2 * ALAC (Apple Lossless Audio Codec) decoder
3 * Copyright (c) 2005 David Hammerton
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * ALAC (Apple Lossless Audio Codec) decoder
24 * @author 2005 David Hammerton
26 * For more information on the ALAC format, visit:
27 * http://crazney.net/programs/itunes/alac.html
29 * Note: This decoder expects a 36- (0x24-)byte QuickTime atom to be
30 * passed through the extradata[_size] fields. This atom is tacked onto
31 * the end of an 'alac' stsd atom and has the following format:
32 * bytes 0-3 atom size (0x24), big-endian
33 * bytes 4-7 atom type ('alac', not the 'alac' tag from start of stsd)
34 * bytes 8-35 data bytes needed by decoder
39 #include "bitstream.h"
41 #define ALAC_EXTRADATA_SIZE 36
45 AVCodecContext
*avctx
;
47 /* init to 0; first frame decode should initialize from extradata and
49 int context_initialized
;
56 int32_t *predicterror_buffer_a
;
57 int32_t *predicterror_buffer_b
;
59 int32_t *outputsamples_buffer_a
;
60 int32_t *outputsamples_buffer_b
;
62 /* stuff from setinfo */
63 uint32_t setinfo_max_samples_per_frame
; /* 0x1000 = 4096 */ /* max samples per frame? */
64 uint8_t setinfo_7a
; /* 0x00 */
65 uint8_t setinfo_sample_size
; /* 0x10 */
66 uint8_t setinfo_rice_historymult
; /* 0x28 */
67 uint8_t setinfo_rice_initialhistory
; /* 0x0a */
68 uint8_t setinfo_rice_kmodifier
; /* 0x0e */
69 uint8_t setinfo_7f
; /* 0x02 */
70 uint16_t setinfo_80
; /* 0x00ff */
71 uint32_t setinfo_82
; /* 0x000020e7 */
72 uint32_t setinfo_86
; /* 0x00069fe4 */
73 uint32_t setinfo_8a_rate
; /* 0x0000ac44 */
74 /* end setinfo stuff */
78 static void allocate_buffers(ALACContext
*alac
)
80 alac
->predicterror_buffer_a
= av_malloc(alac
->setinfo_max_samples_per_frame
* 4);
81 alac
->predicterror_buffer_b
= av_malloc(alac
->setinfo_max_samples_per_frame
* 4);
83 alac
->outputsamples_buffer_a
= av_malloc(alac
->setinfo_max_samples_per_frame
* 4);
84 alac
->outputsamples_buffer_b
= av_malloc(alac
->setinfo_max_samples_per_frame
* 4);
87 void alac_set_info(ALACContext
*alac
)
89 unsigned char *ptr
= alac
->avctx
->extradata
;
95 alac
->setinfo_max_samples_per_frame
= BE_32(ptr
); /* buffer size / 2 ? */
97 alac
->setinfo_7a
= *ptr
++;
98 alac
->setinfo_sample_size
= *ptr
++;
99 alac
->setinfo_rice_historymult
= *ptr
++;
100 alac
->setinfo_rice_initialhistory
= *ptr
++;
101 alac
->setinfo_rice_kmodifier
= *ptr
++;
102 alac
->setinfo_7f
= *ptr
++;
103 alac
->setinfo_80
= BE_16(ptr
);
105 alac
->setinfo_82
= BE_32(ptr
);
107 alac
->setinfo_86
= BE_32(ptr
);
109 alac
->setinfo_8a_rate
= BE_32(ptr
);
112 allocate_buffers(alac
);
115 /* hideously inefficient. could use a bitmask search,
116 * alternatively bsr on x86,
118 static int count_leading_zeros(int32_t input
)
121 while (!(0x80000000 & input
) && i
< 32) {
128 void bastardized_rice_decompress(ALACContext
*alac
,
129 int32_t *output_buffer
,
131 int readsamplesize
, /* arg_10 */
132 int rice_initialhistory
, /* arg424->b */
133 int rice_kmodifier
, /* arg424->d */
134 int rice_historymult
, /* arg424->c */
135 int rice_kmodifier_mask
/* arg424->e */
139 unsigned int history
= rice_initialhistory
;
140 int sign_modifier
= 0;
142 for (output_count
= 0; output_count
< output_size
; output_count
++) {
147 /* read x - number of 1s before 0 represent the rice */
148 while (x
<= 8 && get_bits1(&alac
->gb
)) {
153 if (x
> 8) { /* RICE THRESHOLD */
154 /* use alternative encoding */
157 value
= get_bits(&alac
->gb
, readsamplesize
);
159 /* mask value to readsamplesize size */
160 if (readsamplesize
!= 32)
161 value
&= (0xffffffff >> (32 - readsamplesize
));
165 /* standard rice encoding */
167 int k
; /* size of extra bits */
169 /* read k, that is bits as is */
170 k
= 31 - rice_kmodifier
- count_leading_zeros((history
>> 9) + 3);
178 extrabits
= show_bits(&alac
->gb
, k
);
180 /* multiply x by 2^k - 1, as part of their strange algorithm */
185 get_bits(&alac
->gb
, k
);
187 get_bits(&alac
->gb
, k
- 1);
192 x_modified
= sign_modifier
+ x
;
193 final_val
= (x_modified
+ 1) / 2;
194 if (x_modified
& 1) final_val
*= -1;
196 output_buffer
[output_count
] = final_val
;
200 /* now update the history */
201 history
+= (x_modified
* rice_historymult
)
202 - ((history
* rice_historymult
) >> 9);
204 if (x_modified
> 0xffff)
207 /* special case: there may be compressed blocks of 0 */
208 if ((history
< 128) && (output_count
+1 < output_size
)) {
214 while (x
<= 8 && get_bits1(&alac
->gb
)) {
219 block_size
= get_bits(&alac
->gb
, 16);
220 block_size
&= 0xffff;
225 k
= count_leading_zeros(history
) + ((history
+ 16) >> 6 /* / 64 */) - 24;
227 extrabits
= show_bits(&alac
->gb
, k
);
229 block_size
= (((1 << k
) - 1) & rice_kmodifier_mask
) * x
235 get_bits(&alac
->gb
, k
- 1);
237 get_bits(&alac
->gb
, k
);
241 if (block_size
> 0) {
242 memset(&output_buffer
[output_count
+1], 0, block_size
* 4);
243 output_count
+= block_size
;
247 if (block_size
> 0xffff)
255 #define SIGN_EXTENDED32(val, bits) ((val << (32 - bits)) >> (32 - bits))
257 #define SIGN_ONLY(v) \
262 static void predictor_decompress_fir_adapt(int32_t *error_buffer
,
266 int16_t *predictor_coef_table
,
267 int predictor_coef_num
,
268 int predictor_quantitization
)
272 /* first sample always copies */
273 *buffer_out
= *error_buffer
;
275 if (!predictor_coef_num
) {
276 if (output_size
<= 1) return;
277 memcpy(buffer_out
+1, error_buffer
+1, (output_size
-1) * 4);
281 if (predictor_coef_num
== 0x1f) { /* 11111 - max value of predictor_coef_num */
282 /* second-best case scenario for fir decompression,
283 * error describes a small difference from the previous sample only
285 if (output_size
<= 1) return;
286 for (i
= 0; i
< output_size
- 1; i
++) {
290 prev_value
= buffer_out
[i
];
291 error_value
= error_buffer
[i
+1];
292 buffer_out
[i
+1] = SIGN_EXTENDED32((prev_value
+ error_value
), readsamplesize
);
297 /* read warm-up samples */
298 if (predictor_coef_num
> 0) {
300 for (i
= 0; i
< predictor_coef_num
; i
++) {
303 val
= buffer_out
[i
] + error_buffer
[i
+1];
305 val
= SIGN_EXTENDED32(val
, readsamplesize
);
307 buffer_out
[i
+1] = val
;
312 /* 4 and 8 are very common cases (the only ones i've seen). these
313 * should be unrolled and optimised
315 if (predictor_coef_num
== 4) {
316 /* FIXME: optimised general case */
320 if (predictor_coef_table
== 8) {
321 /* FIXME: optimised general case */
328 if (predictor_coef_num
> 0) {
329 for (i
= predictor_coef_num
+ 1;
335 int error_val
= error_buffer
[i
];
337 for (j
= 0; j
< predictor_coef_num
; j
++) {
338 sum
+= (buffer_out
[predictor_coef_num
-j
] - buffer_out
[0]) *
339 predictor_coef_table
[j
];
342 outval
= (1 << (predictor_quantitization
-1)) + sum
;
343 outval
= outval
>> predictor_quantitization
;
344 outval
= outval
+ buffer_out
[0] + error_val
;
345 outval
= SIGN_EXTENDED32(outval
, readsamplesize
);
347 buffer_out
[predictor_coef_num
+1] = outval
;
350 int predictor_num
= predictor_coef_num
- 1;
352 while (predictor_num
>= 0 && error_val
> 0) {
353 int val
= buffer_out
[0] - buffer_out
[predictor_coef_num
- predictor_num
];
354 int sign
= SIGN_ONLY(val
);
356 predictor_coef_table
[predictor_num
] -= sign
;
358 val
*= sign
; /* absolute value */
360 error_val
-= ((val
>> predictor_quantitization
) *
361 (predictor_coef_num
- predictor_num
));
365 } else if (error_val
< 0) {
366 int predictor_num
= predictor_coef_num
- 1;
368 while (predictor_num
>= 0 && error_val
< 0) {
369 int val
= buffer_out
[0] - buffer_out
[predictor_coef_num
- predictor_num
];
370 int sign
= - SIGN_ONLY(val
);
372 predictor_coef_table
[predictor_num
] -= sign
;
374 val
*= sign
; /* neg value */
376 error_val
-= ((val
>> predictor_quantitization
) *
377 (predictor_coef_num
- predictor_num
));
388 void deinterlace_16(int32_t *buffer_a
, int32_t *buffer_b
,
390 int numchannels
, int numsamples
,
391 uint8_t interlacing_shift
,
392 uint8_t interlacing_leftweight
)
395 if (numsamples
<= 0) return;
397 /* weighted interlacing */
398 if (interlacing_leftweight
) {
399 for (i
= 0; i
< numsamples
; i
++) {
400 int32_t difference
, midright
;
404 midright
= buffer_a
[i
];
405 difference
= buffer_b
[i
];
408 right
= midright
- ((difference
* interlacing_leftweight
) >> interlacing_shift
);
409 left
= (midright
- ((difference
* interlacing_leftweight
) >> interlacing_shift
))
412 buffer_out
[i
*numchannels
] = left
;
413 buffer_out
[i
*numchannels
+ 1] = right
;
419 /* otherwise basic interlacing took place */
420 for (i
= 0; i
< numsamples
; i
++) {
426 buffer_out
[i
*numchannels
] = left
;
427 buffer_out
[i
*numchannels
+ 1] = right
;
431 static int alac_decode_frame(AVCodecContext
*avctx
,
432 void *outbuffer
, int *outputsize
,
433 uint8_t *inbuffer
, int input_buffer_size
)
435 ALACContext
*alac
= avctx
->priv_data
;
438 int32_t outputsamples
;
440 /* short-circuit null buffers */
441 if (!inbuffer
|| !input_buffer_size
)
442 return input_buffer_size
;
444 /* initialize from the extradata */
445 if (!alac
->context_initialized
) {
446 if (alac
->avctx
->extradata_size
!= ALAC_EXTRADATA_SIZE
) {
447 av_log(NULL
, AV_LOG_ERROR
, "alac: expected %d extradata bytes\n",
448 ALAC_EXTRADATA_SIZE
);
449 return input_buffer_size
;
452 alac
->context_initialized
= 1;
455 outputsamples
= alac
->setinfo_max_samples_per_frame
;
457 init_get_bits(&alac
->gb
, inbuffer
, input_buffer_size
* 8);
459 channels
= get_bits(&alac
->gb
, 3);
461 *outputsize
= outputsamples
* alac
->bytespersample
;
464 case 0: { /* 1 channel */
473 /* 2^result = something to do with output waiting.
474 * perhaps matters if we read > 1 frame in a pass?
476 get_bits(&alac
->gb
, 4);
478 get_bits(&alac
->gb
, 12); /* unknown, skip 12 bits */
480 hassize
= get_bits(&alac
->gb
, 1); /* the output sample size is stored soon */
482 wasted_bytes
= get_bits(&alac
->gb
, 2); /* unknown ? */
484 isnotcompressed
= get_bits(&alac
->gb
, 1); /* whether the frame is compressed */
487 /* now read the number of samples,
488 * as a 32bit integer */
489 outputsamples
= get_bits(&alac
->gb
, 32);
490 *outputsize
= outputsamples
* alac
->bytespersample
;
493 readsamplesize
= alac
->setinfo_sample_size
- (wasted_bytes
* 8);
495 if (!isnotcompressed
) {
496 /* so it is compressed */
497 int16_t predictor_coef_table
[32];
498 int predictor_coef_num
;
500 int prediction_quantitization
;
503 /* skip 16 bits, not sure what they are. seem to be used in
504 * two channel case */
505 get_bits(&alac
->gb
, 8);
506 get_bits(&alac
->gb
, 8);
508 prediction_type
= get_bits(&alac
->gb
, 4);
509 prediction_quantitization
= get_bits(&alac
->gb
, 4);
511 ricemodifier
= get_bits(&alac
->gb
, 3);
512 predictor_coef_num
= get_bits(&alac
->gb
, 5);
514 /* read the predictor table */
515 for (i
= 0; i
< predictor_coef_num
; i
++) {
516 predictor_coef_table
[i
] = (int16_t)get_bits(&alac
->gb
, 16);
520 /* these bytes seem to have something to do with
523 av_log(NULL
, AV_LOG_ERROR
, "FIXME: unimplemented, unhandling of wasted_bytes\n");
526 bastardized_rice_decompress(alac
,
527 alac
->predicterror_buffer_a
,
530 alac
->setinfo_rice_initialhistory
,
531 alac
->setinfo_rice_kmodifier
,
532 ricemodifier
* alac
->setinfo_rice_historymult
/ 4,
533 (1 << alac
->setinfo_rice_kmodifier
) - 1);
535 if (prediction_type
== 0) {
537 predictor_decompress_fir_adapt(alac
->predicterror_buffer_a
,
538 alac
->outputsamples_buffer_a
,
541 predictor_coef_table
,
543 prediction_quantitization
);
545 av_log(NULL
, AV_LOG_ERROR
, "FIXME: unhandled prediction type: %i\n", prediction_type
);
546 /* i think the only other prediction type (or perhaps this is just a
547 * boolean?) runs adaptive fir twice.. like:
548 * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
549 * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
555 /* not compressed, easy case */
556 if (readsamplesize
<= 16) {
558 for (i
= 0; i
< outputsamples
; i
++) {
559 int32_t audiobits
= get_bits(&alac
->gb
, readsamplesize
);
561 audiobits
= SIGN_EXTENDED32(audiobits
, readsamplesize
);
563 alac
->outputsamples_buffer_a
[i
] = audiobits
;
567 for (i
= 0; i
< outputsamples
; i
++) {
570 audiobits
= get_bits(&alac
->gb
, 16);
571 /* special case of sign extension..
572 * as we'll be ORing the low 16bits into this */
573 audiobits
= audiobits
<< 16;
574 audiobits
= audiobits
>> (32 - readsamplesize
);
576 audiobits
|= get_bits(&alac
->gb
, readsamplesize
- 16);
578 alac
->outputsamples_buffer_a
[i
] = audiobits
;
581 /* wasted_bytes = 0; // unused */
584 switch(alac
->setinfo_sample_size
) {
587 for (i
= 0; i
< outputsamples
; i
++) {
588 int16_t sample
= alac
->outputsamples_buffer_a
[i
];
589 sample
= be2me_16(sample
);
590 ((int16_t*)outbuffer
)[i
* alac
->numchannels
] = sample
;
597 av_log(NULL
, AV_LOG_ERROR
, "FIXME: unimplemented sample size %i\n", alac
->setinfo_sample_size
);
604 case 1: { /* 2 channels */
611 uint8_t interlacing_shift
;
612 uint8_t interlacing_leftweight
;
614 /* 2^result = something to do with output waiting.
615 * perhaps matters if we read > 1 frame in a pass?
617 get_bits(&alac
->gb
, 4);
619 get_bits(&alac
->gb
, 12); /* unknown, skip 12 bits */
621 hassize
= get_bits(&alac
->gb
, 1); /* the output sample size is stored soon */
623 wasted_bytes
= get_bits(&alac
->gb
, 2); /* unknown ? */
625 isnotcompressed
= get_bits(&alac
->gb
, 1); /* whether the frame is compressed */
628 /* now read the number of samples,
629 * as a 32bit integer */
630 outputsamples
= get_bits(&alac
->gb
, 32);
631 *outputsize
= outputsamples
* alac
->bytespersample
;
634 readsamplesize
= alac
->setinfo_sample_size
- (wasted_bytes
* 8) + 1;
636 if (!isnotcompressed
) {
638 int16_t predictor_coef_table_a
[32];
639 int predictor_coef_num_a
;
640 int prediction_type_a
;
641 int prediction_quantitization_a
;
644 int16_t predictor_coef_table_b
[32];
645 int predictor_coef_num_b
;
646 int prediction_type_b
;
647 int prediction_quantitization_b
;
652 interlacing_shift
= get_bits(&alac
->gb
, 8);
653 interlacing_leftweight
= get_bits(&alac
->gb
, 8);
655 /******** channel 1 ***********/
656 prediction_type_a
= get_bits(&alac
->gb
, 4);
657 prediction_quantitization_a
= get_bits(&alac
->gb
, 4);
659 ricemodifier_a
= get_bits(&alac
->gb
, 3);
660 predictor_coef_num_a
= get_bits(&alac
->gb
, 5);
662 /* read the predictor table */
663 for (i
= 0; i
< predictor_coef_num_a
; i
++) {
664 predictor_coef_table_a
[i
] = (int16_t)get_bits(&alac
->gb
, 16);
667 /******** channel 2 *********/
668 prediction_type_b
= get_bits(&alac
->gb
, 4);
669 prediction_quantitization_b
= get_bits(&alac
->gb
, 4);
671 ricemodifier_b
= get_bits(&alac
->gb
, 3);
672 predictor_coef_num_b
= get_bits(&alac
->gb
, 5);
674 /* read the predictor table */
675 for (i
= 0; i
< predictor_coef_num_b
; i
++) {
676 predictor_coef_table_b
[i
] = (int16_t)get_bits(&alac
->gb
, 16);
679 /*********************/
682 av_log(NULL
, AV_LOG_ERROR
, "FIXME: unimplemented, unhandling of wasted_bytes\n");
686 bastardized_rice_decompress(alac
,
687 alac
->predicterror_buffer_a
,
690 alac
->setinfo_rice_initialhistory
,
691 alac
->setinfo_rice_kmodifier
,
692 ricemodifier_a
* alac
->setinfo_rice_historymult
/ 4,
693 (1 << alac
->setinfo_rice_kmodifier
) - 1);
695 if (prediction_type_a
== 0) {
697 predictor_decompress_fir_adapt(alac
->predicterror_buffer_a
,
698 alac
->outputsamples_buffer_a
,
701 predictor_coef_table_a
,
702 predictor_coef_num_a
,
703 prediction_quantitization_a
);
706 av_log(NULL
, AV_LOG_ERROR
, "FIXME: unhandled prediction type: %i\n", prediction_type_a
);
710 bastardized_rice_decompress(alac
,
711 alac
->predicterror_buffer_b
,
714 alac
->setinfo_rice_initialhistory
,
715 alac
->setinfo_rice_kmodifier
,
716 ricemodifier_b
* alac
->setinfo_rice_historymult
/ 4,
717 (1 << alac
->setinfo_rice_kmodifier
) - 1);
719 if (prediction_type_b
== 0) {
721 predictor_decompress_fir_adapt(alac
->predicterror_buffer_b
,
722 alac
->outputsamples_buffer_b
,
725 predictor_coef_table_b
,
726 predictor_coef_num_b
,
727 prediction_quantitization_b
);
729 av_log(NULL
, AV_LOG_ERROR
, "FIXME: unhandled prediction type: %i\n", prediction_type_b
);
732 /* not compressed, easy case */
733 if (alac
->setinfo_sample_size
<= 16) {
735 for (i
= 0; i
< outputsamples
; i
++) {
736 int32_t audiobits_a
, audiobits_b
;
738 audiobits_a
= get_bits(&alac
->gb
, alac
->setinfo_sample_size
);
739 audiobits_b
= get_bits(&alac
->gb
, alac
->setinfo_sample_size
);
741 audiobits_a
= SIGN_EXTENDED32(audiobits_a
, alac
->setinfo_sample_size
);
742 audiobits_b
= SIGN_EXTENDED32(audiobits_b
, alac
->setinfo_sample_size
);
744 alac
->outputsamples_buffer_a
[i
] = audiobits_a
;
745 alac
->outputsamples_buffer_b
[i
] = audiobits_b
;
749 for (i
= 0; i
< outputsamples
; i
++) {
750 int32_t audiobits_a
, audiobits_b
;
752 audiobits_a
= get_bits(&alac
->gb
, 16);
753 audiobits_a
= audiobits_a
<< 16;
754 audiobits_a
= audiobits_a
>> (32 - alac
->setinfo_sample_size
);
755 audiobits_a
|= get_bits(&alac
->gb
, alac
->setinfo_sample_size
- 16);
757 audiobits_b
= get_bits(&alac
->gb
, 16);
758 audiobits_b
= audiobits_b
<< 16;
759 audiobits_b
= audiobits_b
>> (32 - alac
->setinfo_sample_size
);
760 audiobits_b
|= get_bits(&alac
->gb
, alac
->setinfo_sample_size
- 16);
762 alac
->outputsamples_buffer_a
[i
] = audiobits_a
;
763 alac
->outputsamples_buffer_b
[i
] = audiobits_b
;
766 /* wasted_bytes = 0; */
767 interlacing_shift
= 0;
768 interlacing_leftweight
= 0;
771 switch(alac
->setinfo_sample_size
) {
773 deinterlace_16(alac
->outputsamples_buffer_a
,
774 alac
->outputsamples_buffer_b
,
779 interlacing_leftweight
);
785 av_log(NULL
, AV_LOG_ERROR
, "FIXME: unimplemented sample size %i\n", alac
->setinfo_sample_size
);
795 return input_buffer_size
;
798 static int alac_decode_init(AVCodecContext
* avctx
)
800 ALACContext
*alac
= avctx
->priv_data
;
802 alac
->context_initialized
= 0;
804 alac
->samplesize
= alac
->avctx
->bits_per_sample
;
805 alac
->numchannels
= alac
->avctx
->channels
;
806 alac
->bytespersample
= (alac
->samplesize
/ 8) * alac
->numchannels
;
811 static int alac_decode_close(AVCodecContext
*avctx
)
813 ALACContext
*alac
= avctx
->priv_data
;
815 av_free(alac
->predicterror_buffer_a
);
816 av_free(alac
->predicterror_buffer_b
);
818 av_free(alac
->outputsamples_buffer_a
);
819 av_free(alac
->outputsamples_buffer_b
);
824 AVCodec alac_decoder
= {