2023-12-27 17:26:44 +08:00
|
|
|
//
|
|
|
|
|
// ConvBufLowMemoryExecution.hpp
|
|
|
|
|
// MNN
|
|
|
|
|
//
|
|
|
|
|
// Created by MNN on 2023/10/12.
|
|
|
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
|
|
|
//
|
|
|
|
|
|
|
|
|
|
#ifdef MNN_LOW_MEMORY
|
|
|
|
|
#ifndef MNN_OPENCL_BUFFER_CLOSED
|
|
|
|
|
#ifndef ConvBufLowMemoryExecution_hpp
|
|
|
|
|
#define ConvBufLowMemoryExecution_hpp
|
|
|
|
|
#include "core/ConvolutionCommon.hpp"
|
|
|
|
|
#include "ConvBufExecution.hpp"
|
|
|
|
|
|
|
|
|
|
namespace MNN {
|
|
|
|
|
namespace OpenCL {
|
|
|
|
|
|
|
|
|
|
struct ConvBufResource {
|
|
|
|
|
const Convolution2DCommon *conv2dCommonParams;
|
|
|
|
|
std::shared_ptr<cl::Buffer> kernelBuffer;
|
|
|
|
|
std::shared_ptr<Tensor> filter;
|
|
|
|
|
std::shared_ptr<Tensor> dequantScale;
|
|
|
|
|
std::shared_ptr<Tensor> dequantOffset;
|
|
|
|
|
std::shared_ptr<Tensor> bias;
|
2024-02-29 16:21:40 +08:00
|
|
|
int mKernelWidth;
|
|
|
|
|
int mKernelHeight;
|
|
|
|
|
std::vector<int> mStrides{1, 1};
|
|
|
|
|
std::vector<int> mDilations{1, 1};
|
2023-12-27 17:26:44 +08:00
|
|
|
std::set<std::string> buildOptions;
|
|
|
|
|
bool conv1x1Opt = false;
|
|
|
|
|
bool gemmOpt = false;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
class ConvBufLowMemoryExecution : public ConvBufCommonExecution {
|
|
|
|
|
public:
|
|
|
|
|
ConvBufLowMemoryExecution(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs, const MNN::Op *op, Backend *backend);
|
|
|
|
|
ConvBufLowMemoryExecution(std::shared_ptr<ConvBufResource> resource, const Op* op, Backend* b);
|
|
|
|
|
virtual ~ConvBufLowMemoryExecution();
|
|
|
|
|
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;
|
|
|
|
|
virtual bool onClone(Backend* bn, const Op* op, Execution** dst) override;
|
|
|
|
|
private:
|
|
|
|
|
void getInfoFromOpLowMemory(std::shared_ptr<ConvolutionCommon::Int8Common> & quanCommon);
|
|
|
|
|
void set1x1WeightLowMemory(int packCout, int packCin, void * filterDataPtr, std::shared_ptr<ConvolutionCommon::Int8Common> & quanCommon);
|
|
|
|
|
void setGeneralWeightLowMemory(void * filterDataPtr, std::shared_ptr<ConvolutionCommon::Int8Common> & quanCommon);
|
|
|
|
|
void tune1x1CaseLowMemory(Tensor * input, Tensor * output);
|
|
|
|
|
void tuneGeneralCaseLowMemory(Tensor * input, Tensor * output);
|
|
|
|
|
void tuneGemmLowMemory(Tensor * input, Tensor * output);
|
|
|
|
|
std::shared_ptr<ConvBufResource> mResource;
|
|
|
|
|
const Convolution2D *mConv2dParams;
|
|
|
|
|
std::vector<int> mPaddings{0, 0};
|
|
|
|
|
std::vector<uint32_t> mGlobalWorkSize{1, 1, 1};
|
|
|
|
|
std::vector<uint32_t> mLocalWorkSize{1, 1, 1, 1};
|
|
|
|
|
cl::Kernel mKernel;
|
|
|
|
|
uint32_t mMaxWorkGroupSize;
|
|
|
|
|
int mOutputChannel;
|
|
|
|
|
int mInputChannel;
|
|
|
|
|
void *mFilterDataPtr = nullptr;
|
|
|
|
|
bool mLowMemoryFlag = false;
|
|
|
|
|
int mNumQuantBit = 0;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
} // namespace OpenCL
|
|
|
|
|
} // namespace MNN
|
|
|
|
|
#endif /* ConvBufLowMemoryExecution_hpp */
|
|
|
|
|
#endif /* MNN_OPENCL_BUFFER_CLOSED */
|
|
|
|
|
#endif /* MNN_LOW_MEMORY */
|
