GPU Context Deallocation APIs for Memory-Efficient Resource Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer programs face challenges in optimizing memory and resource usage, leading to inefficiencies in performance.
Innovation Solution
Implementing application programming interfaces (APIs) to manage resources by creating, destroying, and subdividing contexts and sub-contexts, as well as obtaining and generating resource descriptors, which allow for more efficient allocation and management of resources on graphics processing units (GPUs).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If computer programs use significant memory and resources to perform computations, then computational tasks can be executed, but memory efficiency and resource utilization deteriorate
Solution Approach 1:
The patent segments GPU resources by creating multiple contexts and sub-contexts, each managing its own data structures and resources independently. This segmentation allows different computational tasks to share the same physical memory through virtual addressing, improving memory efficiency while maintaining program performance
Solution Approach 2:
The patent introduces context descriptors and resource descriptors as intermediary data structures that manage the mapping between virtual and physical memory. These descriptors act as mediators that enable efficient resource allocation and deallocation without direct memory management overhead, resolving the contradiction between memory efficiency and performance
2Adaptability or versatility
If data structures are allocated for GPU resource management, then resource allocation capability improves, but memory consumption increases
Solution Approach 1:
The patent implements a nested hierarchy where context descriptors contain arrays of resource descriptors, which in turn contain pointers to actual GPU resources. This nested structure allows compact representation of complex resource relationships, providing high adaptability with minimal memory overhead
Solution Approach 2:
The patent implements automatic deallocation of data structures when contexts are destroyed. The system recovers memory by deallocating context descriptors and their associated resource descriptors, ensuring that memory consumption does not permanently increase even as resource management capability improves
Data Source
AI summary
Apparatuses, systems, and techniques to perform computing operations. In at least one embodiment, a processor performs an application programming interface to deallocate one or more data structures to indicate which of one or more streaming multiprocessors of one or more processors are to be used to perform one or more software threads.