c989d075eb17c33fe77aca2552b1e103b2213dba
[libav.git] / libavcodec / g726.c
1 /*
2 * G.726 ADPCM audio codec
3 * Copyright (c) 2004 Roman Shaposhnik
4 *
5 * This is a very straightforward rendition of the G.726
6 * Section 4 "Computational Details".
7 *
8 * This file is part of Libav.
9 *
10 * Libav is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
14 *
15 * Libav is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
19 *
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with Libav; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 */
24 #include <limits.h>
25 #include "libavutil/avassert.h"
26 #include "libavutil/opt.h"
27 #include "avcodec.h"
28 #include "internal.h"
29 #include "get_bits.h"
30 #include "put_bits.h"
31
32 /**
33 * G.726 11bit float.
34 * G.726 Standard uses rather odd 11bit floating point arithmentic for
35 * numerous occasions. It's a mistery to me why they did it this way
36 * instead of simply using 32bit integer arithmetic.
37 */
38 typedef struct Float11 {
39 uint8_t sign; /**< 1bit sign */
40 uint8_t exp; /**< 4bit exponent */
41 uint8_t mant; /**< 6bit mantissa */
42 } Float11;
43
44 static inline Float11* i2f(int i, Float11* f)
45 {
46 f->sign = (i < 0);
47 if (f->sign)
48 i = -i;
49 f->exp = av_log2_16bit(i) + !!i;
50 f->mant = i? (i<<6) >> f->exp : 1<<5;
51 return f;
52 }
53
54 static inline int16_t mult(Float11* f1, Float11* f2)
55 {
56 int res, exp;
57
58 exp = f1->exp + f2->exp;
59 res = (((f1->mant * f2->mant) + 0x30) >> 4);
60 res = exp > 19 ? res << (exp - 19) : res >> (19 - exp);
61 return (f1->sign ^ f2->sign) ? -res : res;
62 }
63
64 static inline int sgn(int value)
65 {
66 return (value < 0) ? -1 : 1;
67 }
68
69 typedef struct G726Tables {
70 const int* quant; /**< quantization table */
71 const int16_t* iquant; /**< inverse quantization table */
72 const int16_t* W; /**< special table #1 ;-) */
73 const uint8_t* F; /**< special table #2 */
74 } G726Tables;
75
76 typedef struct G726Context {
77 AVClass *class;
78 AVFrame frame;
79 G726Tables tbls; /**< static tables needed for computation */
80
81 Float11 sr[2]; /**< prev. reconstructed samples */
82 Float11 dq[6]; /**< prev. difference */
83 int a[2]; /**< second order predictor coeffs */
84 int b[6]; /**< sixth order predictor coeffs */
85 int pk[2]; /**< signs of prev. 2 sez + dq */
86
87 int ap; /**< scale factor control */
88 int yu; /**< fast scale factor */
89 int yl; /**< slow scale factor */
90 int dms; /**< short average magnitude of F[i] */
91 int dml; /**< long average magnitude of F[i] */
92 int td; /**< tone detect */
93
94 int se; /**< estimated signal for the next iteration */
95 int sez; /**< estimated second order prediction */
96 int y; /**< quantizer scaling factor for the next iteration */
97 int code_size;
98 } G726Context;
99
100 static const int quant_tbl16[] = /**< 16kbit/s 2bits per sample */
101 { 260, INT_MAX };
102 static const int16_t iquant_tbl16[] =
103 { 116, 365, 365, 116 };
104 static const int16_t W_tbl16[] =
105 { -22, 439, 439, -22 };
106 static const uint8_t F_tbl16[] =
107 { 0, 7, 7, 0 };
108
109 static const int quant_tbl24[] = /**< 24kbit/s 3bits per sample */
110 { 7, 217, 330, INT_MAX };
111 static const int16_t iquant_tbl24[] =
112 { INT16_MIN, 135, 273, 373, 373, 273, 135, INT16_MIN };
113 static const int16_t W_tbl24[] =
114 { -4, 30, 137, 582, 582, 137, 30, -4 };
115 static const uint8_t F_tbl24[] =
116 { 0, 1, 2, 7, 7, 2, 1, 0 };
117
118 static const int quant_tbl32[] = /**< 32kbit/s 4bits per sample */
119 { -125, 79, 177, 245, 299, 348, 399, INT_MAX };
120 static const int16_t iquant_tbl32[] =
121 { INT16_MIN, 4, 135, 213, 273, 323, 373, 425,
122 425, 373, 323, 273, 213, 135, 4, INT16_MIN };
123 static const int16_t W_tbl32[] =
124 { -12, 18, 41, 64, 112, 198, 355, 1122,
125 1122, 355, 198, 112, 64, 41, 18, -12};
126 static const uint8_t F_tbl32[] =
127 { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 };
128
129 static const int quant_tbl40[] = /**< 40kbit/s 5bits per sample */
130 { -122, -16, 67, 138, 197, 249, 297, 338,
131 377, 412, 444, 474, 501, 527, 552, INT_MAX };
132 static const int16_t iquant_tbl40[] =
133 { INT16_MIN, -66, 28, 104, 169, 224, 274, 318,
134 358, 395, 429, 459, 488, 514, 539, 566,
135 566, 539, 514, 488, 459, 429, 395, 358,
136 318, 274, 224, 169, 104, 28, -66, INT16_MIN };
137 static const int16_t W_tbl40[] =
138 { 14, 14, 24, 39, 40, 41, 58, 100,
139 141, 179, 219, 280, 358, 440, 529, 696,
140 696, 529, 440, 358, 280, 219, 179, 141,
141 100, 58, 41, 40, 39, 24, 14, 14 };
142 static const uint8_t F_tbl40[] =
143 { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6,
144 6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };
145
146 static const G726Tables G726Tables_pool[] =
147 {{ quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
148 { quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
149 { quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
150 { quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }};
151
152
153 /**
154 * Para 4.2.2 page 18: Adaptive quantizer.
155 */
156 static inline uint8_t quant(G726Context* c, int d)
157 {
158 int sign, exp, i, dln;
159
160 sign = i = 0;
161 if (d < 0) {
162 sign = 1;
163 d = -d;
164 }
165 exp = av_log2_16bit(d);
166 dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2);
167
168 while (c->tbls.quant[i] < INT_MAX && c->tbls.quant[i] < dln)
169 ++i;
170
171 if (sign)
172 i = ~i;
173 if (c->code_size != 2 && i == 0) /* I'm not sure this is a good idea */
174 i = 0xff;
175
176 return i;
177 }
178
179 /**
180 * Para 4.2.3 page 22: Inverse adaptive quantizer.
181 */
182 static inline int16_t inverse_quant(G726Context* c, int i)
183 {
184 int dql, dex, dqt;
185
186 dql = c->tbls.iquant[i] + (c->y >> 2);
187 dex = (dql>>7) & 0xf; /* 4bit exponent */
188 dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */
189 return (dql < 0) ? 0 : ((dqt<<dex) >> 7);
190 }
191
192 static int16_t g726_decode(G726Context* c, int I)
193 {
194 int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
195 Float11 f;
196 int I_sig= I >> (c->code_size - 1);
197
198 dq = inverse_quant(c, I);
199
200 /* Transition detect */
201 ylint = (c->yl >> 15);
202 ylfrac = (c->yl >> 10) & 0x1f;
203 thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint;
204 tr= (c->td == 1 && dq > ((3*thr2)>>2));
205
206 if (I_sig) /* get the sign */
207 dq = -dq;
208 re_signal = c->se + dq;
209
210 /* Update second order predictor coefficient A2 and A1 */
211 pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
212 dq0 = dq ? sgn(dq) : 0;
213 if (tr) {
214 c->a[0] = 0;
215 c->a[1] = 0;
216 for (i=0; i<6; i++)
217 c->b[i] = 0;
218 } else {
219 /* This is a bit crazy, but it really is +255 not +256 */
220 fa1 = av_clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255);
221
222 c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7);
223 c->a[1] = av_clip(c->a[1], -12288, 12288);
224 c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8);
225 c->a[0] = av_clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]);
226
227 for (i=0; i<6; i++)
228 c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8);
229 }
230
231 /* Update Dq and Sr and Pk */
232 c->pk[1] = c->pk[0];
233 c->pk[0] = pk0 ? pk0 : 1;
234 c->sr[1] = c->sr[0];
235 i2f(re_signal, &c->sr[0]);
236 for (i=5; i>0; i--)
237 c->dq[i] = c->dq[i-1];
238 i2f(dq, &c->dq[0]);
239 c->dq[0].sign = I_sig; /* Isn't it crazy ?!?! */
240
241 c->td = c->a[1] < -11776;
242
243 /* Update Ap */
244 c->dms += (c->tbls.F[I]<<4) + ((- c->dms) >> 5);
245 c->dml += (c->tbls.F[I]<<4) + ((- c->dml) >> 7);
246 if (tr)
247 c->ap = 256;
248 else {
249 c->ap += (-c->ap) >> 4;
250 if (c->y <= 1535 || c->td || abs((c->dms << 2) - c->dml) >= (c->dml >> 3))
251 c->ap += 0x20;
252 }
253
254 /* Update Yu and Yl */
255 c->yu = av_clip(c->y + c->tbls.W[I] + ((-c->y)>>5), 544, 5120);
256 c->yl += c->yu + ((-c->yl)>>6);
257
258 /* Next iteration for Y */
259 al = (c->ap >= 256) ? 1<<6 : c->ap >> 2;
260 c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6;
261
262 /* Next iteration for SE and SEZ */
263 c->se = 0;
264 for (i=0; i<6; i++)
265 c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]);
266 c->sez = c->se >> 1;
267 for (i=0; i<2; i++)
268 c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]);
269 c->se >>= 1;
270
271 return av_clip(re_signal << 2, -0xffff, 0xffff);
272 }
273
274 static av_cold int g726_reset(G726Context *c)
275 {
276 int i;
277
278 c->tbls = G726Tables_pool[c->code_size - 2];
279 for (i=0; i<2; i++) {
280 c->sr[i].mant = 1<<5;
281 c->pk[i] = 1;
282 }
283 for (i=0; i<6; i++) {
284 c->dq[i].mant = 1<<5;
285 }
286 c->yu = 544;
287 c->yl = 34816;
288
289 c->y = 544;
290
291 return 0;
292 }
293
294 #if CONFIG_ADPCM_G726_ENCODER
295 static int16_t g726_encode(G726Context* c, int16_t sig)
296 {
297 uint8_t i;
298
299 i = quant(c, sig/4 - c->se) & ((1<<c->code_size) - 1);
300 g726_decode(c, i);
301 return i;
302 }
303
304 /* Interfacing to the libavcodec */
305
306 static av_cold int g726_encode_init(AVCodecContext *avctx)
307 {
308 G726Context* c = avctx->priv_data;
309
310 if (avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL &&
311 avctx->sample_rate != 8000) {
312 av_log(avctx, AV_LOG_ERROR, "Sample rates other than 8kHz are not "
313 "allowed when the compliance level is higher than unofficial. "
314 "Resample or reduce the compliance level.\n");
315 return AVERROR(EINVAL);
316 }
317 av_assert0(avctx->sample_rate > 0);
318
319 if(avctx->channels != 1){
320 av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
321 return AVERROR(EINVAL);
322 }
323
324 if (avctx->bit_rate)
325 c->code_size = (avctx->bit_rate + avctx->sample_rate/2) / avctx->sample_rate;
326
327 c->code_size = av_clip(c->code_size, 2, 5);
328 avctx->bit_rate = c->code_size * avctx->sample_rate;
329 avctx->bits_per_coded_sample = c->code_size;
330
331 g726_reset(c);
332
333 #if FF_API_OLD_ENCODE_AUDIO
334 avctx->coded_frame = avcodec_alloc_frame();
335 if (!avctx->coded_frame)
336 return AVERROR(ENOMEM);
337 avctx->coded_frame->key_frame = 1;
338 #endif
339
340 /* select a frame size that will end on a byte boundary and have a size of
341 approximately 1024 bytes */
342 avctx->frame_size = ((int[]){ 4096, 2736, 2048, 1640 })[c->code_size - 2];
343
344 return 0;
345 }
346
347 #if FF_API_OLD_ENCODE_AUDIO
348 static av_cold int g726_encode_close(AVCodecContext *avctx)
349 {
350 av_freep(&avctx->coded_frame);
351 return 0;
352 }
353 #endif
354
355 static int g726_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
356 const AVFrame *frame, int *got_packet_ptr)
357 {
358 G726Context *c = avctx->priv_data;
359 const int16_t *samples = (const int16_t *)frame->data[0];
360 PutBitContext pb;
361 int i, ret, out_size;
362
363 out_size = (frame->nb_samples * c->code_size + 7) / 8;
364 if ((ret = ff_alloc_packet(avpkt, out_size))) {
365 av_log(avctx, AV_LOG_ERROR, "Error getting output packet\n");
366 return ret;
367 }
368 init_put_bits(&pb, avpkt->data, avpkt->size);
369
370 for (i = 0; i < frame->nb_samples; i++)
371 put_bits(&pb, c->code_size, g726_encode(c, *samples++));
372
373 flush_put_bits(&pb);
374
375 avpkt->size = out_size;
376 *got_packet_ptr = 1;
377 return 0;
378 }
379
380 #define OFFSET(x) offsetof(G726Context, x)
381 #define AE AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
382 static const AVOption options[] = {
383 { "code_size", "Bits per code", OFFSET(code_size), AV_OPT_TYPE_INT, { 4 }, 2, 5, AE },
384 { NULL },
385 };
386
387 static const AVClass class = {
388 .class_name = "g726",
389 .item_name = av_default_item_name,
390 .option = options,
391 .version = LIBAVUTIL_VERSION_INT,
392 };
393
394 static const AVCodecDefault defaults[] = {
395 { "b", "0" },
396 { NULL },
397 };
398
399 AVCodec ff_adpcm_g726_encoder = {
400 .name = "g726",
401 .type = AVMEDIA_TYPE_AUDIO,
402 .id = CODEC_ID_ADPCM_G726,
403 .priv_data_size = sizeof(G726Context),
404 .init = g726_encode_init,
405 .encode2 = g726_encode_frame,
406 #if FF_API_OLD_ENCODE_AUDIO
407 .close = g726_encode_close,
408 #endif
409 .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
410 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
411 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
412 .priv_class = &class,
413 .defaults = defaults,
414 };
415 #endif
416
417 #if CONFIG_ADPCM_G726_DECODER
418 static av_cold int g726_decode_init(AVCodecContext *avctx)
419 {
420 G726Context* c = avctx->priv_data;
421
422 if (avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
423 avctx->sample_rate != 8000) {
424 av_log(avctx, AV_LOG_ERROR, "Only 8kHz sample rate is allowed when "
425 "the compliance level is strict. Reduce the compliance level "
426 "if you wish to decode the stream anyway.\n");
427 return AVERROR(EINVAL);
428 }
429
430 if(avctx->channels != 1){
431 av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
432 return AVERROR(EINVAL);
433 }
434
435 c->code_size = avctx->bits_per_coded_sample;
436 if (c->code_size < 2 || c->code_size > 5) {
437 av_log(avctx, AV_LOG_ERROR, "Invalid number of bits %d\n", c->code_size);
438 return AVERROR(EINVAL);
439 }
440 g726_reset(c);
441
442 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
443
444 avcodec_get_frame_defaults(&c->frame);
445 avctx->coded_frame = &c->frame;
446
447 return 0;
448 }
449
450 static int g726_decode_frame(AVCodecContext *avctx, void *data,
451 int *got_frame_ptr, AVPacket *avpkt)
452 {
453 const uint8_t *buf = avpkt->data;
454 int buf_size = avpkt->size;
455 G726Context *c = avctx->priv_data;
456 int16_t *samples;
457 GetBitContext gb;
458 int out_samples, ret;
459
460 out_samples = buf_size * 8 / c->code_size;
461
462 /* get output buffer */
463 c->frame.nb_samples = out_samples;
464 if ((ret = avctx->get_buffer(avctx, &c->frame)) < 0) {
465 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
466 return ret;
467 }
468 samples = (int16_t *)c->frame.data[0];
469
470 init_get_bits(&gb, buf, buf_size * 8);
471
472 while (out_samples--)
473 *samples++ = g726_decode(c, get_bits(&gb, c->code_size));
474
475 if (get_bits_left(&gb) > 0)
476 av_log(avctx, AV_LOG_ERROR, "Frame invalidly split, missing parser?\n");
477
478 *got_frame_ptr = 1;
479 *(AVFrame *)data = c->frame;
480
481 return buf_size;
482 }
483
484 static void g726_decode_flush(AVCodecContext *avctx)
485 {
486 G726Context *c = avctx->priv_data;
487 g726_reset(c);
488 }
489
490 AVCodec ff_adpcm_g726_decoder = {
491 .name = "g726",
492 .type = AVMEDIA_TYPE_AUDIO,
493 .id = CODEC_ID_ADPCM_G726,
494 .priv_data_size = sizeof(G726Context),
495 .init = g726_decode_init,
496 .decode = g726_decode_frame,
497 .flush = g726_decode_flush,
498 .capabilities = CODEC_CAP_DR1,
499 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
500 };
501 #endif