METHOD AND APPARATUS FOR SHARED VIRTUAL MEMORY TO MANAGE DATA COHERENCY IN A HETEROGENEOUS PROCESSING SYSTEM

Smart overview of the Invention

Innovative methods for managing data coherency in heterogeneous processing systems are presented, focusing on a scalable solution that employs shared virtual memory. This approach aims to enhance the granularity of coherency tracking by transitioning storage from dedicated hardware to system memory, specifically through the use of page table metadata. By doing so, a virtually limitless directory structure for coherency states can be established.

Challenges with Existing Solutions

Current hardware-based coherency tracking methods rely on inclusive caches, snoop filters, and other costly structures that consume significant die area and power. These solutions often struggle with scalability in high bandwidth processing environments, leading to inefficiencies. For instance, tracking ownership per cache line incurs overhead due to multiple passes through ownership pipelines, resulting in performance bottlenecks.

Benefits of the Proposed Method

The proposed system allows for the complete removal of hardware coherency tracking, which eliminates capacity-related penalties found in traditional hardware caches. By utilizing page tables to locate coherency information, the method integrates local coherency caching within CPUs and GPUs as part of a translation lookaside buffer (TLB), thereby optimizing resource usage and improving performance.

System Architecture

A typical processing system incorporating this method may include multiple processors and graphics processors, suitable for various applications such as gaming platforms or mobile devices. The architecture is designed to facilitate efficient communication between components through a processor bus and hubs that manage memory and input/output operations, enhancing overall system functionality.

Potential Applications

This technology can be applied across a range of devices beyond graphics processors, including general-purpose computing systems and embedded devices. Its versatility means it can support various processing needs while improving data coherency management, making it an essential advancement in modern computing technology.