yading@10: /*
yading@10:  * TwinVQ decoder
yading@10:  * Copyright (c) 2009 Vitor Sessak
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include "libavutil/channel_layout.h"
yading@10: #include "libavutil/float_dsp.h"
yading@10: #include "avcodec.h"
yading@10: #include "get_bits.h"
yading@10: #include "fft.h"
yading@10: #include "internal.h"
yading@10: #include "lsp.h"
yading@10: #include "sinewin.h"
yading@10: 
yading@10: #include <math.h>
yading@10: #include <stdint.h>
yading@10: 
yading@10: #include "twinvq_data.h"
yading@10: 
yading@10: enum FrameType {
yading@10:     FT_SHORT = 0,  ///< Short frame  (divided in n   sub-blocks)
yading@10:     FT_MEDIUM,     ///< Medium frame (divided in m<n sub-blocks)
yading@10:     FT_LONG,       ///< Long frame   (single sub-block + PPC)
yading@10:     FT_PPC,        ///< Periodic Peak Component (part of the long frame)
yading@10: };
yading@10: 
yading@10: /**
yading@10:  * Parameters and tables that are different for each frame type
yading@10:  */
yading@10: struct FrameMode {
yading@10:     uint8_t         sub;      ///< Number subblocks in each frame
yading@10:     const uint16_t *bark_tab;
yading@10: 
yading@10:     /** number of distinct bark scale envelope values */
yading@10:     uint8_t         bark_env_size;
yading@10: 
yading@10:     const int16_t  *bark_cb;    ///< codebook for the bark scale envelope (BSE)
yading@10:     uint8_t         bark_n_coef;///< number of BSE CB coefficients to read
yading@10:     uint8_t         bark_n_bit; ///< number of bits of the BSE coefs
yading@10: 
yading@10:     //@{
yading@10:     /** main codebooks for spectrum data */
yading@10:     const int16_t    *cb0;
yading@10:     const int16_t    *cb1;
yading@10:     //@}
yading@10: 
yading@10:     uint8_t         cb_len_read; ///< number of spectrum coefficients to read
yading@10: };
yading@10: 
yading@10: /**
yading@10:  * Parameters and tables that are different for every combination of
yading@10:  * bitrate/sample rate
yading@10:  */
yading@10: typedef struct {
yading@10:     struct FrameMode fmode[3]; ///< frame type-dependant parameters
yading@10: 
yading@10:     uint16_t     size;        ///< frame size in samples
yading@10:     uint8_t      n_lsp;       ///< number of lsp coefficients
yading@10:     const float *lspcodebook;
yading@10: 
yading@10:     /* number of bits of the different LSP CB coefficients */
yading@10:     uint8_t      lsp_bit0;
yading@10:     uint8_t      lsp_bit1;
yading@10:     uint8_t      lsp_bit2;
yading@10: 
yading@10:     uint8_t      lsp_split;      ///< number of CB entries for the LSP decoding
yading@10:     const int16_t *ppc_shape_cb; ///< PPC shape CB
yading@10: 
yading@10:     /** number of the bits for the PPC period value */
yading@10:     uint8_t      ppc_period_bit;
yading@10: 
yading@10:     uint8_t      ppc_shape_bit;  ///< number of bits of the PPC shape CB coeffs
yading@10:     uint8_t      ppc_shape_len;  ///< size of PPC shape CB
yading@10:     uint8_t      pgain_bit;      ///< bits for PPC gain
yading@10: 
yading@10:     /** constant for peak period to peak width conversion */
yading@10:     uint16_t     peak_per2wid;
yading@10: } ModeTab;
yading@10: 
yading@10: static const ModeTab mode_08_08 = {
yading@10:     {
yading@10:         { 8, bark_tab_s08_64,  10, tab.fcb08s  , 1, 5, tab.cb0808s0, tab.cb0808s1, 18},
yading@10:         { 2, bark_tab_m08_256, 20, tab.fcb08m  , 2, 5, tab.cb0808m0, tab.cb0808m1, 16},
yading@10:         { 1, bark_tab_l08_512, 30, tab.fcb08l  , 3, 6, tab.cb0808l0, tab.cb0808l1, 17}
yading@10:     },
yading@10:     512 , 12, tab.lsp08,   1, 5, 3, 3, tab.shape08  , 8, 28, 20, 6, 40
yading@10: };
yading@10: 
yading@10: static const ModeTab mode_11_08 = {
yading@10:     {
yading@10:         { 8, bark_tab_s11_64,  10, tab.fcb11s  , 1, 5, tab.cb1108s0, tab.cb1108s1, 29},
yading@10:         { 2, bark_tab_m11_256, 20, tab.fcb11m  , 2, 5, tab.cb1108m0, tab.cb1108m1, 24},
yading@10:         { 1, bark_tab_l11_512, 30, tab.fcb11l  , 3, 6, tab.cb1108l0, tab.cb1108l1, 27}
yading@10:     },
yading@10:     512 , 16, tab.lsp11,   1, 6, 4, 3, tab.shape11  , 9, 36, 30, 7, 90
yading@10: };
yading@10: 
yading@10: static const ModeTab mode_11_10 = {
yading@10:     {
yading@10:         { 8, bark_tab_s11_64,  10, tab.fcb11s  , 1, 5, tab.cb1110s0, tab.cb1110s1, 21},
yading@10:         { 2, bark_tab_m11_256, 20, tab.fcb11m  , 2, 5, tab.cb1110m0, tab.cb1110m1, 18},
yading@10:         { 1, bark_tab_l11_512, 30, tab.fcb11l  , 3, 6, tab.cb1110l0, tab.cb1110l1, 20}
yading@10:     },
yading@10:     512 , 16, tab.lsp11,   1, 6, 4, 3, tab.shape11  , 9, 36, 30, 7, 90
yading@10: };
yading@10: 
yading@10: static const ModeTab mode_16_16 = {
yading@10:     {
yading@10:         { 8, bark_tab_s16_128, 10, tab.fcb16s  , 1, 5, tab.cb1616s0, tab.cb1616s1, 16},
yading@10:         { 2, bark_tab_m16_512, 20, tab.fcb16m  , 2, 5, tab.cb1616m0, tab.cb1616m1, 15},
yading@10:         { 1, bark_tab_l16_1024,30, tab.fcb16l  , 3, 6, tab.cb1616l0, tab.cb1616l1, 16}
yading@10:     },
yading@10:     1024, 16, tab.lsp16,   1, 6, 4, 3, tab.shape16  , 9, 56, 60, 7, 180
yading@10: };
yading@10: 
yading@10: static const ModeTab mode_22_20 = {
yading@10:     {
yading@10:         { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2220s0, tab.cb2220s1, 18},
yading@10:         { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2220m0, tab.cb2220m1, 17},
yading@10:         { 1, bark_tab_l22_1024,32, tab.fcb22l_1, 4, 6, tab.cb2220l0, tab.cb2220l1, 18}
yading@10:     },
yading@10:     1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
yading@10: };
yading@10: 
yading@10: static const ModeTab mode_22_24 = {
yading@10:     {
yading@10:         { 8, bark_tab_s22_128, 10, tab.fcb22s_1, 1, 6, tab.cb2224s0, tab.cb2224s1, 15},
yading@10:         { 2, bark_tab_m22_512, 20, tab.fcb22m_1, 2, 6, tab.cb2224m0, tab.cb2224m1, 14},
yading@10:         { 1, bark_tab_l22_1024,32, tab.fcb22l_1, 4, 6, tab.cb2224l0, tab.cb2224l1, 15}
yading@10:     },
yading@10:     1024, 16, tab.lsp22_1, 1, 6, 4, 3, tab.shape22_1, 9, 56, 36, 7, 144
yading@10: };
yading@10: 
yading@10: static const ModeTab mode_22_32 = {
yading@10:     {
yading@10:         { 4, bark_tab_s22_128, 10, tab.fcb22s_2, 1, 6, tab.cb2232s0, tab.cb2232s1, 11},
yading@10:         { 2, bark_tab_m22_256, 20, tab.fcb22m_2, 2, 6, tab.cb2232m0, tab.cb2232m1, 11},
yading@10:         { 1, bark_tab_l22_512, 32, tab.fcb22l_2, 4, 6, tab.cb2232l0, tab.cb2232l1, 12}
yading@10:     },
yading@10:     512 , 16, tab.lsp22_2, 1, 6, 4, 4, tab.shape22_2, 9, 56, 36, 7, 72
yading@10: };
yading@10: 
yading@10: static const ModeTab mode_44_40 = {
yading@10:     {
yading@10:         {16, bark_tab_s44_128, 10, tab.fcb44s  , 1, 6, tab.cb4440s0, tab.cb4440s1, 18},
yading@10:         { 4, bark_tab_m44_512, 20, tab.fcb44m  , 2, 6, tab.cb4440m0, tab.cb4440m1, 17},
yading@10:         { 1, bark_tab_l44_2048,40, tab.fcb44l  , 4, 6, tab.cb4440l0, tab.cb4440l1, 17}
yading@10:     },
yading@10:     2048, 20, tab.lsp44,   1, 6, 4, 4, tab.shape44  , 9, 84, 54, 7, 432
yading@10: };
yading@10: 
yading@10: static const ModeTab mode_44_48 = {
yading@10:     {
yading@10:         {16, bark_tab_s44_128, 10, tab.fcb44s  , 1, 6, tab.cb4448s0, tab.cb4448s1, 15},
yading@10:         { 4, bark_tab_m44_512, 20, tab.fcb44m  , 2, 6, tab.cb4448m0, tab.cb4448m1, 14},
yading@10:         { 1, bark_tab_l44_2048,40, tab.fcb44l  , 4, 6, tab.cb4448l0, tab.cb4448l1, 14}
yading@10:     },
yading@10:     2048, 20, tab.lsp44,   1, 6, 4, 4, tab.shape44  , 9, 84, 54, 7, 432
yading@10: };
yading@10: 
yading@10: typedef struct TwinContext {
yading@10:     AVCodecContext *avctx;
yading@10:     AVFloatDSPContext fdsp;
yading@10:     FFTContext mdct_ctx[3];
yading@10: 
yading@10:     const ModeTab *mtab;
yading@10: 
yading@10:     // history
yading@10:     float lsp_hist[2][20];           ///< LSP coefficients of the last frame
yading@10:     float bark_hist[3][2][40];       ///< BSE coefficients of last frame
yading@10: 
yading@10:     // bitstream parameters
yading@10:     int16_t permut[4][4096];
yading@10:     uint8_t length[4][2];            ///< main codebook stride
yading@10:     uint8_t length_change[4];
yading@10:     uint8_t bits_main_spec[2][4][2]; ///< bits for the main codebook
yading@10:     int bits_main_spec_change[4];
yading@10:     int n_div[4];
yading@10: 
yading@10:     float *spectrum;
yading@10:     float *curr_frame;               ///< non-interleaved output
yading@10:     float *prev_frame;               ///< non-interleaved previous frame
yading@10:     int last_block_pos[2];
yading@10:     int discarded_packets;
yading@10: 
yading@10:     float *cos_tabs[3];
yading@10: 
yading@10:     // scratch buffers
yading@10:     float *tmp_buf;
yading@10: } TwinContext;
yading@10: 
yading@10: #define PPC_SHAPE_CB_SIZE 64
yading@10: #define PPC_SHAPE_LEN_MAX 60
yading@10: #define SUB_AMP_MAX       4500.0
yading@10: #define MULAW_MU          100.0
yading@10: #define GAIN_BITS         8
yading@10: #define AMP_MAX           13000.0
yading@10: #define SUB_GAIN_BITS     5
yading@10: #define WINDOW_TYPE_BITS  4
yading@10: #define PGAIN_MU          200
yading@10: #define LSP_COEFS_MAX     20
yading@10: #define LSP_SPLIT_MAX     4
yading@10: #define CHANNELS_MAX      2
yading@10: #define SUBBLOCKS_MAX     16
yading@10: #define BARK_N_COEF_MAX   4
yading@10: 
yading@10: /** @note not speed critical, hence not optimized */
yading@10: static void memset_float(float *buf, float val, int size)
yading@10: {
yading@10:     while (size--)
yading@10:         *buf++ = val;
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Evaluate a single LPC amplitude spectrum envelope coefficient from the line
yading@10:  * spectrum pairs.
yading@10:  *
yading@10:  * @param lsp a vector of the cosinus of the LSP values
yading@10:  * @param cos_val cos(PI*i/N) where i is the index of the LPC amplitude
yading@10:  * @param order the order of the LSP (and the size of the *lsp buffer). Must
yading@10:  *        be a multiple of four.
yading@10:  * @return the LPC value
yading@10:  *
yading@10:  * @todo reuse code from Vorbis decoder: vorbis_floor0_decode
yading@10:  */
yading@10: static float eval_lpc_spectrum(const float *lsp, float cos_val, int order)
yading@10: {
yading@10:     int j;
yading@10:     float p = 0.5f;
yading@10:     float q = 0.5f;
yading@10:     float two_cos_w = 2.0f*cos_val;
yading@10: 
yading@10:     for (j = 0; j + 1 < order; j += 2*2) {
yading@10:         // Unroll the loop once since order is a multiple of four
yading@10:         q *= lsp[j  ] - two_cos_w;
yading@10:         p *= lsp[j+1] - two_cos_w;
yading@10: 
yading@10:         q *= lsp[j+2] - two_cos_w;
yading@10:         p *= lsp[j+3] - two_cos_w;
yading@10:     }
yading@10: 
yading@10:     p *= p * (2.0f - two_cos_w);
yading@10:     q *= q * (2.0f + two_cos_w);
yading@10: 
yading@10:     return 0.5 / (p + q);
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Evaluate the LPC amplitude spectrum envelope from the line spectrum pairs.
yading@10:  */
yading@10: static void eval_lpcenv(TwinContext *tctx, const float *cos_vals, float *lpc)
yading@10: {
yading@10:     int i;
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;
yading@10: 
yading@10:     for (i = 0; i < size_s/2; i++) {
yading@10:         float cos_i = tctx->cos_tabs[0][i];
yading@10:         lpc[i]          = eval_lpc_spectrum(cos_vals,  cos_i, mtab->n_lsp);
yading@10:         lpc[size_s-i-1] = eval_lpc_spectrum(cos_vals, -cos_i, mtab->n_lsp);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void interpolate(float *out, float v1, float v2, int size)
yading@10: {
yading@10:     int i;
yading@10:     float step = (v1 - v2)/(size + 1);
yading@10: 
yading@10:     for (i = 0; i < size; i++) {
yading@10:         v2 += step;
yading@10:         out[i] = v2;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static inline float get_cos(int idx, int part, const float *cos_tab, int size)
yading@10: {
yading@10:     return part ? -cos_tab[size - idx - 1] :
yading@10:                    cos_tab[       idx    ];
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Evaluate the LPC amplitude spectrum envelope from the line spectrum pairs.
yading@10:  * Probably for speed reasons, the coefficients are evaluated as
yading@10:  * siiiibiiiisiiiibiiiisiiiibiiiisiiiibiiiis ...
yading@10:  * where s is an evaluated value, i is a value interpolated from the others
yading@10:  * and b might be either calculated or interpolated, depending on an
yading@10:  * unexplained condition.
yading@10:  *
yading@10:  * @param step the size of a block "siiiibiiii"
yading@10:  * @param in the cosinus of the LSP data
yading@10:  * @param part is 0 for 0...PI (positive cossinus values) and 1 for PI...2PI
yading@10:           (negative cossinus values)
yading@10:  * @param size the size of the whole output
yading@10:  */
yading@10: static inline void eval_lpcenv_or_interp(TwinContext *tctx,
yading@10:                                          enum FrameType ftype,
yading@10:                                          float *out, const float *in,
yading@10:                                          int size, int step, int part)
yading@10: {
yading@10:     int i;
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     const float *cos_tab = tctx->cos_tabs[ftype];
yading@10: 
yading@10:     // Fill the 's'
yading@10:     for (i = 0; i < size; i += step)
yading@10:         out[i] =
yading@10:             eval_lpc_spectrum(in,
yading@10:                               get_cos(i, part, cos_tab, size),
yading@10:                               mtab->n_lsp);
yading@10: 
yading@10:     // Fill the 'iiiibiiii'
yading@10:     for (i = step; i <= size - 2*step; i += step) {
yading@10:         if (out[i + step] + out[i - step] >  1.95*out[i] ||
yading@10:             out[i + step]                 >=  out[i - step]) {
yading@10:             interpolate(out + i - step + 1, out[i], out[i-step], step - 1);
yading@10:         } else {
yading@10:             out[i - step/2] =
yading@10:                 eval_lpc_spectrum(in,
yading@10:                                   get_cos(i-step/2, part, cos_tab, size),
yading@10:                                   mtab->n_lsp);
yading@10:             interpolate(out + i - step   + 1, out[i-step/2], out[i-step  ], step/2 - 1);
yading@10:             interpolate(out + i - step/2 + 1, out[i       ], out[i-step/2], step/2 - 1);
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     interpolate(out + size - 2*step + 1, out[size-step], out[size - 2*step], step - 1);
yading@10: }
yading@10: 
yading@10: static void eval_lpcenv_2parts(TwinContext *tctx, enum FrameType ftype,
yading@10:                                const float *buf, float *lpc,
yading@10:                                int size, int step)
yading@10: {
yading@10:     eval_lpcenv_or_interp(tctx, ftype, lpc         , buf, size/2,   step, 0);
yading@10:     eval_lpcenv_or_interp(tctx, ftype, lpc + size/2, buf, size/2, 2*step, 1);
yading@10: 
yading@10:     interpolate(lpc+size/2-step+1, lpc[size/2], lpc[size/2-step], step);
yading@10: 
yading@10:     memset_float(lpc + size - 2*step + 1, lpc[size - 2*step], 2*step - 1);
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Inverse quantization. Read CB coefficients for cb1 and cb2 from the
yading@10:  * bitstream, sum the corresponding vectors and write the result to *out
yading@10:  * after permutation.
yading@10:  */
yading@10: static void dequant(TwinContext *tctx, GetBitContext *gb, float *out,
yading@10:                     enum FrameType ftype,
yading@10:                     const int16_t *cb0, const int16_t *cb1, int cb_len)
yading@10: {
yading@10:     int pos = 0;
yading@10:     int i, j;
yading@10: 
yading@10:     for (i = 0; i < tctx->n_div[ftype]; i++) {
yading@10:         int tmp0, tmp1;
yading@10:         int sign0 = 1;
yading@10:         int sign1 = 1;
yading@10:         const int16_t *tab0, *tab1;
yading@10:         int length = tctx->length[ftype][i >= tctx->length_change[ftype]];
yading@10:         int bitstream_second_part = (i >= tctx->bits_main_spec_change[ftype]);
yading@10: 
yading@10:         int bits = tctx->bits_main_spec[0][ftype][bitstream_second_part];
yading@10:         if (bits == 7) {
yading@10:             if (get_bits1(gb))
yading@10:                 sign0 = -1;
yading@10:             bits = 6;
yading@10:         }
yading@10:         tmp0 = get_bits(gb, bits);
yading@10: 
yading@10:         bits = tctx->bits_main_spec[1][ftype][bitstream_second_part];
yading@10: 
yading@10:         if (bits == 7) {
yading@10:             if (get_bits1(gb))
yading@10:                 sign1 = -1;
yading@10: 
yading@10:             bits = 6;
yading@10:         }
yading@10:         tmp1 = get_bits(gb, bits);
yading@10: 
yading@10:         tab0 = cb0 + tmp0*cb_len;
yading@10:         tab1 = cb1 + tmp1*cb_len;
yading@10: 
yading@10:         for (j = 0; j < length; j++)
yading@10:             out[tctx->permut[ftype][pos+j]] = sign0*tab0[j] + sign1*tab1[j];
yading@10: 
yading@10:         pos += length;
yading@10:     }
yading@10: 
yading@10: }
yading@10: 
yading@10: static inline float mulawinv(float y, float clip, float mu)
yading@10: {
yading@10:     y = av_clipf(y/clip, -1, 1);
yading@10:     return clip * FFSIGN(y) * (exp(log(1+mu) * fabs(y)) - 1) / mu;
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Evaluate a*b/400 rounded to the nearest integer. When, for example,
yading@10:  * a*b == 200 and the nearest integer is ill-defined, use a table to emulate
yading@10:  * the following broken float-based implementation used by the binary decoder:
yading@10:  *
yading@10:  * @code
yading@10:  * static int very_broken_op(int a, int b)
yading@10:  * {
yading@10:  *    static float test; // Ugh, force gcc to do the division first...
yading@10:  *
yading@10:  *    test = a/400.;
yading@10:  *    return b * test +  0.5;
yading@10:  * }
yading@10:  * @endcode
yading@10:  *
yading@10:  * @note if this function is replaced by just ROUNDED_DIV(a*b,400.), the stddev
yading@10:  * between the original file (before encoding with Yamaha encoder) and the
yading@10:  * decoded output increases, which leads one to believe that the encoder expects
yading@10:  * exactly this broken calculation.
yading@10:  */
yading@10: static int very_broken_op(int a, int b)
yading@10: {
yading@10:     int x = a*b + 200;
yading@10:     int size;
yading@10:     const uint8_t *rtab;
yading@10: 
yading@10:     if (x%400 || b%5)
yading@10:         return x/400;
yading@10: 
yading@10:     x /= 400;
yading@10: 
yading@10:     size = tabs[b/5].size;
yading@10:     rtab = tabs[b/5].tab;
yading@10:     return x - rtab[size*av_log2(2*(x - 1)/size)+(x - 1)%size];
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Sum to data a periodic peak of a given period, width and shape.
yading@10:  *
yading@10:  * @param period the period of the peak divised by 400.0
yading@10:  */
yading@10: static void add_peak(int period, int width, const float *shape,
yading@10:                      float ppc_gain, float *speech, int len)
yading@10: {
yading@10:     int i, j;
yading@10: 
yading@10:     const float *shape_end = shape + len;
yading@10:     int center;
yading@10: 
yading@10:     // First peak centered around zero
yading@10:     for (i = 0; i < width/2; i++)
yading@10:         speech[i] += ppc_gain * *shape++;
yading@10: 
yading@10:     for (i = 1; i < ROUNDED_DIV(len,width) ; i++) {
yading@10:         center = very_broken_op(period, i);
yading@10:         for (j = -width/2; j < (width+1)/2; j++)
yading@10:             speech[j+center] += ppc_gain * *shape++;
yading@10:     }
yading@10: 
yading@10:     // For the last block, be careful not to go beyond the end of the buffer
yading@10:     center = very_broken_op(period, i);
yading@10:     for (j = -width/2; j < (width + 1)/2 && shape < shape_end; j++)
yading@10:         speech[j+center] += ppc_gain * *shape++;
yading@10: }
yading@10: 
yading@10: static void decode_ppc(TwinContext *tctx, int period_coef, const float *shape,
yading@10:                        float ppc_gain, float *speech)
yading@10: {
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     int isampf = tctx->avctx->sample_rate/1000;
yading@10:     int ibps = tctx->avctx->bit_rate/(1000 * tctx->avctx->channels);
yading@10:     int min_period = ROUNDED_DIV(  40*2*mtab->size, isampf);
yading@10:     int max_period = ROUNDED_DIV(6*40*2*mtab->size, isampf);
yading@10:     int period_range = max_period - min_period;
yading@10: 
yading@10:     // This is actually the period multiplied by 400. It is just linearly coded
yading@10:     // between its maximum and minimum value.
yading@10:     int period = min_period +
yading@10:         ROUNDED_DIV(period_coef*period_range, (1 << mtab->ppc_period_bit) - 1);
yading@10:     int width;
yading@10: 
yading@10:     if (isampf == 22 && ibps == 32) {
yading@10:         // For some unknown reason, NTT decided to code this case differently...
yading@10:         width = ROUNDED_DIV((period + 800)* mtab->peak_per2wid, 400*mtab->size);
yading@10:     } else
yading@10:         width =             (period      )* mtab->peak_per2wid/(400*mtab->size);
yading@10: 
yading@10:     add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len);
yading@10: }
yading@10: 
yading@10: static void dec_gain(TwinContext *tctx, GetBitContext *gb, enum FrameType ftype,
yading@10:                      float *out)
yading@10: {
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     int i, j;
yading@10:     int sub = mtab->fmode[ftype].sub;
yading@10:     float step     = AMP_MAX     / ((1 <<     GAIN_BITS) - 1);
yading@10:     float sub_step = SUB_AMP_MAX / ((1 << SUB_GAIN_BITS) - 1);
yading@10: 
yading@10:     if (ftype == FT_LONG) {
yading@10:         for (i = 0; i < tctx->avctx->channels; i++)
yading@10:             out[i] = (1./(1<<13)) *
yading@10:                 mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
yading@10:                          AMP_MAX, MULAW_MU);
yading@10:     } else {
yading@10:         for (i = 0; i < tctx->avctx->channels; i++) {
yading@10:             float val = (1./(1<<23)) *
yading@10:                 mulawinv(step * 0.5 + step * get_bits(gb, GAIN_BITS),
yading@10:                          AMP_MAX, MULAW_MU);
yading@10: 
yading@10:             for (j = 0; j < sub; j++) {
yading@10:                 out[i*sub + j] =
yading@10:                     val*mulawinv(sub_step* 0.5 +
yading@10:                                  sub_step* get_bits(gb, SUB_GAIN_BITS),
yading@10:                                  SUB_AMP_MAX, MULAW_MU);
yading@10:             }
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Rearrange the LSP coefficients so that they have a minimum distance of
yading@10:  * min_dist. This function does it exactly as described in section of 3.2.4
yading@10:  * of the G.729 specification (but interestingly is different from what the
yading@10:  * reference decoder actually does).
yading@10:  */
yading@10: static void rearrange_lsp(int order, float *lsp, float min_dist)
yading@10: {
yading@10:     int i;
yading@10:     float min_dist2 = min_dist * 0.5;
yading@10:     for (i = 1; i < order; i++)
yading@10:         if (lsp[i] - lsp[i-1] < min_dist) {
yading@10:             float avg = (lsp[i] + lsp[i-1]) * 0.5;
yading@10: 
yading@10:             lsp[i-1] = avg - min_dist2;
yading@10:             lsp[i  ] = avg + min_dist2;
yading@10:         }
yading@10: }
yading@10: 
yading@10: static void decode_lsp(TwinContext *tctx, int lpc_idx1, uint8_t *lpc_idx2,
yading@10:                        int lpc_hist_idx, float *lsp, float *hist)
yading@10: {
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     int i, j;
yading@10: 
yading@10:     const float *cb  =  mtab->lspcodebook;
yading@10:     const float *cb2 =  cb  + (1 << mtab->lsp_bit1)*mtab->n_lsp;
yading@10:     const float *cb3 =  cb2 + (1 << mtab->lsp_bit2)*mtab->n_lsp;
yading@10: 
yading@10:     const int8_t funny_rounding[4] = {
yading@10:         -2,
yading@10:         mtab->lsp_split == 4 ? -2 : 1,
yading@10:         mtab->lsp_split == 4 ? -2 : 1,
yading@10:         0
yading@10:     };
yading@10: 
yading@10:     j = 0;
yading@10:     for (i = 0; i < mtab->lsp_split; i++) {
yading@10:         int chunk_end = ((i + 1)*mtab->n_lsp + funny_rounding[i])/mtab->lsp_split;
yading@10:         for (; j < chunk_end; j++)
yading@10:             lsp[j] = cb [lpc_idx1    * mtab->n_lsp + j] +
yading@10:                      cb2[lpc_idx2[i] * mtab->n_lsp + j];
yading@10:     }
yading@10: 
yading@10:     rearrange_lsp(mtab->n_lsp, lsp, 0.0001);
yading@10: 
yading@10:     for (i = 0; i < mtab->n_lsp; i++) {
yading@10:         float tmp1 = 1. -          cb3[lpc_hist_idx*mtab->n_lsp + i];
yading@10:         float tmp2 =     hist[i] * cb3[lpc_hist_idx*mtab->n_lsp + i];
yading@10:         hist[i] = lsp[i];
yading@10:         lsp[i]  = lsp[i] * tmp1 + tmp2;
yading@10:     }
yading@10: 
yading@10:     rearrange_lsp(mtab->n_lsp, lsp, 0.0001);
yading@10:     rearrange_lsp(mtab->n_lsp, lsp, 0.000095);
yading@10:     ff_sort_nearly_sorted_floats(lsp, mtab->n_lsp);
yading@10: }
yading@10: 
yading@10: static void dec_lpc_spectrum_inv(TwinContext *tctx, float *lsp,
yading@10:                                  enum FrameType ftype, float *lpc)
yading@10: {
yading@10:     int i;
yading@10:     int size = tctx->mtab->size / tctx->mtab->fmode[ftype].sub;
yading@10: 
yading@10:     for (i = 0; i < tctx->mtab->n_lsp; i++)
yading@10:         lsp[i] =  2*cos(lsp[i]);
yading@10: 
yading@10:     switch (ftype) {
yading@10:     case FT_LONG:
yading@10:         eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 8);
yading@10:         break;
yading@10:     case FT_MEDIUM:
yading@10:         eval_lpcenv_2parts(tctx, ftype, lsp, lpc, size, 2);
yading@10:         break;
yading@10:     case FT_SHORT:
yading@10:         eval_lpcenv(tctx, lsp, lpc);
yading@10:         break;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void imdct_and_window(TwinContext *tctx, enum FrameType ftype, int wtype,
yading@10:                             float *in, float *prev, int ch)
yading@10: {
yading@10:     FFTContext *mdct = &tctx->mdct_ctx[ftype];
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     int bsize = mtab->size / mtab->fmode[ftype].sub;
yading@10:     int size  = mtab->size;
yading@10:     float *buf1 = tctx->tmp_buf;
yading@10:     int j;
yading@10:     int wsize; // Window size
yading@10:     float *out = tctx->curr_frame + 2*ch*mtab->size;
yading@10:     float *out2 = out;
yading@10:     float *prev_buf;
yading@10:     int first_wsize;
yading@10: 
yading@10:     static const uint8_t wtype_to_wsize[]      = {0, 0, 2, 2, 2, 1, 0, 1, 1};
yading@10:     int types_sizes[] = {
yading@10:         mtab->size /    mtab->fmode[FT_LONG  ].sub,
yading@10:         mtab->size /    mtab->fmode[FT_MEDIUM].sub,
yading@10:         mtab->size / (2*mtab->fmode[FT_SHORT ].sub),
yading@10:     };
yading@10: 
yading@10:     wsize = types_sizes[wtype_to_wsize[wtype]];
yading@10:     first_wsize = wsize;
yading@10:     prev_buf = prev + (size - bsize)/2;
yading@10: 
yading@10:     for (j = 0; j < mtab->fmode[ftype].sub; j++) {
yading@10:         int sub_wtype = ftype == FT_MEDIUM ? 8 : wtype;
yading@10: 
yading@10:         if (!j && wtype == 4)
yading@10:             sub_wtype = 4;
yading@10:         else if (j == mtab->fmode[ftype].sub-1 && wtype == 7)
yading@10:             sub_wtype = 7;
yading@10: 
yading@10:         wsize = types_sizes[wtype_to_wsize[sub_wtype]];
yading@10: 
yading@10:         mdct->imdct_half(mdct, buf1 + bsize*j, in + bsize*j);
yading@10: 
yading@10:         tctx->fdsp.vector_fmul_window(out2, prev_buf + (bsize-wsize) / 2,
yading@10:                                       buf1 + bsize * j,
yading@10:                                       ff_sine_windows[av_log2(wsize)],
yading@10:                                       wsize / 2);
yading@10:         out2 += wsize;
yading@10: 
yading@10:         memcpy(out2, buf1 + bsize*j + wsize/2, (bsize - wsize/2)*sizeof(float));
yading@10: 
yading@10:         out2 += ftype == FT_MEDIUM ? (bsize-wsize)/2 : bsize - wsize;
yading@10: 
yading@10:         prev_buf = buf1 + bsize*j + bsize/2;
yading@10:     }
yading@10: 
yading@10:     tctx->last_block_pos[ch] = (size + first_wsize)/2;
yading@10: }
yading@10: 
yading@10: static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype,
yading@10:                          float **out)
yading@10: {
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     int size1, size2;
yading@10:     float *prev_buf = tctx->prev_frame + tctx->last_block_pos[0];
yading@10:     int i;
yading@10: 
yading@10:     for (i = 0; i < tctx->avctx->channels; i++) {
yading@10:         imdct_and_window(tctx, ftype, wtype,
yading@10:                          tctx->spectrum + i*mtab->size,
yading@10:                          prev_buf + 2*i*mtab->size,
yading@10:                          i);
yading@10:     }
yading@10: 
yading@10:     if (!out)
yading@10:         return;
yading@10: 
yading@10:     size2 = tctx->last_block_pos[0];
yading@10:     size1 = mtab->size - size2;
yading@10: 
yading@10:     memcpy(&out[0][0    ], prev_buf,         size1 * sizeof(out[0][0]));
yading@10:     memcpy(&out[0][size1], tctx->curr_frame, size2 * sizeof(out[0][0]));
yading@10: 
yading@10:     if (tctx->avctx->channels == 2) {
yading@10:         memcpy(&out[1][0],     &prev_buf[2*mtab->size],         size1 * sizeof(out[1][0]));
yading@10:         memcpy(&out[1][size1], &tctx->curr_frame[2*mtab->size], size2 * sizeof(out[1][0]));
yading@10:         tctx->fdsp.butterflies_float(out[0], out[1], mtab->size);
yading@10:     }
yading@10: }
yading@10: 
yading@10: static void dec_bark_env(TwinContext *tctx, const uint8_t *in, int use_hist,
yading@10:                          int ch, float *out, float gain, enum FrameType ftype)
yading@10: {
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     int i,j;
yading@10:     float *hist = tctx->bark_hist[ftype][ch];
yading@10:     float val = ((const float []) {0.4, 0.35, 0.28})[ftype];
yading@10:     int bark_n_coef  = mtab->fmode[ftype].bark_n_coef;
yading@10:     int fw_cb_len = mtab->fmode[ftype].bark_env_size / bark_n_coef;
yading@10:     int idx = 0;
yading@10: 
yading@10:     for (i = 0; i < fw_cb_len; i++)
yading@10:         for (j = 0; j < bark_n_coef; j++, idx++) {
yading@10:             float tmp2 =
yading@10:                 mtab->fmode[ftype].bark_cb[fw_cb_len*in[j] + i] * (1./4096);
yading@10:             float st = use_hist ?
yading@10:                 (1. - val) * tmp2 + val*hist[idx] + 1. : tmp2 + 1.;
yading@10: 
yading@10:             hist[idx] = tmp2;
yading@10:             if (st < -1.) st = 1.;
yading@10: 
yading@10:             memset_float(out, st * gain, mtab->fmode[ftype].bark_tab[idx]);
yading@10:             out += mtab->fmode[ftype].bark_tab[idx];
yading@10:         }
yading@10: 
yading@10: }
yading@10: 
yading@10: static void read_and_decode_spectrum(TwinContext *tctx, GetBitContext *gb,
yading@10:                                      float *out, enum FrameType ftype)
yading@10: {
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     int channels = tctx->avctx->channels;
yading@10:     int sub = mtab->fmode[ftype].sub;
yading@10:     int block_size = mtab->size / sub;
yading@10:     float gain[CHANNELS_MAX*SUBBLOCKS_MAX];
yading@10:     float ppc_shape[PPC_SHAPE_LEN_MAX * CHANNELS_MAX * 4];
yading@10:     uint8_t bark1[CHANNELS_MAX][SUBBLOCKS_MAX][BARK_N_COEF_MAX];
yading@10:     uint8_t bark_use_hist[CHANNELS_MAX][SUBBLOCKS_MAX];
yading@10: 
yading@10:     uint8_t lpc_idx1[CHANNELS_MAX];
yading@10:     uint8_t lpc_idx2[CHANNELS_MAX][LSP_SPLIT_MAX];
yading@10:     uint8_t lpc_hist_idx[CHANNELS_MAX];
yading@10: 
yading@10:     int i, j, k;
yading@10: 
yading@10:     dequant(tctx, gb, out, ftype,
yading@10:             mtab->fmode[ftype].cb0, mtab->fmode[ftype].cb1,
yading@10:             mtab->fmode[ftype].cb_len_read);
yading@10: 
yading@10:     for (i = 0; i < channels; i++)
yading@10:         for (j = 0; j < sub; j++)
yading@10:             for (k = 0; k < mtab->fmode[ftype].bark_n_coef; k++)
yading@10:                 bark1[i][j][k] =
yading@10:                     get_bits(gb, mtab->fmode[ftype].bark_n_bit);
yading@10: 
yading@10:     for (i = 0; i < channels; i++)
yading@10:         for (j = 0; j < sub; j++)
yading@10:             bark_use_hist[i][j] = get_bits1(gb);
yading@10: 
yading@10:     dec_gain(tctx, gb, ftype, gain);
yading@10: 
yading@10:     for (i = 0; i < channels; i++) {
yading@10:         lpc_hist_idx[i] = get_bits(gb, tctx->mtab->lsp_bit0);
yading@10:         lpc_idx1    [i] = get_bits(gb, tctx->mtab->lsp_bit1);
yading@10: 
yading@10:         for (j = 0; j < tctx->mtab->lsp_split; j++)
yading@10:             lpc_idx2[i][j] = get_bits(gb, tctx->mtab->lsp_bit2);
yading@10:     }
yading@10: 
yading@10:     if (ftype == FT_LONG) {
yading@10:         int cb_len_p = (tctx->n_div[3] + mtab->ppc_shape_len*channels - 1)/
yading@10:             tctx->n_div[3];
yading@10:         dequant(tctx, gb, ppc_shape, FT_PPC, mtab->ppc_shape_cb,
yading@10:                 mtab->ppc_shape_cb + cb_len_p*PPC_SHAPE_CB_SIZE, cb_len_p);
yading@10:     }
yading@10: 
yading@10:     for (i = 0; i < channels; i++) {
yading@10:         float *chunk = out + mtab->size * i;
yading@10:         float lsp[LSP_COEFS_MAX];
yading@10: 
yading@10:         for (j = 0; j < sub; j++) {
yading@10:             dec_bark_env(tctx, bark1[i][j], bark_use_hist[i][j], i,
yading@10:                          tctx->tmp_buf, gain[sub*i+j], ftype);
yading@10: 
yading@10:             tctx->fdsp.vector_fmul(chunk + block_size*j, chunk + block_size*j,
yading@10:                                    tctx->tmp_buf, block_size);
yading@10: 
yading@10:         }
yading@10: 
yading@10:         if (ftype == FT_LONG) {
yading@10:             float pgain_step = 25000. / ((1 << mtab->pgain_bit) - 1);
yading@10:             int p_coef = get_bits(gb, tctx->mtab->ppc_period_bit);
yading@10:             int g_coef = get_bits(gb, tctx->mtab->pgain_bit);
yading@10:             float v = 1./8192*
yading@10:                 mulawinv(pgain_step*g_coef+ pgain_step/2, 25000., PGAIN_MU);
yading@10: 
yading@10:             decode_ppc(tctx, p_coef, ppc_shape + i*mtab->ppc_shape_len, v,
yading@10:                        chunk);
yading@10:         }
yading@10: 
yading@10:         decode_lsp(tctx, lpc_idx1[i], lpc_idx2[i], lpc_hist_idx[i], lsp,
yading@10:                    tctx->lsp_hist[i]);
yading@10: 
yading@10:         dec_lpc_spectrum_inv(tctx, lsp, ftype, tctx->tmp_buf);
yading@10: 
yading@10:         for (j = 0; j < mtab->fmode[ftype].sub; j++) {
yading@10:             tctx->fdsp.vector_fmul(chunk, chunk, tctx->tmp_buf, block_size);
yading@10:             chunk += block_size;
yading@10:         }
yading@10:     }
yading@10: }
yading@10: 
yading@10: static int twin_decode_frame(AVCodecContext * avctx, void *data,
yading@10:                              int *got_frame_ptr, AVPacket *avpkt)
yading@10: {
yading@10:     AVFrame *frame     = data;
yading@10:     const uint8_t *buf = avpkt->data;
yading@10:     int buf_size = avpkt->size;
yading@10:     TwinContext *tctx = avctx->priv_data;
yading@10:     GetBitContext gb;
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     float **out = NULL;
yading@10:     enum FrameType ftype;
yading@10:     int window_type, ret;
yading@10:     static const enum FrameType wtype_to_ftype_table[] = {
yading@10:         FT_LONG,   FT_LONG, FT_SHORT, FT_LONG,
yading@10:         FT_MEDIUM, FT_LONG, FT_LONG,  FT_MEDIUM, FT_MEDIUM
yading@10:     };
yading@10: 
yading@10:     if (buf_size*8 < avctx->bit_rate*mtab->size/avctx->sample_rate + 8) {
yading@10:         av_log(avctx, AV_LOG_ERROR,
yading@10:                "Frame too small (%d bytes). Truncated file?\n", buf_size);
yading@10:         return AVERROR(EINVAL);
yading@10:     }
yading@10: 
yading@10:     /* get output buffer */
yading@10:     if (tctx->discarded_packets >= 2) {
yading@10:         frame->nb_samples = mtab->size;
yading@10:         if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
yading@10:             return ret;
yading@10:         out = (float **)frame->extended_data;
yading@10:     }
yading@10: 
yading@10:     init_get_bits(&gb, buf, buf_size * 8);
yading@10:     skip_bits(&gb, get_bits(&gb, 8));
yading@10:     window_type = get_bits(&gb, WINDOW_TYPE_BITS);
yading@10: 
yading@10:     if (window_type > 8) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
yading@10:         return -1;
yading@10:     }
yading@10: 
yading@10:     ftype = wtype_to_ftype_table[window_type];
yading@10: 
yading@10:     read_and_decode_spectrum(tctx, &gb, tctx->spectrum, ftype);
yading@10: 
yading@10:     imdct_output(tctx, ftype, window_type, out);
yading@10: 
yading@10:     FFSWAP(float*, tctx->curr_frame, tctx->prev_frame);
yading@10: 
yading@10:     if (tctx->discarded_packets < 2) {
yading@10:         tctx->discarded_packets++;
yading@10:         *got_frame_ptr = 0;
yading@10:         return buf_size;
yading@10:     }
yading@10: 
yading@10:     *got_frame_ptr = 1;
yading@10: 
yading@10:     return buf_size;
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Init IMDCT and windowing tables
yading@10:  */
yading@10: static av_cold int init_mdct_win(TwinContext *tctx)
yading@10: {
yading@10:     int i, j, ret;
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     int size_s = mtab->size / mtab->fmode[FT_SHORT].sub;
yading@10:     int size_m = mtab->size / mtab->fmode[FT_MEDIUM].sub;
yading@10:     int channels = tctx->avctx->channels;
yading@10:     float norm = channels == 1 ? 2. : 1.;
yading@10: 
yading@10:     for (i = 0; i < 3; i++) {
yading@10:         int bsize = tctx->mtab->size/tctx->mtab->fmode[i].sub;
yading@10:         if ((ret = ff_mdct_init(&tctx->mdct_ctx[i], av_log2(bsize) + 1, 1,
yading@10:                                 -sqrt(norm/bsize) / (1<<15))))
yading@10:             return ret;
yading@10:     }
yading@10: 
yading@10:     FF_ALLOC_OR_GOTO(tctx->avctx, tctx->tmp_buf,
yading@10:                      mtab->size * sizeof(*tctx->tmp_buf), alloc_fail);
yading@10: 
yading@10:     FF_ALLOC_OR_GOTO(tctx->avctx, tctx->spectrum,
yading@10:                      2 * mtab->size * channels * sizeof(*tctx->spectrum),
yading@10:                      alloc_fail);
yading@10:     FF_ALLOC_OR_GOTO(tctx->avctx, tctx->curr_frame,
yading@10:                      2 * mtab->size * channels * sizeof(*tctx->curr_frame),
yading@10:                      alloc_fail);
yading@10:     FF_ALLOC_OR_GOTO(tctx->avctx, tctx->prev_frame,
yading@10:                      2 * mtab->size * channels * sizeof(*tctx->prev_frame),
yading@10:                      alloc_fail);
yading@10: 
yading@10:     for (i = 0; i < 3; i++) {
yading@10:         int m = 4*mtab->size/mtab->fmode[i].sub;
yading@10:         double freq = 2*M_PI/m;
yading@10:         FF_ALLOC_OR_GOTO(tctx->avctx, tctx->cos_tabs[i],
yading@10:                          (m / 4) * sizeof(*tctx->cos_tabs[i]), alloc_fail);
yading@10: 
yading@10:         for (j = 0; j <= m/8; j++)
yading@10:             tctx->cos_tabs[i][j] = cos((2*j + 1)*freq);
yading@10:         for (j = 1; j <  m/8; j++)
yading@10:             tctx->cos_tabs[i][m/4-j] = tctx->cos_tabs[i][j];
yading@10:     }
yading@10: 
yading@10: 
yading@10:     ff_init_ff_sine_windows(av_log2(size_m));
yading@10:     ff_init_ff_sine_windows(av_log2(size_s/2));
yading@10:     ff_init_ff_sine_windows(av_log2(mtab->size));
yading@10: 
yading@10:     return 0;
yading@10: alloc_fail:
yading@10:     return AVERROR(ENOMEM);
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Interpret the data as if it were a num_blocks x line_len[0] matrix and for
yading@10:  * each line do a cyclic permutation, i.e.
yading@10:  * abcdefghijklm -> defghijklmabc
yading@10:  * where the amount to be shifted is evaluated depending on the column.
yading@10:  */
yading@10: static void permutate_in_line(int16_t *tab, int num_vect, int num_blocks,
yading@10:                               int block_size,
yading@10:                               const uint8_t line_len[2], int length_div,
yading@10:                               enum FrameType ftype)
yading@10: 
yading@10: {
yading@10:     int i,j;
yading@10: 
yading@10:     for (i = 0; i < line_len[0]; i++) {
yading@10:         int shift;
yading@10: 
yading@10:         if (num_blocks == 1 ||
yading@10:             (ftype == FT_LONG && num_vect % num_blocks) ||
yading@10:             (ftype != FT_LONG && num_vect & 1         ) ||
yading@10:             i == line_len[1]) {
yading@10:             shift = 0;
yading@10:         } else if (ftype == FT_LONG) {
yading@10:             shift = i;
yading@10:         } else
yading@10:             shift = i*i;
yading@10: 
yading@10:         for (j = 0; j < num_vect && (j+num_vect*i < block_size*num_blocks); j++)
yading@10:             tab[i*num_vect+j] = i*num_vect + (j + shift) % num_vect;
yading@10:     }
yading@10: }
yading@10: 
yading@10: /**
yading@10:  * Interpret the input data as in the following table:
yading@10:  *
yading@10:  * @verbatim
yading@10:  *
yading@10:  * abcdefgh
yading@10:  * ijklmnop
yading@10:  * qrstuvw
yading@10:  * x123456
yading@10:  *
yading@10:  * @endverbatim
yading@10:  *
yading@10:  * and transpose it, giving the output
yading@10:  * aiqxbjr1cks2dlt3emu4fvn5gow6hp
yading@10:  */
yading@10: static void transpose_perm(int16_t *out, int16_t *in, int num_vect,
yading@10:                            const uint8_t line_len[2], int length_div)
yading@10: {
yading@10:     int i,j;
yading@10:     int cont= 0;
yading@10:     for (i = 0; i < num_vect; i++)
yading@10:         for (j = 0; j < line_len[i >= length_div]; j++)
yading@10:             out[cont++] = in[j*num_vect + i];
yading@10: }
yading@10: 
yading@10: static void linear_perm(int16_t *out, int16_t *in, int n_blocks, int size)
yading@10: {
yading@10:     int block_size = size/n_blocks;
yading@10:     int i;
yading@10: 
yading@10:     for (i = 0; i < size; i++)
yading@10:         out[i] = block_size * (in[i] % n_blocks) + in[i] / n_blocks;
yading@10: }
yading@10: 
yading@10: static av_cold void construct_perm_table(TwinContext *tctx,enum FrameType ftype)
yading@10: {
yading@10:     int block_size;
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     int size;
yading@10:     int16_t *tmp_perm = (int16_t *) tctx->tmp_buf;
yading@10: 
yading@10:     if (ftype == FT_PPC) {
yading@10:         size  = tctx->avctx->channels;
yading@10:         block_size = mtab->ppc_shape_len;
yading@10:     } else {
yading@10:         size       = tctx->avctx->channels * mtab->fmode[ftype].sub;
yading@10:         block_size = mtab->size / mtab->fmode[ftype].sub;
yading@10:     }
yading@10: 
yading@10:     permutate_in_line(tmp_perm, tctx->n_div[ftype], size,
yading@10:                       block_size, tctx->length[ftype],
yading@10:                       tctx->length_change[ftype], ftype);
yading@10: 
yading@10:     transpose_perm(tctx->permut[ftype], tmp_perm, tctx->n_div[ftype],
yading@10:                    tctx->length[ftype], tctx->length_change[ftype]);
yading@10: 
yading@10:     linear_perm(tctx->permut[ftype], tctx->permut[ftype], size,
yading@10:                 size*block_size);
yading@10: }
yading@10: 
yading@10: static av_cold void init_bitstream_params(TwinContext *tctx)
yading@10: {
yading@10:     const ModeTab *mtab = tctx->mtab;
yading@10:     int n_ch = tctx->avctx->channels;
yading@10:     int total_fr_bits = tctx->avctx->bit_rate*mtab->size/
yading@10:                              tctx->avctx->sample_rate;
yading@10: 
yading@10:     int lsp_bits_per_block = n_ch*(mtab->lsp_bit0 + mtab->lsp_bit1 +
yading@10:                                    mtab->lsp_split*mtab->lsp_bit2);
yading@10: 
yading@10:     int ppc_bits = n_ch*(mtab->pgain_bit + mtab->ppc_shape_bit +
yading@10:                          mtab->ppc_period_bit);
yading@10: 
yading@10:     int bsize_no_main_cb[3];
yading@10:     int bse_bits[3];
yading@10:     int i;
yading@10:     enum FrameType frametype;
yading@10: 
yading@10:     for (i = 0; i < 3; i++)
yading@10:         // +1 for history usage switch
yading@10:         bse_bits[i] = n_ch *
yading@10:             (mtab->fmode[i].bark_n_coef * mtab->fmode[i].bark_n_bit + 1);
yading@10: 
yading@10:     bsize_no_main_cb[2] = bse_bits[2] + lsp_bits_per_block + ppc_bits +
yading@10:                           WINDOW_TYPE_BITS + n_ch*GAIN_BITS;
yading@10: 
yading@10:     for (i = 0; i < 2; i++)
yading@10:         bsize_no_main_cb[i] =
yading@10:             lsp_bits_per_block + n_ch*GAIN_BITS + WINDOW_TYPE_BITS +
yading@10:             mtab->fmode[i].sub*(bse_bits[i] + n_ch*SUB_GAIN_BITS);
yading@10: 
yading@10:     // The remaining bits are all used for the main spectrum coefficients
yading@10:     for (i = 0; i < 4; i++) {
yading@10:         int bit_size;
yading@10:         int vect_size;
yading@10:         int rounded_up, rounded_down, num_rounded_down, num_rounded_up;
yading@10:         if (i == 3) {
yading@10:             bit_size  = n_ch * mtab->ppc_shape_bit;
yading@10:             vect_size = n_ch * mtab->ppc_shape_len;
yading@10:         } else {
yading@10:             bit_size = total_fr_bits - bsize_no_main_cb[i];
yading@10:             vect_size = n_ch * mtab->size;
yading@10:         }
yading@10: 
yading@10:         tctx->n_div[i] = (bit_size + 13) / 14;
yading@10: 
yading@10:         rounded_up   = (bit_size + tctx->n_div[i] - 1)/tctx->n_div[i];
yading@10:         rounded_down = (bit_size           )/tctx->n_div[i];
yading@10:         num_rounded_down = rounded_up * tctx->n_div[i] - bit_size;
yading@10:         num_rounded_up = tctx->n_div[i] - num_rounded_down;
yading@10:         tctx->bits_main_spec[0][i][0] = (rounded_up   + 1)/2;
yading@10:         tctx->bits_main_spec[1][i][0] = (rounded_up      )/2;
yading@10:         tctx->bits_main_spec[0][i][1] = (rounded_down + 1)/2;
yading@10:         tctx->bits_main_spec[1][i][1] = (rounded_down    )/2;
yading@10:         tctx->bits_main_spec_change[i] = num_rounded_up;
yading@10: 
yading@10:         rounded_up   = (vect_size + tctx->n_div[i] - 1)/tctx->n_div[i];
yading@10:         rounded_down = (vect_size                     )/tctx->n_div[i];
yading@10:         num_rounded_down = rounded_up * tctx->n_div[i] - vect_size;
yading@10:         num_rounded_up = tctx->n_div[i] - num_rounded_down;
yading@10:         tctx->length[i][0] = rounded_up;
yading@10:         tctx->length[i][1] = rounded_down;
yading@10:         tctx->length_change[i] = num_rounded_up;
yading@10:     }
yading@10: 
yading@10:     for (frametype = FT_SHORT; frametype <= FT_PPC; frametype++)
yading@10:         construct_perm_table(tctx, frametype);
yading@10: }
yading@10: 
yading@10: static av_cold int twin_decode_close(AVCodecContext *avctx)
yading@10: {
yading@10:     TwinContext *tctx = avctx->priv_data;
yading@10:     int i;
yading@10: 
yading@10:     for (i = 0; i < 3; i++) {
yading@10:         ff_mdct_end(&tctx->mdct_ctx[i]);
yading@10:         av_free(tctx->cos_tabs[i]);
yading@10:     }
yading@10: 
yading@10: 
yading@10:     av_free(tctx->curr_frame);
yading@10:     av_free(tctx->spectrum);
yading@10:     av_free(tctx->prev_frame);
yading@10:     av_free(tctx->tmp_buf);
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: static av_cold int twin_decode_init(AVCodecContext *avctx)
yading@10: {
yading@10:     int ret;
yading@10:     TwinContext *tctx = avctx->priv_data;
yading@10:     int isampf, ibps;
yading@10: 
yading@10:     tctx->avctx       = avctx;
yading@10:     avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
yading@10: 
yading@10:     if (!avctx->extradata || avctx->extradata_size < 12) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n");
yading@10:         return AVERROR_INVALIDDATA;
yading@10:     }
yading@10:     avctx->channels = AV_RB32(avctx->extradata    ) + 1;
yading@10:     avctx->bit_rate = AV_RB32(avctx->extradata + 4) * 1000;
yading@10:     isampf          = AV_RB32(avctx->extradata + 8);
yading@10: 
yading@10:     if (isampf < 8 || isampf > 44) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "Unsupported sample rate\n");
yading@10:         return AVERROR_INVALIDDATA;
yading@10:     }
yading@10:     switch (isampf) {
yading@10:     case 44: avctx->sample_rate = 44100;         break;
yading@10:     case 22: avctx->sample_rate = 22050;         break;
yading@10:     case 11: avctx->sample_rate = 11025;         break;
yading@10:     default: avctx->sample_rate = isampf * 1000; break;
yading@10:     }
yading@10: 
yading@10:     if (avctx->channels <= 0 || avctx->channels > CHANNELS_MAX) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n",
yading@10:                avctx->channels);
yading@10:         return -1;
yading@10:     }
yading@10:     avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO :
yading@10:                                                    AV_CH_LAYOUT_STEREO;
yading@10: 
yading@10:     ibps = avctx->bit_rate / (1000 * avctx->channels);
yading@10: 
yading@10:     if (ibps > 255U) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "unsupported per channel bitrate %dkbps\n", ibps);
yading@10:         return AVERROR_INVALIDDATA;
yading@10:     }
yading@10: 
yading@10:     switch ((isampf << 8) +  ibps) {
yading@10:     case (8 <<8) +  8: tctx->mtab = &mode_08_08; break;
yading@10:     case (11<<8) +  8: tctx->mtab = &mode_11_08; break;
yading@10:     case (11<<8) + 10: tctx->mtab = &mode_11_10; break;
yading@10:     case (16<<8) + 16: tctx->mtab = &mode_16_16; break;
yading@10:     case (22<<8) + 20: tctx->mtab = &mode_22_20; break;
yading@10:     case (22<<8) + 24: tctx->mtab = &mode_22_24; break;
yading@10:     case (22<<8) + 32: tctx->mtab = &mode_22_32; break;
yading@10:     case (44<<8) + 40: tctx->mtab = &mode_44_40; break;
yading@10:     case (44<<8) + 48: tctx->mtab = &mode_44_48; break;
yading@10:     default:
yading@10:         av_log(avctx, AV_LOG_ERROR, "This version does not support %d kHz - %d kbit/s/ch mode.\n", isampf, isampf);
yading@10:         return -1;
yading@10:     }
yading@10: 
yading@10:     avpriv_float_dsp_init(&tctx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
yading@10:     if ((ret = init_mdct_win(tctx))) {
yading@10:         av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
yading@10:         twin_decode_close(avctx);
yading@10:         return ret;
yading@10:     }
yading@10:     init_bitstream_params(tctx);
yading@10: 
yading@10:     memset_float(tctx->bark_hist[0][0], 0.1, FF_ARRAY_ELEMS(tctx->bark_hist));
yading@10: 
yading@10:     return 0;
yading@10: }
yading@10: 
yading@10: AVCodec ff_twinvq_decoder = {
yading@10:     .name           = "twinvq",
yading@10:     .type           = AVMEDIA_TYPE_AUDIO,
yading@10:     .id             = AV_CODEC_ID_TWINVQ,
yading@10:     .priv_data_size = sizeof(TwinContext),
yading@10:     .init           = twin_decode_init,
yading@10:     .close          = twin_decode_close,
yading@10:     .decode         = twin_decode_frame,
yading@10:     .capabilities   = CODEC_CAP_DR1,
yading@10:     .long_name      = NULL_IF_CONFIG_SMALL("VQF TwinVQ"),
yading@10:     .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
yading@10:                                                       AV_SAMPLE_FMT_NONE },
yading@10: };