncompressed_bytes * 8); | |
746 | ||
747 | left |= uncompressed_bytes_buffer_a[i] & mask; | |
748 | right |= uncompressed_bytes_buffer_b[i] & mask; | |
749 | } | |
750 | ||
751 | ((uint8_t*)buffer_out)[i * numchannels * 3] = (left) & 0xFF; | |
752 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 1] = (left >> 8) & 0xFF; | |
753 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 2] = (left >> 16) & 0xFF; | |
754 | ||
755 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 3] = (right) & 0xFF; | |
756 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 4] = (right >> 8) & 0xFF; | |
757 | ((uint8_t*)buffer_out)[i * numchannels * 3 + 5] = (right >> 16) & 0xFF; | |
758 | ||
759 | } | |
760 | ||
761 | } | |
762 | ||
763 | void decode_frame(alac_file *alac, | |
764 | unsigned char *inbuffer, | |
765 | void *outbuffer, int *outputsize) | |
766 | { | |
767 | int channels; | |
768 | int32_t outputsamples = alac->setinfo_max_samples_per_frame; | |
769 | ||
770 | /* setup the stream */ | |
771 | alac->input_buffer = inbuffer; | |
772 | alac->input_buffer_bitaccumulator = 0; | |
773 | ||
774 | channels = readbits(alac, 3); | |
775 | ||
776 | *outputsize = outputsamples * alac->bytespersample; | |
777 | ||
778 | switch(channels) | |
779 | { | |
780 | case 0: /* 1 channel */ | |
781 | { | |
782 | int hassize; | |
783 | int isnotcompressed; | |
784 | int readsamplesize; | |
785 | ||
786 | int uncompressed_bytes; | |
787 | int ricemodifier; | |
788 | ||
789 | /* 2^result = something to do with output waiting. | |
790 | * perhaps matters if we read > 1 frame in a pass? | |
791 | */ | |
792 | readbits(alac, 4); | |
793 | ||
794 | readbits(alac, 12); /* unknown, skip 12 bits */ | |
795 | ||
796 | hassize = readbits(alac, 1); /* the output sample size is stored soon */ | |
797 | ||
798 | uncompressed_bytes = readbits(alac, 2); /* number of bytes in the (compressed) stream that are not compressed */ | |
799 | ||
800 | isnotcompressed = readbits(alac, 1); /* whether the frame is compressed */ | |
801 | ||
802 | if (hassize) | |
803 | { | |
804 | /* now read the number of samples, | |
805 | * as a 32bit integer */ | |
806 | outputsamples = readbits(alac, 32); | |
807 | *outputsize = outputsamples * alac->bytespersample; | |
808 | } | |
809 | ||
810 | readsamplesize = alac->setinfo_sample_size - (uncompressed_bytes * 8); | |
811 | ||
812 | if (!isnotcompressed) | |
813 | { /* so it is compressed */ | |
814 | int16_t predictor_coef_table[32]; | |
815 | int predictor_coef_num; | |
816 | int prediction_type; | |
817 | int prediction_quantitization; | |
818 | int i; | |
819 | ||
820 | /* skip 16 bits, not sure what they are. seem to be used in | |
821 | * two channel case */ | |
822 | readbits(alac, 8); | |
823 | readbits(alac, 8); | |
824 | ||
825 | prediction_type = readbits(alac, 4); | |
826 | prediction_quantitization = readbits(alac, 4); | |
827 | ||
828 | ricemodifier = readbits(alac, 3); | |
829 | predictor_coef_num = readbits(alac, 5); | |
830 | ||
831 | /* read the predictor table */ | |
832 | for (i = 0; i < predictor_coef_num; i++) | |
833 | { | |
834 | predictor_coef_table[i] = (int16_t)readbits(alac, 16); | |
835 | } | |
836 | ||
837 | if (uncompressed_bytes) | |
838 | { | |
839 | int i; | |
840 | for (i = 0; i < outputsamples; i++) | |
841 | { | |
842 | alac->uncompressed_bytes_buffer_a[i] = readbits(alac, uncompressed_bytes * 8); | |
843 | } | |
844 | } | |
845 | ||
846 | entropy_rice_decode(alac, | |
847 | alac->predicterror_buffer_a, | |
848 | outputsamples, | |
849 | readsamplesize, | |
850 | alac->setinfo_rice_initialhistory, | |
851 | alac->setinfo_rice_kmodifier, | |
852 | ricemodifier * alac->setinfo_rice_historymult / 4, | |
853 | (1 << alac->setinfo_rice_kmodifier) - 1); | |
854 | ||
855 | if (prediction_type == 0) | |
856 | { /* adaptive fir */ | |
857 | predictor_decompress_fir_adapt(alac->predicterror_buffer_a, | |
858 | alac->outputsamples_buffer_a, | |
859 | outputsamples, | |
860 | readsamplesize, | |
861 | predictor_coef_table, | |
862 | predictor_coef_num, | |
863 | prediction_quantitization); | |
864 | } | |
865 | else | |
866 | { | |
867 | fprintf(stderr, "FIXME: unhandled predicition type: %i\n", prediction_type); | |
868 | /* i think the only other prediction type (or perhaps this is just a | |
869 | * boolean?) runs adaptive fir twice.. like: | |
870 | * predictor_decompress_fir_adapt(predictor_error, tempout, ...) | |
871 | * predictor_decompress_fir_adapt(predictor_error, outputsamples ...) | |
872 | * little strange.. | |
873 | */ | |
874 | } | |
875 | ||
876 | } | |
877 | else | |
878 | { /* not compressed, easy case */ | |
879 | if (alac->setinfo_sample_size <= 16) | |
880 | { | |
881 | int i; | |
882 | for (i = 0; i < outputsamples; i++) | |
883 | { | |
884 | int32_t audiobits = readbits(alac, alac->setinfo_sample_size); | |
885 | ||
886 | audiobits = SIGN_EXTENDED32(audiobits, alac->setinfo_sample_size); | |
887 | ||
888 | alac->outputsamples_buffer_a[i] = audiobits; | |
889 | } | |
890 | } | |
891 | else | |
892 | { | |
893 | int i; | |
894 | for (i = 0; i < outputsamples; i++) | |
895 | { | |
896 | int32_t audiobits; | |
897 | ||
898 | audiobits = readbits(alac, 16); | |
899 | /* special case of sign extension.. | |
900 | * as we'll be ORing the low 16bits into this */ | |
901 | audiobits = audiobits << (alac->setinfo_sample_size - 16); | |
902 | audiobits |= readbits(alac, alac->setinfo_sample_size - 16); | |
903 | audiobits = SignExtend24(audiobits); | |
904 | ||
905 | alac->outputsamples_buffer_a[i] = audiobits; | |
906 | } | |
907 | } | |
908 | uncompressed_bytes = 0; // always 0 for uncompressed | |
909 | } | |
910 | ||
911 | switch(alac->setinfo_sample_size) | |
912 | { | |
913 | case 16: | |
914 | { | |
915 | int i; | |
916 | for (i = 0; i < outputsamples; i++) | |
917 | { | |
918 | int16_t sample = alac->outputsamples_buffer_a[i]; | |
919 | if (host_bigendian) | |
920 | _Swap16(sample); | |
921 | ((int16_t*)outbuffer)[i * alac->numchannels] = sample; | |
922 | } | |
923 | break; | |
924 | } | |
925 | case 24: | |
926 | { | |
927 | int i; | |
928 | for (i = 0; i < outputsamples; i++) | |
929 | { | |
930 | int32_t sample = alac->outputsamples_buffer_a[i]; | |
931 | ||
932 | if (uncompressed_bytes) | |
933 | { | |
934 | uint32_t mask; | |
935 | sample = sample << (uncompressed_bytes * 8); | |
936 | mask = ~(0xFFFFFFFF << (uncompressed_bytes * 8)); | |
937 | sample |= alac->uncompressed_bytes_buffer_a[i] & mask; | |
938 | } | |
939 | ||
940 | ((uint8_t*)outbuffer)[i * alac->numchannels * 3] = (sample) & 0xFF; | |
941 | ((uint8_t*)outbuffer)[i * alac->numchannels * 3 + 1] = (sample >> 8) & 0xFF; | |
942 | ((uint8_t*)outbuffer)[i * alac->numchannels * 3 + 2] = (sample >> 16) & 0xFF; | |
943 | } | |
944 | break; | |
945 | } | |
946 | case 20: | |
947 | case 32: | |
948 | fprintf(stderr, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size); | |
949 | break; | |
950 | default: | |
951 | break; | |
952 | } | |
953 | break; | |
954 | } | |
955 | case 1: /* 2 channels */ | |
956 | { | |
957 | int hassize; | |
958 | int isnotcompressed; | |
959 | int readsamplesize; | |
960 | ||
961 | int uncompressed_bytes; | |
962 | ||
963 | uint8_t interlacing_shift; | |
964 | uint8_t interlacing_leftweight; | |
965 | ||
966 | /* 2^result = something to do with output waiting. | |
967 | * perhaps matters if we read > 1 frame in a pass? | |
968 | */ | |
969 | readbits(alac, 4); | |
970 | ||
971 | readbits(alac, 12); /* unknown, skip 12 bits */ | |
972 | ||
973 | hassize = readbits(alac, 1); /* the output sample size is stored soon */ | |
974 | ||
975 | uncompressed_bytes = readbits(alac, 2); /* the number of bytes in the (compressed) stream that are not compressed */ | |
976 | ||
977 | isnotcompressed = readbits(alac, 1); /* whether the frame is compressed */ | |
978 | ||
979 | if (hassize) | |
980 | { | |
981 | /* now read the number of samples, | |
982 | * as a 32bit integer */ | |
983 | outputsamples = readbits(alac, 32); | |
984 | *outputsize = outputsamples * alac->bytespersample; | |
985 | } | |
986 | ||
987 | readsamplesize = alac->setinfo_sample_size - (uncompressed_bytes * 8) + 1; | |
988 | ||
989 | if (!isnotcompressed) | |
990 | { /* compressed */ | |
991 | int16_t predictor_coef_table_a[32]; | |
992 | int predictor_coef_num_a; | |
993 | int prediction_type_a; | |
994 | int prediction_quantitization_a; | |
995 | int ricemodifier_a; | |
996 | ||
997 | int16_t predictor_coef_table_b[32]; | |
998 | int predictor_coef_num_b; | |
999 | int prediction_type_b; | |
1000 | int prediction_quantitization_b; | |
1001 | int ricemodifier_b; | |
1002 | ||
1003 | int i; | |
1004 | ||
1005 | interlacing_shift = readbits(alac, 8); | |
1006 | interlacing_leftweight = readbits(alac, 8); | |
1007 | ||
1008 | /******** channel 1 ***********/ | |
1009 | prediction_type_a = readbits(alac, 4); | |
1010 | prediction_quantitization_a = readbits(alac, 4); | |
1011 | ||
1012 | ricemodifier_a = readbits(alac, 3); | |
1013 | predictor_coef_num_a = readbits(alac, 5); | |
1014 | ||
1015 | /* read the predictor table */ | |
1016 | for (i = 0; i < predictor_coef_num_a; i++) | |
1017 | { | |
1018 | predictor_coef_table_a[i] = (int16_t)readbits(alac, 16); | |
1019 | } | |
1020 | ||
1021 | /******** channel 2 *********/ | |
1022 | prediction_type_b = readbits(alac, 4); | |
1023 | prediction_quantitization_b = readbits(alac, 4); | |
1024 | ||
1025 | ricemodifier_b = readbits(alac, 3); | |
1026 | predictor_coef_num_b = readbits(alac, 5); | |
1027 | ||
1028 | /* read the predictor table */ | |
1029 | for (i = 0; i < predictor_coef_num_b; i++) | |
1030 | { | |
1031 | predictor_coef_table_b[i] = (int16_t)readbits(alac, 16); | |
1032 | } | |
1033 | ||
1034 | /*********************/ | |
1035 | if (uncompressed_bytes) | |
1036 | { /* see mono case */ | |
1037 | int i; | |
1038 | for (i = 0; i < outputsamples; i++) | |
1039 | { | |
1040 | alac->uncompressed_bytes_buffer_a[i] = readbits(alac, uncompressed_bytes * 8); | |
1041 | alac->uncompressed_bytes_buffer_b[i] = readbits(alac, uncompressed_bytes * 8); | |
1042 | } | |
1043 | } | |
1044 | ||
1045 | /* channel 1 */ | |
1046 | entropy_rice_decode(alac, | |
1047 | alac->predicterror_buffer_a, | |
1048 | outputsamples, | |
1049 | readsamplesize, | |
1050 | alac->setinfo_rice_initialhistory, | |
1051 | alac->setinfo_rice_kmodifier, | |
1052 | ricemodifier_a * alac->setinfo_rice_historymult / 4, | |
1053 | (1 << alac->setinfo_rice_kmodifier) - 1); | |
1054 | ||
1055 | if (prediction_type_a == 0) | |
1056 | { /* adaptive fir */ | |
1057 | predictor_decompress_fir_adapt(alac->predicterror_buffer_a, | |
1058 | alac->outputsamples_buffer_a, | |
1059 | outputsamples, | |
1060 | readsamplesize, | |
1061 | predictor_coef_table_a, | |
1062 | predictor_coef_num_a, | |
1063 | prediction_quantitization_a); | |
1064 | } | |
1065 | else | |
1066 | { /* see mono case */ | |
1067 | fprintf(stderr, "FIXME: unhandled predicition type: %i\n", prediction_type_a); | |
1068 | } | |
1069 | ||
1070 | /* channel 2 */ | |
1071 | entropy_rice_decode(alac, | |
1072 | alac->predicterror_buffer_b, | |
1073 | outputsamples, | |
1074 | readsamplesize, | |
1075 | alac->setinfo_rice_initialhistory, | |
1076 | alac->setinfo_rice_kmodifier, | |
1077 | ricemodifier_b * alac->setinfo_rice_historymult / 4, | |
1078 | (1 << alac->setinfo_rice_kmodifier) - 1); | |
1079 | ||
1080 | if (prediction_type_b == 0) | |
1081 | { /* adaptive fir */ | |
1082 | predictor_decompress_fir_adapt(alac->predicterror_buffer_b, | |
1083 | alac->outputsamples_buffer_b, | |
1084 | outputsamples, | |
1085 | readsamplesize, | |
1086 | predictor_coef_table_b, | |
1087 | predictor_coef_num_b, | |
1088 | prediction_quantitization_b); | |
1089 | } | |
1090 | else | |
1091 | { | |
1092 | fprintf(stderr, "FIXME: unhandled predicition type: %i\n", prediction_type_b); | |
1093 | } | |
1094 | } | |
1095 | else | |
1096 | { /* not compressed, easy case */ | |
1097 | if (alac->setinfo_sample_size <= 16) | |
1098 | { | |
1099 | int i; | |
1100 | for (i = 0; i < outputsamples; i++) | |
1101 | { | |
1102 | int32_t audiobits_a, audiobits_b; | |
1103 | ||
1104 | audiobits_a = readbits(alac, alac->setinfo_sample_size); | |
1105 | audiobits_b = readbits(alac, alac->setinfo_sample_size); | |
1106 | ||
1107 | audiobits_a = SIGN_EXTENDED32(audiobits_a, alac->setinfo_sample_size); | |
1108 | audiobits_b = SIGN_EXTENDED32(audiobits_b, alac->setinfo_sample_size); | |
1109 | ||
1110 | alac->outputsamples_buffer_a[i] = audiobits_a; | |
1111 | alac->outputsamples_buffer_b[i] = audiobits_b; | |
1112 | } | |
1113 | } | |
1114 | else | |
1115 | { | |
1116 | int i; | |
1117 | for (i = 0; i < outputsamples; i++) | |
1118 | { | |
1119 | int32_t audiobits_a, audiobits_b; | |
1120 | ||
1121 | audiobits_a = readbits(alac, 16); | |
1122 | audiobits_a = audiobits_a << (alac->setinfo_sample_size - 16); | |
1123 | audiobits_a |= readbits(alac, alac->setinfo_sample_size - 16); | |
1124 | audiobits_a = SignExtend24(audiobits_a); | |
1125 | ||
1126 | audiobits_b = readbits(alac, 16); | |
1127 | audiobits_b = audiobits_b << (alac->setinfo_sample_size - 16); | |
1128 | audiobits_b |= readbits(alac, alac->setinfo_sample_size - 16); | |
1129 | audiobits_b = SignExtend24(audiobits_b); | |
1130 | ||
1131 | alac->outputsamples_buffer_a[i] = audiobits_a; | |
1132 | alac->outputsamples_buffer_b[i] = audiobits_b; | |
1133 | } | |
1134 | } | |
1135 | uncompressed_bytes = 0; // always 0 for uncompressed | |
1136 | interlacing_shift = 0; | |
1137 | interlacing_leftweight = 0; | |
1138 | } | |
1139 | ||
1140 | switch(alac->setinfo_sample_size) | |
1141 | { | |
1142 | case 16: | |
1143 | { | |
1144 | deinterlace_16(alac->outputsamples_buffer_a, | |
1145 | alac->outputsamples_buffer_b, | |
1146 | (int16_t*)outbuffer, | |
1147 | alac->numchannels, | |
1148 | outputsamples, | |
1149 | interlacing_shift, | |
1150 | interlacing_leftweight); | |
1151 | break; | |
1152 | } | |
1153 | case 24: | |
1154 | { | |
1155 | deinterlace_24(alac->outputsamples_buffer_a, | |
1156 | alac->outputsamples_buffer_b, | |
1157 | uncompressed_bytes, | |
1158 | alac->uncompressed_bytes_buffer_a, | |
1159 | alac->uncompressed_bytes_buffer_b, | |
1160 | (int16_t*)outbuffer, | |
1161 | alac->numchannels, | |
1162 | outputsamples, | |
1163 | interlacing_shift, | |
1164 | interlacing_leftweight); | |
1165 | break; | |
1166 | } | |
1167 | case 20: | |
1168 | case 32: | |
1169 | fprintf(stderr, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size); | |
1170 | break; | |
1171 | default: | |
1172 | break; | |
1173 | } | |
1174 | ||
1175 | break; | |
1176 | } | |
1177 | } | |
1178 | } | |
1179 | ||
1180 | alac_file *create_alac(int samplesize, int numchannels) | |
1181 | { | |
1182 | alac_file *newfile = malloc(sizeof(alac_file)); | |
1183 | ||
1184 | newfile->samplesize = samplesize; | |
1185 | newfile->numchannels = numchannels; | |
1186 | newfile->bytespersample = (samplesize / 8) * numchannels; | |
1187 | ||
1188 | return newfile; | |
1189 | } | |
1190 | ||
982a0c26 JVH |
1191 | void destroy_alac(alac_file *alac) |
1192 | { | |
1193 | if (!alac) return; | |
1194 | deallocate_buffers(alac); | |
1195 | free(alac); | |
1196 | } |