TheRock/rocm-libraries/projects/rocwmma/test/dlrm/detail/dlrm_dot_lds.hpp

/*******************************************************************************
 *
 * MIT License
 *
 * Copyright (C) 2021-2025 Advanced Micro Devices, Inc. All rights reserved.
 *
 * 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.
 *
 *******************************************************************************/

#ifndef DLRM_DOT_LDS_DETAIL_HPP
#define DLRM_DOT_LDS_DETAIL_HPP

#include "../helper_macros.hpp"
#include "device/dlrm_dot_bwd_lds.hpp"
#include "device/dlrm_dot_fwd_lds.hpp"
#include "dlrm_kernel_base.hpp"

namespace rocwmma
{

    // Wrapper into the actual device function
    template <uint32_t TileSize, typename DataT, typename MappingLds>
    struct DlrmDotLdsKernel final : public DlrmKernelBase<TileSize, DataT>
    {
    private:
        using Base = DlrmKernelBase<TileSize, DataT>;

    public:
        DlrmDotLdsKernel() {}
        ~DlrmDotLdsKernel() final {}

        uint32_t ldsUsage() const final
        {
            auto blockDims = this->blockDim();
            return 2 * sizeof(DataT)
                   * ((blockDims.x / Base::DeviceInfo::instance()->warpSize()) + blockDims.y)
                   * (TileSize * TileSize);
        }
        typename Base::KernelFwdFunc kernelFwdImpl() const final
        {
// Generate a nested switch case
// for TBlockX = [32, 64, 128, 256]
// and TBlockY = [1, 2, 4]
#define CASE_IMPL_ASSIGN2(TBLOCK_X, TBLOCK_Y) \
    return typename Base::KernelFwdFunc(      \
        dlrmDotFwdLds<DataT, TileSize, MappingLds, TBLOCK_X, TBLOCK_Y>);

#define SWITCH_BODY_TBLOCK_X(TBLOCK_Y) \
    ROCWMMA_SWITCH_BODY3_ARG2(Base::mTBlockX, CASE_IMPL_ASSIGN2, 32u, 64u, 128u, TBLOCK_Y)

#define DISPATCH_BODY ROCWMMA_SWITCH_BODY3_ARG1(Base::mTBlockY, SWITCH_BODY_TBLOCK_X, 1u, 2u, 4u)

            // Invoke dispatch
            DISPATCH_BODY

// Clean up macro space
#undef CASE_IMPL_ASSIGN2
#undef SWITCH_BODY_TBLOCK_X
#undef DISPATCH_BODY

            // Default case
            return [](const DataT* __restrict, // input
                      DataT* __restrict, // output
                      float*, // acc
                      uint32_t, // m
                      uint32_t, // k
                      uint32_t, // b
                      uint32_t, // inputBatchOffset
                      uint32_t, // outputBatchOffset
                      uint32_t) {};
        }

        typename Base::KernelBwdFunc kernelBwdImpl() const final
        {
// Generate a nested switch case
// for TBlockX = [32, 64, 128, 256]
// and TBlockY = [1, 2, 4]
#define CASE_IMPL_ASSIGN2(TBLOCK_X, TBLOCK_Y) \
    return typename Base::KernelBwdFunc(      \
        dlrmDotBwdLds<DataT, TileSize, MappingLds, TBLOCK_X, TBLOCK_Y>);

#define SWITCH_BODY_TBLOCK_X(TBLOCK_Y) \
    ROCWMMA_SWITCH_BODY3_ARG2(Base::mTBlockX, CASE_IMPL_ASSIGN2, 32u, 64u, 128u, TBLOCK_Y)

#define DISPATCH_BODY ROCWMMA_SWITCH_BODY3_ARG1(Base::mTBlockY, SWITCH_BODY_TBLOCK_X, 1u, 2u, 4u)

            // Invoke dispatch
            DISPATCH_BODY

// Clean up macro space
#undef CASE_IMPL_ASSIGN2
#undef SWITCH_BODY_TBLOCK_X
#undef DISPATCH_BODY

            // Default case
            return [](const DataT* __restrict, // input
                      const DataT* __restrict, // upstreamGrad
                      DataT* __restrict, // grad
                      DataT* __restrict, // bottomMlpGrad
                      DataT* __restrict, // acc
                      uint32_t, // m
                      uint32_t, // k
                      uint32_t, // b
                      uint32_t, // inputBatchOffset
                      uint32_t, // upstreamBatchOffset
                      uint32_t) {}; // accBatchOffset
        }

        typename Base::KernelTrilFunc kernelTrilImpl() const final
        {
            return typename Base::KernelTrilFunc(trilReconstructLds<DataT>);
        }
    };

    // This is the GeneratorImpl class
    struct DlrmDotLdsGenerator
    {
        // Indices to test parameters
        enum : uint32_t
        {
            DataT      = 0,
            TileSize   = 1,
            MappingLds = 2
        };

        using ResultT = std::shared_ptr<KernelI>;

        template <typename... Ts>
        static ResultT generate(std::tuple<Ts...> testParams)
        {
            // Map GTest params to Kernel params
            using TestParamsT = std::tuple<Ts...>;
            using KernelT     = DlrmDotLdsKernel<std::tuple_element_t<TileSize, TestParamsT>::value,
                                             std::tuple_element_t<DataT, TestParamsT>,
                                             std::tuple_element_t<MappingLds, TestParamsT>>;

            return std::make_shared<KernelT>();
        }
    };

} // namespace rocwmma

#endif // DLRM_DOT_LDS_DETAIL_HPP