AAX_FastInterpolatedTableLookup.hpp

Go to the documentation of this file.

/*================================================================================================*/
/*
 *  Copyright 2009-2015, 2023 Avid Technology, Inc.
 *  All rights reserved.
 *  
 *  CONFIDENTIAL: this document contains confidential information of Avid. Do
 *  not disclose to any third party. Use of the information contained in this
 *  document is subject to an Avid SDK license.
 */
/*================================================================================================*/
 
#ifndef AAX_FASTINTERPOLATEDTABLELOOKUP_HPP
#define AAX_FASTINTERPOLATEDTABLELOOKUP_HPP
 
#include "AAX_Quantize.h"
 
//This version requires that the lookup table is padded with one extra location so we 
//can avoid one of the checks to see if ours pointers are out of bounds.
template<class TFLOAT, class DFLOAT>
inline DFLOAT AAX_FastInterpolatedTableLookup<TFLOAT,DFLOAT>::DoTableLookupExtraFast(const TFLOAT* iTable, DFLOAT iValue) const
{
    const TFLOAT aScaledValue=(iValue - mMin)*mTableSizeM1DivMaxMinusMin;
 
    //Clamp between 0.0 and table size, so we don't go out of the table bounds.  One thing that's not obvious is this clamp was written 
    //so that it will return 0.0 if a NaN is given as the table input.  This keeps us within bounds even if we're feed garbage.;
    const TFLOAT aScaledValueLimited = aScaledValue > mTableSizeM1 ? mTableSizeM1 : (aScaledValue > 0.0f ? aScaledValue : 0.0f);
 
    int aTableIndex=AAX::FastTrunc2Int32(aScaledValueLimited);
 
    TFLOAT aFracIndex=aScaledValueLimited - TFLOAT(aTableIndex);
    TFLOAT aFracIndexM1=1.0f-aFracIndex;
 
    return (DFLOAT)(iTable[aTableIndex]*aFracIndexM1 + iTable[aTableIndex+1]*aFracIndex);
}
 
//This version requires that the lookup table is padded with one extra location so we
//can avoid one of the checks to see if ours pointers are out of bounds.
template<class TFLOAT, class DFLOAT>
inline void AAX_FastInterpolatedTableLookup<TFLOAT,DFLOAT>::DoTableLookupExtraFastMulti(const TFLOAT* iTable, DFLOAT iValue, DFLOAT* oValues) const
{
    TFLOAT aScaledValue=(iValue - mMin)*mTableSizeM1DivMaxMinusMin;
    
    //Clamp between 0.0 and table size, so we don't go out of the table bounds.  One thing that's not obvious is this clamp was written
    //so that it will return 0.0 if a NaN is given as the table input.  This keeps us within bounds even if we're feed garbage.
    const TFLOAT aScaledValueLimited = aScaledValue > mTableSizeM1 ? mTableSizeM1 : (aScaledValue > 0.0f ? aScaledValue : 0.0f);
    
    int aTableIndex=AAX::FastTrunc2Int32(aScaledValueLimited);
    
    TFLOAT aFracIndex=aScaledValueLimited - TFLOAT(aTableIndex);
    TFLOAT aFracIndexM1=1.0f-aFracIndex;
    
    aTableIndex=aTableIndex*mNumTables;
    int aTableIndexPlus1=aTableIndex+mNumTables;
    
    for(int i=0; i<mNumTables; i++)
    {
        oValues[i]=(DFLOAT)(iTable[aTableIndex++]*aFracIndexM1 + iTable[aTableIndexPlus1++]*aFracIndex);
    }
}
 
//Block version of table lookup
template<class TFLOAT, class DFLOAT>
inline void AAX_FastInterpolatedTableLookup<TFLOAT,DFLOAT>::DoTableLookupExtraFast(const TFLOAT* const iTable, const TFLOAT* const inpBuf, DFLOAT* const outBuf, int blockSize)
{
#if defined( _TMS320C6700 )
    const int r = 16;
    #pragma MUST_ITERATE(r, , r);
#endif
    for (int i = 0; i < blockSize; i++)
    {
        outBuf[i] = DoTableLookupExtraFast(iTable, inpBuf[i]);
    }
    return;
}
 
#endif // AAX_FASTINTERPOLATEDTABLELOOKUP_HPP