| 1 | /* |
| 2 | * FLAC audio encoder |
| 3 | * Copyright (c) 2006 Justin Ruggles <justin.ruggles@gmail.com> |
| 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 "libavutil/avassert.h" |
| 23 | #include "libavutil/crc.h" |
| 24 | #include "libavutil/intmath.h" |
| 25 | #include "libavutil/md5.h" |
| 26 | #include "libavutil/opt.h" |
| 27 | #include "avcodec.h" |
| 28 | #include "bswapdsp.h" |
| 29 | #include "put_bits.h" |
| 30 | #include "golomb.h" |
| 31 | #include "internal.h" |
| 32 | #include "lpc.h" |
| 33 | #include "flac.h" |
| 34 | #include "flacdata.h" |
| 35 | #include "flacdsp.h" |
| 36 | |
| 37 | #define FLAC_SUBFRAME_CONSTANT 0 |
| 38 | #define FLAC_SUBFRAME_VERBATIM 1 |
| 39 | #define FLAC_SUBFRAME_FIXED 8 |
| 40 | #define FLAC_SUBFRAME_LPC 32 |
| 41 | |
| 42 | #define MAX_FIXED_ORDER 4 |
| 43 | #define MAX_PARTITION_ORDER 8 |
| 44 | #define MAX_PARTITIONS (1 << MAX_PARTITION_ORDER) |
| 45 | #define MAX_LPC_PRECISION 15 |
| 46 | #define MAX_LPC_SHIFT 15 |
| 47 | |
| 48 | enum CodingMode { |
| 49 | CODING_MODE_RICE = 4, |
| 50 | CODING_MODE_RICE2 = 5, |
| 51 | }; |
| 52 | |
| 53 | typedef struct CompressionOptions { |
| 54 | int compression_level; |
| 55 | int block_time_ms; |
| 56 | enum FFLPCType lpc_type; |
| 57 | int lpc_passes; |
| 58 | int lpc_coeff_precision; |
| 59 | int min_prediction_order; |
| 60 | int max_prediction_order; |
| 61 | int prediction_order_method; |
| 62 | int min_partition_order; |
| 63 | int max_partition_order; |
| 64 | int ch_mode; |
| 65 | } CompressionOptions; |
| 66 | |
| 67 | typedef struct RiceContext { |
| 68 | enum CodingMode coding_mode; |
| 69 | int porder; |
| 70 | int params[MAX_PARTITIONS]; |
| 71 | } RiceContext; |
| 72 | |
| 73 | typedef struct FlacSubframe { |
| 74 | int type; |
| 75 | int type_code; |
| 76 | int obits; |
| 77 | int wasted; |
| 78 | int order; |
| 79 | int32_t coefs[MAX_LPC_ORDER]; |
| 80 | int shift; |
| 81 | RiceContext rc; |
| 82 | int32_t samples[FLAC_MAX_BLOCKSIZE]; |
| 83 | int32_t residual[FLAC_MAX_BLOCKSIZE+11]; |
| 84 | } FlacSubframe; |
| 85 | |
| 86 | typedef struct FlacFrame { |
| 87 | FlacSubframe subframes[FLAC_MAX_CHANNELS]; |
| 88 | int blocksize; |
| 89 | int bs_code[2]; |
| 90 | uint8_t crc8; |
| 91 | int ch_mode; |
| 92 | int verbatim_only; |
| 93 | } FlacFrame; |
| 94 | |
| 95 | typedef struct FlacEncodeContext { |
| 96 | AVClass *class; |
| 97 | PutBitContext pb; |
| 98 | int channels; |
| 99 | int samplerate; |
| 100 | int sr_code[2]; |
| 101 | int bps_code; |
| 102 | int max_blocksize; |
| 103 | int min_framesize; |
| 104 | int max_framesize; |
| 105 | int max_encoded_framesize; |
| 106 | uint32_t frame_count; |
| 107 | uint64_t sample_count; |
| 108 | uint8_t md5sum[16]; |
| 109 | FlacFrame frame; |
| 110 | CompressionOptions options; |
| 111 | AVCodecContext *avctx; |
| 112 | LPCContext lpc_ctx; |
| 113 | struct AVMD5 *md5ctx; |
| 114 | uint8_t *md5_buffer; |
| 115 | unsigned int md5_buffer_size; |
| 116 | BswapDSPContext bdsp; |
| 117 | FLACDSPContext flac_dsp; |
| 118 | |
| 119 | int flushed; |
| 120 | int64_t next_pts; |
| 121 | } FlacEncodeContext; |
| 122 | |
| 123 | |
| 124 | /** |
| 125 | * Write streaminfo metadata block to byte array. |
| 126 | */ |
| 127 | static void write_streaminfo(FlacEncodeContext *s, uint8_t *header) |
| 128 | { |
| 129 | PutBitContext pb; |
| 130 | |
| 131 | memset(header, 0, FLAC_STREAMINFO_SIZE); |
| 132 | init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE); |
| 133 | |
| 134 | /* streaminfo metadata block */ |
| 135 | put_bits(&pb, 16, s->max_blocksize); |
| 136 | put_bits(&pb, 16, s->max_blocksize); |
| 137 | put_bits(&pb, 24, s->min_framesize); |
| 138 | put_bits(&pb, 24, s->max_framesize); |
| 139 | put_bits(&pb, 20, s->samplerate); |
| 140 | put_bits(&pb, 3, s->channels-1); |
| 141 | put_bits(&pb, 5, s->avctx->bits_per_raw_sample - 1); |
| 142 | /* write 36-bit sample count in 2 put_bits() calls */ |
| 143 | put_bits(&pb, 24, (s->sample_count & 0xFFFFFF000LL) >> 12); |
| 144 | put_bits(&pb, 12, s->sample_count & 0x000000FFFLL); |
| 145 | flush_put_bits(&pb); |
| 146 | memcpy(&header[18], s->md5sum, 16); |
| 147 | } |
| 148 | |
| 149 | |
| 150 | /** |
| 151 | * Set blocksize based on samplerate. |
| 152 | * Choose the closest predefined blocksize >= BLOCK_TIME_MS milliseconds. |
| 153 | */ |
| 154 | static int select_blocksize(int samplerate, int block_time_ms) |
| 155 | { |
| 156 | int i; |
| 157 | int target; |
| 158 | int blocksize; |
| 159 | |
| 160 | av_assert0(samplerate > 0); |
| 161 | blocksize = ff_flac_blocksize_table[1]; |
| 162 | target = (samplerate * block_time_ms) / 1000; |
| 163 | for (i = 0; i < 16; i++) { |
| 164 | if (target >= ff_flac_blocksize_table[i] && |
| 165 | ff_flac_blocksize_table[i] > blocksize) { |
| 166 | blocksize = ff_flac_blocksize_table[i]; |
| 167 | } |
| 168 | } |
| 169 | return blocksize; |
| 170 | } |
| 171 | |
| 172 | |
| 173 | static av_cold void dprint_compression_options(FlacEncodeContext *s) |
| 174 | { |
| 175 | AVCodecContext *avctx = s->avctx; |
| 176 | CompressionOptions *opt = &s->options; |
| 177 | |
| 178 | av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", opt->compression_level); |
| 179 | |
| 180 | switch (opt->lpc_type) { |
| 181 | case FF_LPC_TYPE_NONE: |
| 182 | av_log(avctx, AV_LOG_DEBUG, " lpc type: None\n"); |
| 183 | break; |
| 184 | case FF_LPC_TYPE_FIXED: |
| 185 | av_log(avctx, AV_LOG_DEBUG, " lpc type: Fixed pre-defined coefficients\n"); |
| 186 | break; |
| 187 | case FF_LPC_TYPE_LEVINSON: |
| 188 | av_log(avctx, AV_LOG_DEBUG, " lpc type: Levinson-Durbin recursion with Welch window\n"); |
| 189 | break; |
| 190 | case FF_LPC_TYPE_CHOLESKY: |
| 191 | av_log(avctx, AV_LOG_DEBUG, " lpc type: Cholesky factorization, %d pass%s\n", |
| 192 | opt->lpc_passes, opt->lpc_passes == 1 ? "" : "es"); |
| 193 | break; |
| 194 | } |
| 195 | |
| 196 | av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n", |
| 197 | opt->min_prediction_order, opt->max_prediction_order); |
| 198 | |
| 199 | switch (opt->prediction_order_method) { |
| 200 | case ORDER_METHOD_EST: |
| 201 | av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "estimate"); |
| 202 | break; |
| 203 | case ORDER_METHOD_2LEVEL: |
| 204 | av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "2-level"); |
| 205 | break; |
| 206 | case ORDER_METHOD_4LEVEL: |
| 207 | av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "4-level"); |
| 208 | break; |
| 209 | case ORDER_METHOD_8LEVEL: |
| 210 | av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "8-level"); |
| 211 | break; |
| 212 | case ORDER_METHOD_SEARCH: |
| 213 | av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "full search"); |
| 214 | break; |
| 215 | case ORDER_METHOD_LOG: |
| 216 | av_log(avctx, AV_LOG_DEBUG, " order method: %s\n", "log search"); |
| 217 | break; |
| 218 | } |
| 219 | |
| 220 | |
| 221 | av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n", |
| 222 | opt->min_partition_order, opt->max_partition_order); |
| 223 | |
| 224 | av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", avctx->frame_size); |
| 225 | |
| 226 | av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n", |
| 227 | opt->lpc_coeff_precision); |
| 228 | } |
| 229 | |
| 230 | |
| 231 | static av_cold int flac_encode_init(AVCodecContext *avctx) |
| 232 | { |
| 233 | int freq = avctx->sample_rate; |
| 234 | int channels = avctx->channels; |
| 235 | FlacEncodeContext *s = avctx->priv_data; |
| 236 | int i, level, ret; |
| 237 | uint8_t *streaminfo; |
| 238 | |
| 239 | s->avctx = avctx; |
| 240 | |
| 241 | switch (avctx->sample_fmt) { |
| 242 | case AV_SAMPLE_FMT_S16: |
| 243 | avctx->bits_per_raw_sample = 16; |
| 244 | s->bps_code = 4; |
| 245 | break; |
| 246 | case AV_SAMPLE_FMT_S32: |
| 247 | if (avctx->bits_per_raw_sample != 24) |
| 248 | av_log(avctx, AV_LOG_WARNING, "encoding as 24 bits-per-sample\n"); |
| 249 | avctx->bits_per_raw_sample = 24; |
| 250 | s->bps_code = 6; |
| 251 | break; |
| 252 | } |
| 253 | |
| 254 | if (channels < 1 || channels > FLAC_MAX_CHANNELS) { |
| 255 | av_log(avctx, AV_LOG_ERROR, "%d channels not supported (max %d)\n", |
| 256 | channels, FLAC_MAX_CHANNELS); |
| 257 | return AVERROR(EINVAL); |
| 258 | } |
| 259 | s->channels = channels; |
| 260 | |
| 261 | /* find samplerate in table */ |
| 262 | if (freq < 1) |
| 263 | return -1; |
| 264 | for (i = 4; i < 12; i++) { |
| 265 | if (freq == ff_flac_sample_rate_table[i]) { |
| 266 | s->samplerate = ff_flac_sample_rate_table[i]; |
| 267 | s->sr_code[0] = i; |
| 268 | s->sr_code[1] = 0; |
| 269 | break; |
| 270 | } |
| 271 | } |
| 272 | /* if not in table, samplerate is non-standard */ |
| 273 | if (i == 12) { |
| 274 | if (freq % 1000 == 0 && freq < 255000) { |
| 275 | s->sr_code[0] = 12; |
| 276 | s->sr_code[1] = freq / 1000; |
| 277 | } else if (freq % 10 == 0 && freq < 655350) { |
| 278 | s->sr_code[0] = 14; |
| 279 | s->sr_code[1] = freq / 10; |
| 280 | } else if (freq < 65535) { |
| 281 | s->sr_code[0] = 13; |
| 282 | s->sr_code[1] = freq; |
| 283 | } else { |
| 284 | av_log(avctx, AV_LOG_ERROR, "%d Hz not supported\n", freq); |
| 285 | return AVERROR(EINVAL); |
| 286 | } |
| 287 | s->samplerate = freq; |
| 288 | } |
| 289 | |
| 290 | /* set compression option defaults based on avctx->compression_level */ |
| 291 | if (avctx->compression_level < 0) |
| 292 | s->options.compression_level = 5; |
| 293 | else |
| 294 | s->options.compression_level = avctx->compression_level; |
| 295 | |
| 296 | level = s->options.compression_level; |
| 297 | if (level > 12) { |
| 298 | av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n", |
| 299 | s->options.compression_level); |
| 300 | return AVERROR(EINVAL); |
| 301 | } |
| 302 | |
| 303 | s->options.block_time_ms = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level]; |
| 304 | |
| 305 | if (s->options.lpc_type == FF_LPC_TYPE_DEFAULT) |
| 306 | s->options.lpc_type = ((int[]){ FF_LPC_TYPE_FIXED, FF_LPC_TYPE_FIXED, FF_LPC_TYPE_FIXED, |
| 307 | FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, |
| 308 | FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, |
| 309 | FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, FF_LPC_TYPE_LEVINSON, |
| 310 | FF_LPC_TYPE_LEVINSON})[level]; |
| 311 | |
| 312 | s->options.min_prediction_order = ((int[]){ 2, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1})[level]; |
| 313 | s->options.max_prediction_order = ((int[]){ 3, 4, 4, 6, 8, 8, 8, 8, 12, 12, 12, 32, 32})[level]; |
| 314 | |
| 315 | if (s->options.prediction_order_method < 0) |
| 316 | s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST, |
| 317 | ORDER_METHOD_EST, ORDER_METHOD_EST, ORDER_METHOD_EST, |
| 318 | ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG, ORDER_METHOD_4LEVEL, |
| 319 | ORDER_METHOD_LOG, ORDER_METHOD_SEARCH, ORDER_METHOD_LOG, |
| 320 | ORDER_METHOD_SEARCH})[level]; |
| 321 | |
| 322 | if (s->options.min_partition_order > s->options.max_partition_order) { |
| 323 | av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n", |
| 324 | s->options.min_partition_order, s->options.max_partition_order); |
| 325 | return AVERROR(EINVAL); |
| 326 | } |
| 327 | if (s->options.min_partition_order < 0) |
| 328 | s->options.min_partition_order = ((int[]){ 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0})[level]; |
| 329 | if (s->options.max_partition_order < 0) |
| 330 | s->options.max_partition_order = ((int[]){ 2, 2, 3, 3, 3, 8, 8, 8, 8, 8, 8, 8, 8})[level]; |
| 331 | |
| 332 | if (s->options.lpc_type == FF_LPC_TYPE_NONE) { |
| 333 | s->options.min_prediction_order = 0; |
| 334 | } else if (avctx->min_prediction_order >= 0) { |
| 335 | if (s->options.lpc_type == FF_LPC_TYPE_FIXED) { |
| 336 | if (avctx->min_prediction_order > MAX_FIXED_ORDER) { |
| 337 | av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", |
| 338 | avctx->min_prediction_order); |
| 339 | return AVERROR(EINVAL); |
| 340 | } |
| 341 | } else if (avctx->min_prediction_order < MIN_LPC_ORDER || |
| 342 | avctx->min_prediction_order > MAX_LPC_ORDER) { |
| 343 | av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n", |
| 344 | avctx->min_prediction_order); |
| 345 | return AVERROR(EINVAL); |
| 346 | } |
| 347 | s->options.min_prediction_order = avctx->min_prediction_order; |
| 348 | } |
| 349 | if (s->options.lpc_type == FF_LPC_TYPE_NONE) { |
| 350 | s->options.max_prediction_order = 0; |
| 351 | } else if (avctx->max_prediction_order >= 0) { |
| 352 | if (s->options.lpc_type == FF_LPC_TYPE_FIXED) { |
| 353 | if (avctx->max_prediction_order > MAX_FIXED_ORDER) { |
| 354 | av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", |
| 355 | avctx->max_prediction_order); |
| 356 | return AVERROR(EINVAL); |
| 357 | } |
| 358 | } else if (avctx->max_prediction_order < MIN_LPC_ORDER || |
| 359 | avctx->max_prediction_order > MAX_LPC_ORDER) { |
| 360 | av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n", |
| 361 | avctx->max_prediction_order); |
| 362 | return AVERROR(EINVAL); |
| 363 | } |
| 364 | s->options.max_prediction_order = avctx->max_prediction_order; |
| 365 | } |
| 366 | if (s->options.max_prediction_order < s->options.min_prediction_order) { |
| 367 | av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n", |
| 368 | s->options.min_prediction_order, s->options.max_prediction_order); |
| 369 | return AVERROR(EINVAL); |
| 370 | } |
| 371 | |
| 372 | if (avctx->frame_size > 0) { |
| 373 | if (avctx->frame_size < FLAC_MIN_BLOCKSIZE || |
| 374 | avctx->frame_size > FLAC_MAX_BLOCKSIZE) { |
| 375 | av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n", |
| 376 | avctx->frame_size); |
| 377 | return AVERROR(EINVAL); |
| 378 | } |
| 379 | } else { |
| 380 | s->avctx->frame_size = select_blocksize(s->samplerate, s->options.block_time_ms); |
| 381 | } |
| 382 | s->max_blocksize = s->avctx->frame_size; |
| 383 | |
| 384 | /* set maximum encoded frame size in verbatim mode */ |
| 385 | s->max_framesize = ff_flac_get_max_frame_size(s->avctx->frame_size, |
| 386 | s->channels, |
| 387 | s->avctx->bits_per_raw_sample); |
| 388 | |
| 389 | /* initialize MD5 context */ |
| 390 | s->md5ctx = av_md5_alloc(); |
| 391 | if (!s->md5ctx) |
| 392 | return AVERROR(ENOMEM); |
| 393 | av_md5_init(s->md5ctx); |
| 394 | |
| 395 | streaminfo = av_malloc(FLAC_STREAMINFO_SIZE); |
| 396 | if (!streaminfo) |
| 397 | return AVERROR(ENOMEM); |
| 398 | write_streaminfo(s, streaminfo); |
| 399 | avctx->extradata = streaminfo; |
| 400 | avctx->extradata_size = FLAC_STREAMINFO_SIZE; |
| 401 | |
| 402 | s->frame_count = 0; |
| 403 | s->min_framesize = s->max_framesize; |
| 404 | |
| 405 | if (channels == 3 && |
| 406 | avctx->channel_layout != (AV_CH_LAYOUT_STEREO|AV_CH_FRONT_CENTER) || |
| 407 | channels == 4 && |
| 408 | avctx->channel_layout != AV_CH_LAYOUT_2_2 && |
| 409 | avctx->channel_layout != AV_CH_LAYOUT_QUAD || |
| 410 | channels == 5 && |
| 411 | avctx->channel_layout != AV_CH_LAYOUT_5POINT0 && |
| 412 | avctx->channel_layout != AV_CH_LAYOUT_5POINT0_BACK || |
| 413 | channels == 6 && |
| 414 | avctx->channel_layout != AV_CH_LAYOUT_5POINT1 && |
| 415 | avctx->channel_layout != AV_CH_LAYOUT_5POINT1_BACK) { |
| 416 | if (avctx->channel_layout) { |
| 417 | av_log(avctx, AV_LOG_ERROR, "Channel layout not supported by Flac, " |
| 418 | "output stream will have incorrect " |
| 419 | "channel layout.\n"); |
| 420 | } else { |
| 421 | av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The encoder " |
| 422 | "will use Flac channel layout for " |
| 423 | "%d channels.\n", channels); |
| 424 | } |
| 425 | } |
| 426 | |
| 427 | ret = ff_lpc_init(&s->lpc_ctx, avctx->frame_size, |
| 428 | s->options.max_prediction_order, FF_LPC_TYPE_LEVINSON); |
| 429 | |
| 430 | ff_bswapdsp_init(&s->bdsp); |
| 431 | ff_flacdsp_init(&s->flac_dsp, avctx->sample_fmt, |
| 432 | avctx->bits_per_raw_sample); |
| 433 | |
| 434 | dprint_compression_options(s); |
| 435 | |
| 436 | return ret; |
| 437 | } |
| 438 | |
| 439 | |
| 440 | static void init_frame(FlacEncodeContext *s, int nb_samples) |
| 441 | { |
| 442 | int i, ch; |
| 443 | FlacFrame *frame; |
| 444 | |
| 445 | frame = &s->frame; |
| 446 | |
| 447 | for (i = 0; i < 16; i++) { |
| 448 | if (nb_samples == ff_flac_blocksize_table[i]) { |
| 449 | frame->blocksize = ff_flac_blocksize_table[i]; |
| 450 | frame->bs_code[0] = i; |
| 451 | frame->bs_code[1] = 0; |
| 452 | break; |
| 453 | } |
| 454 | } |
| 455 | if (i == 16) { |
| 456 | frame->blocksize = nb_samples; |
| 457 | if (frame->blocksize <= 256) { |
| 458 | frame->bs_code[0] = 6; |
| 459 | frame->bs_code[1] = frame->blocksize-1; |
| 460 | } else { |
| 461 | frame->bs_code[0] = 7; |
| 462 | frame->bs_code[1] = frame->blocksize-1; |
| 463 | } |
| 464 | } |
| 465 | |
| 466 | for (ch = 0; ch < s->channels; ch++) { |
| 467 | FlacSubframe *sub = &frame->subframes[ch]; |
| 468 | |
| 469 | sub->wasted = 0; |
| 470 | sub->obits = s->avctx->bits_per_raw_sample; |
| 471 | |
| 472 | if (sub->obits > 16) |
| 473 | sub->rc.coding_mode = CODING_MODE_RICE2; |
| 474 | else |
| 475 | sub->rc.coding_mode = CODING_MODE_RICE; |
| 476 | } |
| 477 | |
| 478 | frame->verbatim_only = 0; |
| 479 | } |
| 480 | |
| 481 | |
| 482 | /** |
| 483 | * Copy channel-interleaved input samples into separate subframes. |
| 484 | */ |
| 485 | static void copy_samples(FlacEncodeContext *s, const void *samples) |
| 486 | { |
| 487 | int i, j, ch; |
| 488 | FlacFrame *frame; |
| 489 | int shift = av_get_bytes_per_sample(s->avctx->sample_fmt) * 8 - |
| 490 | s->avctx->bits_per_raw_sample; |
| 491 | |
| 492 | #define COPY_SAMPLES(bits) do { \ |
| 493 | const int ## bits ## _t *samples0 = samples; \ |
| 494 | frame = &s->frame; \ |
| 495 | for (i = 0, j = 0; i < frame->blocksize; i++) \ |
| 496 | for (ch = 0; ch < s->channels; ch++, j++) \ |
| 497 | frame->subframes[ch].samples[i] = samples0[j] >> shift; \ |
| 498 | } while (0) |
| 499 | |
| 500 | if (s->avctx->sample_fmt == AV_SAMPLE_FMT_S16) |
| 501 | COPY_SAMPLES(16); |
| 502 | else |
| 503 | COPY_SAMPLES(32); |
| 504 | } |
| 505 | |
| 506 | |
| 507 | static uint64_t rice_count_exact(int32_t *res, int n, int k) |
| 508 | { |
| 509 | int i; |
| 510 | uint64_t count = 0; |
| 511 | |
| 512 | for (i = 0; i < n; i++) { |
| 513 | int32_t v = -2 * res[i] - 1; |
| 514 | v ^= v >> 31; |
| 515 | count += (v >> k) + 1 + k; |
| 516 | } |
| 517 | return count; |
| 518 | } |
| 519 | |
| 520 | |
| 521 | static uint64_t subframe_count_exact(FlacEncodeContext *s, FlacSubframe *sub, |
| 522 | int pred_order) |
| 523 | { |
| 524 | int p, porder, psize; |
| 525 | int i, part_end; |
| 526 | uint64_t count = 0; |
| 527 | |
| 528 | /* subframe header */ |
| 529 | count += 8; |
| 530 | |
| 531 | /* subframe */ |
| 532 | if (sub->type == FLAC_SUBFRAME_CONSTANT) { |
| 533 | count += sub->obits; |
| 534 | } else if (sub->type == FLAC_SUBFRAME_VERBATIM) { |
| 535 | count += s->frame.blocksize * sub->obits; |
| 536 | } else { |
| 537 | /* warm-up samples */ |
| 538 | count += pred_order * sub->obits; |
| 539 | |
| 540 | /* LPC coefficients */ |
| 541 | if (sub->type == FLAC_SUBFRAME_LPC) |
| 542 | count += 4 + 5 + pred_order * s->options.lpc_coeff_precision; |
| 543 | |
| 544 | /* rice-encoded block */ |
| 545 | count += 2; |
| 546 | |
| 547 | /* partition order */ |
| 548 | porder = sub->rc.porder; |
| 549 | psize = s->frame.blocksize >> porder; |
| 550 | count += 4; |
| 551 | |
| 552 | /* residual */ |
| 553 | i = pred_order; |
| 554 | part_end = psize; |
| 555 | for (p = 0; p < 1 << porder; p++) { |
| 556 | int k = sub->rc.params[p]; |
| 557 | count += sub->rc.coding_mode; |
| 558 | count += rice_count_exact(&sub->residual[i], part_end - i, k); |
| 559 | i = part_end; |
| 560 | part_end = FFMIN(s->frame.blocksize, part_end + psize); |
| 561 | } |
| 562 | } |
| 563 | |
| 564 | return count; |
| 565 | } |
| 566 | |
| 567 | |
| 568 | #define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k))) |
| 569 | |
| 570 | /** |
| 571 | * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0. |
| 572 | */ |
| 573 | static int find_optimal_param(uint64_t sum, int n, int max_param) |
| 574 | { |
| 575 | int k; |
| 576 | uint64_t sum2; |
| 577 | |
| 578 | if (sum <= n >> 1) |
| 579 | return 0; |
| 580 | sum2 = sum - (n >> 1); |
| 581 | k = av_log2(av_clipl_int32(sum2 / n)); |
| 582 | return FFMIN(k, max_param); |
| 583 | } |
| 584 | |
| 585 | |
| 586 | static uint64_t calc_optimal_rice_params(RiceContext *rc, int porder, |
| 587 | uint64_t *sums, int n, int pred_order) |
| 588 | { |
| 589 | int i; |
| 590 | int k, cnt, part, max_param; |
| 591 | uint64_t all_bits; |
| 592 | |
| 593 | max_param = (1 << rc->coding_mode) - 2; |
| 594 | |
| 595 | part = (1 << porder); |
| 596 | all_bits = 4 * part; |
| 597 | |
| 598 | cnt = (n >> porder) - pred_order; |
| 599 | for (i = 0; i < part; i++) { |
| 600 | k = find_optimal_param(sums[i], cnt, max_param); |
| 601 | rc->params[i] = k; |
| 602 | all_bits += rice_encode_count(sums[i], cnt, k); |
| 603 | cnt = n >> porder; |
| 604 | } |
| 605 | |
| 606 | rc->porder = porder; |
| 607 | |
| 608 | return all_bits; |
| 609 | } |
| 610 | |
| 611 | |
| 612 | static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order, |
| 613 | uint64_t sums[][MAX_PARTITIONS]) |
| 614 | { |
| 615 | int i, j; |
| 616 | int parts; |
| 617 | uint32_t *res, *res_end; |
| 618 | |
| 619 | /* sums for highest level */ |
| 620 | parts = (1 << pmax); |
| 621 | res = &data[pred_order]; |
| 622 | res_end = &data[n >> pmax]; |
| 623 | for (i = 0; i < parts; i++) { |
| 624 | uint64_t sum = 0; |
| 625 | while (res < res_end) |
| 626 | sum += *(res++); |
| 627 | sums[pmax][i] = sum; |
| 628 | res_end += n >> pmax; |
| 629 | } |
| 630 | /* sums for lower levels */ |
| 631 | for (i = pmax - 1; i >= pmin; i--) { |
| 632 | parts = (1 << i); |
| 633 | for (j = 0; j < parts; j++) |
| 634 | sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1]; |
| 635 | } |
| 636 | } |
| 637 | |
| 638 | |
| 639 | static uint64_t calc_rice_params(RiceContext *rc, int pmin, int pmax, |
| 640 | int32_t *data, int n, int pred_order) |
| 641 | { |
| 642 | int i; |
| 643 | uint64_t bits[MAX_PARTITION_ORDER+1]; |
| 644 | int opt_porder; |
| 645 | RiceContext tmp_rc; |
| 646 | uint32_t *udata; |
| 647 | uint64_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS]; |
| 648 | |
| 649 | av_assert1(pmin >= 0 && pmin <= MAX_PARTITION_ORDER); |
| 650 | av_assert1(pmax >= 0 && pmax <= MAX_PARTITION_ORDER); |
| 651 | av_assert1(pmin <= pmax); |
| 652 | |
| 653 | tmp_rc.coding_mode = rc->coding_mode; |
| 654 | |
| 655 | udata = av_malloc_array(n, sizeof(uint32_t)); |
| 656 | for (i = 0; i < n; i++) |
| 657 | udata[i] = (2*data[i]) ^ (data[i]>>31); |
| 658 | |
| 659 | calc_sums(pmin, pmax, udata, n, pred_order, sums); |
| 660 | |
| 661 | opt_porder = pmin; |
| 662 | bits[pmin] = UINT32_MAX; |
| 663 | for (i = pmin; i <= pmax; i++) { |
| 664 | bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order); |
| 665 | if (bits[i] <= bits[opt_porder]) { |
| 666 | opt_porder = i; |
| 667 | *rc = tmp_rc; |
| 668 | } |
| 669 | } |
| 670 | |
| 671 | av_freep(&udata); |
| 672 | return bits[opt_porder]; |
| 673 | } |
| 674 | |
| 675 | |
| 676 | static int get_max_p_order(int max_porder, int n, int order) |
| 677 | { |
| 678 | int porder = FFMIN(max_porder, av_log2(n^(n-1))); |
| 679 | if (order > 0) |
| 680 | porder = FFMIN(porder, av_log2(n/order)); |
| 681 | return porder; |
| 682 | } |
| 683 | |
| 684 | |
| 685 | static uint64_t find_subframe_rice_params(FlacEncodeContext *s, |
| 686 | FlacSubframe *sub, int pred_order) |
| 687 | { |
| 688 | int pmin = get_max_p_order(s->options.min_partition_order, |
| 689 | s->frame.blocksize, pred_order); |
| 690 | int pmax = get_max_p_order(s->options.max_partition_order, |
| 691 | s->frame.blocksize, pred_order); |
| 692 | |
| 693 | uint64_t bits = 8 + pred_order * sub->obits + 2 + sub->rc.coding_mode; |
| 694 | if (sub->type == FLAC_SUBFRAME_LPC) |
| 695 | bits += 4 + 5 + pred_order * s->options.lpc_coeff_precision; |
| 696 | bits += calc_rice_params(&sub->rc, pmin, pmax, sub->residual, |
| 697 | s->frame.blocksize, pred_order); |
| 698 | return bits; |
| 699 | } |
| 700 | |
| 701 | |
| 702 | static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n, |
| 703 | int order) |
| 704 | { |
| 705 | int i; |
| 706 | |
| 707 | for (i = 0; i < order; i++) |
| 708 | res[i] = smp[i]; |
| 709 | |
| 710 | if (order == 0) { |
| 711 | for (i = order; i < n; i++) |
| 712 | res[i] = smp[i]; |
| 713 | } else if (order == 1) { |
| 714 | for (i = order; i < n; i++) |
| 715 | res[i] = smp[i] - smp[i-1]; |
| 716 | } else if (order == 2) { |
| 717 | int a = smp[order-1] - smp[order-2]; |
| 718 | for (i = order; i < n; i += 2) { |
| 719 | int b = smp[i ] - smp[i-1]; |
| 720 | res[i] = b - a; |
| 721 | a = smp[i+1] - smp[i ]; |
| 722 | res[i+1] = a - b; |
| 723 | } |
| 724 | } else if (order == 3) { |
| 725 | int a = smp[order-1] - smp[order-2]; |
| 726 | int c = smp[order-1] - 2*smp[order-2] + smp[order-3]; |
| 727 | for (i = order; i < n; i += 2) { |
| 728 | int b = smp[i ] - smp[i-1]; |
| 729 | int d = b - a; |
| 730 | res[i] = d - c; |
| 731 | a = smp[i+1] - smp[i ]; |
| 732 | c = a - b; |
| 733 | res[i+1] = c - d; |
| 734 | } |
| 735 | } else { |
| 736 | int a = smp[order-1] - smp[order-2]; |
| 737 | int c = smp[order-1] - 2*smp[order-2] + smp[order-3]; |
| 738 | int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4]; |
| 739 | for (i = order; i < n; i += 2) { |
| 740 | int b = smp[i ] - smp[i-1]; |
| 741 | int d = b - a; |
| 742 | int f = d - c; |
| 743 | res[i ] = f - e; |
| 744 | a = smp[i+1] - smp[i ]; |
| 745 | c = a - b; |
| 746 | e = c - d; |
| 747 | res[i+1] = e - f; |
| 748 | } |
| 749 | } |
| 750 | } |
| 751 | |
| 752 | |
| 753 | static int encode_residual_ch(FlacEncodeContext *s, int ch) |
| 754 | { |
| 755 | int i, n; |
| 756 | int min_order, max_order, opt_order, omethod; |
| 757 | FlacFrame *frame; |
| 758 | FlacSubframe *sub; |
| 759 | int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER]; |
| 760 | int shift[MAX_LPC_ORDER]; |
| 761 | int32_t *res, *smp; |
| 762 | |
| 763 | frame = &s->frame; |
| 764 | sub = &frame->subframes[ch]; |
| 765 | res = sub->residual; |
| 766 | smp = sub->samples; |
| 767 | n = frame->blocksize; |
| 768 | |
| 769 | /* CONSTANT */ |
| 770 | for (i = 1; i < n; i++) |
| 771 | if(smp[i] != smp[0]) |
| 772 | break; |
| 773 | if (i == n) { |
| 774 | sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT; |
| 775 | res[0] = smp[0]; |
| 776 | return subframe_count_exact(s, sub, 0); |
| 777 | } |
| 778 | |
| 779 | /* VERBATIM */ |
| 780 | if (frame->verbatim_only || n < 5) { |
| 781 | sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM; |
| 782 | memcpy(res, smp, n * sizeof(int32_t)); |
| 783 | return subframe_count_exact(s, sub, 0); |
| 784 | } |
| 785 | |
| 786 | min_order = s->options.min_prediction_order; |
| 787 | max_order = s->options.max_prediction_order; |
| 788 | omethod = s->options.prediction_order_method; |
| 789 | |
| 790 | /* FIXED */ |
| 791 | sub->type = FLAC_SUBFRAME_FIXED; |
| 792 | if (s->options.lpc_type == FF_LPC_TYPE_NONE || |
| 793 | s->options.lpc_type == FF_LPC_TYPE_FIXED || n <= max_order) { |
| 794 | uint64_t bits[MAX_FIXED_ORDER+1]; |
| 795 | if (max_order > MAX_FIXED_ORDER) |
| 796 | max_order = MAX_FIXED_ORDER; |
| 797 | opt_order = 0; |
| 798 | bits[0] = UINT32_MAX; |
| 799 | for (i = min_order; i <= max_order; i++) { |
| 800 | encode_residual_fixed(res, smp, n, i); |
| 801 | bits[i] = find_subframe_rice_params(s, sub, i); |
| 802 | if (bits[i] < bits[opt_order]) |
| 803 | opt_order = i; |
| 804 | } |
| 805 | sub->order = opt_order; |
| 806 | sub->type_code = sub->type | sub->order; |
| 807 | if (sub->order != max_order) { |
| 808 | encode_residual_fixed(res, smp, n, sub->order); |
| 809 | find_subframe_rice_params(s, sub, sub->order); |
| 810 | } |
| 811 | return subframe_count_exact(s, sub, sub->order); |
| 812 | } |
| 813 | |
| 814 | /* LPC */ |
| 815 | sub->type = FLAC_SUBFRAME_LPC; |
| 816 | opt_order = ff_lpc_calc_coefs(&s->lpc_ctx, smp, n, min_order, max_order, |
| 817 | s->options.lpc_coeff_precision, coefs, shift, s->options.lpc_type, |
| 818 | s->options.lpc_passes, omethod, |
| 819 | MAX_LPC_SHIFT, 0); |
| 820 | |
| 821 | if (omethod == ORDER_METHOD_2LEVEL || |
| 822 | omethod == ORDER_METHOD_4LEVEL || |
| 823 | omethod == ORDER_METHOD_8LEVEL) { |
| 824 | int levels = 1 << omethod; |
| 825 | uint64_t bits[1 << ORDER_METHOD_8LEVEL]; |
| 826 | int order = -1; |
| 827 | int opt_index = levels-1; |
| 828 | opt_order = max_order-1; |
| 829 | bits[opt_index] = UINT32_MAX; |
| 830 | for (i = levels-1; i >= 0; i--) { |
| 831 | int last_order = order; |
| 832 | order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1; |
| 833 | order = av_clip(order, min_order - 1, max_order - 1); |
| 834 | if (order == last_order) |
| 835 | continue; |
| 836 | s->flac_dsp.lpc_encode(res, smp, n, order+1, coefs[order], |
| 837 | shift[order]); |
| 838 | bits[i] = find_subframe_rice_params(s, sub, order+1); |
| 839 | if (bits[i] < bits[opt_index]) { |
| 840 | opt_index = i; |
| 841 | opt_order = order; |
| 842 | } |
| 843 | } |
| 844 | opt_order++; |
| 845 | } else if (omethod == ORDER_METHOD_SEARCH) { |
| 846 | // brute-force optimal order search |
| 847 | uint64_t bits[MAX_LPC_ORDER]; |
| 848 | opt_order = 0; |
| 849 | bits[0] = UINT32_MAX; |
| 850 | for (i = min_order-1; i < max_order; i++) { |
| 851 | s->flac_dsp.lpc_encode(res, smp, n, i+1, coefs[i], shift[i]); |
| 852 | bits[i] = find_subframe_rice_params(s, sub, i+1); |
| 853 | if (bits[i] < bits[opt_order]) |
| 854 | opt_order = i; |
| 855 | } |
| 856 | opt_order++; |
| 857 | } else if (omethod == ORDER_METHOD_LOG) { |
| 858 | uint64_t bits[MAX_LPC_ORDER]; |
| 859 | int step; |
| 860 | |
| 861 | opt_order = min_order - 1 + (max_order-min_order)/3; |
| 862 | memset(bits, -1, sizeof(bits)); |
| 863 | |
| 864 | for (step = 16; step; step >>= 1) { |
| 865 | int last = opt_order; |
| 866 | for (i = last-step; i <= last+step; i += step) { |
| 867 | if (i < min_order-1 || i >= max_order || bits[i] < UINT32_MAX) |
| 868 | continue; |
| 869 | s->flac_dsp.lpc_encode(res, smp, n, i+1, coefs[i], shift[i]); |
| 870 | bits[i] = find_subframe_rice_params(s, sub, i+1); |
| 871 | if (bits[i] < bits[opt_order]) |
| 872 | opt_order = i; |
| 873 | } |
| 874 | } |
| 875 | opt_order++; |
| 876 | } |
| 877 | |
| 878 | sub->order = opt_order; |
| 879 | sub->type_code = sub->type | (sub->order-1); |
| 880 | sub->shift = shift[sub->order-1]; |
| 881 | for (i = 0; i < sub->order; i++) |
| 882 | sub->coefs[i] = coefs[sub->order-1][i]; |
| 883 | |
| 884 | s->flac_dsp.lpc_encode(res, smp, n, sub->order, sub->coefs, sub->shift); |
| 885 | |
| 886 | find_subframe_rice_params(s, sub, sub->order); |
| 887 | |
| 888 | return subframe_count_exact(s, sub, sub->order); |
| 889 | } |
| 890 | |
| 891 | |
| 892 | static int count_frame_header(FlacEncodeContext *s) |
| 893 | { |
| 894 | uint8_t av_unused tmp; |
| 895 | int count; |
| 896 | |
| 897 | /* |
| 898 | <14> Sync code |
| 899 | <1> Reserved |
| 900 | <1> Blocking strategy |
| 901 | <4> Block size in inter-channel samples |
| 902 | <4> Sample rate |
| 903 | <4> Channel assignment |
| 904 | <3> Sample size in bits |
| 905 | <1> Reserved |
| 906 | */ |
| 907 | count = 32; |
| 908 | |
| 909 | /* coded frame number */ |
| 910 | PUT_UTF8(s->frame_count, tmp, count += 8;) |
| 911 | |
| 912 | /* explicit block size */ |
| 913 | if (s->frame.bs_code[0] == 6) |
| 914 | count += 8; |
| 915 | else if (s->frame.bs_code[0] == 7) |
| 916 | count += 16; |
| 917 | |
| 918 | /* explicit sample rate */ |
| 919 | count += ((s->sr_code[0] == 12) + (s->sr_code[0] > 12)) * 8; |
| 920 | |
| 921 | /* frame header CRC-8 */ |
| 922 | count += 8; |
| 923 | |
| 924 | return count; |
| 925 | } |
| 926 | |
| 927 | |
| 928 | static int encode_frame(FlacEncodeContext *s) |
| 929 | { |
| 930 | int ch; |
| 931 | uint64_t count; |
| 932 | |
| 933 | count = count_frame_header(s); |
| 934 | |
| 935 | for (ch = 0; ch < s->channels; ch++) |
| 936 | count += encode_residual_ch(s, ch); |
| 937 | |
| 938 | count += (8 - (count & 7)) & 7; // byte alignment |
| 939 | count += 16; // CRC-16 |
| 940 | |
| 941 | count >>= 3; |
| 942 | if (count > INT_MAX) |
| 943 | return AVERROR_BUG; |
| 944 | return count; |
| 945 | } |
| 946 | |
| 947 | |
| 948 | static void remove_wasted_bits(FlacEncodeContext *s) |
| 949 | { |
| 950 | int ch, i; |
| 951 | |
| 952 | for (ch = 0; ch < s->channels; ch++) { |
| 953 | FlacSubframe *sub = &s->frame.subframes[ch]; |
| 954 | int32_t v = 0; |
| 955 | |
| 956 | for (i = 0; i < s->frame.blocksize; i++) { |
| 957 | v |= sub->samples[i]; |
| 958 | if (v & 1) |
| 959 | break; |
| 960 | } |
| 961 | |
| 962 | if (v && !(v & 1)) { |
| 963 | v = av_ctz(v); |
| 964 | |
| 965 | for (i = 0; i < s->frame.blocksize; i++) |
| 966 | sub->samples[i] >>= v; |
| 967 | |
| 968 | sub->wasted = v; |
| 969 | sub->obits -= v; |
| 970 | |
| 971 | /* for 24-bit, check if removing wasted bits makes the range better |
| 972 | suited for using RICE instead of RICE2 for entropy coding */ |
| 973 | if (sub->obits <= 17) |
| 974 | sub->rc.coding_mode = CODING_MODE_RICE; |
| 975 | } |
| 976 | } |
| 977 | } |
| 978 | |
| 979 | |
| 980 | static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n, |
| 981 | int max_rice_param) |
| 982 | { |
| 983 | int i, best; |
| 984 | int32_t lt, rt; |
| 985 | uint64_t sum[4]; |
| 986 | uint64_t score[4]; |
| 987 | int k; |
| 988 | |
| 989 | /* calculate sum of 2nd order residual for each channel */ |
| 990 | sum[0] = sum[1] = sum[2] = sum[3] = 0; |
| 991 | for (i = 2; i < n; i++) { |
| 992 | lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2]; |
| 993 | rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2]; |
| 994 | sum[2] += FFABS((lt + rt) >> 1); |
| 995 | sum[3] += FFABS(lt - rt); |
| 996 | sum[0] += FFABS(lt); |
| 997 | sum[1] += FFABS(rt); |
| 998 | } |
| 999 | /* estimate bit counts */ |
| 1000 | for (i = 0; i < 4; i++) { |
| 1001 | k = find_optimal_param(2 * sum[i], n, max_rice_param); |
| 1002 | sum[i] = rice_encode_count( 2 * sum[i], n, k); |
| 1003 | } |
| 1004 | |
| 1005 | /* calculate score for each mode */ |
| 1006 | score[0] = sum[0] + sum[1]; |
| 1007 | score[1] = sum[0] + sum[3]; |
| 1008 | score[2] = sum[1] + sum[3]; |
| 1009 | score[3] = sum[2] + sum[3]; |
| 1010 | |
| 1011 | /* return mode with lowest score */ |
| 1012 | best = 0; |
| 1013 | for (i = 1; i < 4; i++) |
| 1014 | if (score[i] < score[best]) |
| 1015 | best = i; |
| 1016 | |
| 1017 | return best; |
| 1018 | } |
| 1019 | |
| 1020 | |
| 1021 | /** |
| 1022 | * Perform stereo channel decorrelation. |
| 1023 | */ |
| 1024 | static void channel_decorrelation(FlacEncodeContext *s) |
| 1025 | { |
| 1026 | FlacFrame *frame; |
| 1027 | int32_t *left, *right; |
| 1028 | int i, n; |
| 1029 | |
| 1030 | frame = &s->frame; |
| 1031 | n = frame->blocksize; |
| 1032 | left = frame->subframes[0].samples; |
| 1033 | right = frame->subframes[1].samples; |
| 1034 | |
| 1035 | if (s->channels != 2) { |
| 1036 | frame->ch_mode = FLAC_CHMODE_INDEPENDENT; |
| 1037 | return; |
| 1038 | } |
| 1039 | |
| 1040 | if (s->options.ch_mode < 0) { |
| 1041 | int max_rice_param = (1 << frame->subframes[0].rc.coding_mode) - 2; |
| 1042 | frame->ch_mode = estimate_stereo_mode(left, right, n, max_rice_param); |
| 1043 | } else |
| 1044 | frame->ch_mode = s->options.ch_mode; |
| 1045 | |
| 1046 | /* perform decorrelation and adjust bits-per-sample */ |
| 1047 | if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT) |
| 1048 | return; |
| 1049 | if (frame->ch_mode == FLAC_CHMODE_MID_SIDE) { |
| 1050 | int32_t tmp; |
| 1051 | for (i = 0; i < n; i++) { |
| 1052 | tmp = left[i]; |
| 1053 | left[i] = (tmp + right[i]) >> 1; |
| 1054 | right[i] = tmp - right[i]; |
| 1055 | } |
| 1056 | frame->subframes[1].obits++; |
| 1057 | } else if (frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) { |
| 1058 | for (i = 0; i < n; i++) |
| 1059 | right[i] = left[i] - right[i]; |
| 1060 | frame->subframes[1].obits++; |
| 1061 | } else { |
| 1062 | for (i = 0; i < n; i++) |
| 1063 | left[i] -= right[i]; |
| 1064 | frame->subframes[0].obits++; |
| 1065 | } |
| 1066 | } |
| 1067 | |
| 1068 | |
| 1069 | static void write_utf8(PutBitContext *pb, uint32_t val) |
| 1070 | { |
| 1071 | uint8_t tmp; |
| 1072 | PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);) |
| 1073 | } |
| 1074 | |
| 1075 | |
| 1076 | static void write_frame_header(FlacEncodeContext *s) |
| 1077 | { |
| 1078 | FlacFrame *frame; |
| 1079 | int crc; |
| 1080 | |
| 1081 | frame = &s->frame; |
| 1082 | |
| 1083 | put_bits(&s->pb, 16, 0xFFF8); |
| 1084 | put_bits(&s->pb, 4, frame->bs_code[0]); |
| 1085 | put_bits(&s->pb, 4, s->sr_code[0]); |
| 1086 | |
| 1087 | if (frame->ch_mode == FLAC_CHMODE_INDEPENDENT) |
| 1088 | put_bits(&s->pb, 4, s->channels-1); |
| 1089 | else |
| 1090 | put_bits(&s->pb, 4, frame->ch_mode + FLAC_MAX_CHANNELS - 1); |
| 1091 | |
| 1092 | put_bits(&s->pb, 3, s->bps_code); |
| 1093 | put_bits(&s->pb, 1, 0); |
| 1094 | write_utf8(&s->pb, s->frame_count); |
| 1095 | |
| 1096 | if (frame->bs_code[0] == 6) |
| 1097 | put_bits(&s->pb, 8, frame->bs_code[1]); |
| 1098 | else if (frame->bs_code[0] == 7) |
| 1099 | put_bits(&s->pb, 16, frame->bs_code[1]); |
| 1100 | |
| 1101 | if (s->sr_code[0] == 12) |
| 1102 | put_bits(&s->pb, 8, s->sr_code[1]); |
| 1103 | else if (s->sr_code[0] > 12) |
| 1104 | put_bits(&s->pb, 16, s->sr_code[1]); |
| 1105 | |
| 1106 | flush_put_bits(&s->pb); |
| 1107 | crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0, s->pb.buf, |
| 1108 | put_bits_count(&s->pb) >> 3); |
| 1109 | put_bits(&s->pb, 8, crc); |
| 1110 | } |
| 1111 | |
| 1112 | |
| 1113 | static void write_subframes(FlacEncodeContext *s) |
| 1114 | { |
| 1115 | int ch; |
| 1116 | |
| 1117 | for (ch = 0; ch < s->channels; ch++) { |
| 1118 | FlacSubframe *sub = &s->frame.subframes[ch]; |
| 1119 | int i, p, porder, psize; |
| 1120 | int32_t *part_end; |
| 1121 | int32_t *res = sub->residual; |
| 1122 | int32_t *frame_end = &sub->residual[s->frame.blocksize]; |
| 1123 | |
| 1124 | /* subframe header */ |
| 1125 | put_bits(&s->pb, 1, 0); |
| 1126 | put_bits(&s->pb, 6, sub->type_code); |
| 1127 | put_bits(&s->pb, 1, !!sub->wasted); |
| 1128 | if (sub->wasted) |
| 1129 | put_bits(&s->pb, sub->wasted, 1); |
| 1130 | |
| 1131 | /* subframe */ |
| 1132 | if (sub->type == FLAC_SUBFRAME_CONSTANT) { |
| 1133 | put_sbits(&s->pb, sub->obits, res[0]); |
| 1134 | } else if (sub->type == FLAC_SUBFRAME_VERBATIM) { |
| 1135 | while (res < frame_end) |
| 1136 | put_sbits(&s->pb, sub->obits, *res++); |
| 1137 | } else { |
| 1138 | /* warm-up samples */ |
| 1139 | for (i = 0; i < sub->order; i++) |
| 1140 | put_sbits(&s->pb, sub->obits, *res++); |
| 1141 | |
| 1142 | /* LPC coefficients */ |
| 1143 | if (sub->type == FLAC_SUBFRAME_LPC) { |
| 1144 | int cbits = s->options.lpc_coeff_precision; |
| 1145 | put_bits( &s->pb, 4, cbits-1); |
| 1146 | put_sbits(&s->pb, 5, sub->shift); |
| 1147 | for (i = 0; i < sub->order; i++) |
| 1148 | put_sbits(&s->pb, cbits, sub->coefs[i]); |
| 1149 | } |
| 1150 | |
| 1151 | /* rice-encoded block */ |
| 1152 | put_bits(&s->pb, 2, sub->rc.coding_mode - 4); |
| 1153 | |
| 1154 | /* partition order */ |
| 1155 | porder = sub->rc.porder; |
| 1156 | psize = s->frame.blocksize >> porder; |
| 1157 | put_bits(&s->pb, 4, porder); |
| 1158 | |
| 1159 | /* residual */ |
| 1160 | part_end = &sub->residual[psize]; |
| 1161 | for (p = 0; p < 1 << porder; p++) { |
| 1162 | int k = sub->rc.params[p]; |
| 1163 | put_bits(&s->pb, sub->rc.coding_mode, k); |
| 1164 | while (res < part_end) |
| 1165 | set_sr_golomb_flac(&s->pb, *res++, k, INT32_MAX, 0); |
| 1166 | part_end = FFMIN(frame_end, part_end + psize); |
| 1167 | } |
| 1168 | } |
| 1169 | } |
| 1170 | } |
| 1171 | |
| 1172 | |
| 1173 | static void write_frame_footer(FlacEncodeContext *s) |
| 1174 | { |
| 1175 | int crc; |
| 1176 | flush_put_bits(&s->pb); |
| 1177 | crc = av_bswap16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, s->pb.buf, |
| 1178 | put_bits_count(&s->pb)>>3)); |
| 1179 | put_bits(&s->pb, 16, crc); |
| 1180 | flush_put_bits(&s->pb); |
| 1181 | } |
| 1182 | |
| 1183 | |
| 1184 | static int write_frame(FlacEncodeContext *s, AVPacket *avpkt) |
| 1185 | { |
| 1186 | init_put_bits(&s->pb, avpkt->data, avpkt->size); |
| 1187 | write_frame_header(s); |
| 1188 | write_subframes(s); |
| 1189 | write_frame_footer(s); |
| 1190 | return put_bits_count(&s->pb) >> 3; |
| 1191 | } |
| 1192 | |
| 1193 | |
| 1194 | static int update_md5_sum(FlacEncodeContext *s, const void *samples) |
| 1195 | { |
| 1196 | const uint8_t *buf; |
| 1197 | int buf_size = s->frame.blocksize * s->channels * |
| 1198 | ((s->avctx->bits_per_raw_sample + 7) / 8); |
| 1199 | |
| 1200 | if (s->avctx->bits_per_raw_sample > 16 || HAVE_BIGENDIAN) { |
| 1201 | av_fast_malloc(&s->md5_buffer, &s->md5_buffer_size, buf_size); |
| 1202 | if (!s->md5_buffer) |
| 1203 | return AVERROR(ENOMEM); |
| 1204 | } |
| 1205 | |
| 1206 | if (s->avctx->bits_per_raw_sample <= 16) { |
| 1207 | buf = (const uint8_t *)samples; |
| 1208 | #if HAVE_BIGENDIAN |
| 1209 | s->bdsp.bswap16_buf((uint16_t *) s->md5_buffer, |
| 1210 | (const uint16_t *) samples, buf_size / 2); |
| 1211 | buf = s->md5_buffer; |
| 1212 | #endif |
| 1213 | } else { |
| 1214 | int i; |
| 1215 | const int32_t *samples0 = samples; |
| 1216 | uint8_t *tmp = s->md5_buffer; |
| 1217 | |
| 1218 | for (i = 0; i < s->frame.blocksize * s->channels; i++) { |
| 1219 | int32_t v = samples0[i] >> 8; |
| 1220 | *tmp++ = (v ) & 0xFF; |
| 1221 | *tmp++ = (v >> 8) & 0xFF; |
| 1222 | *tmp++ = (v >> 16) & 0xFF; |
| 1223 | } |
| 1224 | buf = s->md5_buffer; |
| 1225 | } |
| 1226 | av_md5_update(s->md5ctx, buf, buf_size); |
| 1227 | |
| 1228 | return 0; |
| 1229 | } |
| 1230 | |
| 1231 | |
| 1232 | static int flac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, |
| 1233 | const AVFrame *frame, int *got_packet_ptr) |
| 1234 | { |
| 1235 | FlacEncodeContext *s; |
| 1236 | int frame_bytes, out_bytes, ret; |
| 1237 | |
| 1238 | s = avctx->priv_data; |
| 1239 | |
| 1240 | /* when the last block is reached, update the header in extradata */ |
| 1241 | if (!frame) { |
| 1242 | s->max_framesize = s->max_encoded_framesize; |
| 1243 | av_md5_final(s->md5ctx, s->md5sum); |
| 1244 | write_streaminfo(s, avctx->extradata); |
| 1245 | |
| 1246 | if (avctx->side_data_only_packets && !s->flushed) { |
| 1247 | uint8_t *side_data = av_packet_new_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA, |
| 1248 | avctx->extradata_size); |
| 1249 | if (!side_data) |
| 1250 | return AVERROR(ENOMEM); |
| 1251 | memcpy(side_data, avctx->extradata, avctx->extradata_size); |
| 1252 | |
| 1253 | avpkt->pts = s->next_pts; |
| 1254 | |
| 1255 | *got_packet_ptr = 1; |
| 1256 | s->flushed = 1; |
| 1257 | } |
| 1258 | |
| 1259 | return 0; |
| 1260 | } |
| 1261 | |
| 1262 | /* change max_framesize for small final frame */ |
| 1263 | if (frame->nb_samples < s->frame.blocksize) { |
| 1264 | s->max_framesize = ff_flac_get_max_frame_size(frame->nb_samples, |
| 1265 | s->channels, |
| 1266 | avctx->bits_per_raw_sample); |
| 1267 | } |
| 1268 | |
| 1269 | init_frame(s, frame->nb_samples); |
| 1270 | |
| 1271 | copy_samples(s, frame->data[0]); |
| 1272 | |
| 1273 | channel_decorrelation(s); |
| 1274 | |
| 1275 | remove_wasted_bits(s); |
| 1276 | |
| 1277 | frame_bytes = encode_frame(s); |
| 1278 | |
| 1279 | /* Fall back on verbatim mode if the compressed frame is larger than it |
| 1280 | would be if encoded uncompressed. */ |
| 1281 | if (frame_bytes < 0 || frame_bytes > s->max_framesize) { |
| 1282 | s->frame.verbatim_only = 1; |
| 1283 | frame_bytes = encode_frame(s); |
| 1284 | if (frame_bytes < 0) { |
| 1285 | av_log(avctx, AV_LOG_ERROR, "Bad frame count\n"); |
| 1286 | return frame_bytes; |
| 1287 | } |
| 1288 | } |
| 1289 | |
| 1290 | if ((ret = ff_alloc_packet2(avctx, avpkt, frame_bytes)) < 0) |
| 1291 | return ret; |
| 1292 | |
| 1293 | out_bytes = write_frame(s, avpkt); |
| 1294 | |
| 1295 | s->frame_count++; |
| 1296 | s->sample_count += frame->nb_samples; |
| 1297 | if ((ret = update_md5_sum(s, frame->data[0])) < 0) { |
| 1298 | av_log(avctx, AV_LOG_ERROR, "Error updating MD5 checksum\n"); |
| 1299 | return ret; |
| 1300 | } |
| 1301 | if (out_bytes > s->max_encoded_framesize) |
| 1302 | s->max_encoded_framesize = out_bytes; |
| 1303 | if (out_bytes < s->min_framesize) |
| 1304 | s->min_framesize = out_bytes; |
| 1305 | |
| 1306 | avpkt->pts = frame->pts; |
| 1307 | avpkt->duration = ff_samples_to_time_base(avctx, frame->nb_samples); |
| 1308 | avpkt->size = out_bytes; |
| 1309 | |
| 1310 | s->next_pts = avpkt->pts + avpkt->duration; |
| 1311 | |
| 1312 | *got_packet_ptr = 1; |
| 1313 | return 0; |
| 1314 | } |
| 1315 | |
| 1316 | |
| 1317 | static av_cold int flac_encode_close(AVCodecContext *avctx) |
| 1318 | { |
| 1319 | if (avctx->priv_data) { |
| 1320 | FlacEncodeContext *s = avctx->priv_data; |
| 1321 | av_freep(&s->md5ctx); |
| 1322 | av_freep(&s->md5_buffer); |
| 1323 | ff_lpc_end(&s->lpc_ctx); |
| 1324 | } |
| 1325 | av_freep(&avctx->extradata); |
| 1326 | avctx->extradata_size = 0; |
| 1327 | return 0; |
| 1328 | } |
| 1329 | |
| 1330 | #define FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM |
| 1331 | static const AVOption options[] = { |
| 1332 | { "lpc_coeff_precision", "LPC coefficient precision", offsetof(FlacEncodeContext, options.lpc_coeff_precision), AV_OPT_TYPE_INT, {.i64 = 15 }, 0, MAX_LPC_PRECISION, FLAGS }, |
| 1333 | { "lpc_type", "LPC algorithm", offsetof(FlacEncodeContext, options.lpc_type), AV_OPT_TYPE_INT, {.i64 = FF_LPC_TYPE_DEFAULT }, FF_LPC_TYPE_DEFAULT, FF_LPC_TYPE_NB-1, FLAGS, "lpc_type" }, |
| 1334 | { "none", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_NONE }, INT_MIN, INT_MAX, FLAGS, "lpc_type" }, |
| 1335 | { "fixed", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_FIXED }, INT_MIN, INT_MAX, FLAGS, "lpc_type" }, |
| 1336 | { "levinson", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_LEVINSON }, INT_MIN, INT_MAX, FLAGS, "lpc_type" }, |
| 1337 | { "cholesky", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_LPC_TYPE_CHOLESKY }, INT_MIN, INT_MAX, FLAGS, "lpc_type" }, |
| 1338 | { "lpc_passes", "Number of passes to use for Cholesky factorization during LPC analysis", offsetof(FlacEncodeContext, options.lpc_passes), AV_OPT_TYPE_INT, {.i64 = 2 }, 1, INT_MAX, FLAGS }, |
| 1339 | { "min_partition_order", NULL, offsetof(FlacEncodeContext, options.min_partition_order), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, MAX_PARTITION_ORDER, FLAGS }, |
| 1340 | { "max_partition_order", NULL, offsetof(FlacEncodeContext, options.max_partition_order), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, MAX_PARTITION_ORDER, FLAGS }, |
| 1341 | { "prediction_order_method", "Search method for selecting prediction order", offsetof(FlacEncodeContext, options.prediction_order_method), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, ORDER_METHOD_LOG, FLAGS, "predm" }, |
| 1342 | { "estimation", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_EST }, INT_MIN, INT_MAX, FLAGS, "predm" }, |
| 1343 | { "2level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_2LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" }, |
| 1344 | { "4level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_4LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" }, |
| 1345 | { "8level", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_8LEVEL }, INT_MIN, INT_MAX, FLAGS, "predm" }, |
| 1346 | { "search", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_SEARCH }, INT_MIN, INT_MAX, FLAGS, "predm" }, |
| 1347 | { "log", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = ORDER_METHOD_LOG }, INT_MIN, INT_MAX, FLAGS, "predm" }, |
| 1348 | { "ch_mode", "Stereo decorrelation mode", offsetof(FlacEncodeContext, options.ch_mode), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, FLAC_CHMODE_MID_SIDE, FLAGS, "ch_mode" }, |
| 1349 | { "auto", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = -1 }, INT_MIN, INT_MAX, FLAGS, "ch_mode" }, |
| 1350 | { "indep", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_INDEPENDENT }, INT_MIN, INT_MAX, FLAGS, "ch_mode" }, |
| 1351 | { "left_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_LEFT_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" }, |
| 1352 | { "right_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_RIGHT_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" }, |
| 1353 | { "mid_side", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = FLAC_CHMODE_MID_SIDE }, INT_MIN, INT_MAX, FLAGS, "ch_mode" }, |
| 1354 | { NULL }, |
| 1355 | }; |
| 1356 | |
| 1357 | static const AVClass flac_encoder_class = { |
| 1358 | "FLAC encoder", |
| 1359 | av_default_item_name, |
| 1360 | options, |
| 1361 | LIBAVUTIL_VERSION_INT, |
| 1362 | }; |
| 1363 | |
| 1364 | AVCodec ff_flac_encoder = { |
| 1365 | .name = "flac", |
| 1366 | .long_name = NULL_IF_CONFIG_SMALL("FLAC (Free Lossless Audio Codec)"), |
| 1367 | .type = AVMEDIA_TYPE_AUDIO, |
| 1368 | .id = AV_CODEC_ID_FLAC, |
| 1369 | .priv_data_size = sizeof(FlacEncodeContext), |
| 1370 | .init = flac_encode_init, |
| 1371 | .encode2 = flac_encode_frame, |
| 1372 | .close = flac_encode_close, |
| 1373 | .capabilities = CODEC_CAP_SMALL_LAST_FRAME | CODEC_CAP_DELAY | CODEC_CAP_LOSSLESS, |
| 1374 | .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16, |
| 1375 | AV_SAMPLE_FMT_S32, |
| 1376 | AV_SAMPLE_FMT_NONE }, |
| 1377 | .priv_class = &flac_encoder_class, |
| 1378 | }; |