Node-Based Arithmetic Processing for Tensor-Aware Data Distribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing arithmetic processing devices require software developers to create different programs for each node, increasing development burden and production costs due to the need for tensor division considerations.
Innovation Solution
An arithmetic processing device with a network of nodes, each containing multiple processor elements, utilizes a write-out, change, and read-out processing unit to manage data distribution and correspondence without deeply considering tensor division, allowing a common program to be used across nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If different programs are created for each node to handle tensor division, then the arithmetic processing device can process high-resolution image information, but the development burden and production costs increase
Solution Approach 1:
The patent implements a universal data distribution mechanism where a single program can be executed across all nodes. The controller automatically performs data division and transposition operations, making the system multi-functional in handling different tensor shapes without requiring node-specific software variations. This universal approach resolves the contradiction by maintaining processing capability while eliminating the need for multiple specialized programs.
Solution Approach 2:
The controller acts as an intermediary between the input data and the nodes. It handles the complex tensor division and transposition operations centrally, then distributes the processed data to appropriate nodes. This intermediary approach allows nodes to execute simple, identical programs while the controller manages the complexity of data distribution according to different tensor shapes.
2Manufacturing precision
If different programs are created for each node, then tensor division can be handled properly, but the development period increases due to verification requirements
Solution Approach 1:
The system implements self-service through automated data distribution. The controller automatically divides and transposes data according to the tensor shape, and nodes automatically receive and process their assigned data portions. This self-service mechanism eliminates the need for manual program customization and verification for each node, maintaining tensor division accuracy while significantly reducing the development period.
Solution Approach 2:
The patent changes the approach from modifying program code for each node to changing data parameters and distribution patterns. The controller adjusts data division and transposition based on tensor shape parameters, while nodes execute the same program with different data inputs. This parameter-based approach maintains processing accuracy while reducing development time.
3Productivity
If high-resolution image information is processed by dividing and arranging data across multiple nodes, then processing capability is improved, but the correspondence relationship between memory device and nodes becomes complex
Solution Approach 1:
The patent applies segmentation by dividing the data processing task into distinct phases: data division, transposition, and distribution. The controller segments the high-resolution image information into appropriate portions for each node based on tensor shape, establishing clear correspondence relationships without requiring complex manual configuration. This segmented approach maintains processing capability while simplifying the memory-node correspondence management.
Data Source
AI summary
An arithmetic processing device is configured from a network having a plurality of nodes, each of which includes a plurality of processor elements. The arithmetic processing device includes: a write-out processing unit that writes out data of image information, which is input, divided and transposed for each node, to a predetermined area of a memory device; a change processing unit that changes a correspondence relationship between the predetermined area of the memory device and the node in accordance with a tensor shape of the image information; and a read-out processing unit that reads out the data stored in the memory device to a corresponding node.