Execution-Time Profiling for NVIDIA Jetson Platforms in Simulink

Since R2025a

You can create execution-time profiles for Simulink^® generated code running on NVIDIA^® Jetson™ platforms to measure execution times on the Jetson hardware. To create an execution-time profile, you specify configuration parameters for the profile, then generate the profile using processor-in-the-loop (PIL) or external mode simulations. You can analyze the profiling data from the PIL or external mode simulation using the Code Profile Analyzer.

Open a Simulink Model for Profiling

This example uses the Simulink model TopModelPil.slx from the example Verify Generated Code on NVIDIA Targets Using PIL in Simulink. The model accepts an input signal from the MATLAB^® workspace and counts the number of peaks.

To open the model, execute this command in the MATLAB Command Window:

openExample("nvidia/PILExecutionOnNVIDIATargetsUsingSimulinkExample",...
"supportingFile","TopModelPil.slx")

To generate the input data, use the model stop time and step size to create a signal with the correct dimensions. The model uses a stop time of 10 seconds and a step size of 0.2 seconds. Create a struct named Input with two fields:

• time, which contains the time of each step in the model

• signals, which contains the value of the input at each step

StopTime = 10;
StepSize = 0.2;
Input.time = (0:StepSize:StopTime)';
Input.signals.values = sin(Input.time);
Input.signals.dimensions = 1;

To load inputs from the MATLAB workspace, enable the LoadExternalInput parameter. Specify the ExternalInput parameter to load the Input variable from the MATLAB workspace.

set_param("TopModelPil","LoadExternalInput","on");
set_param("TopModelPil","ExternalInput","Input");

Connect to NVIDIA Jetson

To communicate with the NVIDIA Jetson hardware, create a live hardware connection object using the jetson function. When connecting to the target board for the first time, you must input the host name or IP address, user name, and password of the target board.

hwobj = jetson("jetson-deviceaddress","username","password");

On subsequent connections, the Jetson hardware connection object reuses the settings from the most recent successful connection to an NVIDIA board.

hwobj = jetson;

Configure the Simulink Model for Execution-Time Profiling

Before you create an execution-time profile, you first enable profiling and set the configuration for the profile in the Configuration Parameters dialog box.

1. In the model, in the toolstrip, open the Apps tab. Under the Setup to Run on Hardware section, click Run on Hardware Board.

2. In the toolstrip, in the Hardware tab, click Hardware Settings. In the Configuration Parameters dialog box, in the left pane, open the Code Generation > Verification parameters.

3. In the Code execution time profiling section, select Measure task execution time.

4. To profile the reference model and subsystem functions in the generated code, set Measure function execution times to Coarse (referenced models and subsystems only).

Alternatively, use the set_param function to configure the model programmatically.

set_param("TopModelPil","CodeExecutionProfiling","on");
set_param("TopModelPil","CodeProfilingInstrumentation","coarse");

Generate the Execution-Time Profile

To generate the execution-time profile, simulate the model using PIL or external mode simulation.

Generate Execution-Time Profile Using PIL

You can generate an execution-time profile by generating code for the model and running the code on NVIDIA Jetson using PIL. Simulink measures task execution times during the PIL simulation and creates an execution-time profile.

Set the SimulationMode parameter to processor-in-the-loop (pil), and start PIL execution by using the sim command.

set_param("TopModelPil","SimulationMode","processor-in-the-loop (pil)");
simOutput = sim("TopModelPil",StopTime);

The simulation output simOutput contains the execution profile from the PIL simulation. By default, the execution profile is named executionProfile. Access the execution profile using dot notation.

simOutput.executionProfile

Generate Execution-Time Profile Using External Mode

Alternatively, you can generate an execution-time profile for the model using external mode simulation. To run the model in external mode, disable PIL block creation. Additionally, configure the model to return the simulation output as one or more workspace variables.

set_param("TopModelPil","CreateSILPILBlock","None");
set_param("TopModelPil","ReturnWorkspaceOutputs","off");

In the model, in the Hardware tab, click Monitor & Tune to run the model in external mode.

The simulation returns the executionProfile variable in the MATLAB workspace.

View Profiling Data for the Model

To examine the profiling data from the model, use the Code Profile Analyzer. Open the execution-time profile in the analyzer by using the coder.profile.show function.

coder.profile.show(simOutput.executionProfile);

The analyzer contains the profiling data for the generated code, including:

• The function execution times for:

  • The model initialize function, TopModelPil_initialize

  • The model step function, TopModelPil_step

  • The model terminate function, TopModelPil_terminate

• The CPU utilization

The Code Profile Analyzer displays the time in ticks by default. To change the unit, in the toolstrip, use the Time unit parameter.