mirror of https://github.com/alibaba/MNN.git
51 lines
1.6 KiB
C++
51 lines
1.6 KiB
C++
|
//
|
||
|
|
// ShapeTile.cpp
|
||
|
|
// MNN
|
||
|
|
//
|
||
|
|
// Created by MNN on 2019/01/10.
|
||
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
||
|
|
//
|
||
|
|
|
||
|
|
#include "Macro.h"
|
||
|
|
#include "SizeComputer.hpp"
|
||
|
|
|
||
|
|
namespace MNN {
|
||
|
|
|
||
|
|
class TileComputer : public SizeComputer {
|
||
|
|
public:
|
||
|
|
virtual bool onComputeSize(const MNN::Op* op, const std::vector<Tensor*>& inputs,
|
||
|
|
const std::vector<Tensor*>& outputs) const override {
|
||
|
|
MNN_ASSERT(2 == inputs.size());
|
||
|
|
MNN_ASSERT(1 == outputs.size());
|
||
|
|
auto& input = inputs[0]->buffer();
|
||
|
|
auto multiples = inputs[1];
|
||
|
|
MNN_ASSERT(multiples->getType().code == halide_type_int);
|
||
|
|
auto& output = outputs[0]->buffer();
|
||
|
|
// Expected multiples argument to be a 1-D vector of length input.dimensions
|
||
|
|
MNN_ASSERT(1 == multiples->buffer().dimensions)
|
||
|
|
MNN_ASSERT(input.dimensions == multiples->buffer().dim[0].extent);
|
||
|
|
const int inputDims = input.dimensions;
|
||
|
|
::memcpy(output.dim, input.dim, input.dimensions * sizeof(halide_dimension_t));
|
||
|
|
output.dimensions = inputDims;
|
||
|
|
output.type = input.type;
|
||
|
|
|
||
|
|
std::shared_ptr<Tensor> multipleTemp;
|
||
|
|
|
||
|
|
// copy data from device to host if needed
|
||
|
|
if (!multiples->host<int32_t>() && multiples->deviceId()) {
|
||
|
|
multipleTemp.reset(Tensor::createHostTensorFromDevice(multiples, true));
|
||
|
|
multiples = multipleTemp.get();
|
||
|
|
}
|
||
|
|
|
||
|
|
for (int i = 0; i < inputDims; ++i) {
|
||
|
|
output.dim[i].extent = input.dim[i].extent * multiples->host<int32_t>()[i];
|
||
|
|
}
|
||
|
|
|
||
|
|
return true;
|
||
|
|
}
|
||
|
|
};
|
||
|
|
|
||
|
|
REGISTER_SHAPE(TileComputer, OpType_Tile);
|
||
|
|
|
||
|
|
} // namespace MNN
|