ffmpeg/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 FFmpeg.
   9  *
  10  * FFmpeg 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  * FFmpeg 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 FFmpeg; 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     int little_endian;  /**< little-endian bitstream as used in aiff and Sun AU */
 100 } G726Context;
 101
 102 static const int quant_tbl16[] =                  /**< 16kbit/s 2bits per sample */
 103            { 260, INT_MAX };
 104 static const int16_t iquant_tbl16[] =
 105            { 116, 365, 365, 116 };
 106 static const int16_t W_tbl16[] =
 107            { -22, 439, 439, -22 };
 108 static const uint8_t F_tbl16[] =
 109            { 0, 7, 7, 0 };
 110
 111 static const int quant_tbl24[] =                  /**< 24kbit/s 3bits per sample */
 112            {  7, 217, 330, INT_MAX };
 113 static const int16_t iquant_tbl24[] =
 114            { INT16_MIN, 135, 273, 373, 373, 273, 135, INT16_MIN };
 115 static const int16_t W_tbl24[] =
 116            { -4,  30, 137, 582, 582, 137,  30, -4 };
 117 static const uint8_t F_tbl24[] =
 118            { 0, 1, 2, 7, 7, 2, 1, 0 };
 119
 120 static const int quant_tbl32[] =                  /**< 32kbit/s 4bits per sample */
 121            { -125,  79, 177, 245, 299, 348, 399, INT_MAX };
 122 static const int16_t iquant_tbl32[] =
 123          { INT16_MIN,   4, 135, 213, 273, 323, 373, 425,
 124                  425, 373, 323, 273, 213, 135,   4, INT16_MIN };
 125 static const int16_t W_tbl32[] =
 126            { -12,  18,  41,  64, 112, 198, 355, 1122,
 127             1122, 355, 198, 112,  64,  41,  18, -12};
 128 static const uint8_t F_tbl32[] =
 129            { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 };
 130
 131 static const int quant_tbl40[] =                  /**< 40kbit/s 5bits per sample */
 132            { -122, -16,  67, 138, 197, 249, 297, 338,
 133               377, 412, 444, 474, 501, 527, 552, INT_MAX };
 134 static const int16_t iquant_tbl40[] =
 135          { INT16_MIN, -66,  28, 104, 169, 224, 274, 318,
 136                  358, 395, 429, 459, 488, 514, 539, 566,
 137                  566, 539, 514, 488, 459, 429, 395, 358,
 138                  318, 274, 224, 169, 104,  28, -66, INT16_MIN };
 139 static const int16_t W_tbl40[] =
 140            {   14,  14,  24,  39,  40,  41,   58,  100,
 141               141, 179, 219, 280, 358, 440,  529,  696,
 142               696, 529, 440, 358, 280, 219,  179,  141,
 143               100,  58,  41,  40,  39,  24,   14,   14 };
 144 static const uint8_t F_tbl40[] =
 145            { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6,
 146              6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };
 147
 148 static const G726Tables G726Tables_pool[] =
 149            {{ quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
 150             { quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
 151             { quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
 152             { quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }};
 153
 154
 155 /**
 156  * Para 4.2.2 page 18: Adaptive quantizer.
 157  */
 158 static inline uint8_t quant(G726Context* c, int d)
 159 {
 160     int sign, exp, i, dln;
 161
 162     sign = i = 0;
 163     if (d < 0) {
 164         sign = 1;
 165         d = -d;
 166     }
 167     exp = av_log2_16bit(d);
 168     dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2);
 169
 170     while (c->tbls.quant[i] < INT_MAX && c->tbls.quant[i] < dln)
 171         ++i;
 172
 173     if (sign)
 174         i = ~i;
 175     if (c->code_size != 2 && i == 0) /* I'm not sure this is a good idea */
 176         i = 0xff;
 177
 178     return i;
 179 }
 180
 181 /**
 182  * Para 4.2.3 page 22: Inverse adaptive quantizer.
 183  */
 184 static inline int16_t inverse_quant(G726Context* c, int i)
 185 {
 186     int dql, dex, dqt;
 187
 188     dql = c->tbls.iquant[i] + (c->y >> 2);
 189     dex = (dql>>7) & 0xf;        /* 4bit exponent */
 190     dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */
 191     return (dql < 0) ? 0 : ((dqt<<dex) >> 7);
 192 }
 193
 194 static int16_t g726_decode(G726Context* c, int I)
 195 {
 196     int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
 197     Float11 f;
 198     int I_sig= I >> (c->code_size - 1);
 199
 200     dq = inverse_quant(c, I);
 201
 202     /* Transition detect */
 203     ylint = (c->yl >> 15);
 204     ylfrac = (c->yl >> 10) & 0x1f;
 205     thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint;
 206     tr= (c->td == 1 && dq > ((3*thr2)>>2));
 207
 208     if (I_sig)  /* get the sign */
 209         dq = -dq;
 210     re_signal = c->se + dq;
 211
 212     /* Update second order predictor coefficient A2 and A1 */
 213     pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
 214     dq0 = dq ? sgn(dq) : 0;
 215     if (tr) {
 216         c->a[0] = 0;
 217         c->a[1] = 0;
 218         for (i=0; i<6; i++)
 219             c->b[i] = 0;
 220     } else {
 221         /* This is a bit crazy, but it really is +255 not +256 */
 222         fa1 = av_clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255);
 223
 224         c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7);
 225         c->a[1] = av_clip(c->a[1], -12288, 12288);
 226         c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8);
 227         c->a[0] = av_clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]);
 228
 229         for (i=0; i<6; i++)
 230             c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8);
 231     }
 232
 233     /* Update Dq and Sr and Pk */
 234     c->pk[1] = c->pk[0];
 235     c->pk[0] = pk0 ? pk0 : 1;
 236     c->sr[1] = c->sr[0];
 237     i2f(re_signal, &c->sr[0]);
 238     for (i=5; i>0; i--)
 239         c->dq[i] = c->dq[i-1];
 240     i2f(dq, &c->dq[0]);
 241     c->dq[0].sign = I_sig; /* Isn't it crazy ?!?! */
 242
 243     c->td = c->a[1] < -11776;
 244
 245     /* Update Ap */
 246     c->dms += (c->tbls.F[I]<<4) + ((- c->dms) >> 5);
 247     c->dml += (c->tbls.F[I]<<4) + ((- c->dml) >> 7);
 248     if (tr)
 249         c->ap = 256;
 250     else {
 251         c->ap += (-c->ap) >> 4;
 252         if (c->y <= 1535 || c->td || abs((c->dms << 2) - c->dml) >= (c->dml >> 3))
 253             c->ap += 0x20;
 254     }
 255
 256     /* Update Yu and Yl */
 257     c->yu = av_clip(c->y + c->tbls.W[I] + ((-c->y)>>5), 544, 5120);
 258     c->yl += c->yu + ((-c->yl)>>6);
 259
 260     /* Next iteration for Y */
 261     al = (c->ap >= 256) ? 1<<6 : c->ap >> 2;
 262     c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6;
 263
 264     /* Next iteration for SE and SEZ */
 265     c->se = 0;
 266     for (i=0; i<6; i++)
 267         c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]);
 268     c->sez = c->se >> 1;
 269     for (i=0; i<2; i++)
 270         c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]);
 271     c->se >>= 1;
 272
 273     return av_clip(re_signal << 2, -0xffff, 0xffff);
 274 }
 275
 276 static av_cold int g726_reset(G726Context *c)
 277 {
 278     int i;
 279
 280     c->tbls = G726Tables_pool[c->code_size - 2];
 281     for (i=0; i<2; i++) {
 282         c->sr[i].mant = 1<<5;
 283         c->pk[i] = 1;
 284     }
 285     for (i=0; i<6; i++) {
 286         c->dq[i].mant = 1<<5;
 287     }
 288     c->yu = 544;
 289     c->yl = 34816;
 290
 291     c->y = 544;
 292
 293     return 0;
 294 }
 295
 296 #if CONFIG_ADPCM_G726_ENCODER
 297 static int16_t g726_encode(G726Context* c, int16_t sig)
 298 {
 299     uint8_t i;
 300
 301     i = quant(c, sig/4 - c->se) & ((1<<c->code_size) - 1);
 302     g726_decode(c, i);
 303     return i;
 304 }
 305
 306 /* Interfacing to the libavcodec */
 307
 308 static av_cold int g726_encode_init(AVCodecContext *avctx)
 309 {
 310     G726Context* c = avctx->priv_data;
 311
 312     if (avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL &&
 313         avctx->sample_rate != 8000) {
 314         av_log(avctx, AV_LOG_ERROR, "Sample rates other than 8kHz are not "
 315                "allowed when the compliance level is higher than unofficial. "
 316                "Resample or reduce the compliance level.\n");
 317         return AVERROR(EINVAL);
 318     }
 319     av_assert0(avctx->sample_rate > 0);
 320
 321     if(avctx->channels != 1){
 322         av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
 323         return AVERROR(EINVAL);
 324     }
 325
 326     if (avctx->bit_rate)
 327         c->code_size = (avctx->bit_rate + avctx->sample_rate/2) / avctx->sample_rate;
 328
 329     c->code_size = av_clip(c->code_size, 2, 5);
 330     avctx->bit_rate = c->code_size * avctx->sample_rate;
 331     avctx->bits_per_coded_sample = c->code_size;
 332
 333     g726_reset(c);
 334
 335     /* select a frame size that will end on a byte boundary and have a size of
 336        approximately 1024 bytes */
 337     avctx->frame_size = ((int[]){ 4096, 2736, 2048, 1640 })[c->code_size - 2];
 338
 339     return 0;
 340 }
 341
 342 static int g726_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
 343                              const AVFrame *frame, int *got_packet_ptr)
 344 {
 345     G726Context *c = avctx->priv_data;
 346     const int16_t *samples = (const int16_t *)frame->data[0];
 347     PutBitContext pb;
 348     int i, ret, out_size;
 349
 350     out_size = (frame->nb_samples * c->code_size + 7) / 8;
 351     if ((ret = ff_alloc_packet2(avctx, avpkt, out_size)) < 0)
 352         return ret;
 353     init_put_bits(&pb, avpkt->data, avpkt->size);
 354
 355     for (i = 0; i < frame->nb_samples; i++)
 356         put_bits(&pb, c->code_size, g726_encode(c, *samples++));
 357
 358     flush_put_bits(&pb);
 359
 360     avpkt->size = out_size;
 361     *got_packet_ptr = 1;
 362     return 0;
 363 }
 364
 365 #define OFFSET(x) offsetof(G726Context, x)
 366 #define AE AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
 367 static const AVOption options[] = {
 368     { "code_size", "Bits per code", OFFSET(code_size), AV_OPT_TYPE_INT, { .i64 = 4 }, 2, 5, AE },
 369     { NULL },
 370 };
 371
 372 static const AVClass g726_class = {
 373     .class_name = "g726",
 374     .item_name  = av_default_item_name,
 375     .option     = options,
 376     .version    = LIBAVUTIL_VERSION_INT,
 377 };
 378
 379 static const AVCodecDefault defaults[] = {
 380     { "b", "0" },
 381     { NULL },
 382 };
 383
 384 AVCodec ff_adpcm_g726_encoder = {
 385     .name           = "g726",
 386     .long_name      = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
 387     .type           = AVMEDIA_TYPE_AUDIO,
 388     .id             = AV_CODEC_ID_ADPCM_G726,
 389     .priv_data_size = sizeof(G726Context),
 390     .init           = g726_encode_init,
 391     .encode2        = g726_encode_frame,
 392     .capabilities   = CODEC_CAP_SMALL_LAST_FRAME,
 393     .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
 394                                                      AV_SAMPLE_FMT_NONE },
 395     .priv_class     = &g726_class,
 396     .defaults       = defaults,
 397 };
 398 #endif
 399
 400 #if CONFIG_ADPCM_G726_DECODER || CONFIG_ADPCM_G726LE_DECODER
 401 static av_cold int g726_decode_init(AVCodecContext *avctx)
 402 {
 403     G726Context* c = avctx->priv_data;
 404
 405     if(avctx->channels > 1){
 406         avpriv_request_sample(avctx, "Decoding more than one channel");
 407         return AVERROR_PATCHWELCOME;
 408     }
 409     avctx->channels       = 1;
 410     avctx->channel_layout = AV_CH_LAYOUT_MONO;
 411
 412     c->little_endian = !strcmp(avctx->codec->name, "g726le");
 413
 414     c->code_size = avctx->bits_per_coded_sample;
 415     if (c->code_size < 2 || c->code_size > 5) {
 416         av_log(avctx, AV_LOG_ERROR, "Invalid number of bits %d\n", c->code_size);
 417         return AVERROR(EINVAL);
 418     }
 419     g726_reset(c);
 420
 421     avctx->sample_fmt = AV_SAMPLE_FMT_S16;
 422
 423     return 0;
 424 }
 425
 426 static int g726_decode_frame(AVCodecContext *avctx, void *data,
 427                              int *got_frame_ptr, AVPacket *avpkt)
 428 {
 429     AVFrame *frame     = data;
 430     const uint8_t *buf = avpkt->data;
 431     int buf_size = avpkt->size;
 432     G726Context *c = avctx->priv_data;
 433     int16_t *samples;
 434     GetBitContext gb;
 435     int out_samples, ret;
 436
 437     out_samples = buf_size * 8 / c->code_size;
 438
 439     /* get output buffer */
 440     frame->nb_samples = out_samples;
 441     if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
 442         return ret;
 443     samples = (int16_t *)frame->data[0];
 444
 445     init_get_bits(&gb, buf, buf_size * 8);
 446
 447     while (out_samples--)
 448         *samples++ = g726_decode(c, c->little_endian ?
 449                                     get_bits_le(&gb, c->code_size) :
 450                                     get_bits(&gb, c->code_size));
 451
 452     if (get_bits_left(&gb) > 0)
 453         av_log(avctx, AV_LOG_ERROR, "Frame invalidly split, missing parser?\n");
 454
 455     *got_frame_ptr = 1;
 456
 457     return buf_size;
 458 }
 459
 460 static void g726_decode_flush(AVCodecContext *avctx)
 461 {
 462     G726Context *c = avctx->priv_data;
 463     g726_reset(c);
 464 }
 465 #endif
 466
 467 #if CONFIG_ADPCM_G726_DECODER
 468 AVCodec ff_adpcm_g726_decoder = {
 469     .name           = "g726",
 470     .long_name      = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
 471     .type           = AVMEDIA_TYPE_AUDIO,
 472     .id             = AV_CODEC_ID_ADPCM_G726,
 473     .priv_data_size = sizeof(G726Context),
 474     .init           = g726_decode_init,
 475     .decode         = g726_decode_frame,
 476     .flush          = g726_decode_flush,
 477     .capabilities   = CODEC_CAP_DR1,
 478 };
 479 #endif
 480
 481 #if CONFIG_ADPCM_G726LE_DECODER
 482 AVCodec ff_adpcm_g726le_decoder = {
 483     .name           = "g726le",
 484     .type           = AVMEDIA_TYPE_AUDIO,
 485     .id             = AV_CODEC_ID_ADPCM_G726LE,
 486     .priv_data_size = sizeof(G726Context),
 487     .init           = g726_decode_init,
 488     .decode         = g726_decode_frame,
 489     .flush          = g726_decode_flush,
 490     .capabilities   = CODEC_CAP_DR1,
 491     .long_name      = NULL_IF_CONFIG_SMALL("G.726 ADPCM little-endian"),
 492 };
 493 #endif