aacps.c 34.3 KB
Newer Older
1 2 3 4
/*
 * MPEG-4 Parametric Stereo decoding functions
 * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com>
 *
5
 * This file is part of Libav.
6
 *
7
 * Libav is free software; you can redistribute it and/or
8 9 10 11
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
12
 * Libav 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 Libav; if not, write to the Free Software
19 20 21 22
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include <stdint.h>
23
#include "libavutil/common.h"
24 25 26
#include "libavutil/mathematics.h"
#include "avcodec.h"
#include "get_bits.h"
27 28 29
#include "aacps.h"
#include "aacps_tablegen.h"
#include "aacpsdata.c"
Mans Rullgard's avatar
Mans Rullgard committed
30
#include "dsputil.h"
31

32 33 34
#define PS_BASELINE 0  ///< Operate in Baseline PS mode
                       ///< Baseline implies 10 or 20 stereo bands,
                       ///< mixing mode A, and no ipd/opd
35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72

#define numQMFSlots 32 //numTimeSlots * RATE

static const int8_t num_env_tab[2][4] = {
    { 0, 1, 2, 4, },
    { 1, 2, 3, 4, },
};

static const int8_t nr_iidicc_par_tab[] = {
    10, 20, 34, 10, 20, 34,
};

static const int8_t nr_iidopd_par_tab[] = {
     5, 11, 17,  5, 11, 17,
};

enum {
    huff_iid_df1,
    huff_iid_dt1,
    huff_iid_df0,
    huff_iid_dt0,
    huff_icc_df,
    huff_icc_dt,
    huff_ipd_df,
    huff_ipd_dt,
    huff_opd_df,
    huff_opd_dt,
};

static const int huff_iid[] = {
    huff_iid_df0,
    huff_iid_df1,
    huff_iid_dt0,
    huff_iid_dt1,
};

static VLC vlc_ps[10];

73
#define READ_PAR_DATA(PAR, OFFSET, MASK, ERR_CONDITION) \
74 75 76 77 78 79 80 81 82 83 84 85
/** \
 * Read Inter-channel Intensity Difference/Inter-Channel Coherence/ \
 * Inter-channel Phase Difference/Overall Phase Difference parameters from the \
 * bitstream. \
 * \
 * @param avctx contains the current codec context \
 * @param gb    pointer to the input bitstream \
 * @param ps    pointer to the Parametric Stereo context \
 * @param PAR   pointer to the parameter to be read \
 * @param e     envelope to decode \
 * @param dt    1: time delta-coded, 0: frequency delta-coded \
 */ \
86
static int read_ ## PAR ## _data(AVCodecContext *avctx, GetBitContext *gb, PSContext *ps, \
87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114
                        int8_t (*PAR)[PS_MAX_NR_IIDICC], int table_idx, int e, int dt) \
{ \
    int b, num = ps->nr_ ## PAR ## _par; \
    VLC_TYPE (*vlc_table)[2] = vlc_ps[table_idx].table; \
    if (dt) { \
        int e_prev = e ? e - 1 : ps->num_env_old - 1; \
        e_prev = FFMAX(e_prev, 0); \
        for (b = 0; b < num; b++) { \
            int val = PAR[e_prev][b] + get_vlc2(gb, vlc_table, 9, 3) - OFFSET; \
            if (MASK) val &= MASK; \
            PAR[e][b] = val; \
            if (ERR_CONDITION) \
                goto err; \
        } \
    } else { \
        int val = 0; \
        for (b = 0; b < num; b++) { \
            val += get_vlc2(gb, vlc_table, 9, 3) - OFFSET; \
            if (MASK) val &= MASK; \
            PAR[e][b] = val; \
            if (ERR_CONDITION) \
                goto err; \
        } \
    } \
    return 0; \
err: \
    av_log(avctx, AV_LOG_ERROR, "illegal "#PAR"\n"); \
    return -1; \
115 116
}

117 118 119
READ_PAR_DATA(iid,    huff_offset[table_idx],    0, FFABS(ps->iid_par[e][b]) > 7 + 8 * ps->iid_quant)
READ_PAR_DATA(icc,    huff_offset[table_idx],    0, ps->icc_par[e][b] > 7U)
READ_PAR_DATA(ipdopd,                      0, 0x07, 0)
120

121
static int ps_read_extension_data(GetBitContext *gb, PSContext *ps, int ps_extension_id)
122 123 124 125 126 127 128 129 130 131 132
{
    int e;
    int count = get_bits_count(gb);

    if (ps_extension_id)
        return 0;

    ps->enable_ipdopd = get_bits1(gb);
    if (ps->enable_ipdopd) {
        for (e = 0; e < ps->num_env; e++) {
            int dt = get_bits1(gb);
133
            read_ipdopd_data(NULL, gb, ps, ps->ipd_par, dt ? huff_ipd_dt : huff_ipd_df, e, dt);
134
            dt = get_bits1(gb);
135
            read_ipdopd_data(NULL, gb, ps, ps->opd_par, dt ? huff_opd_dt : huff_opd_df, e, dt);
136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162
        }
    }
    skip_bits1(gb);      //reserved_ps
    return get_bits_count(gb) - count;
}

static void ipdopd_reset(int8_t *opd_hist, int8_t *ipd_hist)
{
    int i;
    for (i = 0; i < PS_MAX_NR_IPDOPD; i++) {
        opd_hist[i] = 0;
        ipd_hist[i] = 0;
    }
}

int ff_ps_read_data(AVCodecContext *avctx, GetBitContext *gb_host, PSContext *ps, int bits_left)
{
    int e;
    int bit_count_start = get_bits_count(gb_host);
    int header;
    int bits_consumed;
    GetBitContext gbc = *gb_host, *gb = &gbc;

    header = get_bits1(gb);
    if (header) {     //enable_ps_header
        ps->enable_iid = get_bits1(gb);
        if (ps->enable_iid) {
163 164
            int iid_mode = get_bits(gb, 3);
            if (iid_mode > 5) {
165
                av_log(avctx, AV_LOG_ERROR, "iid_mode %d is reserved.\n",
166
                       iid_mode);
167 168
                goto err;
            }
169 170 171
            ps->nr_iid_par    = nr_iidicc_par_tab[iid_mode];
            ps->iid_quant     = iid_mode > 2;
            ps->nr_ipdopd_par = nr_iidopd_par_tab[iid_mode];
172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195
        }
        ps->enable_icc = get_bits1(gb);
        if (ps->enable_icc) {
            ps->icc_mode = get_bits(gb, 3);
            if (ps->icc_mode > 5) {
                av_log(avctx, AV_LOG_ERROR, "icc_mode %d is reserved.\n",
                       ps->icc_mode);
                goto err;
            }
            ps->nr_icc_par = nr_iidicc_par_tab[ps->icc_mode];
        }
        ps->enable_ext = get_bits1(gb);
    }

    ps->frame_class = get_bits1(gb);
    ps->num_env_old = ps->num_env;
    ps->num_env     = num_env_tab[ps->frame_class][get_bits(gb, 2)];

    ps->border_position[0] = -1;
    if (ps->frame_class) {
        for (e = 1; e <= ps->num_env; e++)
            ps->border_position[e] = get_bits(gb, 5);
    } else
        for (e = 1; e <= ps->num_env; e++)
196
            ps->border_position[e] = (e * numQMFSlots >> ff_log2_tab[ps->num_env]) - 1;
197 198 199 200

    if (ps->enable_iid) {
        for (e = 0; e < ps->num_env; e++) {
            int dt = get_bits1(gb);
201
            if (read_iid_data(avctx, gb, ps, ps->iid_par, huff_iid[2*dt+ps->iid_quant], e, dt))
202 203 204 205 206 207 208 209
                goto err;
        }
    } else
        memset(ps->iid_par, 0, sizeof(ps->iid_par));

    if (ps->enable_icc)
        for (e = 0; e < ps->num_env; e++) {
            int dt = get_bits1(gb);
210
            if (read_icc_data(avctx, gb, ps, ps->icc_par, dt ? huff_icc_dt : huff_icc_df, e, dt))
211 212 213 214 215 216 217 218 219 220 221 222 223
                goto err;
        }
    else
        memset(ps->icc_par, 0, sizeof(ps->icc_par));

    if (ps->enable_ext) {
        int cnt = get_bits(gb, 4);
        if (cnt == 15) {
            cnt += get_bits(gb, 8);
        }
        cnt *= 8;
        while (cnt > 7) {
            int ps_extension_id = get_bits(gb, 2);
224
            cnt -= 2 + ps_read_extension_data(gb, ps, ps_extension_id);
225 226
        }
        if (cnt < 0) {
227
            av_log(avctx, AV_LOG_ERROR, "ps extension overflow %d\n", cnt);
228 229 230 231 232 233 234 235 236 237 238 239
            goto err;
        }
        skip_bits(gb, cnt);
    }

    ps->enable_ipdopd &= !PS_BASELINE;

    //Fix up envelopes
    if (!ps->num_env || ps->border_position[ps->num_env] < numQMFSlots - 1) {
        //Create a fake envelope
        int source = ps->num_env ? ps->num_env - 1 : ps->num_env_old - 1;
        if (source >= 0 && source != ps->num_env) {
240
            if (ps->enable_iid) {
241 242
                memcpy(ps->iid_par+ps->num_env, ps->iid_par+source, sizeof(ps->iid_par[0]));
            }
243
            if (ps->enable_icc) {
244 245
                memcpy(ps->icc_par+ps->num_env, ps->icc_par+source, sizeof(ps->icc_par[0]));
            }
246
            if (ps->enable_ipdopd) {
247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278
                memcpy(ps->ipd_par+ps->num_env, ps->ipd_par+source, sizeof(ps->ipd_par[0]));
                memcpy(ps->opd_par+ps->num_env, ps->opd_par+source, sizeof(ps->opd_par[0]));
            }
        }
        ps->num_env++;
        ps->border_position[ps->num_env] = numQMFSlots - 1;
    }


    ps->is34bands_old = ps->is34bands;
    if (!PS_BASELINE && (ps->enable_iid || ps->enable_icc))
        ps->is34bands = (ps->enable_iid && ps->nr_iid_par == 34) ||
                        (ps->enable_icc && ps->nr_icc_par == 34);

    //Baseline
    if (!ps->enable_ipdopd) {
        memset(ps->ipd_par, 0, sizeof(ps->ipd_par));
        memset(ps->opd_par, 0, sizeof(ps->opd_par));
    }

    if (header)
        ps->start = 1;

    bits_consumed = get_bits_count(gb) - bit_count_start;
    if (bits_consumed <= bits_left) {
        skip_bits_long(gb_host, bits_consumed);
        return bits_consumed;
    }
    av_log(avctx, AV_LOG_ERROR, "Expected to read %d PS bits actually read %d.\n", bits_left, bits_consumed);
err:
    ps->start = 0;
    skip_bits_long(gb_host, bits_left);
279 280 281 282
    memset(ps->iid_par, 0, sizeof(ps->iid_par));
    memset(ps->icc_par, 0, sizeof(ps->icc_par));
    memset(ps->ipd_par, 0, sizeof(ps->ipd_par));
    memset(ps->opd_par, 0, sizeof(ps->opd_par));
283 284 285 286 287
    return bits_left;
}

/** Split one subband into 2 subsubbands with a symmetric real filter.
 * The filter must have its non-center even coefficients equal to zero. */
Mans Rullgard's avatar
Mans Rullgard committed
288
static void hybrid2_re(float (*in)[2], float (*out)[32][2], const float filter[8], int len, int reverse)
289 290
{
    int i, j;
291 292
    for (i = 0; i < len; i++, in++) {
        float re_in = filter[6] * in[6][0];          //real inphase
293
        float re_op = 0.0f;                          //real out of phase
294
        float im_in = filter[6] * in[6][1];          //imag inphase
295 296
        float im_op = 0.0f;                          //imag out of phase
        for (j = 0; j < 6; j += 2) {
297 298
            re_op += filter[j+1] * (in[j+1][0] + in[12-j-1][0]);
            im_op += filter[j+1] * (in[j+1][1] + in[12-j-1][1]);
299 300 301 302 303 304 305 306 307
        }
        out[ reverse][i][0] = re_in + re_op;
        out[ reverse][i][1] = im_in + im_op;
        out[!reverse][i][0] = re_in - re_op;
        out[!reverse][i][1] = im_in - im_op;
    }
}

/** Split one subband into 6 subsubbands with a complex filter */
Mans Rullgard's avatar
Mans Rullgard committed
308
static void hybrid6_cx(PSDSPContext *dsp, float (*in)[2], float (*out)[32][2], const float (*filter)[8][2], int len)
309
{
310
    int i;
311
    int N = 8;
Mans Rullgard's avatar
Mans Rullgard committed
312
    LOCAL_ALIGNED_16(float, temp, [8], [2]);
313

314
    for (i = 0; i < len; i++, in++) {
315
        dsp->hybrid_analysis(temp, in, filter, 1, N);
316 317 318 319 320 321 322 323 324 325 326 327 328 329 330
        out[0][i][0] = temp[6][0];
        out[0][i][1] = temp[6][1];
        out[1][i][0] = temp[7][0];
        out[1][i][1] = temp[7][1];
        out[2][i][0] = temp[0][0];
        out[2][i][1] = temp[0][1];
        out[3][i][0] = temp[1][0];
        out[3][i][1] = temp[1][1];
        out[4][i][0] = temp[2][0] + temp[5][0];
        out[4][i][1] = temp[2][1] + temp[5][1];
        out[5][i][0] = temp[3][0] + temp[4][0];
        out[5][i][1] = temp[3][1] + temp[4][1];
    }
}

Mans Rullgard's avatar
Mans Rullgard committed
331
static void hybrid4_8_12_cx(PSDSPContext *dsp, float (*in)[2], float (*out)[32][2], const float (*filter)[8][2], int N, int len)
332
{
333
    int i;
334

335
    for (i = 0; i < len; i++, in++) {
336
        dsp->hybrid_analysis(out[0] + i, in, filter, 32, N);
337 338 339
    }
}

340 341 342
static void hybrid_analysis(PSDSPContext *dsp, float out[91][32][2],
                            float in[5][44][2], float L[2][38][64],
                            int is34, int len)
343 344 345 346 347 348 349 350
{
    int i, j;
    for (i = 0; i < 5; i++) {
        for (j = 0; j < 38; j++) {
            in[i][j+6][0] = L[0][j][i];
            in[i][j+6][1] = L[1][j][i];
        }
    }
Alex Converse's avatar
Alex Converse committed
351
    if (is34) {
352 353 354 355 356 357
        hybrid4_8_12_cx(dsp, in[0], out,    f34_0_12, 12, len);
        hybrid4_8_12_cx(dsp, in[1], out+12, f34_1_8,   8, len);
        hybrid4_8_12_cx(dsp, in[2], out+20, f34_2_4,   4, len);
        hybrid4_8_12_cx(dsp, in[3], out+24, f34_2_4,   4, len);
        hybrid4_8_12_cx(dsp, in[4], out+28, f34_2_4,   4, len);
        dsp->hybrid_analysis_ileave(out + 27, L, 5, len);
358
    } else {
359
        hybrid6_cx(dsp, in[0], out, f20_0_8, len);
360 361
        hybrid2_re(in[1], out+6, g1_Q2, len, 1);
        hybrid2_re(in[2], out+8, g1_Q2, len, 0);
362
        dsp->hybrid_analysis_ileave(out + 7, L, 3, len);
363 364 365 366 367 368 369
    }
    //update in_buf
    for (i = 0; i < 5; i++) {
        memcpy(in[i], in[i]+32, 6 * sizeof(in[i][0]));
    }
}

370 371
static void hybrid_synthesis(PSDSPContext *dsp, float out[2][38][64],
                             float in[91][32][2], int is34, int len)
372 373
{
    int i, n;
Alex Converse's avatar
Alex Converse committed
374
    if (is34) {
375 376 377
        for (n = 0; n < len; n++) {
            memset(out[0][n], 0, 5*sizeof(out[0][n][0]));
            memset(out[1][n], 0, 5*sizeof(out[1][n][0]));
Alex Converse's avatar
Alex Converse committed
378
            for (i = 0; i < 12; i++) {
379 380 381
                out[0][n][0] += in[   i][n][0];
                out[1][n][0] += in[   i][n][1];
            }
Alex Converse's avatar
Alex Converse committed
382
            for (i = 0; i < 8; i++) {
383 384 385
                out[0][n][1] += in[12+i][n][0];
                out[1][n][1] += in[12+i][n][1];
            }
Alex Converse's avatar
Alex Converse committed
386
            for (i = 0; i < 4; i++) {
387 388 389 390 391 392 393 394
                out[0][n][2] += in[20+i][n][0];
                out[1][n][2] += in[20+i][n][1];
                out[0][n][3] += in[24+i][n][0];
                out[1][n][3] += in[24+i][n][1];
                out[0][n][4] += in[28+i][n][0];
                out[1][n][4] += in[28+i][n][1];
            }
        }
395
        dsp->hybrid_synthesis_deint(out, in + 27, 5, len);
396 397 398 399 400 401 402 403 404 405 406
    } else {
        for (n = 0; n < len; n++) {
            out[0][n][0] = in[0][n][0] + in[1][n][0] + in[2][n][0] +
                           in[3][n][0] + in[4][n][0] + in[5][n][0];
            out[1][n][0] = in[0][n][1] + in[1][n][1] + in[2][n][1] +
                           in[3][n][1] + in[4][n][1] + in[5][n][1];
            out[0][n][1] = in[6][n][0] + in[7][n][0];
            out[1][n][1] = in[6][n][1] + in[7][n][1];
            out[0][n][2] = in[8][n][0] + in[9][n][0];
            out[1][n][2] = in[8][n][1] + in[9][n][1];
        }
407
        dsp->hybrid_synthesis_deint(out, in + 7, 3, len);
408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610
    }
}

/// All-pass filter decay slope
#define DECAY_SLOPE      0.05f
/// Number of frequency bands that can be addressed by the parameter index, b(k)
static const int   NR_PAR_BANDS[]      = { 20, 34 };
/// Number of frequency bands that can be addressed by the sub subband index, k
static const int   NR_BANDS[]          = { 71, 91 };
/// Start frequency band for the all-pass filter decay slope
static const int   DECAY_CUTOFF[]      = { 10, 32 };
/// Number of all-pass filer bands
static const int   NR_ALLPASS_BANDS[]  = { 30, 50 };
/// First stereo band using the short one sample delay
static const int   SHORT_DELAY_BAND[]  = { 42, 62 };

/** Table 8.46 */
static void map_idx_10_to_20(int8_t *par_mapped, const int8_t *par, int full)
{
    int b;
    if (full)
        b = 9;
    else {
        b = 4;
        par_mapped[10] = 0;
    }
    for (; b >= 0; b--) {
        par_mapped[2*b+1] = par_mapped[2*b] = par[b];
    }
}

static void map_idx_34_to_20(int8_t *par_mapped, const int8_t *par, int full)
{
    par_mapped[ 0] = (2*par[ 0] +   par[ 1]) / 3;
    par_mapped[ 1] = (  par[ 1] + 2*par[ 2]) / 3;
    par_mapped[ 2] = (2*par[ 3] +   par[ 4]) / 3;
    par_mapped[ 3] = (  par[ 4] + 2*par[ 5]) / 3;
    par_mapped[ 4] = (  par[ 6] +   par[ 7]) / 2;
    par_mapped[ 5] = (  par[ 8] +   par[ 9]) / 2;
    par_mapped[ 6] =    par[10];
    par_mapped[ 7] =    par[11];
    par_mapped[ 8] = (  par[12] +   par[13]) / 2;
    par_mapped[ 9] = (  par[14] +   par[15]) / 2;
    par_mapped[10] =    par[16];
    if (full) {
        par_mapped[11] =    par[17];
        par_mapped[12] =    par[18];
        par_mapped[13] =    par[19];
        par_mapped[14] = (  par[20] +   par[21]) / 2;
        par_mapped[15] = (  par[22] +   par[23]) / 2;
        par_mapped[16] = (  par[24] +   par[25]) / 2;
        par_mapped[17] = (  par[26] +   par[27]) / 2;
        par_mapped[18] = (  par[28] +   par[29] +   par[30] +   par[31]) / 4;
        par_mapped[19] = (  par[32] +   par[33]) / 2;
    }
}

static void map_val_34_to_20(float par[PS_MAX_NR_IIDICC])
{
    par[ 0] = (2*par[ 0] +   par[ 1]) * 0.33333333f;
    par[ 1] = (  par[ 1] + 2*par[ 2]) * 0.33333333f;
    par[ 2] = (2*par[ 3] +   par[ 4]) * 0.33333333f;
    par[ 3] = (  par[ 4] + 2*par[ 5]) * 0.33333333f;
    par[ 4] = (  par[ 6] +   par[ 7]) * 0.5f;
    par[ 5] = (  par[ 8] +   par[ 9]) * 0.5f;
    par[ 6] =    par[10];
    par[ 7] =    par[11];
    par[ 8] = (  par[12] +   par[13]) * 0.5f;
    par[ 9] = (  par[14] +   par[15]) * 0.5f;
    par[10] =    par[16];
    par[11] =    par[17];
    par[12] =    par[18];
    par[13] =    par[19];
    par[14] = (  par[20] +   par[21]) * 0.5f;
    par[15] = (  par[22] +   par[23]) * 0.5f;
    par[16] = (  par[24] +   par[25]) * 0.5f;
    par[17] = (  par[26] +   par[27]) * 0.5f;
    par[18] = (  par[28] +   par[29] +   par[30] +   par[31]) * 0.25f;
    par[19] = (  par[32] +   par[33]) * 0.5f;
}

static void map_idx_10_to_34(int8_t *par_mapped, const int8_t *par, int full)
{
    if (full) {
        par_mapped[33] = par[9];
        par_mapped[32] = par[9];
        par_mapped[31] = par[9];
        par_mapped[30] = par[9];
        par_mapped[29] = par[9];
        par_mapped[28] = par[9];
        par_mapped[27] = par[8];
        par_mapped[26] = par[8];
        par_mapped[25] = par[8];
        par_mapped[24] = par[8];
        par_mapped[23] = par[7];
        par_mapped[22] = par[7];
        par_mapped[21] = par[7];
        par_mapped[20] = par[7];
        par_mapped[19] = par[6];
        par_mapped[18] = par[6];
        par_mapped[17] = par[5];
        par_mapped[16] = par[5];
    } else {
        par_mapped[16] =      0;
    }
    par_mapped[15] = par[4];
    par_mapped[14] = par[4];
    par_mapped[13] = par[4];
    par_mapped[12] = par[4];
    par_mapped[11] = par[3];
    par_mapped[10] = par[3];
    par_mapped[ 9] = par[2];
    par_mapped[ 8] = par[2];
    par_mapped[ 7] = par[2];
    par_mapped[ 6] = par[2];
    par_mapped[ 5] = par[1];
    par_mapped[ 4] = par[1];
    par_mapped[ 3] = par[1];
    par_mapped[ 2] = par[0];
    par_mapped[ 1] = par[0];
    par_mapped[ 0] = par[0];
}

static void map_idx_20_to_34(int8_t *par_mapped, const int8_t *par, int full)
{
    if (full) {
        par_mapped[33] =  par[19];
        par_mapped[32] =  par[19];
        par_mapped[31] =  par[18];
        par_mapped[30] =  par[18];
        par_mapped[29] =  par[18];
        par_mapped[28] =  par[18];
        par_mapped[27] =  par[17];
        par_mapped[26] =  par[17];
        par_mapped[25] =  par[16];
        par_mapped[24] =  par[16];
        par_mapped[23] =  par[15];
        par_mapped[22] =  par[15];
        par_mapped[21] =  par[14];
        par_mapped[20] =  par[14];
        par_mapped[19] =  par[13];
        par_mapped[18] =  par[12];
        par_mapped[17] =  par[11];
    }
    par_mapped[16] =  par[10];
    par_mapped[15] =  par[ 9];
    par_mapped[14] =  par[ 9];
    par_mapped[13] =  par[ 8];
    par_mapped[12] =  par[ 8];
    par_mapped[11] =  par[ 7];
    par_mapped[10] =  par[ 6];
    par_mapped[ 9] =  par[ 5];
    par_mapped[ 8] =  par[ 5];
    par_mapped[ 7] =  par[ 4];
    par_mapped[ 6] =  par[ 4];
    par_mapped[ 5] =  par[ 3];
    par_mapped[ 4] = (par[ 2] + par[ 3]) / 2;
    par_mapped[ 3] =  par[ 2];
    par_mapped[ 2] =  par[ 1];
    par_mapped[ 1] = (par[ 0] + par[ 1]) / 2;
    par_mapped[ 0] =  par[ 0];
}

static void map_val_20_to_34(float par[PS_MAX_NR_IIDICC])
{
    par[33] =  par[19];
    par[32] =  par[19];
    par[31] =  par[18];
    par[30] =  par[18];
    par[29] =  par[18];
    par[28] =  par[18];
    par[27] =  par[17];
    par[26] =  par[17];
    par[25] =  par[16];
    par[24] =  par[16];
    par[23] =  par[15];
    par[22] =  par[15];
    par[21] =  par[14];
    par[20] =  par[14];
    par[19] =  par[13];
    par[18] =  par[12];
    par[17] =  par[11];
    par[16] =  par[10];
    par[15] =  par[ 9];
    par[14] =  par[ 9];
    par[13] =  par[ 8];
    par[12] =  par[ 8];
    par[11] =  par[ 7];
    par[10] =  par[ 6];
    par[ 9] =  par[ 5];
    par[ 8] =  par[ 5];
    par[ 7] =  par[ 4];
    par[ 6] =  par[ 4];
    par[ 5] =  par[ 3];
    par[ 4] = (par[ 2] + par[ 3]) * 0.5f;
    par[ 3] =  par[ 2];
    par[ 2] =  par[ 1];
    par[ 1] = (par[ 0] + par[ 1]) * 0.5f;
    par[ 0] =  par[ 0];
}

static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[32][2], int is34)
{
Mans Rullgard's avatar
Mans Rullgard committed
611 612
    LOCAL_ALIGNED_16(float, power, [34], [PS_QMF_TIME_SLOTS]);
    LOCAL_ALIGNED_16(float, transient_gain, [34], [PS_QMF_TIME_SLOTS]);
613 614 615 616 617 618 619 620
    float *peak_decay_nrg = ps->peak_decay_nrg;
    float *power_smooth = ps->power_smooth;
    float *peak_decay_diff_smooth = ps->peak_decay_diff_smooth;
    float (*delay)[PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2] = ps->delay;
    float (*ap_delay)[PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2] = ps->ap_delay;
    const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20;
    const float peak_decay_factor = 0.76592833836465f;
    const float transient_impact  = 1.5f;
621
    const float a_smooth          = 0.25f; ///< Smoothing coefficient
622 623 624
    int i, k, m, n;
    int n0 = 0, nL = 32;

Mans Rullgard's avatar
Mans Rullgard committed
625 626
    memset(power, 0, 34 * sizeof(*power));

627 628 629 630 631 632 633 634
    if (is34 != ps->is34bands_old) {
        memset(ps->peak_decay_nrg,         0, sizeof(ps->peak_decay_nrg));
        memset(ps->power_smooth,           0, sizeof(ps->power_smooth));
        memset(ps->peak_decay_diff_smooth, 0, sizeof(ps->peak_decay_diff_smooth));
        memset(ps->delay,                  0, sizeof(ps->delay));
        memset(ps->ap_delay,               0, sizeof(ps->ap_delay));
    }

635 636 637
    for (k = 0; k < NR_BANDS[is34]; k++) {
        int i = k_to_i[k];
        ps->dsp.add_squares(power[i], s[k], nL - n0);
638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670
    }

    //Transient detection
    for (i = 0; i < NR_PAR_BANDS[is34]; i++) {
        for (n = n0; n < nL; n++) {
            float decayed_peak = peak_decay_factor * peak_decay_nrg[i];
            float denom;
            peak_decay_nrg[i] = FFMAX(decayed_peak, power[i][n]);
            power_smooth[i] += a_smooth * (power[i][n] - power_smooth[i]);
            peak_decay_diff_smooth[i] += a_smooth * (peak_decay_nrg[i] - power[i][n] - peak_decay_diff_smooth[i]);
            denom = transient_impact * peak_decay_diff_smooth[i];
            transient_gain[i][n]   = (denom > power_smooth[i]) ?
                                         power_smooth[i] / denom : 1.0f;
        }
    }

    //Decorrelation and transient reduction
    //                         PS_AP_LINKS - 1
    //                               -----
    //                                | |  Q_fract_allpass[k][m]*z^-link_delay[m] - a[m]*g_decay_slope[k]
    //H[k][z] = z^-2 * phi_fract[k] * | | ----------------------------------------------------------------
    //                                | | 1 - a[m]*g_decay_slope[k]*Q_fract_allpass[k][m]*z^-link_delay[m]
    //                               m = 0
    //d[k][z] (out) = transient_gain_mapped[k][z] * H[k][z] * s[k][z]
    for (k = 0; k < NR_ALLPASS_BANDS[is34]; k++) {
        int b = k_to_i[k];
        float g_decay_slope = 1.f - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]);
        g_decay_slope = av_clipf(g_decay_slope, 0.f, 1.f);
        memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
        memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
        for (m = 0; m < PS_AP_LINKS; m++) {
            memcpy(ap_delay[k][m],   ap_delay[k][m]+numQMFSlots,           5*sizeof(ap_delay[k][m][0]));
        }
671 672 673
        ps->dsp.decorrelate(out[k], delay[k] + PS_MAX_DELAY - 2, ap_delay[k],
                            phi_fract[is34][k], Q_fract_allpass[is34][k],
                            transient_gain[b], g_decay_slope, nL - n0);
674 675
    }
    for (; k < SHORT_DELAY_BAND[is34]; k++) {
676
        int i = k_to_i[k];
677 678
        memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
        memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
679 680 681
        //H = delay 14
        ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 14,
                                transient_gain[i], nL - n0);
682 683
    }
    for (; k < NR_BANDS[is34]; k++) {
684
        int i = k_to_i[k];
685 686
        memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0]));
        memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0]));
687 688 689
        //H = delay 1
        ps->dsp.mul_pair_single(out[k], delay[k] + PS_MAX_DELAY - 1,
                                transient_gain[i], nL - n0);
690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732
    }
}

static void remap34(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC],
                    int8_t           (*par)[PS_MAX_NR_IIDICC],
                    int num_par, int num_env, int full)
{
    int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped;
    int e;
    if (num_par == 20 || num_par == 11) {
        for (e = 0; e < num_env; e++) {
            map_idx_20_to_34(par_mapped[e], par[e], full);
        }
    } else if (num_par == 10 || num_par == 5) {
        for (e = 0; e < num_env; e++) {
            map_idx_10_to_34(par_mapped[e], par[e], full);
        }
    } else {
        *p_par_mapped = par;
    }
}

static void remap20(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC],
                    int8_t           (*par)[PS_MAX_NR_IIDICC],
                    int num_par, int num_env, int full)
{
    int8_t (*par_mapped)[PS_MAX_NR_IIDICC] = *p_par_mapped;
    int e;
    if (num_par == 34 || num_par == 17) {
        for (e = 0; e < num_env; e++) {
            map_idx_34_to_20(par_mapped[e], par[e], full);
        }
    } else if (num_par == 10 || num_par == 5) {
        for (e = 0; e < num_env; e++) {
            map_idx_10_to_20(par_mapped[e], par[e], full);
        }
    } else {
        *p_par_mapped = par;
    }
}

static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2], int is34)
{
733
    int e, b, k;
734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752

    float (*H11)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H11;
    float (*H12)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H12;
    float (*H21)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H21;
    float (*H22)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H22;
    int8_t *opd_hist = ps->opd_hist;
    int8_t *ipd_hist = ps->ipd_hist;
    int8_t iid_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
    int8_t icc_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
    int8_t ipd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
    int8_t opd_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC];
    int8_t (*iid_mapped)[PS_MAX_NR_IIDICC] = iid_mapped_buf;
    int8_t (*icc_mapped)[PS_MAX_NR_IIDICC] = icc_mapped_buf;
    int8_t (*ipd_mapped)[PS_MAX_NR_IIDICC] = ipd_mapped_buf;
    int8_t (*opd_mapped)[PS_MAX_NR_IIDICC] = opd_mapped_buf;
    const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20;
    const float (*H_LUT)[8][4] = (PS_BASELINE || ps->icc_mode < 3) ? HA : HB;

    //Remapping
753 754 755 756 757 758 759 760 761 762 763
    if (ps->num_env_old) {
        memcpy(H11[0][0], H11[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H11[0][0][0]));
        memcpy(H11[1][0], H11[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H11[1][0][0]));
        memcpy(H12[0][0], H12[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H12[0][0][0]));
        memcpy(H12[1][0], H12[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H12[1][0][0]));
        memcpy(H21[0][0], H21[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H21[0][0][0]));
        memcpy(H21[1][0], H21[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H21[1][0][0]));
        memcpy(H22[0][0], H22[0][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H22[0][0][0]));
        memcpy(H22[1][0], H22[1][ps->num_env_old], PS_MAX_NR_IIDICC*sizeof(H22[1][0][0]));
    }

764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844
    if (is34) {
        remap34(&iid_mapped, ps->iid_par, ps->nr_iid_par, ps->num_env, 1);
        remap34(&icc_mapped, ps->icc_par, ps->nr_icc_par, ps->num_env, 1);
        if (ps->enable_ipdopd) {
            remap34(&ipd_mapped, ps->ipd_par, ps->nr_ipdopd_par, ps->num_env, 0);
            remap34(&opd_mapped, ps->opd_par, ps->nr_ipdopd_par, ps->num_env, 0);
        }
        if (!ps->is34bands_old) {
            map_val_20_to_34(H11[0][0]);
            map_val_20_to_34(H11[1][0]);
            map_val_20_to_34(H12[0][0]);
            map_val_20_to_34(H12[1][0]);
            map_val_20_to_34(H21[0][0]);
            map_val_20_to_34(H21[1][0]);
            map_val_20_to_34(H22[0][0]);
            map_val_20_to_34(H22[1][0]);
            ipdopd_reset(ipd_hist, opd_hist);
        }
    } else {
        remap20(&iid_mapped, ps->iid_par, ps->nr_iid_par, ps->num_env, 1);
        remap20(&icc_mapped, ps->icc_par, ps->nr_icc_par, ps->num_env, 1);
        if (ps->enable_ipdopd) {
            remap20(&ipd_mapped, ps->ipd_par, ps->nr_ipdopd_par, ps->num_env, 0);
            remap20(&opd_mapped, ps->opd_par, ps->nr_ipdopd_par, ps->num_env, 0);
        }
        if (ps->is34bands_old) {
            map_val_34_to_20(H11[0][0]);
            map_val_34_to_20(H11[1][0]);
            map_val_34_to_20(H12[0][0]);
            map_val_34_to_20(H12[1][0]);
            map_val_34_to_20(H21[0][0]);
            map_val_34_to_20(H21[1][0]);
            map_val_34_to_20(H22[0][0]);
            map_val_34_to_20(H22[1][0]);
            ipdopd_reset(ipd_hist, opd_hist);
        }
    }

    //Mixing
    for (e = 0; e < ps->num_env; e++) {
        for (b = 0; b < NR_PAR_BANDS[is34]; b++) {
            float h11, h12, h21, h22;
            h11 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][0];
            h12 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][1];
            h21 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][2];
            h22 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][3];
            if (!PS_BASELINE && ps->enable_ipdopd && b < ps->nr_ipdopd_par) {
                //The spec say says to only run this smoother when enable_ipdopd
                //is set but the reference decoder appears to run it constantly
                float h11i, h12i, h21i, h22i;
                float ipd_adj_re, ipd_adj_im;
                int opd_idx = opd_hist[b] * 8 + opd_mapped[e][b];
                int ipd_idx = ipd_hist[b] * 8 + ipd_mapped[e][b];
                float opd_re = pd_re_smooth[opd_idx];
                float opd_im = pd_im_smooth[opd_idx];
                float ipd_re = pd_re_smooth[ipd_idx];
                float ipd_im = pd_im_smooth[ipd_idx];
                opd_hist[b] = opd_idx & 0x3F;
                ipd_hist[b] = ipd_idx & 0x3F;

                ipd_adj_re = opd_re*ipd_re + opd_im*ipd_im;
                ipd_adj_im = opd_im*ipd_re - opd_re*ipd_im;
                h11i = h11 * opd_im;
                h11  = h11 * opd_re;
                h12i = h12 * ipd_adj_im;
                h12  = h12 * ipd_adj_re;
                h21i = h21 * opd_im;
                h21  = h21 * opd_re;
                h22i = h22 * ipd_adj_im;
                h22  = h22 * ipd_adj_re;
                H11[1][e+1][b] = h11i;
                H12[1][e+1][b] = h12i;
                H21[1][e+1][b] = h21i;
                H22[1][e+1][b] = h22i;
            }
            H11[0][e+1][b] = h11;
            H12[0][e+1][b] = h12;
            H21[0][e+1][b] = h21;
            H22[0][e+1][b] = h22;
        }
        for (k = 0; k < NR_BANDS[is34]; k++) {
845 846
            float h[2][4];
            float h_step[2][4];
847 848 849 850
            int start = ps->border_position[e];
            int stop  = ps->border_position[e+1];
            float width = 1.f / (stop - start);
            b = k_to_i[k];
851 852 853 854
            h[0][0] = H11[0][e][b];
            h[0][1] = H12[0][e][b];
            h[0][2] = H21[0][e][b];
            h[0][3] = H22[0][e][b];
855 856 857
            if (!PS_BASELINE && ps->enable_ipdopd) {
            //Is this necessary? ps_04_new seems unchanged
            if ((is34 && k <= 13 && k >= 9) || (!is34 && k <= 1)) {
858 859 860 861
                h[1][0] = -H11[1][e][b];
                h[1][1] = -H12[1][e][b];
                h[1][2] = -H21[1][e][b];
                h[1][3] = -H22[1][e][b];
862
            } else {
863 864 865 866
                h[1][0] = H11[1][e][b];
                h[1][1] = H12[1][e][b];
                h[1][2] = H21[1][e][b];
                h[1][3] = H22[1][e][b];
867 868 869
            }
            }
            //Interpolation
870 871 872 873
            h_step[0][0] = (H11[0][e+1][b] - h[0][0]) * width;
            h_step[0][1] = (H12[0][e+1][b] - h[0][1]) * width;
            h_step[0][2] = (H21[0][e+1][b] - h[0][2]) * width;
            h_step[0][3] = (H22[0][e+1][b] - h[0][3]) * width;
874
            if (!PS_BASELINE && ps->enable_ipdopd) {
875 876 877 878
                h_step[1][0] = (H11[1][e+1][b] - h[1][0]) * width;
                h_step[1][1] = (H12[1][e+1][b] - h[1][1]) * width;
                h_step[1][2] = (H21[1][e+1][b] - h[1][2]) * width;
                h_step[1][3] = (H22[1][e+1][b] - h[1][3]) * width;
879
            }
880 881 882
            ps->dsp.stereo_interpolate[!PS_BASELINE && ps->enable_ipdopd](
                l[k] + start + 1, r[k] + start + 1,
                h, h_step, stop - start);
883 884 885 886 887 888
        }
    }
}

int ff_ps_apply(AVCodecContext *avctx, PSContext *ps, float L[2][38][64], float R[2][38][64], int top)
{
Mans Rullgard's avatar
Mans Rullgard committed
889 890
    LOCAL_ALIGNED_16(float, Lbuf, [91], [32][2]);
    LOCAL_ALIGNED_16(float, Rbuf, [91], [32][2]);
891 892 893 894 895 896 897 898
    const int len = 32;
    int is34 = ps->is34bands;

    top += NR_BANDS[is34] - 64;
    memset(ps->delay+top, 0, (NR_BANDS[is34] - top)*sizeof(ps->delay[0]));
    if (top < NR_ALLPASS_BANDS[is34])
        memset(ps->ap_delay + top, 0, (NR_ALLPASS_BANDS[is34] - top)*sizeof(ps->ap_delay[0]));

899
    hybrid_analysis(&ps->dsp, Lbuf, ps->in_buf, L, is34, len);
900 901
    decorrelation(ps, Rbuf, Lbuf, is34);
    stereo_processing(ps, Lbuf, Rbuf, is34);
902 903
    hybrid_synthesis(&ps->dsp, L, Lbuf, is34, len);
    hybrid_synthesis(&ps->dsp, R, Rbuf, is34, len);
904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950

    return 0;
}

#define PS_INIT_VLC_STATIC(num, size) \
    INIT_VLC_STATIC(&vlc_ps[num], 9, ps_tmp[num].table_size / ps_tmp[num].elem_size,    \
                    ps_tmp[num].ps_bits, 1, 1,                                          \
                    ps_tmp[num].ps_codes, ps_tmp[num].elem_size, ps_tmp[num].elem_size, \
                    size);

#define PS_VLC_ROW(name) \
    { name ## _codes, name ## _bits, sizeof(name ## _codes), sizeof(name ## _codes[0]) }

av_cold void ff_ps_init(void) {
    // Syntax initialization
    static const struct {
        const void *ps_codes, *ps_bits;
        const unsigned int table_size, elem_size;
    } ps_tmp[] = {
        PS_VLC_ROW(huff_iid_df1),
        PS_VLC_ROW(huff_iid_dt1),
        PS_VLC_ROW(huff_iid_df0),
        PS_VLC_ROW(huff_iid_dt0),
        PS_VLC_ROW(huff_icc_df),
        PS_VLC_ROW(huff_icc_dt),
        PS_VLC_ROW(huff_ipd_df),
        PS_VLC_ROW(huff_ipd_dt),
        PS_VLC_ROW(huff_opd_df),
        PS_VLC_ROW(huff_opd_dt),
    };

    PS_INIT_VLC_STATIC(0, 1544);
    PS_INIT_VLC_STATIC(1,  832);
    PS_INIT_VLC_STATIC(2, 1024);
    PS_INIT_VLC_STATIC(3, 1036);
    PS_INIT_VLC_STATIC(4,  544);
    PS_INIT_VLC_STATIC(5,  544);
    PS_INIT_VLC_STATIC(6,  512);
    PS_INIT_VLC_STATIC(7,  512);
    PS_INIT_VLC_STATIC(8,  512);
    PS_INIT_VLC_STATIC(9,  512);

    ps_tableinit();
}

av_cold void ff_ps_ctx_init(PSContext *ps)
{
951
    ff_psdsp_init(&ps->dsp);
952
}