MNN/source/backend/hiai/execution/NPUReduction.cpp

107 lines
4.2 KiB
C++

//
// NPUReduction.cpp
// MNN
//
// Created by MNN on b'2020/10/15'.
// Copyright © 2018, Alibaba Group Holding Limited
//
#include "NPUReduction.hpp"
#include "NPUBackend.hpp"
using namespace std;
namespace MNN {
NPUReduction::NPUReduction(MNN::Backend *b, const MNN::Op *op, const std::vector<Tensor *> &inputs, const std::vector<MNN::Tensor *> &outputs) : NPUCommonExecution(b, op) {
}
ErrorCode NPUReduction::onResize(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
mNpuBackend->setNetworkInput(inputs, mOp);
auto opName = mOp->name()->str();
auto type = mOp->main_as_ReductionParam()->operation();
auto xOp = mNpuBackend->getInputOps(mOp);
vector<int32_t> origAxis;
auto reduct = mOp->main_as_ReductionParam();
if (inputs.size() >= 2) {
for (int i = 0; i < inputs[1]->elementSize(); ++i) {
int32_t *reduce_dim = inputs[1]->host<int32_t>();
origAxis.push_back(reduce_dim[i]);
}
} else if (nullptr != reduct->dim()) {
for (int i = 0; i < reduct->dim()->size(); ++i) {
origAxis.push_back(reduct->dim()->data()[i]);
}
} else {
MNN_ASSERT(false);
}
mConstAxis = hiai::op::Const(opName + "_axis");
{
ge::TensorDesc fdesc(ge::Shape({static_cast<long>(origAxis.size())}), ge::FORMAT_ND, ge::DT_INT32);
ge::TensorPtr constTensor = std::make_shared<ge::Tensor>();
constTensor->SetTensorDesc(fdesc);
constTensor->SetData((uint8_t *)(origAxis.data()), origAxis.size()*sizeof(int32_t));
mConstAxis.set_attr_value(constTensor);
}
vector<int64_t> dims;
for (int32_t i = 0; i < outputs[0]->buffer().dimensions; i++) {
dims.push_back(outputs[0]->buffer().dim[i].extent);
}
shapeConst = hiai::op::Const(opName + "_shape_const");
{
ge::TensorDesc fdesc(ge::Shape({static_cast<int64_t>(dims.size())}),
ge::FORMAT_NCHW, ge::DT_INT32);
ge::TensorPtr filter = std::make_shared<ge::Tensor>();
filter->SetTensorDesc(fdesc);
filter->SetData((uint8_t *)dims.data(), dims.size() * sizeof(int32_t));
shapeConst.set_attr_value(filter);
}
if(type == ReductionType_MAXIMUM) {
shared_ptr<hiai::op::ReduceMax> reduction(new hiai::op::ReduceMax(opName));
(*reduction)
.set_input_x(*xOp.get()).set_input_axes(mConstAxis)
.set_attr_keep_dims(mOp->main_as_ReductionParam()->keepDims());
mNpuBackend->setOutputOps(mOp, {reduction}, outputs);
}else if(type == ReductionType_SUM) {
shared_ptr<hiai::op::ReduceSum> reduction(new hiai::op::ReduceSum(opName));
(*reduction)
.set_input_x(*xOp.get()).set_input_axes(mConstAxis)
.set_attr_keep_dims(mOp->main_as_ReductionParam()->keepDims());
mNpuBackend->setOutputOps(mOp, {reduction}, outputs);
}else if(type == ReductionType_MEAN) {
shared_ptr<hiai::op::ReduceMean> reduction(new hiai::op::ReduceMean(opName));
(*reduction)
.set_input_x(*xOp.get()).set_input_axes(mConstAxis)
.set_attr_keep_dims(reduct->keepDims());
if(reduct->keepDims() == false) {
shared_ptr<hiai::op::Reshape> reshape1(new hiai::op::Reshape(opName+"reshape1"));
(*reshape1).set_input_x(*reduction.get()).set_input_shape(shapeConst);
mNpuBackend->setOutputOps(mOp, {reduction,reshape1}, outputs);
} else {
mNpuBackend->setOutputOps(mOp, {reduction}, outputs);
}
} else if(type == ReductionType_ANY) {
shared_ptr<ge::op::ReduceAll> reduction(new ge::op::ReduceAll(opName));
vector<int64_t> axis;
for (int32_t j = 0; j < origAxis.size(); j++) {
axis.push_back(static_cast<int64_t>(origAxis[j]));
}
(*reduction)
.set_input_x(*xOp.get()).set_attr_axes(axis)
.set_attr_keep_dims(mOp->main_as_ReductionParam()->keepDims());
mNpuBackend->setOutputOps(mOp, {reduction}, outputs);
}else{
MNN_ERROR("npu reducton not support type : %d \n", type);
}
return NO_ERROR;
}
NPUCreatorRegister<TypedCreator<NPUReduction>> __reduction_op(OpType_Reduction);
} // namespace MNN