Improve accuracy of SIMD exp for small args

[gromacs.git] / src / gromacs / simd / impl_reference / impl_reference_simd_double.h

/*
 * This file is part of the GROMACS molecular simulation package.
 *
 * Copyright (c) 2014,2015,2016,2017, by the GROMACS development team, led by
 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
 * and including many others, as listed in the AUTHORS file in the
 * top-level source directory and at http://www.gromacs.org.
 *
 * GROMACS is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public License
 * as published by the Free Software Foundation; either version 2.1
 * of the License, or (at your option) any later version.
 *
 * GROMACS 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 for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with GROMACS; if not, see
 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
 *
 * If you want to redistribute modifications to GROMACS, please
 * consider that scientific software is very special. Version
 * control is crucial - bugs must be traceable. We will be happy to
 * consider code for inclusion in the official distribution, but
 * derived work must not be called official GROMACS. Details are found
 * in the README & COPYING files - if they are missing, get the
 * official version at http://www.gromacs.org.
 *
 * To help us fund GROMACS development, we humbly ask that you cite
 * the research papers on the package. Check out http://www.gromacs.org.
 */

#ifndef GMX_SIMD_IMPL_REFERENCE_SIMD_DOUBLE_H
#define GMX_SIMD_IMPL_REFERENCE_SIMD_DOUBLE_H

/*! \libinternal \file
 *
 * \brief Reference implementation, SIMD double precision.
 *
 * \author Erik Lindahl <erik.lindahl@scilifelab.se>
 *
 * \ingroup module_simd
 */

#include "config.h"

#include <cassert>
#include <cmath>
#include <cstddef>
#include <cstdint>

#include <algorithm>
#include <array>

#include "gromacs/math/utilities.h"
#include "gromacs/utility/fatalerror.h"

#include "impl_reference_definitions.h"
#include "impl_reference_simd_float.h"

namespace gmx
{

/*! \cond libapi */
/*! \addtogroup module_simd */
/*! \{ */

/* \name SIMD implementation data types
 * \{
 */

/*! \libinternal \brief Double SIMD variable. Available if GMX_SIMD_HAVE_DOUBLE is 1.
 *
 * \note This variable cannot be placed inside other structures or classes, since
 *       some compilers (including at least clang-3.7) appear to lose the
 *       alignment. This is likely particularly severe when allocating such
 *       memory on the heap, but it occurs for stack structures too.
 */
class SimdDouble
{
    public:
        SimdDouble() {}

        //! \brief Construct from scalar
        SimdDouble(double d) { simdInternal_.fill(d); }

        /*! \brief Internal SIMD data. Implementation dependent, don't touch.
         *
         * This has to be public to enable usage in combination with static inline
         * functions, but it should never, EVER, be accessed by any code outside
         * the corresponding implementation directory since the type will depend
         * on the architecture.
         */
        std::array<double, GMX_SIMD_DOUBLE_WIDTH>  simdInternal_;
};

/*! \libinternal \brief Integer SIMD variable type to use for conversions to/from double.
 *
 * Available if GMX_SIMD_HAVE_DOUBLE is 1.
 *
 * \note The integer SIMD type will always be available, but on architectures
 * that do not have any real integer SIMD support it might be defined as the
 * floating-point type. This will work fine, since there are separate defines
 * for whether the implementation can actually do any operations on integer
 * SIMD types.
 *
 * \note This variable cannot be placed inside other structures or classes, since
 *       some compilers (including at least clang-3.7) appear to lose the
 *       alignment. This is likely particularly severe when allocating such
 *       memory on the heap, but it occurs for stack structures too.
 */
class SimdDInt32
{
    public:
        SimdDInt32() {}

        //! \brief Construct from scalar
        SimdDInt32(std::int32_t i) { simdInternal_.fill(i); }

        /*! \brief Internal SIMD data. Implementation dependent, don't touch.
         *
         * This has to be public to enable usage in combination with static inline
         * functions, but it should never, EVER, be accessed by any code outside
         * the corresponding implementation directory since the type will depend
         * on the architecture.
         */
        std::array<std::int32_t, GMX_SIMD_DINT32_WIDTH>  simdInternal_;
};

/*! \libinternal \brief Boolean type for double SIMD data.
 *
 *  Available if GMX_SIMD_HAVE_DOUBLE is 1.
 *
 * \note This variable cannot be placed inside other structures or classes, since
 *       some compilers (including at least clang-3.7) appear to lose the
 *       alignment. This is likely particularly severe when allocating such
 *       memory on the heap, but it occurs for stack structures too.
 */
class SimdDBool
{
    public:
        SimdDBool() {}

        //! \brief Construct from scalar bool
        SimdDBool(bool b) { simdInternal_.fill(b); }

        /*! \brief Internal SIMD data. Implementation dependent, don't touch.
         *
         * This has to be public to enable usage in combination with static inline
         * functions, but it should never, EVER, be accessed by any code outside
         * the corresponding implementation directory since the type will depend
         * on the architecture.
         */
        std::array<bool, GMX_SIMD_DOUBLE_WIDTH>  simdInternal_;
};

/*! \libinternal \brief Boolean type for integer datatypes corresponding to double SIMD.
 *
 * Available if GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \note This variable cannot be placed inside other structures or classes, since
 *       some compilers (including at least clang-3.7) appear to lose the
 *       alignment. This is likely particularly severe when allocating such
 *       memory on the heap, but it occurs for stack structures too.
 */
class SimdDIBool
{
    public:
        SimdDIBool() {}

        //! \brief Construct from scalar
        SimdDIBool(bool b) { simdInternal_.fill(b); }

        /*! \brief Internal SIMD data. Implementation dependent, don't touch.
         *
         * This has to be public to enable usage in combination with static inline
         * functions, but it should never, EVER, be accessed by any code outside
         * the corresponding implementation directory since the type will depend
         * on the architecture.
         */
        std::array<bool, GMX_SIMD_DINT32_WIDTH>  simdInternal_;
};

/*! \}
 *
 * \name SIMD implementation load/store operations for double precision floating point
 * \{
 */

/*! \brief Load \ref GMX_SIMD_DOUBLE_WIDTH numbers from aligned memory.
 *
 * \param m Pointer to memory aligned to the SIMD width.
 * \return SIMD variable with data loaded.
 */
static inline SimdDouble gmx_simdcall
simdLoad(const double *m)
{
    SimdDouble a;

    assert(std::size_t(m) % (a.simdInternal_.size()*sizeof(double)) == 0);

    std::copy(m, m+a.simdInternal_.size(), a.simdInternal_.begin());
    return a;
}

/*! \brief Store the contents of SIMD double variable to aligned memory m.
 *
 * \param[out] m Pointer to memory, aligned to SIMD width.
 * \param a SIMD variable to store
 */
static inline void gmx_simdcall
store(double *m, SimdDouble a)
{
    assert(std::size_t(m) % (a.simdInternal_.size()*sizeof(double)) == 0);

    std::copy(a.simdInternal_.begin(), a.simdInternal_.end(), m);
}

/*! \brief Load SIMD double from unaligned memory.
 *
 * Available if \ref GMX_SIMD_HAVE_LOADU is 1.
 *
 * \param m Pointer to memory, no alignment requirement.
 * \return SIMD variable with data loaded.
 */
static inline SimdDouble gmx_simdcall
simdLoadU(const double *m)
{
    SimdDouble a;
    std::copy(m, m+a.simdInternal_.size(), a.simdInternal_.begin());
    return a;
}

/*! \brief Store SIMD double to unaligned memory.
 *
 * Available if \ref GMX_SIMD_HAVE_STOREU is 1.
 *
 * \param[out] m Pointer to memory, no alignment requirement.
 * \param a SIMD variable to store.
 */
static inline void gmx_simdcall
storeU(double *m, SimdDouble a)
{
    std::copy(a.simdInternal_.begin(), a.simdInternal_.end(), m);
}

/*! \brief Set all SIMD double variable elements to 0.0.
 *
 * You should typically just call \ref gmx::setZero(), which uses proxy objects
 * internally to handle all types rather than adding the suffix used here.
 *
 * \return SIMD 0.0
 */
static inline SimdDouble gmx_simdcall
setZeroD()
{
    return SimdDouble(0.0);
}

/*! \}
 *
 * \name SIMD implementation load/store operations for integers (corresponding to double)
 * \{
 */

/*! \brief Load aligned SIMD integer data, width corresponds to \ref gmx::SimdDouble.
 *
 * You should typically just call \ref gmx::load(), which uses proxy objects
 * internally to handle all types rather than adding the suffix used here.
 *
 * \param m Pointer to memory, aligned to (double) integer SIMD width.
 * \return SIMD integer variable.
 */
static inline SimdDInt32 gmx_simdcall
simdLoadDI(const std::int32_t * m)
{
    SimdDInt32 a;

    assert(std::size_t(m) % (a.simdInternal_.size()*sizeof(std::int32_t)) == 0);

    std::copy(m, m+a.simdInternal_.size(), a.simdInternal_.begin());
    return a;
};

/*! \brief Store aligned SIMD integer data, width corresponds to \ref gmx::SimdDouble.
 *
 * \param m Memory aligned to (double) integer SIMD width.
 * \param a SIMD (double) integer variable to store.
 */
static inline void gmx_simdcall
store(std::int32_t * m, SimdDInt32 a)
{
    assert(std::size_t(m) % (a.simdInternal_.size()*sizeof(std::int32_t)) == 0);

    std::copy(a.simdInternal_.begin(), a.simdInternal_.end(), m);
};

/*! \brief Load unaligned integer SIMD data, width corresponds to \ref gmx::SimdDouble.
 *
 * You should typically just call \ref gmx::loadU(), which uses proxy objects
 * internally to handle all types rather than adding the suffix used here.
 *
 * Available if \ref GMX_SIMD_HAVE_LOADU is 1.
 *
 * \param m Pointer to memory, no alignment requirements.
 * \return SIMD integer variable.
 */
static inline SimdDInt32 gmx_simdcall
simdLoadUDI(const std::int32_t *m)
{
    SimdDInt32 a;
    std::copy(m, m+a.simdInternal_.size(), a.simdInternal_.begin());
    return a;
}

/*! \brief Store unaligned SIMD integer data, width corresponds to \ref gmx::SimdDouble.
 *
 * Available if \ref GMX_SIMD_HAVE_STOREU is 1.
 *
 * \param m Memory pointer, no alignment requirements.
 * \param a SIMD (double) integer variable to store.
 */
static inline void gmx_simdcall
storeU(std::int32_t * m, SimdDInt32 a)
{
    std::copy(a.simdInternal_.begin(), a.simdInternal_.end(), m);
}

/*! \brief Set all SIMD (double) integer variable elements to 0.
 *
 * You should typically just call \ref gmx::setZero(), which uses proxy objects
 * internally to handle all types rather than adding the suffix used here.
 *
 * \return SIMD 0
 */
static inline SimdDInt32 gmx_simdcall
setZeroDI()
{
    return SimdDInt32(0);
}

/*! \brief Extract element with index i from \ref gmx::SimdDInt32.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_EXTRACT is 1.
 *
 * \tparam index Compile-time constant, position to extract (first position is 0)
 * \param  a     SIMD variable from which to extract value.
 * \return Single integer from position index in SIMD variable.
 */
template<int index>
static inline std::int32_t gmx_simdcall
extract(SimdDInt32 a)
{
    return a.simdInternal_[index];
}

/*! \}
 *
 * \name SIMD implementation double precision floating-point bitwise logical operations
 * \{
 */

/*! \brief Bitwise and for two SIMD double variables.
 *
 * Supported if \ref GMX_SIMD_HAVE_LOGICAL is 1.
 *
 * \param a data1
 * \param b data2
 * \return data1 & data2
 */
static inline SimdDouble gmx_simdcall
operator&(SimdDouble a, SimdDouble b)
{
    SimdDouble         res;

    union
    {
        double        r;
        std::int64_t  i;
    }
    conv1, conv2;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        conv1.r              = a.simdInternal_[i];
        conv2.r              = b.simdInternal_[i];
        conv1.i              = conv1.i & conv2.i;
        res.simdInternal_[i] = conv1.r;
    }
    return res;
}

/*! \brief Bitwise andnot for SIMD double.
 *
 * Available if \ref GMX_SIMD_HAVE_LOGICAL is 1.
 *
 * \param a data1
 * \param b data2
 * \return (~data1) & data2
 */
static inline SimdDouble gmx_simdcall
andNot(SimdDouble a, SimdDouble b)
{
    SimdDouble         res;

    union
    {
        double        r;
        std::int64_t  i;
    }
    conv1, conv2;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        conv1.r              = a.simdInternal_[i];
        conv2.r              = b.simdInternal_[i];
        conv1.i              = ~conv1.i & conv2.i;
        res.simdInternal_[i] = conv1.r;
    }
    return res;
}

/*! \brief Bitwise or for SIMD double.
 *
 * Available if \ref GMX_SIMD_HAVE_LOGICAL is 1.
 *
 * \param a data1
 * \param b data2
 * \return data1 | data2
 */
static inline SimdDouble gmx_simdcall
operator|(SimdDouble a, SimdDouble b)
{
    SimdDouble         res;

    union
    {
        double        r;
        std::int64_t  i;
    }
    conv1, conv2;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        conv1.r              = a.simdInternal_[i];
        conv2.r              = b.simdInternal_[i];
        conv1.i              = conv1.i | conv2.i;
        res.simdInternal_[i] = conv1.r;
    }
    return res;
}

/*! \brief Bitwise xor for SIMD double.
 *
 * Available if \ref GMX_SIMD_HAVE_LOGICAL is 1.
 *
 * \param a data1
 * \param b data2
 * \return data1 ^ data2
 */
static inline SimdDouble gmx_simdcall
operator^(SimdDouble a, SimdDouble b)
{
    SimdDouble         res;

    union
    {
        double        r;
        std::int64_t  i;
    }
    conv1, conv2;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        conv1.r              = a.simdInternal_[i];
        conv2.r              = b.simdInternal_[i];
        conv1.i              = conv1.i ^ conv2.i;
        res.simdInternal_[i] = conv1.r;
    }
    return res;
}

/*! \}
 *
 * \name SIMD implementation double precision floating-point arithmetics
 * \{
 */

/*! \brief Add two double SIMD variables.
 *
 * \param a term1
 * \param b term2
 * \return a+b
 */
static inline SimdDouble gmx_simdcall
operator+(SimdDouble a, SimdDouble b)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] + b.simdInternal_[i];
    }
    return res;
}

/*! \brief Subtract two double SIMD variables.
 *
 * \param a term1
 * \param b term2
 * \return a-b
 */
static inline SimdDouble gmx_simdcall
operator-(SimdDouble a, SimdDouble b)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] - b.simdInternal_[i];
    }
    return res;
}

/*! \brief SIMD double precision negate.
 *
 * \param a SIMD double precision value
 * \return -a
 */
static inline SimdDouble gmx_simdcall
operator-(SimdDouble a)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = -a.simdInternal_[i];
    }
    return res;
}

/*! \brief Multiply two double SIMD variables.
 *
 * \param a factor1
 * \param b factor2
 * \return a*b.
 */
static inline SimdDouble gmx_simdcall
operator*(SimdDouble a, SimdDouble b)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] * b.simdInternal_[i];
    }
    return res;
}

/*! \brief SIMD double Fused-multiply-add. Result is a*b+c.
 *
 * \param a factor1
 * \param b factor2
 * \param c term
 * \return a*b+c
 */
static inline SimdDouble gmx_simdcall
fma(SimdDouble a, SimdDouble b, SimdDouble c)
{
    return a*b+c;
}

/*! \brief SIMD double Fused-multiply-subtract. Result is a*b-c.
 *
 * \param a factor1
 * \param b factor2
 * \param c term
 * \return a*b-c
 */
static inline SimdDouble gmx_simdcall
fms(SimdDouble a, SimdDouble b, SimdDouble c)
{
    return a*b-c;
}

/*! \brief SIMD double Fused-negated-multiply-add. Result is -a*b+c.
 *
 * \param a factor1
 * \param b factor2
 * \param c term
 * \return -a*b+c
 */
static inline SimdDouble gmx_simdcall
fnma(SimdDouble a, SimdDouble b, SimdDouble c)
{
    return c-a*b;
}

/*! \brief SIMD double Fused-negated-multiply-subtract. Result is -a*b-c.
 *
 * \param a factor1
 * \param b factor2
 * \param c term
 * \return -a*b-c
 */
static inline SimdDouble gmx_simdcall
fnms(SimdDouble a, SimdDouble b, SimdDouble c)
{
    return -a*b-c;
}

/*! \brief double SIMD 1.0/sqrt(x) lookup.
 *
 * This is a low-level instruction that should only be called from routines
 * implementing the inverse square root in simd_math.h.
 *
 * \param x Argument, x>0
 * \return Approximation of 1/sqrt(x), accuracy is \ref GMX_SIMD_RSQRT_BITS.
 */
static inline SimdDouble gmx_simdcall
rsqrt(SimdDouble x)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        // sic - we only use single precision for the lookup
        res.simdInternal_[i] = 1.0f / std::sqrt(static_cast<float>(x.simdInternal_[i]));
    }
    return res;
};

/*! \brief SIMD double 1.0/x lookup.
 *
 * This is a low-level instruction that should only be called from routines
 * implementing the reciprocal in simd_math.h.
 *
 * \param x Argument, x!=0
 * \return Approximation of 1/x, accuracy is \ref GMX_SIMD_RCP_BITS.
 */
static inline SimdDouble gmx_simdcall
rcp(SimdDouble x)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        // sic - we only use single precision for the lookup
        res.simdInternal_[i] = 1.0f / static_cast<float>(x.simdInternal_[i]);
    }
    return res;
};

/*! \brief Add two double SIMD variables, masked version.
 *
 * \param a term1
 * \param b term2
 * \param m mask
 * \return a+b where mask is true, 0.0 otherwise.
 */
static inline SimdDouble gmx_simdcall
maskAdd(SimdDouble a, SimdDouble b, SimdDBool m)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] + (m.simdInternal_[i] ? b.simdInternal_[i] : 0.0);
    }
    return res;
}

/*! \brief Multiply two double SIMD variables, masked version.
 *
 * \param a factor1
 * \param b factor2
 * \param m mask
 * \return a*b where mask is true, 0.0 otherwise.
 */
static inline SimdDouble gmx_simdcall
maskzMul(SimdDouble a, SimdDouble b, SimdDBool m)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = m.simdInternal_[i] ? (a.simdInternal_[i] * b.simdInternal_[i]) : 0.0;
    }
    return res;
}

/*! \brief SIMD double fused multiply-add, masked version.
 *
 * \param a factor1
 * \param b factor2
 * \param c term
 * \param m mask
 * \return a*b+c where mask is true, 0.0 otherwise.
 */
static inline SimdDouble gmx_simdcall
maskzFma(SimdDouble a, SimdDouble b, SimdDouble c, SimdDBool m)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = m.simdInternal_[i] ? (a.simdInternal_[i] * b.simdInternal_[i] + c.simdInternal_[i]) : 0.0;
    }
    return res;
}

/*! \brief SIMD double 1.0/sqrt(x) lookup, masked version.
 *
 * This is a low-level instruction that should only be called from routines
 * implementing the inverse square root in simd_math.h.
 *
 * \param x Argument, x>0 for entries where mask is true.
 * \param m Mask
 * \return Approximation of 1/sqrt(x), accuracy is \ref GMX_SIMD_RSQRT_BITS.
 *         The result for masked-out entries will be 0.0.
 */
static inline SimdDouble gmx_simdcall
maskzRsqrt(SimdDouble x, SimdDBool m)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        // sic - we only use single precision for the lookup
        res.simdInternal_[i] = (m.simdInternal_[i] != 0) ? 1.0f / std::sqrt(static_cast<float>(x.simdInternal_[i])) : 0.0;
    }
    return res;
}

/*! \brief SIMD double 1.0/x lookup, masked version.
 *
 * This is a low-level instruction that should only be called from routines
 * implementing the reciprocal in simd_math.h.
 *
 * \param x Argument, x>0 for entries where mask is true.
 * \param m Mask
 * \return Approximation of 1/x, accuracy is \ref GMX_SIMD_RCP_BITS.
 *         The result for masked-out entries will be 0.0.
 */
static inline SimdDouble gmx_simdcall
maskzRcp(SimdDouble x, SimdDBool m)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (m.simdInternal_[i] != 0) ? 1.0f / static_cast<float>(x.simdInternal_[i]) : 0.0;
    }
    return res;
}

/*! \brief SIMD double floating-point fabs().
 *
 * \param a any floating point values
 * \return fabs(a) for each element.
 */
static inline SimdDouble gmx_simdcall
abs(SimdDouble a)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = std::abs(a.simdInternal_[i]);
    }
    return res;
}

/*! \brief Set each SIMD double element to the largest from two variables.
 *
 * \param a Any floating-point value
 * \param b Any floating-point value
 * \return max(a,b) for each element.
 */
static inline SimdDouble gmx_simdcall
max(SimdDouble a, SimdDouble b)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = std::max(a.simdInternal_[i], b.simdInternal_[i]);
    }
    return res;
}

/*! \brief Set each SIMD double element to the smallest from two variables.
 *
 * \param a Any floating-point value
 * \param b Any floating-point value
 * \return min(a,b) for each element.
 */
static inline SimdDouble gmx_simdcall
min(SimdDouble a, SimdDouble b)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = std::min(a.simdInternal_[i], b.simdInternal_[i]);
    }
    return res;
}

/*! \brief SIMD double round to nearest integer value (in floating-point format).
 *
 * \param a Any floating-point value
 * \return The nearest integer, represented in floating-point format.
 */
static inline SimdDouble gmx_simdcall
round(SimdDouble a)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = std::round(a.simdInternal_[i]);
    }
    return res;
}

/*! \brief Truncate SIMD double, i.e. round towards zero - common hardware instruction.
 *
 * \param a Any floating-point value
 * \return Integer rounded towards zero, represented in floating-point format.
 *
 * \note This is truncation towards zero, not floor(). The reason for this
 * is that truncation is virtually always present as a dedicated hardware
 * instruction, but floor() frequently isn't.
 */
static inline SimdDouble gmx_simdcall
trunc(SimdDouble a)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = std::trunc(a.simdInternal_[i]);
    }
    return res;
}

/*! \brief Extract (integer) exponent and fraction from double precision SIMD.
 *
 * \param       value     Floating-point value to extract from
 * \param[out]  exponent  Returned exponent of value, integer SIMD format.
 * \return      Fraction of value, floating-point SIMD format.
 */
static inline SimdDouble gmx_simdcall
frexp(SimdDouble value, SimdDInt32 * exponent)
{
    SimdDouble fraction;

    for (std::size_t i = 0; i < fraction.simdInternal_.size(); i++)
    {
        fraction.simdInternal_[i] = std::frexp(value.simdInternal_[i], &exponent->simdInternal_[i]);
    }
    return fraction;
}

/*! \brief Multiply a SIMD double value by the number 2 raised to an exp power.
 *
 * \tparam opt By default, this routine will return zero for input arguments
 *             that are so small they cannot be reproduced in the current
 *             precision. If the unsafe math optimization template parameter
 *             setting is used, these tests are skipped, and the result will
 *             be undefined (possible even NaN). This might happen below -127
 *             in single precision or -1023 in double, although some
 *             might use denormal support to extend the range.
 *
 * \param value Floating-point number to multiply with new exponent
 * \param exponent Integer that will not overflow as 2^exponent.
 * \return value*2^exponent
 */
template <MathOptimization opt = MathOptimization::Safe>
static inline SimdDouble gmx_simdcall
ldexp(SimdDouble value, SimdDInt32 exponent)
{
    SimdDouble           res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        // std::ldexp already takes care of clamping arguments, so we do not
        // need to do anything in the reference implementation
        res.simdInternal_[i] = std::ldexp(value.simdInternal_[i], exponent.simdInternal_[i]);
    }
    return res;
}

/*! \brief Return sum of all elements in SIMD double variable.
 *
 * \param a SIMD variable to reduce/sum.
 * \return The sum of all elements in the argument variable.
 *
 */
static inline double gmx_simdcall
reduce(SimdDouble a)
{
    double sum = 0.0;

    for (std::size_t i = 0; i < a.simdInternal_.size(); i++)
    {
        sum += a.simdInternal_[i];
    }
    return sum;
}

/*! \}
 *
 * \name SIMD implementation double precision floating-point comparison, boolean, selection.
 * \{
 */

/*! \brief SIMD a==b for double SIMD.
 *
 * \param a value1
 * \param b value2
 * \return Each element of the boolean will be set to true if a==b.
 *
 * Beware that exact floating-point comparisons are difficult.
 */
static inline SimdDBool gmx_simdcall
operator==(SimdDouble a, SimdDouble b)
{
    SimdDBool         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (a.simdInternal_[i] == b.simdInternal_[i]);
    }
    return res;
}

/*! \brief SIMD a!=b for double SIMD.
 *
 * \param a value1
 * \param b value2
 * \return Each element of the boolean will be set to true if a!=b.
 *
 * Beware that exact floating-point comparisons are difficult.
 */
static inline SimdDBool gmx_simdcall
operator!=(SimdDouble a, SimdDouble b)
{
    SimdDBool         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (a.simdInternal_[i] != b.simdInternal_[i]);
    }
    return res;
}

/*! \brief SIMD a<b for double SIMD.
 *
 * \param a value1
 * \param b value2
 * \return Each element of the boolean will be set to true if a<b.
 */
static inline SimdDBool gmx_simdcall
operator<(SimdDouble a, SimdDouble b)
{
    SimdDBool          res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (a.simdInternal_[i] < b.simdInternal_[i]);
    }
    return res;
}

/*! \brief SIMD a<=b for double SIMD.
 *
 * \param a value1
 * \param b value2
 * \return Each element of the boolean will be set to true if a<=b.
 */
static inline SimdDBool gmx_simdcall
operator<=(SimdDouble a, SimdDouble b)
{
    SimdDBool          res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (a.simdInternal_[i] <= b.simdInternal_[i]);
    }
    return res;
}

/*! \brief Return true if any bits are set in the single precision SIMD.
 *
 * This function is used to handle bitmasks, mainly for exclusions in the
 * inner kernels. Note that it will return true even for -0.0 (sign bit set),
 * so it is not identical to not-equal.
 *
 * \param a value
 * \return Each element of the boolean will be true if any bit in a is nonzero.
 */
static inline SimdDBool gmx_simdcall
testBits(SimdDouble a)
{
    SimdDBool         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        union
        {
            std::uint64_t i;
            double        d;
        } conv;

        conv.d               = a.simdInternal_[i];
        res.simdInternal_[i] = (conv.i != 0);
    }
    return res;
}

/*! \brief Logical \a and on double precision SIMD booleans.
 *
 * \param a logical vars 1
 * \param b logical vars 2
 * \return For each element, the result boolean is true if a \& b are true.
 *
 * \note This is not necessarily a bitwise operation - the storage format
 * of booleans is implementation-dependent.
 */
static inline SimdDBool gmx_simdcall
operator&&(SimdDBool a, SimdDBool b)
{
    SimdDBool         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (a.simdInternal_[i] && b.simdInternal_[i]);
    }
    return res;
}

/*! \brief Logical \a or on double precision SIMD booleans.
 *
 * \param a logical vars 1
 * \param b logical vars 2
 * \return For each element, the result boolean is true if a or b is true.
 *
 * Note that this is not necessarily a bitwise operation - the storage format
 * of booleans is implementation-dependent.
 *
 \ */
static inline SimdDBool gmx_simdcall
operator||(SimdDBool a, SimdDBool b)
{
    SimdDBool         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (a.simdInternal_[i] || b.simdInternal_[i]);
    }
    return res;
}

/*! \brief Returns non-zero if any of the boolean in SIMD a is True, otherwise 0.
 *
 * \param a Logical variable.
 * \return true if any element in a is true, otherwise false.
 *
 * The actual return value for truth will depend on the architecture,
 * so any non-zero value is considered truth.
 */
static inline bool gmx_simdcall
anyTrue(SimdDBool a)
{
    bool res = false;

    for (std::size_t i = 0; i < a.simdInternal_.size(); i++)
    {
        res = res || a.simdInternal_[i];
    }
    return res;
}

/*! \brief Select from double precision SIMD variable where boolean is true.
 *
 * \param a Floating-point variable to select from
 * \param mask Boolean selector
 * \return  For each element, a is selected for true, 0 for false.
 */
static inline SimdDouble gmx_simdcall
selectByMask(SimdDouble a, SimdDBool mask)
{
    SimdDouble          res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = mask.simdInternal_[i] ? a.simdInternal_[i] : 0.0;
    }
    return res;
}

/*! \brief Select from double precision SIMD variable where boolean is false.
 *
 * \param a Floating-point variable to select from
 * \param mask Boolean selector
 * \return  For each element, a is selected for false, 0 for true (sic).
 */
static inline SimdDouble gmx_simdcall
selectByNotMask(SimdDouble a, SimdDBool mask)
{
    SimdDouble          res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = mask.simdInternal_[i] ? 0.0 : a.simdInternal_[i];
    }
    return res;
}

/*! \brief Vector-blend SIMD double selection.
 *
 * \param a First source
 * \param b Second source
 * \param sel Boolean selector
 * \return For each element, select b if sel is true, a otherwise.
 */
static inline SimdDouble gmx_simdcall
blend(SimdDouble a, SimdDouble b, SimdDBool sel)
{
    SimdDouble         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = sel.simdInternal_[i] ? b.simdInternal_[i] : a.simdInternal_[i];
    }
    return res;
}

/*! \}
 *
 * \name SIMD implementation integer (corresponding to double) bitwise logical operations
 * \{
 */

/*! \brief SIMD integer shift left logical, based on immediate value.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_LOGICAL is 1.
 *
 *  Logical shift. Each element is shifted (independently) up to 32 positions
 *  left, while zeros are shifted in from the right.
 *
 * \param a integer data to shift
 * \param n number of positions to shift left. n<=32.
 * \return shifted values
 */
static inline SimdDInt32 gmx_simdcall
operator<<(SimdDInt32 a, int n)
{
    SimdDInt32         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] << n;
    }
    return res;
}

/*! \brief SIMD integer shift right logical, based on immediate value.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_LOGICAL is 1.
 *
 *  Logical shift. Each element is shifted (independently) up to 32 positions
 *  right, while zeros are shifted in from the left.
 *
 * \param a integer data to shift
 * \param n number of positions to shift right. n<=32.
 * \return shifted values
 */
static inline SimdDInt32 gmx_simdcall
operator>>(SimdDInt32 a, int n)
{
    SimdDInt32         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] >> n;
    }
    return res;
}

/*! \brief Integer SIMD bitwise and.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_LOGICAL is 1.
 *
 * \note You can \a not use this operation directly to select based on a boolean
 * SIMD variable, since booleans are separate from integer SIMD. If that
 * is what you need, have a look at \ref gmx::selectByMask instead.
 *
 * \param a first integer SIMD
 * \param b second integer SIMD
 * \return a \& b (bitwise and)
 */
static inline SimdDInt32 gmx_simdcall
operator&(SimdDInt32 a, SimdDInt32 b)
{
    SimdDInt32         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] & b.simdInternal_[i];
    }
    return res;
}

/*! \brief Integer SIMD bitwise not/complement.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_LOGICAL is 1.
 *
 * \note You can \a not use this operation directly to select based on a boolean
 * SIMD variable, since booleans are separate from integer SIMD. If that
 * is what you need, have a look at \ref gmx::selectByMask instead.
 *
 * \param a integer SIMD
 * \param b integer SIMD
 * \return (~a) & b
 */
static inline SimdDInt32 gmx_simdcall
andNot(SimdDInt32 a, SimdDInt32 b)
{
    SimdDInt32         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = ~a.simdInternal_[i] & b.simdInternal_[i];
    }
    return res;
}

/*! \brief Integer SIMD bitwise or.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_LOGICAL is 1.
 *
 * \param a first integer SIMD
 * \param b second integer SIMD
 * \return a \| b (bitwise or)
 */
static inline SimdDInt32 gmx_simdcall
operator|(SimdDInt32 a, SimdDInt32 b)
{
    SimdDInt32         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] | b.simdInternal_[i];
    }
    return res;
}

/*! \brief Integer SIMD bitwise xor.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_LOGICAL is 1.
 *
 * \param a first integer SIMD
 * \param b second integer SIMD
 * \return a ^ b (bitwise xor)
 */
static inline SimdDInt32 gmx_simdcall
operator^(SimdDInt32 a, SimdDInt32 b)
{
    SimdDInt32         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] ^ b.simdInternal_[i];
    }
    return res;
}

/*! \}
 *
 * \name SIMD implementation integer (corresponding to double) arithmetics
 * \{
 */

/*! \brief Add SIMD integers.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \param a term1
 * \param b term2
 * \return a+b
 */
static inline SimdDInt32 gmx_simdcall
operator+(SimdDInt32 a, SimdDInt32 b)
{
    SimdDInt32         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] + b.simdInternal_[i];
    }
    return res;
}

/*! \brief Subtract SIMD integers.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \param a term1
 * \param b term2
 * \return a-b
 */
static inline SimdDInt32 gmx_simdcall
operator-(SimdDInt32 a, SimdDInt32 b)
{
    SimdDInt32         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] - b.simdInternal_[i];
    }
    return res;
}

/*! \brief Multiply SIMD integers.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \param a factor1
 * \param b factor2
 * \return a*b.
 *
 * \note Only the low 32 bits are retained, so this can overflow.
 */
static inline SimdDInt32 gmx_simdcall
operator*(SimdDInt32 a, SimdDInt32 b)
{
    SimdDInt32         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = a.simdInternal_[i] * b.simdInternal_[i];
    }
    return res;
}

/*! \}
 *
 * \name SIMD implementation integer (corresponding to double) comparisons, boolean selection
 * \{
 */

/*! \brief Equality comparison of two integers corresponding to double values.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \param a SIMD integer1
 * \param b SIMD integer2
 * \return SIMD integer boolean with true for elements where a==b
 */
static inline SimdDIBool gmx_simdcall
operator==(SimdDInt32 a, SimdDInt32 b)
{
    SimdDIBool         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (a.simdInternal_[i] == b.simdInternal_[i]);
    }
    return res;
}

/*! \brief Less-than comparison of two SIMD integers corresponding to double values.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \param a SIMD integer1
 * \param b SIMD integer2
 * \return SIMD integer boolean with true for elements where a<b
 */
static inline SimdDIBool gmx_simdcall
operator<(SimdDInt32 a, SimdDInt32 b)
{
    SimdDIBool         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (a.simdInternal_[i] < b.simdInternal_[i]);
    }
    return res;
}

/*! \brief Check if any bit is set in each element
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \param a SIMD integer
 * \return SIMD integer boolean with true for elements where any bit is set
 */
static inline SimdDIBool gmx_simdcall
testBits(SimdDInt32 a)
{
    SimdDIBool         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (a.simdInternal_[i] != 0);
    }
    return res;
}

/*! \brief Logical AND on SimdDIBool.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \param a SIMD boolean 1
 * \param b SIMD boolean 2
 * \return True for elements where both a and b are true.
 */
static inline SimdDIBool gmx_simdcall
operator&&(SimdDIBool a, SimdDIBool b)
{
    SimdDIBool        res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (a.simdInternal_[i] && b.simdInternal_[i]);
    }
    return res;
}

/*! \brief Logical OR on SimdDIBool.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \param a SIMD boolean 1
 * \param b SIMD boolean 2
 * \return True for elements where both a and b are true.
 */
static inline SimdDIBool gmx_simdcall
operator||(SimdDIBool a, SimdDIBool b)
{
    SimdDIBool         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = (a.simdInternal_[i] || b.simdInternal_[i]);
    }
    return res;
}

/*! \brief Returns true if any of the boolean in x is True, otherwise 0.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * The actual return value for "any true" will depend on the architecture.
 * Any non-zero value should be considered truth.
 *
 * \param a SIMD boolean
 * \return True if any of the elements in a is true, otherwise 0.
 */
static inline bool gmx_simdcall
anyTrue(SimdDIBool a)
{
    bool res = false;

    for (std::size_t i = 0; i < a.simdInternal_.size(); i++)
    {
        res = res || a.simdInternal_[i];
    }
    return res;
}

/*! \brief Select from \ref gmx::SimdDInt32 variable where boolean is true.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \param a SIMD integer to select from
 * \param mask Boolean selector
 * \return Elements from a where sel is true, 0 otherwise.
 */
static inline SimdDInt32 gmx_simdcall
selectByMask(SimdDInt32 a, SimdDIBool mask)
{
    SimdDInt32         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = mask.simdInternal_[i] ? a.simdInternal_[i] : 0;
    }
    return res;
}

/*! \brief Select from \ref gmx::SimdDInt32 variable where boolean is false.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \param a SIMD integer to select from
 * \param mask Boolean selector
 * \return Elements from a where sel is false, 0 otherwise (sic).
 */
static inline SimdDInt32 gmx_simdcall
selectByNotMask(SimdDInt32 a, SimdDIBool mask)
{
    SimdDInt32         res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = mask.simdInternal_[i] ? 0 : a.simdInternal_[i];
    }
    return res;
}

/*! \brief Vector-blend SIMD integer selection.
 *
 * Available if \ref GMX_SIMD_HAVE_DINT32_ARITHMETICS is 1.
 *
 * \param a First source
 * \param b Second source
 * \param sel Boolean selector
 * \return For each element, select b if sel is true, a otherwise.
 */
static inline SimdDInt32 gmx_simdcall
blend(SimdDInt32 a, SimdDInt32 b, SimdDIBool sel)
{
    SimdDInt32        res;

    for (std::size_t i = 0; i < res.simdInternal_.size(); i++)
    {
        res.simdInternal_[i] = sel.simdInternal_[i] ? b.simdInternal_[i] : a.simdInternal_[i];
    }
    return res;
}

/*! \}
 *
 * \name SIMD implementation conversion operations
 * \{
 */

/*! \brief Round double precision floating point to integer.
 *
 * \param a SIMD floating-point
 * \return SIMD integer, rounded to nearest integer.
 */
static inline SimdDInt32 gmx_simdcall
cvtR2I(SimdDouble a)
{
    SimdDInt32         b;

    for (std::size_t i = 0; i < b.simdInternal_.size(); i++)
    {
        b.simdInternal_[i] = std::round(a.simdInternal_[i]);
    }
    return b;
};

/*! \brief Truncate double precision floating point to integer.
 *
 * \param a SIMD floating-point
 * \return SIMD integer, truncated to nearest integer.
 */
static inline SimdDInt32 gmx_simdcall
cvttR2I(SimdDouble a)
{
    SimdDInt32         b;

    for (std::size_t i = 0; i < b.simdInternal_.size(); i++)
    {
        b.simdInternal_[i] = std::trunc(a.simdInternal_[i]);
    }
    return b;
};

/*! \brief Convert integer to double precision floating point.
 *
 * \param a SIMD integer
 * \return SIMD floating-point
 */
static inline SimdDouble gmx_simdcall
cvtI2R(SimdDInt32 a)
{
    SimdDouble         b;

    for (std::size_t i = 0; i < b.simdInternal_.size(); i++)
    {
        b.simdInternal_[i] = a.simdInternal_[i];
    }
    return b;
};

/*! \brief Convert from double precision boolean to corresponding integer boolean
 *
 * \param a SIMD floating-point boolean
 * \return SIMD integer boolean
 */
static inline SimdDIBool gmx_simdcall
cvtB2IB(SimdDBool a)
{
    SimdDIBool         b;

    for (std::size_t i = 0; i < b.simdInternal_.size(); i++)
    {
        b.simdInternal_[i] = a.simdInternal_[i];
    }
    return b;
};

/*! \brief Convert from integer boolean to corresponding double precision boolean
 *
 * \param a SIMD integer boolean
 * \return SIMD floating-point boolean
 */
static inline SimdDBool gmx_simdcall
cvtIB2B(SimdDIBool a)
{
    SimdDBool         b;

    for (std::size_t i = 0; i < b.simdInternal_.size(); i++)
    {
        b.simdInternal_[i] = a.simdInternal_[i];
    }
    return b;
};

/*! \brief Convert SIMD float to double.
 *
 * This version is available if \ref GMX_SIMD_FLOAT_WIDTH is identical to
 * \ref GMX_SIMD_DOUBLE_WIDTH.
 *
 * Float/double conversions are complex since the SIMD width could either
 * be different (e.g. on x86) or identical (e.g. IBM QPX). This means you will
 * need to check for the width in the code, and have different code paths.
 *
 * \param f Single-precision SIMD variable
 * \return Double-precision SIMD variable of the same width
 */
static inline SimdDouble gmx_simdcall
cvtF2D(SimdFloat gmx_unused f)
{
#if (GMX_SIMD_FLOAT_WIDTH == GMX_SIMD_DOUBLE_WIDTH)
    SimdDouble        d;
    for (std::size_t i = 0; i < d.simdInternal_.size(); i++)
    {
        d.simdInternal_[i] = f.simdInternal_[i];
    }
    return d;
#else
    gmx_fatal(FARGS, "cvtF2D() requires GMX_SIMD_FLOAT_WIDTH==GMX_SIMD_DOUBLE_WIDTH");
#endif
}

/*! \brief Convert SIMD double to float.
 *
 * This version is available if \ref GMX_SIMD_FLOAT_WIDTH is identical to
 * \ref GMX_SIMD_DOUBLE_WIDTH.
 *
 * Float/double conversions are complex since the SIMD width could either
 * be different (e.g. on x86) or identical (e.g. IBM QPX). This means you will
 * need to check for the width in the code, and have different code paths.
 *
 * \param d Double-precision SIMD variable
 * \return Single-precision SIMD variable of the same width
 */
static inline SimdFloat gmx_simdcall
cvtD2F(SimdDouble gmx_unused d)
{
#if (GMX_SIMD_FLOAT_WIDTH == GMX_SIMD_DOUBLE_WIDTH)
    SimdFloat        f;
    for (std::size_t i = 0; i < f.simdInternal_.size(); i++)
    {
        f.simdInternal_[i] = d.simdInternal_[i];
    }
    return f;
#else
    gmx_fatal(FARGS, "cvtD2F() requires GMX_SIMD_FLOAT_WIDTH==GMX_SIMD_DOUBLE_WIDTH");
#endif
}

/*! \brief Convert SIMD float to double.
 *
 * This version is available if \ref GMX_SIMD_FLOAT_WIDTH is twice as large
 * as \ref GMX_SIMD_DOUBLE_WIDTH.
 *
 * Float/double conversions are complex since the SIMD width could either
 * be different (e.g. on x86) or identical (e.g. IBM QPX). This means you will
 * need to check for the width in the code, and have different code paths.
 *
 * \param f Single-precision SIMD variable
 * \param[out] d0 Double-precision SIMD variable, first half of values from f.
 * \param[out] d1 Double-precision SIMD variable, second half of values from f.
 */
static inline void gmx_simdcall
cvtF2DD(SimdFloat gmx_unused f, SimdDouble gmx_unused * d0, SimdDouble gmx_unused * d1)
{
#if (GMX_SIMD_FLOAT_WIDTH == 2*GMX_SIMD_DOUBLE_WIDTH)
    for (std::size_t i = 0; i < d0->simdInternal_.size(); i++)
    {
        d0->simdInternal_[i] = f.simdInternal_[i];
        d1->simdInternal_[i] = f.simdInternal_[f.simdInternal_.size()/2 + i];
    }
#else
    gmx_fatal(FARGS, "simdCvtF2DD() requires GMX_SIMD_FLOAT_WIDTH==2*GMX_SIMD_DOUBLE_WIDTH");
#endif
}

/*! \brief Convert SIMD double to float.
 *
 * This version is available if \ref GMX_SIMD_FLOAT_WIDTH is twice as large
 * as \ref GMX_SIMD_DOUBLE_WIDTH.
 *
 * Float/double conversions are complex since the SIMD width could either
 * be different (e.g. on x86) or identical (e.g. IBM QPX). This means you will
 * need to check for the width in the code, and have different code paths.
 *
 * \param d0 Double-precision SIMD variable, first half of values to put in f.
 * \param d1 Double-precision SIMD variable, second half of values to put in f.
 * \return Single-precision SIMD variable with all values.
 */
static inline SimdFloat gmx_simdcall
cvtDD2F(SimdDouble gmx_unused d0, SimdDouble gmx_unused d1)
{
#if (GMX_SIMD_FLOAT_WIDTH == 2*GMX_SIMD_DOUBLE_WIDTH)
    SimdFloat        f;
    for (std::size_t i = 0; i < d0.simdInternal_.size(); i++)
    {
        f.simdInternal_[i]                            = d0.simdInternal_[i];
        f.simdInternal_[f.simdInternal_.size()/2 + i] = d1.simdInternal_[i];
    }
    return f;
#else
    gmx_fatal(FARGS, "simdCvtDD2F() requires GMX_SIMD_FLOAT_WIDTH==2*GMX_SIMD_DOUBLE_WIDTH");
#endif
}

/*! \} */

/*! \} */
/*! \endcond */

}      // namespace gmx

#endif // GMX_SIMD_IMPL_REFERENCE_SIMD_DOUBLE_H