9ad91f238e0baecf3fd064b27caec84fb6a305e9
[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
26 #include "libavutil/avassert.h"
27 #include "libavutil/channel_layout.h"
28 #include "libavutil/opt.h"
29 #include "avcodec.h"
30 #include "internal.h"
31 #include "get_bits.h"
32 #include "put_bits.h"
33
34 /**
35 * G.726 11bit float.
36 * G.726 Standard uses rather odd 11bit floating point arithmentic for
37 * numerous occasions. It's a mystery to me why they did it this way
38 * instead of simply using 32bit integer arithmetic.
39 */
40 typedef struct Float11 {
41 uint8_t sign; /**< 1bit sign */
42 uint8_t exp; /**< 4bit exponent */
43 uint8_t mant; /**< 6bit mantissa */
44 } Float11;
45
46 static inline Float11* i2f(int i, Float11* f)
47 {
48 f->sign = (i < 0);
49 if (f->sign)
50 i = -i;
51 f->exp = av_log2_16bit(i) + !!i;
52 f->mant = i? (i<<6) >> f->exp : 1<<5;
53 return f;
54 }
55
56 static inline int16_t mult(Float11* f1, Float11* f2)
57 {
58 int res, exp;
59
60 exp = f1->exp + f2->exp;
61 res = (((f1->mant * f2->mant) + 0x30) >> 4);
62 res = exp > 19 ? res << (exp - 19) : res >> (19 - exp);
63 return (f1->sign ^ f2->sign) ? -res : res;
64 }
65
66 static inline int sgn(int value)
67 {
68 return (value < 0) ? -1 : 1;
69 }
70
71 typedef struct G726Tables {
72 const int* quant; /**< quantization table */
73 const int16_t* iquant; /**< inverse quantization table */
74 const int16_t* W; /**< special table #1 ;-) */
75 const uint8_t* F; /**< special table #2 */
76 } G726Tables;
77
78 typedef struct G726Context {
79 AVClass *class;
80 G726Tables tbls; /**< static tables needed for computation */
81
82 Float11 sr[2]; /**< prev. reconstructed samples */
83 Float11 dq[6]; /**< prev. difference */
84 int a[2]; /**< second order predictor coeffs */
85 int b[6]; /**< sixth order predictor coeffs */
86 int pk[2]; /**< signs of prev. 2 sez + dq */
87
88 int ap; /**< scale factor control */
89 int yu; /**< fast scale factor */
90 int yl; /**< slow scale factor */
91 int dms; /**< short average magnitude of F[i] */
92 int dml; /**< long average magnitude of F[i] */
93 int td; /**< tone detect */
94
95 int se; /**< estimated signal for the next iteration */
96 int sez; /**< estimated second order prediction */
97 int y; /**< quantizer scaling factor for the next iteration */
98 int code_size;
99 } G726Context;
100
101 static const int quant_tbl16[] = /**< 16kbit/s 2bits per sample */
102 { 260, INT_MAX };
103 static const int16_t iquant_tbl16[] =
104 { 116, 365, 365, 116 };
105 static const int16_t W_tbl16[] =
106 { -22, 439, 439, -22 };
107 static const uint8_t F_tbl16[] =
108 { 0, 7, 7, 0 };
109
110 static const int quant_tbl24[] = /**< 24kbit/s 3bits per sample */
111 { 7, 217, 330, INT_MAX };
112 static const int16_t iquant_tbl24[] =
113 { INT16_MIN, 135, 273, 373, 373, 273, 135, INT16_MIN };
114 static const int16_t W_tbl24[] =
115 { -4, 30, 137, 582, 582, 137, 30, -4 };
116 static const uint8_t F_tbl24[] =
117 { 0, 1, 2, 7, 7, 2, 1, 0 };
118
119 static const int quant_tbl32[] = /**< 32kbit/s 4bits per sample */
120 { -125, 79, 177, 245, 299, 348, 399, INT_MAX };
121 static const int16_t iquant_tbl32[] =
122 { INT16_MIN, 4, 135, 213, 273, 323, 373, 425,
123 425, 373, 323, 273, 213, 135, 4, INT16_MIN };
124 static const int16_t W_tbl32[] =
125 { -12, 18, 41, 64, 112, 198, 355, 1122,
126 1122, 355, 198, 112, 64, 41, 18, -12};
127 static const uint8_t F_tbl32[] =
128 { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 };
129
130 static const int quant_tbl40[] = /**< 40kbit/s 5bits per sample */
131 { -122, -16, 67, 138, 197, 249, 297, 338,
132 377, 412, 444, 474, 501, 527, 552, INT_MAX };
133 static const int16_t iquant_tbl40[] =
134 { INT16_MIN, -66, 28, 104, 169, 224, 274, 318,
135 358, 395, 429, 459, 488, 514, 539, 566,
136 566, 539, 514, 488, 459, 429, 395, 358,
137 318, 274, 224, 169, 104, 28, -66, INT16_MIN };
138 static const int16_t W_tbl40[] =
139 { 14, 14, 24, 39, 40, 41, 58, 100,
140 141, 179, 219, 280, 358, 440, 529, 696,
141 696, 529, 440, 358, 280, 219, 179, 141,
142 100, 58, 41, 40, 39, 24, 14, 14 };
143 static const uint8_t F_tbl40[] =
144 { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6,
145 6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };
146
147 static const G726Tables G726Tables_pool[] =
148 {{ quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
149 { quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
150 { quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
151 { quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }};
152
153
154 /**
155 * Para 4.2.2 page 18: Adaptive quantizer.
156 */
157 static inline uint8_t quant(G726Context* c, int d)
158 {
159 int sign, exp, i, dln;
160
161 sign = i = 0;
162 if (d < 0) {
163 sign = 1;
164 d = -d;
165 }
166 exp = av_log2_16bit(d);
167 dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2);
168
169 while (c->tbls.quant[i] < INT_MAX && c->tbls.quant[i] < dln)
170 ++i;
171
172 if (sign)
173 i = ~i;
174 if (c->code_size != 2 && i == 0) /* I'm not sure this is a good idea */
175 i = 0xff;
176
177 return i;
178 }
179
180 /**
181 * Para 4.2.3 page 22: Inverse adaptive quantizer.
182 */
183 static inline int16_t inverse_quant(G726Context* c, int i)
184 {
185 int dql, dex, dqt;
186
187 dql = c->tbls.iquant[i] + (c->y >> 2);
188 dex = (dql>>7) & 0xf; /* 4bit exponent */
189 dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */
190 return (dql < 0) ? 0 : ((dqt<<dex) >> 7);
191 }
192
193 static int16_t g726_decode(G726Context* c, int I)
194 {
195 int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
196 Float11 f;
197 int I_sig= I >> (c->code_size - 1);
198
199 dq = inverse_quant(c, I);
200
201 /* Transition detect */
202 ylint = (c->yl >> 15);
203 ylfrac = (c->yl >> 10) & 0x1f;
204 thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint;
205 tr= (c->td == 1 && dq > ((3*thr2)>>2));
206
207 if (I_sig) /* get the sign */
208 dq = -dq;
209 re_signal = c->se + dq;
210
211 /* Update second order predictor coefficient A2 and A1 */
212 pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
213 dq0 = dq ? sgn(dq) : 0;
214 if (tr) {
215 c->a[0] = 0;
216 c->a[1] = 0;
217 for (i=0; i<6; i++)
218 c->b[i] = 0;
219 } else {
220 /* This is a bit crazy, but it really is +255 not +256 */
221 fa1 = av_clip_intp2((-c->a[0]*c->pk[0]*pk0)>>5, 8);
222
223 c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7);
224 c->a[1] = av_clip(c->a[1], -12288, 12288);
225 c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8);
226 c->a[0] = av_clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]);
227
228 for (i=0; i<6; i++)
229 c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8);
230 }
231
232 /* Update Dq and Sr and Pk */
233 c->pk[1] = c->pk[0];
234 c->pk[0] = pk0 ? pk0 : 1;
235 c->sr[1] = c->sr[0];
236 i2f(re_signal, &c->sr[0]);
237 for (i=5; i>0; i--)
238 c->dq[i] = c->dq[i-1];
239 i2f(dq, &c->dq[0]);
240 c->dq[0].sign = I_sig; /* Isn't it crazy ?!?! */
241
242 c->td = c->a[1] < -11776;
243
244 /* Update Ap */
245 c->dms += (c->tbls.F[I]<<4) + ((- c->dms) >> 5);
246 c->dml += (c->tbls.F[I]<<4) + ((- c->dml) >> 7);
247 if (tr)
248 c->ap = 256;
249 else {
250 c->ap += (-c->ap) >> 4;
251 if (c->y <= 1535 || c->td || abs((c->dms << 2) - c->dml) >= (c->dml >> 3))
252 c->ap += 0x20;
253 }
254
255 /* Update Yu and Yl */
256 c->yu = av_clip(c->y + c->tbls.W[I] + ((-c->y)>>5), 544, 5120);
257 c->yl += c->yu + ((-c->yl)>>6);
258
259 /* Next iteration for Y */
260 al = (c->ap >= 256) ? 1<<6 : c->ap >> 2;
261 c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6;
262
263 /* Next iteration for SE and SEZ */
264 c->se = 0;
265 for (i=0; i<6; i++)
266 c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]);
267 c->sez = c->se >> 1;
268 for (i=0; i<2; i++)
269 c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]);
270 c->se >>= 1;
271
272 return av_clip(re_signal << 2, -0xffff, 0xffff);
273 }
274
275 static av_cold int g726_reset(G726Context *c)
276 {
277 int i;
278
279 c->tbls = G726Tables_pool[c->code_size - 2];
280 for (i=0; i<2; i++) {
281 c->sr[i].mant = 1<<5;
282 c->pk[i] = 1;
283 }
284 for (i=0; i<6; i++) {
285 c->dq[i].mant = 1<<5;
286 }
287 c->yu = 544;
288 c->yl = 34816;
289
290 c->y = 544;
291
292 return 0;
293 }
294
295 #if CONFIG_ADPCM_G726_ENCODER
296 static int16_t g726_encode(G726Context* c, int16_t sig)
297 {
298 uint8_t i;
299
300 i = quant(c, sig/4 - c->se) & ((1<<c->code_size) - 1);
301 g726_decode(c, i);
302 return i;
303 }
304
305 /* Interfacing to the libavcodec */
306
307 static av_cold int g726_encode_init(AVCodecContext *avctx)
308 {
309 G726Context* c = avctx->priv_data;
310
311 if (avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL &&
312 avctx->sample_rate != 8000) {
313 av_log(avctx, AV_LOG_ERROR, "Sample rates other than 8kHz are not "
314 "allowed when the compliance level is higher than unofficial. "
315 "Resample or reduce the compliance level.\n");
316 return AVERROR(EINVAL);
317 }
318 av_assert0(avctx->sample_rate > 0);
319
320 if(avctx->channels != 1){
321 av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
322 return AVERROR(EINVAL);
323 }
324
325 if (avctx->bit_rate)
326 c->code_size = (avctx->bit_rate + avctx->sample_rate/2) / avctx->sample_rate;
327
328 c->code_size = av_clip(c->code_size, 2, 5);
329 avctx->bit_rate = c->code_size * avctx->sample_rate;
330 avctx->bits_per_coded_sample = c->code_size;
331
332 g726_reset(c);
333
334 /* select a frame size that will end on a byte boundary and have a size of
335 approximately 1024 bytes */
336 avctx->frame_size = ((int[]){ 4096, 2736, 2048, 1640 })[c->code_size - 2];
337
338 return 0;
339 }
340
341 static int g726_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
342 const AVFrame *frame, int *got_packet_ptr)
343 {
344 G726Context *c = avctx->priv_data;
345 const int16_t *samples = (const int16_t *)frame->data[0];
346 PutBitContext pb;
347 int i, ret, out_size;
348
349 out_size = (frame->nb_samples * c->code_size + 7) / 8;
350 if ((ret = ff_alloc_packet(avpkt, out_size))) {
351 av_log(avctx, AV_LOG_ERROR, "Error getting output packet\n");
352 return ret;
353 }
354 init_put_bits(&pb, avpkt->data, avpkt->size);
355
356 for (i = 0; i < frame->nb_samples; i++)
357 put_bits(&pb, c->code_size, g726_encode(c, *samples++));
358
359 flush_put_bits(&pb);
360
361 avpkt->size = out_size;
362 *got_packet_ptr = 1;
363 return 0;
364 }
365
366 #define OFFSET(x) offsetof(G726Context, x)
367 #define AE AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
368 static const AVOption options[] = {
369 { "code_size", "Bits per code", OFFSET(code_size), AV_OPT_TYPE_INT, { .i64 = 4 }, 2, 5, AE },
370 { NULL },
371 };
372
373 static const AVClass class = {
374 .class_name = "g726",
375 .item_name = av_default_item_name,
376 .option = options,
377 .version = LIBAVUTIL_VERSION_INT,
378 };
379
380 static const AVCodecDefault defaults[] = {
381 { "b", "0" },
382 { NULL },
383 };
384
385 AVCodec ff_adpcm_g726_encoder = {
386 .name = "g726",
387 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
388 .type = AVMEDIA_TYPE_AUDIO,
389 .id = AV_CODEC_ID_ADPCM_G726,
390 .priv_data_size = sizeof(G726Context),
391 .init = g726_encode_init,
392 .encode2 = g726_encode_frame,
393 .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
394 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
395 AV_SAMPLE_FMT_NONE },
396 .priv_class = &class,
397 .defaults = defaults,
398 };
399 #endif
400
401 #if CONFIG_ADPCM_G726_DECODER
402 static av_cold int g726_decode_init(AVCodecContext *avctx)
403 {
404 G726Context* c = avctx->priv_data;
405
406 avctx->channels = 1;
407 avctx->channel_layout = AV_CH_LAYOUT_MONO;
408
409 c->code_size = avctx->bits_per_coded_sample;
410 if (c->code_size < 2 || c->code_size > 5) {
411 av_log(avctx, AV_LOG_ERROR, "Invalid number of bits %d\n", c->code_size);
412 return AVERROR(EINVAL);
413 }
414 g726_reset(c);
415
416 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
417
418 return 0;
419 }
420
421 static int g726_decode_frame(AVCodecContext *avctx, void *data,
422 int *got_frame_ptr, AVPacket *avpkt)
423 {
424 AVFrame *frame = data;
425 const uint8_t *buf = avpkt->data;
426 int buf_size = avpkt->size;
427 G726Context *c = avctx->priv_data;
428 int16_t *samples;
429 GetBitContext gb;
430 int out_samples, ret;
431
432 out_samples = buf_size * 8 / c->code_size;
433
434 /* get output buffer */
435 frame->nb_samples = out_samples;
436 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
437 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
438 return ret;
439 }
440 samples = (int16_t *)frame->data[0];
441
442 init_get_bits(&gb, buf, buf_size * 8);
443
444 while (out_samples--)
445 *samples++ = g726_decode(c, get_bits(&gb, c->code_size));
446
447 if (get_bits_left(&gb) > 0)
448 av_log(avctx, AV_LOG_ERROR, "Frame invalidly split, missing parser?\n");
449
450 *got_frame_ptr = 1;
451
452 return buf_size;
453 }
454
455 static void g726_decode_flush(AVCodecContext *avctx)
456 {
457 G726Context *c = avctx->priv_data;
458 g726_reset(c);
459 }
460
461 AVCodec ff_adpcm_g726_decoder = {
462 .name = "g726",
463 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
464 .type = AVMEDIA_TYPE_AUDIO,
465 .id = AV_CODEC_ID_ADPCM_G726,
466 .priv_data_size = sizeof(G726Context),
467 .init = g726_decode_init,
468 .decode = g726_decode_frame,
469 .flush = g726_decode_flush,
470 .capabilities = CODEC_CAP_DR1,
471 };
472 #endif