Compressed On-Chip Vertex Data in Extended Local Memory
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current graphics processing systems face limitations in efficiently processing vertex data due to constraints in memory capacity and bandwidth, particularly in handling large datasets required for advanced graphics rendering techniques like tile-based immediate mode rendering.
Innovation Solution
The implementation of compressed on-chip vertex data within an extended local memory architecture, which includes a memory extender that compresses and decompresses vertex attribute information, allowing for increased memory efficiency and improved processing capabilities by reducing the memory footprint while maintaining or enhancing the number of triangles that can be processed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If vertex data is stored in uncompressed format in local memory, then processing speed is improved, but memory capacity is exceeded
Solution Approach 1:
The patent applies compression algorithms to change the storage format parameter of vertex data from uncompressed to compressed form. This reduces the memory footprint by a factor of 2-4x while maintaining the same functional capability, allowing the system to store larger vertex buffers in the fixed-size local memory without sacrificing processing capability
Solution Approach 2:
The patent introduces a compression/decompression intermediary layer between the vertex buffer storage and the rendering pipeline. The vertex data is compressed before being stored in local memory and decompressed when needed for processing, acting as a mediator that enables efficient use of limited memory resources while maintaining processing performance
2Productivity
If more vertex data is stored in local memory, then the number of triangles that can be processed is increased, but bandwidth requirements increase
Solution Approach 1:
By changing the storage parameter from uncompressed to compressed format, the patent reduces the bandwidth required to transfer vertex data between memory and processing units. The compressed data requires 2-4x less bandwidth for the same amount of geometric content, enabling the system to process more triangles within the same bandwidth constraints
Solution Approach 2:
The patent segments the vertex data storage into compressed blocks within local memory, allowing selective loading and processing of only the necessary vertex data for current rendering operations. This reduces the overall bandwidth requirements by avoiding transfer of unnecessary data
3Productivity
If local memory size is increased to handle larger vertex buffers, then processing capability is improved, but hardware complexity increases
Solution Approach 1:
Rather than increasing the physical size of local memory, the patent changes the storage parameter to compressed format, effectively multiplying the available memory capacity by 2-4x without any physical expansion. This avoids the complexity of larger memory structures while achieving the same functional improvement
Solution Approach 2:
The patent implements software-based compression and decompression routines that replicate the functionality of hardware compression without requiring additional physical memory resources. This copying of data in compressed form achieves enhanced processing capability without increasing hardware complexity
Data Source
AI summary
An embodiment of a graphics apparatus may include an embedded local memory, and a memory extender communicatively coupled to the embedded local memory to extend the embedded local memory. The memory extender may be configured to compress information and store the compressed information in the embedded local memory. Additionally, or alternatively, the memory extender may be configured to expose the embedded local memory for non-local access. Other embodiments are disclosed and claimed.