wehub-resource-sync
82 lines
2.0 KiB
C++
82 lines
2.0 KiB
C++
//
|
|
// ConcatTorch.cpp
|
|
// MNNConverter
|
|
//
|
|
// Created by MNN on 2021/05/12.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#include <stdio.h>
|
|
#include "torchOpConverter.hpp"
|
|
|
|
DECLARE_OP_CONVERTER(ListTorch);
|
|
|
|
MNN::OpType ListTorch::opType() {
|
|
return MNN::OpType_Pack;
|
|
}
|
|
MNN::OpParameter ListTorch::type() {
|
|
return MNN::OpParameter_PackParam;
|
|
}
|
|
std::vector<int> ListTorch::inputTensorIdx() {
|
|
return {-1};
|
|
}
|
|
|
|
void ListTorch::run(MNN::OpT* dstOp, const torch::jit::Node* node, TorchScope* scope) {
|
|
auto param = new MNN::PackParamT;
|
|
param->axis = 0;
|
|
if (getRealOpType(node) == "stack") {
|
|
dstOp->inputIndexes.pop_back();
|
|
auto axis = node->inputs().back();
|
|
param->axis = getValue<int64_t>(axis);
|
|
}
|
|
dstOp->main.value = param;
|
|
}
|
|
|
|
REGISTER_CONVERTER(ListTorch, stack);
|
|
REGISTER_CONVERTER(ListTorch, ListConstruct);
|
|
|
|
DECLARE_OP_CONVERTER(TupleTorch);
|
|
|
|
MNN::OpType TupleTorch::opType() {
|
|
return MNN::OpType_Concat;
|
|
}
|
|
MNN::OpParameter TupleTorch::type() {
|
|
return MNN::OpParameter_Axis;
|
|
}
|
|
std::vector<int> TupleTorch::inputTensorIdx() {
|
|
return {-1};
|
|
}
|
|
|
|
void TupleTorch::run(MNN::OpT* dstOp, const torch::jit::Node* node, TorchScope* scope) {
|
|
auto param = new MNN::AxisT;
|
|
param->axis = 0;
|
|
dstOp->main.value = param;
|
|
}
|
|
|
|
REGISTER_CONVERTER(TupleTorch, TupleConstruct);
|
|
|
|
DECLARE_OP_CONVERTER(ConcatTorch);
|
|
|
|
MNN::OpType ConcatTorch::opType() {
|
|
return MNN::OpType_Concat;
|
|
}
|
|
MNN::OpParameter ConcatTorch::type() {
|
|
return MNN::OpParameter_Axis;
|
|
}
|
|
std::vector<int> ConcatTorch::inputTensorIdx() {
|
|
return {};
|
|
}
|
|
|
|
void ConcatTorch::run(MNN::OpT* dstOp, const torch::jit::Node* node, TorchScope* scope) {
|
|
auto param = new MNN::AxisT;
|
|
const auto inputs = node->inputs();
|
|
auto tensorlist = inputs[0];
|
|
for (const auto input : tensorlist->node()->inputs()) {
|
|
dstOp->inputIndexes.push_back(scope->lookupTensor(input->debugName()));
|
|
}
|
|
param->axis = getValue<int64_t>(inputs[1]);
|
|
dstOp->main.value = param;
|
|
}
|
|
|
|
REGISTER_CONVERTER(ConcatTorch, cat);
|