MNN/source/shape/ShapeTopKV2.cpp

49 lines
2.0 KiB
C++

//
// ShapeTopKV2.cpp
// MNN
//
// Created by MNN on 2019/01/10.
// Copyright © 2018, Alibaba Group Holding Limited
//
#include "shape/SizeComputer.hpp"
#include "core/Macro.h"
namespace MNN {
class TopKV2SizeComputer : public SizeComputer {
virtual bool onComputeSize(const MNN::Op* op, const std::vector<Tensor*>& inputs,
const std::vector<Tensor*>& outputs) const override {
MNN_ASSERT(2 == inputs.size() || 3 == inputs.size());
MNN_ASSERT(2 == outputs.size());
auto input = inputs[0];
auto kTensor = inputs[1];
MNN_ASSERT(kTensor->elementSize() == 1); // Scalar
MNN_ASSERT(kTensor->getType().code == halide_type_int);
const int k = kTensor->host<int32_t>()[0];
const int inputDimension = input->buffer().dimensions;
int axis = (inputs.size() == 3 ? inputs[2]->host<int32_t>()[0] : inputDimension - 1);
if (axis < 0) axis += input->dimensions();
// outputs: 0 --> data, 1 --> index
auto outputData = outputs[0];
outputData->buffer().dimensions = inputDimension;
memcpy(outputData->buffer().dim, input->buffer().dim, inputDimension * sizeof(halide_dimension_t));
outputData->buffer().dim[axis].extent = k;
outputData->buffer().type = input->buffer().type;
auto outputIndices = outputs[1];
outputIndices->buffer().dimensions = inputDimension;
memcpy(outputIndices->buffer().dim, input->buffer().dim, inputDimension * sizeof(halide_dimension_t));
outputIndices->buffer().dim[axis].extent = k;
outputIndices->setType(MNN::DataType_DT_INT32);
TensorUtils::getDescribe(outputs[0])->dimensionFormat = TensorUtils::getDescribe(inputs[0])->dimensionFormat;
TensorUtils::getDescribe(outputs[1])->dimensionFormat = TensorUtils::getDescribe(inputs[1])->dimensionFormat;
return true;
}
};
REGISTER_SHAPE_INPUTS(TopKV2SizeComputer, OpType_TopKV2, (std::vector<int>{1,2}));
} // namespace MNN