• Main Page
  • Related Pages
  • Modules
  • Data Structures
  • Files
  • File List
  • Globals

libavcodec/g726.c

Go to the documentation of this file.
00001 /*
00002  * G.726 ADPCM audio codec
00003  * Copyright (c) 2004 Roman Shaposhnik
00004  *
00005  * This is a very straightforward rendition of the G.726
00006  * Section 4 "Computational Details".
00007  *
00008  * This file is part of FFmpeg.
00009  *
00010  * FFmpeg is free software; you can redistribute it and/or
00011  * modify it under the terms of the GNU Lesser General Public
00012  * License as published by the Free Software Foundation; either
00013  * version 2.1 of the License, or (at your option) any later version.
00014  *
00015  * FFmpeg is distributed in the hope that it will be useful,
00016  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00017  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00018  * Lesser General Public License for more details.
00019  *
00020  * You should have received a copy of the GNU Lesser General Public
00021  * License along with FFmpeg; if not, write to the Free Software
00022  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00023  */
00024 #include <limits.h>
00025 #include "libavutil/avassert.h"
00026 #include "libavutil/opt.h"
00027 #include "avcodec.h"
00028 #include "internal.h"
00029 #include "get_bits.h"
00030 #include "put_bits.h"
00031 
00038 typedef struct Float11 {
00039     uint8_t sign;   
00040     uint8_t exp;    
00041     uint8_t mant;   
00042 } Float11;
00043 
00044 static inline Float11* i2f(int i, Float11* f)
00045 {
00046     f->sign = (i < 0);
00047     if (f->sign)
00048         i = -i;
00049     f->exp = av_log2_16bit(i) + !!i;
00050     f->mant = i? (i<<6) >> f->exp : 1<<5;
00051     return f;
00052 }
00053 
00054 static inline int16_t mult(Float11* f1, Float11* f2)
00055 {
00056         int res, exp;
00057 
00058         exp = f1->exp + f2->exp;
00059         res = (((f1->mant * f2->mant) + 0x30) >> 4);
00060         res = exp > 19 ? res << (exp - 19) : res >> (19 - exp);
00061         return (f1->sign ^ f2->sign) ? -res : res;
00062 }
00063 
00064 static inline int sgn(int value)
00065 {
00066     return (value < 0) ? -1 : 1;
00067 }
00068 
00069 typedef struct G726Tables {
00070     const int* quant;         
00071     const int16_t* iquant;    
00072     const int16_t* W;         
00073     const uint8_t* F;         
00074 } G726Tables;
00075 
00076 typedef struct G726Context {
00077     AVClass *class;
00078     AVFrame frame;
00079     G726Tables tbls;    
00081     Float11 sr[2];      
00082     Float11 dq[6];      
00083     int a[2];           
00084     int b[6];           
00085     int pk[2];          
00087     int ap;             
00088     int yu;             
00089     int yl;             
00090     int dms;            
00091     int dml;            
00092     int td;             
00094     int se;             
00095     int sez;            
00096     int y;              
00097     int code_size;
00098 } G726Context;
00099 
00100 static const int quant_tbl16[] =                  
00101            { 260, INT_MAX };
00102 static const int16_t iquant_tbl16[] =
00103            { 116, 365, 365, 116 };
00104 static const int16_t W_tbl16[] =
00105            { -22, 439, 439, -22 };
00106 static const uint8_t F_tbl16[] =
00107            { 0, 7, 7, 0 };
00108 
00109 static const int quant_tbl24[] =                  
00110            {  7, 217, 330, INT_MAX };
00111 static const int16_t iquant_tbl24[] =
00112            { INT16_MIN, 135, 273, 373, 373, 273, 135, INT16_MIN };
00113 static const int16_t W_tbl24[] =
00114            { -4,  30, 137, 582, 582, 137,  30, -4 };
00115 static const uint8_t F_tbl24[] =
00116            { 0, 1, 2, 7, 7, 2, 1, 0 };
00117 
00118 static const int quant_tbl32[] =                  
00119            { -125,  79, 177, 245, 299, 348, 399, INT_MAX };
00120 static const int16_t iquant_tbl32[] =
00121          { INT16_MIN,   4, 135, 213, 273, 323, 373, 425,
00122                  425, 373, 323, 273, 213, 135,   4, INT16_MIN };
00123 static const int16_t W_tbl32[] =
00124            { -12,  18,  41,  64, 112, 198, 355, 1122,
00125             1122, 355, 198, 112,  64,  41,  18, -12};
00126 static const uint8_t F_tbl32[] =
00127            { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 };
00128 
00129 static const int quant_tbl40[] =                  
00130            { -122, -16,  67, 138, 197, 249, 297, 338,
00131               377, 412, 444, 474, 501, 527, 552, INT_MAX };
00132 static const int16_t iquant_tbl40[] =
00133          { INT16_MIN, -66,  28, 104, 169, 224, 274, 318,
00134                  358, 395, 429, 459, 488, 514, 539, 566,
00135                  566, 539, 514, 488, 459, 429, 395, 358,
00136                  318, 274, 224, 169, 104,  28, -66, INT16_MIN };
00137 static const int16_t W_tbl40[] =
00138            {   14,  14,  24,  39,  40,  41,   58,  100,
00139               141, 179, 219, 280, 358, 440,  529,  696,
00140               696, 529, 440, 358, 280, 219,  179,  141,
00141               100,  58,  41,  40,  39,  24,   14,   14 };
00142 static const uint8_t F_tbl40[] =
00143            { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6,
00144              6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 };
00145 
00146 static const G726Tables G726Tables_pool[] =
00147            {{ quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 },
00148             { quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 },
00149             { quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 },
00150             { quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }};
00151 
00152 
00156 static inline uint8_t quant(G726Context* c, int d)
00157 {
00158     int sign, exp, i, dln;
00159 
00160     sign = i = 0;
00161     if (d < 0) {
00162         sign = 1;
00163         d = -d;
00164     }
00165     exp = av_log2_16bit(d);
00166     dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2);
00167 
00168     while (c->tbls.quant[i] < INT_MAX && c->tbls.quant[i] < dln)
00169         ++i;
00170 
00171     if (sign)
00172         i = ~i;
00173     if (c->code_size != 2 && i == 0) /* I'm not sure this is a good idea */
00174         i = 0xff;
00175 
00176     return i;
00177 }
00178 
00182 static inline int16_t inverse_quant(G726Context* c, int i)
00183 {
00184     int dql, dex, dqt;
00185 
00186     dql = c->tbls.iquant[i] + (c->y >> 2);
00187     dex = (dql>>7) & 0xf;        /* 4bit exponent */
00188     dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */
00189     return (dql < 0) ? 0 : ((dqt<<dex) >> 7);
00190 }
00191 
00192 static int16_t g726_decode(G726Context* c, int I)
00193 {
00194     int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0;
00195     Float11 f;
00196     int I_sig= I >> (c->code_size - 1);
00197 
00198     dq = inverse_quant(c, I);
00199 
00200     /* Transition detect */
00201     ylint = (c->yl >> 15);
00202     ylfrac = (c->yl >> 10) & 0x1f;
00203     thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint;
00204     tr= (c->td == 1 && dq > ((3*thr2)>>2));
00205 
00206     if (I_sig)  /* get the sign */
00207         dq = -dq;
00208     re_signal = c->se + dq;
00209 
00210     /* Update second order predictor coefficient A2 and A1 */
00211     pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0;
00212     dq0 = dq ? sgn(dq) : 0;
00213     if (tr) {
00214         c->a[0] = 0;
00215         c->a[1] = 0;
00216         for (i=0; i<6; i++)
00217             c->b[i] = 0;
00218     } else {
00219         /* This is a bit crazy, but it really is +255 not +256 */
00220         fa1 = av_clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255);
00221 
00222         c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7);
00223         c->a[1] = av_clip(c->a[1], -12288, 12288);
00224         c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8);
00225         c->a[0] = av_clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]);
00226 
00227         for (i=0; i<6; i++)
00228             c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8);
00229     }
00230 
00231     /* Update Dq and Sr and Pk */
00232     c->pk[1] = c->pk[0];
00233     c->pk[0] = pk0 ? pk0 : 1;
00234     c->sr[1] = c->sr[0];
00235     i2f(re_signal, &c->sr[0]);
00236     for (i=5; i>0; i--)
00237         c->dq[i] = c->dq[i-1];
00238     i2f(dq, &c->dq[0]);
00239     c->dq[0].sign = I_sig; /* Isn't it crazy ?!?! */
00240 
00241     c->td = c->a[1] < -11776;
00242 
00243     /* Update Ap */
00244     c->dms += (c->tbls.F[I]<<4) + ((- c->dms) >> 5);
00245     c->dml += (c->tbls.F[I]<<4) + ((- c->dml) >> 7);
00246     if (tr)
00247         c->ap = 256;
00248     else {
00249         c->ap += (-c->ap) >> 4;
00250         if (c->y <= 1535 || c->td || abs((c->dms << 2) - c->dml) >= (c->dml >> 3))
00251             c->ap += 0x20;
00252     }
00253 
00254     /* Update Yu and Yl */
00255     c->yu = av_clip(c->y + c->tbls.W[I] + ((-c->y)>>5), 544, 5120);
00256     c->yl += c->yu + ((-c->yl)>>6);
00257 
00258     /* Next iteration for Y */
00259     al = (c->ap >= 256) ? 1<<6 : c->ap >> 2;
00260     c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6;
00261 
00262     /* Next iteration for SE and SEZ */
00263     c->se = 0;
00264     for (i=0; i<6; i++)
00265         c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]);
00266     c->sez = c->se >> 1;
00267     for (i=0; i<2; i++)
00268         c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]);
00269     c->se >>= 1;
00270 
00271     return av_clip(re_signal << 2, -0xffff, 0xffff);
00272 }
00273 
00274 static av_cold int g726_reset(G726Context *c)
00275 {
00276     int i;
00277 
00278     c->tbls = G726Tables_pool[c->code_size - 2];
00279     for (i=0; i<2; i++) {
00280         c->sr[i].mant = 1<<5;
00281         c->pk[i] = 1;
00282     }
00283     for (i=0; i<6; i++) {
00284         c->dq[i].mant = 1<<5;
00285     }
00286     c->yu = 544;
00287     c->yl = 34816;
00288 
00289     c->y = 544;
00290 
00291     return 0;
00292 }
00293 
00294 #if CONFIG_ADPCM_G726_ENCODER
00295 static int16_t g726_encode(G726Context* c, int16_t sig)
00296 {
00297     uint8_t i;
00298 
00299     i = quant(c, sig/4 - c->se) & ((1<<c->code_size) - 1);
00300     g726_decode(c, i);
00301     return i;
00302 }
00303 
00304 /* Interfacing to the libavcodec */
00305 
00306 static av_cold int g726_encode_init(AVCodecContext *avctx)
00307 {
00308     G726Context* c = avctx->priv_data;
00309 
00310     if (avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL &&
00311         avctx->sample_rate != 8000) {
00312         av_log(avctx, AV_LOG_ERROR, "Sample rates other than 8kHz are not "
00313                "allowed when the compliance level is higher than unofficial. "
00314                "Resample or reduce the compliance level.\n");
00315         return AVERROR(EINVAL);
00316     }
00317     av_assert0(avctx->sample_rate > 0);
00318 
00319     if(avctx->channels != 1){
00320         av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
00321         return AVERROR(EINVAL);
00322     }
00323 
00324     if (avctx->bit_rate)
00325         c->code_size = (avctx->bit_rate + avctx->sample_rate/2) / avctx->sample_rate;
00326 
00327     c->code_size = av_clip(c->code_size, 2, 5);
00328     avctx->bit_rate = c->code_size * avctx->sample_rate;
00329     avctx->bits_per_coded_sample = c->code_size;
00330 
00331     g726_reset(c);
00332 
00333     avctx->coded_frame = avcodec_alloc_frame();
00334     if (!avctx->coded_frame)
00335         return AVERROR(ENOMEM);
00336     avctx->coded_frame->key_frame = 1;
00337 
00338     /* select a frame size that will end on a byte boundary and have a size of
00339        approximately 1024 bytes */
00340     avctx->frame_size = ((int[]){ 4096, 2736, 2048, 1640 })[c->code_size - 2];
00341 
00342     return 0;
00343 }
00344 
00345 static av_cold int g726_encode_close(AVCodecContext *avctx)
00346 {
00347     av_freep(&avctx->coded_frame);
00348     return 0;
00349 }
00350 
00351 static int g726_encode_frame(AVCodecContext *avctx,
00352                             uint8_t *dst, int buf_size, void *data)
00353 {
00354     G726Context *c = avctx->priv_data;
00355     const int16_t *samples = data;
00356     PutBitContext pb;
00357     int i;
00358 
00359     init_put_bits(&pb, dst, 1024*1024);
00360 
00361     for (i = 0; i < avctx->frame_size; i++)
00362         put_bits(&pb, c->code_size, g726_encode(c, *samples++));
00363 
00364     flush_put_bits(&pb);
00365 
00366     return put_bits_count(&pb)>>3;
00367 }
00368 
00369 #define OFFSET(x) offsetof(G726Context, x)
00370 #define AE AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
00371 static const AVOption options[] = {
00372     { "code_size", "Bits per code", OFFSET(code_size), AV_OPT_TYPE_INT, { 4 }, 2, 5, AE },
00373     { NULL },
00374 };
00375 
00376 static const AVClass class = {
00377     .class_name = "g726",
00378     .item_name  = av_default_item_name,
00379     .option     = options,
00380     .version    = LIBAVUTIL_VERSION_INT,
00381 };
00382 
00383 static const AVCodecDefault defaults[] = {
00384     { "b", "0" },
00385     { NULL },
00386 };
00387 
00388 AVCodec ff_adpcm_g726_encoder = {
00389     .name           = "g726",
00390     .type           = AVMEDIA_TYPE_AUDIO,
00391     .id             = CODEC_ID_ADPCM_G726,
00392     .priv_data_size = sizeof(G726Context),
00393     .init           = g726_encode_init,
00394     .encode         = g726_encode_frame,
00395     .close          = g726_encode_close,
00396     .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
00397     .sample_fmts = (const enum AVSampleFormat[]){AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE},
00398     .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
00399     .priv_class     = &class,
00400     .defaults       = defaults,
00401 };
00402 #endif
00403 
00404 #if CONFIG_ADPCM_G726_DECODER
00405 static av_cold int g726_decode_init(AVCodecContext *avctx)
00406 {
00407     G726Context* c = avctx->priv_data;
00408 
00409     if (avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT &&
00410         avctx->sample_rate != 8000) {
00411         av_log(avctx, AV_LOG_ERROR, "Only 8kHz sample rate is allowed when "
00412                "the compliance level is strict. Reduce the compliance level "
00413                "if you wish to decode the stream anyway.\n");
00414         return AVERROR(EINVAL);
00415     }
00416 
00417     if(avctx->channels != 1){
00418         av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n");
00419         return AVERROR(EINVAL);
00420     }
00421 
00422     c->code_size = avctx->bits_per_coded_sample;
00423     if (c->code_size < 2 || c->code_size > 5) {
00424         av_log(avctx, AV_LOG_ERROR, "Invalid number of bits %d\n", c->code_size);
00425         return AVERROR(EINVAL);
00426     }
00427     g726_reset(c);
00428 
00429     avctx->sample_fmt = AV_SAMPLE_FMT_S16;
00430 
00431     avcodec_get_frame_defaults(&c->frame);
00432     avctx->coded_frame = &c->frame;
00433 
00434     return 0;
00435 }
00436 
00437 static int g726_decode_frame(AVCodecContext *avctx, void *data,
00438                              int *got_frame_ptr, AVPacket *avpkt)
00439 {
00440     const uint8_t *buf = avpkt->data;
00441     int buf_size = avpkt->size;
00442     G726Context *c = avctx->priv_data;
00443     int16_t *samples;
00444     GetBitContext gb;
00445     int out_samples, ret;
00446 
00447     out_samples = buf_size * 8 / c->code_size;
00448 
00449     /* get output buffer */
00450     c->frame.nb_samples = out_samples;
00451     if ((ret = avctx->get_buffer(avctx, &c->frame)) < 0) {
00452         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
00453         return ret;
00454     }
00455     samples = (int16_t *)c->frame.data[0];
00456 
00457     init_get_bits(&gb, buf, buf_size * 8);
00458 
00459     while (out_samples--)
00460         *samples++ = g726_decode(c, get_bits(&gb, c->code_size));
00461 
00462     if (get_bits_left(&gb) > 0)
00463         av_log(avctx, AV_LOG_ERROR, "Frame invalidly split, missing parser?\n");
00464 
00465     *got_frame_ptr   = 1;
00466     *(AVFrame *)data = c->frame;
00467 
00468     return buf_size;
00469 }
00470 
00471 static void g726_decode_flush(AVCodecContext *avctx)
00472 {
00473     G726Context *c = avctx->priv_data;
00474     g726_reset(c);
00475 }
00476 
00477 AVCodec ff_adpcm_g726_decoder = {
00478     .name           = "g726",
00479     .type           = AVMEDIA_TYPE_AUDIO,
00480     .id             = CODEC_ID_ADPCM_G726,
00481     .priv_data_size = sizeof(G726Context),
00482     .init           = g726_decode_init,
00483     .decode         = g726_decode_frame,
00484     .flush          = g726_decode_flush,
00485     .capabilities   = CODEC_CAP_DR1,
00486     .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"),
00487 };
00488 #endif
Generated on Fri Feb 1 2013 14:34:35 for FFmpeg by doxygen 1.7.1