/*
********************************************************************************
*
*      GSM AMR-NB speech codec   R98   Version 7.6.0   December 12, 2001
*                                R99   Version 3.3.0                
*                                REL-4 Version 4.1.0                
*
********************************************************************************
*
*      File             : qgain795.c
*      Purpose          : pitch and codebook gain quantization for MR795
*
********************************************************************************
*/

/*
********************************************************************************
*                         MODULE INCLUDE FILE AND VERSION ID
********************************************************************************
*/
#include "namespace.h"
#include "qgain795.h"

/*
********************************************************************************
*                         INCLUDE FILES
********************************************************************************
*/
#include "typedef.h"
#include "basic_op.h"
#include "oper_32b.h"
#include "no_count.h"
#include "cnst.h"
#include "log2.h"
#include "pow2.h"
#include "sqrt_l.h"
#include "g_adapt.h"
#include "calc_en.h"
#include "q_gain_p.h"
#include "mac_32.h"
#include "gains_tab.h"

/*
********************************************************************************
*                         LOCAL PROGRAM CODE
********************************************************************************
*/

/*************************************************************************
 *
 * FUNCTION:  MR795_gain_code_quant3
 *
 * PURPOSE: Pre-quantization of codebook gains, given three possible
 *          LTP gains (using predicted codebook gain)
 *
 *************************************************************************/
static void
MR795_gain_code_quant3(
    Word16 exp_gcode0,        /* i  : predicted CB gain (exponent), Q0  */
    Word16 gcode0,            /* i  : predicted CB gain (norm.),    Q14 */
    Word16 g_pitch_cand[],    /* i  : Pitch gain candidates (3),    Q14 */
    Word16 g_pitch_cind[],    /* i  : Pitch gain cand. indices (3), Q0  */
    Word16 frac_coeff[],      /* i  : coefficients (5),             Q15 */
    Word16 exp_coeff[],       /* i  : energy coefficients (5),      Q0  */
                              /*      coefficients from calc_filt_ener()*/
    Word16 *gain_pit,         /* o  : Pitch gain,                   Q14 */
    Word16 *gain_pit_ind,     /* o  : Pitch gain index,             Q0  */
    Word16 *gain_cod,         /* o  : Code gain,                    Q1  */
    Word16 *gain_cod_ind,     /* o  : Code gain index,              Q0  */
    Word16 *qua_ener_MR122,   /* o  : quantized energy error,       Q10 */
                              /*      (for MR122 MA predictor update)   */
    Word16 *qua_ener          /* o  : quantized energy error,       Q10 */
                              /*      (for other MA predictor update)   */
)
{
    const Word16 *p;
    Word16 i, j, cod_ind, pit_ind;
    Word16 e_max, exp_code;
    Word16 g_pitch, g2_pitch, g_code, g2_code_h, g2_code_l;
    Word16 g_pit_cod_h, g_pit_cod_l;
    Word16 coeff[5], coeff_lo[5];
    Word16 exp_max[5];
    Word32 L_tmp, L_tmp0, dist_min;

    /*
     * The error energy (sum) to be minimized consists of five terms, t[0..4].
     *
     *                      t[0] =    gp^2  * <y1 y1>
     *                      t[1] = -2*gp    * <xn y1>
     *                      t[2] =    gc^2  * <y2 y2>
     *                      t[3] = -2*gc    * <xn y2>
     *                      t[4] =  2*gp*gc * <y1 y2>
     *
     */

    /* determine the scaling exponent for g_code: ec = ec0 - 10 */
    exp_code = sub(exp_gcode0, 10);

    /* calculate exp_max[i] = s[i]-1 */
    exp_max[0] = sub(exp_coeff[0], 13);                        move16 ();
    exp_max[1] = sub(exp_coeff[1], 14);                        move16 ();
    exp_max[2] = add(exp_coeff[2], add(15, shl(exp_code, 1))); move16 ();
    exp_max[3] = add(exp_coeff[3], exp_code);                  move16 ();
    exp_max[4] = add(exp_coeff[4], add(exp_code,1));           move16 ();


    /*-------------------------------------------------------------------*
     *  Find maximum exponent:                                           *
     *  ~~~~~~~~~~~~~~~~~~~~~~                                           *
     *                                                                   *
     *  For the sum operation, all terms must have the same scaling;     *
     *  that scaling should be low enough to prevent overflow. There-    *
     *  fore, the maximum scale is determined and all coefficients are   *
     *  re-scaled:                                                       *
     *                                                                   *
     *    e_max = max(exp_max[i]) + 1;                                   *
     *    e = exp_max[i]-e_max;         e <= 0!                          *
     *    c[i] = c[i]*2^e                                                *
     *-------------------------------------------------------------------*/

    e_max = exp_max[0];                                        move16 ();
    for (i = 1; i < 5; i++)     /* implemented flattened */
    {
        move16(); test();
        if (sub(exp_max[i], e_max) > 0)
        {
            e_max = exp_max[i];                                move16 ();
        }
    }

    e_max = add(e_max, 1);      /* To avoid overflow */

    for (i = 0; i < 5; i++) {
        j = sub(e_max, exp_max[i]);
        L_tmp = L_deposit_h(frac_coeff[i]);
        L_tmp = L_shr(L_tmp, j);
        L_Extract(L_tmp, &coeff[i], &coeff_lo[i]);
    }


    /*-------------------------------------------------------------------*
     *  Codebook search:                                                 *
     *  ~~~~~~~~~~~~~~~~                                                 *
     *                                                                   *
     *  For each of the candiates LTP gains in g_pitch_cand[], the terms *
     *  t[0..4] are calculated from the values in the table (and the     *
     *  pitch gain candidate) and summed up; the result is the mean      *
     *  squared error for the LPT/CB gain pair. The index for the mini-  *
     *  mum MSE is stored and finally used to retrieve the quantized CB  *
     *  gain                                                             *
     *-------------------------------------------------------------------*/

    /* start with "infinite" MSE */
    dist_min = MAX_32;        move16 ();
    cod_ind = 0;              move16 ();
    pit_ind = 0;              move16 ();

    /* loop through LTP gain candidates */
    for (j = 0; j < 3; j++)
    {
        /* pre-calculate terms only dependent on pitch gain */
        g_pitch = g_pitch_cand[j];    move16 ();
        g2_pitch = mult(g_pitch, g_pitch);
        L_tmp0 = Mpy_32_16(        coeff[0], coeff_lo[0], g2_pitch);
        L_tmp0 = Mac_32_16(L_tmp0, coeff[1], coeff_lo[1], g_pitch);

        p = &qua_gain_code[0];
        for (i = 0; i < NB_QUA_CODE; i++)
        {
            g_code = *p++;        move16 (); /* this is g_fac        Q11 */
            p++;                             /* skip log2(g_fac)         */
            p++;                             /* skip 20*log10(g_fac)     */

            g_code = mult(g_code, gcode0);

            L_tmp = L_mult (g_code, g_code);
            L_Extract (L_tmp, &g2_code_h, &g2_code_l);

            L_tmp = L_mult(g_code, g_pitch);
            L_Extract (L_tmp, &g_pit_cod_h, &g_pit_cod_l);

            L_tmp = Mac_32  (L_tmp0, coeff[2], coeff_lo[2],
                                     g2_code_h, g2_code_l);
            L_tmp = Mac_32_16(L_tmp, coeff[3], coeff_lo[3],
                                     g_code);
            L_tmp = Mac_32   (L_tmp, coeff[4], coeff_lo[4],
                                     g_pit_cod_h, g_pit_cod_l);

            /* store table index if MSE for this index is lower
               than the minimum MSE seen so far; also store the
               pitch gain for this (so far) lowest MSE          */
            test ();
            if (L_sub(L_tmp, dist_min) < (Word32) 0)
            {
                dist_min = L_tmp;                move32 ();
                cod_ind = i;                     move16 ();
                pit_ind = j;                     move16 ();
            }
        }
    }

    /*------------------------------------------------------------------*
     *  read quantized gains and new values for MA predictor memories   *
     *  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   *
     *------------------------------------------------------------------*/

    /* Read the quantized gains */
    p = &qua_gain_code[add (add (cod_ind, cod_ind), cod_ind)]; move16 ();
    g_code = *p++;            move16();
    *qua_ener_MR122 = *p++;   move16();
    *qua_ener = *p;           move16();

    /*------------------------------------------------------------------*
     *  calculate final fixed codebook gain:                            *
     *  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                            *
     *                                                                  *
     *   gc = gc0 * g                                                   *
     *------------------------------------------------------------------*/

    L_tmp = L_mult(g_code, gcode0);
    L_tmp = L_shr(L_tmp, sub(9, exp_gcode0));
    *gain_cod = extract_h(L_tmp);
    *gain_cod_ind = cod_ind;                move16 ();
    *gain_pit = g_pitch_cand[pit_ind];      move16 ();
    *gain_pit_ind = g_pitch_cind[pit_ind];  move16 ();
}


/*************************************************************************
 *
 * FUNCTION:  MR795_gain_code_quant_mod
 *
 * PURPOSE: Modified quantization of the MR795 codebook gain
 *
 * Uses pre-computed energy coefficients in frac_en[]/exp_en[]
 *
 *       frac_en[0]*2^exp_en[0] = <res res>   // LP residual energy
 *       frac_en[1]*2^exp_en[1] = <exc exc>   // LTP residual energy
 *       frac_en[2]*2^exp_en[2] = <exc code>  // LTP/CB innovation dot product
 *       frac_en[3]*2^exp_en[3] = <code code> // CB innovation energy
 *
 *************************************************************************/
static Word16
MR795_gain_code_quant_mod(  /* o  : index of quantization.            */
    Word16 gain_pit,        /* i  : pitch gain,                   Q14 */
    Word16 exp_gcode0,      /* i  : predicted CB gain (exponent), Q0  */
    Word16 gcode0,          /* i  : predicted CB gain (norm.),    Q14 */
    Word16 frac_en[],       /* i  : energy coefficients (4),
                                    fraction part,                Q15 */
    Word16 exp_en[],        /* i  : energy coefficients (4),
                                    eponent part,                 Q0  */
    Word16 alpha,           /* i  : gain adaptor factor (>0),     Q15 */
    Word16 gain_cod_unq,    /* i  : Code gain (unquantized)           */
                            /*      (scaling: Q10 - exp_gcode0)       */
    Word16 *gain_cod,       /* i/o: Code gain (pre-/quantized),   Q1  */
    Word16 *qua_ener_MR122, /* o  : quantized energy error,       Q10 */
                            /*      (for MR122 MA predictor update)   */
    Word16 *qua_ener        /* o  : quantized energy error,       Q10 */
                            /*      (for other MA predictor update)   */
 )
{
    const Word16 *p;
    Word16 i, index, tmp;
    Word16 one_alpha;
    Word16 exp, e_max;
    Word16 g2_pitch, g_code;
    Word16 g2_code_h, g2_code_l;
    Word16 d2_code_h, d2_code_l;
    Word16 coeff[5], coeff_lo[5], exp_coeff[5];
    Word32 L_tmp, L_t0, L_t1, dist_min;
    Word16 gain_code;

    /*
      Steps in calculation of the error criterion (dist):
      ---------------------------------------------------

      underlined = constant; alp = FLP value of alpha, alpha = FIP
      ----------


        ExEn = gp^2 * LtpEn + 2.0*gp*gc[i] * XC + gc[i]^2 * InnEn;
               ------------   ------         --             -----

        aExEn= alp * ExEn
             = alp*gp^2*LtpEn + 2.0*alp*gp*XC* gc[i] + alp*InnEn* gc[i]^2
               --------------   -------------          ---------

             =         t[1]   +              t[2]    +          t[3]

        dist = d1 + d2;

          d1 = (1.0 - alp) * InnEn * (gcu - gc[i])^2 = t[4]
               -------------------    ---

          d2 =        alp  * (ResEn - 2.0 * sqrt(ResEn*ExEn) + ExEn);
                      ---     -----   ---        -----

             =        alp  * (sqrt(ExEn) - sqrt(ResEn))^2
                      ---                  -----------

             =               (sqrt(aExEn) - sqrt(alp*ResEn))^2
                                            ---------------

             =               (sqrt(aExEn) -       t[0]     )^2
                                                  ----

     */

    /*
     * calculate scalings of the constant terms
     */
    gain_code = shl (*gain_cod, sub (10, exp_gcode0)); /* Q1  -> Q11 (-ec0) */
    g2_pitch = mult (gain_pit, gain_pit);              /* Q14 -> Q13        */
    /* 0 < alpha <= 0.5 => 0.5 <= 1-alpha < 1, i.e one_alpha is normalized  */
    one_alpha = add (sub (32767, alpha), 1); /* 32768 - alpha */


    /*  alpha <= 0.5 -> mult. by 2 to keep precision; compensate in exponent */
    tmp = extract_h (L_shl (L_mult (alpha, frac_en[1]), 1));
    /* directly store in 32 bit variable because no further mult. required */
    L_t1 = L_mult (tmp, g2_pitch);                    move16 ();
    exp_coeff[1] = sub (exp_en[1], 15);               move16 ();


    tmp = extract_h (L_shl (L_mult (alpha, frac_en[2]), 1));
    coeff[2] = mult (tmp, gain_pit);                  move16 ();
    exp = sub (exp_gcode0, 10);
    exp_coeff[2] = add (exp_en[2], exp);              move16 ();


    /* alpha <= 0.5 -> mult. by 2 to keep precision; compensate in exponent */
    coeff[3] = extract_h (L_shl (L_mult (alpha, frac_en[3]), 1));
    exp = sub (shl (exp_gcode0, 1), 7);
    exp_coeff[3] = add (exp_en[3], exp);              move16 ();


    coeff[4] = mult (one_alpha, frac_en[3]);          move16 ();
    exp_coeff[4] = add (exp_coeff[3], 1);             move16 ();


    L_tmp = L_mult (alpha, frac_en[0]);
    /* sqrt_l returns normalized value and 2*exponent
       -> result = val >> (exp/2)
       exp_coeff holds 2*exponent for c[0]            */
    /* directly store in 32 bit variable because no further mult. required */
    L_t0 = sqrt_l_exp (L_tmp, &exp); /* normalization included in sqrt_l_exp */
                                     move32 (); /* function result */
    exp = add (exp, 47);
    exp_coeff[0] = sub (exp_en[0], exp);              move16 ();

    /*
     * Determine the maximum exponent occuring in the distance calculation
     * and adjust all fractions accordingly (including a safety margin)
     *
     */

    /* find max(e[1..4],e[0]+31) */
    e_max = add (exp_coeff[0], 31);
    for (i = 1; i <= 4; i++)
    {
        test ();
        if (sub (exp_coeff[i], e_max) > 0)
        {
            e_max = exp_coeff[i];                     move16 ();
        }
    }

    /* scale c[1]         (requires no further multiplication) */
    tmp = sub (e_max, exp_coeff[1]);
    L_t1 = L_shr(L_t1, tmp);

    /* scale c[2..4] (used in Mpy_32_16 in the quantizer loop) */
    for (i = 2; i <= 4; i++)
    {
        tmp = sub (e_max, exp_coeff[i]);
        L_tmp = L_deposit_h(coeff[i]);
        L_tmp = L_shr(L_tmp, tmp);
        L_Extract(L_tmp, &coeff[i], &coeff_lo[i]);
    }

    /* scale c[0]         (requires no further multiplication) */
    exp = sub (e_max, 31);             /* new exponent */
    tmp = sub (exp, exp_coeff[0]);
    L_t0 = L_shr (L_t0, shr (tmp, 1));
    /* perform correction by 1/sqrt(2) if exponent difference is odd */
    test (); logic16 ();
    if ((tmp & 0x1) != 0)
    {
        L_Extract(L_t0, &coeff[0], &coeff_lo[0]);
        L_t0 = Mpy_32_16(coeff[0], coeff_lo[0],
                         23170);                    /* 23170 Q15 = 1/sqrt(2)*/
    }

    /* search the quantizer table for the lowest value
       of the search criterion                           */
    dist_min = MAX_32; move32 ();
    index = 0;         move16 ();
    p = &qua_gain_code[0]; move16 ();

    for (i = 0; i < NB_QUA_CODE; i++)
    {
        g_code = *p++;        move16 (); /* this is g_fac (Q11)  */
        p++;                             /* skip log2(g_fac)     */
        p++;                             /* skip 20*log10(g_fac) */
        g_code = mult (g_code, gcode0);

        /* only continue if    gc[i]            < 2.0*gc
           which is equiv. to  g_code (Q10-ec0) < gain_code (Q11-ec0) */
        test ();
        if (sub (g_code, gain_code) >= 0)
            break;

        L_tmp = L_mult (g_code, g_code);
        L_Extract (L_tmp, &g2_code_h, &g2_code_l);

        tmp = sub (g_code, gain_cod_unq);
        L_tmp = L_mult (tmp, tmp);
        L_Extract (L_tmp, &d2_code_h, &d2_code_l);

        /* t2, t3, t4 */
        L_tmp = Mac_32_16 (L_t1, coeff[2], coeff_lo[2], g_code);
        L_tmp = Mac_32(L_tmp,    coeff[3], coeff_lo[3], g2_code_h, g2_code_l);

        L_tmp = sqrt_l_exp (L_tmp, &exp);
        L_tmp = L_shr (L_tmp, shr (exp, 1));

        /* d2 */
        tmp = round (L_sub (L_tmp, L_t0));
        L_tmp = L_mult (tmp, tmp);

        /* dist */
        L_tmp = Mac_32(L_tmp, coeff[4], coeff_lo[4], d2_code_h, d2_code_l);

        /* store table index if distance measure for this
            index is lower than the minimum seen so far   */
        test ();
        if (L_sub (L_tmp, dist_min) < (Word32) 0)
        {
            dist_min = L_tmp;                move16 ();
            index = i;                       move16 ();
        }
    }

    /*------------------------------------------------------------------*
     *  read quantized gains and new values for MA predictor memories   *
     *  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   *
     *------------------------------------------------------------------*/

    /* Read the quantized gains */
    p = &qua_gain_code[add (add (index, index), index)]; move16 ();
    g_code = *p++;            move16();
    *qua_ener_MR122 = *p++;   move16();
    *qua_ener = *p;           move16();

    /*------------------------------------------------------------------*
     *  calculate final fixed codebook gain:                            *
     *  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~                            *
     *                                                                  *
     *   gc = gc0 * g                                                   *
     *------------------------------------------------------------------*/

    L_tmp = L_mult(g_code, gcode0);
    L_tmp = L_shr(L_tmp, sub(9, exp_gcode0));
    *gain_cod = extract_h(L_tmp);

    return index;
}

/*
********************************************************************************
*                         PUBLIC PROGRAM CODE
********************************************************************************
*/

/*************************************************************************
 *
 * FUNCTION:  MR795_gain_quant
 *
 * PURPOSE:   pitch and codebook quantization for MR795
 *
 *************************************************************************/
void
MR795_gain_quant(
    GainAdaptState *adapt_st, /* i/o: gain adapter state structure       */
    Word16 res[],             /* i  : LP residual,                  Q0   */
    Word16 exc[],             /* i  : LTP excitation (unfiltered),  Q0   */
    Word16 code[],            /* i  : CB innovation (unfiltered),   Q13  */
    Word16 frac_coeff[],      /* i  : coefficients (5),             Q15  */
    Word16 exp_coeff[],       /* i  : energy coefficients (5),      Q0   */
                              /*      coefficients from calc_filt_ener() */
    Word16 exp_code_en,       /* i  : innovation energy (exponent), Q0   */
    Word16 frac_code_en,      /* i  : innovation energy (fraction), Q15  */
    Word16 exp_gcode0,        /* i  : predicted CB gain (exponent), Q0   */
    Word16 frac_gcode0,       /* i  : predicted CB gain (fraction), Q15  */
    Word16 L_subfr,           /* i  : Subframe length                    */
    Word16 cod_gain_frac,     /* i  : opt. codebook gain (fraction),Q15  */
    Word16 cod_gain_exp,      /* i  : opt. codebook gain (exponent), Q0  */
    Word16 gp_limit,          /* i  : pitch gain limit                   */
    Word16 *gain_pit,         /* i/o: Pitch gain,                   Q14  */
    Word16 *gain_cod,         /* o  : Code gain,                    Q1   */
    Word16 *qua_ener_MR122,   /* o  : quantized energy error,       Q10  */
                              /*      (for MR122 MA predictor update)    */
    Word16 *qua_ener,         /* o  : quantized energy error,       Q10  */
                              /*      (for other MA predictor update)    */
    Word16 **anap             /* o  : Index of quantization              */
                              /*      (first gain pitch, then code pitch)*/
)
{
    Word16 frac_en[4];
    Word16 exp_en[4];
    Word16 ltpg, alpha, gcode0;
    Word16 g_pitch_cand[3];      /* pitch gain candidates   Q14 */
    Word16 g_pitch_cind[3];      /* pitch gain indices      Q0  */
    Word16 gain_pit_index;
    Word16 gain_cod_index;
    Word16 exp;
    Word16 gain_cod_unq;         /* code gain (unq.) Q(10-exp_gcode0)  */


    /* get list of candidate quantized pitch gain values
     * and corresponding quantization indices
     */
    gain_pit_index = q_gain_pitch (MR795, gp_limit, gain_pit,
                                   g_pitch_cand, g_pitch_cind); 
                     move16 (); /* function result */

    /*-------------------------------------------------------------------*
     *  predicted codebook gain                                          *
     *  ~~~~~~~~~~~~~~~~~~~~~~~                                          *
     *  gc0     = 2^exp_gcode0 + 2^frac_gcode0                           *
     *                                                                   *
     *  gcode0 (Q14) = 2^14*2^frac_gcode0 = gc0 * 2^(14-exp_gcode0)      *
     *-------------------------------------------------------------------*/
    gcode0 = extract_l(Pow2(14, frac_gcode0));          /* Q14 */

    /* pre-quantization of codebook gain
     * (using three pitch gain candidates);
     * result: best guess of pitch gain and code gain
     */
    MR795_gain_code_quant3(
        exp_gcode0, gcode0, g_pitch_cand, g_pitch_cind,
        frac_coeff, exp_coeff,
        gain_pit, &gain_pit_index, gain_cod, &gain_cod_index,
        qua_ener_MR122, qua_ener);

    /* calculation of energy coefficients and LTP coding gain */
    calc_unfilt_energies(res, exc, code, *gain_pit, L_subfr,
                         frac_en, exp_en, &ltpg);

    /* run gain adaptor, calculate alpha factor to balance LTP/CB gain
     * (this includes the gain adaptor update)
     * Note: ltpg = 0 if frac_en[0] == 0, so the update is OK in that case
     */
    gain_adapt(adapt_st, ltpg, *gain_cod, &alpha);

    /* if this is a very low energy signal (threshold: see
     * calc_unfilt_energies) or alpha <= 0 then don't run the modified quantizer
     */
    test (); move16 (); test ();
    if (frac_en[0] != 0 && alpha > 0)
    {
        /* innovation energy <cod cod> was already computed in gc_pred() */
        /* (this overwrites the LtpResEn which is no longer needed)      */
        frac_en[3] = frac_code_en;          move16 ();
        exp_en[3] = exp_code_en;            move16 ();
        
        /* store optimum codebook gain in Q(10-exp_gcode0) */
        exp = add (sub (cod_gain_exp, exp_gcode0), 10);
        gain_cod_unq = shl (cod_gain_frac, exp);
        
        /* run quantization with modified criterion */
        gain_cod_index = MR795_gain_code_quant_mod(
            *gain_pit, exp_gcode0, gcode0,
            frac_en, exp_en, alpha, gain_cod_unq,
            gain_cod, qua_ener_MR122, qua_ener); move16 (); /* function result */
    }

    *(*anap)++ = gain_pit_index;        move16 ();
    *(*anap)++ = gain_cod_index;        move16 ();
}