mirror of https://github.com/alibaba/MNN.git
54 lines
1.8 KiB
C++
54 lines
1.8 KiB
C++
//
|
|
// NPUInstanceNorm.cpp
|
|
// MNN
|
|
//
|
|
// Created by MNN on b'2020/10/15'.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#include "NPUInstanceNorm.hpp"
|
|
#include "NPUBackend.hpp"
|
|
|
|
using namespace std;
|
|
|
|
namespace MNN {
|
|
|
|
NPUInstanceNorm::NPUInstanceNorm(MNN::Backend *b, const MNN::Op *op, const std::vector<Tensor *> &inputs, const std::vector<MNN::Tensor *> &outputs) : NPUCommonExecution(b, op) {}
|
|
|
|
ErrorCode NPUInstanceNorm::onResize(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
|
|
mNpuBackend->setNetworkInput(inputs, mOp);
|
|
|
|
auto xOp = mNpuBackend->getInputOps(mOp);
|
|
auto opName = mOp->name()->str();
|
|
|
|
auto slope = mOp->main_as_BatchNorm()->slopeData();
|
|
mScale = hiai::op::Const(opName + "_scale");
|
|
{
|
|
ge::TensorDesc fdesc(ge::Shape({1,slope->size(),1,1}),ge::DT_FLOAT);
|
|
ge::TensorPtr filter = std::make_shared<ge::Tensor>();
|
|
filter->SetTensorDesc(fdesc);
|
|
filter->SetData((uint8_t *)slope->data(), slope->size() * sizeof(float));
|
|
mScale.set_attr_value(filter);
|
|
}
|
|
|
|
auto bias = mOp->main_as_BatchNorm()->biasData();
|
|
mBias = hiai::op::Const(opName + "_bias");
|
|
{
|
|
ge::TensorDesc fdesc(ge::Shape({1,bias->size(),1,1}),ge::DT_FLOAT);
|
|
ge::TensorPtr filter = std::make_shared<ge::Tensor>();
|
|
filter->SetTensorDesc(fdesc);
|
|
filter->SetData((uint8_t *)bias->data(), bias->size() * sizeof(float));
|
|
mBias.set_attr_value(filter);
|
|
}
|
|
shared_ptr<hiai::op::InstanceNorm> insNorm(new hiai::op::InstanceNorm(opName));
|
|
(*insNorm).set_input_x(*xOp.get())
|
|
.set_input_gamma(mScale)
|
|
.set_input_beta(mBias);
|
|
|
|
mNpuBackend->setOutputOps(mOp, {insNorm}, outputs);
|
|
return NO_ERROR;
|
|
}
|
|
|
|
NPUCreatorRegister<TypedCreator<NPUInstanceNorm>> __instanceNorm_op(OpType_InstanceNorm);
|
|
|
|
} // namespace MNN
|