BVH Traversal Pruning Using Local Ray Parameters in GPUs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Ray tracing in graphics processing units (GPUs) is computationally expensive due to the need for extensive intersection tests between rays and primitives, which consumes significant power and reduces performance, particularly in battery-powered devices.
Innovation Solution
Implementing prune control circuitry in GPUs to determine and prune intersection tests based on local ray parameter values (tLocal) and max ray parameter values (tMax) during BVH traversal, specifically for certain types of nodes like instance nodes, to reduce unnecessary ray transform operations and intersection tests.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional ray tracing techniques traverse all nodes in the bounding volume hierarchy, then complete scene coverage is achieved, but computational expense and power consumption increase significantly
Solution Approach 1:
The patent performs preliminary comparison of tLocal with tMax before executing intersection tests. By checking whether the local ray parameter exceeds the maximum ray parameter in advance, the system determines whether traversal is necessary, preventing unnecessary computational operations and reducing power consumption while maintaining complete scene coverage where needed
Solution Approach 2:
The patent applies different traversal strategies to different nodes based on their local characteristics. Nodes where tLocal > tMax are pruned with zero intersection tests, while other nodes undergo full traversal. This localized differentiation optimizes power consumption by applying computational effort only where geometrically necessary
2Measurement precision
If ray tracing performs intersection tests for all bounding regions, then accurate visibility determination is achieved, but the number of operations increases leading to performance bottlenecks
Solution Approach 1:
The system performs a preliminary check comparing tLocal against tMax before conducting intersection tests. This pre-evaluation identifies nodes that cannot possibly contain visible geometry, allowing the system to skip intersection tests for those nodes while maintaining accuracy for nodes where tests are actually needed, thus improving rendering performance without sacrificing visibility determination accuracy
Solution Approach 2:
The patent extracts and removes unnecessary intersection test operations by identifying nodes where tLocal > tMax. These nodes are excluded from the traversal process entirely, eliminating redundant computational operations while preserving the accuracy of visibility determination for the remaining necessary nodes
3Manufacturing precision
If the graphics processor traverses more nodes in the bounding volume hierarchy, then more primitives are tested for intersection, but power consumption increases especially in battery-powered devices
Solution Approach 1:
Before traversing any node in the bounding volume hierarchy, the system performs a preliminary comparison of tLocal with tMax. Nodes where tLocal exceeds tMax are identified in advance and pruned, preventing the processor from consuming power on unnecessary traversal and intersection tests while maintaining complete testing of all geometrically relevant primitives
Solution Approach 2:
The patent applies pruning selectively to individual nodes based on their local ray parameter characteristics. Each node is evaluated independently, and only those nodes where tLocal > tMax are pruned. This localized approach ensures that intersection test completeness is maintained for all nodes where it matters geometrically while eliminating energy waste on nodes where it cannot matter
Data Source
AI summary
Techniques are disclosed relating to graphics processors that support ray tracing. In disclosed embodiments, ray intersect circuitry is configured to access a traversal stack used for traversal of multiple levels of a bounding volume hierarchy (BVH) acceleration data structure (ADS) according to a depth-first search to retrieve: coordinates of a first bounding region for a child node and a local ray parameter value that indicates a point along a ray at which an intersection with a second bounding region for the child node's parent node was detected. The accelerator circuitry may compare the local ray parameter value with an end ray parameter value to determine whether to traverse to the child node as part of traversal of the BVH.