Module zero_copy 

Zero-copy data loading utilities for efficient memory management

This module provides zero-copy tensor operations and memory management utilities for high-performance data processing pipelines. It enables efficient handling of large datasets without unnecessary memory allocations.

Features

• Zero-copy tensors: Work directly with existing memory without copying
• Memory pools: Reuse allocated tensors to reduce allocation overhead
• Buffer management: Efficient buffer reuse in data pipelines
• Memory mapping: Direct access to file data without loading into memory
• Thread-safe operations: Concurrent access to shared memory pools

Structs

BufferManager

Buffer manager for efficient buffer reuse in data pipelines

MemoryMappedLoader

Memory-mapped data loader for large datasets

TensorPool

Memory pool for reusing allocated tensors to avoid allocation/deallocation overhead

ZeroCopyTensor

Zero-copy tensor wrapper that avoids unnecessary memory allocation

Functions

create_buffer_manager

Convenience function to create a buffer manager

create_tensor_pool

Convenience function to create a tensor pool

zero_copy_from_slice

Convenience function to create a zero-copy tensor from a slice

zero_copy_from_vec

Convenience function to create a zero-copy tensor from a vector