Make sure frexp() returns correct for argument 0.0

[gromacs.git] / src / gromacs / simd / impl_x86_avx_512 / impl_x86_avx_512_simd_double.h

/*
 * This file is part of the GROMACS molecular simulation package.
 *
 * Copyright (c) 2014,2015,2016,2017,2018 by the GROMACS development team.
 * Copyright (c) 2019,2020, 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_X86_AVX_512_SIMD_DOUBLE_H
#define GMX_SIMD_IMPL_X86_AVX_512_SIMD_DOUBLE_H

#include "config.h"

#include <cassert>
#include <cstdint>

#include <immintrin.h>

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

#include "impl_x86_avx_512_general.h"
#include "impl_x86_avx_512_simd_float.h"

namespace gmx
{

class SimdDouble
{
public:
    SimdDouble() {}

    SimdDouble(double d) : simdInternal_(_mm512_set1_pd(d)) {}

    // Internal utility constructor to simplify return statements
    SimdDouble(__m512d simd) : simdInternal_(simd) {}

    __m512d simdInternal_;
};

class SimdDInt32
{
public:
    SimdDInt32() {}

    SimdDInt32(std::int32_t i) : simdInternal_(_mm256_set1_epi32(i)) {}

    // Internal utility constructor to simplify return statements
    SimdDInt32(__m256i simd) : simdInternal_(simd) {}

    __m256i simdInternal_;
};

class SimdDBool
{
public:
    SimdDBool() {}

    // Internal utility constructor to simplify return statements
    SimdDBool(__mmask8 simd) : simdInternal_(simd) {}

    __mmask8 simdInternal_;
};

class SimdDIBool
{
public:
    SimdDIBool() {}

    // Internal utility constructor to simplify return statements
    SimdDIBool(__mmask16 simd) : simdInternal_(simd) {}

    __mmask16 simdInternal_;
};

static inline SimdDouble gmx_simdcall simdLoad(const double* m, SimdDoubleTag = {})
{
    assert(std::size_t(m) % 64 == 0);
    return { _mm512_load_pd(m) };
}

static inline void gmx_simdcall store(double* m, SimdDouble a)
{
    assert(std::size_t(m) % 64 == 0);
    _mm512_store_pd(m, a.simdInternal_);
}

static inline SimdDouble gmx_simdcall simdLoadU(const double* m, SimdDoubleTag = {})
{
    return { _mm512_loadu_pd(m) };
}

static inline void gmx_simdcall storeU(double* m, SimdDouble a)
{
    _mm512_storeu_pd(m, a.simdInternal_);
}

static inline SimdDouble gmx_simdcall setZeroD()
{
    return { _mm512_setzero_pd() };
}

static inline SimdDInt32 gmx_simdcall simdLoad(const std::int32_t* m, SimdDInt32Tag)
{
    assert(std::size_t(m) % 32 == 0);
    return { _mm256_load_si256(reinterpret_cast<const __m256i*>(m)) };
}

static inline void gmx_simdcall store(std::int32_t* m, SimdDInt32 a)
{
    assert(std::size_t(m) % 32 == 0);
    _mm256_store_si256(reinterpret_cast<__m256i*>(m), a.simdInternal_);
}

static inline SimdDInt32 gmx_simdcall simdLoadU(const std::int32_t* m, SimdDInt32Tag)
{
    return { _mm256_loadu_si256(reinterpret_cast<const __m256i*>(m)) };
}

static inline void gmx_simdcall storeU(std::int32_t* m, SimdDInt32 a)
{
    _mm256_storeu_si256(reinterpret_cast<__m256i*>(m), a.simdInternal_);
}

static inline SimdDInt32 gmx_simdcall setZeroDI()
{
    return { _mm256_setzero_si256() };
}

static inline SimdDouble gmx_simdcall operator&(SimdDouble a, SimdDouble b)
{
    return { _mm512_castsi512_pd(_mm512_and_epi32(_mm512_castpd_si512(a.simdInternal_),
                                                  _mm512_castpd_si512(b.simdInternal_))) };
}

static inline SimdDouble gmx_simdcall andNot(SimdDouble a, SimdDouble b)
{
    return { _mm512_castsi512_pd(_mm512_andnot_epi32(_mm512_castpd_si512(a.simdInternal_),
                                                     _mm512_castpd_si512(b.simdInternal_))) };
}

static inline SimdDouble gmx_simdcall operator|(SimdDouble a, SimdDouble b)
{
    return { _mm512_castsi512_pd(_mm512_or_epi32(_mm512_castpd_si512(a.simdInternal_),
                                                 _mm512_castpd_si512(b.simdInternal_))) };
}

static inline SimdDouble gmx_simdcall operator^(SimdDouble a, SimdDouble b)
{
    return { _mm512_castsi512_pd(_mm512_xor_epi32(_mm512_castpd_si512(a.simdInternal_),
                                                  _mm512_castpd_si512(b.simdInternal_))) };
}

static inline SimdDouble gmx_simdcall operator+(SimdDouble a, SimdDouble b)
{
    return { _mm512_add_pd(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDouble gmx_simdcall operator-(SimdDouble a, SimdDouble b)
{
    return { _mm512_sub_pd(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDouble gmx_simdcall operator-(SimdDouble x)
{
    return { _mm512_castsi512_pd(_mm512_xor_epi32(_mm512_castpd_si512(x.simdInternal_),
                                                  _mm512_castpd_si512(_mm512_set1_pd(GMX_DOUBLE_NEGZERO)))) };
}

static inline SimdDouble gmx_simdcall operator*(SimdDouble a, SimdDouble b)
{
    return { _mm512_mul_pd(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDouble gmx_simdcall fma(SimdDouble a, SimdDouble b, SimdDouble c)
{
    return { _mm512_fmadd_pd(a.simdInternal_, b.simdInternal_, c.simdInternal_) };
}

static inline SimdDouble gmx_simdcall fms(SimdDouble a, SimdDouble b, SimdDouble c)
{
    return { _mm512_fmsub_pd(a.simdInternal_, b.simdInternal_, c.simdInternal_) };
}

static inline SimdDouble gmx_simdcall fnma(SimdDouble a, SimdDouble b, SimdDouble c)
{
    return { _mm512_fnmadd_pd(a.simdInternal_, b.simdInternal_, c.simdInternal_) };
}

static inline SimdDouble gmx_simdcall fnms(SimdDouble a, SimdDouble b, SimdDouble c)
{
    return { _mm512_fnmsub_pd(a.simdInternal_, b.simdInternal_, c.simdInternal_) };
}

// Override for AVX-512-KNL
#if GMX_SIMD_X86_AVX_512
static inline SimdDouble gmx_simdcall rsqrt(SimdDouble x)
{
    return { _mm512_rsqrt14_pd(x.simdInternal_) };
}

static inline SimdDouble gmx_simdcall rcp(SimdDouble x)
{
    return { _mm512_rcp14_pd(x.simdInternal_) };
}
#endif

static inline SimdDouble gmx_simdcall maskAdd(SimdDouble a, SimdDouble b, SimdDBool m)
{
    return { _mm512_mask_add_pd(a.simdInternal_, m.simdInternal_, a.simdInternal_, b.simdInternal_) };
}

static inline SimdDouble gmx_simdcall maskzMul(SimdDouble a, SimdDouble b, SimdDBool m)
{
    return { _mm512_maskz_mul_pd(m.simdInternal_, a.simdInternal_, b.simdInternal_) };
}

static inline SimdDouble gmx_simdcall maskzFma(SimdDouble a, SimdDouble b, SimdDouble c, SimdDBool m)
{
    return { _mm512_maskz_fmadd_pd(m.simdInternal_, a.simdInternal_, b.simdInternal_, c.simdInternal_) };
}

// Override for AVX-512-KNL
#if GMX_SIMD_X86_AVX_512
static inline SimdDouble gmx_simdcall maskzRsqrt(SimdDouble x, SimdDBool m)
{
    return { _mm512_maskz_rsqrt14_pd(m.simdInternal_, x.simdInternal_) };
}

static inline SimdDouble gmx_simdcall maskzRcp(SimdDouble x, SimdDBool m)
{
    return { _mm512_maskz_rcp14_pd(m.simdInternal_, x.simdInternal_) };
}
#endif

static inline SimdDouble gmx_simdcall abs(SimdDouble x)
{
    return { _mm512_castsi512_pd(_mm512_andnot_epi32(_mm512_castpd_si512(_mm512_set1_pd(GMX_DOUBLE_NEGZERO)),
                                                     _mm512_castpd_si512(x.simdInternal_))) };
}

static inline SimdDouble gmx_simdcall max(SimdDouble a, SimdDouble b)
{
    return { _mm512_max_pd(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDouble gmx_simdcall min(SimdDouble a, SimdDouble b)
{
    return { _mm512_min_pd(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDouble gmx_simdcall round(SimdDouble x)
{
    return { _mm512_roundscale_pd(x.simdInternal_, 0) };
}

static inline SimdDouble gmx_simdcall trunc(SimdDouble x)
{
#if defined(__INTEL_COMPILER) || defined(__ECC)
    return { _mm512_trunc_pd(x.simdInternal_) };
#else
    return { _mm512_cvtepi32_pd(_mm512_cvttpd_epi32(x.simdInternal_)) };
#endif
}

template<MathOptimization opt = MathOptimization::Safe>
static inline SimdDouble frexp(SimdDouble value, SimdDInt32* exponent)
{
    __m512d rExponent;
    __m256i iExponent;
    __m512d result;

    if (opt == MathOptimization::Safe)
    {
        // For the safe branch, we use the masked operations to only assign results if the
        // input value was nonzero, and otherwise set exponent to 0, and the fraction to the input (+-0).
        __mmask8 valueIsNonZero =
                _mm512_cmp_pd_mask(_mm512_setzero_pd(), value.simdInternal_, _CMP_NEQ_OQ);
        rExponent = _mm512_maskz_getexp_pd(valueIsNonZero, value.simdInternal_);

        // AVX512F does not contain any function to use masking when adding 1 to a 256-bit register
        // (that comes with AVX512VL), so to work around this we create an integer -1 value, and
        // use masking in the _conversion_ instruction where it is supported. When we later add
        // 1 to all fields, the files that were formerly -1 (corresponding to zero exponent)
        // will be assigned -1 + 1 = 0.
        iExponent = _mm512_mask_cvtpd_epi32(_mm256_set1_epi32(-1), valueIsNonZero, rExponent);
        iExponent = _mm256_add_epi32(iExponent, _mm256_set1_epi32(1));

        // Set result to value (+-0) when it is zero.
        result = _mm512_mask_getmant_pd(value.simdInternal_, valueIsNonZero, value.simdInternal_,
                                        _MM_MANT_NORM_p5_1, _MM_MANT_SIGN_src);
    }
    else
    {
        // For the fast branch, it's the user's responsibility to make sure never to call the
        // function with input values of +-0.0
        rExponent = _mm512_getexp_pd(value.simdInternal_);
        iExponent = _mm512_cvtpd_epi32(rExponent);
        iExponent = _mm256_add_epi32(iExponent, _mm256_set1_epi32(1));

        result = _mm512_getmant_pd(value.simdInternal_, _MM_MANT_NORM_p5_1, _MM_MANT_SIGN_src);
    }

    exponent->simdInternal_ = iExponent;
    return { result };
}

template<MathOptimization opt = MathOptimization::Safe>
static inline SimdDouble ldexp(SimdDouble value, SimdDInt32 exponent)
{
    const __m256i exponentBias = _mm256_set1_epi32(1023);
    __m256i       iExponent    = _mm256_add_epi32(exponent.simdInternal_, exponentBias);
    __m512i       iExponent512;

    if (opt == MathOptimization::Safe)
    {
        // Make sure biased argument is not negative
        iExponent = _mm256_max_epi32(iExponent, _mm256_setzero_si256());
    }

    iExponent512 =
            _mm512_permutexvar_epi32(_mm512_set_epi32(7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2, 2, 1, 1, 0, 0),
                                     _mm512_castsi256_si512(iExponent));
    iExponent512 =
            _mm512_mask_slli_epi32(_mm512_setzero_epi32(), avx512Int2Mask(0xAAAA), iExponent512, 20);
    return _mm512_mul_pd(_mm512_castsi512_pd(iExponent512), value.simdInternal_);
}

static inline double gmx_simdcall reduce(SimdDouble a)
{
    __m512d x = a.simdInternal_;
    x         = _mm512_add_pd(x, _mm512_shuffle_f64x2(x, x, 0xEE));
    x         = _mm512_add_pd(x, _mm512_shuffle_f64x2(x, x, 0x11));
    x         = _mm512_add_pd(x, _mm512_permute_pd(x, 0x01));
    return *reinterpret_cast<double*>(&x);
}

static inline SimdDBool gmx_simdcall operator==(SimdDouble a, SimdDouble b)
{
    return { _mm512_cmp_pd_mask(a.simdInternal_, b.simdInternal_, _CMP_EQ_OQ) };
}

static inline SimdDBool gmx_simdcall operator!=(SimdDouble a, SimdDouble b)
{
    return { _mm512_cmp_pd_mask(a.simdInternal_, b.simdInternal_, _CMP_NEQ_OQ) };
}

static inline SimdDBool gmx_simdcall operator<(SimdDouble a, SimdDouble b)
{
    return { _mm512_cmp_pd_mask(a.simdInternal_, b.simdInternal_, _CMP_LT_OQ) };
}

static inline SimdDBool gmx_simdcall operator<=(SimdDouble a, SimdDouble b)
{
    return { _mm512_cmp_pd_mask(a.simdInternal_, b.simdInternal_, _CMP_LE_OQ) };
}

static inline SimdDBool gmx_simdcall testBits(SimdDouble a)
{
    return { _mm512_test_epi64_mask(_mm512_castpd_si512(a.simdInternal_),
                                    _mm512_castpd_si512(a.simdInternal_)) };
}

static inline SimdDBool gmx_simdcall operator&&(SimdDBool a, SimdDBool b)
{
    return { static_cast<__mmask8>(_mm512_kand(a.simdInternal_, b.simdInternal_)) };
}

static inline SimdDBool gmx_simdcall operator||(SimdDBool a, SimdDBool b)
{
    return { static_cast<__mmask8>(_mm512_kor(a.simdInternal_, b.simdInternal_)) };
}

static inline bool gmx_simdcall anyTrue(SimdDBool a)
{
    return (avx512Mask2Int(a.simdInternal_) != 0);
}

static inline SimdDouble gmx_simdcall selectByMask(SimdDouble a, SimdDBool m)
{
    return { _mm512_mask_mov_pd(_mm512_setzero_pd(), m.simdInternal_, a.simdInternal_) };
}

static inline SimdDouble gmx_simdcall selectByNotMask(SimdDouble a, SimdDBool m)
{
    return { _mm512_mask_mov_pd(a.simdInternal_, m.simdInternal_, _mm512_setzero_pd()) };
}

static inline SimdDouble gmx_simdcall blend(SimdDouble a, SimdDouble b, SimdDBool sel)
{
    return { _mm512_mask_blend_pd(sel.simdInternal_, a.simdInternal_, b.simdInternal_) };
}

static inline SimdDouble gmx_simdcall copysign(SimdDouble a, SimdDouble b)
{
    return { _mm512_castsi512_pd(_mm512_ternarylogic_epi64(_mm512_castpd_si512(a.simdInternal_),
                                                           _mm512_castpd_si512(b.simdInternal_),
                                                           _mm512_set1_epi64(INT64_MIN), 0xD8)) };
}

static inline SimdDInt32 gmx_simdcall operator&(SimdDInt32 a, SimdDInt32 b)
{
    return { _mm256_and_si256(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDInt32 gmx_simdcall andNot(SimdDInt32 a, SimdDInt32 b)
{
    return { _mm256_andnot_si256(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDInt32 gmx_simdcall operator|(SimdDInt32 a, SimdDInt32 b)
{
    return { _mm256_or_si256(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDInt32 gmx_simdcall operator^(SimdDInt32 a, SimdDInt32 b)
{
    return { _mm256_xor_si256(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDInt32 gmx_simdcall operator+(SimdDInt32 a, SimdDInt32 b)
{
    return { _mm256_add_epi32(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDInt32 gmx_simdcall operator-(SimdDInt32 a, SimdDInt32 b)
{
    return { _mm256_sub_epi32(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDInt32 gmx_simdcall operator*(SimdDInt32 a, SimdDInt32 b)
{
    return { _mm256_mullo_epi32(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDIBool gmx_simdcall operator==(SimdDInt32 a, SimdDInt32 b)
{
    return { _mm512_mask_cmp_epi32_mask(avx512Int2Mask(0xFF), _mm512_castsi256_si512(a.simdInternal_),
                                        _mm512_castsi256_si512(b.simdInternal_), _MM_CMPINT_EQ) };
}

static inline SimdDIBool gmx_simdcall testBits(SimdDInt32 a)
{
    return { _mm512_mask_test_epi32_mask(avx512Int2Mask(0xFF), _mm512_castsi256_si512(a.simdInternal_),
                                         _mm512_castsi256_si512(a.simdInternal_)) };
}

static inline SimdDIBool gmx_simdcall operator<(SimdDInt32 a, SimdDInt32 b)
{
    return { _mm512_mask_cmp_epi32_mask(avx512Int2Mask(0xFF), _mm512_castsi256_si512(a.simdInternal_),
                                        _mm512_castsi256_si512(b.simdInternal_), _MM_CMPINT_LT) };
}

static inline SimdDIBool gmx_simdcall operator&&(SimdDIBool a, SimdDIBool b)
{
    return { _mm512_kand(a.simdInternal_, b.simdInternal_) };
}

static inline SimdDIBool gmx_simdcall operator||(SimdDIBool a, SimdDIBool b)
{
    return { _mm512_kor(a.simdInternal_, b.simdInternal_) };
}

static inline bool gmx_simdcall anyTrue(SimdDIBool a)
{
    return (avx512Mask2Int(a.simdInternal_) & 0xFF) != 0;
}

static inline SimdDInt32 gmx_simdcall selectByMask(SimdDInt32 a, SimdDIBool m)
{
    return { _mm512_castsi512_si256(_mm512_mask_mov_epi32(
            _mm512_setzero_si512(), m.simdInternal_, _mm512_castsi256_si512(a.simdInternal_))) };
}

static inline SimdDInt32 gmx_simdcall selectByNotMask(SimdDInt32 a, SimdDIBool m)
{
    return { _mm512_castsi512_si256(_mm512_mask_mov_epi32(
            _mm512_castsi256_si512(a.simdInternal_), m.simdInternal_, _mm512_setzero_si512())) };
}

static inline SimdDInt32 gmx_simdcall blend(SimdDInt32 a, SimdDInt32 b, SimdDIBool sel)
{
    return { _mm512_castsi512_si256(
            _mm512_mask_blend_epi32(sel.simdInternal_, _mm512_castsi256_si512(a.simdInternal_),
                                    _mm512_castsi256_si512(b.simdInternal_))) };
}

static inline SimdDInt32 gmx_simdcall cvtR2I(SimdDouble a)
{
    return { _mm512_cvtpd_epi32(a.simdInternal_) };
}

static inline SimdDInt32 gmx_simdcall cvttR2I(SimdDouble a)
{
    return { _mm512_cvttpd_epi32(a.simdInternal_) };
}

static inline SimdDouble gmx_simdcall cvtI2R(SimdDInt32 a)
{
    return { _mm512_cvtepi32_pd(a.simdInternal_) };
}

static inline SimdDIBool gmx_simdcall cvtB2IB(SimdDBool a)
{
    return { a.simdInternal_ };
}

static inline SimdDBool gmx_simdcall cvtIB2B(SimdDIBool a)
{
    return { static_cast<__mmask8>(a.simdInternal_) };
}

static inline void gmx_simdcall cvtF2DD(SimdFloat f, SimdDouble* d0, SimdDouble* d1)
{
    d0->simdInternal_ = _mm512_cvtps_pd(_mm512_castps512_ps256(f.simdInternal_));
    d1->simdInternal_ = _mm512_cvtps_pd(
            _mm512_castps512_ps256(_mm512_shuffle_f32x4(f.simdInternal_, f.simdInternal_, 0xEE)));
}

static inline SimdFloat gmx_simdcall cvtDD2F(SimdDouble d0, SimdDouble d1)
{
    __m512 f0 = _mm512_castps256_ps512(_mm512_cvtpd_ps(d0.simdInternal_));
    __m512 f1 = _mm512_castps256_ps512(_mm512_cvtpd_ps(d1.simdInternal_));
    return { _mm512_shuffle_f32x4(f0, f1, 0x44) };
}

} // namespace gmx

#endif // GMX_SIMD_IMPL_X86_AVX_512_SIMD_DOUBLE_H