Knight

Overview

Knight is an efficient code generator which can effectively optimizing loop semantics within the target Simulink model.

Basic idea: Knight precisely analyses the interactions between model blocks to identify state-dependent blocks, and then effectively optimizes the redundant calculations caused by state-dependent blocks.

Requirements

• Simulink (Matlab R2022b)
• GCC (Version 11.3.0)
• Clang (Version 14.0.6)

The above versions of Simulink, GCC, and Clang are utilized in our evaluation.

Repository Structure

Experiments

The evaluation results on x86 and ARM architectures. Each sub-folder named by "Model_Name" contains the generated code and compiled executable files among different compilers and optimization levels.

For quick start and reproducing our evaluation results, you can simply run the following instructions.

sh build.sh
sh run.sh $execution_times$

Model_IR

This folder contains the parsed IR (Intermediate Representation). For a target Simulink model, we first translate it into readable MIR, and then convert the MIR into IL for code generation. The IL files with suffix (_ Optimized) is the IL files optimized by Knight.

Models

This folder contains the benchmark Simulink models. Each of the model is collected from academic works and our industry partner (HUAWEI).

Tools

The implementation of our tools. This folder contains different executable files (.exe) for different purpose. We will introduce how to use these files in detail in Usage Section.

Usage

Workflow of Toolsets

To generate efficient code, you should use the following executable files.

1. ParserForSimulink.exe

   Convert the Simulink model files (SLX) into customized IR (MIR), which contains the detailed model information, such as inport, outport, etc.

   Useful Flags:

   • -help: help information
   • -i: input file path of target Simulink model
   • -o: output file path of the generated MIR file

   Instruction Example:

   ParserForSimulink.exe -i $Simulink_Model_Path$ -o $Output_MIR_File_Path$

2. ILGenerator.exe

   Convert the MIR file into IL file, which contains simplified code information and can be used for redundancy elimination.

   Useful Flags:

   • -help: help information
   • -i: input file path of MIR file
   • -o: output file path of the generated IL file

   Instruction Example:

   ILGenerator.exe -i $MIR_File_Path$ -O $Output_IL_File_Path$

3. ILOptimizer.exe

   Optimize the obtained IL file

   Useful Flags:

   • -help: help information
   • -i: input file path of IL file
   • -m: input file path of MIR file
   • -o: output file path of optimized IL file
   • -lp: loop reshaping (default is false)

   For enabling optimization of loop reshaping, you should use the following instructions.

   ILOptimizer.exe -i $IL_File_Path$ -m $MIR_File_Path$ -o $Output_File_Path$ -lp true

4. ILTranslator.exe

   Translate the IL file into efficient embedded C code for deployment

   Useful Flags:

   • -help: help information
   • -i: input file path of IL file or optimized IL file
   • -o: output directory of the generated code
   • -l: the language of the generated code
   • -c: generate comment (default is true)

   Instruction Example:

   ILTranslator.exe -i $IL_File_Path$ -o $Output_Code_Path$ -c -l c

In general, the building workflow is:

Simulink model -> ParserForSimulink.exe -> ILGenerator.exe -> ILOptimizer.exe -> ILTranslator.exe