mirror of https://github.com/alibaba/MNN.git
66 lines
2.5 KiB
C++
66 lines
2.5 KiB
C++
//
|
|
// ShapeSpaceToBatchND.cpp
|
|
// MNN
|
|
//
|
|
// Created by MNN on 2019/01/10.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#include "shape/SizeComputer.hpp"
|
|
#include "core/TensorUtils.hpp"
|
|
|
|
namespace MNN {
|
|
class SpaceToBatchNDSizeComputer : public SizeComputer {
|
|
public:
|
|
virtual bool onComputeSize(const MNN::Op* op, const std::vector<Tensor*>& inputs,
|
|
const std::vector<Tensor*>& outputs) const override {
|
|
MNN_ASSERT(outputs.size() == 1);
|
|
MNN_ASSERT(inputs.size() == 1 || inputs.size() == 3);
|
|
auto input = inputs[0];
|
|
auto output = outputs[0];
|
|
|
|
int blockSize = 0;
|
|
const int *blockData, *paddingData;
|
|
if (inputs.size() == 3) {
|
|
blockSize = inputs[1]->length(0);
|
|
blockData = inputs[1]->host<int32_t>();
|
|
paddingData = inputs[2]->host<int32_t>();
|
|
} else {
|
|
auto paramter = op->main_as_SpaceBatch();
|
|
const auto blockShape = paramter->blockShape();
|
|
const auto paddings = paramter->padding();
|
|
blockSize = blockShape->dims()->data()[0];
|
|
blockData = blockShape->int32s()->data();
|
|
paddingData = paddings->int32s()->data();
|
|
}
|
|
int batch = input->batch();
|
|
for (int i = 0; i < blockSize; ++i) {
|
|
batch *= blockData[i];
|
|
}
|
|
auto format = TensorUtils::getDescribe(input)->dimensionFormat;
|
|
output->buffer().type = input->buffer().type;
|
|
output->buffer().dimensions = input->buffer().dimensions;
|
|
output->setLength(0, batch);
|
|
TensorUtils::getDescribe(output)->dimensionFormat = format;
|
|
if (MNN_DATA_FORMAT_NHWC != format) {
|
|
output->setLength(1, input->length(1));
|
|
for (int i = 0; i < blockSize; ++i) {
|
|
int paddedLength = input->length(2+i) + paddingData[2 * i] + paddingData[2 * i+1];
|
|
int outputLength = paddedLength / blockData[i];
|
|
output->setLength(i+2, outputLength);
|
|
}
|
|
} else {
|
|
for (int i = 0; i < blockSize; ++i) {
|
|
int paddedLength = input->length(1 + i) + paddingData[2 * i] + paddingData[2 * i+1];
|
|
int outputLength = paddedLength / blockData[i];
|
|
output->setLength(i+1, outputLength);
|
|
}
|
|
output->setLength(1+blockSize, input->length(1+blockSize));
|
|
}
|
|
return true;
|
|
}
|
|
};
|
|
|
|
REGISTER_SHAPE_INPUTS(SpaceToBatchNDSizeComputer, OpType_SpaceToBatchND, std::vector<int>({1, 2}));
|
|
} // namespace MNN
|