wehub-resource-sync
72 lines
1.5 KiB
C++
72 lines
1.5 KiB
C++
//
|
|
// ShapeTorch.cpp
|
|
// MNNConverter
|
|
//
|
|
// Created by MNN on 2021/05/11.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#include <stdio.h>
|
|
#include "torchOpConverter.hpp"
|
|
|
|
// size -> Shape
|
|
DECLARE_OP_CONVERTER(ShapeTorch);
|
|
|
|
MNN::OpType ShapeTorch::opType() {
|
|
return MNN::OpType_Extra;
|
|
}
|
|
MNN::OpParameter ShapeTorch::type() {
|
|
return MNN::OpParameter_Extra;
|
|
}
|
|
std::vector<int> ShapeTorch::inputTensorIdx() {
|
|
return {-1};
|
|
}
|
|
|
|
void ShapeTorch::run(MNN::OpT* dstOp, const torch::jit::Node* node, TorchScope* scope) {
|
|
auto extra = new MNN::ExtraT;
|
|
dstOp->main.value = extra;
|
|
extra->engine = "Torch";
|
|
extra->type = "size";
|
|
}
|
|
|
|
REGISTER_CONVERTER(ShapeTorch, size);
|
|
|
|
// dim -> Rank
|
|
DECLARE_OP_CONVERTER(RankTorch);
|
|
|
|
MNN::OpType RankTorch::opType() {
|
|
return MNN::OpType_Rank;
|
|
}
|
|
MNN::OpParameter RankTorch::type() {
|
|
return MNN::OpParameter_NONE;
|
|
}
|
|
std::vector<int> RankTorch::inputTensorIdx() {
|
|
return {0};
|
|
}
|
|
|
|
void RankTorch::run(MNN::OpT* dstOp, const torch::jit::Node* node, TorchScope* scope) {
|
|
dstOp->main.value = nullptr;
|
|
}
|
|
|
|
REGISTER_CONVERTER(RankTorch, dim);
|
|
|
|
// len -> Size
|
|
DECLARE_OP_CONVERTER(SizeTorch);
|
|
|
|
MNN::OpType SizeTorch::opType() {
|
|
return MNN::OpType_Size;
|
|
}
|
|
MNN::OpParameter SizeTorch::type() {
|
|
return MNN::OpParameter_NONE;
|
|
}
|
|
std::vector<int> SizeTorch::inputTensorIdx() {
|
|
return {0};
|
|
}
|
|
|
|
void SizeTorch::run(MNN::OpT* dstOp, const torch::jit::Node* node, TorchScope* scope) {
|
|
dstOp->main.value = nullptr;
|
|
}
|
|
|
|
REGISTER_CONVERTER(SizeTorch, len);
|
|
REGISTER_CONVERTER(SizeTorch, numel);
|