wehub-resource-sync
42 lines
1.1 KiB
C++
42 lines
1.1 KiB
C++
//
|
|
// QNNLayerNorm.hpp
|
|
// MNN
|
|
//
|
|
// Created by MNN on b'2025/04/10'.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#ifndef MNN_QNNLAYERNORM_HPP
|
|
#define MNN_QNNLAYERNORM_HPP
|
|
|
|
#include "QNNCommonExecution.hpp"
|
|
|
|
namespace MNN {
|
|
namespace QNN {
|
|
#ifdef ENABLE_QNN_ONLINE_FINALIZE
|
|
|
|
class QNNLayerNorm : public QNNCommonExecution {
|
|
public:
|
|
QNNLayerNorm(Backend *backend, const Op *op, Tensor * input);
|
|
virtual ErrorCode onResize(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) override;
|
|
virtual ErrorCode onEncode(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) override;
|
|
private:
|
|
ErrorCode onEncodeNormWithPermute(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs);
|
|
void createGammaBeta(Qnn_DataType_t dataType);
|
|
private:
|
|
Qnn_DataType_t mQnnDataType;
|
|
int mInputDim;
|
|
Tensor::DimensionType mDimType;
|
|
float mEpsilon;
|
|
bool mUseRMSNorm;
|
|
int mRealAxis;
|
|
int mGammaBetaSize = 0;
|
|
std::vector<float> mGammaData;
|
|
std::vector<float> mBetaData;
|
|
};
|
|
#endif
|
|
} // end namespace MNN
|
|
} // end namespace QNN
|
|
|
|
#endif // end MNN_QNNLAYERNORM_HPP
|