wehub-resource-sync
55 lines
1.3 KiB
C++
55 lines
1.3 KiB
C++
//
|
|
// SqueezeTorch.cpp
|
|
// MNNConverter
|
|
//
|
|
// Created by MNN on 2021/05/10.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#include <stdio.h>
|
|
#include "torchOpConverter.hpp"
|
|
|
|
DECLARE_OP_CONVERTER(UnSqueezeTorch);
|
|
|
|
MNN::OpType UnSqueezeTorch::opType() {
|
|
return MNN::OpType_Unsqueeze;
|
|
}
|
|
MNN::OpParameter UnSqueezeTorch::type() {
|
|
return MNN::OpParameter_SqueezeParam;
|
|
}
|
|
std::vector<int> UnSqueezeTorch::inputTensorIdx() {
|
|
return {0};
|
|
}
|
|
|
|
void UnSqueezeTorch::run(MNN::OpT* dstOp, const torch::jit::Node* node, TorchScope* scope) {
|
|
auto param = new MNN::SqueezeParamT;
|
|
if (node->inputs().size() > 1) {
|
|
param->squeezeDims.push_back(getValue<int64_t>(node->input(1)));
|
|
}
|
|
dstOp->main.value = param;
|
|
}
|
|
|
|
REGISTER_CONVERTER(UnSqueezeTorch, unsqueeze);
|
|
|
|
DECLARE_OP_CONVERTER(SqueezeTorch);
|
|
|
|
MNN::OpType SqueezeTorch::opType() {
|
|
return MNN::OpType_Squeeze;
|
|
}
|
|
MNN::OpParameter SqueezeTorch::type() {
|
|
return MNN::OpParameter_SqueezeParam;
|
|
}
|
|
std::vector<int> SqueezeTorch::inputTensorIdx() {
|
|
return {0};
|
|
}
|
|
|
|
void SqueezeTorch::run(MNN::OpT* dstOp, const torch::jit::Node* node, TorchScope* scope) {
|
|
auto param = new MNN::SqueezeParamT;
|
|
if (node->inputs().size() > 1) {
|
|
param->squeezeDims.push_back(getValue<int64_t>(node->input(1)));
|
|
}
|
|
dstOp->main.value = param;
|
|
}
|
|
|
|
REGISTER_CONVERTER(SqueezeTorch, squeeze);
|