Mercurial > audlegacy-plugins
comparison src/ffmpeg/libavcodec/flacenc.c @ 808:e8776388b02a trunk
[svn] - add ffmpeg
| author | nenolod |
|---|---|
| date | Mon, 12 Mar 2007 11:18:54 -0700 |
| parents | |
| children |
comparison
equal
deleted
inserted
replaced
| 807:0f9c8d4d3ac4 | 808:e8776388b02a |
|---|---|
| 1 /** | |
| 2 * FLAC audio encoder | |
| 3 * Copyright (c) 2006 Justin Ruggles <jruggle@earthlink.net> | |
| 4 * | |
| 5 * This file is part of FFmpeg. | |
| 6 * | |
| 7 * FFmpeg is free software; you can redistribute it and/or | |
| 8 * modify it under the terms of the GNU Lesser General Public | |
| 9 * License as published by the Free Software Foundation; either | |
| 10 * version 2.1 of the License, or (at your option) any later version. | |
| 11 * | |
| 12 * FFmpeg is distributed in the hope that it will be useful, | |
| 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
| 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
| 15 * Lesser General Public License for more details. | |
| 16 * | |
| 17 * You should have received a copy of the GNU Lesser General Public | |
| 18 * License along with FFmpeg; if not, write to the Free Software | |
| 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
| 20 */ | |
| 21 | |
| 22 #include "avcodec.h" | |
| 23 #include "bitstream.h" | |
| 24 #include "crc.h" | |
| 25 #include "golomb.h" | |
| 26 #include "lls.h" | |
| 27 | |
| 28 #define FLAC_MAX_CH 8 | |
| 29 #define FLAC_MIN_BLOCKSIZE 16 | |
| 30 #define FLAC_MAX_BLOCKSIZE 65535 | |
| 31 | |
| 32 #define FLAC_SUBFRAME_CONSTANT 0 | |
| 33 #define FLAC_SUBFRAME_VERBATIM 1 | |
| 34 #define FLAC_SUBFRAME_FIXED 8 | |
| 35 #define FLAC_SUBFRAME_LPC 32 | |
| 36 | |
| 37 #define FLAC_CHMODE_NOT_STEREO 0 | |
| 38 #define FLAC_CHMODE_LEFT_RIGHT 1 | |
| 39 #define FLAC_CHMODE_LEFT_SIDE 8 | |
| 40 #define FLAC_CHMODE_RIGHT_SIDE 9 | |
| 41 #define FLAC_CHMODE_MID_SIDE 10 | |
| 42 | |
| 43 #define ORDER_METHOD_EST 0 | |
| 44 #define ORDER_METHOD_2LEVEL 1 | |
| 45 #define ORDER_METHOD_4LEVEL 2 | |
| 46 #define ORDER_METHOD_8LEVEL 3 | |
| 47 #define ORDER_METHOD_SEARCH 4 | |
| 48 #define ORDER_METHOD_LOG 5 | |
| 49 | |
| 50 #define FLAC_STREAMINFO_SIZE 34 | |
| 51 | |
| 52 #define MIN_LPC_ORDER 1 | |
| 53 #define MAX_LPC_ORDER 32 | |
| 54 #define MAX_FIXED_ORDER 4 | |
| 55 #define MAX_PARTITION_ORDER 8 | |
| 56 #define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER) | |
| 57 #define MAX_LPC_PRECISION 15 | |
| 58 #define MAX_LPC_SHIFT 15 | |
| 59 #define MAX_RICE_PARAM 14 | |
| 60 | |
| 61 typedef struct CompressionOptions { | |
| 62 int compression_level; | |
| 63 int block_time_ms; | |
| 64 int use_lpc; | |
| 65 int lpc_coeff_precision; | |
| 66 int min_prediction_order; | |
| 67 int max_prediction_order; | |
| 68 int prediction_order_method; | |
| 69 int min_partition_order; | |
| 70 int max_partition_order; | |
| 71 } CompressionOptions; | |
| 72 | |
| 73 typedef struct RiceContext { | |
| 74 int porder; | |
| 75 int params[MAX_PARTITIONS]; | |
| 76 } RiceContext; | |
| 77 | |
| 78 typedef struct FlacSubframe { | |
| 79 int type; | |
| 80 int type_code; | |
| 81 int obits; | |
| 82 int order; | |
| 83 int32_t coefs[MAX_LPC_ORDER]; | |
| 84 int shift; | |
| 85 RiceContext rc; | |
| 86 int32_t samples[FLAC_MAX_BLOCKSIZE]; | |
| 87 int32_t residual[FLAC_MAX_BLOCKSIZE]; | |
| 88 } FlacSubframe; | |
| 89 | |
| 90 typedef struct FlacFrame { | |
| 91 FlacSubframe subframes[FLAC_MAX_CH]; | |
| 92 int blocksize; | |
| 93 int bs_code[2]; | |
| 94 uint8_t crc8; | |
| 95 int ch_mode; | |
| 96 } FlacFrame; | |
| 97 | |
| 98 typedef struct FlacEncodeContext { | |
| 99 PutBitContext pb; | |
| 100 int channels; | |
| 101 int ch_code; | |
| 102 int samplerate; | |
| 103 int sr_code[2]; | |
| 104 int blocksize; | |
| 105 int max_framesize; | |
| 106 uint32_t frame_count; | |
| 107 FlacFrame frame; | |
| 108 CompressionOptions options; | |
| 109 AVCodecContext *avctx; | |
| 110 } FlacEncodeContext; | |
| 111 | |
| 112 static const int flac_samplerates[16] = { | |
| 113 0, 0, 0, 0, | |
| 114 8000, 16000, 22050, 24000, 32000, 44100, 48000, 96000, | |
| 115 0, 0, 0, 0 | |
| 116 }; | |
| 117 | |
| 118 static const int flac_blocksizes[16] = { | |
| 119 0, | |
| 120 192, | |
| 121 576, 1152, 2304, 4608, | |
| 122 0, 0, | |
| 123 256, 512, 1024, 2048, 4096, 8192, 16384, 32768 | |
| 124 }; | |
| 125 | |
| 126 /** | |
| 127 * Writes streaminfo metadata block to byte array | |
| 128 */ | |
| 129 static void write_streaminfo(FlacEncodeContext *s, uint8_t *header) | |
| 130 { | |
| 131 PutBitContext pb; | |
| 132 | |
| 133 memset(header, 0, FLAC_STREAMINFO_SIZE); | |
| 134 init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE); | |
| 135 | |
| 136 /* streaminfo metadata block */ | |
| 137 put_bits(&pb, 16, s->blocksize); | |
| 138 put_bits(&pb, 16, s->blocksize); | |
| 139 put_bits(&pb, 24, 0); | |
| 140 put_bits(&pb, 24, s->max_framesize); | |
| 141 put_bits(&pb, 20, s->samplerate); | |
| 142 put_bits(&pb, 3, s->channels-1); | |
| 143 put_bits(&pb, 5, 15); /* bits per sample - 1 */ | |
| 144 flush_put_bits(&pb); | |
| 145 /* total samples = 0 */ | |
| 146 /* MD5 signature = 0 */ | |
| 147 } | |
| 148 | |
| 149 /** | |
| 150 * Sets blocksize based on samplerate | |
| 151 * Chooses the closest predefined blocksize >= BLOCK_TIME_MS milliseconds | |
| 152 */ | |
| 153 static int select_blocksize(int samplerate, int block_time_ms) | |
| 154 { | |
| 155 int i; | |
| 156 int target; | |
| 157 int blocksize; | |
| 158 | |
| 159 assert(samplerate > 0); | |
| 160 blocksize = flac_blocksizes[1]; | |
| 161 target = (samplerate * block_time_ms) / 1000; | |
| 162 for(i=0; i<16; i++) { | |
| 163 if(target >= flac_blocksizes[i] && flac_blocksizes[i] > blocksize) { | |
| 164 blocksize = flac_blocksizes[i]; | |
| 165 } | |
| 166 } | |
| 167 return blocksize; | |
| 168 } | |
| 169 | |
| 170 static int flac_encode_init(AVCodecContext *avctx) | |
| 171 { | |
| 172 int freq = avctx->sample_rate; | |
| 173 int channels = avctx->channels; | |
| 174 FlacEncodeContext *s = avctx->priv_data; | |
| 175 int i, level; | |
| 176 uint8_t *streaminfo; | |
| 177 | |
| 178 s->avctx = avctx; | |
| 179 | |
| 180 if(avctx->sample_fmt != SAMPLE_FMT_S16) { | |
| 181 return -1; | |
| 182 } | |
| 183 | |
| 184 if(channels < 1 || channels > FLAC_MAX_CH) { | |
| 185 return -1; | |
| 186 } | |
| 187 s->channels = channels; | |
| 188 s->ch_code = s->channels-1; | |
| 189 | |
| 190 /* find samplerate in table */ | |
| 191 if(freq < 1) | |
| 192 return -1; | |
| 193 for(i=4; i<12; i++) { | |
| 194 if(freq == flac_samplerates[i]) { | |
| 195 s->samplerate = flac_samplerates[i]; | |
| 196 s->sr_code[0] = i; | |
| 197 s->sr_code[1] = 0; | |
| 198 break; | |
| 199 } | |
| 200 } | |
| 201 /* if not in table, samplerate is non-standard */ | |
| 202 if(i == 12) { | |
| 203 if(freq % 1000 == 0 && freq < 255000) { | |
| 204 s->sr_code[0] = 12; | |
| 205 s->sr_code[1] = freq / 1000; | |
| 206 } else if(freq % 10 == 0 && freq < 655350) { | |
| 207 s->sr_code[0] = 14; | |
| 208 s->sr_code[1] = freq / 10; | |
| 209 } else if(freq < 65535) { | |
| 210 s->sr_code[0] = 13; | |
| 211 s->sr_code[1] = freq; | |
| 212 } else { | |
| 213 return -1; | |
| 214 } | |
| 215 s->samplerate = freq; | |
| 216 } | |
| 217 | |
| 218 /* set compression option defaults based on avctx->compression_level */ | |
| 219 if(avctx->compression_level < 0) { | |
| 220 s->options.compression_level = 5; | |
| 221 } else { | |
| 222 s->options.compression_level = avctx->compression_level; | |
| 223 } | |
| 224 av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", s->options.compression_level); | |
| 225 | |
| 226 level= s->options.compression_level; | |
| 227 if(level > 12) { | |
| 228 av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n", | |
| 229 s->options.compression_level); | |
| 230 return -1; | |
| 231 } | |
| 232 | |
| 233 s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level]; | |
| 234 s->options.use_lpc = ((int[]){ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level]; | |
| 235 s->options.min_prediction_order= ((int[]){ 2, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level]; | |
| 236 s->options.max_prediction_order= ((int[]){ 3, 4, 4, 6, 8, 8, 8, 8, 12, 12, 12, 32, 32})[level]; | |
| 237 s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST, | |
| 238 ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST, | |
| 239 ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG, ORDER_METHOD_4LEVEL, | |
| 240 ORDER_METHOD_LOG, ORDER_METHOD_SEARCH, ORDER_METHOD_LOG, | |
| 241 ORDER_METHOD_SEARCH})[level]; | |
| 242 s->options.min_partition_order = ((int[]){ 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0})[level]; | |
| 243 s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8, 8, 8, 8, 8, 8, 8, 8})[level]; | |
| 244 | |
| 245 /* set compression option overrides from AVCodecContext */ | |
| 246 if(avctx->use_lpc >= 0) { | |
| 247 s->options.use_lpc = clip(avctx->use_lpc, 0, 11); | |
| 248 } | |
| 249 if(s->options.use_lpc == 1) | |
| 250 av_log(avctx, AV_LOG_DEBUG, " use lpc: Levinson-Durbin recursion with Welch window\n"); | |
| 251 else if(s->options.use_lpc > 1) | |
| 252 av_log(avctx, AV_LOG_DEBUG, " use lpc: Cholesky factorization\n"); | |
| 253 | |
| 254 if(avctx->min_prediction_order >= 0) { | |
| 255 if(s->options.use_lpc) { | |
| 256 if(avctx->min_prediction_order < MIN_LPC_ORDER || | |
| 257 avctx->min_prediction_order > MAX_LPC_ORDER) { | |
| 258 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", | |
| 259 avctx->min_prediction_order); | |
| 260 return -1; | |
| 261 } | |
| 262 } else { | |
| 263 if(avctx->min_prediction_order > MAX_FIXED_ORDER) { | |
| 264 av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", | |
| 265 avctx->min_prediction_order); | |
| 266 return -1; | |
| 267 } | |
| 268 } | |
| 269 s->options.min_prediction_order = avctx->min_prediction_order; | |
| 270 } | |
| 271 if(avctx->max_prediction_order >= 0) { | |
| 272 if(s->options.use_lpc) { | |
| 273 if(avctx->max_prediction_order < MIN_LPC_ORDER || | |
| 274 avctx->max_prediction_order > MAX_LPC_ORDER) { | |
| 275 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", | |
| 276 avctx->max_prediction_order); | |
| 277 return -1; | |
| 278 } | |
| 279 } else { | |
| 280 if(avctx->max_prediction_order > MAX_FIXED_ORDER) { | |
| 281 av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", | |
| 282 avctx->max_prediction_order); | |
| 283 return -1; | |
| 284 } | |
| 285 } | |
| 286 s->options.max_prediction_order = avctx->max_prediction_order; | |
| 287 } | |
| 288 if(s->options.max_prediction_order < s->options.min_prediction_order) { | |
| 289 av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n", | |
| 290 s->options.min_prediction_order, s->options.max_prediction_order); | |
| 291 return -1; | |
| 292 } | |
| 293 av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n", | |
| 294 s->options.min_prediction_order, s->options.max_prediction_order); | |
| 295 | |
| 296 if(avctx->prediction_order_method >= 0) { | |
| 297 if(avctx->prediction_order_method > ORDER_METHOD_LOG) { | |
| 298 av_log(avctx, AV_LOG_ERROR, "invalid prediction order method: %d\n", | |
| 299 avctx->prediction_order_method); | |
| 300 return -1; | |
| 301 } | |
| 302 s->options.prediction_order_method = avctx->prediction_order_method; | |
| 303 } | |
| 304 switch(s->options.prediction_order_method) { | |
| 305 case ORDER_METHOD_EST: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", | |
| 306 "estimate"); break; | |
| 307 case ORDER_METHOD_2LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", | |
| 308 "2-level"); break; | |
| 309 case ORDER_METHOD_4LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", | |
| 310 "4-level"); break; | |
| 311 case ORDER_METHOD_8LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", | |
| 312 "8-level"); break; | |
| 313 case ORDER_METHOD_SEARCH: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", | |
| 314 "full search"); break; | |
| 315 case ORDER_METHOD_LOG: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", | |
| 316 "log search"); break; | |
| 317 } | |
| 318 | |
| 319 if(avctx->min_partition_order >= 0) { | |
| 320 if(avctx->min_partition_order > MAX_PARTITION_ORDER) { | |
| 321 av_log(avctx, AV_LOG_ERROR, "invalid min partition order: %d\n", | |
| 322 avctx->min_partition_order); | |
| 323 return -1; | |
| 324 } | |
| 325 s->options.min_partition_order = avctx->min_partition_order; | |
| 326 } | |
| 327 if(avctx->max_partition_order >= 0) { | |
| 328 if(avctx->max_partition_order > MAX_PARTITION_ORDER) { | |
| 329 av_log(avctx, AV_LOG_ERROR, "invalid max partition order: %d\n", | |
| 330 avctx->max_partition_order); | |
| 331 return -1; | |
| 332 } | |
| 333 s->options.max_partition_order = avctx->max_partition_order; | |
| 334 } | |
| 335 if(s->options.max_partition_order < s->options.min_partition_order) { | |
| 336 av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n", | |
| 337 s->options.min_partition_order, s->options.max_partition_order); | |
| 338 return -1; | |
| 339 } | |
| 340 av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n", | |
| 341 s->options.min_partition_order, s->options.max_partition_order); | |
| 342 | |
| 343 if(avctx->frame_size > 0) { | |
| 344 if(avctx->frame_size < FLAC_MIN_BLOCKSIZE || | |
| 345 avctx->frame_size > FLAC_MAX_BLOCKSIZE) { | |
| 346 av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n", | |
| 347 avctx->frame_size); | |
| 348 return -1; | |
| 349 } | |
| 350 s->blocksize = avctx->frame_size; | |
| 351 } else { | |
| 352 s->blocksize = select_blocksize(s->samplerate, s->options.block_time_ms); | |
| 353 avctx->frame_size = s->blocksize; | |
| 354 } | |
| 355 av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", s->blocksize); | |
| 356 | |
| 357 /* set LPC precision */ | |
| 358 if(avctx->lpc_coeff_precision > 0) { | |
| 359 if(avctx->lpc_coeff_precision > MAX_LPC_PRECISION) { | |
| 360 av_log(avctx, AV_LOG_ERROR, "invalid lpc coeff precision: %d\n", | |
| 361 avctx->lpc_coeff_precision); | |
| 362 return -1; | |
| 363 } | |
| 364 s->options.lpc_coeff_precision = avctx->lpc_coeff_precision; | |
| 365 } else { | |
| 366 /* select LPC precision based on block size */ | |
| 367 if( s->blocksize <= 192) s->options.lpc_coeff_precision = 7; | |
| 368 else if(s->blocksize <= 384) s->options.lpc_coeff_precision = 8; | |
| 369 else if(s->blocksize <= 576) s->options.lpc_coeff_precision = 9; | |
| 370 else if(s->blocksize <= 1152) s->options.lpc_coeff_precision = 10; | |
| 371 else if(s->blocksize <= 2304) s->options.lpc_coeff_precision = 11; | |
| 372 else if(s->blocksize <= 4608) s->options.lpc_coeff_precision = 12; | |
| 373 else if(s->blocksize <= 8192) s->options.lpc_coeff_precision = 13; | |
| 374 else if(s->blocksize <= 16384) s->options.lpc_coeff_precision = 14; | |
| 375 else s->options.lpc_coeff_precision = 15; | |
| 376 } | |
| 377 av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n", | |
| 378 s->options.lpc_coeff_precision); | |
| 379 | |
| 380 /* set maximum encoded frame size in verbatim mode */ | |
| 381 if(s->channels == 2) { | |
| 382 s->max_framesize = 14 + ((s->blocksize * 33 + 7) >> 3); | |
| 383 } else { | |
| 384 s->max_framesize = 14 + (s->blocksize * s->channels * 2); | |
| 385 } | |
| 386 | |
| 387 streaminfo = av_malloc(FLAC_STREAMINFO_SIZE); | |
| 388 write_streaminfo(s, streaminfo); | |
| 389 avctx->extradata = streaminfo; | |
| 390 avctx->extradata_size = FLAC_STREAMINFO_SIZE; | |
| 391 | |
| 392 s->frame_count = 0; | |
| 393 | |
| 394 avctx->coded_frame = avcodec_alloc_frame(); | |
| 395 avctx->coded_frame->key_frame = 1; | |
| 396 | |
| 397 return 0; | |
| 398 } | |
| 399 | |
| 400 static void init_frame(FlacEncodeContext *s) | |
| 401 { | |
| 402 int i, ch; | |
| 403 FlacFrame *frame; | |
| 404 | |
| 405 frame = &s->frame; | |
| 406 | |
| 407 for(i=0; i<16; i++) { | |
| 408 if(s->blocksize == flac_blocksizes[i]) { | |
| 409 frame->blocksize = flac_blocksizes[i]; | |
| 410 frame->bs_code[0] = i; | |
| 411 frame->bs_code[1] = 0; | |
| 412 break; | |
| 413 } | |
| 414 } | |
| 415 if(i == 16) { | |
| 416 frame->blocksize = s->blocksize; | |
| 417 if(frame->blocksize <= 256) { | |
| 418 frame->bs_code[0] = 6; | |
| 419 frame->bs_code[1] = frame->blocksize-1; | |
| 420 } else { | |
| 421 frame->bs_code[0] = 7; | |
| 422 frame->bs_code[1] = frame->blocksize-1; | |
| 423 } | |
| 424 } | |
| 425 | |
| 426 for(ch=0; ch<s->channels; ch++) { | |
| 427 frame->subframes[ch].obits = 16; | |
| 428 } | |
| 429 } | |
| 430 | |
| 431 /** | |
| 432 * Copy channel-interleaved input samples into separate subframes | |
| 433 */ | |
| 434 static void copy_samples(FlacEncodeContext *s, int16_t *samples) | |
| 435 { | |
| 436 int i, j, ch; | |
| 437 FlacFrame *frame; | |
| 438 | |
| 439 frame = &s->frame; | |
| 440 for(i=0,j=0; i<frame->blocksize; i++) { | |
| 441 for(ch=0; ch<s->channels; ch++,j++) { | |
| 442 frame->subframes[ch].samples[i] = samples[j]; | |
| 443 } | |
| 444 } | |
| 445 } | |
| 446 | |
| 447 | |
| 448 #define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k))) | |
| 449 | |
| 450 static int find_optimal_param(uint32_t sum, int n) | |
| 451 { | |
| 452 int k, k_opt; | |
| 453 uint32_t nbits[MAX_RICE_PARAM+1]; | |
| 454 | |
| 455 k_opt = 0; | |
| 456 nbits[0] = UINT32_MAX; | |
| 457 for(k=0; k<=MAX_RICE_PARAM; k++) { | |
| 458 nbits[k] = rice_encode_count(sum, n, k); | |
| 459 if(nbits[k] < nbits[k_opt]) { | |
| 460 k_opt = k; | |
| 461 } | |
| 462 } | |
| 463 return k_opt; | |
| 464 } | |
| 465 | |
| 466 static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder, | |
| 467 uint32_t *sums, int n, int pred_order) | |
| 468 { | |
| 469 int i; | |
| 470 int k, cnt, part; | |
| 471 uint32_t all_bits; | |
| 472 | |
| 473 part = (1 << porder); | |
| 474 all_bits = 0; | |
| 475 | |
| 476 cnt = (n >> porder) - pred_order; | |
| 477 for(i=0; i<part; i++) { | |
| 478 if(i == 1) cnt = (n >> porder); | |
| 479 k = find_optimal_param(sums[i], cnt); | |
| 480 rc->params[i] = k; | |
| 481 all_bits += rice_encode_count(sums[i], cnt, k); | |
| 482 } | |
| 483 all_bits += (4 * part); | |
| 484 | |
| 485 rc->porder = porder; | |
| 486 | |
| 487 return all_bits; | |
| 488 } | |
| 489 | |
| 490 static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order, | |
| 491 uint32_t sums[][MAX_PARTITIONS]) | |
| 492 { | |
| 493 int i, j; | |
| 494 int parts; | |
| 495 uint32_t *res, *res_end; | |
| 496 | |
| 497 /* sums for highest level */ | |
| 498 parts = (1 << pmax); | |
| 499 res = &data[pred_order]; | |
| 500 res_end = &data[n >> pmax]; | |
| 501 for(i=0; i<parts; i++) { | |
| 502 sums[pmax][i] = 0; | |
| 503 while(res < res_end){ | |
| 504 sums[pmax][i] += *(res++); | |
| 505 } | |
| 506 res_end+= n >> pmax; | |
| 507 } | |
| 508 /* sums for lower levels */ | |
| 509 for(i=pmax-1; i>=pmin; i--) { | |
| 510 parts = (1 << i); | |
| 511 for(j=0; j<parts; j++) { | |
| 512 sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1]; | |
| 513 } | |
| 514 } | |
| 515 } | |
| 516 | |
| 517 static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax, | |
| 518 int32_t *data, int n, int pred_order) | |
| 519 { | |
| 520 int i; | |
| 521 uint32_t bits[MAX_PARTITION_ORDER+1]; | |
| 522 int opt_porder; | |
| 523 RiceContext tmp_rc; | |
| 524 uint32_t *udata; | |
| 525 uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS]; | |
| 526 | |
| 527 assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER); | |
| 528 assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER); | |
| 529 assert(pmin <= pmax); | |
| 530 | |
| 531 udata = av_malloc(n * sizeof(uint32_t)); | |
| 532 for(i=0; i<n; i++) { | |
| 533 udata[i] = (2*data[i]) ^ (data[i]>>31); | |
| 534 } | |
| 535 | |
| 536 calc_sums(pmin, pmax, udata, n, pred_order, sums); | |
| 537 | |
| 538 opt_porder = pmin; | |
| 539 bits[pmin] = UINT32_MAX; | |
| 540 for(i=pmin; i<=pmax; i++) { | |
| 541 bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order); | |
| 542 if(bits[i] <= bits[opt_porder]) { | |
| 543 opt_porder = i; | |
| 544 *rc= tmp_rc; | |
| 545 } | |
| 546 } | |
| 547 | |
| 548 av_freep(&udata); | |
| 549 return bits[opt_porder]; | |
| 550 } | |
| 551 | |
| 552 static int get_max_p_order(int max_porder, int n, int order) | |
| 553 { | |
| 554 int porder = FFMIN(max_porder, av_log2(n^(n-1))); | |
| 555 if(order > 0) | |
| 556 porder = FFMIN(porder, av_log2(n/order)); | |
| 557 return porder; | |
| 558 } | |
| 559 | |
| 560 static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmin, int pmax, | |
| 561 int32_t *data, int n, int pred_order, | |
| 562 int bps) | |
| 563 { | |
| 564 uint32_t bits; | |
| 565 pmin = get_max_p_order(pmin, n, pred_order); | |
| 566 pmax = get_max_p_order(pmax, n, pred_order); | |
| 567 bits = pred_order*bps + 6; | |
| 568 bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order); | |
| 569 return bits; | |
| 570 } | |
| 571 | |
| 572 static uint32_t calc_rice_params_lpc(RiceContext *rc, int pmin, int pmax, | |
| 573 int32_t *data, int n, int pred_order, | |
| 574 int bps, int precision) | |
| 575 { | |
| 576 uint32_t bits; | |
| 577 pmin = get_max_p_order(pmin, n, pred_order); | |
| 578 pmax = get_max_p_order(pmax, n, pred_order); | |
| 579 bits = pred_order*bps + 4 + 5 + pred_order*precision + 6; | |
| 580 bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order); | |
| 581 return bits; | |
| 582 } | |
| 583 | |
| 584 /** | |
| 585 * Apply Welch window function to audio block | |
| 586 */ | |
| 587 static void apply_welch_window(const int32_t *data, int len, double *w_data) | |
| 588 { | |
| 589 int i, n2; | |
| 590 double w; | |
| 591 double c; | |
| 592 | |
| 593 n2 = (len >> 1); | |
| 594 c = 2.0 / (len - 1.0); | |
| 595 for(i=0; i<n2; i++) { | |
| 596 w = c - i - 1.0; | |
| 597 w = 1.0 - (w * w); | |
| 598 w_data[i] = data[i] * w; | |
| 599 w_data[len-1-i] = data[len-1-i] * w; | |
| 600 } | |
| 601 } | |
| 602 | |
| 603 /** | |
| 604 * Calculates autocorrelation data from audio samples | |
| 605 * A Welch window function is applied before calculation. | |
| 606 */ | |
| 607 static void compute_autocorr(const int32_t *data, int len, int lag, | |
| 608 double *autoc) | |
| 609 { | |
| 610 int i, lag_ptr; | |
| 611 double tmp[len + lag]; | |
| 612 double *data1= tmp + lag; | |
| 613 | |
| 614 apply_welch_window(data, len, data1); | |
| 615 | |
| 616 for(i=0; i<lag; i++){ | |
| 617 autoc[i] = 1.0; | |
| 618 data1[i-lag]= 0.0; | |
| 619 } | |
| 620 | |
| 621 for(i=0; i<len; i++){ | |
| 622 for(lag_ptr= i-lag; lag_ptr<=i; lag_ptr++){ | |
| 623 autoc[i-lag_ptr] += data1[i] * data1[lag_ptr]; | |
| 624 } | |
| 625 } | |
| 626 } | |
| 627 | |
| 628 /** | |
| 629 * Levinson-Durbin recursion. | |
| 630 * Produces LPC coefficients from autocorrelation data. | |
| 631 */ | |
| 632 static void compute_lpc_coefs(const double *autoc, int max_order, | |
| 633 double lpc[][MAX_LPC_ORDER], double *ref) | |
| 634 { | |
| 635 int i, j, i2; | |
| 636 double r, err, tmp; | |
| 637 double lpc_tmp[MAX_LPC_ORDER]; | |
| 638 | |
| 639 for(i=0; i<max_order; i++) lpc_tmp[i] = 0; | |
| 640 err = autoc[0]; | |
| 641 | |
| 642 for(i=0; i<max_order; i++) { | |
| 643 r = -autoc[i+1]; | |
| 644 for(j=0; j<i; j++) { | |
| 645 r -= lpc_tmp[j] * autoc[i-j]; | |
| 646 } | |
| 647 r /= err; | |
| 648 ref[i] = fabs(r); | |
| 649 | |
| 650 err *= 1.0 - (r * r); | |
| 651 | |
| 652 i2 = (i >> 1); | |
| 653 lpc_tmp[i] = r; | |
| 654 for(j=0; j<i2; j++) { | |
| 655 tmp = lpc_tmp[j]; | |
| 656 lpc_tmp[j] += r * lpc_tmp[i-1-j]; | |
| 657 lpc_tmp[i-1-j] += r * tmp; | |
| 658 } | |
| 659 if(i & 1) { | |
| 660 lpc_tmp[j] += lpc_tmp[j] * r; | |
| 661 } | |
| 662 | |
| 663 for(j=0; j<=i; j++) { | |
| 664 lpc[i][j] = -lpc_tmp[j]; | |
| 665 } | |
| 666 } | |
| 667 } | |
| 668 | |
| 669 /** | |
| 670 * Quantize LPC coefficients | |
| 671 */ | |
| 672 static void quantize_lpc_coefs(double *lpc_in, int order, int precision, | |
| 673 int32_t *lpc_out, int *shift) | |
| 674 { | |
| 675 int i; | |
| 676 double cmax, error; | |
| 677 int32_t qmax; | |
| 678 int sh; | |
| 679 | |
| 680 /* define maximum levels */ | |
| 681 qmax = (1 << (precision - 1)) - 1; | |
| 682 | |
| 683 /* find maximum coefficient value */ | |
| 684 cmax = 0.0; | |
| 685 for(i=0; i<order; i++) { | |
| 686 cmax= FFMAX(cmax, fabs(lpc_in[i])); | |
| 687 } | |
| 688 | |
| 689 /* if maximum value quantizes to zero, return all zeros */ | |
| 690 if(cmax * (1 << MAX_LPC_SHIFT) < 1.0) { | |
| 691 *shift = 0; | |
| 692 memset(lpc_out, 0, sizeof(int32_t) * order); | |
| 693 return; | |
| 694 } | |
| 695 | |
| 696 /* calculate level shift which scales max coeff to available bits */ | |
| 697 sh = MAX_LPC_SHIFT; | |
| 698 while((cmax * (1 << sh) > qmax) && (sh > 0)) { | |
| 699 sh--; | |
| 700 } | |
| 701 | |
| 702 /* since negative shift values are unsupported in decoder, scale down | |
| 703 coefficients instead */ | |
| 704 if(sh == 0 && cmax > qmax) { | |
| 705 double scale = ((double)qmax) / cmax; | |
| 706 for(i=0; i<order; i++) { | |
| 707 lpc_in[i] *= scale; | |
| 708 } | |
| 709 } | |
| 710 | |
| 711 /* output quantized coefficients and level shift */ | |
| 712 error=0; | |
| 713 for(i=0; i<order; i++) { | |
| 714 error += lpc_in[i] * (1 << sh); | |
| 715 lpc_out[i] = clip(lrintf(error), -qmax, qmax); | |
| 716 error -= lpc_out[i]; | |
| 717 } | |
| 718 *shift = sh; | |
| 719 } | |
| 720 | |
| 721 static int estimate_best_order(double *ref, int max_order) | |
| 722 { | |
| 723 int i, est; | |
| 724 | |
| 725 est = 1; | |
| 726 for(i=max_order-1; i>=0; i--) { | |
| 727 if(ref[i] > 0.10) { | |
| 728 est = i+1; | |
| 729 break; | |
| 730 } | |
| 731 } | |
| 732 return est; | |
| 733 } | |
| 734 | |
| 735 /** | |
| 736 * Calculate LPC coefficients for multiple orders | |
| 737 */ | |
| 738 static int lpc_calc_coefs(const int32_t *samples, int blocksize, int max_order, | |
| 739 int precision, int32_t coefs[][MAX_LPC_ORDER], | |
| 740 int *shift, int use_lpc, int omethod) | |
| 741 { | |
| 742 double autoc[MAX_LPC_ORDER+1]; | |
| 743 double ref[MAX_LPC_ORDER]; | |
| 744 double lpc[MAX_LPC_ORDER][MAX_LPC_ORDER]; | |
| 745 int i, j, pass; | |
| 746 int opt_order; | |
| 747 | |
| 748 assert(max_order >= MIN_LPC_ORDER && max_order <= MAX_LPC_ORDER); | |
| 749 | |
| 750 if(use_lpc == 1){ | |
| 751 compute_autocorr(samples, blocksize, max_order+1, autoc); | |
| 752 | |
| 753 compute_lpc_coefs(autoc, max_order, lpc, ref); | |
| 754 }else{ | |
| 755 LLSModel m[2]; | |
| 756 double var[MAX_LPC_ORDER+1], eval, weight; | |
| 757 | |
| 758 for(pass=0; pass<use_lpc-1; pass++){ | |
| 759 av_init_lls(&m[pass&1], max_order); | |
| 760 | |
| 761 weight=0; | |
| 762 for(i=max_order; i<blocksize; i++){ | |
| 763 for(j=0; j<=max_order; j++) | |
| 764 var[j]= samples[i-j]; | |
| 765 | |
| 766 if(pass){ | |
| 767 eval= av_evaluate_lls(&m[(pass-1)&1], var+1, max_order-1); | |
| 768 eval= (512>>pass) + fabs(eval - var[0]); | |
| 769 for(j=0; j<=max_order; j++) | |
| 770 var[j]/= sqrt(eval); | |
| 771 weight += 1/eval; | |
| 772 }else | |
| 773 weight++; | |
| 774 | |
| 775 av_update_lls(&m[pass&1], var, 1.0); | |
| 776 } | |
| 777 av_solve_lls(&m[pass&1], 0.001, 0); | |
| 778 } | |
| 779 | |
| 780 for(i=0; i<max_order; i++){ | |
| 781 for(j=0; j<max_order; j++) | |
| 782 lpc[i][j]= m[(pass-1)&1].coeff[i][j]; | |
| 783 ref[i]= sqrt(m[(pass-1)&1].variance[i] / weight) * (blocksize - max_order) / 4000; | |
| 784 } | |
| 785 for(i=max_order-1; i>0; i--) | |
| 786 ref[i] = ref[i-1] - ref[i]; | |
| 787 } | |
| 788 opt_order = max_order; | |
| 789 | |
| 790 if(omethod == ORDER_METHOD_EST) { | |
| 791 opt_order = estimate_best_order(ref, max_order); | |
| 792 i = opt_order-1; | |
| 793 quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i]); | |
| 794 } else { | |
| 795 for(i=0; i<max_order; i++) { | |
| 796 quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i]); | |
| 797 } | |
| 798 } | |
| 799 | |
| 800 return opt_order; | |
| 801 } | |
| 802 | |
| 803 | |
| 804 static void encode_residual_verbatim(int32_t *res, int32_t *smp, int n) | |
| 805 { | |
| 806 assert(n > 0); | |
| 807 memcpy(res, smp, n * sizeof(int32_t)); | |
| 808 } | |
| 809 | |
| 810 static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n, | |
| 811 int order) | |
| 812 { | |
| 813 int i; | |
| 814 | |
| 815 for(i=0; i<order; i++) { | |
| 816 res[i] = smp[i]; | |
| 817 } | |
| 818 | |
| 819 if(order==0){ | |
| 820 for(i=order; i<n; i++) | |
| 821 res[i]= smp[i]; | |
| 822 }else if(order==1){ | |
| 823 for(i=order; i<n; i++) | |
| 824 res[i]= smp[i] - smp[i-1]; | |
| 825 }else if(order==2){ | |
| 826 for(i=order; i<n; i++) | |
| 827 res[i]= smp[i] - 2*smp[i-1] + smp[i-2]; | |
| 828 }else if(order==3){ | |
| 829 for(i=order; i<n; i++) | |
| 830 res[i]= smp[i] - 3*smp[i-1] + 3*smp[i-2] - smp[i-3]; | |
| 831 }else{ | |
| 832 for(i=order; i<n; i++) | |
| 833 res[i]= smp[i] - 4*smp[i-1] + 6*smp[i-2] - 4*smp[i-3] + smp[i-4]; | |
| 834 } | |
| 835 } | |
| 836 | |
| 837 static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n, | |
| 838 int order, const int32_t *coefs, int shift) | |
| 839 { | |
| 840 int i, j; | |
| 841 int32_t pred; | |
| 842 | |
| 843 for(i=0; i<order; i++) { | |
| 844 res[i] = smp[i]; | |
| 845 } | |
| 846 for(i=order; i<n; i++) { | |
| 847 pred = 0; | |
| 848 for(j=0; j<order; j++) { | |
| 849 pred += coefs[j] * smp[i-j-1]; | |
| 850 } | |
| 851 res[i] = smp[i] - (pred >> shift); | |
| 852 } | |
| 853 } | |
| 854 | |
| 855 static int encode_residual(FlacEncodeContext *ctx, int ch) | |
| 856 { | |
| 857 int i, n; | |
| 858 int min_order, max_order, opt_order, precision, omethod; | |
| 859 int min_porder, max_porder; | |
| 860 FlacFrame *frame; | |
| 861 FlacSubframe *sub; | |
| 862 int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER]; | |
| 863 int shift[MAX_LPC_ORDER]; | |
| 864 int32_t *res, *smp; | |
| 865 | |
| 866 frame = &ctx->frame; | |
| 867 sub = &frame->subframes[ch]; | |
| 868 res = sub->residual; | |
| 869 smp = sub->samples; | |
| 870 n = frame->blocksize; | |
| 871 | |
| 872 /* CONSTANT */ | |
| 873 for(i=1; i<n; i++) { | |
| 874 if(smp[i] != smp[0]) break; | |
| 875 } | |
| 876 if(i == n) { | |
| 877 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT; | |
| 878 res[0] = smp[0]; | |
| 879 return sub->obits; | |
| 880 } | |
| 881 | |
| 882 /* VERBATIM */ | |
| 883 if(n < 5) { | |
| 884 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM; | |
| 885 encode_residual_verbatim(res, smp, n); | |
| 886 return sub->obits * n; | |
| 887 } | |
| 888 | |
| 889 min_order = ctx->options.min_prediction_order; | |
| 890 max_order = ctx->options.max_prediction_order; | |
| 891 min_porder = ctx->options.min_partition_order; | |
| 892 max_porder = ctx->options.max_partition_order; | |
| 893 precision = ctx->options.lpc_coeff_precision; | |
| 894 omethod = ctx->options.prediction_order_method; | |
| 895 | |
| 896 /* FIXED */ | |
| 897 if(!ctx->options.use_lpc || max_order == 0 || (n <= max_order)) { | |
| 898 uint32_t bits[MAX_FIXED_ORDER+1]; | |
| 899 if(max_order > MAX_FIXED_ORDER) max_order = MAX_FIXED_ORDER; | |
| 900 opt_order = 0; | |
| 901 bits[0] = UINT32_MAX; | |
| 902 for(i=min_order; i<=max_order; i++) { | |
| 903 encode_residual_fixed(res, smp, n, i); | |
| 904 bits[i] = calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res, | |
| 905 n, i, sub->obits); | |
| 906 if(bits[i] < bits[opt_order]) { | |
| 907 opt_order = i; | |
| 908 } | |
| 909 } | |
| 910 sub->order = opt_order; | |
| 911 sub->type = FLAC_SUBFRAME_FIXED; | |
| 912 sub->type_code = sub->type | sub->order; | |
| 913 if(sub->order != max_order) { | |
| 914 encode_residual_fixed(res, smp, n, sub->order); | |
| 915 return calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res, n, | |
| 916 sub->order, sub->obits); | |
| 917 } | |
| 918 return bits[sub->order]; | |
| 919 } | |
| 920 | |
| 921 /* LPC */ | |
| 922 opt_order = lpc_calc_coefs(smp, n, max_order, precision, coefs, shift, ctx->options.use_lpc, omethod); | |
| 923 | |
| 924 if(omethod == ORDER_METHOD_2LEVEL || | |
| 925 omethod == ORDER_METHOD_4LEVEL || | |
| 926 omethod == ORDER_METHOD_8LEVEL) { | |
| 927 int levels = 1 << omethod; | |
| 928 uint32_t bits[levels]; | |
| 929 int order; | |
| 930 int opt_index = levels-1; | |
| 931 opt_order = max_order-1; | |
| 932 bits[opt_index] = UINT32_MAX; | |
| 933 for(i=levels-1; i>=0; i--) { | |
| 934 order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1; | |
| 935 if(order < 0) order = 0; | |
| 936 encode_residual_lpc(res, smp, n, order+1, coefs[order], shift[order]); | |
| 937 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder, | |
| 938 res, n, order+1, sub->obits, precision); | |
| 939 if(bits[i] < bits[opt_index]) { | |
| 940 opt_index = i; | |
| 941 opt_order = order; | |
| 942 } | |
| 943 } | |
| 944 opt_order++; | |
| 945 } else if(omethod == ORDER_METHOD_SEARCH) { | |
| 946 // brute-force optimal order search | |
| 947 uint32_t bits[MAX_LPC_ORDER]; | |
| 948 opt_order = 0; | |
| 949 bits[0] = UINT32_MAX; | |
| 950 for(i=min_order-1; i<max_order; i++) { | |
| 951 encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]); | |
| 952 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder, | |
| 953 res, n, i+1, sub->obits, precision); | |
| 954 if(bits[i] < bits[opt_order]) { | |
| 955 opt_order = i; | |
| 956 } | |
| 957 } | |
| 958 opt_order++; | |
| 959 } else if(omethod == ORDER_METHOD_LOG) { | |
| 960 uint32_t bits[MAX_LPC_ORDER]; | |
| 961 int step; | |
| 962 | |
| 963 opt_order= min_order - 1 + (max_order-min_order)/3; | |
| 964 memset(bits, -1, sizeof(bits)); | |
| 965 | |
| 966 for(step=16 ;step; step>>=1){ | |
| 967 int last= opt_order; | |
| 968 for(i=last-step; i<=last+step; i+= step){ | |
| 969 if(i<min_order-1 || i>=max_order || bits[i] < UINT32_MAX) | |
| 970 continue; | |
| 971 encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]); | |
| 972 bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder, | |
| 973 res, n, i+1, sub->obits, precision); | |
| 974 if(bits[i] < bits[opt_order]) | |
| 975 opt_order= i; | |
| 976 } | |
| 977 } | |
| 978 opt_order++; | |
| 979 } | |
| 980 | |
| 981 sub->order = opt_order; | |
| 982 sub->type = FLAC_SUBFRAME_LPC; | |
| 983 sub->type_code = sub->type | (sub->order-1); | |
| 984 sub->shift = shift[sub->order-1]; | |
| 985 for(i=0; i<sub->order; i++) { | |
| 986 sub->coefs[i] = coefs[sub->order-1][i]; | |
| 987 } | |
| 988 encode_residual_lpc(res, smp, n, sub->order, sub->coefs, sub->shift); | |
| 989 return calc_rice_params_lpc(&sub->rc, min_porder, max_porder, res, n, sub->order, | |
| 990 sub->obits, precision); | |
| 991 } | |
| 992 | |
| 993 static int encode_residual_v(FlacEncodeContext *ctx, int ch) | |
| 994 { | |
| 995 int i, n; | |
| 996 FlacFrame *frame; | |
| 997 FlacSubframe *sub; | |
| 998 int32_t *res, *smp; | |
| 999 | |
| 1000 frame = &ctx->frame; | |
| 1001 sub = &frame->subframes[ch]; | |
| 1002 res = sub->residual; | |
| 1003 smp = sub->samples; | |
| 1004 n = frame->blocksize; | |
| 1005 | |
| 1006 /* CONSTANT */ | |
| 1007 for(i=1; i<n; i++) { | |
| 1008 if(smp[i] != smp[0]) break; | |
| 1009 } | |
| 1010 if(i == n) { | |
| 1011 sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT; | |
| 1012 res[0] = smp[0]; | |
| 1013 return sub->obits; | |
| 1014 } | |
| 1015 | |
| 1016 /* VERBATIM */ | |
| 1017 sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM; | |
| 1018 encode_residual_verbatim(res, smp, n); | |
| 1019 return sub->obits * n; | |
| 1020 } | |
| 1021 | |
| 1022 static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n) | |
| 1023 { | |
| 1024 int i, best; | |
| 1025 int32_t lt, rt; | |
| 1026 uint64_t sum[4]; | |
| 1027 uint64_t score[4]; | |
| 1028 int k; | |
| 1029 | |
| 1030 /* calculate sum of 2nd order residual for each channel */ | |
| 1031 sum[0] = sum[1] = sum[2] = sum[3] = 0; | |
| 1032 for(i=2; i<n; i++) { | |
| 1033 lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2]; | |
| 1034 rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2]; | |
| 1035 sum[2] += FFABS((lt + rt) >> 1); | |
| 1036 sum[3] += FFABS(lt - rt); | |
| 1037 sum[0] += FFABS(lt); | |
| 1038 sum[1] += FFABS(rt); | |
| 1039 } | |
| 1040 /* estimate bit counts */ | |
| 1041 for(i=0; i<4; i++) { | |
| 1042 k = find_optimal_param(2*sum[i], n); | |
| 1043 sum[i] = rice_encode_count(2*sum[i], n, k); | |
| 1044 } | |
| 1045 | |
| 1046 /* calculate score for each mode */ | |
| 1047 score[0] = sum[0] + sum[1]; | |
| 1048 score[1] = sum[0] + sum[3]; | |
| 1049 score[2] = sum[1] + sum[3]; | |
| 1050 score[3] = sum[2] + sum[3]; | |
| 1051 | |
| 1052 /* return mode with lowest score */ | |
| 1053 best = 0; | |
| 1054 for(i=1; i<4; i++) { | |
| 1055 if(score[i] < score[best]) { | |
| 1056 best = i; | |
| 1057 } | |
| 1058 } | |
| 1059 if(best == 0) { | |
| 1060 return FLAC_CHMODE_LEFT_RIGHT; | |
| 1061 } else if(best == 1) { | |
| 1062 return FLAC_CHMODE_LEFT_SIDE; | |
| 1063 } else if(best == 2) { | |
| 1064 return FLAC_CHMODE_RIGHT_SIDE; | |
| 1065 } else { | |
| 1066 return FLAC_CHMODE_MID_SIDE; | |
| 1067 } | |
| 1068 } | |
| 1069 | |
| 1070 /** | |
| 1071 * Perform stereo channel decorrelation | |
| 1072 */ | |
| 1073 static void channel_decorrelation(FlacEncodeContext *ctx) | |
| 1074 { | |
| 1075 FlacFrame *frame; | |
| 1076 int32_t *left, *right; | |
| 1077 int i, n; | |
| 1078 | |
| 1079 frame = &ctx->frame; | |
| 1080 n = frame->blocksize; | |
| 1081 left = frame->subframes[0].samples; | |
| 1082 right = frame->subframes[1].samples; | |
| 1083 | |
| 1084 if(ctx->channels != 2) { | |
| 1085 frame->ch_mode = FLAC_CHMODE_NOT_STEREO; | |
| 1086 return; | |
| 1087 } | |
| 1088 | |
| 1089 frame->ch_mode = estimate_stereo_mode(left, right, n); | |
| 1090 | |
| 1091 /* perform decorrelation and adjust bits-per-sample */ | |
| 1092 if(frame->ch_mode == FLAC_CHMODE_LEFT_RIGHT) { | |
| 1093 return; | |
| 1094 } | |
| 1095 if(frame->ch_mode == FLAC_CHMODE_MID_SIDE) { | |
| 1096 int32_t tmp; | |
| 1097 for(i=0; i<n; i++) { | |
| 1098 tmp = left[i]; | |
| 1099 left[i] = (tmp + right[i]) >> 1; | |
| 1100 right[i] = tmp - right[i]; | |
| 1101 } | |
| 1102 frame->subframes[1].obits++; | |
| 1103 } else if(frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) { | |
| 1104 for(i=0; i<n; i++) { | |
| 1105 right[i] = left[i] - right[i]; | |
| 1106 } | |
| 1107 frame->subframes[1].obits++; | |
| 1108 } else { | |
| 1109 for(i=0; i<n; i++) { | |
| 1110 left[i] -= right[i]; | |
| 1111 } | |
| 1112 frame->subframes[0].obits++; | |
| 1113 } | |
| 1114 } | |
| 1115 | |
| 1116 static void put_sbits(PutBitContext *pb, int bits, int32_t val) | |
| 1117 { | |
| 1118 assert(bits >= 0 && bits <= 31); | |
| 1119 | |
| 1120 put_bits(pb, bits, val & ((1<<bits)-1)); | |
| 1121 } | |
| 1122 | |
| 1123 static void write_utf8(PutBitContext *pb, uint32_t val) | |
| 1124 { | |
| 1125 int bytes, shift; | |
| 1126 | |
| 1127 if(val < 0x80){ | |
| 1128 put_bits(pb, 8, val); | |
| 1129 return; | |
| 1130 } | |
| 1131 | |
| 1132 bytes= (av_log2(val)+4) / 5; | |
| 1133 shift = (bytes - 1) * 6; | |
| 1134 put_bits(pb, 8, (256 - (256>>bytes)) | (val >> shift)); | |
| 1135 while(shift >= 6){ | |
| 1136 shift -= 6; | |
| 1137 put_bits(pb, 8, 0x80 | ((val >> shift) & 0x3F)); | |
| 1138 } | |
| 1139 } | |
| 1140 | |
| 1141 static void output_frame_header(FlacEncodeContext *s) | |
| 1142 { | |
| 1143 FlacFrame *frame; | |
| 1144 int crc; | |
| 1145 | |
| 1146 frame = &s->frame; | |
| 1147 | |
| 1148 put_bits(&s->pb, 16, 0xFFF8); | |
| 1149 put_bits(&s->pb, 4, frame->bs_code[0]); | |
| 1150 put_bits(&s->pb, 4, s->sr_code[0]); | |
| 1151 if(frame->ch_mode == FLAC_CHMODE_NOT_STEREO) { | |
| 1152 put_bits(&s->pb, 4, s->ch_code); | |
| 1153 } else { | |
| 1154 put_bits(&s->pb, 4, frame->ch_mode); | |
| 1155 } | |
| 1156 put_bits(&s->pb, 3, 4); /* bits-per-sample code */ | |
| 1157 put_bits(&s->pb, 1, 0); | |
| 1158 write_utf8(&s->pb, s->frame_count); | |
| 1159 if(frame->bs_code[0] == 6) { | |
| 1160 put_bits(&s->pb, 8, frame->bs_code[1]); | |
| 1161 } else if(frame->bs_code[0] == 7) { | |
| 1162 put_bits(&s->pb, 16, frame->bs_code[1]); | |
| 1163 } | |
| 1164 if(s->sr_code[0] == 12) { | |
| 1165 put_bits(&s->pb, 8, s->sr_code[1]); | |
| 1166 } else if(s->sr_code[0] > 12) { | |
| 1167 put_bits(&s->pb, 16, s->sr_code[1]); | |
| 1168 } | |
| 1169 flush_put_bits(&s->pb); | |
| 1170 crc = av_crc(av_crc07, 0, s->pb.buf, put_bits_count(&s->pb)>>3); | |
| 1171 put_bits(&s->pb, 8, crc); | |
| 1172 } | |
| 1173 | |
| 1174 static void output_subframe_constant(FlacEncodeContext *s, int ch) | |
| 1175 { | |
| 1176 FlacSubframe *sub; | |
| 1177 int32_t res; | |
| 1178 | |
| 1179 sub = &s->frame.subframes[ch]; | |
| 1180 res = sub->residual[0]; | |
| 1181 put_sbits(&s->pb, sub->obits, res); | |
| 1182 } | |
| 1183 | |
| 1184 static void output_subframe_verbatim(FlacEncodeContext *s, int ch) | |
| 1185 { | |
| 1186 int i; | |
| 1187 FlacFrame *frame; | |
| 1188 FlacSubframe *sub; | |
| 1189 int32_t res; | |
| 1190 | |
| 1191 frame = &s->frame; | |
| 1192 sub = &frame->subframes[ch]; | |
| 1193 | |
| 1194 for(i=0; i<frame->blocksize; i++) { | |
| 1195 res = sub->residual[i]; | |
| 1196 put_sbits(&s->pb, sub->obits, res); | |
| 1197 } | |
| 1198 } | |
| 1199 | |
| 1200 static void output_residual(FlacEncodeContext *ctx, int ch) | |
| 1201 { | |
| 1202 int i, j, p, n, parts; | |
| 1203 int k, porder, psize, res_cnt; | |
| 1204 FlacFrame *frame; | |
| 1205 FlacSubframe *sub; | |
| 1206 int32_t *res; | |
| 1207 | |
| 1208 frame = &ctx->frame; | |
| 1209 sub = &frame->subframes[ch]; | |
| 1210 res = sub->residual; | |
| 1211 n = frame->blocksize; | |
| 1212 | |
| 1213 /* rice-encoded block */ | |
| 1214 put_bits(&ctx->pb, 2, 0); | |
| 1215 | |
| 1216 /* partition order */ | |
| 1217 porder = sub->rc.porder; | |
| 1218 psize = n >> porder; | |
| 1219 parts = (1 << porder); | |
| 1220 put_bits(&ctx->pb, 4, porder); | |
| 1221 res_cnt = psize - sub->order; | |
| 1222 | |
| 1223 /* residual */ | |
| 1224 j = sub->order; | |
| 1225 for(p=0; p<parts; p++) { | |
| 1226 k = sub->rc.params[p]; | |
| 1227 put_bits(&ctx->pb, 4, k); | |
| 1228 if(p == 1) res_cnt = psize; | |
| 1229 for(i=0; i<res_cnt && j<n; i++, j++) { | |
| 1230 set_sr_golomb_flac(&ctx->pb, res[j], k, INT32_MAX, 0); | |
| 1231 } | |
| 1232 } | |
| 1233 } | |
| 1234 | |
| 1235 static void output_subframe_fixed(FlacEncodeContext *ctx, int ch) | |
| 1236 { | |
| 1237 int i; | |
| 1238 FlacFrame *frame; | |
| 1239 FlacSubframe *sub; | |
| 1240 | |
| 1241 frame = &ctx->frame; | |
| 1242 sub = &frame->subframes[ch]; | |
| 1243 | |
| 1244 /* warm-up samples */ | |
| 1245 for(i=0; i<sub->order; i++) { | |
| 1246 put_sbits(&ctx->pb, sub->obits, sub->residual[i]); | |
| 1247 } | |
| 1248 | |
| 1249 /* residual */ | |
| 1250 output_residual(ctx, ch); | |
| 1251 } | |
| 1252 | |
| 1253 static void output_subframe_lpc(FlacEncodeContext *ctx, int ch) | |
| 1254 { | |
| 1255 int i, cbits; | |
| 1256 FlacFrame *frame; | |
| 1257 FlacSubframe *sub; | |
| 1258 | |
| 1259 frame = &ctx->frame; | |
| 1260 sub = &frame->subframes[ch]; | |
| 1261 | |
| 1262 /* warm-up samples */ | |
| 1263 for(i=0; i<sub->order; i++) { | |
| 1264 put_sbits(&ctx->pb, sub->obits, sub->residual[i]); | |
| 1265 } | |
| 1266 | |
| 1267 /* LPC coefficients */ | |
| 1268 cbits = ctx->options.lpc_coeff_precision; | |
| 1269 put_bits(&ctx->pb, 4, cbits-1); | |
| 1270 put_sbits(&ctx->pb, 5, sub->shift); | |
| 1271 for(i=0; i<sub->order; i++) { | |
| 1272 put_sbits(&ctx->pb, cbits, sub->coefs[i]); | |
| 1273 } | |
| 1274 | |
| 1275 /* residual */ | |
| 1276 output_residual(ctx, ch); | |
| 1277 } | |
| 1278 | |
| 1279 static void output_subframes(FlacEncodeContext *s) | |
| 1280 { | |
| 1281 FlacFrame *frame; | |
| 1282 FlacSubframe *sub; | |
| 1283 int ch; | |
| 1284 | |
| 1285 frame = &s->frame; | |
| 1286 | |
| 1287 for(ch=0; ch<s->channels; ch++) { | |
| 1288 sub = &frame->subframes[ch]; | |
| 1289 | |
| 1290 /* subframe header */ | |
| 1291 put_bits(&s->pb, 1, 0); | |
| 1292 put_bits(&s->pb, 6, sub->type_code); | |
| 1293 put_bits(&s->pb, 1, 0); /* no wasted bits */ | |
| 1294 | |
| 1295 /* subframe */ | |
| 1296 if(sub->type == FLAC_SUBFRAME_CONSTANT) { | |
| 1297 output_subframe_constant(s, ch); | |
| 1298 } else if(sub->type == FLAC_SUBFRAME_VERBATIM) { | |
| 1299 output_subframe_verbatim(s, ch); | |
| 1300 } else if(sub->type == FLAC_SUBFRAME_FIXED) { | |
| 1301 output_subframe_fixed(s, ch); | |
| 1302 } else if(sub->type == FLAC_SUBFRAME_LPC) { | |
| 1303 output_subframe_lpc(s, ch); | |
| 1304 } | |
| 1305 } | |
| 1306 } | |
| 1307 | |
| 1308 static void output_frame_footer(FlacEncodeContext *s) | |
| 1309 { | |
| 1310 int crc; | |
| 1311 flush_put_bits(&s->pb); | |
| 1312 crc = bswap_16(av_crc(av_crc8005, 0, s->pb.buf, put_bits_count(&s->pb)>>3)); | |
| 1313 put_bits(&s->pb, 16, crc); | |
| 1314 flush_put_bits(&s->pb); | |
| 1315 } | |
| 1316 | |
| 1317 static int flac_encode_frame(AVCodecContext *avctx, uint8_t *frame, | |
| 1318 int buf_size, void *data) | |
| 1319 { | |
| 1320 int ch; | |
| 1321 FlacEncodeContext *s; | |
| 1322 int16_t *samples = data; | |
| 1323 int out_bytes; | |
| 1324 | |
| 1325 s = avctx->priv_data; | |
| 1326 | |
| 1327 s->blocksize = avctx->frame_size; | |
| 1328 init_frame(s); | |
| 1329 | |
| 1330 copy_samples(s, samples); | |
| 1331 | |
| 1332 channel_decorrelation(s); | |
| 1333 | |
| 1334 for(ch=0; ch<s->channels; ch++) { | |
| 1335 encode_residual(s, ch); | |
| 1336 } | |
| 1337 init_put_bits(&s->pb, frame, buf_size); | |
| 1338 output_frame_header(s); | |
| 1339 output_subframes(s); | |
| 1340 output_frame_footer(s); | |
| 1341 out_bytes = put_bits_count(&s->pb) >> 3; | |
| 1342 | |
| 1343 if(out_bytes > s->max_framesize || out_bytes >= buf_size) { | |
| 1344 /* frame too large. use verbatim mode */ | |
| 1345 for(ch=0; ch<s->channels; ch++) { | |
| 1346 encode_residual_v(s, ch); | |
| 1347 } | |
| 1348 init_put_bits(&s->pb, frame, buf_size); | |
| 1349 output_frame_header(s); | |
| 1350 output_subframes(s); | |
| 1351 output_frame_footer(s); | |
| 1352 out_bytes = put_bits_count(&s->pb) >> 3; | |
| 1353 | |
| 1354 if(out_bytes > s->max_framesize || out_bytes >= buf_size) { | |
| 1355 /* still too large. must be an error. */ | |
| 1356 av_log(avctx, AV_LOG_ERROR, "error encoding frame\n"); | |
| 1357 return -1; | |
| 1358 } | |
| 1359 } | |
| 1360 | |
| 1361 s->frame_count++; | |
| 1362 return out_bytes; | |
| 1363 } | |
| 1364 | |
| 1365 static int flac_encode_close(AVCodecContext *avctx) | |
| 1366 { | |
| 1367 av_freep(&avctx->extradata); | |
| 1368 avctx->extradata_size = 0; | |
| 1369 av_freep(&avctx->coded_frame); | |
| 1370 return 0; | |
| 1371 } | |
| 1372 | |
| 1373 AVCodec flac_encoder = { | |
| 1374 "flac", | |
| 1375 CODEC_TYPE_AUDIO, | |
| 1376 CODEC_ID_FLAC, | |
| 1377 sizeof(FlacEncodeContext), | |
| 1378 flac_encode_init, | |
| 1379 flac_encode_frame, | |
| 1380 flac_encode_close, | |
| 1381 NULL, | |
| 1382 .capabilities = CODEC_CAP_SMALL_LAST_FRAME, | |
| 1383 }; |
