g726dec: remove the sample_rate validation
[libav.git] / libavcodec / g726.c
CommitLineData
e5966052 1/*
115329f1 2 * G.726 ADPCM audio codec
406792e7 3 * Copyright (c) 2004 Roman Shaposhnik
e5966052
RS
4 *
5 * This is a very straightforward rendition of the G.726
115329f1 6 * Section 4 "Computational Details".
e5966052 7 *
2912e87a 8 * This file is part of Libav.
b78e7197 9 *
2912e87a 10 * Libav 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 *
2912e87a 15 * Libav 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
2912e87a 21 * License along with Libav; 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"
9106a698 26#include "get_bits.h"
b2755007 27#include "put_bits.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 {
deec52cc
MN
36 uint8_t sign; /**< 1bit sign */
37 uint8_t exp; /**< 4bit exponent */
38 uint8_t mant; /**< 6bit mantissa */
e5966052
RS
39} Float11;
40
5171fd9d 41static inline Float11* i2f(int i, Float11* f)
e5966052 42{
e344c1ea
SH
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 : 1<<5;
48 return f;
e5966052
RS
49}
50
51static inline int16_t mult(Float11* f1, Float11* f2)
52{
bb270c08 53 int res, exp;
e5966052 54
bb270c08 55 exp = f1->exp + f2->exp;
7d7a9e7f
MN
56 res = (((f1->mant * f2->mant) + 0x30) >> 4);
57 res = exp > 19 ? res << (exp - 19) : res >> (19 - exp);
bb270c08 58 return (f1->sign ^ f2->sign) ? -res : res;
e5966052
RS
59}
60
e5966052
RS
61static inline int sgn(int value)
62{
e344c1ea 63 return (value < 0) ? -1 : 1;
e5966052
RS
64}
65
66typedef struct G726Tables {
9307db13 67 const int* quant; /**< quantization table */
a0e5830a 68 const int16_t* iquant; /**< inverse quantization table */
e3cafac6
MN
69 const int16_t* W; /**< special table #1 ;-) */
70 const uint8_t* F; /**< special table #2 */
e5966052
RS
71} G726Tables;
72
73typedef struct G726Context {
05c9f351 74 G726Tables tbls; /**< static tables needed for computation */
e344c1ea
SH
75
76 Float11 sr[2]; /**< prev. reconstructed samples */
77 Float11 dq[6]; /**< prev. difference */
78 int a[2]; /**< second order predictor coeffs */
79 int b[6]; /**< sixth order predictor coeffs */
80 int pk[2]; /**< signs of prev. 2 sez + dq */
81
82 int ap; /**< scale factor control */
83 int yu; /**< fast scale factor */
84 int yl; /**< slow scale factor */
85 int dms; /**< short average magnitude of F[i] */
86 int dml; /**< long average magnitude of F[i] */
87 int td; /**< tone detect */
88
89 int se; /**< estimated signal for the next iteration */
90 int sez; /**< estimated second order prediction */
91 int y; /**< quantizer scaling factor for the next iteration */
c79c1a01 92 int code_size;
e5966052
RS
93} G726Context;
94
9307db13 95static const int quant_tbl16[] = /**< 16kbit/s 2bits per sample */
115329f1 96 { 260, INT_MAX };
a0e5830a 97static const int16_t iquant_tbl16[] =
e5966052 98 { 116, 365, 365, 116 };
e3cafac6 99static const int16_t W_tbl16[] =
e5966052 100 { -22, 439, 439, -22 };
e3cafac6 101static const uint8_t F_tbl16[] =
e5966052 102 { 0, 7, 7, 0 };
115329f1 103
9307db13 104static const int quant_tbl24[] = /**< 24kbit/s 3bits per sample */
e5966052 105 { 7, 217, 330, INT_MAX };
a0e5830a
MN
106static const int16_t iquant_tbl24[] =
107 { INT16_MIN, 135, 273, 373, 373, 273, 135, INT16_MIN };
e3cafac6 108static const int16_t W_tbl24[] =
115329f1 109 { -4, 30, 137, 582, 582, 137, 30, -4 };
e3cafac6 110static const uint8_t F_tbl24[] =
e5966052 111 { 0, 1, 2, 7, 7, 2, 1, 0 };
115329f1 112
9307db13 113static const int quant_tbl32[] = /**< 32kbit/s 4bits per sample */
e5966052 114 { -125, 79, 177, 245, 299, 348, 399, INT_MAX };
a0e5830a
MN
115static const int16_t iquant_tbl32[] =
116 { INT16_MIN, 4, 135, 213, 273, 323, 373, 425,
117 425, 373, 323, 273, 213, 135, 4, INT16_MIN };
e3cafac6 118static const int16_t W_tbl32[] =
e5966052 119 { -12, 18, 41, 64, 112, 198, 355, 1122,
bb270c08 120 1122, 355, 198, 112, 64, 41, 18, -12};
e3cafac6 121static const uint8_t F_tbl32[] =
e5966052 122 { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 };
115329f1 123
9307db13 124static const int quant_tbl40[] = /**< 40kbit/s 5bits per sample */
e5966052 125 { -122, -16, 67, 138, 197, 249, 297, 338,
bb270c08 126 377, 412, 444, 474, 501, 527, 552, INT_MAX };
a0e5830a
MN
127static const int16_t iquant_tbl40[] =
128 { INT16_MIN, -66, 28, 104, 169, 224, 274, 318,
bb270c08
DB
129 358, 395, 429, 459, 488, 514, 539, 566,
130 566, 539, 514, 488, 459, 429, 395, 358,
a0e5830a 131 318, 274, 224, 169, 104, 28, -66, INT16_MIN };
e3cafac6 132static const int16_t W_tbl40[] =
e5966052 133 { 14, 14, 24, 39, 40, 41, 58, 100,
bb270c08
DB
134 141, 179, 219, 280, 358, 440, 529, 696,
135 696, 529, 440, 358, 280, 219, 179, 141,
136 100, 58, 41, 40, 39, 24, 14, 14 };
e3cafac6 137static const uint8_t F_tbl40[] =
e5966052 138 { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6,
bb270c08 139 6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };
e5966052 140
9307db13 141static const G726Tables G726Tables_pool[] =
0e0d6cfd
MN
142 {{ quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
143 { quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
144 { quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
145 { quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }};
115329f1 146
e5966052 147
ef9c5d84 148/**
115329f1 149 * Para 4.2.2 page 18: Adaptive quantizer.
e5966052
RS
150 */
151static inline uint8_t quant(G726Context* c, int d)
152{
e344c1ea 153 int sign, exp, i, dln;
115329f1 154
e344c1ea
SH
155 sign = i = 0;
156 if (d < 0) {
157 sign = 1;
158 d = -d;
159 }
160 exp = av_log2_16bit(d);
161 dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2);
115329f1 162
05c9f351 163 while (c->tbls.quant[i] < INT_MAX && c->tbls.quant[i] < dln)
e5966052 164 ++i;
115329f1 165
e344c1ea
SH
166 if (sign)
167 i = ~i;
0e0d6cfd 168 if (c->code_size != 2 && i == 0) /* I'm not sure this is a good idea */
e344c1ea 169 i = 0xff;
e5966052 170
e344c1ea 171 return i;
e5966052
RS
172}
173
ef9c5d84 174/**
e5966052
RS
175 * Para 4.2.3 page 22: Inverse adaptive quantizer.
176 */
177static inline int16_t inverse_quant(G726Context* c, int i)
178{
179 int dql, dex, dqt;
115329f1 180
05c9f351 181 dql = c->tbls.iquant[i] + (c->y >> 2);
e5966052
RS
182 dex = (dql>>7) & 0xf; /* 4bit exponent */
183 dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */
6aa37bcf 184 return (dql < 0) ? 0 : ((dqt<<dex) >> 7);
e5966052
RS
185}
186
fc234250 187static int16_t g726_decode(G726Context* c, int I)
e5966052
RS
188{
189 int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
190 Float11 f;
0e0d6cfd 191 int I_sig= I >> (c->code_size - 1);
115329f1 192
e5966052 193 dq = inverse_quant(c, I);
e5966052
RS
194
195 /* Transition detect */
196 ylint = (c->yl >> 15);
197 ylfrac = (c->yl >> 10) & 0x1f;
198 thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint;
73ff4f83
MN
199 tr= (c->td == 1 && dq > ((3*thr2)>>2));
200
cb26c9d6 201 if (I_sig) /* get the sign */
73ff4f83
MN
202 dq = -dq;
203 re_signal = c->se + dq;
115329f1 204
e5966052
RS
205 /* Update second order predictor coefficient A2 and A1 */
206 pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
207 dq0 = dq ? sgn(dq) : 0;
208 if (tr) {
209 c->a[0] = 0;
bb270c08 210 c->a[1] = 0;
e5966052 211 for (i=0; i<6; i++)
e344c1ea 212 c->b[i] = 0;
e5966052 213 } else {
bb270c08 214 /* This is a bit crazy, but it really is +255 not +256 */
f66e4f5f 215 fa1 = av_clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255);
115329f1 216
bb270c08 217 c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7);
f66e4f5f 218 c->a[1] = av_clip(c->a[1], -12288, 12288);
e5966052 219 c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8);
f66e4f5f 220 c->a[0] = av_clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]);
e5966052
RS
221
222 for (i=0; i<6; i++)
e344c1ea 223 c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8);
e5966052
RS
224 }
225
226 /* Update Dq and Sr and Pk */
227 c->pk[1] = c->pk[0];
228 c->pk[0] = pk0 ? pk0 : 1;
229 c->sr[1] = c->sr[0];
230 i2f(re_signal, &c->sr[0]);
231 for (i=5; i>0; i--)
e344c1ea 232 c->dq[i] = c->dq[i-1];
e5966052 233 i2f(dq, &c->dq[0]);
cb26c9d6 234 c->dq[0].sign = I_sig; /* Isn't it crazy ?!?! */
115329f1 235
428c82cb 236 c->td = c->a[1] < -11776;
115329f1 237
e5966052 238 /* Update Ap */
74d94417
MN
239 c->dms += (c->tbls.F[I]<<4) + ((- c->dms) >> 5);
240 c->dml += (c->tbls.F[I]<<4) + ((- c->dml) >> 7);
115329f1 241 if (tr)
e344c1ea 242 c->ap = 256;
50c52d22 243 else {
e344c1ea 244 c->ap += (-c->ap) >> 4;
50c52d22
MN
245 if (c->y <= 1535 || c->td || abs((c->dms << 2) - c->dml) >= (c->dml >> 3))
246 c->ap += 0x20;
247 }
e5966052
RS
248
249 /* Update Yu and Yl */
05c9f351 250 c->yu = av_clip(c->y + c->tbls.W[I] + ((-c->y)>>5), 544, 5120);
e5966052 251 c->yl += c->yu + ((-c->yl)>>6);
115329f1 252
e5966052
RS
253 /* Next iteration for Y */
254 al = (c->ap >= 256) ? 1<<6 : c->ap >> 2;
255 c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6;
115329f1 256
e5966052
RS
257 /* Next iteration for SE and SEZ */
258 c->se = 0;
259 for (i=0; i<6; i++)
e344c1ea 260 c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]);
e5966052
RS
261 c->sez = c->se >> 1;
262 for (i=0; i<2; i++)
e344c1ea 263 c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]);
e5966052
RS
264 c->se >>= 1;
265
f66e4f5f 266 return av_clip(re_signal << 2, -0xffff, 0xffff);
e5966052
RS
267}
268
8161c28c 269static av_cold int g726_reset(G726Context* c, int index)
e5966052
RS
270{
271 int i;
272
05c9f351 273 c->tbls = G726Tables_pool[index];
e5966052 274 for (i=0; i<2; i++) {
aeaa7c3d 275 c->sr[i].mant = 1<<5;
e344c1ea 276 c->pk[i] = 1;
e5966052
RS
277 }
278 for (i=0; i<6; i++) {
aeaa7c3d 279 c->dq[i].mant = 1<<5;
e5966052 280 }
e5966052
RS
281 c->yu = 544;
282 c->yl = 34816;
e5966052 283
e5966052
RS
284 c->y = 544;
285
286 return 0;
287}
288
b250f9c6 289#if CONFIG_ADPCM_G726_ENCODER
e5966052
RS
290static int16_t g726_encode(G726Context* c, int16_t sig)
291{
e344c1ea 292 uint8_t i;
115329f1 293
0e0d6cfd 294 i = quant(c, sig/4 - c->se) & ((1<<c->code_size) - 1);
cf409a6f 295 g726_decode(c, i);
e344c1ea 296 return i;
e5966052
RS
297}
298
299/* Interfacing to the libavcodec */
300
d405237b 301static av_cold int g726_encode_init(AVCodecContext *avctx)
e5966052 302{
c79c1a01 303 G726Context* c = avctx->priv_data;
bd10f6e1
LA
304
305 if (avctx->sample_rate <= 0) {
306 av_log(avctx, AV_LOG_ERROR, "Samplerate is invalid\n");
307 return -1;
308 }
309
8b470cc5
MN
310 if(avctx->channels != 1){
311 av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
312 return -1;
313 }
6ac34eed
JR
314
315 if (avctx->bit_rate % avctx->sample_rate) {
316 av_log(avctx, AV_LOG_ERROR, "Bitrate - Samplerate combination is invalid\n");
317 return AVERROR(EINVAL);
318 }
319 c->code_size = (avctx->bit_rate + avctx->sample_rate/2) / avctx->sample_rate;
320 if (c->code_size < 2 || c->code_size > 5) {
321 av_log(avctx, AV_LOG_ERROR, "Invalid number of bits %d\n", c->code_size);
322 return AVERROR(EINVAL);
35d0e16b 323 }
6ac34eed
JR
324 avctx->bits_per_coded_sample = c->code_size;
325
326 g726_reset(c, c->code_size - 2);
e5966052 327
826ca104
RS
328 avctx->coded_frame = avcodec_alloc_frame();
329 if (!avctx->coded_frame)
8fa36ae0 330 return AVERROR(ENOMEM);
826ca104
RS
331 avctx->coded_frame->key_frame = 1;
332
c7d89948
JR
333 /* select a frame size that will end on a byte boundary and have a size of
334 approximately 1024 bytes */
6ac34eed 335 avctx->frame_size = ((int[]){ 4096, 2736, 2048, 1640 })[c->code_size - 2];
c7d89948 336
826ca104
RS
337 return 0;
338}
339
d405237b 340static av_cold int g726_encode_close(AVCodecContext *avctx)
826ca104
RS
341{
342 av_freep(&avctx->coded_frame);
e5966052
RS
343 return 0;
344}
345
346static int g726_encode_frame(AVCodecContext *avctx,
347 uint8_t *dst, int buf_size, void *data)
348{
c79c1a01 349 G726Context *c = avctx->priv_data;
e61a670b 350 const int16_t *samples = data;
e5966052 351 PutBitContext pb;
c7d89948 352 int i;
e5966052
RS
353
354 init_put_bits(&pb, dst, 1024*1024);
355
c7d89948 356 for (i = 0; i < avctx->frame_size; i++)
c79c1a01 357 put_bits(&pb, c->code_size, g726_encode(c, *samples++));
e5966052
RS
358
359 flush_put_bits(&pb);
360
115329f1
DB
361 return put_bits_count(&pb)>>3;
362}
62bb489b 363#endif
e5966052 364
d405237b
JR
365static av_cold int g726_decode_init(AVCodecContext *avctx)
366{
367 G726Context* c = avctx->priv_data;
d405237b 368
d405237b
JR
369 if(avctx->channels != 1){
370 av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
371 return -1;
372 }
6ac34eed
JR
373
374 c->code_size = avctx->bits_per_coded_sample;
375 if (c->code_size < 2 || c->code_size > 5) {
376 av_log(avctx, AV_LOG_ERROR, "Invalid number of bits %d\n", c->code_size);
377 return AVERROR(EINVAL);
d405237b 378 }
6ac34eed 379 g726_reset(c, c->code_size - 2);
d405237b
JR
380
381 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
382
383 return 0;
384}
385
e5966052
RS
386static int g726_decode_frame(AVCodecContext *avctx,
387 void *data, int *data_size,
7a00bbad 388 AVPacket *avpkt)
e5966052 389{
7a00bbad
TB
390 const uint8_t *buf = avpkt->data;
391 int buf_size = avpkt->size;
c79c1a01 392 G726Context *c = avctx->priv_data;
e61a670b 393 int16_t *samples = data;
115329f1 394 GetBitContext gb;
c8d36d25
JR
395 int out_samples, out_size;
396
397 out_samples = buf_size * 8 / c->code_size;
398 out_size = out_samples * av_get_bytes_per_sample(avctx->sample_fmt);
399 if (*data_size < out_size) {
400 av_log(avctx, AV_LOG_ERROR, "Output buffer is too small\n");
401 return AVERROR(EINVAL);
402 }
115329f1 403
e5966052 404 init_get_bits(&gb, buf, buf_size * 8);
115329f1 405
c8d36d25 406 while (out_samples--)
c79c1a01 407 *samples++ = g726_decode(c, get_bits(&gb, c->code_size));
115329f1 408
c8d36d25 409 if (get_bits_left(&gb) > 0)
ef4c5c6d 410 av_log(avctx, AV_LOG_ERROR, "Frame invalidly split, missing parser?\n");
115329f1 411
c8d36d25 412 *data_size = out_size;
e5966052
RS
413 return buf_size;
414}
415
b250f9c6 416#if CONFIG_ADPCM_G726_ENCODER
d36beb3f 417AVCodec ff_adpcm_g726_encoder = {
ec6402b7
AK
418 .name = "g726",
419 .type = AVMEDIA_TYPE_AUDIO,
420 .id = CODEC_ID_ADPCM_G726,
421 .priv_data_size = sizeof(G726Context),
d405237b 422 .init = g726_encode_init,
ec6402b7 423 .encode = g726_encode_frame,
d405237b 424 .close = g726_encode_close,
c7d89948 425 .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
5d6e4c16 426 .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
fe4bf374 427 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
e5966052 428};
f544a5fc 429#endif
e5966052 430
d36beb3f 431AVCodec ff_adpcm_g726_decoder = {
ec6402b7
AK
432 .name = "g726",
433 .type = AVMEDIA_TYPE_AUDIO,
434 .id = CODEC_ID_ADPCM_G726,
435 .priv_data_size = sizeof(G726Context),
d405237b 436 .init = g726_decode_init,
ec6402b7 437 .decode = g726_decode_frame,
fe4bf374 438 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
e5966052 439};