twinvq.c 36.6 KB
Newer Older
Vitor Sessak's avatar
Vitor Sessak committed
1 2 3 4
/*
 * TwinVQ decoder
 * Copyright (c) 2009 Vitor Sessak
 *
5
 * This file is part of Libav.
Vitor Sessak's avatar
Vitor Sessak committed
6
 *
7
 * Libav is free software; you can redistribute it and/or
Vitor Sessak's avatar
Vitor Sessak committed
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,
Vitor Sessak's avatar
Vitor Sessak committed
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
Vitor Sessak's avatar
Vitor Sessak committed
19 20 21 22 23 24
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "avcodec.h"
#include "get_bits.h"
#include "dsputil.h"
25
#include "fft.h"
26
#include "lsp.h"
27
#include "sinewin.h"
Vitor Sessak's avatar
Vitor Sessak committed
28 29 30 31 32 33 34 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 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177

#include <math.h>
#include <stdint.h>

#include "twinvq_data.h"

enum FrameType {
    FT_SHORT = 0,  ///< Short frame  (divided in n   sub-blocks)
    FT_MEDIUM,     ///< Medium frame (divided in m<n sub-blocks)
    FT_LONG,       ///< Long frame   (single sub-block + PPC)
    FT_PPC,        ///< Periodic Peak Component (part of the long frame)
};

/**
 * Parameters and tables that are different for each frame type
 */
struct FrameMode {
    uint8_t         sub;      ///< Number subblocks in each frame
    const uint16_t *bark_tab;

    /** number of distinct bark scale envelope values */
    uint8_t         bark_env_size;

    const int16_t  *bark_cb;    ///< codebook for the bark scale envelope (BSE)
    uint8_t         bark_n_coef;///< number of BSE CB coefficients to read
    uint8_t         bark_n_bit; ///< number of bits of the BSE coefs

    //@{
    /** main codebooks for spectrum data */
    const int16_t    *cb0;
    const int16_t    *cb1;
    //@}

    uint8_t         cb_len_read; ///< number of spectrum coefficients to read
};

/**
 * Parameters and tables that are different for every combination of
 * bitrate/sample rate
 */
typedef struct {
    struct FrameMode fmode[3]; ///< frame type-dependant parameters

    uint16_t     size;        ///< frame size in samples
    uint8_t      n_lsp;       ///< number of lsp coefficients
    const float *lspcodebook;

    /* number of bits of the different LSP CB coefficients */
    uint8_t      lsp_bit0;
    uint8_t      lsp_bit1;
    uint8_t      lsp_bit2;

    uint8_t      lsp_split;      ///< number of CB entries for the LSP decoding
    const int16_t *ppc_shape_cb; ///< PPC shape CB

    /** number of the bits for the PPC period value */
    uint8_t      ppc_period_bit;

    uint8_t      ppc_shape_bit;  ///< number of bits of the PPC shape CB coeffs
    uint8_t      ppc_shape_len;  ///< size of PPC shape CB
    uint8_t      pgain_bit;      ///< bits for PPC gain

    /** constant for peak period to peak width conversion */
    uint16_t     peak_per2wid;
} ModeTab;

static const ModeTab mode_08_08 = {
    {
        { 8, bark_tab_s08_64,  10, tab.fcb08s  , 1, 5, tab.cb0808s0, tab.cb0808s1, 18},
        { 2, bark_tab_m08_256, 20, tab.fcb08m  , 2, 5, tab.cb0808m0, tab.cb0808m1, 16},
        { 1, bark_tab_l08_512, 30, tab.fcb08l  , 3, 6, tab.cb0808l0, tab.cb0808l1, 17}
    },
    512 , 12, tab.lsp08,   1, 5, 3, 3, tab.shape08  , 8, 28, 20, 6, 40
};

static const ModeTab mode_11_08 = {
    {
        { 8, bark_tab_s11_64,  10, tab.fcb11s  , 1, 5, tab.cb1108s0, tab.cb1108s1, 29},
        { 2, bark_tab_m11_256, 20, tab.fcb11m  , 2, 5, tab.cb1108m0, tab.cb1108m1, 24},
        { 1, bark_tab_l11_512, 30, tab.fcb11l  , 3, 6, tab.cb1108l0, tab.cb1108l1, 27}
    },
    512 , 16, tab.lsp11,   1, 6, 4, 3, tab.shape11  , 9, 36, 30, 7, 90
};

static const ModeTab mode_11_10 = {
    {
        { 8, bark_tab_s11_64,  10, tab.fcb11s  , 1, 5, tab.cb1110s0, tab.cb1110s1, 21},
        { 2, bark_tab_m11_256, 20, tab.fcb11m  , 2, 5, tab.cb1110m0, tab.cb1110m1, 18},
        { 1, bark_tab_l11_512, 30, tab.fcb11l  , 3, 6, tab.cb1110l0, tab.cb1110l1, 20}
    },
    512 , 16, tab.lsp11,   1, 6, 4, 3, tab.shape11  , 9, 36, 30, 7, 90
};

static const ModeTab mode_16_16 = {
    {
        { 8, bark_tab_s16_128, 10, tab.fcb16s  , 1, 5, tab.cb1616s0, tab.cb1616s1, 16},
        { 2, bark_tab_m16_512, 20, tab.fcb16m  , 2, 5, tab.cb1616m0, tab.cb1616m1, 15},
        { 1, bark_tab_l16_1024,30, tab.fcb16l  , 3, 6, tab.cb1616l0, tab.cb1616l1, 16}
    },
    1024, 16, tab.lsp16,   1, 6, 4, 3, tab.shape16  , 9, 56, 60, 7, 180
};

static const ModeTab mode_22_20 = {
    {
        { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2220s0, tab.cb2220s1, 18},
        { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2220m0, tab.cb2220m1, 17},
        { 1, bark_tab_l22_1024,32, tab.fcb22l_1, 4, 6, tab.cb2220l0, tab.cb2220l1, 18}
    },
    1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
};

static const ModeTab mode_22_24 = {
    {
        { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2224s0, tab.cb2224s1, 15},
        { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2224m0, tab.cb2224m1, 14},
        { 1, bark_tab_l22_1024,32, tab.fcb22l_1, 4, 6, tab.cb2224l0, tab.cb2224l1, 15}
    },
    1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
};

static const ModeTab mode_22_32 = {
    {
        { 4, bark_tab_s22_128, 10, tab.fcb22s_2, 1, 6, tab.cb2232s0, tab.cb2232s1, 11},
        { 2, bark_tab_m22_256, 20, tab.fcb22m_2, 2, 6, tab.cb2232m0, tab.cb2232m1, 11},
        { 1, bark_tab_l22_512, 32, tab.fcb22l_2, 4, 6, tab.cb2232l0, tab.cb2232l1, 12}
    },
    512 , 16, tab.lsp22_2, 1, 6, 4, 4, tab.shape22_2, 9, 56, 36, 7, 72
};

static const ModeTab mode_44_40 = {
    {
        {16, bark_tab_s44_128, 10, tab.fcb44s  , 1, 6, tab.cb4440s0, tab.cb4440s1, 18},
        { 4, bark_tab_m44_512, 20, tab.fcb44m  , 2, 6, tab.cb4440m0, tab.cb4440m1, 17},
        { 1, bark_tab_l44_2048,40, tab.fcb44l  , 4, 6, tab.cb4440l0, tab.cb4440l1, 17}
    },
    2048, 20, tab.lsp44,   1, 6, 4, 4, tab.shape44  , 9, 84, 54, 7, 432
};

static const ModeTab mode_44_48 = {
    {
        {16, bark_tab_s44_128, 10, tab.fcb44s  , 1, 6, tab.cb4448s0, tab.cb4448s1, 15},
        { 4, bark_tab_m44_512, 20, tab.fcb44m  , 2, 6, tab.cb4448m0, tab.cb4448m1, 14},
        { 1, bark_tab_l44_2048,40, tab.fcb44l  , 4, 6, tab.cb4448l0, tab.cb4448l1, 14}
    },
    2048, 20, tab.lsp44,   1, 6, 4, 4, tab.shape44  , 9, 84, 54, 7, 432
};

typedef struct TwinContext {
    AVCodecContext *avctx;
    DSPContext      dsp;
178
    FFTContext mdct_ctx[3];
Vitor Sessak's avatar
Vitor Sessak committed
179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205

    const ModeTab *mtab;

    // history
    float lsp_hist[2][20];           ///< LSP coefficients of the last frame
    float bark_hist[3][2][40];       ///< BSE coefficients of last frame

    // bitstream parameters
    int16_t permut[4][4096];
    uint8_t length[4][2];            ///< main codebook stride
    uint8_t length_change[4];
    uint8_t bits_main_spec[2][4][2]; ///< bits for the main codebook
    int bits_main_spec_change[4];
    int n_div[4];

    float *spectrum;
    float *curr_frame;               ///< non-interleaved output
    float *prev_frame;               ///< non-interleaved previous frame
    int last_block_pos[2];

    float *cos_tabs[3];

    // scratch buffers
    float *tmp_buf;
} TwinContext;

#define PPC_SHAPE_CB_SIZE 64
Måns Rullgård's avatar
Måns Rullgård committed
206
#define PPC_SHAPE_LEN_MAX 60
Vitor Sessak's avatar
Vitor Sessak committed
207 208 209 210 211 212 213
#define SUB_AMP_MAX       4500.0
#define MULAW_MU          100.0
#define GAIN_BITS         8
#define AMP_MAX           13000.0
#define SUB_GAIN_BITS     5
#define WINDOW_TYPE_BITS  4
#define PGAIN_MU          200
Måns Rullgård's avatar
Måns Rullgård committed
214 215 216 217 218
#define LSP_COEFS_MAX     20
#define LSP_SPLIT_MAX     4
#define CHANNELS_MAX      2
#define SUBBLOCKS_MAX     16
#define BARK_N_COEF_MAX   4
Vitor Sessak's avatar
Vitor Sessak committed
219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245

/** @note not speed critical, hence not optimized */
static void memset_float(float *buf, float val, int size)
{
    while (size--)
        *buf++ = val;
}

/**
 * Evaluate a single LPC amplitude spectrum envelope coefficient from the line
 * spectrum pairs.
 *
 * @param lsp a vector of the cosinus of the LSP values
 * @param cos_val cos(PI*i/N) where i is the index of the LPC amplitude
 * @param order the order of the LSP (and the size of the *lsp buffer). Must
 *        be a multiple of four.
 * @return the LPC value
 *
 * @todo reuse code from vorbis_dec.c: vorbis_floor0_decode
 */
static float eval_lpc_spectrum(const float *lsp, float cos_val, int order)
{
    int j;
    float p = 0.5f;
    float q = 0.5f;
    float two_cos_w = 2.0f*cos_val;

Vitor Sessak's avatar
Vitor Sessak committed
246
    for (j = 0; j + 1 < order; j += 2*2) {
Vitor Sessak's avatar
Vitor Sessak committed
247 248 249 250 251 252 253 254 255 256 257 258 259 260 261
        // Unroll the loop once since order is a multiple of four
        q *= lsp[j  ] - two_cos_w;
        p *= lsp[j+1] - two_cos_w;

        q *= lsp[j+2] - two_cos_w;
        p *= lsp[j+3] - two_cos_w;
    }

    p *= p * (2.0f - two_cos_w);
    q *= q * (2.0f + two_cos_w);

    return 0.5 / (p + q);
}

/**
262
 * Evaluate the LPC amplitude spectrum envelope from the line spectrum pairs.
Vitor Sessak's avatar
Vitor Sessak committed
263 264 265 266 267 268 269
 */
static void eval_lpcenv(TwinContext *tctx, const float *cos_vals, float *lpc)
{
    int i;
    const ModeTab *mtab = tctx->mtab;
    int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;

Vitor Sessak's avatar
Vitor Sessak committed
270
    for (i = 0; i < size_s/2; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
271 272 273 274 275 276 277 278 279 280 281
        float cos_i = tctx->cos_tabs[0][i];
        lpc[i]          = eval_lpc_spectrum(cos_vals,  cos_i, mtab->n_lsp);
        lpc[size_s-i-1] = eval_lpc_spectrum(cos_vals, -cos_i, mtab->n_lsp);
    }
}

static void interpolate(float *out, float v1, float v2, int size)
{
    int i;
    float step = (v1 - v2)/(size + 1);

Vitor Sessak's avatar
Vitor Sessak committed
282
    for (i = 0; i < size; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
283 284 285 286 287 288 289 290 291 292 293 294
        v2 += step;
        out[i] = v2;
    }
}

static inline float get_cos(int idx, int part, const float *cos_tab, int size)
{
    return part ? -cos_tab[size - idx - 1] :
                   cos_tab[       idx    ];
}

/**
295
 * Evaluate the LPC amplitude spectrum envelope from the line spectrum pairs.
Vitor Sessak's avatar
Vitor Sessak committed
296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317
 * Probably for speed reasons, the coefficients are evaluated as
 * siiiibiiiisiiiibiiiisiiiibiiiisiiiibiiiis ...
 * where s is an evaluated value, i is a value interpolated from the others
 * and b might be either calculated or interpolated, depending on an
 * unexplained condition.
 *
 * @param step the size of a block "siiiibiiii"
 * @param in the cosinus of the LSP data
 * @param part is 0 for 0...PI (positive cossinus values) and 1 for PI...2PI
          (negative cossinus values)
 * @param size the size of the whole output
 */
static inline void eval_lpcenv_or_interp(TwinContext *tctx,
                                         enum FrameType ftype,
                                         float *out, const float *in,
                                         int size, int step, int part)
{
    int i;
    const ModeTab *mtab = tctx->mtab;
    const float *cos_tab = tctx->cos_tabs[ftype];

    // Fill the 's'
Vitor Sessak's avatar
Vitor Sessak committed
318
    for (i = 0; i < size; i += step)
Vitor Sessak's avatar
Vitor Sessak committed
319 320 321 322 323 324
        out[i] =
            eval_lpc_spectrum(in,
                              get_cos(i, part, cos_tab, size),
                              mtab->n_lsp);

    // Fill the 'iiiibiiii'
Vitor Sessak's avatar
Vitor Sessak committed
325
    for (i = step; i <= size - 2*step; i += step) {
Vitor Sessak's avatar
Vitor Sessak committed
326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365
        if (out[i + step] + out[i - step] >  1.95*out[i] ||
            out[i + step]                 >=  out[i - step]) {
            interpolate(out + i - step + 1, out[i], out[i-step], step - 1);
        } else {
            out[i - step/2] =
                eval_lpc_spectrum(in,
                                  get_cos(i-step/2, part, cos_tab, size),
                                  mtab->n_lsp);
            interpolate(out + i - step   + 1, out[i-step/2], out[i-step  ], step/2 - 1);
            interpolate(out + i - step/2 + 1, out[i       ], out[i-step/2], step/2 - 1);
        }
    }

    interpolate(out + size - 2*step + 1, out[size-step], out[size - 2*step], step - 1);
}

static void eval_lpcenv_2parts(TwinContext *tctx, enum FrameType ftype,
                               const float *buf, float *lpc,
                               int size, int step)
{
    eval_lpcenv_or_interp(tctx, ftype, lpc         , buf, size/2,   step, 0);
    eval_lpcenv_or_interp(tctx, ftype, lpc + size/2, buf, size/2, 2*step, 1);

    interpolate(lpc+size/2-step+1, lpc[size/2], lpc[size/2-step], step);

    memset_float(lpc + size - 2*step + 1, lpc[size - 2*step], 2*step - 1);
}

/**
 * Inverse quantization. Read CB coefficients for cb1 and cb2 from the
 * bitstream, sum the corresponding vectors and write the result to *out
 * after permutation.
 */
static void dequant(TwinContext *tctx, GetBitContext *gb, float *out,
                    enum FrameType ftype,
                    const int16_t *cb0, const int16_t *cb1, int cb_len)
{
    int pos = 0;
    int i, j;

Vitor Sessak's avatar
Vitor Sessak committed
366
    for (i = 0; i < tctx->n_div[ftype]; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394
        int tmp0, tmp1;
        int sign0 = 1;
        int sign1 = 1;
        const int16_t *tab0, *tab1;
        int length = tctx->length[ftype][i >= tctx->length_change[ftype]];
        int bitstream_second_part = (i >= tctx->bits_main_spec_change[ftype]);

        int bits = tctx->bits_main_spec[0][ftype][bitstream_second_part];
        if (bits == 7) {
            if (get_bits1(gb))
                sign0 = -1;
            bits = 6;
        }
        tmp0 = get_bits(gb, bits);

        bits = tctx->bits_main_spec[1][ftype][bitstream_second_part];

        if (bits == 7) {
            if (get_bits1(gb))
                sign1 = -1;

            bits = 6;
        }
        tmp1 = get_bits(gb, bits);

        tab0 = cb0 + tmp0*cb_len;
        tab1 = cb1 + tmp1*cb_len;

Vitor Sessak's avatar
Vitor Sessak committed
395
        for (j = 0; j < length; j++)
Vitor Sessak's avatar
Vitor Sessak committed
396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458
            out[tctx->permut[ftype][pos+j]] = sign0*tab0[j] + sign1*tab1[j];

        pos += length;
    }

}

static inline float mulawinv(float y, float clip, float mu)
{
    y = av_clipf(y/clip, -1, 1);
    return clip * FFSIGN(y) * (exp(log(1+mu) * fabs(y)) - 1) / mu;
}

/**
 * Evaluate a*b/400 rounded to the nearest integer. When, for example,
 * a*b == 200 and the nearest integer is ill-defined, use a table to emulate
 * the following broken float-based implementation used by the binary decoder:
 *
 * \code
 * static int very_broken_op(int a, int b)
 * {
 *    static float test; // Ugh, force gcc to do the division first...
 *
 *    test = a/400.;
 *    return b * test +  0.5;
 * }
 * \endcode
 *
 * @note if this function is replaced by just ROUNDED_DIV(a*b,400.), the stddev
 * between the original file (before encoding with Yamaha encoder) and the
 * decoded output increases, which leads one to believe that the encoder expects
 * exactly this broken calculation.
 */
static int very_broken_op(int a, int b)
{
    int x = a*b + 200;
    int size;
    const uint8_t *rtab;

    if (x%400 || b%5)
        return x/400;

    x /= 400;

    size = tabs[b/5].size;
    rtab = tabs[b/5].tab;
    return x - rtab[size*av_log2(2*(x - 1)/size)+(x - 1)%size];
}

/**
 * Sum to data a periodic peak of a given period, width and shape.
 *
 * @param period the period of the peak divised by 400.0
 */
static void add_peak(int period, int width, const float *shape,
                     float ppc_gain, float *speech, int len)
{
    int i, j;

    const float *shape_end = shape + len;
    int center;

    // First peak centered around zero
Vitor Sessak's avatar
Vitor Sessak committed
459
    for (i = 0; i < width/2; i++)
Vitor Sessak's avatar
Vitor Sessak committed
460 461
        speech[i] += ppc_gain * *shape++;

Vitor Sessak's avatar
Vitor Sessak committed
462
    for (i = 1; i < ROUNDED_DIV(len,width) ; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
463
        center = very_broken_op(period, i);
Vitor Sessak's avatar
Vitor Sessak committed
464
        for (j = -width/2; j < (width+1)/2; j++)
Vitor Sessak's avatar
Vitor Sessak committed
465 466 467 468 469
            speech[j+center] += ppc_gain * *shape++;
    }

    // For the last block, be careful not to go beyond the end of the buffer
    center = very_broken_op(period, i);
Vitor Sessak's avatar
Vitor Sessak committed
470
    for (j = -width/2; j < (width + 1)/2 && shape < shape_end; j++)
Vitor Sessak's avatar
Vitor Sessak committed
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
        speech[j+center] += ppc_gain * *shape++;
}

static void decode_ppc(TwinContext *tctx, int period_coef, const float *shape,
                       float ppc_gain, float *speech)
{
    const ModeTab *mtab = tctx->mtab;
    int isampf = tctx->avctx->sample_rate/1000;
    int ibps = tctx->avctx->bit_rate/(1000 * tctx->avctx->channels);
    int min_period = ROUNDED_DIV(  40*2*mtab->size, isampf);
    int max_period = ROUNDED_DIV(6*40*2*mtab->size, isampf);
    int period_range = max_period - min_period;

    // This is actually the period multiplied by 400. It is just linearly coded
    // between its maximum and minimum value.
    int period = min_period +
        ROUNDED_DIV(period_coef*period_range, (1 << mtab->ppc_period_bit) - 1);
    int width;

    if (isampf == 22 && ibps == 32) {
        // For some unknown reason, NTT decided to code this case differently...
        width = ROUNDED_DIV((period + 800)* mtab->peak_per2wid, 400*mtab->size);
    } else
        width =             (period      )* mtab->peak_per2wid/(400*mtab->size);

    add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len);
}

static void dec_gain(TwinContext *tctx, GetBitContext *gb, enum FrameType ftype,
                     float *out)
{
    const ModeTab *mtab = tctx->mtab;
    int i, j;
    int sub = mtab->fmode[ftype].sub;
    float step     = AMP_MAX     / ((1 <<     GAIN_BITS) - 1);
    float sub_step = SUB_AMP_MAX / ((1 << SUB_GAIN_BITS) - 1);

    if (ftype == FT_LONG) {
Vitor Sessak's avatar
Vitor Sessak committed
509
        for (i = 0; i < tctx->avctx->channels; i++)
Vitor Sessak's avatar
Vitor Sessak committed
510 511 512 513
            out[i] = (1./(1<<13)) *
                mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
                         AMP_MAX, MULAW_MU);
    } else {
Vitor Sessak's avatar
Vitor Sessak committed
514
        for (i = 0; i < tctx->avctx->channels; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
515 516 517 518
            float val = (1./(1<<23)) *
                mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
                         AMP_MAX, MULAW_MU);

Vitor Sessak's avatar
Vitor Sessak committed
519
            for (j = 0; j < sub; j++) {
Vitor Sessak's avatar
Vitor Sessak committed
520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538
                out[i*sub + j] =
                    val*mulawinv(sub_step* 0.5 +
                                 sub_step* get_bits(gb, SUB_GAIN_BITS),
                                 SUB_AMP_MAX, MULAW_MU);
            }
        }
    }
}

/**
 * Rearrange the LSP coefficients so that they have a minimum distance of
 * min_dist. This function does it exactly as described in section of 3.2.4
 * of the G.729 specification (but interestingly is different from what the
 * reference decoder actually does).
 */
static void rearrange_lsp(int order, float *lsp, float min_dist)
{
    int i;
    float min_dist2 = min_dist * 0.5;
Vitor Sessak's avatar
Vitor Sessak committed
539
    for (i = 1; i < order; i++)
Vitor Sessak's avatar
Vitor Sessak committed
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
        if (lsp[i] - lsp[i-1] < min_dist) {
            float avg = (lsp[i] + lsp[i-1]) * 0.5;

            lsp[i-1] = avg - min_dist2;
            lsp[i  ] = avg + min_dist2;
        }
}

static void decode_lsp(TwinContext *tctx, int lpc_idx1, uint8_t *lpc_idx2,
                       int lpc_hist_idx, float *lsp, float *hist)
{
    const ModeTab *mtab = tctx->mtab;
    int i, j;

    const float *cb  =  mtab->lspcodebook;
    const float *cb2 =  cb  + (1 << mtab->lsp_bit1)*mtab->n_lsp;
    const float *cb3 =  cb2 + (1 << mtab->lsp_bit2)*mtab->n_lsp;

    const int8_t funny_rounding[4] = {
        -2,
        mtab->lsp_split == 4 ? -2 : 1,
        mtab->lsp_split == 4 ? -2 : 1,
        0
    };

Vitor Sessak's avatar
Vitor Sessak committed
565 566
    j = 0;
    for (i = 0; i < mtab->lsp_split; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
567 568 569 570 571 572 573 574
        int chunk_end = ((i + 1)*mtab->n_lsp + funny_rounding[i])/mtab->lsp_split;
        for (; j < chunk_end; j++)
            lsp[j] = cb [lpc_idx1    * mtab->n_lsp + j] +
                     cb2[lpc_idx2[i] * mtab->n_lsp + j];
    }

    rearrange_lsp(mtab->n_lsp, lsp, 0.0001);

Vitor Sessak's avatar
Vitor Sessak committed
575
    for (i = 0; i < mtab->n_lsp; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
576 577 578 579 580 581 582 583
        float tmp1 = 1. -          cb3[lpc_hist_idx*mtab->n_lsp + i];
        float tmp2 =     hist[i] * cb3[lpc_hist_idx*mtab->n_lsp + i];
        hist[i] = lsp[i];
        lsp[i]  = lsp[i] * tmp1 + tmp2;
    }

    rearrange_lsp(mtab->n_lsp, lsp, 0.0001);
    rearrange_lsp(mtab->n_lsp, lsp, 0.000095);
584
    ff_sort_nearly_sorted_floats(lsp, mtab->n_lsp);
Vitor Sessak's avatar
Vitor Sessak committed
585 586 587 588 589 590 591 592
}

static void dec_lpc_spectrum_inv(TwinContext *tctx, float *lsp,
                                 enum FrameType ftype, float *lpc)
{
    int i;
    int size = tctx->mtab->size / tctx->mtab->fmode[ftype].sub;

Vitor Sessak's avatar
Vitor Sessak committed
593
    for (i = 0; i < tctx->mtab->n_lsp; i++)
Vitor Sessak's avatar
Vitor Sessak committed
594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611
        lsp[i] =  2*cos(lsp[i]);

    switch (ftype) {
    case FT_LONG:
        eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 8);
        break;
    case FT_MEDIUM:
        eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 2);
        break;
    case FT_SHORT:
        eval_lpcenv(tctx, lsp, lpc);
        break;
    }
}

static void imdct_and_window(TwinContext *tctx, enum FrameType ftype, int wtype,
                            float *in, float *prev, int ch)
{
612
    FFTContext *mdct = &tctx->mdct_ctx[ftype];
Vitor Sessak's avatar
Vitor Sessak committed
613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634
    const ModeTab *mtab = tctx->mtab;
    int bsize = mtab->size / mtab->fmode[ftype].sub;
    int size  = mtab->size;
    float *buf1 = tctx->tmp_buf;
    int j;
    int wsize; // Window size
    float *out = tctx->curr_frame + 2*ch*mtab->size;
    float *out2 = out;
    float *prev_buf;
    int first_wsize;

    static const uint8_t wtype_to_wsize[]      = {0, 0, 2, 2, 2, 1, 0, 1, 1};
    int types_sizes[] = {
        mtab->size /    mtab->fmode[FT_LONG  ].sub,
        mtab->size /    mtab->fmode[FT_MEDIUM].sub,
        mtab->size / (2*mtab->fmode[FT_SHORT ].sub),
    };

    wsize = types_sizes[wtype_to_wsize[wtype]];
    first_wsize = wsize;
    prev_buf = prev + (size - bsize)/2;

Vitor Sessak's avatar
Vitor Sessak committed
635
    for (j = 0; j < mtab->fmode[ftype].sub; j++) {
Vitor Sessak's avatar
Vitor Sessak committed
636 637 638 639 640 641 642 643 644
        int sub_wtype = ftype == FT_MEDIUM ? 8 : wtype;

        if (!j && wtype == 4)
            sub_wtype = 4;
        else if (j == mtab->fmode[ftype].sub-1 && wtype == 7)
            sub_wtype = 7;

        wsize = types_sizes[wtype_to_wsize[sub_wtype]];

645
        mdct->imdct_half(mdct, buf1 + bsize*j, in + bsize*j);
Vitor Sessak's avatar
Vitor Sessak committed
646 647 648 649

        tctx->dsp.vector_fmul_window(out2,
                                     prev_buf + (bsize-wsize)/2,
                                     buf1 + bsize*j,
650
                                     ff_sine_windows[av_log2(wsize)],
Vitor Sessak's avatar
Vitor Sessak committed
651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670
                                     wsize/2);
        out2 += wsize;

        memcpy(out2, buf1 + bsize*j + wsize/2, (bsize - wsize/2)*sizeof(float));

        out2 += ftype == FT_MEDIUM ? (bsize-wsize)/2 : bsize - wsize;

        prev_buf = buf1 + bsize*j + bsize/2;
    }

    tctx->last_block_pos[ch] = (size + first_wsize)/2;
}

static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype,
                         float *out)
{
    const ModeTab *mtab = tctx->mtab;
    float *prev_buf = tctx->prev_frame + tctx->last_block_pos[0];
    int i, j;

Vitor Sessak's avatar
Vitor Sessak committed
671
    for (i = 0; i < tctx->avctx->channels; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
672 673 674 675 676 677 678
        imdct_and_window(tctx, ftype, wtype,
                         tctx->spectrum + i*mtab->size,
                         prev_buf + 2*i*mtab->size,
                         i);
    }

    if (tctx->avctx->channels == 2) {
Vitor Sessak's avatar
Vitor Sessak committed
679
        for (i = 0; i < mtab->size - tctx->last_block_pos[0]; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
680 681 682 683 684
            float f1 = prev_buf[               i];
            float f2 = prev_buf[2*mtab->size + i];
            out[2*i    ] = f1 + f2;
            out[2*i + 1] = f1 - f2;
        }
Vitor Sessak's avatar
Vitor Sessak committed
685
        for (j = 0; i < mtab->size; j++,i++) {
Vitor Sessak's avatar
Vitor Sessak committed
686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713
            float f1 = tctx->curr_frame[               j];
            float f2 = tctx->curr_frame[2*mtab->size + j];
            out[2*i    ] = f1 + f2;
            out[2*i + 1] = f1 - f2;
        }
    } else {
        memcpy(out, prev_buf,
               (mtab->size - tctx->last_block_pos[0]) * sizeof(*out));

        out +=  mtab->size - tctx->last_block_pos[0];

        memcpy(out, tctx->curr_frame,
               (tctx->last_block_pos[0]) * sizeof(*out));
    }

}

static void dec_bark_env(TwinContext *tctx, const uint8_t *in, int use_hist,
                         int ch, float *out, float gain, enum FrameType ftype)
{
    const ModeTab *mtab = tctx->mtab;
    int i,j;
    float *hist = tctx->bark_hist[ftype][ch];
    float val = ((const float []) {0.4, 0.35, 0.28})[ftype];
    int bark_n_coef  = mtab->fmode[ftype].bark_n_coef;
    int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef;
    int idx = 0;

Vitor Sessak's avatar
Vitor Sessak committed
714 715
    for (i = 0; i < fw_cb_len; i++)
        for (j = 0; j < bark_n_coef; j++, idx++) {
Vitor Sessak's avatar
Vitor Sessak committed
716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736
            float tmp2 =
                mtab->fmode[ftype].bark_cb[fw_cb_len*in[j] + i] * (1./4096);
            float st = use_hist ?
                (1. - val) * tmp2 + val*hist[idx] + 1. : tmp2 + 1.;

            hist[idx] = tmp2;
            if (st < -1.) st = 1.;

            memset_float(out, st * gain, mtab->fmode[ftype].bark_tab[idx]);
            out += mtab->fmode[ftype].bark_tab[idx];
        }

}

static void read_and_decode_spectrum(TwinContext *tctx, GetBitContext *gb,
                                     float *out, enum FrameType ftype)
{
    const ModeTab *mtab = tctx->mtab;
    int channels = tctx->avctx->channels;
    int sub = mtab->fmode[ftype].sub;
    int block_size = mtab->size / sub;
Måns Rullgård's avatar
Måns Rullgård committed
737 738 739 740
    float gain[CHANNELS_MAX*SUBBLOCKS_MAX];
    float ppc_shape[PPC_SHAPE_LEN_MAX * CHANNELS_MAX * 4];
    uint8_t bark1[CHANNELS_MAX][SUBBLOCKS_MAX][BARK_N_COEF_MAX];
    uint8_t bark_use_hist[CHANNELS_MAX][SUBBLOCKS_MAX];
Vitor Sessak's avatar
Vitor Sessak committed
741

Måns Rullgård's avatar
Måns Rullgård committed
742 743 744
    uint8_t lpc_idx1[CHANNELS_MAX];
    uint8_t lpc_idx2[CHANNELS_MAX][LSP_SPLIT_MAX];
    uint8_t lpc_hist_idx[CHANNELS_MAX];
Vitor Sessak's avatar
Vitor Sessak committed
745 746 747 748 749 750 751

    int i, j, k;

    dequant(tctx, gb, out, ftype,
            mtab->fmode[ftype].cb0, mtab->fmode[ftype].cb1,
            mtab->fmode[ftype].cb_len_read);

Vitor Sessak's avatar
Vitor Sessak committed
752 753 754
    for (i = 0; i < channels; i++)
        for (j = 0; j < sub; j++)
            for (k = 0; k < mtab->fmode[ftype].bark_n_coef; k++)
Vitor Sessak's avatar
Vitor Sessak committed
755 756 757
                bark1[i][j][k] =
                    get_bits(gb, mtab->fmode[ftype].bark_n_bit);

Vitor Sessak's avatar
Vitor Sessak committed
758 759
    for (i = 0; i < channels; i++)
        for (j = 0; j < sub; j++)
Vitor Sessak's avatar
Vitor Sessak committed
760 761 762 763
            bark_use_hist[i][j] = get_bits1(gb);

    dec_gain(tctx, gb, ftype, gain);

Vitor Sessak's avatar
Vitor Sessak committed
764
    for (i = 0; i < channels; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
765 766 767
        lpc_hist_idx[i] = get_bits(gb, tctx->mtab->lsp_bit0);
        lpc_idx1    [i] = get_bits(gb, tctx->mtab->lsp_bit1);

Vitor Sessak's avatar
Vitor Sessak committed
768
        for (j = 0; j < tctx->mtab->lsp_split; j++)
Vitor Sessak's avatar
Vitor Sessak committed
769 770 771 772 773 774 775 776 777 778
            lpc_idx2[i][j] = get_bits(gb, tctx->mtab->lsp_bit2);
    }

    if (ftype == FT_LONG) {
        int cb_len_p = (tctx->n_div[3] + mtab->ppc_shape_len*channels - 1)/
            tctx->n_div[3];
        dequant(tctx, gb, ppc_shape, FT_PPC, mtab->ppc_shape_cb,
                mtab->ppc_shape_cb + cb_len_p*PPC_SHAPE_CB_SIZE, cb_len_p);
    }

Vitor Sessak's avatar
Vitor Sessak committed
779
    for (i = 0; i < channels; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
780
        float *chunk = out + mtab->size * i;
Måns Rullgård's avatar
Måns Rullgård committed
781
        float lsp[LSP_COEFS_MAX];
Vitor Sessak's avatar
Vitor Sessak committed
782

Vitor Sessak's avatar
Vitor Sessak committed
783
        for (j = 0; j < sub; j++) {
Vitor Sessak's avatar
Vitor Sessak committed
784 785 786
            dec_bark_env(tctx, bark1[i][j], bark_use_hist[i][j], i,
                         tctx->tmp_buf, gain[sub*i+j], ftype);

787
            tctx->dsp.vector_fmul(chunk + block_size*j, chunk + block_size*j, tctx->tmp_buf,
Vitor Sessak's avatar
Vitor Sessak committed
788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807
                                  block_size);

        }

        if (ftype == FT_LONG) {
            float pgain_step = 25000. / ((1 << mtab->pgain_bit) - 1);
            int p_coef = get_bits(gb, tctx->mtab->ppc_period_bit);
            int g_coef = get_bits(gb, tctx->mtab->pgain_bit);
            float v = 1./8192*
                mulawinv(pgain_step*g_coef+ pgain_step/2, 25000., PGAIN_MU);

            decode_ppc(tctx, p_coef, ppc_shape + i*mtab->ppc_shape_len, v,
                       chunk);
        }

        decode_lsp(tctx, lpc_idx1[i], lpc_idx2[i], lpc_hist_idx[i], lsp,
                   tctx->lsp_hist[i]);

        dec_lpc_spectrum_inv(tctx, lsp, ftype, tctx->tmp_buf);

Vitor Sessak's avatar
Vitor Sessak committed
808
        for (j = 0; j < mtab->fmode[ftype].sub; j++) {
809
            tctx->dsp.vector_fmul(chunk, chunk, tctx->tmp_buf, block_size);
Vitor Sessak's avatar
Vitor Sessak committed
810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876
            chunk += block_size;
        }
    }
}

static int twin_decode_frame(AVCodecContext * avctx, void *data,
                             int *data_size, AVPacket *avpkt)
{
    const uint8_t *buf = avpkt->data;
    int buf_size = avpkt->size;
    TwinContext *tctx = avctx->priv_data;
    GetBitContext gb;
    const ModeTab *mtab = tctx->mtab;
    float *out = data;
    enum FrameType ftype;
    int window_type;
    static const enum FrameType wtype_to_ftype_table[] = {
        FT_LONG,   FT_LONG, FT_SHORT, FT_LONG,
        FT_MEDIUM, FT_LONG, FT_LONG,  FT_MEDIUM, FT_MEDIUM
    };

    if (buf_size*8 < avctx->bit_rate*mtab->size/avctx->sample_rate + 8) {
        av_log(avctx, AV_LOG_ERROR,
               "Frame too small (%d bytes). Truncated file?\n", buf_size);
        *data_size = 0;
        return buf_size;
    }

    init_get_bits(&gb, buf, buf_size * 8);
    skip_bits(&gb, get_bits(&gb, 8));
    window_type = get_bits(&gb, WINDOW_TYPE_BITS);

    if (window_type > 8) {
        av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
        return -1;
    }

    ftype = wtype_to_ftype_table[window_type];

    read_and_decode_spectrum(tctx, &gb, tctx->spectrum, ftype);

    imdct_output(tctx, ftype, window_type, out);

    FFSWAP(float*, tctx->curr_frame, tctx->prev_frame);

    if (tctx->avctx->frame_number < 2) {
        *data_size=0;
        return buf_size;
    }

    *data_size = mtab->size*avctx->channels*4;

    return buf_size;
}

/**
 * Init IMDCT and windowing tables
 */
static av_cold void init_mdct_win(TwinContext *tctx)
{
    int i,j;
    const ModeTab *mtab = tctx->mtab;
    int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;
    int size_m = mtab->size / mtab->fmode[FT_MEDIUM].sub;
    int channels = tctx->avctx->channels;
    float norm = channels == 1 ? 2. : 1.;

Vitor Sessak's avatar
Vitor Sessak committed
877
    for (i = 0; i < 3; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
878 879 880 881 882 883 884 885 886 887 888
        int bsize = tctx->mtab->size/tctx->mtab->fmode[i].sub;
        ff_mdct_init(&tctx->mdct_ctx[i], av_log2(bsize) + 1, 1,
                     -sqrt(norm/bsize) / (1<<15));
    }

    tctx->tmp_buf  = av_malloc(mtab->size            * sizeof(*tctx->tmp_buf));

    tctx->spectrum  = av_malloc(2*mtab->size*channels*sizeof(float));
    tctx->curr_frame = av_malloc(2*mtab->size*channels*sizeof(float));
    tctx->prev_frame  = av_malloc(2*mtab->size*channels*sizeof(float));

Vitor Sessak's avatar
Vitor Sessak committed
889
    for (i = 0; i < 3; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
890 891 892 893
        int m = 4*mtab->size/mtab->fmode[i].sub;
        double freq = 2*M_PI/m;
        tctx->cos_tabs[i] = av_malloc((m/4)*sizeof(*tctx->cos_tabs));

Vitor Sessak's avatar
Vitor Sessak committed
894
        for (j = 0; j <= m/8; j++)
Vitor Sessak's avatar
Vitor Sessak committed
895
            tctx->cos_tabs[i][j] = cos((2*j + 1)*freq);
Vitor Sessak's avatar
Vitor Sessak committed
896
        for (j = 1; j <  m/8; j++)
Vitor Sessak's avatar
Vitor Sessak committed
897 898 899 900
            tctx->cos_tabs[i][m/4-j] = tctx->cos_tabs[i][j];
    }


901 902 903
    ff_init_ff_sine_windows(av_log2(size_m));
    ff_init_ff_sine_windows(av_log2(size_s/2));
    ff_init_ff_sine_windows(av_log2(mtab->size));
Vitor Sessak's avatar
Vitor Sessak committed
904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919
}

/**
 * Interpret the data as if it were a num_blocks x line_len[0] matrix and for
 * each line do a cyclic permutation, i.e.
 * abcdefghijklm -> defghijklmabc
 * where the amount to be shifted is evaluated depending on the column.
 */
static void permutate_in_line(int16_t *tab, int num_vect, int num_blocks,
                              int block_size,
                              const uint8_t line_len[2], int length_div,
                              enum FrameType ftype)

{
    int i,j;

Vitor Sessak's avatar
Vitor Sessak committed
920
    for (i = 0; i < line_len[0]; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
921 922 923 924 925 926 927 928 929 930 931 932
        int shift;

        if (num_blocks == 1 ||
            (ftype == FT_LONG && num_vect % num_blocks) ||
            (ftype != FT_LONG && num_vect & 1         ) ||
            i == line_len[1]) {
            shift = 0;
        } else if (ftype == FT_LONG) {
            shift = i;
        } else
            shift = i*i;

Vitor Sessak's avatar
Vitor Sessak committed
933
        for (j = 0; j < num_vect && (j+num_vect*i < block_size*num_blocks); j++)
Vitor Sessak's avatar
Vitor Sessak committed
934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957
            tab[i*num_vect+j] = i*num_vect + (j + shift) % num_vect;
    }
}

/**
 * Interpret the input data as in the following table:
 *
 * \verbatim
 *
 * abcdefgh
 * ijklmnop
 * qrstuvw
 * x123456
 *
 * \endverbatim
 *
 * and transpose it, giving the output
 * aiqxbjr1cks2dlt3emu4fvn5gow6hp
 */
static void transpose_perm(int16_t *out, int16_t *in, int num_vect,
                           const uint8_t line_len[2], int length_div)
{
    int i,j;
    int cont= 0;
Vitor Sessak's avatar
Vitor Sessak committed
958 959
    for (i = 0; i < num_vect; i++)
        for (j = 0; j < line_len[i >= length_div]; j++)
Vitor Sessak's avatar
Vitor Sessak committed
960 961 962 963 964 965 966 967
            out[cont++] = in[j*num_vect + i];
}

static void linear_perm(int16_t *out, int16_t *in, int n_blocks, int size)
{
    int block_size = size/n_blocks;
    int i;

Vitor Sessak's avatar
Vitor Sessak committed
968
    for (i = 0; i < size; i++)
Vitor Sessak's avatar
Vitor Sessak committed
969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011
        out[i] = block_size * (in[i] % n_blocks) + in[i] / n_blocks;
}

static av_cold void construct_perm_table(TwinContext *tctx,enum FrameType ftype)
{
    int block_size;
    const ModeTab *mtab = tctx->mtab;
    int size = tctx->avctx->channels*mtab->fmode[ftype].sub;
    int16_t *tmp_perm = (int16_t *) tctx->tmp_buf;

    if (ftype == FT_PPC) {
        size  = tctx->avctx->channels;
        block_size = mtab->ppc_shape_len;
    } else
        block_size = mtab->size / mtab->fmode[ftype].sub;

    permutate_in_line(tmp_perm, tctx->n_div[ftype], size,
                      block_size, tctx->length[ftype],
                      tctx->length_change[ftype], ftype);

    transpose_perm(tctx->permut[ftype], tmp_perm, tctx->n_div[ftype],
                   tctx->length[ftype], tctx->length_change[ftype]);

    linear_perm(tctx->permut[ftype], tctx->permut[ftype], size,
                size*block_size);
}

static av_cold void init_bitstream_params(TwinContext *tctx)
{
    const ModeTab *mtab = tctx->mtab;
    int n_ch = tctx->avctx->channels;
    int total_fr_bits = tctx->avctx->bit_rate*mtab->size/
                             tctx->avctx->sample_rate;

    int lsp_bits_per_block = n_ch*(mtab->lsp_bit0 + mtab->lsp_bit1 +
                                   mtab->lsp_split*mtab->lsp_bit2);

    int ppc_bits = n_ch*(mtab->pgain_bit + mtab->ppc_shape_bit +
                         mtab->ppc_period_bit);

    int bsize_no_main_cb[3];
    int bse_bits[3];
    int i;
1012
    enum FrameType frametype;
Vitor Sessak's avatar
Vitor Sessak committed
1013

Vitor Sessak's avatar
Vitor Sessak committed
1014
    for (i = 0; i < 3; i++)
Vitor Sessak's avatar
Vitor Sessak committed
1015 1016 1017 1018 1019 1020 1021
        // +1 for history usage switch
        bse_bits[i] = n_ch *
            (mtab->fmode[i].bark_n_coef * mtab->fmode[i].bark_n_bit + 1);

    bsize_no_main_cb[2] = bse_bits[2] + lsp_bits_per_block + ppc_bits +
                          WINDOW_TYPE_BITS + n_ch*GAIN_BITS;

Vitor Sessak's avatar
Vitor Sessak committed
1022
    for (i = 0; i < 2; i++)
Vitor Sessak's avatar
Vitor Sessak committed
1023 1024 1025 1026 1027
        bsize_no_main_cb[i] =
            lsp_bits_per_block + n_ch*GAIN_BITS + WINDOW_TYPE_BITS +
            mtab->fmode[i].sub*(bse_bits[i] + n_ch*SUB_GAIN_BITS);

    // The remaining bits are all used for the main spectrum coefficients
Vitor Sessak's avatar
Vitor Sessak committed
1028
    for (i = 0; i < 4; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060
        int bit_size;
        int vect_size;
        int rounded_up, rounded_down, num_rounded_down, num_rounded_up;
        if (i == 3) {
            bit_size  = n_ch * mtab->ppc_shape_bit;
            vect_size = n_ch * mtab->ppc_shape_len;
        } else {
            bit_size = total_fr_bits - bsize_no_main_cb[i];
            vect_size = n_ch * mtab->size;
        }

        tctx->n_div[i] = (bit_size + 13) / 14;

        rounded_up   = (bit_size + tctx->n_div[i] - 1)/tctx->n_div[i];
        rounded_down = (bit_size           )/tctx->n_div[i];
        num_rounded_down = rounded_up * tctx->n_div[i] - bit_size;
        num_rounded_up = tctx->n_div[i] - num_rounded_down;
        tctx->bits_main_spec[0][i][0] = (rounded_up   + 1)/2;
        tctx->bits_main_spec[1][i][0] = (rounded_up      )/2;
        tctx->bits_main_spec[0][i][1] = (rounded_down + 1)/2;
        tctx->bits_main_spec[1][i][1] = (rounded_down    )/2;
        tctx->bits_main_spec_change[i] = num_rounded_up;

        rounded_up   = (vect_size + tctx->n_div[i] - 1)/tctx->n_div[i];
        rounded_down = (vect_size                     )/tctx->n_div[i];
        num_rounded_down = rounded_up * tctx->n_div[i] - vect_size;
        num_rounded_up = tctx->n_div[i] - num_rounded_down;
        tctx->length[i][0] = rounded_up;
        tctx->length[i][1] = rounded_down;
        tctx->length_change[i] = num_rounded_up;
    }

1061 1062
    for (frametype = FT_SHORT; frametype <= FT_PPC; frametype++)
        construct_perm_table(tctx, frametype);
Vitor Sessak's avatar
Vitor Sessak committed
1063 1064 1065 1066 1067 1068 1069 1070 1071
}

static av_cold int twin_decode_init(AVCodecContext *avctx)
{
    TwinContext *tctx = avctx->priv_data;
    int isampf = avctx->sample_rate/1000;
    int ibps = avctx->bit_rate/(1000 * avctx->channels);

    tctx->avctx       = avctx;
1072
    avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
Vitor Sessak's avatar
Vitor Sessak committed
1073

Måns Rullgård's avatar
Måns Rullgård committed
1074
    if (avctx->channels > CHANNELS_MAX) {
Vitor Sessak's avatar
Vitor Sessak committed
1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108
        av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n",
               avctx->channels);
        return -1;
    }

    switch ((isampf << 8) +  ibps) {
    case (8 <<8) +  8: tctx->mtab = &mode_08_08; break;
    case (11<<8) +  8: tctx->mtab = &mode_11_08; break;
    case (11<<8) + 10: tctx->mtab = &mode_11_10; break;
    case (16<<8) + 16: tctx->mtab = &mode_16_16; break;
    case (22<<8) + 20: tctx->mtab = &mode_22_20; break;
    case (22<<8) + 24: tctx->mtab = &mode_22_24; break;
    case (22<<8) + 32: tctx->mtab = &mode_22_32; break;
    case (44<<8) + 40: tctx->mtab = &mode_44_40; break;
    case (44<<8) + 48: tctx->mtab = &mode_44_48; break;
    default:
        av_log(avctx, AV_LOG_ERROR, "This version does not support %d kHz - %d kbit/s/ch mode.\n", isampf, isampf);
        return -1;
    }

    dsputil_init(&tctx->dsp, avctx);
    init_mdct_win(tctx);
    init_bitstream_params(tctx);

    memset_float(tctx->bark_hist[0][0], 0.1, FF_ARRAY_ELEMS(tctx->bark_hist));

    return 0;
}

static av_cold int twin_decode_close(AVCodecContext *avctx)
{
    TwinContext *tctx = avctx->priv_data;
    int i;

Vitor Sessak's avatar
Vitor Sessak committed
1109
    for (i = 0; i < 3; i++) {
Vitor Sessak's avatar
Vitor Sessak committed
1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122
        ff_mdct_end(&tctx->mdct_ctx[i]);
        av_free(tctx->cos_tabs[i]);
    }


    av_free(tctx->curr_frame);
    av_free(tctx->spectrum);
    av_free(tctx->prev_frame);
    av_free(tctx->tmp_buf);

    return 0;
}

1123
AVCodec ff_twinvq_decoder =
Vitor Sessak's avatar
Vitor Sessak committed
1124 1125
{
    "twinvq",
1126
    AVMEDIA_TYPE_AUDIO,
Vitor Sessak's avatar
Vitor Sessak committed
1127 1128 1129 1130 1131 1132 1133 1134
    CODEC_ID_TWINVQ,
    sizeof(TwinContext),
    twin_decode_init,
    NULL,
    twin_decode_close,
    twin_decode_frame,
    .long_name = NULL_IF_CONFIG_SMALL("VQF TwinVQ"),
};