GPU Resource APIs for Hierarchical Memory Management

The patent divides memory resources into hierarchical contexts (root context, child contexts, and sub-contexts). Each context manages its own memory space and resources independently, allowing fine-grained control over memory allocation. This segmentation enables programs to access only the memory portions they need at each hierarchical level, reducing overall memory consumption while maintaining execution efficiency.

The patent introduces a hierarchical dimension to memory management by creating multiple levels of contexts (root → child → sub-context). This dimensional approach allows memory to be organized and accessed in a tree-like structure, enabling programs to navigate and allocate memory resources across different hierarchical levels, thereby optimizing both access speed and memory utilization.

The patent pre-allocates and organizes memory resources into hierarchical contexts before program execution begins. By establishing the context hierarchy and allocating memory spaces in advance, the system eliminates the need for dynamic resource negotiation during program runtime, thereby reducing computational overhead while maintaining full computational capability.

The patent introduces contexts as intermediary layers between the program and physical memory resources. These contexts act as mediators that manage memory allocation and access, reducing the computational burden on programs by handling resource management tasks automatically through the hierarchical context structure.