Do not shift F[I] twice, it is also clearer and smaller now.
[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"
caa336b4 26#include "bitstream.h"
e5966052 27
ef9c5d84
MN
28/**
29 * G.726 11bit float.
115329f1 30 * G.726 Standard uses rather odd 11bit floating point arithmentic for
e5966052
RS
31 * numerous occasions. It's a mistery to me why they did it this way
32 * instead of simply using 32bit integer arithmetic.
33 */
34typedef struct Float11 {
deec52cc
MN
35 uint8_t sign; /**< 1bit sign */
36 uint8_t exp; /**< 4bit exponent */
37 uint8_t mant; /**< 6bit mantissa */
e5966052
RS
38} Float11;
39
5171fd9d 40static inline Float11* i2f(int i, Float11* f)
e5966052 41{
e344c1ea
SH
42 f->sign = (i < 0);
43 if (f->sign)
44 i = -i;
45 f->exp = av_log2_16bit(i) + !!i;
46 f->mant = i? (i<<6) >> f->exp : 1<<5;
47 return f;
e5966052
RS
48}
49
50static inline int16_t mult(Float11* f1, Float11* f2)
51{
bb270c08 52 int res, exp;
e5966052 53
bb270c08 54 exp = f1->exp + f2->exp;
7d7a9e7f
MN
55 res = (((f1->mant * f2->mant) + 0x30) >> 4);
56 res = exp > 19 ? res << (exp - 19) : res >> (19 - exp);
bb270c08 57 return (f1->sign ^ f2->sign) ? -res : res;
e5966052
RS
58}
59
e5966052
RS
60static inline int sgn(int value)
61{
e344c1ea 62 return (value < 0) ? -1 : 1;
e5966052
RS
63}
64
65typedef struct G726Tables {
9307db13 66 const int* quant; /**< quantization table */
a0e5830a 67 const int16_t* iquant; /**< inverse quantization table */
e3cafac6
MN
68 const int16_t* W; /**< special table #1 ;-) */
69 const uint8_t* F; /**< special table #2 */
e5966052
RS
70} G726Tables;
71
72typedef struct G726Context {
05c9f351 73 G726Tables tbls; /**< static tables needed for computation */
e344c1ea
SH
74
75 Float11 sr[2]; /**< prev. reconstructed samples */
76 Float11 dq[6]; /**< prev. difference */
77 int a[2]; /**< second order predictor coeffs */
78 int b[6]; /**< sixth order predictor coeffs */
79 int pk[2]; /**< signs of prev. 2 sez + dq */
80
81 int ap; /**< scale factor control */
82 int yu; /**< fast scale factor */
83 int yl; /**< slow scale factor */
84 int dms; /**< short average magnitude of F[i] */
85 int dml; /**< long average magnitude of F[i] */
86 int td; /**< tone detect */
87
88 int se; /**< estimated signal for the next iteration */
89 int sez; /**< estimated second order prediction */
90 int y; /**< quantizer scaling factor for the next iteration */
c79c1a01 91 int code_size;
e5966052
RS
92} G726Context;
93
9307db13 94static const int quant_tbl16[] = /**< 16kbit/s 2bits per sample */
115329f1 95 { 260, INT_MAX };
a0e5830a 96static const int16_t iquant_tbl16[] =
e5966052 97 { 116, 365, 365, 116 };
e3cafac6 98static const int16_t W_tbl16[] =
e5966052 99 { -22, 439, 439, -22 };
e3cafac6 100static const uint8_t F_tbl16[] =
e5966052 101 { 0, 7, 7, 0 };
115329f1 102
9307db13 103static const int quant_tbl24[] = /**< 24kbit/s 3bits per sample */
e5966052 104 { 7, 217, 330, INT_MAX };
a0e5830a
MN
105static const int16_t iquant_tbl24[] =
106 { INT16_MIN, 135, 273, 373, 373, 273, 135, INT16_MIN };
e3cafac6 107static const int16_t W_tbl24[] =
115329f1 108 { -4, 30, 137, 582, 582, 137, 30, -4 };
e3cafac6 109static const uint8_t F_tbl24[] =
e5966052 110 { 0, 1, 2, 7, 7, 2, 1, 0 };
115329f1 111
9307db13 112static const int quant_tbl32[] = /**< 32kbit/s 4bits per sample */
e5966052 113 { -125, 79, 177, 245, 299, 348, 399, INT_MAX };
a0e5830a
MN
114static const int16_t iquant_tbl32[] =
115 { INT16_MIN, 4, 135, 213, 273, 323, 373, 425,
116 425, 373, 323, 273, 213, 135, 4, INT16_MIN };
e3cafac6 117static const int16_t W_tbl32[] =
e5966052 118 { -12, 18, 41, 64, 112, 198, 355, 1122,
bb270c08 119 1122, 355, 198, 112, 64, 41, 18, -12};
e3cafac6 120static const uint8_t F_tbl32[] =
e5966052 121 { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 };
115329f1 122
9307db13 123static const int quant_tbl40[] = /**< 40kbit/s 5bits per sample */
e5966052 124 { -122, -16, 67, 138, 197, 249, 297, 338,
bb270c08 125 377, 412, 444, 474, 501, 527, 552, INT_MAX };
a0e5830a
MN
126static const int16_t iquant_tbl40[] =
127 { INT16_MIN, -66, 28, 104, 169, 224, 274, 318,
bb270c08
DB
128 358, 395, 429, 459, 488, 514, 539, 566,
129 566, 539, 514, 488, 459, 429, 395, 358,
a0e5830a 130 318, 274, 224, 169, 104, 28, -66, INT16_MIN };
e3cafac6 131static const int16_t W_tbl40[] =
e5966052 132 { 14, 14, 24, 39, 40, 41, 58, 100,
bb270c08
DB
133 141, 179, 219, 280, 358, 440, 529, 696,
134 696, 529, 440, 358, 280, 219, 179, 141,
135 100, 58, 41, 40, 39, 24, 14, 14 };
e3cafac6 136static const uint8_t F_tbl40[] =
e5966052 137 { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6,
bb270c08 138 6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };
e5966052 139
9307db13 140static const G726Tables G726Tables_pool[] =
0e0d6cfd
MN
141 {{ quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
142 { quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
143 { quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
144 { quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }};
115329f1 145
e5966052 146
ef9c5d84 147/**
115329f1 148 * Para 4.2.2 page 18: Adaptive quantizer.
e5966052
RS
149 */
150static inline uint8_t quant(G726Context* c, int d)
151{
e344c1ea 152 int sign, exp, i, dln;
115329f1 153
e344c1ea
SH
154 sign = i = 0;
155 if (d < 0) {
156 sign = 1;
157 d = -d;
158 }
159 exp = av_log2_16bit(d);
160 dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2);
115329f1 161
05c9f351 162 while (c->tbls.quant[i] < INT_MAX && c->tbls.quant[i] < dln)
e5966052 163 ++i;
115329f1 164
e344c1ea
SH
165 if (sign)
166 i = ~i;
0e0d6cfd 167 if (c->code_size != 2 && i == 0) /* I'm not sure this is a good idea */
e344c1ea 168 i = 0xff;
e5966052 169
e344c1ea 170 return i;
e5966052
RS
171}
172
ef9c5d84 173/**
e5966052
RS
174 * Para 4.2.3 page 22: Inverse adaptive quantizer.
175 */
176static inline int16_t inverse_quant(G726Context* c, int i)
177{
178 int dql, dex, dqt;
115329f1 179
05c9f351 180 dql = c->tbls.iquant[i] + (c->y >> 2);
e5966052
RS
181 dex = (dql>>7) & 0xf; /* 4bit exponent */
182 dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */
6aa37bcf 183 return (dql < 0) ? 0 : ((dqt<<dex) >> 7);
e5966052
RS
184}
185
fc234250 186static int16_t g726_decode(G726Context* c, int I)
e5966052
RS
187{
188 int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
189 Float11 f;
0e0d6cfd 190 int I_sig= I >> (c->code_size - 1);
115329f1 191
e5966052 192 dq = inverse_quant(c, I);
e5966052
RS
193
194 /* Transition detect */
195 ylint = (c->yl >> 15);
196 ylfrac = (c->yl >> 10) & 0x1f;
197 thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint;
73ff4f83
MN
198 tr= (c->td == 1 && dq > ((3*thr2)>>2));
199
cb26c9d6 200 if (I_sig) /* get the sign */
73ff4f83
MN
201 dq = -dq;
202 re_signal = c->se + dq;
115329f1 203
e5966052
RS
204 /* Update second order predictor coefficient A2 and A1 */
205 pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
206 dq0 = dq ? sgn(dq) : 0;
207 if (tr) {
208 c->a[0] = 0;
bb270c08 209 c->a[1] = 0;
e5966052 210 for (i=0; i<6; i++)
e344c1ea 211 c->b[i] = 0;
e5966052 212 } else {
bb270c08 213 /* This is a bit crazy, but it really is +255 not +256 */
f66e4f5f 214 fa1 = av_clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255);
115329f1 215
bb270c08 216 c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7);
f66e4f5f 217 c->a[1] = av_clip(c->a[1], -12288, 12288);
e5966052 218 c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8);
f66e4f5f 219 c->a[0] = av_clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]);
e5966052
RS
220
221 for (i=0; i<6; i++)
e344c1ea 222 c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8);
e5966052
RS
223 }
224
225 /* Update Dq and Sr and Pk */
226 c->pk[1] = c->pk[0];
227 c->pk[0] = pk0 ? pk0 : 1;
228 c->sr[1] = c->sr[0];
229 i2f(re_signal, &c->sr[0]);
230 for (i=5; i>0; i--)
e344c1ea 231 c->dq[i] = c->dq[i-1];
e5966052 232 i2f(dq, &c->dq[0]);
cb26c9d6 233 c->dq[0].sign = I_sig; /* Isn't it crazy ?!?! */
115329f1 234
428c82cb 235 c->td = c->a[1] < -11776;
115329f1 236
e5966052 237 /* Update Ap */
74d94417
MN
238 c->dms += (c->tbls.F[I]<<4) + ((- c->dms) >> 5);
239 c->dml += (c->tbls.F[I]<<4) + ((- c->dml) >> 7);
115329f1 240 if (tr)
e344c1ea 241 c->ap = 256;
50c52d22 242 else {
e344c1ea 243 c->ap += (-c->ap) >> 4;
50c52d22
MN
244 if (c->y <= 1535 || c->td || abs((c->dms << 2) - c->dml) >= (c->dml >> 3))
245 c->ap += 0x20;
246 }
e5966052
RS
247
248 /* Update Yu and Yl */
05c9f351 249 c->yu = av_clip(c->y + c->tbls.W[I] + ((-c->y)>>5), 544, 5120);
e5966052 250 c->yl += c->yu + ((-c->yl)>>6);
115329f1 251
e5966052
RS
252 /* Next iteration for Y */
253 al = (c->ap >= 256) ? 1<<6 : c->ap >> 2;
254 c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6;
115329f1 255
e5966052
RS
256 /* Next iteration for SE and SEZ */
257 c->se = 0;
258 for (i=0; i<6; i++)
e344c1ea 259 c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]);
e5966052
RS
260 c->sez = c->se >> 1;
261 for (i=0; i<2; i++)
e344c1ea 262 c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]);
e5966052
RS
263 c->se >>= 1;
264
f66e4f5f 265 return av_clip(re_signal << 2, -0xffff, 0xffff);
e5966052
RS
266}
267
8161c28c 268static av_cold int g726_reset(G726Context* c, int index)
e5966052
RS
269{
270 int i;
271
05c9f351 272 c->tbls = G726Tables_pool[index];
e5966052 273 for (i=0; i<2; i++) {
aeaa7c3d 274 c->sr[i].mant = 1<<5;
e344c1ea 275 c->pk[i] = 1;
e5966052
RS
276 }
277 for (i=0; i<6; i++) {
aeaa7c3d 278 c->dq[i].mant = 1<<5;
e5966052 279 }
e5966052
RS
280 c->yu = 544;
281 c->yl = 34816;
e5966052 282
e5966052
RS
283 c->y = 544;
284
285 return 0;
286}
287
62bb489b 288#ifdef CONFIG_ENCODERS
e5966052
RS
289static int16_t g726_encode(G726Context* c, int16_t sig)
290{
e344c1ea 291 uint8_t i;
115329f1 292
0e0d6cfd 293 i = quant(c, sig/4 - c->se) & ((1<<c->code_size) - 1);
cf409a6f 294 g726_decode(c, i);
e344c1ea 295 return i;
e5966052 296}
62bb489b 297#endif
e5966052
RS
298
299/* Interfacing to the libavcodec */
300
98a6fff9 301static av_cold int g726_init(AVCodecContext * avctx)
e5966052 302{
c79c1a01 303 G726Context* c = avctx->priv_data;
8161c28c 304 unsigned int index= (avctx->bit_rate + avctx->sample_rate/2) / avctx->sample_rate - 2;
115329f1 305
aca49077
MN
306 if (avctx->bit_rate % avctx->sample_rate && avctx->codec->encode) {
307 av_log(avctx, AV_LOG_ERROR, "Bitrate - Samplerate combination is invalid\n");
bb270c08 308 return -1;
eb5b0422 309 }
8b470cc5
MN
310 if(avctx->channels != 1){
311 av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
312 return -1;
313 }
35d0e16b
MN
314 if(index>3){
315 av_log(avctx, AV_LOG_ERROR, "Unsupported number of bits %d\n", index+2);
316 return -1;
317 }
c79c1a01 318 g726_reset(c, index);
0e0d6cfd 319 c->code_size = index+2;
e5966052 320
826ca104
RS
321 avctx->coded_frame = avcodec_alloc_frame();
322 if (!avctx->coded_frame)
8fa36ae0 323 return AVERROR(ENOMEM);
826ca104
RS
324 avctx->coded_frame->key_frame = 1;
325
326 return 0;
327}
328
98a6fff9 329static av_cold int g726_close(AVCodecContext *avctx)
826ca104
RS
330{
331 av_freep(&avctx->coded_frame);
e5966052
RS
332 return 0;
333}
334
62bb489b 335#ifdef CONFIG_ENCODERS
e5966052
RS
336static int g726_encode_frame(AVCodecContext *avctx,
337 uint8_t *dst, int buf_size, void *data)
338{
c79c1a01 339 G726Context *c = avctx->priv_data;
e5966052
RS
340 short *samples = data;
341 PutBitContext pb;
342
343 init_put_bits(&pb, dst, 1024*1024);
344
345 for (; buf_size; buf_size--)
c79c1a01 346 put_bits(&pb, c->code_size, g726_encode(c, *samples++));
e5966052
RS
347
348 flush_put_bits(&pb);
349
115329f1
DB
350 return put_bits_count(&pb)>>3;
351}
62bb489b 352#endif
e5966052
RS
353
354static int g726_decode_frame(AVCodecContext *avctx,
355 void *data, int *data_size,
5cbbbf37 356 const uint8_t *buf, int buf_size)
e5966052 357{
c79c1a01 358 G726Context *c = avctx->priv_data;
e5966052 359 short *samples = data;
115329f1
DB
360 GetBitContext gb;
361
e5966052 362 init_get_bits(&gb, buf, buf_size * 8);
115329f1 363
e5966052 364 while (get_bits_count(&gb) + c->code_size <= buf_size*8)
c79c1a01 365 *samples++ = g726_decode(c, get_bits(&gb, c->code_size));
115329f1 366
ef4c5c6d
MN
367 if(buf_size*8 != get_bits_count(&gb))
368 av_log(avctx, AV_LOG_ERROR, "Frame invalidly split, missing parser?\n");
115329f1 369
e5966052
RS
370 *data_size = (uint8_t*)samples - (uint8_t*)data;
371 return buf_size;
372}
373
374#ifdef CONFIG_ENCODERS
375AVCodec adpcm_g726_encoder = {
376 "g726",
377 CODEC_TYPE_AUDIO,
378 CODEC_ID_ADPCM_G726,
c79c1a01 379 sizeof(G726Context),
e5966052
RS
380 g726_init,
381 g726_encode_frame,
826ca104 382 g726_close,
e5966052 383 NULL,
fe4bf374 384 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
e5966052
RS
385};
386#endif //CONFIG_ENCODERS
387
388AVCodec adpcm_g726_decoder = {
389 "g726",
390 CODEC_TYPE_AUDIO,
391 CODEC_ID_ADPCM_G726,
c79c1a01 392 sizeof(G726Context),
e5966052
RS
393 g726_init,
394 NULL,
826ca104 395 g726_close,
e5966052 396 g726_decode_frame,
fe4bf374 397 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
e5966052 398};