pitch_ol.cpp

实现3GPP的GSM中AMR语音的CODECS。

/* ------------------------------------------------------------------
 * Copyright (C) 2008 PacketVideo
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either
 * express or implied.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 * -------------------------------------------------------------------
 */
/****************************************************************************************
Portions of this file are derived from the following 3GPP standard:

    3GPP TS 26.073
    ANSI-C code for the Adaptive Multi-Rate (AMR) speech codec
    Available from http://www.3gpp.org

(C) 2004, 3GPP Organizational Partners (ARIB, ATIS, CCSA, ETSI, TTA, TTC)
Permission to distribute, modify and use this file under the standard license
terms listed above has been obtained from the copyright holder.
****************************************************************************************/
/*
------------------------------------------------------------------------------



 Pathname: ./audio/gsm-amr/c/src/pitch_ol.c
 Funtions: Pitch_ol
           Lag_max

     Date: 06/15/2000
------------------------------------------------------------------------------
 REVISION HISTORY

 Description: Placed into PV template and began optimization.

 Description: Made changes based on review meeting.

 Description: Fixed bug in Pitch_ol that causes function to not be bit-exact
          with the original version. This was done by adding an extra
          call to mult after the first IF statement when comparing the
          three section maximums (towards the end of the function). Fixed
          tabs. Synced-up Lag_max_wrapper with template.

 Description: Synchronized file with UMTS version 3.2.0. Updated coding
              template. Removed unnecessary include file.

 Description: Added code that compares max1 to THRESHOLD value prior to
              comparison with max3.

 Description: Replaced basic_op.h and oper_32b.h with the header files of the
              math functions used in this file. Fixed typecasting issue with
              the TI C compiler.

 Description: Made the following changes per comments from Phase 2/3 review:
              1. Defined one local variable per line.
              2. Added more comments to the code.

 Description: Passing in pointer to overflow flag for EPOC compatibility

 Description:  For Lag_max() and Pitch_ol()
              1. Eliminated unused include files.
              2. Replaced array addressing by pointers
              3. Eliminated math operations that unnecessary checked for
                 saturation, in some cases this by shifting before adding and
                 in other cases by evaluating the operands
              4. Unrolled loops to speed up processing, use decrement loops
              5. Eliminated if-else statements for sign extension when
                 right-shifting
              6. Replaced for-loop with memcpy()

 Description:  Replaced OSCL mem type functions and eliminated include
               files that now are chosen by OSCL definitions

 Description:  Replaced "int" and/or "char" with OSCL defined types.

 Description: Added l_add.h in Include section.

 Description: Using intrinsics from fxp_arithmetic.h .

 Description: Replacing fxp_arithmetic.h with basic_op.h.

 Description:

------------------------------------------------------------------------------
 MODULE DESCRIPTION

 The modules in this file compute the open loop pitch lag.
------------------------------------------------------------------------------
*/


/*----------------------------------------------------------------------------
; INCLUDES
----------------------------------------------------------------------------*/
#include "pitch_ol.h"
#include "typedef.h"
#include "basicop_malloc.h"
#include "cnst.h"
#include "inv_sqrt.h"
#include "vad.h"
#include "calc_cor.h"
#include "hp_max.h"
#include "oscl_mem.h"
#include "basic_op.h"

/*----------------------------------------------------------------------------
; MACROS
; Define module specific macros here
----------------------------------------------------------------------------*/


/*----------------------------------------------------------------------------
; DEFINES
; Include all pre-processor statements here. Include conditional
; compile variables also.
----------------------------------------------------------------------------*/
#define THRESHOLD 27853

/*----------------------------------------------------------------------------
; LOCAL FUNCTION DEFINITIONS
; Function Prototype declaration
----------------------------------------------------------------------------*/

/*----------------------------------------------------------------------------
; LOCAL VARIABLE DEFINITIONS
; Variable declaration - defined here and used outside this module
----------------------------------------------------------------------------*/


/*
------------------------------------------------------------------------------
 FUNCTION NAME: Lag_max
------------------------------------------------------------------------------
 INPUT AND OUTPUT DEFINITIONS (If VAD2 is defined)

 Inputs
    corr = pointer to buffer of correlation values (Word32)
    scal_sig = pointer to buffer of scaled signal values (Word16)
    scal_fac = scaled signal factor (Word16)
    scal_flag = EFR compatible scaling flag (Word16)
    L_frame = length of frame to compute pitch (Word16)
    lag_max = maximum lag (Word16)
    lag_min = minimum lag (Word16)
    cor_max = pointer to the normalized correlation of selected lag (Word16)
    rmax = pointer to max(<s[i]*s[j]>), (Word32)
    r0 = pointer to the residual energy (Word32)
    dtx  = dtx flag; equal to 1, if dtx is enabled, 0, otherwise (Flag)

 Outputs:
    cor_max contains the newly calculated normalized correlation of the
      selected lag
    rmax contains the newly calculated max(<s[i]*s[j]>)
    r0 contains the newly calculated residual energy

 Returns:
    p_max = lag of the max correlation found (Word16)

 Global Variables Used:
    None.

 Local Variables Needed:
    None.

------------------------------------------------------------------------------
 INPUT AND OUTPUT DEFINITIONS (If VAD2 is not defined)

 Inputs
    vadSt = pointer to a vadState structure
    corr = pointer to buffer of correlation values (Word32)
    scal_sig = pointer to buffer of scaled signal values (Word16)
    scal_fac = scaled signal factor (Word16)
    scal_flag = EFR compatible scaling flag (Word16)
    L_frame = length of frame to compute pitch (Word16)
    lag_max = maximum lag (Word16)
    lag_min = minimum lag (Word16)
    cor_max = pointer to the normalized correlation of selected lag (Word16)
    dtx  = dtx flag; equal to 1, if dtx is enabled, 0, otherwise (Flag)
    pOverflow = pointer to overflow indicator (Flag)

 Outputs:
    cor_max contains the newly calculated normalized correlation of the
      selected lag
    vadSt contains the updated VAD state parameters
    pOverflow -> 1 if the math operations called by this routine saturate

 Returns:
    p_max = lag of the max correlation found (Word16)

 Global Variables Used:
    None.

 Local Variables Needed:
    None.

------------------------------------------------------------------------------
 FUNCTION DESCRIPTION

 Find the lag that has maximum correlation of scal_sig in a given delay range.
 The correlation is given by:

         cor[t] = <scal_sig[n],scal_sig[n-t]>,  t=lag_min,...,lag_max

 The function returns the maximum correlation after normalization and the
 corresponding lag.

------------------------------------------------------------------------------
 REQUIREMENTS

 None.

------------------------------------------------------------------------------
 REFERENCES

 pitch_ol.c, UMTS GSM AMR speech codec, R99 - Version 3.2.0, March 2, 2001

------------------------------------------------------------------------------
 PSEUDO-CODE

#ifdef VAD2
static Word16 Lag_max ( // o   : lag found
    Word32 corr[],      // i   : correlation vector.
    Word16 scal_sig[],  // i   : scaled signal.
    Word16 scal_fac,    // i   : scaled signal factor.
    Word16 scal_flag,   // i   : if 1 use EFR compatible scaling
    Word16 L_frame,     // i   : length of frame to compute pitch
    Word16 lag_max,     // i   : maximum lag
    Word16 lag_min,     // i   : minimum lag
    Word16 *cor_max,    // o   : normalized correlation of selected lag
    Word32 *rmax,       // o   : max(<s[i]*s[j]>)
    Word32 *r0,         // o   : residual energy
    Flag dtx            // i   : dtx flag; use dtx=1, do not use dtx=0
    )
#else
static Word16 Lag_max ( // o   : lag found
    vadState *vadSt,    // i/o : VAD state struct
    Word32 corr[],      // i   : correlation vector.
    Word16 scal_sig[],  // i   : scaled signal.
    Word16 scal_fac,    // i   : scaled signal factor.
    Word16 scal_flag,   // i   : if 1 use EFR compatible scaling
    Word16 L_frame,     // i   : length of frame to compute pitch
    Word16 lag_max,     // i   : maximum lag
    Word16 lag_min,     // i   : minimum lag
    Word16 *cor_max,    // o   : normalized correlation of selected lag
    Flag dtx            // i   : dtx flag; use dtx=1, do not use dtx=0
    )
#endif
{
    Word16 i, j;
    Word16 *p;
    Word32 max, t0;
    Word16 max_h, max_l, ener_h, ener_l;
    Word16 p_max = 0; // initialization only needed to keep gcc silent

    max = MIN_32;
    p_max = lag_max;

    for (i = lag_max, j = (PIT_MAX-lag_max-1); i >= lag_min; i--, j--)
    {
       if (L_sub (corr[-i], max) >= 0)
       {
          max = corr[-i];
          p_max = i;
       }
    }

    // compute energy

    t0 = 0;
    p = &scal_sig[-p_max];
    for (i = 0; i < L_frame; i++, p++)
    {
        t0 = L_mac (t0, *p, *p);
    }
    // 1/sqrt(energy)

    if (dtx)
    {  // no test() call since this if is only in simulation env
#ifdef VAD2
       *rmax = max;
       *r0 = t0;
#else
       // check tone
       vad_tone_detection (vadSt, max, t0);
#endif
    }

    t0 = Inv_sqrt (t0);

    if (scal_flag)
    {
       t0 = L_shl (t0, 1);
    }

    // max = max/sqrt(energy)

    L_Extract (max, &max_h, &max_l);
    L_Extract (t0, &ener_h, &ener_l);

    t0 = Mpy_32 (max_h, max_l, ener_h, ener_l);

    if (scal_flag)
    {
      t0 = L_shr (t0, scal_fac);
      *cor_max = extract_h (L_shl (t0, 15)); // divide by 2
    }
    else
    {
      *cor_max = extract_l(t0);
    }

    return (p_max);
}

------------------------------------------------------------------------------
 RESOURCES USED [optional]

 When the code is written for a specific target processor the
 the resources used should be documented below.

 HEAP MEMORY USED: x bytes

 STACK MEMORY USED: x bytes

 CLOCK CYCLES: (cycle count equation for this function) + (variable
                used to represent cycle count for each subroutine
                called)
     where: (cycle count variable) = cycle count for [subroutine
                                     name]

------------------------------------------------------------------------------
 CAUTION [optional]
 [State any special notes, constraints or cautions for users of this function]

------------------------------------------------------------------------------
*/

#ifdef VAD2
static Word16 Lag_max(  /* o   : lag found                               */
    Word32 corr[],      /* i   : correlation vector.                     */
    Word16 scal_sig[],  /* i   : scaled signal.                          */
    Word16 scal_fac,    /* i   : scaled signal factor.                   */
    Word16 scal_flag,   /* i   : if 1 use EFR compatible scaling         */
    Word16 L_frame,     /* i   : length of frame to compute pitch        */
    Word16 lag_max,     /* i   : maximum lag                             */
    Word16 lag_min,     /* i   : minimum lag                             */
    Word16 *cor_max,    /* o   : normalized correlation of selected lag  */
    Word32 *rmax,       /* o   : max(<s[i]*s[j]>)                        */
    Word32 *r0,         /* o   : residual energy                         */
    Flag dtx,           /* i   : dtx flag; use dtx=1, do not use dtx=0   */
    Flag *pOverflow     /* i/o : overflow Flag                           */
)
#else
static Word16 Lag_max(  /* o   : lag found                               */
    vadState *vadSt,    /* i/o : VAD state struct                        */
    Word32 corr[],      /* i   : correlation vector.                     */
    Word16 scal_sig[],  /* i   : scaled signal.                          */
    Word16 scal_fac,    /* i   : scaled signal factor.                   */
    Word16 scal_flag,   /* i   : if 1 use EFR compatible scaling         */
    Word16 L_frame,     /* i   : length of frame to compute pitch        */
    Word16 lag_max,     /* i   : maximum lag                             */
    Word16 lag_min,     /* i   : minimum lag                             */
    Word16 *cor_max,    /* o   : normalized correlation of selected lag  */
    Flag dtx,           /* i   : dtx flag; use dtx=1, do not use dtx=0   */
    Flag *pOverflow     /* i/o : overflow Flag                           */
)
#endif
{
    register Word16 i;
    Word16 *p;
    Word32 max;
    Word32 t0;
    Word16 max_h;
    Word16 max_l;
    Word16 ener_h;
    Word16 ener_l;
    Word16 p_max = 0; /* initialization only needed to keep gcc silent */
    Word32 L_temp;
    Word32 L_temp_2;
    Word32 L_temp_3;
    Word32  *p_corr = &corr[-lag_max];

    max = MIN_32;
    p_max = lag_max;

    for (i = lag_max; i >= lag_min; i--)
    {
        /* The negative array index is equivalent to a negative */
        /* address offset, i.e., corr[-i] == *(corr - i)        */
        if (*(p_corr++) >= max)
        {
            p_corr--;
            max = *(p_corr++);
            p_max = i;
        }
    }

    /* compute energy */

    t0 = 0;

    /* The negative array index is equivalent to a negative          */
    /* address offset, i.e., scal_sig[-p_max] == *(scal_sig - p_max) */
    p = &scal_sig[-p_max];
    for (i = (L_frame >> 2); i != 0; i--)
    {
        t0 = amrnb_fxp_mac_16_by_16bb((Word32) * (p), (Word32) * (p), t0);
        p++;
        t0 = amrnb_fxp_mac_16_by_16bb((Word32) * (p), (Word32) * (p), t0);
        p++;
        t0 = amrnb_fxp_mac_16_by_16bb((Word32) * (p), (Word32) * (p), t0);
        p++;
        t0 = amrnb_fxp_mac_16_by_16bb((Word32) * (p), (Word32) * (p), t0);
        p++;