TheRock/rocm-libraries/projects/miopen/test/gtest/conv_tensor_gen.hpp

/*******************************************************************************
 *
 * MIT License
 *
 * Copyright (c) 2023 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 <gtest/gtest.h>
#include <hip_float8.hpp>

#include "../random.hpp"

template <typename T>
inline T RanGenData()
{
    return prng::gen_canonical<T>();
}

template <>
inline float8_fnuz RanGenData()
{
    return prng::gen_A_to_B(static_cast<float8_fnuz>(-1.0f), static_cast<float8_fnuz>(1.0f));
}

template <>
inline bfloat8_fnuz RanGenData()
{
    return prng::gen_A_to_B(static_cast<bfloat8_fnuz>(-1.0f), static_cast<bfloat8_fnuz>(1.0f));
}

template <typename T>
struct GenData
{
    template <class... Ts>
    T operator()(Ts...) const
    {
        return RanGenData<T>();
    }
};

template <typename T>
T RanGenWeights()
{
    return prng::gen_A_to_B(static_cast<T>(-0.5), static_cast<T>(0.5));
}

// Shift FP16 distribution towards positive numbers,
// otherwise Winograd FP16 validation fails.
template <>
inline half_float::half RanGenWeights()
{
    return prng::gen_A_to_B(static_cast<half_float::half>(-1.0 / 3.0),
                            static_cast<half_float::half>(0.5));
}

template <>
inline float8_fnuz RanGenWeights()
{
    const auto tmp = prng::gen_canonical<float>() > 0.5f ? 0.0f : 2.0f;
    // 1 in 2 chance of number being positive
    const auto sign = prng::gen_canonical<float>() > 0.5f ? -1.0f : 1.0f;
    return static_cast<float8_fnuz>(
        static_cast<float>(std::numeric_limits<float8_fnuz>::epsilon()) * sign * tmp);
}

template <>
inline bfloat8_fnuz RanGenWeights()
{
    const auto tmp = prng::gen_canonical<float>() > 0.5f ? 0.0f : 2.0f;
    // 1 in 2 chance of number being positive
    const auto sign = prng::gen_canonical<float>() > 0.5f ? -1.0f : 1.0f;
    return static_cast<bfloat8_fnuz>(
        static_cast<float>(std::numeric_limits<bfloat8_fnuz>::epsilon()) * sign * tmp);
}

template <typename T>
struct GenWeights
{
    template <class... Ts>
    T operator()(Ts...) const
    {
        return RanGenWeights<T>();
    }
};

template <typename T, typename Tacc>
struct GenConvData
{
    /// \note CHWNc filter layout is not supported (different storage layout)
    GenConvData(const std::vector<std::size_t>& filter, unsigned group_count = 1)
    {
        static_assert(std::is_integral_v<T> == std::is_integral_v<Tacc>);
        static_assert(sizeof(Tacc) >= sizeof(T));

        constexpr auto is_integral = std::is_integral_v<T>;

        // Multiply all dimensions except K to get the number of additions
        const auto num_add = std::accumulate(filter.cbegin() + 1,
                                             filter.cend(),
                                             static_cast<std::size_t>(1),
                                             std::multiplies<std::size_t>()) /
                             group_count;

        constexpr auto max_acc_v = std::numeric_limits<Tacc>::max();
        if constexpr(is_integral)
        {
            // "B" must be > 0
            if(num_add >= max_acc_v)
                throw std::runtime_error("filter is too big");
        }
        const auto tmp_B = static_cast<Tacc>(std::sqrt(max_acc_v / (num_add + 1)));

        if constexpr(std::is_same_v<T, Tacc>)
        {
            B = tmp_B;
        }
        else
        {
            constexpr T max_v = std::numeric_limits<T>::max();
            B                 = (tmp_B >= max_v) ? max_v : tmp_B;
        }

        if constexpr(!is_integral)
        {
            // Limit the range of FP
            constexpr auto limit = static_cast<float>(std::numeric_limits<uint32_t>::max());
            if(B > limit)
                // cppcheck-suppress floatConversionOverflow
                B = limit;
        }

        A = -B;
    }

    template <class... Ts>
    T operator()(Ts...) const
    {
        return prng::gen_A_to_B(A, B);
    }

private:
    T A;
    T B;
};