Unified GPU Memory Access via CPU Page Table Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing application programming interfaces (APIs) for multimedia tasks, such as DirectX, are wasteful in integrated graphical processing units with unified memory architecture, as they require copying data from one memory unit to another, leading to inefficient memory management, performance loss, and increased power consumption.
Innovation Solution
Modifying existing APIs to allow direct access to graphical processing unit memory by designating resources as shared, creating a link between staging and non-staging resources, and mapping this memory into central processing unit page tables, thereby reducing unnecessary memory copies and conserving power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If data is copied from graphical processing unit memory to a resource using existing APIs, then the application can access the data, but memory efficiency decreases and power consumption increases
Solution Approach 1:
The patent merges the staging resource and non-staging resource into a single shared memory space. By allowing the application to directly access graphical processing unit memory through a unified interface, the need for separate copy operations between distinct memory regions is eliminated, thereby reducing power consumption while maintaining data accessibility.
2Ease of operation
If data is copied from graphical processing unit memory to a resource, then the application can access the data, but memory management efficiency decreases
Solution Approach 1:
The patent combines the previously separate staging and non-staging memory regions into a unified memory architecture. This merging allows the application to access graphical processing unit memory directly without intermediate copy operations, significantly improving memory management efficiency while maintaining ease of data access through the modified API interface.
3Ease of operation
If data is copied from graphical processing unit memory to a resource, then the application can access the data, but application speed decreases
Solution Approach 1:
The patent unifies the memory architecture by merging staging and non-staging resources into a single accessible memory space. This elimination of intermediate copy operations removes a significant performance bottleneck, allowing the application to access graphical processing unit memory directly at full speed while maintaining the same ease of access through the modified API.
4Adaptability or versatility
If existing APIs are used with integrated graphical processing units, then compatibility is maintained, but resource wastage increases
Solution Approach 1:
The patent modifies the existing DirectX API to provide universal access to graphical processing unit memory across different GPU architectures. The same API interface can now function efficiently with both traditional discrete GPUs and modern integrated GPUs with unified memory, eliminating the need for architecture-specific implementations and reducing memory wastage through direct access.
Data Source
AI summary
Various embodiments are presented herein that may allow an application direct access to graphical processing unit memory. An apparatus and a computer-implemented method may include accessing allocated graphical processing unit memory of a second resource via a link from a first resource. The allocated graphical processing unit memory may be mapped into one or more page tables of a central processing unit. A virtual address of the graphical processing unit memory from the one or more page tables of the central processing unit may be sent to the application.