Target Language Compiler

TLC Profiler

This section discusses how to use the TLC profiler to analyze your TLC code to improve code generation performance.

The TLC profiler collects timing statistics for TLC code. It collects execution time for functions, scripts, macros, and built-in functions. These results become the basis of HTML reports that are identical in format to MATLAB profiler reports. By analyzing the report, you can identify bottlenecks in your code that make code generation take longer.

Using the Profiler

To access the profiler, select Profile TLC from the TLC debugging category of the Real-Time Workshop pane of the Simulation Parameters dialog box. Apply your changes and click the Build button.

At the end of the build process, the HTML summary and related files are placed in the Real-Time Workshop project directory and the report will be opened in your MATLAB selected Web browser.

Analyzing the Report

The created report is fairly self explanatory. Some points to note are

• Functions are sorted in descending order of their execution time.
• Self-time is the time spent in the function alone and does not include the time spent in calling the function.
• Functions are hyperlinks that take you to the details related to that specific function.

A situation where the profiler report may be helpful is when you have inlined S-functions in your model. You can use the profiler to compare time spent in specific user-written or Lib functions, and then modify your TLC code accordingly.

This is a portion of the profiler report for the simple.log model in matlabroot/toolbox/rtw/rtwdemos/tlctutorial/tlcdebug.

The report shows the time spent in the function FcnEliminateUnnecessaryParams in gain.tlc and other functions, both built-in and library, called during various stages of code generation.

Nonexecutable Directives

TLC considers the following directives to be nonexecutable lines. Therefore, these directives are not counted in TLC Profiler reports:

• %filescope
• %else
• %endif
• %endforeach
• %endfor
• %endroll
• %endwith
• %body
• %endbody
• %endfunction
• %endswitch
• %default
• any type of comment (%% or /% stuff %/)

Improving Performance

Analyzing the profiler results also gives you an overview of which functions are used more often or are more expensive. Then, you can either improve those functions that were written by you, or try alternative methods to improve code generation speed. Two points to consider are

• Reduce usage of EXISTS. Performing an EXISTS on a field is more costly than comparing the field to a value. When possible, create an inert default value for a field. Then, instead of doing an EXISTS on the entity, compare it against the default value.
• Reduce the use of one line functions when they are not really needed. One line functions might be a bottleneck for code generation speed. When readability is not greatly impacted, consider expanding out the function.

TLC Coverage

Inlining S-Functions