Simulink 中的数据类型优化
数据类型优化是一种自动化的数据类型转换工作流，它自动循环访问各种定点配置以选择最佳异构数据类型，同时满足对系统数值行为的容差约束。优化力图最小化目标函数，如总位宽或总运算符计数，以期通过定点数据类型来实现高效设计。
您可以在命令行中使用 fxpopt 函数或在定点工具中使用优化的定点转换工作流来优化数据类型。
工具
| 定点工具 | Convert a floating-point model to a fixed-point model |
函数
fxpopt | Optimize data types of a system |
类
fxpOptimizationOptions | 指定数据类型优化的选项 |
OptimizationSolution | Optimized fixed-point implementation of system |
OptimizationResult | Result after optimizing fixed-point system |
主题
- Data Type Conversion Overview
Convert data types in your model to fixed point in one of three ways.
- Best Practices for Fixed-Point Conversion Workflow
Use best practices when converting a model to fixed point.
- 指定行为约束
使用信号容差和 Model Verification 模块来验证定点实现的行为。
- Optimize Fixed-Point Data Types for a System
Optimize data types in a system based on specified tolerances.
- Optimize the Fixed-Point Data Types of a System Using the Fixed-Point Tool
Use the Fixed-Point Tool to optimize the data types of a system using multiple simulation scenarios.
- Optimize Data Types Using Multiple Simulation Scenarios
Define multiple simulation scenarios for range collection and verification.
- Perform Data Type Optimization with Custom Behavioral Constraints
Use Model Verification blocks to specify custom behavioral constraints for data type optimization with
fxpopt. - Use Custom Cost Functions for Optimized Fixed-Point Conversion
Customize objective function for optimized fixed-point conversion.
- Model Configuration Changes Made During Data Type Optimization
Changes made to model configuration parameters during data type optimization with
fxpopt.
疑难解答
Blocks That Do Not Support Fixed-Point Data Types
Troubleshoot blocks that are not supported for fixed-point conversion.
Data Type Optimization Not Successful
Troubleshoot errors thrown during data type optimization using the
fxpopt function.
Follow best practices and avoid unsupported MATLAB® Function block features.
精选示例
Optimize Data Types for an FPGA with DSP Slices
Use the addSpecification method to achieve better mapping for
product and gain blocks.
Convert Neural Network Algorithms to Fixed-Point Using fxpopt and Generate HDL Code
Convert a neural network to fixed point using fixed-point optimization and lookup table optimization.
Image Denoising Using Fixed-Point Quantized Restricted Boltzmann Machine Algorithm
Convert a Restricted Boltzmann Machine (RBM) algorithm to an embedded-efficient version.
Use Design Evolution Manager with Fixed-Point Tool
Use the Design Evolution Manager app to manage different design versions while using the Fixed-Point Tool and the Lookup Table Optimizer to convert a floating-point model to use fixed-point data types. Although this example demonstrates how to convert models to use fixed-point data types, you can use the Design Evolution Manager app to organize and analyze any work in which you create different versions of files in your design.
