Partial Summed Area Table Generation for Memory Access Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The generation and usage of summed area tables (SATs) in data processing are hindered by the need for frequent access to global memory, leading to slow processing speeds due to the high number of memory accesses required.
Innovation Solution
The method involves dividing data into partial blocks, calculating row and column sums, and performing propagation operations to create partial SATs, which are then used to complete specific areas of the SAT, reducing the need for full SAT generation and minimizing global memory access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the complete SAT is generated using data from global memory, then the processing can be performed using the complete SAT, but the number of global memory accesses increases significantly, slowing down the processing speed
Solution Approach 1:
The patent divides the complete SAT into multiple partial SATs, where each partial SAT corresponds to a specific region or block of the original data. This segmentation allows the system to process and store only the necessary partial SATs rather than generating the entire SAT, thereby reducing the number of global memory accesses while maintaining the reliability of processing for the required regions.
2Productivity
If data is divided into multiple blocks to generate partial SATs, then the number of global memory accesses is reduced, but the complexity of managing and combining partial SATs increases
Solution Approach 1:
The patent introduces a spatial dimension to the SAT management by organizing partial SATs according to their corresponding data blocks in the original image. Each partial SAT is tagged with metadata indicating its position and scope, allowing the system to efficiently locate, retrieve, and combine the necessary partial SATs without complex management overhead. This dimensional organization transforms the complexity from managing arbitrary fragments to managing spatially-organized regions.
Data Source
AI summary
A method of processing data includes: generating at least one partial summed area table (SAT) by dividing data stored in a global memory, and obtaining a row sum and a column sum of each of the at least one partial SAT; performing propagation on the obtained row sums and column sums; when a process of utilizing the SAT is executed, completing at least a partial area of the SAT necessary for the process based on the at least one partial SAT, and the row sums and column sums on which the propagation has been performed; and performing the process by using the completed partial area.