mirror of https://github.com/alibaba/MNN.git
37 lines
1.2 KiB
C++
37 lines
1.2 KiB
C++
|
//
|
||
|
|
// GemmInt8Executor.hpp
|
||
|
|
// MNNCPU
|
||
|
|
//
|
||
|
|
// Created by jbyang on 2023/3/16.
|
||
|
|
//
|
||
|
|
|
||
|
|
#ifndef GemmInt8Executor_hpp
|
||
|
|
#define GemmInt8Executor_hpp
|
||
|
|
|
||
|
|
#include "Int8FunctionsOpt.h"
|
||
|
|
#include "backend/cpu/CPUConvolution.hpp"
|
||
|
|
|
||
|
|
namespace MNN {
|
||
|
|
class GemmInt8Executor : public CPUConvolution {
|
||
|
|
public:
|
||
|
|
GemmInt8Executor(Backend* bn, std::shared_ptr<ResourceInt8> resource, const Convolution2D *conv2D, decltype(CoreInt8Functions::Int8GemmKernel),
|
||
|
|
std::vector<int32_t> bias);
|
||
|
|
virtual ~GemmInt8Executor();
|
||
|
|
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;
|
||
|
|
protected:
|
||
|
|
int mThreadNums;
|
||
|
|
int mTileCnt;
|
||
|
|
int mKernelX;
|
||
|
|
int mKernelY;
|
||
|
|
std::shared_ptr<Tensor> mInputCol;
|
||
|
|
std::vector<float> mScaleData;
|
||
|
|
std::vector<int32_t> mQuantBias;
|
||
|
|
std::shared_ptr<CPUConvolution::ResourceInt8> mResource;
|
||
|
|
ConvolutionCommon::Im2ColParameter mIm2ColParamter;
|
||
|
|
CPUConvolution::MutableResourceInt8 mMutableResource;
|
||
|
|
decltype(CoreInt8Functions::Int8GemmKernel) mGemmKernel;
|
||
|
|
};
|
||
|
|
} // namespace MNN
|
||
|
|
#endif /* DeconvInt8Executor_hpp */
|