TheRock/rocm-libraries/projects/hipblaslt/tensilelite/client/include/ClientProblemFactory.hpp

/*******************************************************************************
 *
 * MIT License
 *
 * Copyright (C) 2022-2026 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.
 *
 *******************************************************************************/

#pragma once

#include <Tensile/Activation.hpp>
#include <Tensile/ContractionProblem.hpp>
#include <Tensile/KernelLanguageTypes.hpp>
#include <Tensile/Tensile.hpp>

#include "ProgramOptions.hpp"

#include <cstddef>

namespace TensileLite
{
    namespace Client
    {

        class ClientProblemFactory
        {
        public:
            ClientProblemFactory(po::variables_map const& args);
            ~ClientProblemFactory();

            std::vector<std::shared_ptr<ContractionProblem>> const& problems() const;

            size_t workspaceSize() const
            {
                return m_maxWorkspaceSize;
            }

        private:
            void createProblems(std::vector<ContractionProblemGemm>& rv);

            std::vector<std::shared_ptr<ContractionProblem>> m_problems;

            ContractionProblemGemm::FreeIndices  m_freeIndices;
            ContractionProblemGemm::BatchIndices m_batchIndices;
            ContractionProblemGemm::BoundIndices m_boundIndices;

            std::vector<rocisa::DataType>                 m_tensorTypes;
            std::vector<std::vector<std::vector<size_t>>> m_tensorStrides;

            std::vector<rocisa::DataType> m_constantTypes;
            std::vector<double>           m_constantValues;

            bool        m_stridedBatched;
            bool        m_groupedGemm;
            bool        m_highPrecisionAccumulate;
            bool        m_deterministicMode;
            bool        m_cEqualsD;
            int         m_useBias;
            int         m_biasSrc;
            std::string m_useScaleAB;
            bool        m_useScaleCD;
            int         m_useScaleAlphaVec;
            bool        m_useSynchronizer;
            bool        m_useE;
            bool        m_useGradient     = false;
            bool        m_useGateResidual = false;
            bool        m_outputAmaxD;

            int                              m_sparse;
            KernelLanguage                   m_kernelLanguage;
            PerformanceMetric                m_performanceMetric;
            ActivationType                   m_activationType;
            std::vector<rocisa::DataType>    m_biasTypeArgs;
            std::vector<rocisa::DataType>    m_gateTypeArgs;
            std::vector<int>                 m_factorDimArgs;
            std::vector<bool>                m_icacheFlushArgs;
            bool                             m_activationNoGuard;
            std::vector<ActivationType>      m_activationEnumArg;
            // StreamK=5 hybrid-mode toggle values to test. Each element
            // generates a separate ContractionProblemGemm variant with
            // setParams().setStreamKTileSchedulingMode(value); the SK5
            // kernel then runs the static (0), dynamic (1), or
            // heuristic-picked AUTO (2) path accordingly. Empty or
            // single-element vectors keep the host's default behavior.
            std::vector<int>                 m_streamKHybridMode;
            size_t                           m_maxWorkspaceSize = 0;
            rocisa::DataType                 m_computeInputTypeA;
            rocisa::DataType                 m_computeInputTypeB;
            rocisa::DataType                 m_f32XdlMathOp;
            rocisa::DataType                 m_activationComputeType;
            std::vector<std::vector<size_t>> m_problemSizes;
            bool                             m_useUserArgs;
            bool                             m_swizzleTensorA;
            bool                             m_swizzleTensorB;
            int                              m_metadataLayout;
            int                              m_mxBlockA;
            int                              m_mxBlockB;
            bool                             m_padMXScaleTensorFreeDim;

            TensorOps m_aOps;
            TensorOps m_bOps;
            TensorOps m_cOps;
            TensorOps m_dOps;
        };

    } // namespace Client
} // namespace TensileLite