Prefetch Table for Ray Tracing Latency Reduction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Ray tracing schemes face inefficiencies due to high computational requirements and memory bandwidth usage in rendering three-dimensional images, particularly in the traversal and intersection test processes, leading to processor stalls and performance degradation.

Innovation Solution

An image processing apparatus and method utilizing a prefetch table to store and manage rendering information, allowing for local memory access instead of global memory, reducing latency and data loading time by verifying local index information and transferring relevant information to a shader for shading and ray generation processes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If ray tracing scheme is used to create high quality image, then image quality is improved, but rendering rate deteriorates due to large amount of calculation

Engineering Contradiction:
Improveimage qualityVSAvoidrendering rate
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent applies preliminary action by prefetching rendering information from global memory to local memory before it is actually needed for ray-triangle intersection tests. The system predicts which rendering information will be needed based on current ray intersection points and prefetches it in advance, thereby reducing memory access latency and improving rendering rate while maintaining image quality.

Inventive Principle:
Principle #10Preliminary action

2Quantity of substance

If rendering information is accessed from global memory, then data availability is improved, but data loading time and latency increase

Engineering Contradiction:
Improvedata availabilityVSAvoiddata loading time
Core Design Contradiction:
Quantity of substanceVSLoss of time

Solution Approach 1:

The patent applies segmentation by dividing the memory system into two parts: global memory for storing all rendering information and local memory for storing frequently accessed rendering information. By segmenting the memory hierarchy and placing hot data in local memory, the system reduces data loading time while maintaining data availability.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces local memory as an intermediary between global memory and the processing units. Local memory acts as a buffer that stores rendering information prefetched from global memory, reducing the time required to access data while maintaining availability of rendering information.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Productivity

If prefetch table is used to store rendering information, then data access efficiency is improved, but memory bandwidth usage increases

Engineering Contradiction:
Improvedata access efficiencyVSAvoidmemory bandwidth usage
Core Design Contradiction:
ProductivityVSQuantity of substance

Solution Approach 1:

The patent applies local quality by making the prefetch table selective rather than universal. Instead of prefetching all possible rendering information, the system uses a prefetch table to identify and prefetch only the specific rendering information that is likely to be needed based on current ray intersection patterns. This selective approach improves data access efficiency while controlling memory bandwidth usage.

Inventive Principle:
Principle #3Local quality

Data Source

PatentUS9449423B2Apparatus and method for image processing
Publication Date: 2016.09.20 SAMSUNG ELECTRONICS CO LTD
  • US9449423B2 patent drawing
  • US9449423B2 patent drawing
  • US9449423B2 patent drawing

AI summary

A method for processing a three-dimensional (3D) image of a ray tracing scheme may be performed by an image processing apparatus by verifying whether local index information matching intersection point information of a ray is present within a prefetch table when intersection point information is received, and by transferring, to a shader, rendering information stored in a local memory based on the local index information, when the local index information matching the intersection point information is present.