reference_gemm.hpp

// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.

#pragma once

#include "ck/utility/common_header.hpp"
#include "ck/library/utility/host_tensor.hpp"

template <typename ADataType, typename BDataType, typename CDataType, typename AccDataType>
void reference_gemm(const Tensor<ADataType>& a_m_k,
                    const Tensor<BDataType>& b_n_k,
                    Tensor<CDataType>& c_m_n)
{
    auto f_mk_kn_mn = [&](auto m, auto n) {
        const int K = a_m_k.mDesc.GetLengths()[1];

        AccDataType v_acc = 0;

        for(int k = 0; k < K; ++k)
        {
            ADataType v_a = a_m_k(m, k);
            BDataType v_b = b_n_k(n, k);

            v_acc += ck::type_convert<AccDataType>(v_a) * ck::type_convert<AccDataType>(v_b);
        }

        c_m_n(m, n) = ck::type_convert<CDataType>(v_acc);
    };

    make_ParallelTensorFunctor(f_mk_kn_mn,
                               c_m_n.mDesc.GetLengths()[0],
                               c_m_n.mDesc.GetLengths()[1])(std::thread::hardware_concurrency());
}