GPU Rasterizer Wavefront Assignment via Locality Caching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for assigning wavefronts to processing elements in GPUs are inefficient, leading to increased memory access and reduced processing efficiency due to improper consideration of locality in graphics processing tasks.
Innovation Solution
The proposed solution involves maintaining a locality table to identify regions of an image frame and assigning wavefronts to processing elements based on their associated locality regions, reducing memory access by ensuring that texture data is cached and reused within the same region.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If wavefronts are assigned to processing elements using conventional techniques, then processing elements can execute operations in parallel, but memory access efficiency deteriorates due to lack of locality consideration
Solution Approach 1:
The patent applies local quality by assigning wavefronts to processing elements based on their spatial locality regions. Each wavefront is associated with specific image frame regions, and processing elements are selected that have cached the corresponding texture data for those regions. This ensures that each processing element operates on data locally available in its cache, improving memory access efficiency while maintaining parallel processing capability.
Solution Approach 2:
The patent implements preliminary action by pre-loading texture data into processing element caches before the wavefronts that need it are assigned. The system identifies which texture data will be needed by upcoming wavefronts and loads it into the appropriate processing element caches in advance, so that when the wavefronts are executed, the data is already available locally without requiring memory bandwidth during execution.
2Ease of operation
If wavefronts are assigned without considering locality regions, then assignment is simple and fast, but data caching efficiency deteriorates leading to repeated memory loading
Solution Approach 1:
The patent uses feedback by implementing a locality table that tracks which processing elements are assigned to which image frame regions. When a wavefront needs to be assigned, the system queries this table to identify processing elements that have the required texture data cached for the wavefront's target regions. This feedback mechanism maintains assignment simplicity while ensuring data caching efficiency, as the table provides direct lookup information without complex computations.
Data Source
AI summary
An apparatus, such as a graphical processing unit (GPU), includes one or more processors configured to determine a plurality of first locality information of a received wave at a processing unit and to select a first processing element of a plurality of processing elements, the first processing unit having a plurality of second locality information from a previous wave that matches the plurality of first locality information to execute the received wave.