view libtwamr/calc_en.c @ 335:03198f6b0427

libtwamr: integrate d8_31pf.c
author Mychaela Falconia <falcon@freecalypso.org>
date Thu, 18 Apr 2024 22:35:05 +0000
parents 2df212a012af
children
line wrap: on
line source

/*
********************************************************************************
*
*      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             : calc_en.c
*      Purpose          : (pre-) quantization of pitch gain for MR795
*
********************************************************************************
*/

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

/*
********************************************************************************
*                         INCLUDE FILES
********************************************************************************
*/
#include "typedef.h"
#include "basic_op.h"
#include "oper_32b.h"
#include "no_count.h"
#include "cnst.h"
#include "log2.h"

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

/*************************************************************************
 *
 * FUNCTION: calc_unfilt_energies
 *
 * PURPOSE:  calculation of several energy coefficients for unfiltered
 *           excitation signals and the LTP coding gain
 *
 *       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] = <lres lres> // LTP residual energy
 *                                            // (lres = res - gain_pit*exc)
 *       ltpg = log2(LP_res_en / LTP_res_en)
 *
 *************************************************************************/
void
calc_unfilt_energies(
    Word16 res[],     /* i  : LP residual,                               Q0  */
    Word16 exc[],     /* i  : LTP excitation (unfiltered),               Q0  */
    Word16 code[],    /* i  : CB innovation (unfiltered),                Q13 */
    Word16 gain_pit,  /* i  : pitch gain,                                Q14 */
    Word16 L_subfr,   /* i  : Subframe length                                */

    Word16 frac_en[], /* o  : energy coefficients (4), fraction part,    Q15 */
    Word16 exp_en[],  /* o  : energy coefficients (4), exponent part,    Q0  */
    Word16 *ltpg      /* o  : LTP coding gain (log2()),                  Q13 */
)
{
    Word32 s, L_temp;
    Word16 i, exp, tmp;
    Word16 ltp_res_en, pred_gain;
    Word16 ltpg_exp, ltpg_frac;

    /* Compute residual energy */
    s = L_mac((Word32) 0, res[0], res[0]);
    for (i = 1; i < L_subfr; i++)
        s = L_mac(s, res[i], res[i]);

    /* ResEn := 0 if ResEn < 200.0 (= 400 Q1) */
    test();
    if (L_sub (s, 400L) < 0)
    {
        frac_en[0] = 0;                      move16 ();
        exp_en[0] = -15;                     move16 ();
    }
    else
    {
        exp = norm_l(s);
        frac_en[0] = extract_h(L_shl(s, exp));   move16 ();
        exp_en[0] = sub(15, exp);                move16 ();
    }
    
    /* Compute ltp excitation energy */
    s = L_mac((Word32) 0, exc[0], exc[0]);
    for (i = 1; i < L_subfr; i++)
        s = L_mac(s, exc[i], exc[i]);

    exp = norm_l(s);
    frac_en[1] = extract_h(L_shl(s, exp));   move16 ();
    exp_en[1] = sub(15, exp);                move16 ();

    /* Compute scalar product <exc[],code[]> */
    s = L_mac((Word32) 0, exc[0], code[0]);
    for (i = 1; i < L_subfr; i++)
        s = L_mac(s, exc[i], code[i]);

    exp = norm_l(s);
    frac_en[2] = extract_h(L_shl(s, exp));   move16 ();
    exp_en[2] = sub(16-14, exp);             move16 ();

    /* Compute energy of LTP residual */
    s = 0L;                                  move32 ();
    for (i = 0; i < L_subfr; i++)
    {
        L_temp = L_mult(exc[i], gain_pit);
        L_temp = L_shl(L_temp, 1);
        tmp = sub(res[i], round(L_temp));           /* LTP residual, Q0 */
        s = L_mac (s, tmp, tmp);
    }

    exp = norm_l(s);
    ltp_res_en = extract_h (L_shl (s, exp));
    exp = sub (15, exp);

    frac_en[3] = ltp_res_en;                 move16 ();
    exp_en[3] = exp;                         move16 ();
    
    /* calculate LTP coding gain, i.e. energy reduction LP res -> LTP res */
    test (); test ();
    if (ltp_res_en > 0 && frac_en[0] != 0)
    {
        /* gain = ResEn / LTPResEn */
        pred_gain = div_s (shr (frac_en[0], 1), ltp_res_en);
        exp = sub (exp, exp_en[0]);

        /* L_temp = ltpGain * 2^(30 + exp) */
        L_temp = L_deposit_h (pred_gain);
        /* L_temp = ltpGain * 2^27 */
        L_temp = L_shr (L_temp, add (exp, 3));

        /* Log2 = log2() + 27 */
        Log2(L_temp, &ltpg_exp, &ltpg_frac);

        /* ltpg = log2(LtpGain) * 2^13 --> range: +- 4 = +- 12 dB */
        L_temp = L_Comp (sub (ltpg_exp, 27), ltpg_frac);
        *ltpg = round (L_shl (L_temp, 13)); /* Q13 */
    }
    else
    {
        *ltpg = 0;                           move16 ();
    }
}

/*************************************************************************
 *
 * FUNCTION: calc_filt_energies
 *
 * PURPOSE:  calculation of several energy coefficients for filtered
 *           excitation signals
 *
 *     Compute coefficients need for the quantization and the optimum
 *     codebook gain gcu (for MR475 only).
 *
 *      coeff[0] =    y1 y1
 *      coeff[1] = -2 xn y1
 *      coeff[2] =    y2 y2
 *      coeff[3] = -2 xn y2
 *      coeff[4] =  2 y1 y2
 *
 *
 *      gcu = <xn2, y2> / <y2, y2> (0 if <xn2, y2> <= 0)
 *
 *     Product <y1 y1> and <xn y1> have been computed in G_pitch() and
 *     are in vector g_coeff[].
 *
 *************************************************************************/
void
calc_filt_energies(
    enum Mode mode,     /* i  : coder mode                                   */
    Word16 xn[],        /* i  : LTP target vector,                       Q0  */
    Word16 xn2[],       /* i  : CB target vector,                        Q0  */
    Word16 y1[],        /* i  : Adaptive codebook,                       Q0  */
    Word16 Y2[],        /* i  : Filtered innovative vector,              Q12 */
    Word16 g_coeff[],   /* i  : Correlations <xn y1> <y1 y1>                 */
                        /*      computed in G_pitch()                        */

    Word16 frac_coeff[],/* o  : energy coefficients (5), fraction part,  Q15 */
    Word16 exp_coeff[], /* o  : energy coefficients (5), exponent part,  Q0  */
    Word16 *cod_gain_frac,/* o: optimum codebook gain (fraction part),   Q15 */
    Word16 *cod_gain_exp  /* o: optimum codebook gain (exponent part),   Q0  */
)
{
    Word32 s, ener_init;
    Word16 i, exp, frac;
    Word16 y2[L_SUBFR];

    if (test(), sub(mode, MR795) == 0 || sub(mode, MR475) == 0)
    {
        ener_init = 0L; move32 ();
    }
    else
    {
        ener_init = 1L; move32 ();
    }
    
    for (i = 0; i < L_SUBFR; i++) {
        y2[i] = shr(Y2[i], 3);         move16 ();
    }

    frac_coeff[0] = g_coeff[0];          move16 ();
    exp_coeff[0] = g_coeff[1];           move16 ();
    frac_coeff[1] = negate(g_coeff[2]);  move16 ();   /* coeff[1] = -2 xn y1 */
    exp_coeff[1] = add(g_coeff[3], 1);   move16 ();


    /* Compute scalar product <y2[],y2[]> */

    s = L_mac(ener_init, y2[0], y2[0]);
    for (i = 1; i < L_SUBFR; i++)
        s = L_mac(s, y2[i], y2[i]);

    exp = norm_l(s);
    frac_coeff[2] = extract_h(L_shl(s, exp)); move16 ();
    exp_coeff[2] = sub(15 - 18, exp);    move16();

    /* Compute scalar product -2*<xn[],y2[]> */

    s = L_mac(ener_init, xn[0], y2[0]);
    for (i = 1; i < L_SUBFR; i++)
        s = L_mac(s, xn[i], y2[i]);

    exp = norm_l(s);
    frac_coeff[3] = negate(extract_h(L_shl(s, exp))); move16 ();
    exp_coeff[3] = sub(15 - 9 + 1, exp);         move16 ();


    /* Compute scalar product 2*<y1[],y2[]> */

    s = L_mac(ener_init, y1[0], y2[0]);
    for (i = 1; i < L_SUBFR; i++)
        s = L_mac(s, y1[i], y2[i]);

    exp = norm_l(s);
    frac_coeff[4] = extract_h(L_shl(s, exp)); move16 ();
    exp_coeff[4] = sub(15 - 9 + 1, exp);  move16();

    if (test(), test (), sub(mode, MR475) == 0 || sub(mode, MR795) == 0)
    {
        /* Compute scalar product <xn2[],y2[]> */

        s = L_mac(ener_init, xn2[0], y2[0]);
        for (i = 1; i < L_SUBFR; i++)
            s = L_mac(s, xn2[i], y2[i]);
        
        exp = norm_l(s);
        frac = extract_h(L_shl(s, exp));
        exp = sub(15 - 9, exp);

        
        if (test (), frac <= 0)
        {
            *cod_gain_frac = 0; move16 ();
            *cod_gain_exp = 0;  move16 ();
        }
        else
        {
            /*
              gcu = <xn2, y2> / c[2]
                  = (frac>>1)/frac[2]             * 2^(exp+1-exp[2])
                  = div_s(frac>>1, frac[2])*2^-15 * 2^(exp+1-exp[2])
                  = div_s * 2^(exp-exp[2]-14)
             */  
            *cod_gain_frac = div_s (shr (frac,1), frac_coeff[2]); move16 ();
            *cod_gain_exp = sub (sub (exp, exp_coeff[2]), 14);    move16 ();

        }
    }
}

/*************************************************************************
 *
 * FUNCTION: calc_target_energy
 *
 * PURPOSE:  calculation of target energy
 *
 *      en = <xn, xn>
 *
 *************************************************************************/
void
calc_target_energy(
    Word16 xn[],     /* i: LTP target vector,                       Q0  */
    Word16 *en_exp,  /* o: optimum codebook gain (exponent part),   Q0  */
    Word16 *en_frac  /* o: optimum codebook gain (fraction part),   Q15 */
)
{
    Word32 s;
    Word16 i, exp;

    /* Compute scalar product <xn[], xn[]> */
    s = L_mac(0L, xn[0], xn[0]);
    for (i = 1; i < L_SUBFR; i++)
        s = L_mac(s, xn[i], xn[i]);

    /* s = SUM 2*xn(i) * xn(i) = <xn xn> * 2 */
    exp = norm_l(s);
    *en_frac = extract_h(L_shl(s, exp));
    *en_exp = sub(16, exp);    move16();
}