Do not mix declarations and statements.
[libav.git] / libavcodec / g726.c
CommitLineData
e5966052 1/*
115329f1 2 * G.726 ADPCM audio codec
e5966052
RS
3 * Copyright (c) 2004 Roman Shaposhnik.
4 *
5 * This is a very straightforward rendition of the G.726
115329f1 6 * Section 4 "Computational Details".
e5966052 7 *
b78e7197
DB
8 * This file is part of FFmpeg.
9 *
10 * FFmpeg is free software; you can redistribute it and/or
e5966052
RS
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
b78e7197 13 * version 2.1 of the License, or (at your option) any later version.
e5966052 14 *
b78e7197 15 * FFmpeg is distributed in the hope that it will be useful,
e5966052
RS
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
b78e7197 21 * License along with FFmpeg; if not, write to the Free Software
5509bffa 22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
e5966052
RS
23 */
24#include <limits.h>
25#include "avcodec.h"
26#include "common.h"
caa336b4 27#include "bitstream.h"
e5966052 28
ef9c5d84
MN
29/**
30 * G.726 11bit float.
115329f1 31 * G.726 Standard uses rather odd 11bit floating point arithmentic for
e5966052
RS
32 * numerous occasions. It's a mistery to me why they did it this way
33 * instead of simply using 32bit integer arithmetic.
34 */
35typedef struct Float11 {
bb270c08
DB
36 int sign; /**< 1bit sign */
37 int exp; /**< 4bit exponent */
38 int mant; /**< 6bit mantissa */
e5966052
RS
39} Float11;
40
41static inline Float11* i2f(int16_t i, Float11* f)
42{
bb270c08
DB
43 f->sign = (i < 0);
44 if (f->sign)
45 i = -i;
46 f->exp = av_log2_16bit(i) + !!i;
47 f->mant = i? (i<<6) >> f->exp :
48 1<<5;
49 return f;
e5966052
RS
50}
51
52static inline int16_t mult(Float11* f1, Float11* f2)
53{
bb270c08 54 int res, exp;
e5966052 55
bb270c08
DB
56 exp = f1->exp + f2->exp;
57 res = (((f1->mant * f2->mant) + 0x30) >> 4) << 7;
58 res = exp > 26 ? res << (exp - 26) : res >> (26 - exp);
59 return (f1->sign ^ f2->sign) ? -res : res;
e5966052
RS
60}
61
e5966052
RS
62static inline int sgn(int value)
63{
64 return (value < 0) ? -1 : 1;
65}
66
67typedef struct G726Tables {
bb270c08
DB
68 int bits; /**< bits per sample */
69 int* quant; /**< quantization table */
70 int* iquant; /**< inverse quantization table */
71 int* W; /**< special table #1 ;-) */
72 int* F; /**< special table #2 */
e5966052
RS
73} G726Tables;
74
75typedef struct G726Context {
bb270c08
DB
76 G726Tables* tbls; /**< static tables needed for computation */
77
78 Float11 sr[2]; /**< prev. reconstructed samples */
79 Float11 dq[6]; /**< prev. difference */
80 int a[2]; /**< second order predictor coeffs */
81 int b[6]; /**< sixth order predictor coeffs */
82 int pk[2]; /**< signs of prev. 2 sez + dq */
83
84 int ap; /**< scale factor control */
85 int yu; /**< fast scale factor */
86 int yl; /**< slow scale factor */
87 int dms; /**< short average magnitude of F[i] */
88 int dml; /**< long average magnitude of F[i] */
89 int td; /**< tone detect */
90
91 int se; /**< estimated signal for the next iteration */
92 int sez; /**< estimated second order prediction */
93 int y; /**< quantizer scaling factor for the next iteration */
e5966052
RS
94} G726Context;
95
ef9c5d84 96static int quant_tbl16[] = /**< 16kbit/s 2bits per sample */
115329f1 97 { 260, INT_MAX };
e5966052
RS
98static int iquant_tbl16[] =
99 { 116, 365, 365, 116 };
115329f1 100static int W_tbl16[] =
e5966052
RS
101 { -22, 439, 439, -22 };
102static int F_tbl16[] =
103 { 0, 7, 7, 0 };
115329f1 104
ef9c5d84 105static int quant_tbl24[] = /**< 24kbit/s 3bits per sample */
e5966052
RS
106 { 7, 217, 330, INT_MAX };
107static int iquant_tbl24[] =
108 { INT_MIN, 135, 273, 373, 373, 273, 135, INT_MIN };
115329f1
DB
109static int W_tbl24[] =
110 { -4, 30, 137, 582, 582, 137, 30, -4 };
e5966052
RS
111static int F_tbl24[] =
112 { 0, 1, 2, 7, 7, 2, 1, 0 };
115329f1 113
ef9c5d84 114static int quant_tbl32[] = /**< 32kbit/s 4bits per sample */
e5966052
RS
115 { -125, 79, 177, 245, 299, 348, 399, INT_MAX };
116static int iquant_tbl32[] =
115329f1 117 { INT_MIN, 4, 135, 213, 273, 323, 373, 425,
bb270c08 118 425, 373, 323, 273, 213, 135, 4, INT_MIN };
115329f1 119static int W_tbl32[] =
e5966052 120 { -12, 18, 41, 64, 112, 198, 355, 1122,
bb270c08 121 1122, 355, 198, 112, 64, 41, 18, -12};
115329f1 122static int F_tbl32[] =
e5966052 123 { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 };
115329f1 124
ef9c5d84 125static int quant_tbl40[] = /**< 40kbit/s 5bits per sample */
e5966052 126 { -122, -16, 67, 138, 197, 249, 297, 338,
bb270c08 127 377, 412, 444, 474, 501, 527, 552, INT_MAX };
e5966052 128static int iquant_tbl40[] =
115329f1 129 { INT_MIN, -66, 28, 104, 169, 224, 274, 318,
bb270c08
DB
130 358, 395, 429, 459, 488, 514, 539, 566,
131 566, 539, 514, 488, 459, 429, 395, 358,
132 318, 274, 224, 169, 104, 28, -66, INT_MIN };
115329f1 133static int W_tbl40[] =
e5966052 134 { 14, 14, 24, 39, 40, 41, 58, 100,
bb270c08
DB
135 141, 179, 219, 280, 358, 440, 529, 696,
136 696, 529, 440, 358, 280, 219, 179, 141,
137 100, 58, 41, 40, 39, 24, 14, 14 };
115329f1 138static int F_tbl40[] =
e5966052 139 { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6,
bb270c08 140 6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };
e5966052 141
115329f1 142static G726Tables G726Tables_pool[] =
e5966052 143 {{ 2, quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
bb270c08
DB
144 { 3, quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
145 { 4, quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
e5966052 146 { 5, quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }};
115329f1 147
e5966052 148
ef9c5d84 149/**
115329f1 150 * Para 4.2.2 page 18: Adaptive quantizer.
e5966052
RS
151 */
152static inline uint8_t quant(G726Context* c, int d)
153{
154 int sign, exp, i, dln;
115329f1 155
e5966052
RS
156 sign = i = 0;
157 if (d < 0) {
158 sign = 1;
159 d = -d;
160 }
161 exp = av_log2_16bit(d);
162 dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2);
115329f1 163
e5966052
RS
164 while (c->tbls->quant[i] < INT_MAX && c->tbls->quant[i] < dln)
165 ++i;
115329f1 166
e5966052
RS
167 if (sign)
168 i = ~i;
169 if (c->tbls->bits != 2 && i == 0) /* I'm not sure this is a good idea */
170 i = 0xff;
171
115329f1 172 return i;
e5966052
RS
173}
174
ef9c5d84 175/**
e5966052
RS
176 * Para 4.2.3 page 22: Inverse adaptive quantizer.
177 */
178static inline int16_t inverse_quant(G726Context* c, int i)
179{
180 int dql, dex, dqt;
115329f1 181
e5966052
RS
182 dql = c->tbls->iquant[i] + (c->y >> 2);
183 dex = (dql>>7) & 0xf; /* 4bit exponent */
184 dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */
115329f1 185 return (dql < 0) ? 0 : ((dqt<<7) >> (14-dex));
e5966052
RS
186}
187
188static inline int16_t g726_iterate(G726Context* c, int16_t I)
189{
190 int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
191 Float11 f;
115329f1 192
e5966052
RS
193 dq = inverse_quant(c, I);
194 if (I >> (c->tbls->bits - 1)) /* get the sign */
195 dq = -dq;
196 re_signal = c->se + dq;
197
198 /* Transition detect */
199 ylint = (c->yl >> 15);
200 ylfrac = (c->yl >> 10) & 0x1f;
201 thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint;
202 if (c->td == 1 && abs(dq) > ((thr2+(thr2>>1))>>1))
203 tr = 1;
204 else
205 tr = 0;
115329f1 206
e5966052
RS
207 /* Update second order predictor coefficient A2 and A1 */
208 pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
209 dq0 = dq ? sgn(dq) : 0;
210 if (tr) {
211 c->a[0] = 0;
bb270c08 212 c->a[1] = 0;
e5966052 213 for (i=0; i<6; i++)
bb270c08 214 c->b[i] = 0;
e5966052 215 } else {
bb270c08
DB
216 /* This is a bit crazy, but it really is +255 not +256 */
217 fa1 = clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255);
115329f1 218
bb270c08
DB
219 c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7);
220 c->a[1] = clip(c->a[1], -12288, 12288);
e5966052 221 c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8);
bb270c08 222 c->a[0] = clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]);
e5966052
RS
223
224 for (i=0; i<6; i++)
bb270c08 225 c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8);
e5966052
RS
226 }
227
228 /* Update Dq and Sr and Pk */
229 c->pk[1] = c->pk[0];
230 c->pk[0] = pk0 ? pk0 : 1;
231 c->sr[1] = c->sr[0];
232 i2f(re_signal, &c->sr[0]);
233 for (i=5; i>0; i--)
234 c->dq[i] = c->dq[i-1];
235 i2f(dq, &c->dq[0]);
236 c->dq[0].sign = I >> (c->tbls->bits - 1); /* Isn't it crazy ?!?! */
115329f1 237
e5966052 238 /* Update tone detect [I'm not sure 'tr == 0' is really needed] */
115329f1
DB
239 c->td = (tr == 0 && c->a[1] < -11776);
240
e5966052
RS
241 /* Update Ap */
242 c->dms += ((c->tbls->F[I]<<9) - c->dms) >> 5;
243 c->dml += ((c->tbls->F[I]<<11) - c->dml) >> 7;
115329f1 244 if (tr)
e5966052
RS
245 c->ap = 256;
246 else if (c->y > 1535 && !c->td && (abs((c->dms << 2) - c->dml) < (c->dml >> 3)))
247 c->ap += (-c->ap) >> 4;
248 else
115329f1 249 c->ap += (0x200 - c->ap) >> 4;
e5966052
RS
250
251 /* Update Yu and Yl */
eb5b0422 252 c->yu = clip(c->y + (((c->tbls->W[I] << 5) - c->y) >> 5), 544, 5120);
e5966052 253 c->yl += c->yu + ((-c->yl)>>6);
115329f1 254
e5966052
RS
255 /* Next iteration for Y */
256 al = (c->ap >= 256) ? 1<<6 : c->ap >> 2;
257 c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6;
115329f1 258
e5966052
RS
259 /* Next iteration for SE and SEZ */
260 c->se = 0;
261 for (i=0; i<6; i++)
262 c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]);
263 c->sez = c->se >> 1;
264 for (i=0; i<2; i++)
265 c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]);
266 c->se >>= 1;
267
eb5b0422 268 return clip(re_signal << 2, -0xffff, 0xffff);
e5966052
RS
269}
270
271static int g726_reset(G726Context* c, int bit_rate)
272{
273 int i;
274
275 c->tbls = &G726Tables_pool[bit_rate/8000 - 2];
276 for (i=0; i<2; i++) {
277 i2f(0, &c->sr[i]);
278 c->a[i] = 0;
279 c->pk[i] = 1;
280 }
281 for (i=0; i<6; i++) {
282 i2f(0, &c->dq[i]);
283 c->b[i] = 0;
284 }
285 c->ap = 0;
286 c->dms = 0;
287 c->dml = 0;
288 c->yu = 544;
289 c->yl = 34816;
290 c->td = 0;
291
292 c->se = 0;
293 c->sez = 0;
294 c->y = 544;
295
296 return 0;
297}
298
299static int16_t g726_decode(G726Context* c, int16_t i)
300{
301 return g726_iterate(c, i);
302}
303
62bb489b 304#ifdef CONFIG_ENCODERS
e5966052
RS
305static int16_t g726_encode(G726Context* c, int16_t sig)
306{
307 uint8_t i;
115329f1 308
e5966052
RS
309 i = quant(c, sig/4 - c->se) & ((1<<c->tbls->bits) - 1);
310 g726_iterate(c, i);
311 return i;
312}
62bb489b 313#endif
e5966052
RS
314
315/* Interfacing to the libavcodec */
316
317typedef struct AVG726Context {
318 G726Context c;
319 int bits_left;
320 int bit_buffer;
321 int code_size;
322} AVG726Context;
323
324static int g726_init(AVCodecContext * avctx)
325{
326 AVG726Context* c = (AVG726Context*)avctx->priv_data;
115329f1 327
eb5b0422 328 if (avctx->channels != 1 ||
e5966052 329 (avctx->bit_rate != 16000 && avctx->bit_rate != 24000 &&
bb270c08 330 avctx->bit_rate != 32000 && avctx->bit_rate != 40000)) {
e5966052 331 av_log(avctx, AV_LOG_ERROR, "G726: unsupported audio format\n");
bb270c08 332 return -1;
e5966052 333 }
9cd81798 334 if (avctx->sample_rate != 8000 && avctx->strict_std_compliance>FF_COMPLIANCE_INOFFICIAL) {
eb5b0422 335 av_log(avctx, AV_LOG_ERROR, "G726: unsupported audio format\n");
bb270c08 336 return -1;
eb5b0422 337 }
e5966052
RS
338 g726_reset(&c->c, avctx->bit_rate);
339 c->code_size = c->c.tbls->bits;
340 c->bit_buffer = 0;
341 c->bits_left = 0;
342
826ca104
RS
343 avctx->coded_frame = avcodec_alloc_frame();
344 if (!avctx->coded_frame)
345 return -ENOMEM;
346 avctx->coded_frame->key_frame = 1;
347
348 return 0;
349}
350
351static int g726_close(AVCodecContext *avctx)
352{
353 av_freep(&avctx->coded_frame);
e5966052
RS
354 return 0;
355}
356
62bb489b 357#ifdef CONFIG_ENCODERS
e5966052
RS
358static int g726_encode_frame(AVCodecContext *avctx,
359 uint8_t *dst, int buf_size, void *data)
360{
361 AVG726Context *c = avctx->priv_data;
362 short *samples = data;
363 PutBitContext pb;
364
365 init_put_bits(&pb, dst, 1024*1024);
366
367 for (; buf_size; buf_size--)
368 put_bits(&pb, c->code_size, g726_encode(&c->c, *samples++));
369
370 flush_put_bits(&pb);
371
115329f1
DB
372 return put_bits_count(&pb)>>3;
373}
62bb489b 374#endif
e5966052
RS
375
376static int g726_decode_frame(AVCodecContext *avctx,
377 void *data, int *data_size,
378 uint8_t *buf, int buf_size)
379{
380 AVG726Context *c = avctx->priv_data;
381 short *samples = data;
382 uint8_t code;
383 uint8_t mask;
115329f1
DB
384 GetBitContext gb;
385
e5966052
RS
386 if (!buf_size)
387 goto out;
115329f1 388
e5966052
RS
389 mask = (1<<c->code_size) - 1;
390 init_get_bits(&gb, buf, buf_size * 8);
391 if (c->bits_left) {
392 int s = c->code_size - c->bits_left;;
bb270c08
DB
393 code = (c->bit_buffer << s) | get_bits(&gb, s);
394 *samples++ = g726_decode(&c->c, code & mask);
e5966052 395 }
115329f1 396
e5966052 397 while (get_bits_count(&gb) + c->code_size <= buf_size*8)
bb270c08 398 *samples++ = g726_decode(&c->c, get_bits(&gb, c->code_size) & mask);
115329f1 399
e5966052
RS
400 c->bits_left = buf_size*8 - get_bits_count(&gb);
401 c->bit_buffer = get_bits(&gb, c->bits_left);
115329f1 402
e5966052
RS
403out:
404 *data_size = (uint8_t*)samples - (uint8_t*)data;
405 return buf_size;
406}
407
408#ifdef CONFIG_ENCODERS
409AVCodec adpcm_g726_encoder = {
410 "g726",
411 CODEC_TYPE_AUDIO,
412 CODEC_ID_ADPCM_G726,
413 sizeof(AVG726Context),
414 g726_init,
415 g726_encode_frame,
826ca104 416 g726_close,
e5966052
RS
417 NULL,
418};
419#endif //CONFIG_ENCODERS
420
421AVCodec adpcm_g726_decoder = {
422 "g726",
423 CODEC_TYPE_AUDIO,
424 CODEC_ID_ADPCM_G726,
425 sizeof(AVG726Context),
426 g726_init,
427 NULL,
826ca104 428 g726_close,
e5966052
RS
429 g726_decode_frame,
430};