Memory allocation using graphs
Patent Information
- Authority / Receiving Office
- GB · GB
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2022-06-15
- Publication Date
- 2026-04-15
AI Technical Summary
Existing memory allocation techniques for data structures representing operations and dependencies are inefficient, requiring additional computing resources and not effectively utilizing memory outside of these structures, particularly in parallel computing platforms like CUDA.
The use of graph code nodes, such as MemAlloc and MemFree nodes, within a data structure to manage memory allocation and deallocation, allowing for efficient memory management by representing operations and dependencies, and enabling reuse and sharing of physical memory across graphs.
This approach improves memory allocation efficiency by reducing synchronization requirements and allowing for effective reuse and sharing of memory, exceeding the available memory on a GPU, while maintaining proper ownership and validation of allocated memory.