Libav
|
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 "avcodec.h" 00026 #include "get_bits.h" 00027 #include "put_bits.h" 00028 00035 typedef struct Float11 { 00036 uint8_t sign; 00037 uint8_t exp; 00038 uint8_t mant; 00039 } Float11; 00040 00041 static inline Float11* i2f(int i, Float11* f) 00042 { 00043 f->sign = (i < 0); 00044 if (f->sign) 00045 i = -i; 00046 f->exp = av_log2_16bit(i) + !!i; 00047 f->mant = i? (i<<6) >> f->exp : 1<<5; 00048 return f; 00049 } 00050 00051 static inline int16_t mult(Float11* f1, Float11* f2) 00052 { 00053 int res, exp; 00054 00055 exp = f1->exp + f2->exp; 00056 res = (((f1->mant * f2->mant) + 0x30) >> 4); 00057 res = exp > 19 ? res << (exp - 19) : res >> (19 - exp); 00058 return (f1->sign ^ f2->sign) ? -res : res; 00059 } 00060 00061 static inline int sgn(int value) 00062 { 00063 return (value < 0) ? -1 : 1; 00064 } 00065 00066 typedef struct G726Tables { 00067 const int* quant; 00068 const int16_t* iquant; 00069 const int16_t* W; 00070 const uint8_t* F; 00071 } G726Tables; 00072 00073 typedef struct G726Context { 00074 G726Tables tbls; 00076 Float11 sr[2]; 00077 Float11 dq[6]; 00078 int a[2]; 00079 int b[6]; 00080 int pk[2]; 00082 int ap; 00083 int yu; 00084 int yl; 00085 int dms; 00086 int dml; 00087 int td; 00089 int se; 00090 int sez; 00091 int y; 00092 int code_size; 00093 } G726Context; 00094 00095 static const int quant_tbl16[] = 00096 { 260, INT_MAX }; 00097 static const int16_t iquant_tbl16[] = 00098 { 116, 365, 365, 116 }; 00099 static const int16_t W_tbl16[] = 00100 { -22, 439, 439, -22 }; 00101 static const uint8_t F_tbl16[] = 00102 { 0, 7, 7, 0 }; 00103 00104 static const int quant_tbl24[] = 00105 { 7, 217, 330, INT_MAX }; 00106 static const int16_t iquant_tbl24[] = 00107 { INT16_MIN, 135, 273, 373, 373, 273, 135, INT16_MIN }; 00108 static const int16_t W_tbl24[] = 00109 { -4, 30, 137, 582, 582, 137, 30, -4 }; 00110 static const uint8_t F_tbl24[] = 00111 { 0, 1, 2, 7, 7, 2, 1, 0 }; 00112 00113 static const int quant_tbl32[] = 00114 { -125, 79, 177, 245, 299, 348, 399, INT_MAX }; 00115 static const int16_t iquant_tbl32[] = 00116 { INT16_MIN, 4, 135, 213, 273, 323, 373, 425, 00117 425, 373, 323, 273, 213, 135, 4, INT16_MIN }; 00118 static const int16_t W_tbl32[] = 00119 { -12, 18, 41, 64, 112, 198, 355, 1122, 00120 1122, 355, 198, 112, 64, 41, 18, -12}; 00121 static const uint8_t F_tbl32[] = 00122 { 0, 0, 0, 1, 1, 1, 3, 7, 7, 3, 1, 1, 1, 0, 0, 0 }; 00123 00124 static const int quant_tbl40[] = 00125 { -122, -16, 67, 138, 197, 249, 297, 338, 00126 377, 412, 444, 474, 501, 527, 552, INT_MAX }; 00127 static const int16_t iquant_tbl40[] = 00128 { INT16_MIN, -66, 28, 104, 169, 224, 274, 318, 00129 358, 395, 429, 459, 488, 514, 539, 566, 00130 566, 539, 514, 488, 459, 429, 395, 358, 00131 318, 274, 224, 169, 104, 28, -66, INT16_MIN }; 00132 static const int16_t W_tbl40[] = 00133 { 14, 14, 24, 39, 40, 41, 58, 100, 00134 141, 179, 219, 280, 358, 440, 529, 696, 00135 696, 529, 440, 358, 280, 219, 179, 141, 00136 100, 58, 41, 40, 39, 24, 14, 14 }; 00137 static const uint8_t F_tbl40[] = 00138 { 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 2, 3, 4, 5, 6, 6, 00139 6, 6, 5, 4, 3, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 }; 00140 00141 static const G726Tables G726Tables_pool[] = 00142 {{ quant_tbl16, iquant_tbl16, W_tbl16, F_tbl16 }, 00143 { quant_tbl24, iquant_tbl24, W_tbl24, F_tbl24 }, 00144 { quant_tbl32, iquant_tbl32, W_tbl32, F_tbl32 }, 00145 { quant_tbl40, iquant_tbl40, W_tbl40, F_tbl40 }}; 00146 00147 00151 static inline uint8_t quant(G726Context* c, int d) 00152 { 00153 int sign, exp, i, dln; 00154 00155 sign = i = 0; 00156 if (d < 0) { 00157 sign = 1; 00158 d = -d; 00159 } 00160 exp = av_log2_16bit(d); 00161 dln = ((exp<<7) + (((d<<7)>>exp)&0x7f)) - (c->y>>2); 00162 00163 while (c->tbls.quant[i] < INT_MAX && c->tbls.quant[i] < dln) 00164 ++i; 00165 00166 if (sign) 00167 i = ~i; 00168 if (c->code_size != 2 && i == 0) /* I'm not sure this is a good idea */ 00169 i = 0xff; 00170 00171 return i; 00172 } 00173 00177 static inline int16_t inverse_quant(G726Context* c, int i) 00178 { 00179 int dql, dex, dqt; 00180 00181 dql = c->tbls.iquant[i] + (c->y >> 2); 00182 dex = (dql>>7) & 0xf; /* 4bit exponent */ 00183 dqt = (1<<7) + (dql & 0x7f); /* log2 -> linear */ 00184 return (dql < 0) ? 0 : ((dqt<<dex) >> 7); 00185 } 00186 00187 static int16_t g726_decode(G726Context* c, int I) 00188 { 00189 int dq, re_signal, pk0, fa1, i, tr, ylint, ylfrac, thr2, al, dq0; 00190 Float11 f; 00191 int I_sig= I >> (c->code_size - 1); 00192 00193 dq = inverse_quant(c, I); 00194 00195 /* Transition detect */ 00196 ylint = (c->yl >> 15); 00197 ylfrac = (c->yl >> 10) & 0x1f; 00198 thr2 = (ylint > 9) ? 0x1f << 10 : (0x20 + ylfrac) << ylint; 00199 tr= (c->td == 1 && dq > ((3*thr2)>>2)); 00200 00201 if (I_sig) /* get the sign */ 00202 dq = -dq; 00203 re_signal = c->se + dq; 00204 00205 /* Update second order predictor coefficient A2 and A1 */ 00206 pk0 = (c->sez + dq) ? sgn(c->sez + dq) : 0; 00207 dq0 = dq ? sgn(dq) : 0; 00208 if (tr) { 00209 c->a[0] = 0; 00210 c->a[1] = 0; 00211 for (i=0; i<6; i++) 00212 c->b[i] = 0; 00213 } else { 00214 /* This is a bit crazy, but it really is +255 not +256 */ 00215 fa1 = av_clip((-c->a[0]*c->pk[0]*pk0)>>5, -256, 255); 00216 00217 c->a[1] += 128*pk0*c->pk[1] + fa1 - (c->a[1]>>7); 00218 c->a[1] = av_clip(c->a[1], -12288, 12288); 00219 c->a[0] += 64*3*pk0*c->pk[0] - (c->a[0] >> 8); 00220 c->a[0] = av_clip(c->a[0], -(15360 - c->a[1]), 15360 - c->a[1]); 00221 00222 for (i=0; i<6; i++) 00223 c->b[i] += 128*dq0*sgn(-c->dq[i].sign) - (c->b[i]>>8); 00224 } 00225 00226 /* Update Dq and Sr and Pk */ 00227 c->pk[1] = c->pk[0]; 00228 c->pk[0] = pk0 ? pk0 : 1; 00229 c->sr[1] = c->sr[0]; 00230 i2f(re_signal, &c->sr[0]); 00231 for (i=5; i>0; i--) 00232 c->dq[i] = c->dq[i-1]; 00233 i2f(dq, &c->dq[0]); 00234 c->dq[0].sign = I_sig; /* Isn't it crazy ?!?! */ 00235 00236 c->td = c->a[1] < -11776; 00237 00238 /* Update Ap */ 00239 c->dms += (c->tbls.F[I]<<4) + ((- c->dms) >> 5); 00240 c->dml += (c->tbls.F[I]<<4) + ((- c->dml) >> 7); 00241 if (tr) 00242 c->ap = 256; 00243 else { 00244 c->ap += (-c->ap) >> 4; 00245 if (c->y <= 1535 || c->td || abs((c->dms << 2) - c->dml) >= (c->dml >> 3)) 00246 c->ap += 0x20; 00247 } 00248 00249 /* Update Yu and Yl */ 00250 c->yu = av_clip(c->y + c->tbls.W[I] + ((-c->y)>>5), 544, 5120); 00251 c->yl += c->yu + ((-c->yl)>>6); 00252 00253 /* Next iteration for Y */ 00254 al = (c->ap >= 256) ? 1<<6 : c->ap >> 2; 00255 c->y = (c->yl + (c->yu - (c->yl>>6))*al) >> 6; 00256 00257 /* Next iteration for SE and SEZ */ 00258 c->se = 0; 00259 for (i=0; i<6; i++) 00260 c->se += mult(i2f(c->b[i] >> 2, &f), &c->dq[i]); 00261 c->sez = c->se >> 1; 00262 for (i=0; i<2; i++) 00263 c->se += mult(i2f(c->a[i] >> 2, &f), &c->sr[i]); 00264 c->se >>= 1; 00265 00266 return av_clip(re_signal << 2, -0xffff, 0xffff); 00267 } 00268 00269 static av_cold int g726_reset(G726Context* c, int index) 00270 { 00271 int i; 00272 00273 c->tbls = G726Tables_pool[index]; 00274 for (i=0; i<2; i++) { 00275 c->sr[i].mant = 1<<5; 00276 c->pk[i] = 1; 00277 } 00278 for (i=0; i<6; i++) { 00279 c->dq[i].mant = 1<<5; 00280 } 00281 c->yu = 544; 00282 c->yl = 34816; 00283 00284 c->y = 544; 00285 00286 return 0; 00287 } 00288 00289 #if CONFIG_ADPCM_G726_ENCODER 00290 static int16_t g726_encode(G726Context* c, int16_t sig) 00291 { 00292 uint8_t i; 00293 00294 i = quant(c, sig/4 - c->se) & ((1<<c->code_size) - 1); 00295 g726_decode(c, i); 00296 return i; 00297 } 00298 #endif 00299 00300 /* Interfacing to the libavcodec */ 00301 00302 static av_cold int g726_init(AVCodecContext * avctx) 00303 { 00304 G726Context* c = avctx->priv_data; 00305 unsigned int index; 00306 00307 if (avctx->sample_rate <= 0) { 00308 av_log(avctx, AV_LOG_ERROR, "Samplerate is invalid\n"); 00309 return -1; 00310 } 00311 00312 index = (avctx->bit_rate + avctx->sample_rate/2) / avctx->sample_rate - 2; 00313 00314 if (avctx->bit_rate % avctx->sample_rate && avctx->codec->encode) { 00315 av_log(avctx, AV_LOG_ERROR, "Bitrate - Samplerate combination is invalid\n"); 00316 return -1; 00317 } 00318 if(avctx->channels != 1){ 00319 av_log(avctx, AV_LOG_ERROR, "Only mono is supported\n"); 00320 return -1; 00321 } 00322 if(index>3){ 00323 av_log(avctx, AV_LOG_ERROR, "Unsupported number of bits %d\n", index+2); 00324 return -1; 00325 } 00326 g726_reset(c, index); 00327 c->code_size = index+2; 00328 00329 avctx->coded_frame = avcodec_alloc_frame(); 00330 if (!avctx->coded_frame) 00331 return AVERROR(ENOMEM); 00332 avctx->coded_frame->key_frame = 1; 00333 00334 if (avctx->codec->decode) 00335 avctx->sample_fmt = SAMPLE_FMT_S16; 00336 00337 return 0; 00338 } 00339 00340 static av_cold int g726_close(AVCodecContext *avctx) 00341 { 00342 av_freep(&avctx->coded_frame); 00343 return 0; 00344 } 00345 00346 #if CONFIG_ADPCM_G726_ENCODER 00347 static int g726_encode_frame(AVCodecContext *avctx, 00348 uint8_t *dst, int buf_size, void *data) 00349 { 00350 G726Context *c = avctx->priv_data; 00351 short *samples = data; 00352 PutBitContext pb; 00353 00354 init_put_bits(&pb, dst, 1024*1024); 00355 00356 for (; buf_size; buf_size--) 00357 put_bits(&pb, c->code_size, g726_encode(c, *samples++)); 00358 00359 flush_put_bits(&pb); 00360 00361 return put_bits_count(&pb)>>3; 00362 } 00363 #endif 00364 00365 static int g726_decode_frame(AVCodecContext *avctx, 00366 void *data, int *data_size, 00367 AVPacket *avpkt) 00368 { 00369 const uint8_t *buf = avpkt->data; 00370 int buf_size = avpkt->size; 00371 G726Context *c = avctx->priv_data; 00372 short *samples = data; 00373 GetBitContext gb; 00374 00375 init_get_bits(&gb, buf, buf_size * 8); 00376 00377 while (get_bits_count(&gb) + c->code_size <= buf_size*8) 00378 *samples++ = g726_decode(c, get_bits(&gb, c->code_size)); 00379 00380 if(buf_size*8 != get_bits_count(&gb)) 00381 av_log(avctx, AV_LOG_ERROR, "Frame invalidly split, missing parser?\n"); 00382 00383 *data_size = (uint8_t*)samples - (uint8_t*)data; 00384 return buf_size; 00385 } 00386 00387 #if CONFIG_ADPCM_G726_ENCODER 00388 AVCodec adpcm_g726_encoder = { 00389 "g726", 00390 AVMEDIA_TYPE_AUDIO, 00391 CODEC_ID_ADPCM_G726, 00392 sizeof(G726Context), 00393 g726_init, 00394 g726_encode_frame, 00395 g726_close, 00396 NULL, 00397 .sample_fmts = (const enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE}, 00398 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"), 00399 }; 00400 #endif 00401 00402 AVCodec adpcm_g726_decoder = { 00403 "g726", 00404 AVMEDIA_TYPE_AUDIO, 00405 CODEC_ID_ADPCM_G726, 00406 sizeof(G726Context), 00407 g726_init, 00408 NULL, 00409 g726_close, 00410 g726_decode_frame, 00411 .long_name = NULL_IF_CONFIG_SMALL("G.726 ADPCM"), 00412 };