TVM Compiler Heterogeneous Platform Deployment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deployment methods are limited to single hardware platforms with a single instruction set and architecture, lacking a viable method for deploying deep learning networks on heterogeneous platforms that consist of multiple types of instruction sets and architecture units.
Innovation Solution
A deployment method and device based on a TVM compiler that divides nodes in a deep learning network's topological structure into target device blocks corresponding to different device types, performs space distribution on these blocks, and deploys the network across a heterogeneous platform using the TVM compiler-generated topological structure, network parameters, and function information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deployment methods are used for single hardware platforms with one instruction set and one architecture calculation unit, then the deployment process is simple, but the method cannot support heterogeneous platforms with multiple device types
Solution Approach 1:
The patent segments the deployment process into distinct modules: topological structure analysis, node classification, space distribution, and deployment execution. Each module handles specific aspects of heterogeneous platform deployment, making the overall complex process manageable and systematic. The segmentation allows independent optimization of each module while maintaining overall coherence.
Solution Approach 2:
The patent introduces a TVM compiler as an intermediary between the deep learning network and the heterogeneous platform. The TVM compiler serves as a mediator that translates high-level network definitions into platform-specific deployment configurations, handling the complexity of mapping nodes to different device types automatically.
2Adaptability or versatility
If nodes are divided into target device blocks and space distribution is performed, then heterogeneous platform deployment is enabled, but the deployment process becomes more complex
Solution Approach 1:
The patent performs preliminary actions by pre-classifying nodes into target device blocks and pre-distributing space before actual deployment. This preparation phase establishes all necessary mappings and configurations in advance, so that the actual deployment execution becomes a straightforward process of executing pre-planned operations.
Solution Approach 2:
The patent dynamically adjusts deployment parameters based on the specific heterogeneous platform configuration. By changing parameters such as device type mappings, space allocation sizes, and node assignments according to the target platform's characteristics, the system adapts to different hardware configurations while maintaining a consistent deployment framework.
Data Source
AI summary
Provided are a deployment method and apparatus for a heterogeneous platform based on a TVM compiler, and a terminal device. The method comprises: obtaining a topological structure diagram, network parameters and function information of each function of a deep learning network generated by a TVM compiler (S101); on the basis of information of a node in the topological structure diagram and the device type of a heterogeneous platform, dividing the node in the topological structure diagram into target device blocks corresponding to the device type (S102); performing space allocation on the target device blocks to obtain space information of the target device blocks (S103); and deploying the deep learning network to the heterogeneous platform on the basis of the topological structure diagram, the network parameters, the function information, the target device blocks and the spatial information of the target device blocks (S104). According to the method, the node is divided into target device blocks of different device types, and deployment of different devices in the heterogeneous platform is completed by means of the target device blocks of different types.