flacenc.c 43.9 KB
Newer Older
1 2 3 4
/**
 * FLAC audio encoder
 * Copyright (c) 2006  Justin Ruggles <jruggle@earthlink.net>
 *
5 6 7
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
8 9
 * modify it under the terms of the GNU Lesser General Public
 * 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 14 15 16 17
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * 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 20 21 22 23 24
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "avcodec.h"
#include "bitstream.h"
#include "crc.h"
25
#include "dsputil.h"
26
#include "golomb.h"
27
#include "lls.h"
28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43

#define FLAC_MAX_CH  8
#define FLAC_MIN_BLOCKSIZE  16
#define FLAC_MAX_BLOCKSIZE  65535

#define FLAC_SUBFRAME_CONSTANT  0
#define FLAC_SUBFRAME_VERBATIM  1
#define FLAC_SUBFRAME_FIXED     8
#define FLAC_SUBFRAME_LPC      32

#define FLAC_CHMODE_NOT_STEREO      0
#define FLAC_CHMODE_LEFT_RIGHT      1
#define FLAC_CHMODE_LEFT_SIDE       8
#define FLAC_CHMODE_RIGHT_SIDE      9
#define FLAC_CHMODE_MID_SIDE       10

44 45 46 47 48
#define ORDER_METHOD_EST     0
#define ORDER_METHOD_2LEVEL  1
#define ORDER_METHOD_4LEVEL  2
#define ORDER_METHOD_8LEVEL  3
#define ORDER_METHOD_SEARCH  4
49
#define ORDER_METHOD_LOG     5
50

51 52
#define FLAC_STREAMINFO_SIZE  34

53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73
#define MIN_LPC_ORDER       1
#define MAX_LPC_ORDER      32
#define MAX_FIXED_ORDER     4
#define MAX_PARTITION_ORDER 8
#define MAX_PARTITIONS     (1 << MAX_PARTITION_ORDER)
#define MAX_LPC_PRECISION  15
#define MAX_LPC_SHIFT      15
#define MAX_RICE_PARAM     14

typedef struct CompressionOptions {
    int compression_level;
    int block_time_ms;
    int use_lpc;
    int lpc_coeff_precision;
    int min_prediction_order;
    int max_prediction_order;
    int prediction_order_method;
    int min_partition_order;
    int max_partition_order;
} CompressionOptions;

74 75
typedef struct RiceContext {
    int porder;
76
    int params[MAX_PARTITIONS];
77 78
} RiceContext;

79 80 81 82 83
typedef struct FlacSubframe {
    int type;
    int type_code;
    int obits;
    int order;
84 85
    int32_t coefs[MAX_LPC_ORDER];
    int shift;
86
    RiceContext rc;
87
    int32_t samples[FLAC_MAX_BLOCKSIZE];
88
    int32_t residual[FLAC_MAX_BLOCKSIZE+1];
89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108
} FlacSubframe;

typedef struct FlacFrame {
    FlacSubframe subframes[FLAC_MAX_CH];
    int blocksize;
    int bs_code[2];
    uint8_t crc8;
    int ch_mode;
} FlacFrame;

typedef struct FlacEncodeContext {
    PutBitContext pb;
    int channels;
    int ch_code;
    int samplerate;
    int sr_code[2];
    int blocksize;
    int max_framesize;
    uint32_t frame_count;
    FlacFrame frame;
109
    CompressionOptions options;
110
    AVCodecContext *avctx;
111
    DSPContext dsp;
112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154
} FlacEncodeContext;

static const int flac_samplerates[16] = {
    0, 0, 0, 0,
    8000, 16000, 22050, 24000, 32000, 44100, 48000, 96000,
    0, 0, 0, 0
};

static const int flac_blocksizes[16] = {
    0,
    192,
    576, 1152, 2304, 4608,
    0, 0,
    256, 512, 1024, 2048, 4096, 8192, 16384, 32768
};

/**
 * Writes streaminfo metadata block to byte array
 */
static void write_streaminfo(FlacEncodeContext *s, uint8_t *header)
{
    PutBitContext pb;

    memset(header, 0, FLAC_STREAMINFO_SIZE);
    init_put_bits(&pb, header, FLAC_STREAMINFO_SIZE);

    /* streaminfo metadata block */
    put_bits(&pb, 16, s->blocksize);
    put_bits(&pb, 16, s->blocksize);
    put_bits(&pb, 24, 0);
    put_bits(&pb, 24, s->max_framesize);
    put_bits(&pb, 20, s->samplerate);
    put_bits(&pb, 3, s->channels-1);
    put_bits(&pb, 5, 15);       /* bits per sample - 1 */
    flush_put_bits(&pb);
    /* total samples = 0 */
    /* MD5 signature = 0 */
}

/**
 * Sets blocksize based on samplerate
 * Chooses the closest predefined blocksize >= BLOCK_TIME_MS milliseconds
 */
155
static int select_blocksize(int samplerate, int block_time_ms)
156 157 158 159 160 161
{
    int i;
    int target;
    int blocksize;

    assert(samplerate > 0);
162
    blocksize = flac_blocksizes[1];
163
    target = (samplerate * block_time_ms) / 1000;
164 165 166
    for(i=0; i<16; i++) {
        if(target >= flac_blocksizes[i] && flac_blocksizes[i] > blocksize) {
            blocksize = flac_blocksizes[i];
167 168 169 170 171 172 173 174 175 176
        }
    }
    return blocksize;
}

static int flac_encode_init(AVCodecContext *avctx)
{
    int freq = avctx->sample_rate;
    int channels = avctx->channels;
    FlacEncodeContext *s = avctx->priv_data;
Michael Niedermayer's avatar
Michael Niedermayer committed
177
    int i, level;
178 179
    uint8_t *streaminfo;

180 181
    s->avctx = avctx;

182 183
    dsputil_init(&s->dsp, avctx);

184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221
    if(avctx->sample_fmt != SAMPLE_FMT_S16) {
        return -1;
    }

    if(channels < 1 || channels > FLAC_MAX_CH) {
        return -1;
    }
    s->channels = channels;
    s->ch_code = s->channels-1;

    /* find samplerate in table */
    if(freq < 1)
        return -1;
    for(i=4; i<12; i++) {
        if(freq == flac_samplerates[i]) {
            s->samplerate = flac_samplerates[i];
            s->sr_code[0] = i;
            s->sr_code[1] = 0;
            break;
        }
    }
    /* if not in table, samplerate is non-standard */
    if(i == 12) {
        if(freq % 1000 == 0 && freq < 255000) {
            s->sr_code[0] = 12;
            s->sr_code[1] = freq / 1000;
        } else if(freq % 10 == 0 && freq < 655350) {
            s->sr_code[0] = 14;
            s->sr_code[1] = freq / 10;
        } else if(freq < 65535) {
            s->sr_code[0] = 13;
            s->sr_code[1] = freq;
        } else {
            return -1;
        }
        s->samplerate = freq;
    }

222 223 224 225 226 227 228 229
    /* set compression option defaults based on avctx->compression_level */
    if(avctx->compression_level < 0) {
        s->options.compression_level = 5;
    } else {
        s->options.compression_level = avctx->compression_level;
    }
    av_log(avctx, AV_LOG_DEBUG, " compression: %d\n", s->options.compression_level);

Michael Niedermayer's avatar
Michael Niedermayer committed
230
    level= s->options.compression_level;
231
    if(level > 12) {
232 233 234 235 236
        av_log(avctx, AV_LOG_ERROR, "invalid compression level: %d\n",
               s->options.compression_level);
        return -1;
    }

237 238 239 240 241 242
    s->options.block_time_ms       = ((int[]){ 27, 27, 27,105,105,105,105,105,105,105,105,105,105})[level];
    s->options.use_lpc             = ((int[]){  0,  0,  0,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1})[level];
    s->options.min_prediction_order= ((int[]){  2,  0,  0,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1})[level];
    s->options.max_prediction_order= ((int[]){  3,  4,  4,  6,  8,  8,  8,  8, 12, 12, 12, 32, 32})[level];
    s->options.prediction_order_method = ((int[]){ ORDER_METHOD_EST,    ORDER_METHOD_EST,    ORDER_METHOD_EST,
                                                   ORDER_METHOD_EST,    ORDER_METHOD_EST,    ORDER_METHOD_EST,
243 244
                                                   ORDER_METHOD_4LEVEL, ORDER_METHOD_LOG,    ORDER_METHOD_4LEVEL,
                                                   ORDER_METHOD_LOG,    ORDER_METHOD_SEARCH, ORDER_METHOD_LOG,
245
                                                   ORDER_METHOD_SEARCH})[level];
246 247
    s->options.min_partition_order = ((int[]){  2,  2,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0})[level];
    s->options.max_partition_order = ((int[]){  2,  2,  3,  3,  3,  8,  8,  8,  8,  8,  8,  8,  8})[level];
Michael Niedermayer's avatar
Michael Niedermayer committed
248

249 250
    /* set compression option overrides from AVCodecContext */
    if(avctx->use_lpc >= 0) {
251
        s->options.use_lpc = av_clip(avctx->use_lpc, 0, 11);
252
    }
253 254 255 256
    if(s->options.use_lpc == 1)
        av_log(avctx, AV_LOG_DEBUG, " use lpc: Levinson-Durbin recursion with Welch window\n");
    else if(s->options.use_lpc > 1)
        av_log(avctx, AV_LOG_DEBUG, " use lpc: Cholesky factorization\n");
257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300

    if(avctx->min_prediction_order >= 0) {
        if(s->options.use_lpc) {
            if(avctx->min_prediction_order < MIN_LPC_ORDER ||
                    avctx->min_prediction_order > MAX_LPC_ORDER) {
                av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
                       avctx->min_prediction_order);
                return -1;
            }
        } else {
            if(avctx->min_prediction_order > MAX_FIXED_ORDER) {
                av_log(avctx, AV_LOG_ERROR, "invalid min prediction order: %d\n",
                       avctx->min_prediction_order);
                return -1;
            }
        }
        s->options.min_prediction_order = avctx->min_prediction_order;
    }
    if(avctx->max_prediction_order >= 0) {
        if(s->options.use_lpc) {
            if(avctx->max_prediction_order < MIN_LPC_ORDER ||
                    avctx->max_prediction_order > MAX_LPC_ORDER) {
                av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
                       avctx->max_prediction_order);
                return -1;
            }
        } else {
            if(avctx->max_prediction_order > MAX_FIXED_ORDER) {
                av_log(avctx, AV_LOG_ERROR, "invalid max prediction order: %d\n",
                       avctx->max_prediction_order);
                return -1;
            }
        }
        s->options.max_prediction_order = avctx->max_prediction_order;
    }
    if(s->options.max_prediction_order < s->options.min_prediction_order) {
        av_log(avctx, AV_LOG_ERROR, "invalid prediction orders: min=%d max=%d\n",
               s->options.min_prediction_order, s->options.max_prediction_order);
        return -1;
    }
    av_log(avctx, AV_LOG_DEBUG, " prediction order: %d, %d\n",
           s->options.min_prediction_order, s->options.max_prediction_order);

    if(avctx->prediction_order_method >= 0) {
301
        if(avctx->prediction_order_method > ORDER_METHOD_LOG) {
302 303 304 305 306 307
            av_log(avctx, AV_LOG_ERROR, "invalid prediction order method: %d\n",
                   avctx->prediction_order_method);
            return -1;
        }
        s->options.prediction_order_method = avctx->prediction_order_method;
    }
308
    switch(s->options.prediction_order_method) {
309 310 311 312 313 314 315 316 317 318
        case ORDER_METHOD_EST:    av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
                                         "estimate"); break;
        case ORDER_METHOD_2LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
                                         "2-level"); break;
        case ORDER_METHOD_4LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
                                         "4-level"); break;
        case ORDER_METHOD_8LEVEL: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
                                         "8-level"); break;
        case ORDER_METHOD_SEARCH: av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
                                         "full search"); break;
319 320
        case ORDER_METHOD_LOG:    av_log(avctx, AV_LOG_DEBUG, " order method: %s\n",
                                         "log search"); break;
321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348
    }

    if(avctx->min_partition_order >= 0) {
        if(avctx->min_partition_order > MAX_PARTITION_ORDER) {
            av_log(avctx, AV_LOG_ERROR, "invalid min partition order: %d\n",
                   avctx->min_partition_order);
            return -1;
        }
        s->options.min_partition_order = avctx->min_partition_order;
    }
    if(avctx->max_partition_order >= 0) {
        if(avctx->max_partition_order > MAX_PARTITION_ORDER) {
            av_log(avctx, AV_LOG_ERROR, "invalid max partition order: %d\n",
                   avctx->max_partition_order);
            return -1;
        }
        s->options.max_partition_order = avctx->max_partition_order;
    }
    if(s->options.max_partition_order < s->options.min_partition_order) {
        av_log(avctx, AV_LOG_ERROR, "invalid partition orders: min=%d max=%d\n",
               s->options.min_partition_order, s->options.max_partition_order);
        return -1;
    }
    av_log(avctx, AV_LOG_DEBUG, " partition order: %d, %d\n",
           s->options.min_partition_order, s->options.max_partition_order);

    if(avctx->frame_size > 0) {
        if(avctx->frame_size < FLAC_MIN_BLOCKSIZE ||
349
                avctx->frame_size > FLAC_MAX_BLOCKSIZE) {
350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382
            av_log(avctx, AV_LOG_ERROR, "invalid block size: %d\n",
                   avctx->frame_size);
            return -1;
        }
        s->blocksize = avctx->frame_size;
    } else {
        s->blocksize = select_blocksize(s->samplerate, s->options.block_time_ms);
        avctx->frame_size = s->blocksize;
    }
    av_log(avctx, AV_LOG_DEBUG, " block size: %d\n", s->blocksize);

    /* set LPC precision */
    if(avctx->lpc_coeff_precision > 0) {
        if(avctx->lpc_coeff_precision > MAX_LPC_PRECISION) {
            av_log(avctx, AV_LOG_ERROR, "invalid lpc coeff precision: %d\n",
                   avctx->lpc_coeff_precision);
            return -1;
        }
        s->options.lpc_coeff_precision = avctx->lpc_coeff_precision;
    } else {
        /* select LPC precision based on block size */
        if(     s->blocksize <=   192) s->options.lpc_coeff_precision =  7;
        else if(s->blocksize <=   384) s->options.lpc_coeff_precision =  8;
        else if(s->blocksize <=   576) s->options.lpc_coeff_precision =  9;
        else if(s->blocksize <=  1152) s->options.lpc_coeff_precision = 10;
        else if(s->blocksize <=  2304) s->options.lpc_coeff_precision = 11;
        else if(s->blocksize <=  4608) s->options.lpc_coeff_precision = 12;
        else if(s->blocksize <=  8192) s->options.lpc_coeff_precision = 13;
        else if(s->blocksize <= 16384) s->options.lpc_coeff_precision = 14;
        else                           s->options.lpc_coeff_precision = 15;
    }
    av_log(avctx, AV_LOG_DEBUG, " lpc precision: %d\n",
           s->options.lpc_coeff_precision);
383

384 385 386 387 388 389
    /* set maximum encoded frame size in verbatim mode */
    if(s->channels == 2) {
        s->max_framesize = 14 + ((s->blocksize * 33 + 7) >> 3);
    } else {
        s->max_framesize = 14 + (s->blocksize * s->channels * 2);
    }
390 391 392 393 394 395 396 397 398 399 400 401 402 403

    streaminfo = av_malloc(FLAC_STREAMINFO_SIZE);
    write_streaminfo(s, streaminfo);
    avctx->extradata = streaminfo;
    avctx->extradata_size = FLAC_STREAMINFO_SIZE;

    s->frame_count = 0;

    avctx->coded_frame = avcodec_alloc_frame();
    avctx->coded_frame->key_frame = 1;

    return 0;
}

404
static void init_frame(FlacEncodeContext *s)
405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450
{
    int i, ch;
    FlacFrame *frame;

    frame = &s->frame;

    for(i=0; i<16; i++) {
        if(s->blocksize == flac_blocksizes[i]) {
            frame->blocksize = flac_blocksizes[i];
            frame->bs_code[0] = i;
            frame->bs_code[1] = 0;
            break;
        }
    }
    if(i == 16) {
        frame->blocksize = s->blocksize;
        if(frame->blocksize <= 256) {
            frame->bs_code[0] = 6;
            frame->bs_code[1] = frame->blocksize-1;
        } else {
            frame->bs_code[0] = 7;
            frame->bs_code[1] = frame->blocksize-1;
        }
    }

    for(ch=0; ch<s->channels; ch++) {
        frame->subframes[ch].obits = 16;
    }
}

/**
 * Copy channel-interleaved input samples into separate subframes
 */
static void copy_samples(FlacEncodeContext *s, int16_t *samples)
{
    int i, j, ch;
    FlacFrame *frame;

    frame = &s->frame;
    for(i=0,j=0; i<frame->blocksize; i++) {
        for(ch=0; ch<s->channels; ch++,j++) {
            frame->subframes[ch].samples[i] = samples[j];
        }
    }
}

451 452 453

#define rice_encode_count(sum, n, k) (((n)*((k)+1))+((sum-(n>>1))>>(k)))

454 455 456
/**
 * Solve for d/dk(rice_encode_count) = n-((sum-(n>>1))>>(k+1)) = 0
 */
457 458
static int find_optimal_param(uint32_t sum, int n)
{
459 460 461 462 463 464 465 466
    int k;
    uint32_t sum2;

    if(sum <= n>>1)
        return 0;
    sum2 = sum-(n>>1);
    k = av_log2(n<256 ? FASTDIV(sum2,n) : sum2/n);
    return FFMIN(k, MAX_RICE_PARAM);
467 468 469 470 471 472 473 474 475 476
}

static uint32_t calc_optimal_rice_params(RiceContext *rc, int porder,
                                         uint32_t *sums, int n, int pred_order)
{
    int i;
    int k, cnt, part;
    uint32_t all_bits;

    part = (1 << porder);
477
    all_bits = 4 * part;
478 479 480 481 482 483

    cnt = (n >> porder) - pred_order;
    for(i=0; i<part; i++) {
        k = find_optimal_param(sums[i], cnt);
        rc->params[i] = k;
        all_bits += rice_encode_count(sums[i], cnt, k);
484
        cnt = n >> porder;
485 486 487 488 489 490 491
    }

    rc->porder = porder;

    return all_bits;
}

492 493
static void calc_sums(int pmin, int pmax, uint32_t *data, int n, int pred_order,
                      uint32_t sums[][MAX_PARTITIONS])
494 495
{
    int i, j;
Michael Niedermayer's avatar
Michael Niedermayer committed
496 497
    int parts;
    uint32_t *res, *res_end;
498 499 500 501

    /* sums for highest level */
    parts = (1 << pmax);
    res = &data[pred_order];
Michael Niedermayer's avatar
Michael Niedermayer committed
502
    res_end = &data[n >> pmax];
503
    for(i=0; i<parts; i++) {
504
        uint32_t sum = 0;
Michael Niedermayer's avatar
Michael Niedermayer committed
505
        while(res < res_end){
506
            sum += *(res++);
507
        }
508
        sums[pmax][i] = sum;
Michael Niedermayer's avatar
Michael Niedermayer committed
509
        res_end+= n >> pmax;
510 511
    }
    /* sums for lower levels */
512
    for(i=pmax-1; i>=pmin; i--) {
513 514 515 516 517 518 519
        parts = (1 << i);
        for(j=0; j<parts; j++) {
            sums[i][j] = sums[i+1][2*j] + sums[i+1][2*j+1];
        }
    }
}

520 521
static uint32_t calc_rice_params(RiceContext *rc, int pmin, int pmax,
                                 int32_t *data, int n, int pred_order)
522 523
{
    int i;
524
    uint32_t bits[MAX_PARTITION_ORDER+1];
525
    int opt_porder;
526
    RiceContext tmp_rc;
527
    uint32_t *udata;
528
    uint32_t sums[MAX_PARTITION_ORDER+1][MAX_PARTITIONS];
529

530 531 532
    assert(pmin >= 0 && pmin <= MAX_PARTITION_ORDER);
    assert(pmax >= 0 && pmax <= MAX_PARTITION_ORDER);
    assert(pmin <= pmax);
533 534 535 536 537 538

    udata = av_malloc(n * sizeof(uint32_t));
    for(i=0; i<n; i++) {
        udata[i] = (2*data[i]) ^ (data[i]>>31);
    }

539
    calc_sums(pmin, pmax, udata, n, pred_order, sums);
540

541 542 543 544 545
    opt_porder = pmin;
    bits[pmin] = UINT32_MAX;
    for(i=pmin; i<=pmax; i++) {
        bits[i] = calc_optimal_rice_params(&tmp_rc, i, sums[i], n, pred_order);
        if(bits[i] <= bits[opt_porder]) {
546
            opt_porder = i;
547
            *rc= tmp_rc;
548 549 550 551
        }
    }

    av_freep(&udata);
552
    return bits[opt_porder];
553 554
}

555 556 557 558 559 560 561 562
static int get_max_p_order(int max_porder, int n, int order)
{
    int porder = FFMIN(max_porder, av_log2(n^(n-1)));
    if(order > 0)
        porder = FFMIN(porder, av_log2(n/order));
    return porder;
}

563 564 565
static uint32_t calc_rice_params_fixed(RiceContext *rc, int pmin, int pmax,
                                       int32_t *data, int n, int pred_order,
                                       int bps)
566 567
{
    uint32_t bits;
568 569
    pmin = get_max_p_order(pmin, n, pred_order);
    pmax = get_max_p_order(pmax, n, pred_order);
570
    bits = pred_order*bps + 6;
571 572 573 574 575 576 577 578 579
    bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
    return bits;
}

static uint32_t calc_rice_params_lpc(RiceContext *rc, int pmin, int pmax,
                                     int32_t *data, int n, int pred_order,
                                     int bps, int precision)
{
    uint32_t bits;
580 581
    pmin = get_max_p_order(pmin, n, pred_order);
    pmax = get_max_p_order(pmax, n, pred_order);
582 583
    bits = pred_order*bps + 4 + 5 + pred_order*precision + 6;
    bits += calc_rice_params(rc, pmin, pmax, data, n, pred_order);
584 585 586
    return bits;
}

587 588 589 590 591 592 593 594 595
/**
 * Apply Welch window function to audio block
 */
static void apply_welch_window(const int32_t *data, int len, double *w_data)
{
    int i, n2;
    double w;
    double c;

596 597 598
    assert(!(len&1)); //the optimization in r11881 does not support odd len
                      //if someone wants odd len extend the change in r11881

599 600
    n2 = (len >> 1);
    c = 2.0 / (len - 1.0);
601 602 603

    w_data+=n2;
      data+=n2;
604
    for(i=0; i<n2; i++) {
605
        w = c - n2 + i;
606
        w = 1.0 - (w * w);
607 608
        w_data[-i-1] = data[-i-1] * w;
        w_data[+i  ] = data[+i  ] * w;
609 610 611 612 613 614 615
    }
}

/**
 * Calculates autocorrelation data from audio samples
 * A Welch window function is applied before calculation.
 */
616 617
void ff_flac_compute_autocorr(const int32_t *data, int len, int lag,
                              double *autoc)
618
{
619
    int i, j;
620
    double tmp[len + lag + 1];
621
    double *data1= tmp + lag;
622 623 624

    apply_welch_window(data, len, data1);

625 626
    for(j=0; j<lag; j++)
        data1[j-lag]= 0.0;
627
    data1[len] = 0.0;
628

629 630 631 632 633
    for(j=0; j<lag; j+=2){
        double sum0 = 1.0, sum1 = 1.0;
        for(i=0; i<len; i++){
            sum0 += data1[i] * data1[i-j];
            sum1 += data1[i] * data1[i-j-1];
634
        }
635 636 637 638 639 640
        autoc[j  ] = sum0;
        autoc[j+1] = sum1;
    }

    if(j==lag){
        double sum = 1.0;
641 642 643 644
        for(i=0; i<len; i+=2){
            sum += data1[i  ] * data1[i-j  ]
                 + data1[i+1] * data1[i-j+1];
        }
645
        autoc[j] = sum;
646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696
    }
}

/**
 * Levinson-Durbin recursion.
 * Produces LPC coefficients from autocorrelation data.
 */
static void compute_lpc_coefs(const double *autoc, int max_order,
                              double lpc[][MAX_LPC_ORDER], double *ref)
{
   int i, j, i2;
   double r, err, tmp;
   double lpc_tmp[MAX_LPC_ORDER];

   for(i=0; i<max_order; i++) lpc_tmp[i] = 0;
   err = autoc[0];

   for(i=0; i<max_order; i++) {
      r = -autoc[i+1];
      for(j=0; j<i; j++) {
          r -= lpc_tmp[j] * autoc[i-j];
      }
      r /= err;
      ref[i] = fabs(r);

      err *= 1.0 - (r * r);

      i2 = (i >> 1);
      lpc_tmp[i] = r;
      for(j=0; j<i2; j++) {
         tmp = lpc_tmp[j];
         lpc_tmp[j] += r * lpc_tmp[i-1-j];
         lpc_tmp[i-1-j] += r * tmp;
      }
      if(i & 1) {
          lpc_tmp[j] += lpc_tmp[j] * r;
      }

      for(j=0; j<=i; j++) {
          lpc[i][j] = -lpc_tmp[j];
      }
   }
}

/**
 * Quantize LPC coefficients
 */
static void quantize_lpc_coefs(double *lpc_in, int order, int precision,
                               int32_t *lpc_out, int *shift)
{
    int i;
Michael Niedermayer's avatar
Michael Niedermayer committed
697
    double cmax, error;
698 699 700 701 702 703 704 705 706
    int32_t qmax;
    int sh;

    /* define maximum levels */
    qmax = (1 << (precision - 1)) - 1;

    /* find maximum coefficient value */
    cmax = 0.0;
    for(i=0; i<order; i++) {
Michael Niedermayer's avatar
Michael Niedermayer committed
707
        cmax= FFMAX(cmax, fabs(lpc_in[i]));
708 709 710 711 712
    }

    /* if maximum value quantizes to zero, return all zeros */
    if(cmax * (1 << MAX_LPC_SHIFT) < 1.0) {
        *shift = 0;
Michael Niedermayer's avatar
Michael Niedermayer committed
713
        memset(lpc_out, 0, sizeof(int32_t) * order);
714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732
        return;
    }

    /* calculate level shift which scales max coeff to available bits */
    sh = MAX_LPC_SHIFT;
    while((cmax * (1 << sh) > qmax) && (sh > 0)) {
        sh--;
    }

    /* since negative shift values are unsupported in decoder, scale down
       coefficients instead */
    if(sh == 0 && cmax > qmax) {
        double scale = ((double)qmax) / cmax;
        for(i=0; i<order; i++) {
            lpc_in[i] *= scale;
        }
    }

    /* output quantized coefficients and level shift */
Michael Niedermayer's avatar
Michael Niedermayer committed
733
    error=0;
734
    for(i=0; i<order; i++) {
Michael Niedermayer's avatar
Michael Niedermayer committed
735
        error += lpc_in[i] * (1 << sh);
736
        lpc_out[i] = av_clip(lrintf(error), -qmax, qmax);
Michael Niedermayer's avatar
Michael Niedermayer committed
737
        error -= lpc_out[i];
738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758
    }
    *shift = sh;
}

static int estimate_best_order(double *ref, int max_order)
{
    int i, est;

    est = 1;
    for(i=max_order-1; i>=0; i--) {
        if(ref[i] > 0.10) {
            est = i+1;
            break;
        }
    }
    return est;
}

/**
 * Calculate LPC coefficients for multiple orders
 */
759 760
static int lpc_calc_coefs(FlacEncodeContext *s,
                          const int32_t *samples, int blocksize, int max_order,
761
                          int precision, int32_t coefs[][MAX_LPC_ORDER],
762
                          int *shift, int use_lpc, int omethod)
763 764 765 766
{
    double autoc[MAX_LPC_ORDER+1];
    double ref[MAX_LPC_ORDER];
    double lpc[MAX_LPC_ORDER][MAX_LPC_ORDER];
767
    int i, j, pass;
768 769 770 771
    int opt_order;

    assert(max_order >= MIN_LPC_ORDER && max_order <= MAX_LPC_ORDER);

772
    if(use_lpc == 1){
773
        s->dsp.flac_compute_autocorr(samples, blocksize, max_order, autoc);
774 775 776 777

        compute_lpc_coefs(autoc, max_order, lpc, ref);
    }else{
        LLSModel m[2];
Loren Merritt's avatar
Loren Merritt committed
778
        double var[MAX_LPC_ORDER+1], weight;
779 780

        for(pass=0; pass<use_lpc-1; pass++){
Michael Niedermayer's avatar
Michael Niedermayer committed
781
            av_init_lls(&m[pass&1], max_order);
782

783
            weight=0;
784 785 786
            for(i=max_order; i<blocksize; i++){
                for(j=0; j<=max_order; j++)
                    var[j]= samples[i-j];
787

788
                if(pass){
Loren Merritt's avatar
Loren Merritt committed
789
                    double eval, inv, rinv;
790
                    eval= av_evaluate_lls(&m[(pass-1)&1], var+1, max_order-1);
791
                    eval= (512>>pass) + fabs(eval - var[0]);
Loren Merritt's avatar
Loren Merritt committed
792 793
                    inv = 1/eval;
                    rinv = sqrt(inv);
794
                    for(j=0; j<=max_order; j++)
Loren Merritt's avatar
Loren Merritt committed
795 796
                        var[j] *= rinv;
                    weight += inv;
797 798
                }else
                    weight++;
799 800 801

                av_update_lls(&m[pass&1], var, 1.0);
            }
802
            av_solve_lls(&m[pass&1], 0.001, 0);
803 804
        }

805 806 807 808 809 810 811
        for(i=0; i<max_order; i++){
            for(j=0; j<max_order; j++)
                lpc[i][j]= m[(pass-1)&1].coeff[i][j];
            ref[i]= sqrt(m[(pass-1)&1].variance[i] / weight) * (blocksize - max_order) / 4000;
        }
        for(i=max_order-1; i>0; i--)
            ref[i] = ref[i-1] - ref[i];
812
    }
813
    opt_order = max_order;
814

815 816 817 818 819 820 821 822 823
    if(omethod == ORDER_METHOD_EST) {
        opt_order = estimate_best_order(ref, max_order);
        i = opt_order-1;
        quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i]);
    } else {
        for(i=0; i<max_order; i++) {
            quantize_lpc_coefs(lpc[i], i+1, precision, coefs[i], &shift[i]);
        }
    }
824 825 826 827 828

    return opt_order;
}


829 830 831 832 833 834
static void encode_residual_verbatim(int32_t *res, int32_t *smp, int n)
{
    assert(n > 0);
    memcpy(res, smp, n * sizeof(int32_t));
}

835 836
static void encode_residual_fixed(int32_t *res, const int32_t *smp, int n,
                                  int order)
837 838 839 840 841 842 843 844 845 846 847 848 849 850
{
    int i;

    for(i=0; i<order; i++) {
        res[i] = smp[i];
    }

    if(order==0){
        for(i=order; i<n; i++)
            res[i]= smp[i];
    }else if(order==1){
        for(i=order; i<n; i++)
            res[i]= smp[i] - smp[i-1];
    }else if(order==2){
851
        int a = smp[order-1] - smp[order-2];
852
        for(i=order; i<n; i+=2) {
853 854
            int b = smp[i] - smp[i-1];
            res[i]= b - a;
855 856
            a = smp[i+1] - smp[i];
            res[i+1]= a - b;
857
        }
858
    }else if(order==3){
859 860
        int a = smp[order-1] - smp[order-2];
        int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
861
        for(i=order; i<n; i+=2) {
862 863 864
            int b = smp[i] - smp[i-1];
            int d = b - a;
            res[i]= d - c;
865 866 867
            a = smp[i+1] - smp[i];
            c = a - b;
            res[i+1]= c - d;
868
        }
869
    }else{
870 871 872
        int a = smp[order-1] - smp[order-2];
        int c = smp[order-1] - 2*smp[order-2] + smp[order-3];
        int e = smp[order-1] - 3*smp[order-2] + 3*smp[order-3] - smp[order-4];
873
        for(i=order; i<n; i+=2) {
874 875 876 877
            int b = smp[i] - smp[i-1];
            int d = b - a;
            int f = d - c;
            res[i]= f - e;
878 879 880 881
            a = smp[i+1] - smp[i];
            c = a - b;
            e = c - d;
            res[i+1]= e - f;
882
        }
883 884 885
    }
}

886
#define LPC1(x) {\
Loren Merritt's avatar
Loren Merritt committed
887
    int c = coefs[(x)-1];\
888
    p0 += c*s;\
Loren Merritt's avatar
Loren Merritt committed
889 890
    s = smp[i-(x)+1];\
    p1 += c*s;\
891 892 893 894 895 896 897 898
}

static av_always_inline void encode_residual_lpc_unrolled(
    int32_t *res, const int32_t *smp, int n,
    int order, const int32_t *coefs, int shift, int big)
{
    int i;
    for(i=order; i<n; i+=2) {
Loren Merritt's avatar
Loren Merritt committed
899 900
        int s = smp[i-order];
        int p0 = 0, p1 = 0;
901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933
        if(big) {
            switch(order) {
                case 32: LPC1(32)
                case 31: LPC1(31)
                case 30: LPC1(30)
                case 29: LPC1(29)
                case 28: LPC1(28)
                case 27: LPC1(27)
                case 26: LPC1(26)
                case 25: LPC1(25)
                case 24: LPC1(24)
                case 23: LPC1(23)
                case 22: LPC1(22)
                case 21: LPC1(21)
                case 20: LPC1(20)
                case 19: LPC1(19)
                case 18: LPC1(18)
                case 17: LPC1(17)
                case 16: LPC1(16)
                case 15: LPC1(15)
                case 14: LPC1(14)
                case 13: LPC1(13)
                case 12: LPC1(12)
                case 11: LPC1(11)
                case 10: LPC1(10)
                case  9: LPC1( 9)
                         LPC1( 8)
                         LPC1( 7)
                         LPC1( 6)
                         LPC1( 5)
                         LPC1( 4)
                         LPC1( 3)
                         LPC1( 2)
Loren Merritt's avatar
Loren Merritt committed
934
                         LPC1( 1)
935 936 937 938 939 940 941 942 943 944
            }
        } else {
            switch(order) {
                case  8: LPC1( 8)
                case  7: LPC1( 7)
                case  6: LPC1( 6)
                case  5: LPC1( 5)
                case  4: LPC1( 4)
                case  3: LPC1( 3)
                case  2: LPC1( 2)
Loren Merritt's avatar
Loren Merritt committed
945
                case  1: LPC1( 1)
946 947 948 949 950 951 952
            }
        }
        res[i  ] = smp[i  ] - (p0 >> shift);
        res[i+1] = smp[i+1] - (p1 >> shift);
    }
}

953 954 955
static void encode_residual_lpc(int32_t *res, const int32_t *smp, int n,
                                int order, const int32_t *coefs, int shift)
{
956
    int i;
957 958 959
    for(i=0; i<order; i++) {
        res[i] = smp[i];
    }
960
#ifdef CONFIG_SMALL
961
    for(i=order; i<n; i+=2) {
962
        int j;
Loren Merritt's avatar
Loren Merritt committed
963 964 965 966
        int s = smp[i];
        int p0 = 0, p1 = 0;
        for(j=0; j<order; j++) {
            int c = coefs[j];
967
            p1 += c*s;
Loren Merritt's avatar
Loren Merritt committed
968 969
            s = smp[i-j-1];
            p0 += c*s;
970
        }
Loren Merritt's avatar
Loren Merritt committed
971
        res[i  ] = smp[i  ] - (p0 >> shift);
972
        res[i+1] = smp[i+1] - (p1 >> shift);
973
    }
974 975 976 977 978 979 980 981 982 983 984 985 986
#else
    switch(order) {
        case  1: encode_residual_lpc_unrolled(res, smp, n, 1, coefs, shift, 0); break;
        case  2: encode_residual_lpc_unrolled(res, smp, n, 2, coefs, shift, 0); break;
        case  3: encode_residual_lpc_unrolled(res, smp, n, 3, coefs, shift, 0); break;
        case  4: encode_residual_lpc_unrolled(res, smp, n, 4, coefs, shift, 0); break;
        case  5: encode_residual_lpc_unrolled(res, smp, n, 5, coefs, shift, 0); break;
        case  6: encode_residual_lpc_unrolled(res, smp, n, 6, coefs, shift, 0); break;
        case  7: encode_residual_lpc_unrolled(res, smp, n, 7, coefs, shift, 0); break;
        case  8: encode_residual_lpc_unrolled(res, smp, n, 8, coefs, shift, 0); break;
        default: encode_residual_lpc_unrolled(res, smp, n, order, coefs, shift, 1); break;
    }
#endif
987 988
}

989 990
static int encode_residual(FlacEncodeContext *ctx, int ch)
{
991
    int i, n;
992
    int min_order, max_order, opt_order, precision, omethod;
993
    int min_porder, max_porder;
994 995
    FlacFrame *frame;
    FlacSubframe *sub;
996 997
    int32_t coefs[MAX_LPC_ORDER][MAX_LPC_ORDER];
    int shift[MAX_LPC_ORDER];
998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022
    int32_t *res, *smp;

    frame = &ctx->frame;
    sub = &frame->subframes[ch];
    res = sub->residual;
    smp = sub->samples;
    n = frame->blocksize;

    /* CONSTANT */
    for(i=1; i<n; i++) {
        if(smp[i] != smp[0]) break;
    }
    if(i == n) {
        sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
        res[0] = smp[0];
        return sub->obits;
    }

    /* VERBATIM */
    if(n < 5) {
        sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
        encode_residual_verbatim(res, smp, n);
        return sub->obits * n;
    }

1023 1024 1025 1026 1027
    min_order = ctx->options.min_prediction_order;
    max_order = ctx->options.max_prediction_order;
    min_porder = ctx->options.min_partition_order;
    max_porder = ctx->options.max_partition_order;
    precision = ctx->options.lpc_coeff_precision;
1028
    omethod = ctx->options.prediction_order_method;
1029 1030

    /* FIXED */
1031 1032 1033 1034 1035 1036 1037
    if(!ctx->options.use_lpc || max_order == 0 || (n <= max_order)) {
        uint32_t bits[MAX_FIXED_ORDER+1];
        if(max_order > MAX_FIXED_ORDER) max_order = MAX_FIXED_ORDER;
        opt_order = 0;
        bits[0] = UINT32_MAX;
        for(i=min_order; i<=max_order; i++) {
            encode_residual_fixed(res, smp, n, i);
1038
            bits[i] = calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res,
1039 1040 1041 1042
                                             n, i, sub->obits);
            if(bits[i] < bits[opt_order]) {
                opt_order = i;
            }
1043
        }
1044 1045 1046 1047 1048
        sub->order = opt_order;
        sub->type = FLAC_SUBFRAME_FIXED;
        sub->type_code = sub->type | sub->order;
        if(sub->order != max_order) {
            encode_residual_fixed(res, smp, n, sub->order);
1049
            return calc_rice_params_fixed(&sub->rc, min_porder, max_porder, res, n,
1050 1051 1052
                                          sub->order, sub->obits);
        }
        return bits[sub->order];
1053
    }
1054 1055

    /* LPC */
1056
    opt_order = lpc_calc_coefs(ctx, smp, n, max_order, precision, coefs, shift, ctx->options.use_lpc, omethod);
1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092

    if(omethod == ORDER_METHOD_2LEVEL ||
       omethod == ORDER_METHOD_4LEVEL ||
       omethod == ORDER_METHOD_8LEVEL) {
        int levels = 1 << omethod;
        uint32_t bits[levels];
        int order;
        int opt_index = levels-1;
        opt_order = max_order-1;
        bits[opt_index] = UINT32_MAX;
        for(i=levels-1; i>=0; i--) {
            order = min_order + (((max_order-min_order+1) * (i+1)) / levels)-1;
            if(order < 0) order = 0;
            encode_residual_lpc(res, smp, n, order+1, coefs[order], shift[order]);
            bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
                                           res, n, order+1, sub->obits, precision);
            if(bits[i] < bits[opt_index]) {
                opt_index = i;
                opt_order = order;
            }
        }
        opt_order++;
    } else if(omethod == ORDER_METHOD_SEARCH) {
        // brute-force optimal order search
        uint32_t bits[MAX_LPC_ORDER];
        opt_order = 0;
        bits[0] = UINT32_MAX;
        for(i=min_order-1; i<max_order; i++) {
            encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
            bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
                                           res, n, i+1, sub->obits, precision);
            if(bits[i] < bits[opt_order]) {
                opt_order = i;
            }
        }
        opt_order++;
1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112
    } else if(omethod == ORDER_METHOD_LOG) {
        uint32_t bits[MAX_LPC_ORDER];
        int step;

        opt_order= min_order - 1 + (max_order-min_order)/3;
        memset(bits, -1, sizeof(bits));

        for(step=16 ;step; step>>=1){
            int last= opt_order;
            for(i=last-step; i<=last+step; i+= step){
                if(i<min_order-1 || i>=max_order || bits[i] < UINT32_MAX)
                    continue;
                encode_residual_lpc(res, smp, n, i+1, coefs[i], shift[i]);
                bits[i] = calc_rice_params_lpc(&sub->rc, min_porder, max_porder,
                                            res, n, i+1, sub->obits, precision);
                if(bits[i] < bits[opt_order])
                    opt_order= i;
            }
        }
        opt_order++;
1113 1114 1115
    }

    sub->order = opt_order;
1116 1117 1118 1119 1120
    sub->type = FLAC_SUBFRAME_LPC;
    sub->type_code = sub->type | (sub->order-1);
    sub->shift = shift[sub->order-1];
    for(i=0; i<sub->order; i++) {
        sub->coefs[i] = coefs[sub->order-1][i];
1121
    }
1122
    encode_residual_lpc(res, smp, n, sub->order, sub->coefs, sub->shift);
1123
    return calc_rice_params_lpc(&sub->rc, min_porder, max_porder, res, n, sub->order,
1124
                                sub->obits, precision);
1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155
}

static int encode_residual_v(FlacEncodeContext *ctx, int ch)
{
    int i, n;
    FlacFrame *frame;
    FlacSubframe *sub;
    int32_t *res, *smp;

    frame = &ctx->frame;
    sub = &frame->subframes[ch];
    res = sub->residual;
    smp = sub->samples;
    n = frame->blocksize;

    /* CONSTANT */
    for(i=1; i<n; i++) {
        if(smp[i] != smp[0]) break;
    }
    if(i == n) {
        sub->type = sub->type_code = FLAC_SUBFRAME_CONSTANT;
        res[0] = smp[0];
        return sub->obits;
    }

    /* VERBATIM */
    sub->type = sub->type_code = FLAC_SUBFRAME_VERBATIM;
    encode_residual_verbatim(res, smp, n);
    return sub->obits * n;
}

1156 1157 1158 1159
static int estimate_stereo_mode(int32_t *left_ch, int32_t *right_ch, int n)
{
    int i, best;
    int32_t lt, rt;
1160
    uint64_t sum[4];
1161
    uint64_t score[4];
1162
    int k;
1163

1164
    /* calculate sum of 2nd order residual for each channel */
1165
    sum[0] = sum[1] = sum[2] = sum[3] = 0;
1166 1167 1168
    for(i=2; i<n; i++) {
        lt = left_ch[i] - 2*left_ch[i-1] + left_ch[i-2];
        rt = right_ch[i] - 2*right_ch[i-1] + right_ch[i-2];
1169 1170 1171 1172
        sum[2] += FFABS((lt + rt) >> 1);
        sum[3] += FFABS(lt - rt);
        sum[0] += FFABS(lt);
        sum[1] += FFABS(rt);
1173
    }
1174
    /* estimate bit counts */
1175 1176 1177
    for(i=0; i<4; i++) {
        k = find_optimal_param(2*sum[i], n);
        sum[i] = rice_encode_count(2*sum[i], n, k);
1178 1179 1180
    }

    /* calculate score for each mode */
1181 1182 1183 1184
    score[0] = sum[0] + sum[1];
    score[1] = sum[0] + sum[3];
    score[2] = sum[1] + sum[3];
    score[3] = sum[2] + sum[3];
1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249

    /* return mode with lowest score */
    best = 0;
    for(i=1; i<4; i++) {
        if(score[i] < score[best]) {
            best = i;
        }
    }
    if(best == 0) {
        return FLAC_CHMODE_LEFT_RIGHT;
    } else if(best == 1) {
        return FLAC_CHMODE_LEFT_SIDE;
    } else if(best == 2) {
        return FLAC_CHMODE_RIGHT_SIDE;
    } else {
        return FLAC_CHMODE_MID_SIDE;
    }
}

/**
 * Perform stereo channel decorrelation
 */
static void channel_decorrelation(FlacEncodeContext *ctx)
{
    FlacFrame *frame;
    int32_t *left, *right;
    int i, n;

    frame = &ctx->frame;
    n = frame->blocksize;
    left  = frame->subframes[0].samples;
    right = frame->subframes[1].samples;

    if(ctx->channels != 2) {
        frame->ch_mode = FLAC_CHMODE_NOT_STEREO;
        return;
    }

    frame->ch_mode = estimate_stereo_mode(left, right, n);

    /* perform decorrelation and adjust bits-per-sample */
    if(frame->ch_mode == FLAC_CHMODE_LEFT_RIGHT) {
        return;
    }
    if(frame->ch_mode == FLAC_CHMODE_MID_SIDE) {
        int32_t tmp;
        for(i=0; i<n; i++) {
            tmp = left[i];
            left[i] = (tmp + right[i]) >> 1;
            right[i] = tmp - right[i];
        }
        frame->subframes[1].obits++;
    } else if(frame->ch_mode == FLAC_CHMODE_LEFT_SIDE) {
        for(i=0; i<n; i++) {
            right[i] = left[i] - right[i];
        }
        frame->subframes[1].obits++;
    } else {
        for(i=0; i<n; i++) {
            left[i] -= right[i];
        }
        frame->subframes[0].obits++;
    }
}

1250
static void put_sbits(PutBitContext *pb, int bits, int32_t val)
1251 1252
{
    assert(bits >= 0 && bits <= 31);
1253 1254

    put_bits(pb, bits, val & ((1<<bits)-1));
1255 1256
}

1257
static void write_utf8(PutBitContext *pb, uint32_t val)
1258
{
1259 1260
    uint8_t tmp;
    PUT_UTF8(val, tmp, put_bits(pb, 8, tmp);)
1261 1262
}

1263
static void output_frame_header(FlacEncodeContext *s)
1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280
{
    FlacFrame *frame;
    int crc;

    frame = &s->frame;

    put_bits(&s->pb, 16, 0xFFF8);
    put_bits(&s->pb, 4, frame->bs_code[0]);
    put_bits(&s->pb, 4, s->sr_code[0]);
    if(frame->ch_mode == FLAC_CHMODE_NOT_STEREO) {
        put_bits(&s->pb, 4, s->ch_code);
    } else {
        put_bits(&s->pb, 4, frame->ch_mode);
    }
    put_bits(&s->pb, 3, 4); /* bits-per-sample code */
    put_bits(&s->pb, 1, 0);
    write_utf8(&s->pb, s->frame_count);
1281 1282 1283 1284
    if(frame->bs_code[0] == 6) {
        put_bits(&s->pb, 8, frame->bs_code[1]);
    } else if(frame->bs_code[0] == 7) {
        put_bits(&s->pb, 16, frame->bs_code[1]);
1285
    }
1286 1287 1288 1289
    if(s->sr_code[0] == 12) {
        put_bits(&s->pb, 8, s->sr_code[1]);
    } else if(s->sr_code[0] > 12) {
        put_bits(&s->pb, 16, s->sr_code[1]);
1290 1291
    }
    flush_put_bits(&s->pb);
Aurelien Jacobs's avatar
Aurelien Jacobs committed
1292 1293
    crc = av_crc(av_crc_get_table(AV_CRC_8_ATM), 0,
                 s->pb.buf, put_bits_count(&s->pb)>>3);
1294 1295 1296
    put_bits(&s->pb, 8, crc);
}

1297 1298 1299 1300 1301 1302 1303 1304 1305 1306
static void output_subframe_constant(FlacEncodeContext *s, int ch)
{
    FlacSubframe *sub;
    int32_t res;

    sub = &s->frame.subframes[ch];
    res = sub->residual[0];
    put_sbits(&s->pb, sub->obits, res);
}

1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322
static void output_subframe_verbatim(FlacEncodeContext *s, int ch)
{
    int i;
    FlacFrame *frame;
    FlacSubframe *sub;
    int32_t res;

    frame = &s->frame;
    sub = &frame->subframes[ch];

    for(i=0; i<frame->blocksize; i++) {
        res = sub->residual[i];
        put_sbits(&s->pb, sub->obits, res);
    }
}

1323
static void output_residual(FlacEncodeContext *ctx, int ch)
1324
{
1325
    int i, j, p, n, parts;
1326 1327 1328
    int k, porder, psize, res_cnt;
    FlacFrame *frame;
    FlacSubframe *sub;
1329
    int32_t *res;
1330 1331 1332

    frame = &ctx->frame;
    sub = &frame->subframes[ch];
1333 1334
    res = sub->residual;
    n = frame->blocksize;
1335 1336 1337 1338 1339

    /* rice-encoded block */
    put_bits(&ctx->pb, 2, 0);

    /* partition order */
1340 1341 1342
    porder = sub->rc.porder;
    psize = n >> porder;
    parts = (1 << porder);
1343 1344 1345 1346 1347
    put_bits(&ctx->pb, 4, porder);
    res_cnt = psize - sub->order;

    /* residual */
    j = sub->order;
1348 1349
    for(p=0; p<parts; p++) {
        k = sub->rc.params[p];
1350 1351
        put_bits(&ctx->pb, 4, k);
        if(p == 1) res_cnt = psize;
1352 1353
        for(i=0; i<res_cnt && j<n; i++, j++) {
            set_sr_golomb_flac(&ctx->pb, res[j], k, INT32_MAX, 0);
1354 1355 1356 1357
        }
    }
}

1358
static void output_subframe_fixed(FlacEncodeContext *ctx, int ch)
1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375
{
    int i;
    FlacFrame *frame;
    FlacSubframe *sub;

    frame = &ctx->frame;
    sub = &frame->subframes[ch];

    /* warm-up samples */
    for(i=0; i<sub->order; i++) {
        put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
    }

    /* residual */
    output_residual(ctx, ch);
}

1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401
static void output_subframe_lpc(FlacEncodeContext *ctx, int ch)
{
    int i, cbits;
    FlacFrame *frame;
    FlacSubframe *sub;

    frame = &ctx->frame;
    sub = &frame->subframes[ch];

    /* warm-up samples */
    for(i=0; i<sub->order; i++) {
        put_sbits(&ctx->pb, sub->obits, sub->residual[i]);
    }

    /* LPC coefficients */
    cbits = ctx->options.lpc_coeff_precision;
    put_bits(&ctx->pb, 4, cbits-1);
    put_sbits(&ctx->pb, 5, sub->shift);
    for(i=0; i<sub->order; i++) {
        put_sbits(&ctx->pb, cbits, sub->coefs[i]);
    }

    /* residual */
    output_residual(ctx, ch);
}

1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418
static void output_subframes(FlacEncodeContext *s)
{
    FlacFrame *frame;
    FlacSubframe *sub;
    int ch;

    frame = &s->frame;

    for(ch=0; ch<s->channels; ch++) {
        sub = &frame->subframes[ch];

        /* subframe header */
        put_bits(&s->pb, 1, 0);
        put_bits(&s->pb, 6, sub->type_code);
        put_bits(&s->pb, 1, 0); /* no wasted bits */

        /* subframe */
1419 1420 1421
        if(sub->type == FLAC_SUBFRAME_CONSTANT) {
            output_subframe_constant(s, ch);
        } else if(sub->type == FLAC_SUBFRAME_VERBATIM) {
1422
            output_subframe_verbatim(s, ch);
1423
        } else if(sub->type == FLAC_SUBFRAME_FIXED) {
1424
            output_subframe_fixed(s, ch);
1425 1426
        } else if(sub->type == FLAC_SUBFRAME_LPC) {
            output_subframe_lpc(s, ch);
1427 1428 1429 1430 1431 1432 1433 1434
        }
    }
}

static void output_frame_footer(FlacEncodeContext *s)
{
    int crc;
    flush_put_bits(&s->pb);
Aurelien Jacobs's avatar
Aurelien Jacobs committed
1435 1436
    crc = bswap_16(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0,
                          s->pb.buf, put_bits_count(&s->pb)>>3));
1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451
    put_bits(&s->pb, 16, crc);
    flush_put_bits(&s->pb);
}

static int flac_encode_frame(AVCodecContext *avctx, uint8_t *frame,
                             int buf_size, void *data)
{
    int ch;
    FlacEncodeContext *s;
    int16_t *samples = data;
    int out_bytes;

    s = avctx->priv_data;

    s->blocksize = avctx->frame_size;
1452
    init_frame(s);
1453 1454 1455

    copy_samples(s, samples);

1456 1457
    channel_decorrelation(s);

1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469
    for(ch=0; ch<s->channels; ch++) {
        encode_residual(s, ch);
    }
    init_put_bits(&s->pb, frame, buf_size);
    output_frame_header(s);
    output_subframes(s);
    output_frame_footer(s);
    out_bytes = put_bits_count(&s->pb) >> 3;

    if(out_bytes > s->max_framesize || out_bytes >= buf_size) {
        /* frame too large. use verbatim mode */
        for(ch=0; ch<s->channels; ch++) {
1470
            encode_residual_v(s, ch);
1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490
        }
        init_put_bits(&s->pb, frame, buf_size);
        output_frame_header(s);
        output_subframes(s);
        output_frame_footer(s);
        out_bytes = put_bits_count(&s->pb) >> 3;

        if(out_bytes > s->max_framesize || out_bytes >= buf_size) {
            /* still too large. must be an error. */
            av_log(avctx, AV_LOG_ERROR, "error encoding frame\n");
            return -1;
        }
    }

    s->frame_count++;
    return out_bytes;
}

static int flac_encode_close(AVCodecContext *avctx)
{
1491 1492
    av_freep(&avctx->extradata);
    avctx->extradata_size = 0;
1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507
    av_freep(&avctx->coded_frame);
    return 0;
}

AVCodec flac_encoder = {
    "flac",
    CODEC_TYPE_AUDIO,
    CODEC_ID_FLAC,
    sizeof(FlacEncodeContext),
    flac_encode_init,
    flac_encode_frame,
    flac_encode_close,
    NULL,
    .capabilities = CODEC_CAP_SMALL_LAST_FRAME,
};