b7b82b2c4943807326df016a7e24dd9aaafac0c9
[libav.git] / libavcodec / alacenc.c
1 /**
2 * ALAC audio encoder
3 * Copyright (c) 2008 Jaikrishnan Menon <realityman@gmx.net>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include "avcodec.h"
23 #include "get_bits.h"
24 #include "put_bits.h"
25 #include "dsputil.h"
26 #include "lpc.h"
27 #include "mathops.h"
28
29 #define DEFAULT_FRAME_SIZE 4096
30 #define DEFAULT_SAMPLE_SIZE 16
31 #define MAX_CHANNELS 8
32 #define ALAC_EXTRADATA_SIZE 36
33 #define ALAC_FRAME_HEADER_SIZE 55
34 #define ALAC_FRAME_FOOTER_SIZE 3
35
36 #define ALAC_ESCAPE_CODE 0x1FF
37 #define ALAC_MAX_LPC_ORDER 30
38 #define DEFAULT_MAX_PRED_ORDER 6
39 #define DEFAULT_MIN_PRED_ORDER 4
40 #define ALAC_MAX_LPC_PRECISION 9
41 #define ALAC_MAX_LPC_SHIFT 9
42
43 #define ALAC_CHMODE_LEFT_RIGHT 0
44 #define ALAC_CHMODE_LEFT_SIDE 1
45 #define ALAC_CHMODE_RIGHT_SIDE 2
46 #define ALAC_CHMODE_MID_SIDE 3
47
48 typedef struct RiceContext {
49 int history_mult;
50 int initial_history;
51 int k_modifier;
52 int rice_modifier;
53 } RiceContext;
54
55 typedef struct LPCContext {
56 int lpc_order;
57 int lpc_coeff[ALAC_MAX_LPC_ORDER+1];
58 int lpc_quant;
59 } LPCContext;
60
61 typedef struct AlacEncodeContext {
62 int compression_level;
63 int min_prediction_order;
64 int max_prediction_order;
65 int max_coded_frame_size;
66 int write_sample_size;
67 int32_t sample_buf[MAX_CHANNELS][DEFAULT_FRAME_SIZE];
68 int32_t predictor_buf[DEFAULT_FRAME_SIZE];
69 int interlacing_shift;
70 int interlacing_leftweight;
71 PutBitContext pbctx;
72 RiceContext rc;
73 LPCContext lpc[MAX_CHANNELS];
74 DSPContext dspctx;
75 AVCodecContext *avctx;
76 } AlacEncodeContext;
77
78
79 static void init_sample_buffers(AlacEncodeContext *s, int16_t *input_samples)
80 {
81 int ch, i;
82
83 for(ch=0;ch<s->avctx->channels;ch++) {
84 int16_t *sptr = input_samples + ch;
85 for(i=0;i<s->avctx->frame_size;i++) {
86 s->sample_buf[ch][i] = *sptr;
87 sptr += s->avctx->channels;
88 }
89 }
90 }
91
92 static void encode_scalar(AlacEncodeContext *s, int x, int k, int write_sample_size)
93 {
94 int divisor, q, r;
95
96 k = FFMIN(k, s->rc.k_modifier);
97 divisor = (1<<k) - 1;
98 q = x / divisor;
99 r = x % divisor;
100
101 if(q > 8) {
102 // write escape code and sample value directly
103 put_bits(&s->pbctx, 9, ALAC_ESCAPE_CODE);
104 put_bits(&s->pbctx, write_sample_size, x);
105 } else {
106 if(q)
107 put_bits(&s->pbctx, q, (1<<q) - 1);
108 put_bits(&s->pbctx, 1, 0);
109
110 if(k != 1) {
111 if(r > 0)
112 put_bits(&s->pbctx, k, r+1);
113 else
114 put_bits(&s->pbctx, k-1, 0);
115 }
116 }
117 }
118
119 static void write_frame_header(AlacEncodeContext *s, int is_verbatim)
120 {
121 put_bits(&s->pbctx, 3, s->avctx->channels-1); // No. of channels -1
122 put_bits(&s->pbctx, 16, 0); // Seems to be zero
123 put_bits(&s->pbctx, 1, 1); // Sample count is in the header
124 put_bits(&s->pbctx, 2, 0); // FIXME: Wasted bytes field
125 put_bits(&s->pbctx, 1, is_verbatim); // Audio block is verbatim
126 put_bits32(&s->pbctx, s->avctx->frame_size); // No. of samples in the frame
127 }
128
129 static void calc_predictor_params(AlacEncodeContext *s, int ch)
130 {
131 int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
132 int shift[MAX_LPC_ORDER];
133 int opt_order;
134
135 if (s->compression_level == 1) {
136 s->lpc[ch].lpc_order = 6;
137 s->lpc[ch].lpc_quant = 6;
138 s->lpc[ch].lpc_coeff[0] = 160;
139 s->lpc[ch].lpc_coeff[1] = -190;
140 s->lpc[ch].lpc_coeff[2] = 170;
141 s->lpc[ch].lpc_coeff[3] = -130;
142 s->lpc[ch].lpc_coeff[4] = 80;
143 s->lpc[ch].lpc_coeff[5] = -25;
144 } else {
145 opt_order = ff_lpc_calc_coefs(&s->dspctx, s->sample_buf[ch],
146 s->avctx->frame_size,
147 s->min_prediction_order,
148 s->max_prediction_order,
149 ALAC_MAX_LPC_PRECISION, coefs, shift, 1,
150 ORDER_METHOD_EST, ALAC_MAX_LPC_SHIFT, 1);
151
152 s->lpc[ch].lpc_order = opt_order;
153 s->lpc[ch].lpc_quant = shift[opt_order-1];
154 memcpy(s->lpc[ch].lpc_coeff, coefs[opt_order-1], opt_order*sizeof(int));
155 }
156 }
157
158 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
159 {
160 int i, best;
161 int32_t lt, rt;
162 uint64_t sum[4];
163 uint64_t score[4];
164
165 /* calculate sum of 2nd order residual for each channel */
166 sum[0] = sum[1] = sum[2] = sum[3] = 0;
167 for(i=2; i<n; i++) {
168 lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
169 rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
170 sum[2] += FFABS((lt + rt) >> 1);
171 sum[3] += FFABS(lt - rt);
172 sum[0] += FFABS(lt);
173 sum[1] += FFABS(rt);
174 }
175
176 /* calculate score for each mode */
177 score[0] = sum[0] + sum[1];
178 score[1] = sum[0] + sum[3];
179 score[2] = sum[1] + sum[3];
180 score[3] = sum[2] + sum[3];
181
182 /* return mode with lowest score */
183 best = 0;
184 for(i=1; i<4; i++) {
185 if(score[i] < score[best]) {
186 best = i;
187 }
188 }
189 return best;
190 }
191
192 static void alac_stereo_decorrelation(AlacEncodeContext *s)
193 {
194 int32_t *left = s->sample_buf[0], *right = s->sample_buf[1];
195 int i, mode, n = s->avctx->frame_size;
196 int32_t tmp;
197
198 mode = estimate_stereo_mode(left, right, n);
199
200 switch(mode)
201 {
202 case ALAC_CHMODE_LEFT_RIGHT:
203 s->interlacing_leftweight = 0;
204 s->interlacing_shift = 0;
205 break;
206
207 case ALAC_CHMODE_LEFT_SIDE:
208 for(i=0; i<n; i++) {
209 right[i] = left[i] - right[i];
210 }
211 s->interlacing_leftweight = 1;
212 s->interlacing_shift = 0;
213 break;
214
215 case ALAC_CHMODE_RIGHT_SIDE:
216 for(i=0; i<n; i++) {
217 tmp = right[i];
218 right[i] = left[i] - right[i];
219 left[i] = tmp + (right[i] >> 31);
220 }
221 s->interlacing_leftweight = 1;
222 s->interlacing_shift = 31;
223 break;
224
225 default:
226 for(i=0; i<n; i++) {
227 tmp = left[i];
228 left[i] = (tmp + right[i]) >> 1;
229 right[i] = tmp - right[i];
230 }
231 s->interlacing_leftweight = 1;
232 s->interlacing_shift = 1;
233 break;
234 }
235 }
236
237 static void alac_linear_predictor(AlacEncodeContext *s, int ch)
238 {
239 int i;
240 LPCContext lpc = s->lpc[ch];
241
242 if(lpc.lpc_order == 31) {
243 s->predictor_buf[0] = s->sample_buf[ch][0];
244
245 for(i=1; i<s->avctx->frame_size; i++)
246 s->predictor_buf[i] = s->sample_buf[ch][i] - s->sample_buf[ch][i-1];
247
248 return;
249 }
250
251 // generalised linear predictor
252
253 if(lpc.lpc_order > 0) {
254 int32_t *samples = s->sample_buf[ch];
255 int32_t *residual = s->predictor_buf;
256
257 // generate warm-up samples
258 residual[0] = samples[0];
259 for(i=1;i<=lpc.lpc_order;i++)
260 residual[i] = samples[i] - samples[i-1];
261
262 // perform lpc on remaining samples
263 for(i = lpc.lpc_order + 1; i < s->avctx->frame_size; i++) {
264 int sum = 1 << (lpc.lpc_quant - 1), res_val, j;
265
266 for (j = 0; j < lpc.lpc_order; j++) {
267 sum += (samples[lpc.lpc_order-j] - samples[0]) *
268 lpc.lpc_coeff[j];
269 }
270
271 sum >>= lpc.lpc_quant;
272 sum += samples[0];
273 residual[i] = sign_extend(samples[lpc.lpc_order+1] - sum,
274 s->write_sample_size);
275 res_val = residual[i];
276
277 if(res_val) {
278 int index = lpc.lpc_order - 1;
279 int neg = (res_val < 0);
280
281 while(index >= 0 && (neg ? (res_val < 0):(res_val > 0))) {
282 int val = samples[0] - samples[lpc.lpc_order - index];
283 int sign = (val ? FFSIGN(val) : 0);
284
285 if(neg)
286 sign*=-1;
287
288 lpc.lpc_coeff[index] -= sign;
289 val *= sign;
290 res_val -= ((val >> lpc.lpc_quant) *
291 (lpc.lpc_order - index));
292 index--;
293 }
294 }
295 samples++;
296 }
297 }
298 }
299
300 static void alac_entropy_coder(AlacEncodeContext *s)
301 {
302 unsigned int history = s->rc.initial_history;
303 int sign_modifier = 0, i, k;
304 int32_t *samples = s->predictor_buf;
305
306 for(i=0;i < s->avctx->frame_size;) {
307 int x;
308
309 k = av_log2((history >> 9) + 3);
310
311 x = -2*(*samples)-1;
312 x ^= (x>>31);
313
314 samples++;
315 i++;
316
317 encode_scalar(s, x - sign_modifier, k, s->write_sample_size);
318
319 history += x * s->rc.history_mult
320 - ((history * s->rc.history_mult) >> 9);
321
322 sign_modifier = 0;
323 if(x > 0xFFFF)
324 history = 0xFFFF;
325
326 if((history < 128) && (i < s->avctx->frame_size)) {
327 unsigned int block_size = 0;
328
329 k = 7 - av_log2(history) + ((history + 16) >> 6);
330
331 while((*samples == 0) && (i < s->avctx->frame_size)) {
332 samples++;
333 i++;
334 block_size++;
335 }
336 encode_scalar(s, block_size, k, 16);
337
338 sign_modifier = (block_size <= 0xFFFF);
339
340 history = 0;
341 }
342
343 }
344 }
345
346 static void write_compressed_frame(AlacEncodeContext *s)
347 {
348 int i, j;
349
350 if(s->avctx->channels == 2)
351 alac_stereo_decorrelation(s);
352 put_bits(&s->pbctx, 8, s->interlacing_shift);
353 put_bits(&s->pbctx, 8, s->interlacing_leftweight);
354
355 for(i=0;i<s->avctx->channels;i++) {
356
357 calc_predictor_params(s, i);
358
359 put_bits(&s->pbctx, 4, 0); // prediction type : currently only type 0 has been RE'd
360 put_bits(&s->pbctx, 4, s->lpc[i].lpc_quant);
361
362 put_bits(&s->pbctx, 3, s->rc.rice_modifier);
363 put_bits(&s->pbctx, 5, s->lpc[i].lpc_order);
364 // predictor coeff. table
365 for(j=0;j<s->lpc[i].lpc_order;j++) {
366 put_sbits(&s->pbctx, 16, s->lpc[i].lpc_coeff[j]);
367 }
368 }
369
370 // apply lpc and entropy coding to audio samples
371
372 for(i=0;i<s->avctx->channels;i++) {
373 alac_linear_predictor(s, i);
374 alac_entropy_coder(s);
375 }
376 }
377
378 static av_cold int alac_encode_init(AVCodecContext *avctx)
379 {
380 AlacEncodeContext *s = avctx->priv_data;
381 uint8_t *alac_extradata = av_mallocz(ALAC_EXTRADATA_SIZE+1);
382
383 avctx->frame_size = DEFAULT_FRAME_SIZE;
384 avctx->bits_per_coded_sample = DEFAULT_SAMPLE_SIZE;
385
386 if(avctx->sample_fmt != SAMPLE_FMT_S16) {
387 av_log(avctx, AV_LOG_ERROR, "only pcm_s16 input samples are supported\n");
388 return -1;
389 }
390
391 // Set default compression level
392 if(avctx->compression_level == FF_COMPRESSION_DEFAULT)
393 s->compression_level = 2;
394 else
395 s->compression_level = av_clip(avctx->compression_level, 0, 2);
396
397 // Initialize default Rice parameters
398 s->rc.history_mult = 40;
399 s->rc.initial_history = 10;
400 s->rc.k_modifier = 14;
401 s->rc.rice_modifier = 4;
402
403 s->max_coded_frame_size = 8 + (avctx->frame_size*avctx->channels*avctx->bits_per_coded_sample>>3);
404
405 s->write_sample_size = avctx->bits_per_coded_sample + avctx->channels - 1; // FIXME: consider wasted_bytes
406
407 AV_WB32(alac_extradata, ALAC_EXTRADATA_SIZE);
408 AV_WB32(alac_extradata+4, MKBETAG('a','l','a','c'));
409 AV_WB32(alac_extradata+12, avctx->frame_size);
410 AV_WB8 (alac_extradata+17, avctx->bits_per_coded_sample);
411 AV_WB8 (alac_extradata+21, avctx->channels);
412 AV_WB32(alac_extradata+24, s->max_coded_frame_size);
413 AV_WB32(alac_extradata+28, avctx->sample_rate*avctx->channels*avctx->bits_per_coded_sample); // average bitrate
414 AV_WB32(alac_extradata+32, avctx->sample_rate);
415
416 // Set relevant extradata fields
417 if(s->compression_level > 0) {
418 AV_WB8(alac_extradata+18, s->rc.history_mult);
419 AV_WB8(alac_extradata+19, s->rc.initial_history);
420 AV_WB8(alac_extradata+20, s->rc.k_modifier);
421 }
422
423 s->min_prediction_order = DEFAULT_MIN_PRED_ORDER;
424 if(avctx->min_prediction_order >= 0) {
425 if(avctx->min_prediction_order < MIN_LPC_ORDER ||
426 avctx->min_prediction_order > ALAC_MAX_LPC_ORDER) {
427 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", avctx->min_prediction_order);
428 return -1;
429 }
430
431 s->min_prediction_order = avctx->min_prediction_order;
432 }
433
434 s->max_prediction_order = DEFAULT_MAX_PRED_ORDER;
435 if(avctx->max_prediction_order >= 0) {
436 if(avctx->max_prediction_order < MIN_LPC_ORDER ||
437 avctx->max_prediction_order > ALAC_MAX_LPC_ORDER) {
438 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", avctx->max_prediction_order);
439 return -1;
440 }
441
442 s->max_prediction_order = avctx->max_prediction_order;
443 }
444
445 if(s->max_prediction_order < s->min_prediction_order) {
446 av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
447 s->min_prediction_order, s->max_prediction_order);
448 return -1;
449 }
450
451 avctx->extradata = alac_extradata;
452 avctx->extradata_size = ALAC_EXTRADATA_SIZE;
453
454 avctx->coded_frame = avcodec_alloc_frame();
455 avctx->coded_frame->key_frame = 1;
456
457 s->avctx = avctx;
458 dsputil_init(&s->dspctx, avctx);
459
460 return 0;
461 }
462
463 static int alac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
464 int buf_size, void *data)
465 {
466 AlacEncodeContext *s = avctx->priv_data;
467 PutBitContext *pb = &s->pbctx;
468 int i, out_bytes, verbatim_flag = 0;
469
470 if(avctx->frame_size > DEFAULT_FRAME_SIZE) {
471 av_log(avctx, AV_LOG_ERROR, "input frame size exceeded\n");
472 return -1;
473 }
474
475 if(buf_size < 2*s->max_coded_frame_size) {
476 av_log(avctx, AV_LOG_ERROR, "buffer size is too small\n");
477 return -1;
478 }
479
480 verbatim:
481 init_put_bits(pb, frame, buf_size);
482
483 if((s->compression_level == 0) || verbatim_flag) {
484 // Verbatim mode
485 int16_t *samples = data;
486 write_frame_header(s, 1);
487 for(i=0; i<avctx->frame_size*avctx->channels; i++) {
488 put_sbits(pb, 16, *samples++);
489 }
490 } else {
491 init_sample_buffers(s, data);
492 write_frame_header(s, 0);
493 write_compressed_frame(s);
494 }
495
496 put_bits(pb, 3, 7);
497 flush_put_bits(pb);
498 out_bytes = put_bits_count(pb) >> 3;
499
500 if(out_bytes > s->max_coded_frame_size) {
501 /* frame too large. use verbatim mode */
502 if(verbatim_flag || (s->compression_level == 0)) {
503 /* still too large. must be an error. */
504 av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
505 return -1;
506 }
507 verbatim_flag = 1;
508 goto verbatim;
509 }
510
511 return out_bytes;
512 }
513
514 static av_cold int alac_encode_close(AVCodecContext *avctx)
515 {
516 av_freep(&avctx->extradata);
517 avctx->extradata_size = 0;
518 av_freep(&avctx->coded_frame);
519 return 0;
520 }
521
522 AVCodec alac_encoder = {
523 "alac",
524 AVMEDIA_TYPE_AUDIO,
525 CODEC_ID_ALAC,
526 sizeof(AlacEncodeContext),
527 alac_encode_init,
528 alac_encode_frame,
529 alac_encode_close,
530 .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
531 .sample_fmts = (const enum SampleFormat[]){ SAMPLE_FMT_S16, SAMPLE_FMT_NONE},
532 .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
533 };