Generate Single-Precision MATLAB Code

This example shows how to generate single-precision MATLAB^® code from double-precision MATLAB code. This example shows the single-precision conversion workflow that you use when you want to see single-precision MATLAB code or use verification options. Optionally, you can also generate single-precision C/C++ code.

Prerequisites

To complete this example, install the following products:

• MATLAB

• MATLAB Coder™

• Fixed-Point Designer™

• C compiler

  See Supported Compilers.

  You can use mex -setup to change the default compiler. See Change Default Compiler.

Create a Folder and Copy Relevant Files

1. In a local, writable folder, create a function ex_2ndOrder_filter.m.

   function y = ex_2ndOrder_filter(x) %#codegen
     persistent z
     if isempty(z)
         z = zeros(2,1);
     end
     % [b,a] = butter(2, 0.25)
     b = [0.0976310729378175,  0.195262145875635,  0.0976310729378175];
     a = [1, -0.942809041582063,  0.3333333333333333];
   
    
     y = zeros(size(x));
     for i = 1:length(x)
         y(i) = b(1)*x(i) + z(1);
         z(1) = b(2)*x(i) + z(2) - a(2) * y(i);
         z(2) = b(3)*x(i)        - a(3) * y(i);
     end
   end
   

2. Create a test file, ex_2ndOrder_filter_test.m, to exercise the ex_2ndOrder_filter algorithm.

   It is a best practice to create a separate test script for preprocessing and postprocessing such as:

   • Setting up input values.

   • Calling the function under test.

   • Outputting the test results.

   To cover the full intended operating range of the system, the test script runs the ex_2ndOrder_filter function with three input signals: chirp, step, and impulse. The script then plots the outputs.

   % ex_2ndOrder_filter_test
   %
   % Define representative inputs
   N = 256;                   % Number of points
   t = linspace(0,1,N);       % Time vector from 0 to 1 second
   f1 = N/2;                  % Target frequency of chirp set to Nyquist
   x_chirp = sin(pi*f1*t.^2); % Linear chirp from 0 to Fs/2 Hz in 1 second
   x_step = ones(1,N);        % Step
   x_impulse = zeros(1,N);    % Impulse
   x_impulse(1) = 1;
   
   % Run the function under test
   x = [x_chirp;x_step;x_impulse];
   y = zeros(size(x));
   for i = 1:size(x,1)
     y(i,:) = ex_2ndOrder_filter(x(i,:));
   end
   
   % Plot the results
   titles = {'Chirp','Step','Impulse'}
   clf
   for i = 1:size(x,1)
     subplot(size(x,1),1,i)
     plot(t,x(i,:),t,y(i,:))
     title(titles{i})
     legend('Input','Output')
   end
   xlabel('Time (s)')
   figure(gcf)
   
   disp('Test complete.')

Type	Name	Description
Function code	ex_2ndOrder_filter.m	Entry-point MATLAB function
Test file	ex_2ndOrder_filter_test.m	MATLAB script that tests ex_2ndOrder_filter.m

Set Up the Single-Precision Configuration Object

Create a single-precision configuration object. Specify the test file name. Verify the single-precision code using the test file. Plot the error between the double-precision code and single-precision code. Use the default values for the other properties.

scfg = coder.config('single');
scfg.TestBenchName = 'ex_2ndOrder_filter_test';
scfg.TestNumerics = true;
scfg.LogIOForComparisonPlotting = true;

Generate Single-Precision MATLAB Code

To convert the double-precision MATLAB function, ex_2ndOrder_filter, to single-precision MATLAB code, use the codegen function with the -double2single option.

codegen -double2single scfg ex_2ndOrder_filter

codegen analyzes the double-precision code. The conversion process infers types by running the test file because you did not specify the input types for the ex_2ndOrder_filter function. The conversion process selects single-precision types for the double-precision variables. It selects int32 for index variables. When the conversion is complete, codegen generates a type proposal report.

View the Type Proposal Report

To see the types that the conversion process selected for the variables, open the type proposal report for the ex_2ndOrder_filter function. Click the link ex_2ndOrder_filter_report.html.

The report opens in a web browser. The conversion process converted:

• Double-precision variables to single.

• The index i to int32. The conversion process casts index and dimension variables to int32.

View Generated Single-Precision MATLAB Code

To view the report for the generation of the single-precision MATLAB code, in the Command Window:

1. Scroll to the Generate Single-Precision Code step. Click the View report link.

2. In the MATLAB Source pane, click ex_2ndOrder_filter_single.

The code generation report displays the single-precision MATLAB code for ex_2ndOrder_filter.

View Potential Data Type Issues

When you generate single-precision code, codegen enables highlighting of potential data type issues in code generation reports. If codegen cannot remove a double-precision operation, the report highlights the MATLAB expression that results in the operation. Click the Code Insights tab. The absence of potential data type issues indicates that no double-precision operations remain.

Compare the Double-Precision and Single-Precision Variables

You can see the comparison plots for the input x and output y because you selected to log inputs and outputs for comparison plots.

Optionally Generate Single-Precision C Code

If you also want to generate single-precision C code, create a code configuration object for C code generation. Use this configuration object with the -config option of the codegen function. For example:

1. Create a code configuration object for generation of a C static library.

   cfg = coder.config('lib');
   

2. Generate the C code. Enable generation of the code generation report.

   codegen -double2single scfg -config cfg ex_2ndOrder_filter -report

3. To view the code generation report for the C code generation, click the View Report link.

   In the Generated Code pane, click ex_2ndOrder_filter.c.

   • Double-precision variables have type float in the C code.

   • The index i is an integer.

   When you generate single-precision code, codegen enables highlighting of potential data type issues in the code generation report. If codegen cannot remove a double-precision operation, the report highlights the MATLAB expression that results in the operation.

   Click the Code Insights tab. Then, expand Potential data type issues. The absence of double-precision operations indicates that no double-precision operations remain.