//
//  CPUMatMul.hpp
//  MNN
//
//  Created by MNN on 2018/08/06.
//  Copyright © 2018, Alibaba Group Holding Limited
//

#ifndef CPUMATMUL_HPP
#define CPUMATMUL_HPP

#include <functional>
#include "core/Execution.hpp"
#include "CPUBackend.hpp"
namespace MNN {

class CPUMatMul : public Execution {
public:
    CPUMatMul(Backend *backend, bool transposeA, bool transposeB, bool transposeC, bool multiThread);
    virtual ~CPUMatMul() = default;
    virtual ErrorCode onResize(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) override;
    virtual ErrorCode onExecute(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) override;
    void execute(const float* APtr, const float* BPtr, float* CPtr, const float* BiasPtr);

private:
    void _scheduleForVec(int e, int l, int h);
    void _scheduleForVecE(int e, int l, int h);
    bool mTransposeA;
    bool mTransposeB;
    bool mTransposeC;
    bool mSupportMultiThread = false;
    std::vector<std::pair<std::function<void(int)>, int>> mPreFunctions;
    bool mUseBiasDirectly = false;
    MemChunk mTempA;
    MemChunk mTempB;
    MemChunk mTempC;
    MemChunk mTempBias;

    int mE;
    int mL;
    int mH;
    std::vector<float> mPostParameters;
    // For Execute Paramters
    const float* mA = nullptr;
    const float* mB = nullptr;
    const float* mBiasPtr = nullptr;
    float* mC = nullptr;

};
} // namespace MNN

#endif // CPUMATMUL_HPP