Module tensor 

Tensor 模块：为图神经网络和高性能计算提供张量支持

本模块实现了 God-Graph 的 tensor 基础设施，包括：

• Dense tensor（密集张量）：基于 ndarray 的 N 维数组
• Sparse tensor（稀疏张量）：COO, CSR, BSR 格式
• Tensor 操作：矩阵乘法、转置、归约等
• 多后端支持：NdArray, Dfdx (GPU), Candle
• 内存池优化：减少分配开销
• 梯度检查点：降低反向传播内存占用

特性

• 后端抽象：通过 trait 系统支持多种 backend（ndarray, dfdx, candle）
• 稀疏格式：COO（坐标格式）、CSR（压缩稀疏行）、BSR（块稀疏行）
• SIMD 优化：使用 wide crate 实现 SIMD 向量化
• 内存对齐：64 字节缓存行对齐，避免 false sharing
• 内存池：可复用的张量分配，适用于迭代算法

示例

use god_gragh::tensor::{DenseTensor, TensorBase};

// 创建 2x3 密集张量
let data = vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0];
let tensor = DenseTensor::from_vec(data, vec![2, 3]);

assert_eq!(tensor.shape(), &[2, 3]);
assert_eq!(tensor.ndim(), 2);

Re-exports

pub use traits::COOView;

pub use traits::DType;

pub use traits::Device;

pub use traits::SparseTensorOps;

pub use traits::TensorBase;

pub use traits::TensorOps;

pub use dense::DenseTensor;

pub use sparse::COOTensor;

pub use sparse::CSRTensor;

pub use sparse::SparseTensor;

pub use error::TensorError;

pub use types::EdgeFeatures;

pub use types::NodeFeatures;

pub use types::TensorEdge;

pub use types::TensorNode;

pub use types::AdjacencyMatrix;

pub use types::DegreeMatrix;

pub use backend::NdArrayStorage;

pub use backend::TensorStorage;

pub use backend::UnifiedStorage;

pub use pool::ArenaStats;

pub use pool::ArenaTensor;

pub use pool::PoolConfig;

pub use pool::PoolStats;

pub use pool::PooledTensor;

pub use pool::TensorArena;

pub use pool::TensorPool;

pub use gnn::Aggregator;

pub use gnn::GATConv;

pub use gnn::GCNConv;

pub use gnn::GraphSAGE;

pub use gnn::IdentityMessage;

pub use gnn::LinearMessage;

pub use gnn::MaxAggregator;

pub use gnn::MeanAggregator;

pub use gnn::MessageFunction;

pub use gnn::MessagePassingLayer;

pub use gnn::SumAggregator;

pub use gnn::UpdateFunction;

pub use graph_tensor::GraphFeatureExtractor;

pub use graph_tensor::GraphReconstructor;

pub use graph_tensor::GraphAdjacencyMatrix;

pub use graph_tensor::GraphBatch;

pub use graph_tensor::GraphTensorExt;

pub use differentiable::DifferentiableEdge;

pub use differentiable::DifferentiableGraph;

pub use differentiable::DifferentiableNode;

pub use differentiable::EdgeEditOp;

pub use differentiable::EdgeEditPolicy;

pub use differentiable::EditOperation;

pub use differentiable::GradientConfig;

pub use differentiable::GradientRecorder;

pub use differentiable::GraphTransformer;

pub use differentiable::GumbelSoftmaxSampler;

pub use differentiable::NodeEditOp;

pub use differentiable::StructureEdit;

pub use differentiable::ThresholdEditPolicy;

pub use unified_graph::EdgeData;

pub use unified_graph::NodeData;

pub use unified_graph::UnifiedConfig;

pub use unified_graph::UnifiedGraph;

pub use decomposition::lie_algebra::lie_exponential;

pub use decomposition::lie_algebra::lie_logarithm;

pub use decomposition::lie_algebra::skew_symmetric_projection;

pub use decomposition::qr::orthogonalize;

pub use decomposition::qr::qr_decompose;

pub use decomposition::svd::low_rank_approx;

pub use decomposition::svd::svd_decompose;

pub use decomposition::tensor_ring::compress_tensor_ring;

pub use decomposition::tensor_ring::tensor_ring_decompose;

pub use decomposition::tensor_ring::TensorRing;

Modules

backend

Tensor 后端抽象：支持多种 backend 实现

decomposition

Tensor Decomposition Tools

dense

密集张量实现

differentiable

可微图结构变换模块

error

Tensor 错误类型

gnn

GNN（图神经网络）原语模块

graph_tensor

Graph-Tensor Integration: Specialized implementations for seamless graph-tensor conversion

ops

Tensor 高级操作

pool

Tensor 内存池：优化张量分配性能

sparse

稀疏张量实现

traits

Tensor Trait 系统：定义张量的抽象接口

types

Tensor 感知的节点和边类型

unified_graph

UnifiedGraph: 统一图结构，集成 DifferentiableGraph 和 ComputeGraph