/*******************************************************************************
 *
 * MIT License
 *
 * Copyright (C) 2022-2024 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in the Software without restriction, including without limitation the rights
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 *******************************************************************************/

#pragma once

#include <cmath>
#include <hip/hip_runtime.h>
#include <hipblaslt/hipblaslt.h>
#include <hipblaslt/hipblaslt_xfloat32.h>
#include <immintrin.h>
#include <type_traits>

/* ============================================================================================ */
// Helper function to truncate float to bfloat16

inline __host__ hip_bfloat16 float_to_bfloat16_truncate(float val)
{
#if defined(__HIP_PLATFORM_AMD__)
    return hip_bfloat16(val, hip_bfloat16::truncate_t::truncate);
#else
    return __float2bfloat16_rd(val);
#endif
}

/* ============================================================================================ */
/*! \brief negate a value */

template <class T>
inline __device__ __host__ T negate(T x)
{
    return -x;
}

template <>
inline __device__ __host__ hip_bfloat16 negate(hip_bfloat16 x)
{

#if defined(__HIP_PLATFORM_AMD__)
    x.data ^= 0x8000;
    return x;
#else
    auto raw = __nv_bfloat16_raw(x);
    raw.x ^= 0x8000;
    return raw;
#endif
}

#if defined(HIPBLASLT_USE_FP4)
template <>
inline __device__ __host__ hipblaslt_f4x2 negate(hipblaslt_f4x2 x)
{
    x.__x ^= 0x8; // left fp4 element
    x.__x ^= 0x80; // right fp4 element
    return x;
}
#endif

#if defined(HIPBLASLT_USE_FP6)
template <>
inline __device__ __host__ hipblaslt_f6x16 negate(hipblaslt_f6x16 x)
{
    union
    {
        hipblaslt_f6x16_storage real;
        uint8_t                 tmp[12];
    } cvt;

    cvt.real = x.data;

    for(int i = 0; i < sizeof(cvt.tmp); i += 3)
    {
        cvt.tmp[i] ^= 0x20;
        cvt.tmp[i + 1] ^= 0x8;
        cvt.tmp[i + 2] ^= 0x82;
    }

    x.data = cvt.real;
    return x;
}
#endif

#if defined(HIPBLASLT_USE_BF6)
template <>
inline __device__ __host__ hipblaslt_bf6x16 negate(hipblaslt_bf6x16 x)
{
    union
    {
        hipblaslt_bf6x16_storage real;
        uint8_t                  tmp[12];
    } cvt;

    cvt.real = x.data;

    for(int i = 0; i < sizeof(cvt.tmp); i += 3)
    {
        cvt.tmp[i] ^= 0x20;
        cvt.tmp[i + 1] ^= 0x8;
        cvt.tmp[i + 2] ^= 0x82;
    }

    x.data = cvt.real;
    return x;
}
#endif

template <>
inline __device__ __host__ hipblaslt_f8_fnuz negate(hipblaslt_f8_fnuz x)
{
    x.__x ^= 0x80;
    return x;
}

template <>
inline __device__ __host__ hipblaslt_bf8_fnuz negate(hipblaslt_bf8_fnuz x)
{
    x.__x ^= 0x80;
    return x;
}

template <>
inline __device__ __host__ hipblaslt_f8 negate(hipblaslt_f8 x)
{
    x.__x ^= 0x80;
    return x;
}

template <>
inline __device__ __host__ hipblaslt_bf8 negate(hipblaslt_bf8 x)
{
    x.__x ^= 0x80;
    return x;
}

// Complex float specialization
template <>
inline __device__ __host__ std::complex<float> negate(std::complex<float> x)
{
    return std::complex<float>(-x. real(), -x. imag());
}

// Complex double specialization
template <>
inline __device__ __host__ std::complex<double> negate(std::complex<double> x)
{
    return std::complex<double>(-x. real(), -x. imag());
}

// Helper function to reduce intermediate precision and the output type are the same as the input type.
template <typename TxDLi, typename TxDLo, typename Ti>
inline void type_to_xdl_math_op_type(Ti* in, size_t s)
{
    //To filter out the case that input type is not supported by xDL Math Op.
    //Currently, xDL Math Op supports in:float -> intermediat:xf32 -> out:float
    constexpr bool needCast = !std::is_same<TxDLi, Ti>() && std::is_same<TxDLo, Ti>();
    if(!needCast)
        return;

    //Cast input type to xDl math op type, using type alians to avoid the casting error.
    using castType = std::conditional_t<needCast, TxDLi, Ti>;
    for(size_t i = 0; i < s; i++)
        in[i] = static_cast<Ti>(static_cast<castType>(in[i]));
}

template <typename T>
inline __host__ __device__ bool hipblaslt_isnan(std::complex<T> arg,
                                                std::enable_if_t<!std::is_integral<T>::value, int> = 0)
{
    return std::isnan(arg.real()) || std::isnan(arg.imag());
}

template <typename T>
inline __host__ __device__ bool hipblaslt_isnan(std::complex<T> arg,
                                                std::enable_if_t<std::is_integral<T>::value, int> = 0)
{
    return false;
}