Mean.h

/*
  ==============================================================================
    DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 
    Copyright 2023 by sonible GmbH.
 
    This file is part of VCTR - Versatile Container Templates Reconceptualized.
 
    VCTR is free software: you can redistribute it and/or modify
    it under the terms of the GNU Lesser General Public License version 3
    only, as published by the Free Software Foundation.
 
    VCTR is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU Lesser General Public License version 3 for more details.
 
    You should have received a copy of the GNU Lesser General Public License
    version 3 along with VCTR.  If not, see <https://www.gnu.org/licenses/>.
  ==============================================================================
*/
 
namespace vctr::expressions
{
 
template <size_t extent, class SrcType>
requires is::number<ValueType<SrcType>>
class Mean : public ExpressionTemplateBase
{
public:
    using value_type = ValueType<SrcType>;
 
    VCTR_COMMON_UNARY_EXPRESSION_MEMBERS (Mean, src)
 
    static constexpr value_type reductionResultInitValue = 0;
 
    VCTR_FORCEDINLINE constexpr void reduceElementWise (value_type& result, size_t i) const
    {
        result += src[i];
    }
 
    //==============================================================================
    VCTR_FORCEDINLINE value_type reduceVectorOp() const
    requires is::suitableForAccelerateRealFloatVectorReductionOp<SrcType, value_type, detail::dontPreferIfIppAndAccelerateAreAvailable>
    {
        return Expression::Accelerate::mean (src.data(), size());
    };
 
    VCTR_FORCEDINLINE value_type reduceVectorOp() const
    requires is::suitableForIppRealOrComplexFloatVectorReductionOp<SrcType, value_type, detail::preferIfIppAndAccelerateAreAvailable>
    {
        return Expression::IPP::mean (src.data(), sizeToInt (size()));
    };
 
    //==============================================================================
    VCTR_FORCEDINLINE void reduceNeonRegisterWise (NeonRegister<value_type>& result, size_t i) const
    requires Config::archARM && has::getNeon<SrcType> && (is::realFloatNumber<value_type> || is::int32Number<value_type>)
    {
        result = Expression::Neon::add (result, src.getNeon (i));
    }
 
    VCTR_FORCEDINLINE VCTR_TARGET ("avx") void reduceAVXRegisterWise (AVXRegister<value_type>& result, size_t i) const
    requires Config::archX64 && has::getAVX<SrcType> && is::realFloatNumber<value_type>
    {
        result = Expression::AVX::add (result, src.getAVX (i));
    }
 
    VCTR_FORCEDINLINE VCTR_TARGET ("avx2") void reduceAVXRegisterWise (AVXRegister<value_type>& result, size_t i) const
    requires Config::archX64 && has::getAVX<SrcType> && is::int32Number<value_type>
    {
        result = Expression::AVX::add (result, src.getAVX (i));
    }
 
    VCTR_FORCEDINLINE VCTR_TARGET ("sse4.1") void reduceSSERegisterWise (SSERegister<value_type>& result, size_t i) const
    requires Config::archX64 && has::getSSE<SrcType> && (is::realFloatNumber<value_type> || is::int32Number<value_type>)
    {
        result = Expression::SSE::add (result, src.getSSE (i));
    }
 
    //==============================================================================
    template <size_t n>
    VCTR_FORCEDINLINE constexpr value_type finalizeReduction (const std::array<value_type, n>& sums) const
    {
        auto sum = n == 1 ? sums[0] : std::reduce (sums.begin(), sums.end());
 
        return value_type (sum / FloatType<RealType<value_type>> (src.size()));
    }
};
 
template <size_t extent, class SrcType>
requires is::number<ValueType<SrcType>>
class MeanSquare : public ExpressionTemplateBase
{
public:
    using value_type = ValueType<SrcType>;
 
    VCTR_COMMON_UNARY_EXPRESSION_MEMBERS (MeanSquare, src)
 
    static constexpr value_type reductionResultInitValue = 0;
 
    VCTR_FORCEDINLINE constexpr void reduceElementWise (value_type& result, size_t i) const
    {
        auto s = src[i];
        result += s * s;
    }
 
    //==============================================================================
    VCTR_FORCEDINLINE value_type reduceVectorOp() const
    requires is::suitableForAccelerateRealFloatVectorReductionOp<SrcType, value_type, detail::preferIfIppAndAccelerateAreAvailable>
    {
        return Expression::Accelerate::meanSquare (src.data(), size());
    };
 
    VCTR_FORCEDINLINE value_type reduceVectorOp() const
    requires is::suitableForIppRealFloatVectorReductionOp<SrcType, value_type, detail::dontPreferIfIppAndAccelerateAreAvailable>
    {
        auto l2Norm = Expression::IPP::l2Norm (src.data(), sizeToInt (size()));
        return (l2Norm * l2Norm) / value_type (size());
    };
 
    //==============================================================================
    VCTR_FORCEDINLINE void reduceNeonRegisterWise (NeonRegister<value_type>& result, size_t i) const
    requires Config::archARM && has::getNeon<SrcType> && (is::realFloatNumber<value_type> || is::int32Number<value_type>)
    {
        auto s = src.getNeon (i);
        s = Expression::Neon::mul (s, s);
        result = Expression::Neon::add (result, s);
    }
 
    VCTR_FORCEDINLINE VCTR_TARGET ("avx") void reduceAVXRegisterWise (AVXRegister<value_type>& result, size_t i) const
    requires Config::archX64 && has::getAVX<SrcType> && is::realFloatNumber<value_type>
    {
        auto s = src.getAVX (i);
        s = Expression::AVX::mul (s, s);
        result = Expression::AVX::add (result, s);
    }
 
    VCTR_FORCEDINLINE VCTR_TARGET ("sse4.1") void reduceSSERegisterWise (SSERegister<value_type>& result, size_t i) const
    requires Config::archX64 && has::getSSE<SrcType> && is::realFloatNumber<value_type>
    {
        auto s = src.getSSE (i);
        s = Expression::SSE::mul (s, s);
        result = Expression::SSE::add (result, s);
    }
 
    //==============================================================================
    template <size_t n>
    VCTR_FORCEDINLINE constexpr value_type finalizeReduction (const std::array<value_type, n>& sums) const
    {
        auto sum = n == 1 ? sums[0] : std::reduce (sums.begin(), sums.end());
 
        return value_type (sum / FloatType<RealType<value_type>> (size()));
    }
};
 
template <size_t extent, class SrcType>
requires is::realOrComplexFloatNumber<ValueType<SrcType>>
class RootMeanSquare : public ExpressionTemplateBase
{
public:
    using value_type = ValueType<SrcType>;
 
    VCTR_COMMON_UNARY_EXPRESSION_MEMBERS (RootMeanSquare, src)
 
    static constexpr value_type reductionResultInitValue = 0;
 
    VCTR_FORCEDINLINE constexpr void reduceElementWise (value_type& result, size_t i) const
    {
        auto s = src[i];
        result += s * s;
    }
 
    //==============================================================================
    VCTR_FORCEDINLINE value_type reduceVectorOp() const
    requires is::suitableForAccelerateRealFloatVectorReductionOp<SrcType, value_type, detail::preferIfIppAndAccelerateAreAvailable>
    {
        return Expression::Accelerate::rms (src.data(), size());
    };
 
    VCTR_FORCEDINLINE value_type reduceVectorOp() const
    requires is::suitableForIppRealFloatVectorReductionOp<SrcType, value_type, detail::dontPreferIfIppAndAccelerateAreAvailable>
    {
        auto l2Norm = Expression::IPP::l2Norm (src.data(), sizeToInt (size()));
        return std::sqrt ((l2Norm * l2Norm) / value_type (size()));
    };
 
    //==============================================================================
    VCTR_FORCEDINLINE void reduceNeonRegisterWise (NeonRegister<value_type>& result, size_t i) const
    requires Config::archARM && has::getNeon<SrcType> && (is::realFloatNumber<value_type> || is::int32Number<value_type>)
    {
        auto s = src.getNeon (i);
        s = Expression::Neon::mul (s, s);
        result = Expression::Neon::add (result, s);
    }
 
    VCTR_FORCEDINLINE VCTR_TARGET ("avx") void reduceAVXRegisterWise (AVXRegister<value_type>& result, size_t i) const
    requires Config::archX64 && has::getAVX<SrcType> && is::realFloatNumber<value_type>
    {
        auto s = src.getAVX (i);
        s = Expression::AVX::mul (s, s);
        result = Expression::AVX::add (result, s);
    }
 
    VCTR_FORCEDINLINE VCTR_TARGET ("sse4.1") void reduceSSERegisterWise (SSERegister<value_type>& result, size_t i) const
    requires Config::archX64 && has::getSSE<SrcType> && is::realFloatNumber<value_type>
    {
        auto s = src.getSSE (i);
        s = Expression::SSE::mul (s, s);
        result = Expression::SSE::add (result, s);
    }
 
    //==============================================================================
    template <size_t n>
    VCTR_FORCEDINLINE constexpr value_type finalizeReduction (const std::array<value_type, n>& squaredSums) const
    {
        auto squaredSum = n == 1 ? squaredSums[0] : std::reduce (squaredSums.begin(), squaredSums.end());
        auto meanSquaredSum = squaredSum / RealType<value_type> (src.size());
 
    #if VCTR_USE_GCEM
        if constexpr (is::realFloatNumber<value_type>)
        {
            if (std::is_constant_evaluated())
            {
                return value_type (gcem::sqrt (meanSquaredSum));
            }
        }
    #endif
 
        return value_type (std::sqrt (meanSquaredSum));
    }
};
 
} // namespace vctr::expressions
 
namespace vctr
{
 
constexpr inline ExpressionChainBuilder<expressions::Mean> mean;
 
constexpr inline ExpressionChainBuilder<expressions::MeanSquare> meanSquare;
 
constexpr inline ExpressionChainBuilder<expressions::RootMeanSquare> rms;
 
} // namespace vctr