Vector Operation Arrays With Multi-Path Computing Flexibility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional vector accelerator engines have limited functions and poor flexibility, while general-purpose graphics processing units (GPGPUs) have lower computational efficiency for vector computing.
Innovation Solution
A vector data computing method and apparatus that determines to-be-computed vector data and operation types, selects a target vector operation array and data path from preset vector operation units, and controls the array to compute the data through the path, utilizing heterogeneous or homogeneous vector accelerator engines to improve computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional vector accelerator engines are used, then computational efficiency is improved, but function richness and flexibility deteriorate
Solution Approach 1:
The vector operation unit is designed to support multiple operation data paths (first operation data path, second operation data path, third operation data path) within a single hardware unit. Each data path corresponds to different computing functions, allowing the same hardware to perform various vector operations including dot product, element-wise operations, and reduction operations, thereby achieving multi-functionality without sacrificing computational efficiency
Solution Approach 2:
The vector operation unit is segmented into multiple independent operation data paths, where each path can be independently configured and activated based on the specific computing function required. This segmentation allows the system to select and activate only the necessary data path for a given operation, maintaining high computational efficiency while providing access to diverse computing functions
2Adaptability or versatility
If general-purpose graphics processing units are used, then function richness is improved, but computational efficiency deteriorates
Solution Approach 1:
The vector operation unit employs dynamic configuration of operation data paths based on the type of vector operation required. The system can dynamically select and switch between different data paths (first for dot product, second for element-wise operations, third for reduction operations) during runtime, providing the flexibility of GPGPU while maintaining the specialized computational efficiency of dedicated hardware accelerators
Data Source
AI summary
Embodiments of this disclosure disclose a vector data computing method, an electronic device, and a storage medium. The method includes: determining to-be-computed vector data and a corresponding first operation type; determining, based on the first operation type, a target vector operation array and a target operation data path corresponding to the target vector operation array from at least one preset vector operation unit, where the preset vector operation unit supports one or more operation data paths, and any of the one or more operation data paths corresponds to one computing function; and controlling the target vector operation array to compute the to-be-computed vector data through the target operation data path, to obtain a computation result corresponding to the to-be-computed vector data. According to the embodiments of this disclosure, richness and diversity of computing functions and computational efficiency can be improved.