User's Guide: Code Generation From MATLAB

Code Generation from MATLAB®
User’s Guide
R2011b
How to Contact MathWorks
www.mathworks.com Web
comp.soft-sys.matlab Newsgroup
www.mathworks.com/contact_TS.html Technical Support
suggest@mathworks.com Product enhancement suggestions

bugs@mathworks.com Bug reports
doc@mathworks.com Documentation error reports
service@mathworks.com Order status, license renewals, passcodes
info@mathworks.com Sales, pricing, and general information
508-647-7000 (Phone)
508-647-7001 (Fax)
The MathWorks, Inc.

3 Apple Hill Drive
Natick, MA 01760-2098
For contact information about worldwide offices, see the MathWorks Web site.
Code Generation from MATLAB User’s Guide
© COPYRIGHT 2007–2011 by The MathWorks, Inc.
The software described in this document is furnished under a license agreement. The software may be used
or copied only under the terms of the license agreement. No part of this manual may be photocopied or
reproduced in any form without prior written consent from The MathWorks, Inc.
FEDERAL ACQUISITION: This provision applies to all acquisitions of the Program and Documentation
by, for, or through the federal government of the United States. By accepting delivery of the Program
or Documentation, the government hereby agrees that this software or documentation qualifies as
commercial computer software or commercial computer software documentation as such terms are used
or defined in FAR 12.212, DFARS Part 227.72, and DFARS 252.227-7014. Accordingly, the terms and
conditions of this Agreement and only those rights specified in this Agreement, shall pertain to and govern
the use, modification, reproduction, release, performance, display, and disclosure of the Program and
Documentation by the federal government (or other entity acquiring for or through the federal government)
and shall supersede any conflicting contractual terms or conditions. If this License fails to meet the
government’s needs or is inconsistent in any respect with federal procurement law, the government agrees
to return the Program and Documentation, unused, to The MathWorks, Inc.
Trademarks
MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See
www.mathworks.com/trademarks for a list of additional trademarks. Other product or brand
names may be trademarks or registered trademarks of their respective holders.
Patents
MathWorks products are protected by one or more U.S. patents. Please see
www.mathworks.com/patents for more information.
Revision History
March 2007 Online only New for Release 2007a
September 2007 Online only Revised for Release 2007b
March 2008 Online only Revised for Release 2008a
October 2008 Online only Revised for Release 2008b
April 2011 Online only Revised for Release 2011a
Contents
About Code Generation from MATLAB

Algorithms
1
Direct Translation of MATLAB Algorithms to C/C++
Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-2
Prerequisites for Code Generation from MATLAB . . . . 1-3
Preparing MATLAB Code for C/C++ and MEX Code

Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-4
Expected Differences in Behavior After Compiling Your

MATLAB Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-5
Why Are There Differences? . . . . . . . . . . . . . . . . . . . . . . . . . 1-5
Character Size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-5
Order of Evaluation in Expressions . . . . . . . . . . . . . . . . . . . 1-5
Termination Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-6
Size of Variable-Size N-D Arrays . . . . . . . . . . . . . . . . . . . . . 1-6
Floating-Point Numerical Results . . . . . . . . . . . . . . . . . . . . 1-7
NaN and Infinity Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . 1-7
Code Generation Target . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-7
MATLAB Language Features Supported for Code

Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-9
MATLAB Language Features Not Supported for Code

Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-11
Related Products That Support Code Generation from

MATLAB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1-12
v
Functions Supported for Code Generation
2
About Code Generation for Supported Functions . . . . . 2-2
Functions Supported for Code Generation —

Alphabetical List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-3
Functions Supported for Code Generation —

Categorical List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-59
Aerospace Toolbox Functions . . . . . . . . . . . . . . . . . . . . . . . . 2-60
Arithmetic Operator Functions . . . . . . . . . . . . . . . . . . . . . . 2-60
Bit-Wise Operation Functions . . . . . . . . . . . . . . . . . . . . . . . 2-61
Casting Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-61
Communications System Toolbox Functions . . . . . . . . . . . . 2-62
Complex Number Functions . . . . . . . . . . . . . . . . . . . . . . . . . 2-62
Computer Vision System Toolbox Functions . . . . . . . . . . . . 2-63
Data Type Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-64
Derivative and Integral Functions . . . . . . . . . . . . . . . . . . . . 2-64
Discrete Math Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-64
Error Handling Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 2-65
Exponential Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-65
Filtering and Convolution Functions . . . . . . . . . . . . . . . . . . 2-66
Fixed-Point Toolbox Functions . . . . . . . . . . . . . . . . . . . . . . . 2-66
Histogram Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-75
Image Processing Toolbox Functions . . . . . . . . . . . . . . . . . . 2-75
Input and Output Functions . . . . . . . . . . . . . . . . . . . . . . . . 2-76
Interpolation and Computational Geometry . . . . . . . . . . . . 2-76
Linear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-77
Logical Operator Functions . . . . . . . . . . . . . . . . . . . . . . . . . 2-77
MATLAB Compiler Functions . . . . . . . . . . . . . . . . . . . . . . . 2-78
Matrix and Array Functions . . . . . . . . . . . . . . . . . . . . . . . . . 2-78
Nonlinear Numerical Methods . . . . . . . . . . . . . . . . . . . . . . . 2-82
Polynomial Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-82
Relational Operator Functions . . . . . . . . . . . . . . . . . . . . . . . 2-82
Rounding and Remainder Functions . . . . . . . . . . . . . . . . . . 2-83
Set Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-83
Signal Processing Functions in MATLAB . . . . . . . . . . . . . . 2-84
Signal Processing Toolbox Functions . . . . . . . . . . . . . . . . . . 2-84
Special Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-89
Specialized Math . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-89
Statistical Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-90
vi Contents
String Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-90
Structure Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-91
Trigonometric Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-91
System Objects Supported for Code Generation

3
About Code Generation for System Objects . . . . . . . . . . 3-2
Computer Vision System Toolbox System Objects .... 3-3
Communications System Toolbox System Objects .... 3-7
DSP System Toolbox System Objects . . . . . . . . . . . . . . . . 3-13
Defining MATLAB Variables for C/C++ Code

Generation
4
Why Define Variables Differently for Code
Generation? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-2
Best Practices for Defining Variables for C/C++ Code

Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-3
Define Variables By Assignment Before Using Them . . . . 4-3
Use Caution When Reassigning Variables . . . . . . . . . . . . . 4-6
Use Type Cast Operators in Variable Definitions . . . . . . . . 4-6
Define Matrices Before Assigning Indexed Variables . . . . . 4-6
When You Can Reassign Variable Properties for C/C++

Code Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-7
Eliminating Redundant Copies of Variables in

Generated Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-8
vii
When Redundant Copies Occur . . . . . . . . . . . . . . . . . . . . . . 4-8
How to Eliminate Redundant Copies by Defining
Uninitialized Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-8
Defining Uninitialized Variables . . . . . . . . . . . . . . . . . . . . . 4-9
Defining and Initializing Persistent Variables . . . . . . . . 4-10
Reusing the Same Variable with Different

Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-11
When You Can Reuse the Same Variable with Different
Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4-11
When You Cannot Reuse Variables . . . . . . . . . . . . . . . . . . . 4-12
Limitations of Variable Reuse . . . . . . . . . . . . . . . . . . . . . . . 4-14
Supported Variable Types . . . . . . . . . . . . . . . . . . . . . . . . . . 4-16
Defining Data for Code Generation

5
How Working with Data is Different for Code
Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-2
Code Generation for Complex Data . . . . . . . . . . . . . . . . . 5-4

Restrictions When Defining Complex Variables . . . . . . . . . 5-4
Expressions Containing Complex Operands Yield Complex
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5-5
Code Generation for Characters . . . . . . . . . . . . . . . . . . . . 5-6
Code Generation for MATLAB Structures

6
How Working with Structures Is Different for Code
Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-2
viii Contents
Structure Operations Allowed for Code Generation . . . 6-3
Defining Scalar Structures for Code Generation . . . . . . 6-4

Restrictions When Using struct . . . . . . . . . . . . . . . . . . . . . 6-4
Restrictions When Defining Scalar Structures by
Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-4
Adding Fields in Consistent Order on Each Control Flow
Path . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6-4
Restriction on Adding New Fields After First Use . . . . . . . 6-5
Defining Arrays of Structures for Code Generation . . . 6-7

Ensuring Consistency of Fields . . . . . . . . . . . . . . . . . . . . . . 6-7
Using repmat to Define an Array of Structures with
Consistent Field Properties . . . . . . . . . . . . . . . . . . . . . . . 6-7
Defining an Array of Structures Using Concatenation . . . . 6-8
Making Structures Persistent . . . . . . . . . . . . . . . . . . . . . . . 6-9
Indexing Substructures and Fields . . . . . . . . . . . . . . . . . . 6-10
Assigning Values to Structures and Fields . . . . . . . . . . . 6-12
Passing Large Structures as Input Parameters . . . . . . . 6-13
Code Generation for Enumerated Data

7
How Working with Enumerated Data Is Different for
Code Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-2
See Also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-2
Enumerated Types Supported for Code Generation . . . 7-3

Enumerated Type Based on int32 . . . . . . . . . . . . . . . . . . . . 7-3
Enumerated Type Based on Simulink.IntEnumType . . . . . 7-4
When to Use Enumerated Data for Code Generation . . 7-6
ix
Workflows for Using Enumerated Data for Code
Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-7
Workflow for Generating Code for Enumerated Data from
MATLAB Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-7
Workflow for Generating Code for Enumerated Data from
MATLAB Function Blocks . . . . . . . . . . . . . . . . . . . . . . . . 7-8
How to Define Enumerated Data for Code

Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-9
Naming Enumerated Types for Code Generation . . . . . . . . 7-10
How to Instantiate Enumerated Types for Code

Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-11
How to Generate Code for Enumerated Data . . . . . . . . . 7-12

See Also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-12
Defining and Using Enumerated Types for Code

Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-13
About the Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-13
Class Definition: sysMode . . . . . . . . . . . . . . . . . . . . . . . . . . 7-13
Class Definition: LEDcolor . . . . . . . . . . . . . . . . . . . . . . . . . . 7-14
Function: displayState . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-14
Operations on Enumerated Data Allowed for Code

Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-15
Assignment Operator, = . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-15
Relational Operators, < > <= >= == ~= . . . . . . . . . . . . . . . . 7-15
Cast Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-16
Indexing Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-16
Control Flow Statements: if, switch, while . . . . . . . . . . . . . 7-17
Using Enumerated Data in Control Flow Statements . . 7-18

Using the if Statement on Enumerated Data Types . . . . . 7-18
Using the switch Statement on Enumerated Data
Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-19
Using the while Statement on Enumerated Data Types . . 7-22
Restrictions on Use of Enumerated Data in

for-Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-24
x Contents
Toolbox Functions That Support Enumerated Types for
Code Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-25
Code Generation for Variable-Size Data

8
What Is Variable-Size Data? . . . . . . . . . . . . . . . . . . . . . . . . 8-2
How Working with Variable-Size Data Is Different for

Code Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-3
See Also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-3
Bounded Versus Unbounded Variable-Size Data . . . . . . 8-4
When to Use Dynamic Allocation for Variable-Size

Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-5
See Also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-5
How to Generate Code for MATLAB Functions with

Variable-Size Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-6
Tutorial: Generating MEX Code for a MATLAB

Function That Expands a Vector in a Loop . . . . . . . . 8-9
About the MATLAB Function emldemo_uniquetol . . . . . . . 8-9
Step 1: Add Compilation Directive for Code Generation . . 8-10
Step 2: Address Issues Detected by the Code Analyzer . . . 8-10
Step 3: Generate MEX Code . . . . . . . . . . . . . . . . . . . . . . . . . 8-10
Step 4: Fix the Size Mismatch Error . . . . . . . . . . . . . . . . . . 8-12
Step 5: Fix the Upper Bounds Error . . . . . . . . . . . . . . . . . . 8-14
Step 6: Generate C/C++ Code . . . . . . . . . . . . . . . . . . . . . . . . 8-16
Variable-Size Data in Code Generation Reports . . . . . . 8-18

What Reports Tell You About Size . . . . . . . . . . . . . . . . . . . . 8-18
How Size Appears in Code Generation Reports . . . . . . . . . 8-19
How to Generate a Code Generation Report . . . . . . . . . . . . 8-19
Defining Variable-Size Data for Code Generation . . . . . 8-20
xi
When to Define Variable-Size Data Explicitly . . . . . . . . . . 8-20
Using a Matrix Constructor with Nonconstant
Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-21
Inferring Variable Size from Multiple Assignments . . . . . . 8-21
Defining Variable-Size Data Explicitly Using
coder.varsize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-23
Specifying Upper Bounds for Variable-Size Data . . . . . 8-27

When to Specify Upper Bounds for Variable-Size Data . . . 8-27
Specifying Upper Bounds on the Command Line for
Variable-Size Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-27
Specifying Unknown Upper Bounds for Variable-Size
Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-27
Specifying Upper Bounds for Local Variable-Size Data . . . 8-28
C Code Interface for Unbounded Arrays and Structure

Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-30
emxArray: Representation of Data with Unknown Upper
Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-30
Utility Functions for Creating emxArray Data
Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-31
Troubleshooting Issues with Variable-Size Data . . . . . . 8-33

Diagnosing and Fixing Size Mismatch Errors . . . . . . . . . . . 8-33
Diagnosing and Fixing Errors in Detecting Upper
Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-35
Limitations with Variable-Size Support for Code

Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-37
Limitation on Scalar Expansion . . . . . . . . . . . . . . . . . . . . . . 8-37
Incompatibility with MATLAB in Determining Size of
Variable-Size N-D Arrays . . . . . . . . . . . . . . . . . . . . . . . . . 8-38
Limitation on Vector-Vector Indexing . . . . . . . . . . . . . . . . . 8-39
Limitations on Matrix Indexing Operations for Code
Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-40
Dynamic Memory Allocation Not Supported for MATLAB
Function Blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-41
Limitation on Mixing Stack and Heap Allocation . . . . . . . . 8-41
Restrictions on Variable Sizing in Toolbox Functions

Supported for Code Generation . . . . . . . . . . . . . . . . . . . 8-42
Common Restrictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8-42
xii Contents
Toolbox Functions with Variable Sizing Restrictions . . . . . 8-43
Code Generation for Function Handles

9
How Working with Function Handles is Different for
Code Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-2
Example: Defining and Passing Function Handles for

Code Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-3
Limitations with Function Handles for Code

Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-6
Defining Functions for Code Generation

10
Specifying Variable Numbers of Arguments . . . . . . . . . . 10-2
Supported Index Expressions . . . . . . . . . . . . . . . . . . . . . . . 10-3
Using varargin and varargout in for-Loops . . . . . . . . . . 10-4

When to Force Loop Unrolling . . . . . . . . . . . . . . . . . . . . . . . 10-4
Example: Using Variable Numbers of Arguments in a
for-Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-5
Implementing Wrapper Functions with varargin and

varargout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-7
Example: Passing Variable Numbers of Arguments from
One Function to Another . . . . . . . . . . . . . . . . . . . . . . . . . 10-7
Passing Property/Value Pairs with varargin . . . . . . . . . 10-8
xiii
Rules for Using Variable Length Argument Lists for
Code Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10-10
Calling Functions for Code Generation

11
How MATLAB Resolves Function Calls in Generated
Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-2
Key Points About Resolving Function Calls . . . . . . . . . . . . 11-4
Compile Path Search Order . . . . . . . . . . . . . . . . . . . . . . . . . 11-4
When to Use the Code Generation Path . . . . . . . . . . . . . . . 11-5
How MATLAB Resolves File Types on the Path for Code

Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-6
Adding the Compilation Directive %#codegen . . . . . . . . 11-8
Calling Subfunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-9
Calling Supported Toolbox Functions . . . . . . . . . . . . . . . 11-10
Calling MATLAB Functions . . . . . . . . . . . . . . . . . . . . . . . . . 11-11

Declaring MATLAB Functions as Extrinsic Functions . . . 11-11
Calling MATLAB Functions Using feval . . . . . . . . . . . . . . . 11-15
How MATLAB Resolves Extrinsic Functions During
Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-15
Working with mxArrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-16
Restrictions on Extrinsic Functions for Code Generation . . 11-17
Limit on Function Arguments . . . . . . . . . . . . . . . . . . . . . . . 11-18
Generating Efficient and Reusable Code

12
Generating Efficient Code . . . . . . . . . . . . . . . . . . . . . . . . . . 12-2
xiv Contents
Unrolling for-Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-2
Inlining Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-2
Eliminating Redundant Copies of Function Inputs . . . . . . 12-2
Generating Reusable Code . . . . . . . . . . . . . . . . . . . . . . . . . 12-4
Examples
A
Data Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A-2
Code Generation for Structures . . . . . . . . . . . . . . . . . . . . . A-2
Code Generation for Enumerated Data . . . . . . . . . . . . . . A-2
Code Generation for Variable-Size Data . . . . . . . . . . . . . A-2
Code Generation for Function Handles . . . . . . . . . . . . . . A-3
Using Variable-Length Argument Lists . . . . . . . . . . . . . . A-3
Optimizing Generated Code . . . . . . . . . . . . . . . . . . . . . . . . A-3
Index
xv
xvi Contents
1
About Code Generation

from MATLAB Algorithms
• “Direct Translation of MATLAB Algorithms to C/C++ Code” on page 1-2

• “Prerequisites for Code Generation from MATLAB” on page 1-3
• “Preparing MATLAB Code for C/C++ and MEX Code Generation” on page
1-4
• “Expected Differences in Behavior After Compiling Your MATLAB Code”
on page 1-5
• “MATLAB Language Features Supported for Code Generation” on page 1-9
• “MATLAB Language Features Not Supported for Code Generation” on
page 1-11
• “Related Products That Support Code Generation from MATLAB” on page
1-12
1 About Code Generation from MATLAB® Algorithms
Direct Translation of MATLAB Algorithms to C/C++ Code

You can automatically generate MEX functions and standalone C/C++ code
from MATLAB® algorithms. With this capability, you can design, implement,
and test software in the MATLAB environment, then automatically translate
the algorithms to readable, efficient, and compact C/C++ code for deployment
to desktop and embedded systems.
The generated code contains optimizations tailored to meet the requirements

of desktop and embedded applications for speed, memory use, and data type
management.
To verify the generated code in the MATLAB environment, you can generate
MEX functions to compare with your original algorithm to determine whether
they are functionally equivalent.
In certain applications, you can also generate MEX functions to accelerate

MATLAB algorithms.
1-2
Prerequisites for Code Generation from MATLAB®
Prerequisites for Code Generation from MATLAB

To generate C/C++ or MEX code from MATLAB algorithms, you must install
the following software:
• MATLAB® Coder™ product

• C/C++ compiler
For more information, see:
• “Why Test MEX Functions in MATLAB?” in the MATLAB Coder

documentation.
• “Generating C/C++ Code from MATLAB Code” in the MATLAB Coder
documentation.
• “Accelerating MATLAB Algorithms” in the MATLAB Coder documentation.
1-3
Preparing MATLAB Code for C/C++ and MEX Code

Generation
By default, the MATLAB language uses dynamic typing. Properties of
dynamically typed variables can change at runtime, allowing a single variable
to hold a value of any class, size, or complexity. However, to generate efficient
code for statically typed languages such as C and C++, the properties of all
MATLAB variables must be determined at compile time. Therefore, to prepare
MATLAB code for C, C++, and MEX code generation, you must specify the
class, size, and complexity of inputs to the primary function (also known as the
top-level or entry-point function). By determining these properties at compile
time, the code generation process translates your MATLAB algorithms into
code that is efficient and tailored to your specific application, rather than
producing generic code that handles every possible set of MATLAB inputs.
1-4
Expected Differences in Behavior After Compiling Your MATLAB® Code
Expected Differences in Behavior After Compiling Your

MATLAB Code
In this section...
“Why Are There Differences?” on page 1-5
“Character Size” on page 1-5
“Order of Evaluation in Expressions” on page 1-5
“Termination Behavior” on page 1-6
“Size of Variable-Size N-D Arrays” on page 1-6
“Floating-Point Numerical Results” on page 1-7
“NaN and Infinity Patterns” on page 1-7
“Code Generation Target” on page 1-7
Why Are There Differences?

To convert MATLAB code to C/C++ code that works efficiently, the code
generation process introduces optimizations that intentionally cause the
generated code to behave differently — and sometimes produce different
results — from the original source code. This section describes these
differences.
Character Size
MATLAB supports 16-bit characters, but the generated code represents
characters in 8 bits, the standard size for most embedded languages like C.
See “Code Generation for Characters” on page 5-6.
Order of Evaluation in Expressions

Generated code does not enforce order of evaluation in expressions. For most
expressions, order of evaluation is not significant. However, for expressions
with side effects, the generated code may produce the side effects in different
order from the original MATLAB code. Expressions that produce side effects
include those that:
1-5
• Modify persistent or global variables

• Display data to the screen
• Write data to files
In addition, the generated code does not enforce order of evaluation of logical
operators that do not short circuit.
For more predictable results, it is good coding practice to split expressions

that depend on the order of evaluation into multiple statements. For example,
rewrite:
A = f1() + f2();
as
A = f1();
A = A + f2();
so that the generated code calls f1 before f2.
Termination Behavior
Generated code does not match the termination behavior of MATLAB source
code. For example, optimizations remove infinite loops from generated code if
they have no side effects. As a result, the generated code may terminate even
though the corresponding MATLAB code does not.
Size of Variable-Size N-D Arrays

For variable-size N-D arrays, the size function might return a different
result in generated code than in MATLAB source code. The size function
sometimes returns trailing ones (singleton dimensions) in generated code, but
always drops trailing ones in MATLAB. For example, for an N-D array X with
dimensions [4 2 1 1], size(X) might return [4 2 1 1] in generated code,
but always returns [4 2] in MATLAB. See “Incompatibility with MATLAB in
Determining Size of Variable-Size N-D Arrays” on page 8-38.
1-6
Expected Differences in Behavior After Compiling Your MATLAB® Code
Floating-Point Numerical Results

The generated code might not produce the same floating-point numerical
results as MATLAB in the following situations:
When computer hardware uses extended precision registers
Results vary depending on how the C/C++ compiler allocates extended

precision floating-point registers. Computation results might not match
MATLAB calculations because of different compiler optimization settings or
different code surrounding the floating-point calculations.
For certain advanced library functions
The generated code might use different algorithms to implement certain

advanced library functions, such as fft, svd, eig, mldivide, and mrdivide.
For example, the generated code uses a simpler algorithm to implement

svd to accommodate a smaller footprint. Results might also vary according
to matrix properties. For example, MATLAB might detect symmetric or
Hermitian matrices at run time and switch to specialized algorithms that
perform computations faster than implementations in the generated code.
For implementation of BLAS library functions
For implementations of BLAS library functions. Generated C/C++ code uses

reference implementations of BLAS functions, which may produce different
results from platform-specific BLAS implementations in MATLAB.
NaN and Infinity Patterns

The generated code might not produce exactly the same pattern of NaN and inf
values as MATLAB code when these values are mathematically meaningless.
For example, if MATLAB output contains a NaN, output from the generated
code should also contain a NaN, but not necessarily in the same place.
Code Generation Target

The coder.target function returns different values in MATLAB than in the
generated code. The intent is to help you determine whether your function
1-7
is executing in MATLAB or has been compiled for a simulation or code

generation target. See coder.target.
1-8
MATLAB® Language Features Supported for Code Generation
MATLAB Language Features Supported for Code

Generation
MATLAB supports the following language features in generated code:
• N-dimensional arrays
• Matrix operations, including deletion of rows and columns
• Variable-sized data (see “How Working with Variable-Size Data Is Different
for Code Generation” on page 8-3)
• Subscripting (see “Limitations on Matrix Indexing Operations for Code
Generation” on page 8-40)
• Complex numbers (see “Code Generation for Complex Data” on page 5-4)
• Numeric classes (see “Supported Variable Types” on page 4-16)
• Double-precision, single-precision, and integer math
• Fixed-point arithmetic (see “Code Acceleration and Code Generation
from MATLAB for Fixed-Point Algorithms” in the Fixed-Point Toolbox™
documentation)
• Program control statements if, switch, for, and while
• All arithmetic, relational, and logical operators
• Subfunctions (see Chapter 11, “Calling Functions for Code Generation”)
• Persistent variables (see “Defining and Initializing Persistent Variables”
on page 4-10)
• Global variables (see “Specifying Global Variable Type and Initial Value in
a Project” in the MATLAB Coder documentation.
• Structures (see Chapter 6, “Code Generation for MATLAB Structures”)
• Characters (see “Code Generation for Characters” on page 5-6)
• Function handles (see Chapter 9, “Code Generation for Function Handles”)
• Frames (see “Working with Frame-Based Signals” in the Simulink®
documentation.
• Variable length input and output argument lists (see Chapter 8, “Code
Generation for Variable-Size Data”)
1-9
• Subset of MATLAB toolbox functions (see Chapter 2, “Functions Supported

for Code Generation”)
• Ability to call functions (see “How MATLAB Resolves Function Calls in
Generated Code” on page 11-2)
1-10
MATLAB® Language Features Not Supported for Code Generation
MATLAB Language Features Not Supported for Code

Generation
MATLAB does not support the following features in generated code:
• Anonymous functions
• Cell arrays
• Java™
• Nested functions
• Objects
• Recursion
• Sparse matrices
• try/catch statements
1-11
Related Products That Support Code Generation from

MATLAB
You can also generate code from MATLAB using other MathWorks® products
that require additional licenses:
To: Do This: Required Details

Licenses
Generate C/C++ Add MATLAB See:
code from Function blocks • Simulink
MATLAB in a and MATLAB • “Using the
• MATLAB
Simulink model Truth Table MATLAB
Coder
blocks to the Function
model. • Simulink® Block” in the
Coder™ Simulink
and/or documentation
Embedded
• “Building a
Coder™
Model with
a Stateflow
Truth
Table” in the
Stateflow®
documentation
Generate C/C++ Add MATLAB See “Using

code from a functions and • Stateflow MATLAB
Stateflow chart MATLAB Truth Functions
• Simulink
Table functions in Stateflow
to the chart. • MATLAB Charts” and
Coder “Truth Table
Functions for
• Simulink
Decision-Making
Coder
Logic” in the
Stateflow
documentation.
1-12
Related Products That Support Code Generation from MATLAB®
To: Do This: Required Details

Licenses
Accelerate Compile the Fixed-Point See “Code
fixed-point MATLAB code Toolbox Acceleration and
algorithms in with the fiaccel Code Generation
your MATLAB function. from MATLAB
code for Fixed-Point
Algorithms” in
the Fixed-Point
Toolbox
documentation.
Write and • SimEvents® See “Working
simulate with Entities” in
• Simulink
MATLAB the SimEvents
functions that • MATLAB documentation.
manipulate data Coder
associated with
• Simulink
entities
Coder
Verify that Use MATLAB • Simulink® See “About
the simulation functions Design Property
behavior of a for proving Verifier™ Proving” and
model satisfies properties and “About Test Case
• Simulink
test objectives generating tests Generation” in
• MATLAB the Simulink
Coder Design Verifier
documentation
• Simulink
Coder
1-13
1-14
2
Functions Supported for

Code Generation
• “About Code Generation for Supported Functions” on page 2-2

• “Functions Supported for Code Generation — Alphabetical List” on page 2-3
• “Functions Supported for Code Generation — Categorical List” on page 2-59
2 Functions Supported for Code Generation
About Code Generation for Supported Functions

You can generate efficient C/C++ code for a subset of MATLAB and toolbox
functions that you call from MATLAB code. In generated code, each supported
function has the same name, arguments, and functionality as its MATLAB or
toolbox counterparts. However, to generate code for these functions, you must
adhere to certain limitations when calling them from your MATLAB source
code. These limitations appear in “Functions Supported for Code Generation
— Alphabetical List” on page 2-3.
To see at a glance whether there is support for a function in a particular

MATLAB category or toolbox, see “Functions Supported for Code Generation
— Categorical List” on page 2-59.
Note For more information on code generation for fixed-point algorithms,

refer to “Code Acceleration and Code Generation from MATLAB for
Fixed-Point Algorithms” in the Fixed-Point Toolbox documentation.
2-2
Functions Supported for Code Generation — Alphabetical List
Functions Supported for Code Generation — Alphabetical

List
Function Product Remarks/Limitations
abs MATLAB —
abs Fixed-Point —
Toolbox
acos MATLAB • Generates an error during simulation and
returns NaN in generated code when the
input value x is real, but the output should
be complex. To get the complex result,
make the input value complex by passing in
complex(x).
acosd MATLAB —
acosh MATLAB • Generates an error during simulation and
complex(x).
acot MATLAB —
acotd MATLAB —
acoth MATLAB —
acsc MATLAB —
acscd MATLAB —
acsch MATLAB —
add Fixed-Point —
Toolbox
all MATLAB —
all Fixed-Point —
Toolbox
and MATLAB —
2-3

angle MATLAB —
any MATLAB —
any Fixed-Point —
Toolbox
asec MATLAB —
asecd MATLAB —
asech MATLAB —
asin MATLAB • Generates an error during simulation and
complex(x).
asind MATLAB —
asinh MATLAB —
assert MATLAB • Generates specified error messages at compile
time only if all input arguments are constants
or depend on constants. Otherwise, generates
specified error messages at run time.
atan MATLAB —
atan2 MATLAB —
atand MATLAB —
atanh MATLAB • Generates an error during simulation and
complex(x).
2-4

barthannwin Signal Processing • Window length must be a constant.
Toolbox™ Expressions or variables are allowed if their
values do not change.
Specifying constants
To specify a constant input for codegen, use
coder.Constant. For more information, see
“Specifying Constant Inputs at the Command
Line”.
• Requires DSP System Toolbox™ license to
generate code.
bartlett Signal Processing • Window length must be a constant.
Toolbox Expressions or variables are allowed if their
Line”.
• Requires DSP System Toolbox license to
generate code.
besselap Signal Processing • Filter order must be a constant. Expressions
Toolbox or variables are allowed if their values do not
change.
Line”.
generate code.
2-5

beta MATLAB —
betainc MATLAB —
betaln MATLAB —
bi2de Communications • Requires a Communications System Toolbox
System Toolbox™ license to generate code.
bin2dec MATLAB —
bitand MATLAB • Does not support floating-point inputs. The
arguments must belong to an integer class.
bitand Fixed-Point • Not supported for slope-bias scaled fi objects.
Toolbox
bitandreduce Fixed-Point —
Toolbox
bitcmp MATLAB • Does not support floating-point input for the
first argument. The first argument must
belong to an integer class.
bitcmp Fixed-Point —
Toolbox
bitconcat Fixed-Point —
Toolbox
bitget MATLAB —
bitget Fixed-Point —
Toolbox
bitmax MATLAB —
bitor MATLAB • Does not support floating-point inputs. The
bitor Fixed-Point • Not supported for slope-bias scaled fi objects.
Toolbox
bitorreduce Fixed-Point —
Toolbox
2-6

bitreplicate Fixed-Point —
Toolbox
bitrevorder Signal Processing • Computation performed at run time.
Toolbox
generate code.
bitrol Fixed-Point —
Toolbox
bitror Fixed-Point —
Toolbox
bitset MATLAB • Does not support floating-point input for the
bitset Fixed-Point —
Toolbox
bitshift MATLAB • Does not support floating-point input for the
bitshift Fixed-Point —
Toolbox
bitsliceget Fixed-Point —
Toolbox
bitsll Fixed-Point —
Toolbox
bitsra Fixed-Point —
Toolbox
bitsrl Fixed-Point —
Toolbox
bitxor MATLAB • Does not support floating-point inputs. The
bitxor Fixed-Point • Not supported for slope-bias scaled fi objects.
Toolbox
2-7

bitxorreduce Fixed-Point —
Toolbox
blackman Signal Processing • Window length must be a constant.
Line”.
generate code.
blackmanharris Signal Processing • Window length must be a constant.
Line”.
generate code.
blanks MATLAB —
2-8

bohmanwin Signal Processing • Window length must be a constant.
Line”.
generate code.
bsxfun MATLAB —
buttap Signal Processing • Filter order must be a constant. Expressions
Toolbox or variables are allowed if their values do not
change.
Line”.
generate code.
butter Signal Processing • Filter coefficients must be constants.
Line”.
2-9

generate code.
buttord Signal Processing • All inputs must be constants. Expressions or
Toolbox variables are allowed if their values do not
change.
Line”.
generate code.
cart2pol MATLAB —
cart2sph MATLAB —
cast MATLAB —
cat MATLAB —
ceil MATLAB —
ceil Fixed-Point —
Toolbox
cfirpm Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
2-10

char MATLAB —
cheb1ap Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
cheb1ord Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
cheb2ap Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
2-11

generate code.
cheb2ord Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
chebwin Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
2-12

cheby1 Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
cheby2 Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
chol MATLAB —
circshift MATLAB —
class MATLAB —
compan MATLAB —
complex MATLAB —
complex Fixed-Point —
Toolbox
cond MATLAB —
conj MATLAB —
2-13

conj Fixed-Point —
Toolbox
conv MATLAB —
conv Fixed-Point • Variable-sized inputs are only supported
Toolbox when the SumMode property of the governing
fimath is set to Specify precision or Keep
LSB.
• For variable-sized signals, you may see
different results between MATLAB and the
generated code.
- In generated code, the output for
variable-sized signals is always computed
using the SumMode property of the
governing fimath.
- In MATLAB, the output for variable-sized
signals is computed using the SumMode
property of the governing fimath when
both inputs are nonscalar. However, if
either input is a scalar, MATLAB computes
the output using the ProductMode of the
governing fimath.
conv2 MATLAB —
convergent Fixed-Point —
Toolbox
convn MATLAB —
cordicabs Fixed-Point • Variable-size signals are not supported.
Toolbox
cordicangle Fixed-Point • Variable-size signals are not supported.
Toolbox
cordicatan2 Fixed-Point • Variable-size signals are not supported.
Toolbox
2-14

cordiccart2pol Fixed-Point • Variable-size signals are not supported.
Toolbox
cordiccexp Fixed-Point • Variable-size signals are not supported.
Toolbox
cordiccos Fixed-Point • Variable-size signals are not supported.
Toolbox
cordicpol2cart Fixed-Point • Variable-size signals are not supported.
Toolbox
cordicrotate Fixed-Point • Variable-size signals are not supported.
Toolbox
cordicsin Fixed-Point • Variable-size signals are not supported.
Toolbox
cordicsincos Fixed-Point • Variable-size signals are not supported.
Toolbox
corrcoef MATLAB • Row-vector input is only supported when the
first two inputs are vectors and nonscalar.
cos MATLAB —
cosd MATLAB —
cosh MATLAB —
cot MATLAB —
cotd MATLAB —
coth MATLAB —
cov MATLAB —
cross MATLAB • If supplied, dim must be a constant.
csc MATLAB —
cscd MATLAB —
csch MATLAB —
ctranspose MATLAB —
2-15

ctranspose Fixed-Point —
Toolbox
cumprod MATLAB • Logical inputs are not supported. Cast input
to double first.
cumsum MATLAB • Logical inputs are not supported. Cast input
to double first.
cumtrapz MATLAB —
dct Signal Processing • Requires DSP System Toolbox license to
Toolbox generate code.
• Length of transform dimension must
be a power of two. If specified, the pad
or truncation value must be constant.
Expressions or variables are allowed if their
Line”.
de2bi Communications • Requires a Communications System Toolbox
System Toolbox license to generate code.
deal —
deconv MATLAB —
del2 MATLAB —
det MATLAB —
2-16

detrend MATLAB • If supplied and not empty, the input argument
bp must satisfy the following requirements:
- Be real
- Be sorted in ascending order
- Restrict elements to integers in the interval
[1, n-2], where n is the number of
elements in a column of input argument X
, or the number of elements in X when X is
a row vector
- Contain all unique values
diag MATLAB • If supplied, the argument representing the
order of the diagonal matrix must be a real
and scalar integer value.
diag Fixed-Point • If supplied, the index, k, must be a real and
Toolbox scalar integer value that is not a fi object.
diff MATLAB • If supplied, the arguments representing
the number of times to apply diff and
the dimension along which to calculate the
difference must be constants.
disp Fixed-Point —
Toolbox
divide Fixed-Point • Any non-fi input must be constant; that is,
Toolbox its value must be known at compile time so
that it can be cast to a fi object.
• Complex and imaginary divisors are not
supported.
• The syntax T.divide(a,b) is not supported.
dot MATLAB —
double MATLAB —
double Fixed-Point —
Toolbox
2-17

downsample Signal Processing —
Toolbox
dpss Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
eig MATLAB • QZ algorithm used in all cases, whereas
MATLAB might use different algorithms
for different inputs. Consequently, V might
represent a different basis of eigenvectors,
and the eigenvalues in D might not be in the
same order as in MATLAB.
• With one input, [V,D] = eig(A), the results
will be similar to those obtained using [V,D]
= eig(A,eye(size(A)),'qz') in MATLAB,
except that for code generation, the columns
of V are normalized.
• Options 'balance', 'nobalance' are not
supported for the standard eigenvalue
problem, and 'chol' is not supported for the
symmetric generalized eigenvalue problem.
• Outputs are always of complex type.
2-18

ellip Signal Processing • Inputs must be constant. Expressions or
change.
Line”.
generate code.
ellipap Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
ellipke MATLAB —
ellipord Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
2-19

generate code.
end Fixed-Point —
Toolbox
eps MATLAB —
eps Fixed-Point • Supported for scalar fixed-point signals only.
Toolbox
• Supported for scalar, vector, and matrix, fi
single and fi double signals.
eq MATLAB —
eq Fixed-Point • Not supported for fixed-point signals with
Toolbox different biases.
erf MATLAB —
erfc MATLAB —
erfcinv MATLAB —
erfcx MATLAB —
erfinv MATLAB —
error MATLAB • This is an extrinsic call.
estimate Computer Vision —
Fundamental System Toolbox™
Matrix
exp MATLAB —
expint MATLAB —
expm MATLAB —
expm1 MATLAB —
eye MATLAB • Dimensions must be real, nonnegative,
integer constants.
2-20

factor MATLAB • For double precision input, the maximum
value of A is 2^32-1.
• For single precision input, the maximum
factorial MATLAB —
false MATLAB • Dimensions must be real, nonnegative,
integer constants.
fft MATLAB • Length of input vector must be a power of 2.
fft2 MATLAB • Length of input matrix dimensions must each
be a power of 2.
fftn MATLAB • Length of input matrix dimensions must each
be a power of 2.
fftshift MATLAB —
fi Fixed-Point • Use to create a fixed-point constant or
Toolbox variable.
• The default constructor syntax without any
input arguments is not supported.
• The syntax
fi('PropertyName',PropertyValue...)
is not supported. To use property
name/property value pairs, you must first
specify the value v of the fi object as in
fi(v,'PropertyName',PropertyValue...).
• Works for all input values when complete
numerictype information of the fi object is
provided.
• Works only for constant input values (value of
input must be known at compile time) when
complete numerictype information of the fi
object is not specified.
• numerictype object information must be
available for non-fixed-point Simulink inputs.
2-21

filter MATLAB —
filter Fixed-Point • Variable-sized inputs are only supported
LSB.
filter2 MATLAB —
filtfilt Signal Processing • Filter coefficients must be constants.
Line”.
generate code.
fimath Fixed-Point • Fixed-point signals coming in to a MATLAB
Toolbox Function block from Simulink are assigned
the fimath object defined in the MATLAB
Function dialog in the Model Explorer.
• Use to create fimath objects in generated
code.
2-22

find MATLAB • Issues an error if a variable-sized input
becomes a row vector at run time.
Note This limitation does not apply when

the input is scalar or a variable-length row
vector.
• For variable-sized inputs, the shape of empty

outputs, 0-by-0, 0-by-1, or 1-by-0, depends on
the upper bounds of the size of the input. The
output might not match MATLAB when the
input array is a scalar or [] at run time. If
the input is a variable-length row vector, the
size of an empty output is 1-by-0, otherwise it
is 0-by-1.
fir1 Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
2-23

fir2 Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
fircls Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
fircls1 Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
2-24

firls Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
firpm Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
firpmord Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
2-25

firrcos Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
fix MATLAB —
fix Fixed-Point —
Toolbox
flattopwin Signal Processing • All inputs must be constants. Expressions or
change.
Line”.
generate code.
flipdim MATLAB —
fliplr MATLAB —
flipud MATLAB —
floor MATLAB —
floor Fixed-Point —
Toolbox
freqspace MATLAB —
2-26

freqz Signal Processing • freqz with no output arguments produces a
Toolbox plot only when the function call terminates
in a semicolon. See “freqz With No Output
Arguments”.
generate code.
fspecial Image Processing All inputs must be constants at compilation
Toolbox™ time. Expressions or variables are allowed if
their values do not change.
full MATLAB —
fzero MATLAB • The first argument must be a function handle.
Does not support structure, inline function, or
string inputs for the first argument.
• Supports up to three output arguments. Does
not support the fourth output argument (the
output structure).
• Only supports the TolX and FunValCheck
fields of an options input structure. Ignores
all other options in an options input
structure. You cannot use the optimset
function to create the options structure.
Create this structure directly, for example,
opt.TolX = tol;
opt.FunValCheck = 'on';
The input structure field names must match

exactly.
gamma MATLAB —
gammainc MATLAB —
gammaln MATLAB —
2-27

gaussfir Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
gausswin Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
gcd MATLAB —
ge MATLAB —
ge Fixed-Point • Not supported for fixed-point signals with
get Fixed-Point • The syntax structure = get(o) is not
Toolbox supported.
getlsb Fixed-Point —
Toolbox
2-28

getmsb Fixed-Point —
Toolbox
gradient MATLAB —
gt MATLAB —
gt Fixed-Point • Not supported for fixed-point signals with
hadamard MATLAB —
hamming Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
hankel MATLAB —
hann Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
2-29

hex2dec MATLAB —
hilb MATLAB —
hist MATLAB • Histogram bar plotting not supported; call
with at least one output argument.
• If supplied, the second argument x must be a
scalar constant.
• Inputs must be real.
histc MATLAB • The output of a variable-size array that
becomes a column vector at run time is a
column-vector, not a row-vector.
horzcat Fixed-Point —
Toolbox
hypot MATLAB —
idct Signal Processing • Length of transform dimension must
Toolbox be a power of two. If specified, the pad
or truncation value must be constant.
Line”.
generate code.
idivide MATLAB • opt string must be in lowercase.
• For efficient generated code, MATLAB rules
for divide by zero are supported only for the
'round' option.
2-30

ifft MATLAB • Length of input vector must be a power of 2.
• Output of ifft block is always complex.
• Does not support the 'symmetric' option.
ifft2 MATLAB • Length of input matrix dimensions must each
be a power of 2.
ifftn MATLAB • Length of input matrix dimensions must each
be a power of 2.
ifftshift MATLAB —
imag MATLAB —
imag Fixed-Point —
Toolbox
ind2sub MATLAB • The first argument should be a valid size
vector. Size vectors for arrays with more than
intmax elements are not supported.
inf MATLAB • Dimensions must be real, nonnegative,
integer constants.
int8, int16, int32 MATLAB —
int, int8, int16, int32 Fixed-Point —
Toolbox
interp1 MATLAB • Supports only linear and nearest
interpolation methods.
• Does not handle evenly spaced X indices
separately.
• X must be strictly monotonically increasing
or strictly monotonically decreasing; does not
reorder indices.
2-31

interp1q, see interp1 MATLAB • X must be strictly monotonically increasing
or strictly monotonically decreasing; does not
reorder indices.
intersect MATLAB • When rows is not specified:
- Inputs must be row vectors.
- If a vector is variable-sized, its first
dimension must have a fixed length of 1.
- The input [] is not supported. Use a 1-by-0
input, for example zeros(1,0), to represent
the empty set.
- Empty outputs are always row vectors,
1-by-0, never 0-by-0.
• When rows is specified, outputs ia and ib are
always column vectors. If these outputs are
empty, they are 0-by-1, never 0-by-0, even if
the output c is 0-by-0.
• Inputs must already be sorted in ascending
order. The first output is always sorted in
ascending order.
• Complex inputs must be single or double.
intfilt Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
intmax MATLAB —
2-32

intmin MATLAB —
inv MATLAB Singular matrix inputs can produce nonfinite
values that differ from MATLAB results.
invhilb MATLAB —
ipermute MATLAB —
isa MATLAB —
iscell MATLAB —
ischar MATLAB —
iscolumn MATLAB —
iscolumn Fixed-Point —
Toolbox
isdeployed MATLAB® • Returns true and false as appropriate for
Compiler™ MEX and SIM targets
• Returns false for all other targets
isempty MATLAB —
isempty Fixed-Point —
Toolbox
isequal MATLAB —
isequal Fixed-Point —
Toolbox
isequalwithequalnans MATLAB —
isfi Fixed-Point —
Toolbox
isfield MATLAB • Does not support cell input for second
argument
isfimath Fixed-Point —
Toolbox
isfimathlocal Fixed-Point —
Toolbox
2-33

isfinite MATLAB —
isfinite Fixed-Point —
Toolbox
isfloat MATLAB —
isinf MATLAB —
isinf Fixed-Point —
Toolbox
isinteger MATLAB —
islogical MATLAB —
ismatrix MATLAB —
ismcc MATLAB • Returns true and false as appropriate for
Compiler MEX and SIM targets.
• Returns false for all other targets.
ismember MATLAB • The second input, S, must be sorted in
ascending order.
isnan MATLAB —
isnan Fixed-Point —
Toolbox
isnumeric MATLAB —
isnumeric Fixed-Point —
Toolbox
isnumerictype Fixed-Point —
Toolbox
2-34

isprime MATLAB • For double precision input, the maximum
• For single precision input, the maximum
isreal MATLAB —
isreal Fixed-Point —
Toolbox
isrow MATLAB —
isrow Fixed-Point —
Toolbox
isscalar MATLAB —
isscalar Fixed-Point —
Toolbox
issigned Fixed-Point —
Toolbox
issorted MATLAB —
issparse MATLAB —
isstruct MATLAB —
istrellis Communications • Requires a Communications System Toolbox
isvector MATLAB —
2-35

isvector Fixed-Point —
Toolbox
kaiser Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
kaiserord Signal Processing • Computation performed at run time.
Toolbox
generate code.
kron MATLAB —
label2rgb Image Processing Referring to the standard syntax:
Toolbox
RGB = label2rgb(L, map, zerocolor, order)
• Submit at least two input arguments: the

label matrix, L, and the colormap matrix, map.
• map must be an n-by-3, double, colormap
matrix. You cannot use a string containing
the name of a MATLAB colormap function or
a function handle of a colormap function.
• If you set the boundary color zerocolor to the
same color as one of the regions, label2rgb
will not issue a warning.
• If you supply a value for order, it must be
'noshuffle'.
2-36

lcm MATLAB —
ldivide MATLAB —
le MATLAB —
le Fixed-Point • Not supported for fixed-point signals with
length MATLAB —
length Fixed-Point —
Toolbox
levinson Signal Processing • If specified, the order of recursion must be
Toolbox a constant. Expressions or variables are
allowed if their values do not change.
Line”.
generate code.
linsolve MATLAB • The option structure must be a constant.
• Supports only a scalar option structure input.
It does not support arrays of option structures.
• Only optimizes these cases:
- UT
- LT
- UHESS = true (the TRANSA can be either
true or false)
- SYM = true and POSDEF = true
All other options are equivalent to using
mldivide.
2-37

linspace MATLAB —
log MATLAB • Generates an error during simulation and

complex(x).
log2 MATLAB —
log10 MATLAB —
log1p MATLAB —
logical MATLAB —
logical Fixed-Point —
Toolbox
logspace MATLAB —
lowerbound Fixed-Point —
Toolbox
lsb Fixed-Point • Supported for scalar fixed-point signals only.
Toolbox
• Supported for scalar, vector, and matrix, fi
single and double signals.
lt MATLAB —
lt Fixed-Point • Not supported for fixed-point signals with
lu MATLAB —
magic MATLAB —
max MATLAB —
2-38

max Fixed-Point —
Toolbox
maxflat Signal Processing • Inputs must be constant. Expressions or
change.
Line”.
generate code.
mean MATLAB —
mean Fixed-Point —
Toolbox
median MATLAB —
median Fixed-Point —
Toolbox
meshgrid MATLAB —
min MATLAB —
min Fixed-Point —
Toolbox
minus MATLAB —
minus Fixed-Point • Any non-fi input must be constant; that is,
mldivide MATLAB —
mod MATLAB • Performs all arithmetic in the output class.
Hence, results might not match MATLAB due
to different rounding errors.
2-39

mode MATLAB • Does not support third output argument C
(cell array)
mpower MATLAB —
mpower Fixed-Point • The exponent input, k, must be constant; that
Toolbox is, its value must be known at compile time.
• Variable-sized inputs are only supported
when the SumMode property of the governing
LSB.
generated code.
governing fimath.
governing fimath.
mpy Fixed-Point • When you provide complex inputs to the
Toolbox mpy function inside a MATLAB Function
block, you must declare the input as complex
before running the simulation. To do so, go
to the Ports and data manager and set
the Complexity parameter for all known
complex inputs to On.
mrdivide MATLAB —
mrdivide Fixed-Point —
Toolbox
2-40

mtimes MATLAB —
mtimes Fixed-Point • Any non-fi input must be constant; that is,
• Variable-sized inputs are only supported
when the SumMode property of the governing
LSB.
generated code.
governing fimath.
governing fimath.
NaN or nan MATLAB • Dimensions must be real, nonnegative,
integer constants
nargchk MATLAB • Output structure does not include stack
information.
nargin MATLAB —
2-41

nargout MATLAB • For a function with no output arguments,
returns 1 if called without a terminating
semicolon.
Note This behavior also affects extrinsic calls

with no terminating semicolon. nargout is 1 for
the called function in MATLAB.
nargoutchk MATLAB • Output structure does not include stack

information.
nchoosek MATLAB —
ndgrid MATLAB —
ndims MATLAB —
ndims Fixed-Point —
Toolbox
ne MATLAB —
ne Fixed-Point • Not supported for fixed-point signals with
nearest Fixed-Point —
Toolbox
nextpow2 MATLAB —
nnz MATLAB —
nonzeros MATLAB —
norm MATLAB —
normest MATLAB —
not MATLAB —
nthroot MATLAB —
2-42

null MATLAB • Might return a different basis than MATLAB
• Does not support rational basis option (second
input)
numberofelements Fixed-Point • Returns the number of elements of fi objects
Toolbox in the generated code (works the same as
numel for fi objects in generated code).
numel MATLAB • Returns the number of elements of fi objects
in the generated code, rather than always
returning 1.
numerictype Fixed-Point • Fixed-point signals coming in to a MATLAB
Toolbox Function block from Simulink are assigned a
numerictype object that is populated with the
signal’s data type and scaling information.
• Returns the data type when the input is a
non-fixed-point signal.
• Use to create numerictype objects in the
generated code.
nuttallwin Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
ones MATLAB • Dimensions must be real, nonnegative,
integer constants
or MATLAB —
orth MATLAB • Might return a different basis than MATLAB
2-43

parzenwin Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
pascal MATLAB —
permute MATLAB —
permute Fixed-Point —
Toolbox
pi MATLAB —
pinv MATLAB —
planerot MATLAB —
plus MATLAB —
plus Fixed-Point • Any non-fi input must be constant; that is,
pol2cart MATLAB —
poly MATLAB • Does not discard nonfinite input values
• Complex input always produces complex
output
poly2trellis Communications • Requires a Communications System Toolbox
polyfit MATLAB —
polyval MATLAB —
2-44

pow2 Fixed-Point —
Toolbox
power MATLAB • Generates an error during simulation and
returns NaN in generated code when both X
and Y are real, but power(X,Y) is complex.
To get the complex result, make the input
value X complex by passing in complex(X).
For example, power(complex(X),Y).
• Generates an error during simulation and
returns NaN in generated code when both X
and Y are real, but X .^ Y is complex. To get
the complex result, make the input value X
complex by using complex(X). For example,
complex(X).^Y.
power Fixed-Point • The exponent input, k, must be constant; that
Toolbox is, its value must be known at compile time.
primes MATLAB —
prod MATLAB —
qr MATLAB —
quad2d MATLAB
• Generates a warning if the size of the internal
storage arrays is not sufficient. If a warning
occurs, a possible workaround is to divide the
region of integration into pieces and sum the
integrals over each piece.
quadgk MATLAB —
quatconj Aerospace —
Toolbox
quatdivide Aerospace —
Toolbox
quatinv Aerospace —
Toolbox
2-45

quatmod Aerospace —
Toolbox
quatmultiply Aerospace —
Toolbox
quatnorm Aerospace —
Toolbox
quatnormalize Aerospace —
Toolbox
rand MATLAB • Supports only the mt19937ar (formerly known
as ‘twister’) and mcg16807 (formerly known
as ‘seed’) generators.
• When seeding the generator, you must
terminate the call to rand with a semicolon
in the generated code.
• May not match MATLAB results if seeded
with negative values.
randn MATLAB • When seeding the generator, you must
terminate the call to rand with a semicolon
in the generated code.
• May not match MATLAB results if seeded
with negative values
randperm MATLAB —
range Fixed-Point —
Toolbox
rank MATLAB —
rcond MATLAB —
rcosfir Communications • Requires a Communications System Toolbox

2-46

rdivide MATLAB —
rdivide Fixed-Point —
Toolbox
real MATLAB —
real Fixed-Point —
Toolbox
reallog MATLAB —
realmax MATLAB —
realmax Fixed-Point —
Toolbox
realmin MATLAB —
realmin Fixed-Point —
Toolbox
realpow MATLAB —
realsqrt MATLAB —
rectwin Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
reinterpretcast Fixed-Point —
Toolbox
rem MATLAB • Performs all arithmetic in the output class.
to different rounding errors.
2-47

repmat MATLAB —
repmat Fixed-Point —
Toolbox
resample Signal Processing • The upsampling and downsampling factors
Toolbox must be specified as constants. Expressions
or variables are allowed if their values do not
change.
Line”.
generate code.
rescale Fixed-Point —
Toolbox
reshape MATLAB —
reshape Fixed-Point —
Toolbox
roots MATLAB • Output is always variable size
• Output is always complex
• Roots may not be in the same order as
MATLAB
• Roots of poorly conditioned polynomials may
not match MATLAB
rosser MATLAB —
rot90 MATLAB —
round MATLAB —
round Fixed-Point —
Toolbox
2-48

rsf2csf MATLAB —
schur MATLAB Might sometimes return a different Schur
decomposition in generated code than in
MATLAB.
sec MATLAB —
secd MATLAB —
sech MATLAB —
setdiff MATLAB • When rows is not specified:
input, for example, zeros(1,0) to represent
the empty set.
• When rows is specified, output i is always
a column vector. If i is empty, it is 0-by-1,
never 0-by-0, even if the output c is 0-by-0.
ascending order.
2-49

setxor MATLAB • When rows is not specified:
input, such as zeros(1,0), to represent the
empty set.
ascending order.
sfi Fixed-Point —
Toolbox
sgolay Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
shiftdim MATLAB Second argument must be a constant.
2-50

sign MATLAB —
sign Fixed-Point —
Toolbox
sin MATLAB —
sind MATLAB —
single MATLAB —
single Fixed-Point —
Toolbox
sinh MATLAB —
size MATLAB —
size Fixed-Point —
Toolbox
sort MATLAB —
sort Fixed-Point —
Toolbox
sortrows MATLAB —
sosfilt Signal Processing
Toolbox • Computation performed at run time.
generate code.
sph2cart MATLAB —
squeeze MATLAB —
sqrt MATLAB • Generates an error during simulation and
complex(x).
sqrt Fixed-Point • Complex and [Slope Bias] inputs error out.
Toolbox
• Negative inputs yield a 0 result.
2-51

sqrtm MATLAB —
std MATLAB —
str2func MATLAB • String must be constant/known at compile
time
strcmp MATLAB • Arguments must be computable at compile
time.
struct MATLAB —
structfun MATLAB • Does not support the ErrorHandler option.
• The number of outputs must be less than or
equal to three.
sub Fixed-Point —
Toolbox
sub2ind MATLAB • The first argument should be a valid size
vector. Size vectors for arrays with more than
intmax elements are not supported.
subsasgn Fixed-Point —
Toolbox
subspace MATLAB —
subsref Fixed-Point —
Toolbox
sum MATLAB —
sum Fixed-Point • Variable-sized inputs are only supported
LSB.
svd MATLAB —
swapbytes MATLAB Inheritance of the class of the input to swapbytes
in a MATLAB Function block is supported only
when the class of the input is double. For
non-double inputs, the input port data types
must be specified, not inherited.
2-52

tan MATLAB —
tand MATLAB —
tanh MATLAB —
taylorwin Signal Processing • Inputs must be constant
Toolbox
generate code.
times MATLAB —
times Fixed-Point • Any non-fi input must be constant; that is,
• When you provide complex inputs to the
times function inside a MATLAB Function
block, you must declare the input as complex
before running the simulation. To do so, go
to the Ports and data manager and set
the Complexity parameter for all known
complex inputs to On.
toeplitz MATLAB —
trace MATLAB —
trapz MATLAB —
transpose MATLAB —
transpose Fixed-Point —
Toolbox
2-53

triang Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
tril MATLAB • If supplied, the argument representing the
tril Fixed-Point • If supplied, the index, k, must be a real and
triu MATLAB • If supplied, the argument representing the
triu Fixed-Point • If supplied, the index, k, must be a real and
true MATLAB • Dimensions must be real, nonnegative,
integer constants
2-54

tukeywin Signal Processing • All inputs must be constant. Expressions or
change.
Line”.
generate code.
typecast MATLAB • Value of string input argument type must be
lower case
• You might receive a size error when you use
typecast with inheritance of input port data
types in MATLAB Function blocks. To avoid
this error, specify the block’s input port data
types explicitly.
ufi Fixed-Point —
Toolbox
uint8, uint16, uint32 MATLAB —
uint8, uint16, uint32 Fixed-Point —
Toolbox
uminus MATLAB —
uminus Fixed-Point —
Toolbox
2-55

union MATLAB • When rows is not specified:
input, such as zeros(1,0) to represent the
empty set.
ascending order.
unique MATLAB • When rows is not specified:
- The first input must be a row vector.
- If the vector is variable-sized, its first
input, such as zeros(1,0), to represent the
empty set.
• When rows is specified, outputs m and n are
the output b is 0-by-0.
2-56

unwrap MATLAB • Row vector input is only supported when the
first two inputs are vectors and nonscalar
• Performs all arithmetic in the output class.
to different rounding errors
upfirdn Signal Processing • Filter coefficients, upsampling factor, and
Toolbox downsampling factor must be constants.
Line”.
• Variable-size inputs are not supported.
generate code
uplus MATLAB —
uplus Fixed-Point —
Toolbox
upperbound Fixed-Point —
Toolbox
upsample Signal Processing Either declare input n as constant, or use the
Toolbox assert function in the calling function to set
upper bounds for n. For example,
assert(n<10)
vander MATLAB —
var MATLAB —
vertcat Fixed-Point —
Toolbox
2-57

wilkinson MATLAB —
xcorr Signal Processing • Does not support the case where A is a matrix
Toolbox
• Does not support partial (abbreviated) strings
of biased, unbiased, coeff, or none
• Computation performed at run time.
generate code
xor MATLAB —
yulewalk Signal Processing • If specified, the order of recursion must be
Toolbox a constant. Expressions or variables are
Line”.
generate code.
zeros MATLAB • Dimensions must be real, nonnegative,
integer constants
zp2tf MATLAB —
2-58
Functions Supported for Code Generation — Categorical List
Functions Supported for Code Generation — Categorical

List
In this section...
“Aerospace Toolbox Functions” on page 2-60
“Arithmetic Operator Functions” on page 2-60
“Bit-Wise Operation Functions” on page 2-61
“Casting Functions” on page 2-61
“Communications System Toolbox Functions” on page 2-62
“Complex Number Functions” on page 2-62
“Computer Vision System Toolbox Functions” on page 2-63
“Data Type Functions” on page 2-64
“Derivative and Integral Functions” on page 2-64
“Discrete Math Functions” on page 2-64
“Error Handling Functions” on page 2-65
“Exponential Functions” on page 2-65
“Filtering and Convolution Functions” on page 2-66
“Fixed-Point Toolbox Functions” on page 2-66
“Histogram Functions” on page 2-75
“Image Processing Toolbox Functions” on page 2-75
“Input and Output Functions” on page 2-76
“Interpolation and Computational Geometry” on page 2-76
“Linear Algebra” on page 2-77
“Logical Operator Functions” on page 2-77
“MATLAB Compiler Functions” on page 2-78
“Matrix and Array Functions” on page 2-78
“Nonlinear Numerical Methods” on page 2-82
“Polynomial Functions” on page 2-82
2-59
In this section...
“Relational Operator Functions” on page 2-82
“Rounding and Remainder Functions” on page 2-83
“Set Functions” on page 2-83
“Signal Processing Functions in MATLAB” on page 2-84
“Signal Processing Toolbox Functions” on page 2-84
“Special Values” on page 2-89
“Specialized Math” on page 2-89
“Statistical Functions” on page 2-90
“String Functions” on page 2-90
“Structure Functions” on page 2-91
“Trigonometric Functions” on page 2-91
Aerospace Toolbox Functions

Function Description
quatconj Calculate conjugate of quaternion
quatdivide Divide quaternion by another quaternion
quatinv Calculate inverse of quaternion
quatmod Calculate modulus of quaternion
quatmultiply Calculate product of two quaternions
quatnorm Calculate norm of quaternion
quatnormalize Normalize quaternion
Arithmetic Operator Functions

See Arithmetic Operators + - * / \ ^ ’ in the MATLAB Function Reference
documentation for detailed descriptions of the following operator equivalent
functions.
2-60
ctranspose Complex conjugate transpose (')
idivide Integer division with rounding option
isa Determine if input is object of given class
ldivide Left array divide
minus Minus (-)
mldivide Left matrix divide (\)
mpower Equivalent of array power operator (.^)
mrdivide Right matrix divide
mtimes Matrix multiply (*)
plus Plus (+)
power Array power
rdivide Right array divide
times Array multiply
transpose Matrix transpose (')
uminus Unary minus (-)
uplus Unary plus (+)
Bit-Wise Operation Functions

swapbytes Swap byte ordering
Casting Functions
Data Type Description
cast Cast variable to different data type
char Create character array (string)
2-61
Data Type Description

class Query class of object argument
double Convert to double-precision floating point
int8, int16, int32 Convert to signed integer data type
logical Convert to Boolean true or false data type
single Convert to single-precision floating point
typecast Convert data types without changing underlying data
uint8, uint16, Convert to unsigned integer data type
uint32
Communications System Toolbox Functions

Function Remarks/Limitations
bi2de —
de2bi —
istrellis —
poly2trellis —
rcosfir —
Complex Number Functions

complex Construct complex data from real and imaginary components
conj Return the conjugate of a complex number
imag Return the imaginary part of a complex number
isnumeric Return true for numeric arrays
isreal Return false (0) for a complex number
isscalar Return true if array is a scalar
2-62
real Return the real part of a complex number
unwrap Correct phase angles to produce smoother phase plots
Computer Vision System Toolbox Functions

epipolarLine Compute epipolar lines for stereo images
estimateFundamentalMatrix Estimate fundamental matrix from corresponding points in

stereo image
Uncalibrated stereo rectification

estimateUncalibratedRectification
extractFeatures Extract interest point descriptors
isEpipoleInImage Determine whether image contains epipole
lineToBorderPoints Intersection points of lines in image and image border
matchFeatures Find matching image features
2-63
Data Type Functions

deal Distribute inputs to outputs
iscell Determine whether input is cell array
nargchk Validate number of input arguments
nargoutchk Validate number of output arguments
str2func Construct function handle from function name string
structfun Apply function to each field of scalar structure
Derivative and Integral Functions

cumtrapz Cumulative trapezoidal numerical integration
del2 Discrete Laplacian
diff Differences and approximate derivatives
gradient Numerical gradient
trapz Trapezoidal numerical integration
Discrete Math Functions

factor Return a row vector containing the prime factors of n
gcd Return an array containing the greatest common divisors of the
corresponding elements of integer arrays
isprime Array elements that are prime numbers
lcm Least common multiple of corresponding elements in arrays
nchoosek Binomial coefficient or all combinations
primes Generate list of prime numbers
2-64
Error Handling Functions

assert Generate error when condition is violated
error Display message and abort function
Exponential Functions
exp Exponential
expm Matrix exponential
expm1 Compute exp(x)-1 accurately for small values of x
factorial Factorial function
log Natural logarithm
log2 Base 2 logarithm and dissect floating-point numbers into exponent and
mantissa
log10 Common (base 10) logarithm
log1p Compute log(1+x) accurately for small values of x
nextpow2 Next higher power of 2
nthroot Real nth root of real numbers
reallog Natural logarithm for nonnegative real arrays
realpow Array power for real-only output
realsqrt Square root for nonnegative real arrays
sqrt Square root
2-65
Filtering and Convolution Functions

conv Convolution and polynomial multiplication
conv2 2-D convolution
convn N-D convolution
deconv Deconvolution and polynomial division
detrend Remove linear trends
filter 1-D digital filter
filter2 2-D digital filter
Fixed-Point Toolbox Functions

In addition to any function-specific limitations listed in the table, the following
general limitations always apply to the use of Fixed-Point Toolbox functions
in generated code or with fiaccel:
• fipref and quantizer objects are not supported.

• Dot notation is only supported for getting the values of fimath and
numerictype properties. Dot notation is not supported for fi objects, and it
is not supported for setting properties.
• Word lengths greater than 128 bits are not supported.
• You cannot change the fimath or numerictype of a given variable after
that variable has been created.
• The boolean and ScaledDouble values of the DataTypeMode and DataType
properties are not supported.
• For all SumMode property settings other than FullPrecision, the
CastBeforeSum property must be set to true.
• The numel function returns the number of elements of fi objects in the
generated code.
• When you compile code containing fi objects with nontrivial slope and bias
scaling, you may see different results in generated code than you achieve
by running the same code in MATLAB.
2-66
• All general limitations of C/C++ code generated from MATLAB apply. See
“MATLAB Language Features Not Supported for Code Generation” on page
1-11 for more information.
abs N/A
add N/A
all N/A
any N/A
bitand Not supported for slope-bias scaled fi objects.
bitandreduce N/A
bitcmp N/A
bitconcat N/A
bitget N/A
bitor Not supported for slope-bias scaled fi objects.
bitorreduce N/A
bitreplicate N/A
bitrol N/A
bitror N/A
bitset N/A
bitshift N/A
bitsliceget N/A
bitsll N/A
bitsra N/A
bitsrl N/A
bitxor Not supported for slope-bias scaled fi objects.
bitxorreduce N/A
ceil N/A
complex N/A
2-67
conj N/A
conv • Variable-sized inputs are only supported when the
SumMode property of the governing fimath is set to
Specify precision or Keep LSB.
• For variable-sized signals, you may see different
results between generated code and MATLAB.
- In the generated code, the output for
variable-sized signals is always computed using
the SumMode property of the governing fimath.
signals is computed using the SumMode property
of the governing fimath when both inputs
are nonscalar. However, if either input is a
scalar, MATLAB computes the output using the
ProductMode of the governing fimath.
convergent N/A
cordicabs Variable-size signals are not supported.
cordicangle Variable-size signals are not supported.
cordicatan2 Variable-size signals are not supported.
cordiccart2pol Variable-size signals are not supported.
cordiccexp Variable-size signals are not supported.
cordiccos Variable-size signals are not supported.
cordicpol2cart Variable-size signals are not supported.
cordicrotate Variable-size signals are not supported.
cordicsin Variable-size signals are not supported.
cordicsincos Variable-size signals are not supported.
ctranspose N/A
diag If supplied, the index, k, must be a real and scalar
integer value that is not a fi object.
disp —
2-68
divide • Any non-fi input must be constant; that is, its
value must be known at compile time so that it can
be cast to a fi object.
• Complex and imaginary divisors are not supported.
• Code generation in MATLAB does not support the
syntax T.divide(a,b).
double N/A
end N/A
eps • Supported for scalar fixed-point signals only.
• Supported for scalar, vector, and matrix, fi single
and fi double signals.
eq Not supported for fixed-point signals with different
biases.
fi • Use to create a fixed-point constant or variable in
the generated code.
• The default constructor syntax without any input
arguments is not supported.
• The syntax
fi('PropertyName',PropertyValue...)
is not supported. To use property
name/property value pairs, you must first
specify the value v of the fi object as in
fi(v,'PropertyName',PropertyValue...).
• Works for all input values when complete
numerictype information of the fi object is
provided.
• Works only for constant input values (value of
input must be known at compile time) when
complete numerictype information of the fi object
is not specified.
2-69
• numerictype object information must be available

for nonfixed-point Simulink inputs.
filter • Variable-sized inputs are only supported when the
fimath • Fixed-point signals coming in to a MATLAB
Function block from Simulink are assigned a
fimath object. You define this object in the
MATLAB Function block dialog in the Model
Explorer.
• Use to create fimath objects in the generated code.
fix N/A
floor N/A
ge Not supported for fixed-point signals with different
biases.
get The syntax structure = get(o) is not supported.
getlsb N/A
getmsb N/A
gt Not supported for fixed-point signals with different
biases.
horzcat N/A
imag N/A
int, int8, int16, int32
N/A
iscolumn N/A
isempty N/A
isequal N/A
isfi N/A
isfimath N/A
isfimathlocal N/A
2-70
isfinite N/A
isinf N/A
isnan N/A
isnumeric N/A
isnumerictype N/A
isreal N/A
isrow N/A
isscalar N/A
issigned N/A
isvector N/A
le Not supported for fixed-point signals with different
biases.
length N/A
logical N/A
lowerbound N/A
lsb • Supported for scalar fixed-point signals only.
• Supported for scalar, vector, and matrix, fi single
and double signals.
lt Not supported for fixed-point signals with different
biases.
max N/A
mean N/A
median N/A
min N/A
minus Any non-fi input must be constant; that is, its value
must be known at compile time so that it can be cast
to a fi object.
2-71
mpower • The exponent input, k, must be constant; that is,
its value must be known at compile time.
• Variable-sized inputs are only supported when the
results between the generated code and MATLAB.
of the governing fimath when the first input,
a, is nonscalar. However, when a is a scalar,
MATLAB computes the output using the
mpy When you provide complex inputs to the mpy function
inside of a MATLAB Function block, you must declare
the input as complex before running the simulation.
To do so, go to the Ports and data manager and set
the Complexity parameter for all known complex
inputs to On.
mrdivide N/A
mtimes • Any non-fi input must be constant; that is, its
• Variable-sized inputs are only supported when the
results between the generated code and MATLAB.
2-72

of the governing fimath when both inputs
are nonscalar. However, if either input is a
scalar, MATLAB computes the output using the
ndims N/A
ne Not supported for fixed-point signals with different
biases.
nearest N/A
numberofelements numberofelements and numel both work the same as
MATLAB numel for fi objects in the generated code.
numerictype • Fixed-point signals coming in to a MATLAB
Function block from Simulink are assigned a
numerictype object that is populated with the
signal’s data type and scaling information.
• Returns the data type when the input is a
nonfixed-point signal.
• Use to create numerictype objects in generated
code.
permute N/A
plus Any non-fi input must be constant; that is, its value
must be known at compile time so that it can be cast
to a fi object.
pow2 N/A
power The exponent input, k, must be constant; that is, its
value must be known at compile time.
range N/A
2-73
rdivide N/A
real N/A
realmax N/A
realmin N/A
reinterpretcast N/A
repmat N/A
rescale N/A
reshape N/A
round N/A
sfi N/A
sign N/A
single N/A
size N/A
sort N/A
sqrt • Complex and [Slope Bias] inputs error out.
• Negative inputs yield a 0 result.
sub N/A
subsasgn N/A
subsref N/A
sum Variable-sized inputs are only supported when the
2-74
times • Any non-fi input must be constant; that is, its
• When you provide complex inputs to the times
function inside of a MATLAB Function block, you
must declare the input as complex before running
the simulation. To do so, go to the Ports and data
manager and set the Complexity parameter for
all known complex inputs to On.
transpose N/A
tril If supplied, the index, k, must be a real and scalar
triu If supplied, the index, k, must be a real and scalar
ufi N/A
uint8, uint16, uint32N/A
uminus N/A
uplus N/A
upperbound N/A
vertcat N/A
Histogram Functions
hist Non-graphical histogram
histc Histogram count
Image Processing Toolbox Functions

You must have the Image Processing Toolbox software installed to generate
C/C++ code from MATLAB for these functions.
2-75
fspecial All inputs must be constants at compilation time. Expressions or
variables are allowed if their values do not change.
label2rgb Referring to the standard syntax:
RGB = label2rgb(L, map, zerocolor, order)
• Submit at least two input arguments: the label matrix, L, and the
colormap matrix, map.
• map must be an n-by-3, double, colormap matrix. You cannot use
a string containing the name of a MATLAB colormap function or a
function handle of a colormap function.
• If you set the boundary color zerocolor to the same color as one of
the regions, label2rgb will not issue a warning.
• If you supply a value for order, it must be 'noshuffle'.
Input and Output Functions

nargin Return the number of input arguments a user has supplied
nargout Return the number of output return values a user has requested
Interpolation and Computational Geometry

cart2pol Transform Cartesian coordinates to polar or cylindrical
cart2sph Transform Cartesian coordinates to spherical
interp1 One-dimensional interpolation (table lookup)
interp1q Quick one-dimensional linear interpolation (table lookup)
meshgrid Generate X and Y arrays for 3-D plots
2-76
pol2cart Transform polar or cylindrical coordinates to Cartesian
sph2cart Transform spherical coordinates to Cartesian
Linear Algebra
linsolve Solve linear system of equations
null Null space

orth Range space of matrix
rsf2csf Convert real Schur form to complex Schur form
schur Schur decomposition
sqrtm Matrix square root
Logical Operator Functions

and Logical AND (&&)
bitand Bitwise AND
bitcmp Bitwise complement
bitget Bit at specified position
bitor Bitwise OR
bitset Set bit at specified position
bitshift Shift bits specified number of places
bitxor Bitwise XOR
not Logical NOT (~)
2-77
or Logical OR (||)
xor Logical exclusive-OR
MATLAB Compiler Functions

isdeployed Determine whether code is running in deployed or MATLAB mode
ismcc Test if code is running during compilation process (using mcc)
Matrix and Array Functions

abs Return absolute value and complex magnitude of an array
all Test if all elements are nonzero
angle Phase angle
any Test for any nonzero elements
bsxfun Applies element-by-element binary operation to two arrays with
singleton expansion enabled
cat Concatenate arrays along specified dimension
circshift Shift array circularly
compan Companion matrix
cond Condition number of a matrix with respect to inversion
cov Covariance matrix
cross Vector cross product
cumprod Cumulative product of array elements
cumsum Cumulative sum of array elements
det Matrix determinant
2-78
diag Return a matrix formed around the specified diagonal vector and the
specified diagonal (0, 1, 2,...) it occupies
diff Differences and approximate derivatives
dot Vector dot product
eig Eigenvalues and eigenvectors
eye Identity matrix
false Return an array of 0s for the specified dimensions
find Find indices and values of nonzero elements
flipdim Flip array along specified dimension
fliplr Flip matrix left to right
flipud Flip matrix up to down
full Convert sparse matrix to full matrix
hadamard Hadamard matrix
hankel Hankel matrix
hilb Hilbert matrix
ind2sub Subscripts from linear index
inv Inverse of a square matrix
invhilb Inverse of Hilbert matrix
ipermute Inverse permute dimensions of array
iscolumn True if input is a column vector
isempty Determine whether array is empty
isequal Test arrays for equality
isequalwithequalnans Test arrays for equality, treating NaNs as equal
isfinite Detect finite elements of an array
isfloat Determine if input is floating-point array
isinf Detect infinite elements of an array
isinteger Determine if input is integer array
2-79
islogical Determine if input is logical array
ismatrix True if input is a matrix
isnan Detect NaN elements of an array
isrow True if input is a row vector
issparse Determine whether input is sparse
isvector Determine whether input is vector
kron Kronecker tensor product
length Return the length of a matrix
linspace Generate linearly spaced vectors
logspace Generate logarithmically spaced vectors
lu Matrix factorization
magic Magic square
max Maximum elements of a matrix

min Minimum elements of a matrix
ndgrid Generate arrays for N-D functions and interpolation
ndims Number of dimensions
nnz Number of nonzero matrix elements
nonzeros Nonzero matrix elements
norm Vector and matrix norms
normest 2-norm estimate
numel Number of elements in array or subscripted array
ones Create a matrix of all 1s
pascal Pascal matrix
permute Rearrange dimensions of array
pinv Pseudoinverse of a matrix
planerot Givens plane rotation
2-80
prod Product of array element
qr Orthogonal-triangular decomposition
randperm Random permutation
rank Rank of matrix
rcond Matrix reciprocal condition number estimate
repmat Replicate and tile an array
reshape Reshape one array into the dimensions of another
rosser Classic symmetric eigenvalue test problem
rot90 Rotate matrix 90 degrees
shiftdim Shift dimensions
sign Signum function
size Return the size of a matrix
sort Sort elements in ascending or descending order
sortrows Sort rows in ascending order
squeeze Remove singleton dimensions
sub2ind Single index from subscripts
subspace Angle between two subspaces
sum Sum of matrix elements
toeplitz Toeplitz matrix
trace Sum of diagonal elements
tril Extract lower triangular part
triu Extract upper triangular part
true Return an array of logical (Boolean) 1s for the specified dimensions
vander Vandermonde matrix
wilkinson Wilkinson’s eigenvalue test matrix
zeros Create a matrix of all zeros
2-81
Nonlinear Numerical Methods

fzero Find root of continuous function of one variable
quad2d Numerically evaluate double integral over planar region
quadgk Numerically evaluate integral, adaptive Gauss-Kronrod quadrature
Polynomial Functions
poly Polynomial with specified roots
polyfit Polynomial curve fitting
polyval Polynomial evaluation
roots Polynomial roots
Relational Operator Functions

eq Equal (==)
ge Greater than or equal to (>=)
gt Greater than (>)
le Less than or equal to (<=)
lt Less than (<)
ne Not equal (~=)
2-82
Rounding and Remainder Functions

ceil Round toward plus infinity
ceil Round toward positive infinity
convergent Round toward nearest integer with ties rounding to nearest even integer
fix Round toward zero
fix Round toward zero
floor Round toward minus infinity
floor Round toward negative infinity
mod Modulus (signed remainder after division)
nearest Round toward nearest integer with ties rounding toward positive infinity
rem Remainder after division
round Round toward nearest integer
round Round fi object toward nearest integer or round input data using
quantizer object
Set Functions
intersect Find set intersection of two vectors
ismember Array elements that are members of set
issorted Determine whether set elements are in sorted order
setdiff Find set difference of two vectors
setxor Find set exclusive OR of two vectors
union Find set union of two vectors
unique Find unique elements of vector
2-83
Signal Processing Functions in MATLAB

chol Cholesky factorization
conv Convolution and polynomial multiplication
fft Discrete Fourier transform
fft2 2-D discrete Fourier transform
fftn N-D discrete Fourier transform
fftshift Shift zero-frequency component to center of spectrum
filter Filter a data sequence using a digital filter that works for both real and
complex inputs
freqspace Frequency spacing for frequency response
ifft Inverse discrete Fourier transform
ifft2 2-D inverse discrete Fourier transform
ifftn N-D inverse discrete Fourier transform
ifftshift Inverse discrete Fourier transform shift
svd Singular value decomposition
zp2tf Convert zero-pole-gain filter parameters to transfer function form
Signal Processing Toolbox Functions
Note All of these functions require a DSP System Toolbox license to generate
code.
2-84
Note Many Signal Processing Toolbox functions require constant inputs in

generated code. To specify a constant input for codegen, use coder.Constant.
For more information, see the “MATLAB Coder” documentation.
barthannwin Window length must be a constant. Expressions or variables are
bartlett Window length must be a constant. Expressions or variables are
besselap Filter order must be a constant. Expressions or variables are allowed if
bitrevorder —
blackman Window length must be a constant. Expressions or variables are
blackmanharris Window length must be a constant. Expressions or variables are
bohmanwin Window length must be a constant. Expressions or variables are
buttap Filter order must be a constant. Expressions or variables are allowed if
butter Filter coefficients must be constants. Expressions or variables are
buttord All inputs must be constants. Expressions or variables are allowed if
cfirpm All inputs must be constants. Expressions or variables are allowed if
cheb1ap All inputs must be constants. Expressions or variables are allowed if
cheb2ap All inputs must be constants. Expressions or variables are allowed if
cheb1ord All inputs must be constants. Expressions or variables are allowed if
2-85
cheb2ord All inputs must be constants. Expressions or variables are allowed if
chebwin All inputs must be constants. Expressions or variables are allowed if
cheby1 All Inputs must be constants. Expressions or variables are allowed if
cheby2 All inputs must be constants. Expressions or variables are allowed if
dct Length of transform dimension must be a power of two. If specified, the
pad or truncation value must be constant. Expressions or variables are
downsample —
dpss All inputs must be constants. Expressions or variables are allowed if
ellip Inputs must be constant. Expressions or variables are allowed if their
ellipap All inputs must be constants. Expressions or variables are allowed if
ellipord All inputs must be constants. Expressions or variables are allowed if
filtfilt Filter coefficients must be constants. Expressions or variables are
fir1 All inputs must be constants. Expressions or variables are allowed if
fir2 All inputs must be constants. Expressions or variables are allowed if
fircls All inputs must be constants. Expressions or variables are allowed if
fircls1 All inputs must be constants. Expressions or variables are allowed if
2-86
firls All inputs must be constants. Expressions or variables are allowed if
firpm All inputs must be constants. Expressions or variables are allowed if
firpmord All inputs must be constants. Expressions or variables are allowed if
firrcos All inputs must be constants. Expressions or variables are allowed if
flattopwin All inputs must be constants. Expressions or variables are allowed if
freqz freqz with no output arguments produces a plot only when the function
call terminates in a semicolon. See “freqz With No Output Arguments”.
gaussfir All inputs must be constant. Expressions or variables are allowed if
gausswin All inputs must be constant. Expressions or variables are allowed if
hamming All inputs must be constant. Expressions or variables are allowed if
hann All inputs must be constant. Expressions or variables are allowed if
idct Length of transform dimension must be a power of two. If specified, the
pad or truncation value must be constant. Expressions or variables are
intfilt All inputs must be constant. Expressions or variables are allowed if
kaiser All inputs must be constant. Expressions or variables are allowed if
kaiserord —
levinson If specified, the order of recursion must be a constant. Expressions or
2-87
maxflat All inputs must be constant. Expressions or variables are allowed if
nuttallwin All inputs must be constant. Expressions or variables are allowed if
parzenwin All inputs must be constant. Expressions or variables are allowed if
rectwin All inputs must be constant. Expressions or variables are allowed if
resample The upsampling and downsampling factors must be specified as
constants. Expressions or variables are allowed if their values do not
change.
sgolay All inputs must be constant. Expressions or variables are allowed if
sosfilt —
taylorwin All inputs must be constant. Expressions or variables are allowed if
triang All inputs must be constant. Expressions or variables are allowed if
tukeywin All inputs must be constant. Expressions or variables are allowed if
upfirdn • Filter coefficients, upsampling factor, and downsampling factor must
be constants. Expressions or variables are allowed if their values
do not change.
• Variable-size inputs are not supported.
upsample Either declare input n as constant, or use the assert function in the
calling function to set upper bounds for n. For example,
assert(n<10)
xcorr —
yulewalk If specified, the order of recursion must be a constant. Expressions or
2-88
Special Values
Symbol Description
eps Floating-point relative accuracy
inf IEEE® arithmetic representation for positive infinity
intmax Largest possible value of specified integer type
intmin Smallest possible value of specified integer type
NaN or nan Not a number
pi Ratio of the circumference to the diameter for a circle
rand Uniformly distributed pseudorandom numbers
randn Normally distributed random numbers
realmax Largest positive floating-point number
realmin Smallest positive floating-point number
Specialized Math
Symbol Description
beta Beta function
betainc Incomplete beta function
betaln Logarithm of beta function
ellipke Complete elliptic integrals of first and second kind
erf Error function
erfc Complementary error function
erfcinv Inverse of complementary error function
erfcx Scaled complementary error function
erfinv Inverse error function
expint Exponential integral
gamma Gamma function
2-89
Symbol Description
gammainc Incomplete gamma function
gammaln Logarithm of the gamma function
Statistical Functions
corrcoef Correlation coefficients
mean Average or mean value of array
median Median value of array
mode Most frequent values in array
std Standard deviation
var Variance
String Functions
bin2dec Convert binary number string to decimal number
bitmax Maximum double-precision floating-point integer
blanks Create string of blank characters
char Create character array (string)
hex2dec Convert hexadecimal number string to decimal number
ischar True for character array (string)
strcmp Return a logical result for the comparison of two strings; limited to
strings known at compile time
2-90
Structure Functions
isfield Determine whether input is structure array field
struct Create structure
isstruct Determine whether input is a structure
Trigonometric Functions
acos Inverse cosine
acosd Inverse cosine; result in degrees
acosh Inverse hyperbolic cosine
acot Inverse cotangent; result in radians
acotd Inverse cotangent; result in degrees
acoth Inverse hyperbolic cotangent
acsc Inverse cosecant; result in radians
acscd Inverse cosecant; result in degrees
acsch Inverse cosecant and inverse hyperbolic cosecant
asec Inverse secant; result in radians
asecd Inverse secant; result in degrees
asech Inverse hyperbolic secant
asin Inverse sine
asinh Inverse hyperbolic sine
atan Inverse tangent
atan2 Four quadrant inverse tangent
atand Inverse tangent; result in degrees
atanh Inverse hyperbolic tangent
2-91
cos Cosine
cosd Cosine; result in degrees
cosh Hyperbolic cosine
cot Cotangent; result in radians
cotd Cotangent; result in degrees
coth Hyperbolic cotangent
csc Cosecant; result in radians
cscd Cosecant; result in degrees
csch Hyperbolic cosecant
hypot Square root of sum of squares
sec Secant; result in radians
secd Secant; result in degrees
sech Hyperbolic secant
sin Sine
sind Sine; result in degrees
sinh Hyperbolic sine
tan Tangent
tand Tangent; result in degrees
tanh Hyperbolic tangent
2-92
3
System Objects Supported

for Code Generation
• “About Code Generation for System Objects” on page 3-2

• “Computer Vision System Toolbox System Objects” on page 3-3
• “Communications System Toolbox System Objects” on page 3-7
• “DSP System Toolbox System Objects” on page 3-13
3 System Objects Supported for Code Generation
About Code Generation for System Objects

You can generate C/C++ code for a subset of System objects provided by
Communications System Toolbox, DSP System Toolbox, and Computer
Vision System Toolbox. To use these System objects, you need to install the
appropriate toolbox.
System objects are MATLAB object-oriented implementations of algorithms.

They extend MATLAB by enabling you to model dynamic systems represented
by time-varying algorithms. System objects are well integrated into the
MATLAB language, regardless of whether you are writing simple functions,
working interactively in the command window, or creating large applications.
In contrast to MATLAB functions, System objects automatically manage

state information, data indexing, and buffering, which is particularly useful
for iterative computations or stream data processing. This enables efficient
processing of long data sets. For general information on MATLAB objects, see
Object-Oriented Programming in the MATLAB documentation.
3-2
Computer Vision System Toolbox™ System Objects
Computer Vision System Toolbox System Objects

If you install Computer Vision System Toolbox software, you can generate
C/C++ code for the following Computer Vision System Toolbox System objects.
For more information on how to use these System objects, see “Use System
Objects for Code Generation from MATLAB” in the Computer Vision System
Toolbox documentation.
Supported Computer Vision System Toolbox System Objects
Object Description
Analysis & Enhancement
vision.BoundaryTracer Trace object boundaries in binary images
vision.ContrastAdjuster Adjust image contrast by linear scaling
vision.Deinterlacer Remove motion artifacts by deinterlacing input
video signal
vision.EdgeDetector Find edges of objects in images
vision.ForegroundDetector Detect foreground using Gaussian Mixture
Models
vision.HistogramEqualizer Enhance contrast of images using histogram
equalization
vision.TemplateMatcher Perform template matching by shifting
template over image
Conversions
vision.Autothresholder Convert intensity image to binary image
vision.ChromaResampler Downsample or upsample chrominance
components of images
vision.ColorSpaceConverter Convert color information between color spaces
vision.DemosaicInterpolator Demosaic Bayer’s format images
vision.GammaCorrector Apply or remove gamma correction from
images or video streams
3-3
Supported Computer Vision System Toolbox System Objects (Continued)
Object Description
vision.ImageComplementer Compute complement of pixel values in binary,
intensity, or RGB images
vision.ImageDataTypeConverter Convert and scale input image to specified
output data type
Filtering
vision.Convolver Compute 2-D discrete convolution of two input
matrices
vision.ImageFilter Perform 2-D FIR filtering of input matrix
vision.MedianFilter 2D median filtering
Geometric Transformations
vision.GeometricRotator Rotate image by specified angle
vision.GeometricRotator Enlarge or shrink image size
vision.GeometricScaler Shift rows or columns of image by linearly

varying offset
vision.GeometricTransformer Apply projective or affine transformation to an

image
vision.GeometricTransformEstimator Estimate geometric transformation from

matching point pairs
vision.GeometricTranslator Translate image in two-dimensional plane
using displacement vector
Morphological Operations
vision.ConnectedComponentLabeler Label and count the connected regions in a
binary image
vision.MorphologicalClose Perform morphological closing on image
vision.MorphologicalDilate Perform morphological dilation on an image
3-4
Computer Vision System Toolbox™ System Objects
Object Description
vision.MorphologicalErode Perform morphological erosion on an image
vision.MorphologicalOpen Perform morphological opening on an image
Sinks
vision.DeployableVideoPlayer Send video data to computer screen
vision.VideoFileWriter Write video frames and audio samples to

multimedia file
Sources
vision.VideoFileReader Read video frames and audio samples from
compressed multimedia file
Statistics
vision.Autocorrelator Compute 2-D autocorrelation of input matrix
vision.BlobAnalysis Compute statistics for connected regions in a
binary image
vision.Crosscorrelator Compute 2-D cross-correlation of two input
matrices
vision.Histogram Generate histogram of each input matrix
vision.LocalMaximaFinder Find local maxima in matrices
vision.Maximum Find maximum values in input or sequence of
inputs
vision.Mean Find mean value of input or sequence of inputs
vision.Median Find median values in an input
vision.Minimum Find minimum values in input or sequence of
inputs
3-5
Object Description
vision.PSNR Compute peak signal-to-noise ratio (PSNR)
between images
vision.StandardDeviation Find standard deviation of input or sequence
of inputs
vision.Variance Find variance values in an input or sequence
of inputs
Text & Graphics
vision.AlphaBlender Combine images, overlay images, or highlight
selected pixels
vision.MarkerInserter Draw markers on output image
vision.ShapeInserter Draw rectangles, lines, polygons, or circles on
images
vision.TextInserter Draw text on image or video stream
Transforms
vision.DCT Compute 2-D discrete cosine transform
vision.FFT Two-dimensional discrete Fourier transform
vision.HoughLines Find Cartesian coordinates of lines that are
described by rho and theta pairs
vision.HoughTransform Find lines in images via Hough transform
vision.IDCT Compute 2-D inverse discrete cosine transform
vision.IFFT Two–dimensional inverse discrete Fourier
transform
vision.Pyramid Perform Gaussian pyramid decomposition
Utilities
vision.ImagePadder Pad or crop input image along its rows,
columns, or both
3-6
Communications System Toolbox™ System Objects
Communications System Toolbox System Objects

If you install Communications System Toolbox software, you can generate
C/C++ code for the following Communications System Toolbox System objects.
For information on how to use these System objects, see “Generating Code
for Communications System Objects” in the Communications System Toolbox
documentation.
Supported Communications System Toolbox System Objects
Object Description
Source Coding
comm.DifferentialDecoder Decode binary signal using differential decoding
comm.DifferentialEncoder Encode binary signal using differential coding
Channels
comm.AWGNChannel Add white Gaussian noise to input signal
comm.BinarySymmetricChannel Introduce binary errors
Equalizers
comm.MLSEEqualizer Equalize using maximum likelihood sequence
estimation
Filters
comm.IntegrateAndDumpFilter Integrate discrete-time signal with periodic resets
Measurements
comm.EVM Measure error vector magnitude
comm.MER Measure modulation error ratio
Sources
comm.BarkerCode Generate Barker code
comm.HadamardCode Generate Hadamard code
comm.KasamiSequence Generate a Kasami sequence
comm.OVSFCode Generate OVSF code
3-7
Supported Communications System Toolbox System Objects (Continued)
Object Description
comm.PNSequence Generate a pseudo-noise (PN) sequence
comm.WalshCode Generate Walsh code from orthogonal set of codes
Error Detection and Correction – Convolutional Coding
comm.ConvolutionalEncoder Convolutionally encode binary data
comm.ViterbiDecoder Decode convolutionally encoded data using Viterbi
algorithm
Error Detection and Correction – Cyclic Redundancy Check Coding
comm.CRCDetector Detect errors in input data using cyclic redundancy
code
comm.CRCGenerator Generate cyclic redundancy code bits and append to
input data
comm.TurboDecoder Decode input signal using parallel concatenated
decoding scheme
comm.TurboEncoder Encode input signal using parallel concatenated
encoding scheme
Interleavers – Block
comm.AlgebraicDeinterleaver Deinterleave input symbols using algebraically
derived permutation vector
comm.AlgebraicInterleaver Permute input symbols using an algebraically
derived permutation vector
comm.BlockDeinterleaver Deinterleave input symbols using permutation
vector
comm.BlockInterleaver Permute input symbols using a permutation vector
comm.MatrixDeinterleaver Deinterleave input symbols using permutation
matrix
comm.MatrixInterleaver Permute input symbols using permutation matrix
comm.MatrixHelicalScanDeinterleaver Deinterleave input symbols by filling a matrix along
diagonals
3-8
Object Description
comm.MatrixHelicalScanInterleaver Permute input symbols by selecting matrix elements
along diagonals
Interleavers – Convolutional
comm.ConvolutionalDeinterleaver Restore ordering of symbols using shift registers
comm.ConvolutionalInterleaver Permute input symbols using shift registers
comm.HelicalDeinterleaver Restore ordering of symbols using a helical array
comm.HelicalInterleaver Permute input symbols using a helical array
comm.MultiplexedDeinterleaver Restore ordering of symbols using a set of shift
registers with specified delays
comm.MultiplexedInterleaver Permute input symbols using a set of shift registers
with specified delays
MIMO
comm.OSTBCCombiner Combine inputs using orthogonal space-time block
code
comm.OSTBCEncoder Encode input message using orthogonal space-time
block code
Digital Baseband Modulation – Phase
comm.BPSKDemodulator Demodulate using binary PSK method
comm.BPSKModulator Modulate using binary PSK method
comm.DBPSKModulator Modulate using differential binary PSK method
comm.DPSKDemodulator Demodulate using M-ary DPSK method
comm.DPSKModulator Modulate using M-ary DPSK method
comm.DQPSKDemodulator Demodulate using differential quadrature PSK
method
comm.DQPSKModulator Modulate using differential quadrature PSK method
comm.DBPSKDemodulator Demodulate using M-ary DPSK method
comm.QPSKDemodulator Demodulate using quadrature PSK method
3-9
Object Description
comm.QPSKModulator Modulate using quadrature PSK method
comm.PSKDemodulator Demodulate using M-ary PSK method
comm.PSKModulator Modulate using M-ary PSK method
comm.OQPSKDemodulator Demodulate offset quadrature PSK modulated data
comm.OQPSKModulator Modulate using offset quadrature PSK method
Digital Baseband Modulation – Amplitude
comm.GeneralQAMDemodulator Demodulate using arbitrary QAM constellation
comm.GeneralQAMModulator Modulate using arbitrary QAM constellation
comm.PAMDemodulator Demodulate using M-ary PAM method
comm.PAMModulator Modulate using M-ary PAM method
comm.RectangularQAMDemodulator Demodulate using rectangular QAM method
comm.RectangularQAMModulator Modulate using rectangular QAM method
Digital Baseband Modulation – Frequency
comm.FSKDemodulator Demodulate using M-ary FSK method
comm.FSKModulator Modulate using M-ary FSK method
Digital Baseband Modulation – Trelllis Coded
comm.GeneralQAMTCMDemodulator Demodulate convolutionally encoded data mapped
to arbitrary QAM constellation
comm.GeneralQAMTCMModulator Convolutionally encode binary data and map using
arbitrary QAM constellation
comm.PSKTCMDemodulator Demodulate convolutionally encoded data mapped
to M-ary PSK constellation
comm.PSKTCMModulator Convolutionally encode binary data and map using
M-ary PSK constellation
comm.RectangularQAMTCMDemodulator Demodulate convolutionally encoded data mapped
to rectangular QAM constellation
3-10
Object Description
comm.RectangularQAMTCMModulator Convolutionally encode binary data and map using
rectangular QAM constellation
Digital Baseband Modulation – Continuous Phase
comm.CPFSKDemodulator Demodulate using CPFSK method and Viterbi
algorithm
comm.CPFSKModulator Modulate using CPFSK method
comm.CPMDemodulator Demodulate using CPM method and Viterbi
algorithm
comm.CPMModulator Modulate using CPM method
comm.GMSKDemodulator Demodulate using GMSK method and the Viterbi
algorithm
comm.GMSKModulator Modulate using GMSK method
comm.MSKDemodulator Demodulate using MSK method and the Viterbi
algorithm
comm.MSKModulator Modulate using MSK method
RF Impairments
comm.PhaseFrequencyOffset Apply phase and frequency offsets to input signal
Synchronization – Timing Phase
comm.EarlyLateGateTimingSynchronizer Recover symbol timing phase using early-late gate
method
comm.GardnerTimingSynchronizer Recover symbol timing phase using Gardner’s
method
comm.GMSKTimingSynchronizer Recover symbol timing phase using fourth-order
nonlinearity method
comm.MSKTimingSynchronizer Recover symbol timing phase using fourth-order
nonlinearity method
comm.MuellerMullerTimingSynchronizer Recover symbol timing phase using Mueller-Muller
method
3-11
Object Description
Synchronization Utilities
comm.DiscreteTimeVCO Generate variable frequency sinusoid
Converters
comm.BitToInteger Convert vector of bits to vector of integers
comm.IntegerToBit Convert vector of integers to vector of bits
Sequence Operators
comm.Descrambler Descramble input signal
comm.Scrambler Scramble input signal
3-12
DSP System Toolbox™ System Objects
DSP System Toolbox System Objects

If you install DSP System Toolbox software, you can generate C/C++ code for
the following DSP System Toolbox System objects. For information on how to
use these System objects, see “Code Generation with System Objects” in the
DSP System Toolbox documentation.
Supported DSP System Toolbox System Objects
Object Description
Estimation
dsp.BurgAREstimator Compute estimate of autoregressive model parameters
using Burg method
dsp.BurgSpectrumEstimator Compute parametric spectral estimate using Burg
method
Note For code generation, you cannot call the reset

method before calling the step method.
dsp.CepstralToLPC Convert cepstral coefficients to linear prediction

coefficients
dsp.LevinsonSolver Solve linear system of equations using
Levinson-Durbin recursion
dsp.LPCToAutocorrelation Convert linear prediction coefficients to autocorrelation
coefficients
dsp.LPCToCepstral Convert linear prediction coefficients to cepstral
coefficients
dsp.LPCToLSF Convert linear prediction coefficients to line spectral
frequencies
dsp.LPCToLSP Convert linear prediction coefficients to line spectral
pairs
dsp.LPCToRC Convert linear prediction coefficients to reflection
coefficients
3-13
Supported DSP System Toolbox System Objects (Continued)
Object Description
dsp.LSFToLPC Convert line spectral frequencies to linear prediction
coefficients
dsp.LSPToLPC Convert line spectral pairs to linear prediction
coefficients
dsp.RCToAutocorrelation Convert reflection coefficients to autocorrelation

coefficients
dsp.RCToLPC Convert reflection coefficients to linear prediction
coefficients
Filters
dsp.BiquadFilter Model biquadratic IIR (SOS) filters
dsp.DigitalFilter Filter each channel of input over time using
discrete-time filter implementations
dsp.FIRInterpolator Upsample and filter input signals
dsp.FIRRateConverter Upsample, filter and downsample input signals
dsp.LMSFilter Compute output, error, and weights using LMS
adaptive algorithm
Math Operations
dsp.ArrayVectorAdder Add vector to array along specified dimension
dsp.ArrayVectorDivider Divide array by vector along specified dimension
dsp.ArrayVectorMultiplier Multiply array by vector along specified dimension
dsp.ArrayVectorSubtractor Subtract vector from array along specified dimension
dsp.CumulativeProduct Compute cumulative product of channel, column, or
row elements
dsp.CumulativeSum Compute cumulative sum of channel, column, or row
elements
dsp.LDLFactor Factor square Hermitian positive definite matrices
into lower, upper, and diagonal components
3-14
Object Description
dsp.LevinsonSolver Solve linear system of equations using
Levinson-Durbin recursion
dsp.LowerTriangularSolver Solve LX = B for X when L is lower triangular matrix
dsp.LUFactor Factor square matrix into lower and upper triangular
matrices
dsp.Normalizer Normalize input
dsp.UpperTriangularSolver Solve UX = B for X when U is upper triangular matrix
Quantizers
dsp.ScalarQuantizerDecoder Convert each index value into quantized output value

dsp.ScalarQuantizerEncoder Perform scalar quantization encoding
dsp.VectorQuantizerDecoder Find vector quantizer codeword for given index value
dsp.VectorQuantizerEncoder Perform vector quantization encoding
Signal Management
dsp.Counter Count up or down through specified range of numbers
dsp.DelayLine Rebuffer sequence of inputs with one-sample shift
Signal Operations
dsp.Convolver Compute convolution of two inputs
dsp.Delay Delay input by specified number of samples or frames
dsp.Interpolator Interpolate values of real input samples
dsp.NCO Generate real or complex sinusoidal signals
dsp.PeakFinder Determine extrema (maxima or minima) in input
signal
dsp.PhaseUnwrapper Unwrap signal phase
dsp.VariableFractionalDelay Delay input by time-varying fractional number of
sample periods
3-15
Object Description
dsp.VariableIntegerDelay Delay input by time-varying integer number of sample
periods
dsp.Window Generate or apply window function
dsp.ZeroCrossingDetector Calculate number of zero crossings of a signal
Signal Processing Sinks
dsp.AudioPlayer Write audio data to computer’s audio device
dsp.AudioFileWriter Write audio file
dsp.UDPSender Send UDP packets to the network
Signal Processing Sources
dsp.AudioFileReader Read audio samples from an audio file
dsp.AudioRecorder Read audio data from computer’s audio device
dsp.UDPReceiver Receive UDP packets from the network

dsp.SineWave Generate discrete sine wave
Statistics
dsp.Autocorrelator Compute autocorrelation of vector inputs
dsp.Crosscorrelator Compute cross-correlation of two inputs
dsp.Histogram Output histogram of an input or sequence of inputs
dsp.Maximum Compute maximum value in input
dsp.Mean Compute average or mean value in input
dsp.Median Compute median value in input
dsp.Minimum Compute minimum value in input
dsp.RMS Compute root-mean-square of vector elements
dsp.StandardDeviation Compute standard deviation of vector elements
3-16
Object Description
dsp.Variance Compute variance of input or sequence of inputs
Transforms
dsp.AnalyticSignal Compute analytic signals of discrete-time inputs
Note For code generation, you cannot call the reset

method before calling the step method.
dsp.DCT Compute discrete cosine transform (DCT) of input

dsp.FFT Compute fast Fourier transform (FFT) of input
dsp.IDCT Compute inverse discrete cosine transform (IDCT) of
input
dsp.IFFT Compute inverse fast Fourier transform (IFFT) of

input
3-17
3-18
4
Defining MATLAB
Variables for C/C++ Code
Generation
• “Why Define Variables Differently for Code Generation?” on page 4-2

• “Best Practices for Defining Variables for C/C++ Code Generation” on
page 4-3
• “When You Can Reassign Variable Properties for C/C++ Code Generation”
on page 4-7
• “Eliminating Redundant Copies of Variables in Generated Code” on page
4-8
• “Defining and Initializing Persistent Variables” on page 4-10
• “Reusing the Same Variable with Different Properties” on page 4-11
• “Supported Variable Types” on page 4-16
4 Defining MATLAB® Variables for C/C++ Code Generation
Why Define Variables Differently for Code Generation?

In the MATLAB language, variables can change their properties dynamically
at run time so you can use the same variable to hold a value of any class, size,
or complexity. For example, the following code works in MATLAB:
function x = foo(c) %#codegen

coder.extrinsic('disp');
if(c>0)
x = int8(0);
else
x = [1 2 3];
end
disp(x);
end
However, statically-typed languages like C must be able to determine variable

properties at compile time. Therefore, for C/C++ code generation, you must
explicitly define the class, size, and complexity of variables in MATLAB
source code before using them. For example, rewrite the above source code
with a definition for x:
function x = foo(c) %#codegen

x = zeros(1,3);
if(c>0)
x = int8(0);
else
x = [1 2 3];
end
disp(x);
end
For more information, see “Best Practices for Defining Variables for C/C++
Code Generation” on page 4-3.
4-2
Best Practices for Defining Variables for C/C++ Code Generation
Best Practices for Defining Variables for C/C++ Code

Generation
In this section...
“Define Variables By Assignment Before Using Them” on page 4-3
“Use Caution When Reassigning Variables” on page 4-6
“Use Type Cast Operators in Variable Definitions” on page 4-6
“Define Matrices Before Assigning Indexed Variables” on page 4-6
Define Variables By Assignment Before Using Them

For C/C++ code generation, you should explicitly and unambiguously define
the class, size, and complexity of variables before using them in operations or
returning them as outputs. Define variables by assignment, but note that the
assignment copies not only the value, but also the size, class, and complexity
represented by that value to the new variable. For example:
Assignment: Defines:
a = 14.7; a as a real double scalar.
b = a; b with properties of a (real double
scalar).
c = zeros(5,2); c as a real 5-by-2 array of doubles.
d = [1 2 3 4 5; 6 7 8 9 0]; d as a real 5-by-2 array of doubles.
y = int16(3); y as a real 16-bit integer scalar.
Define properties this way so that the variable is defined on all execution
paths during C/C++ code generation (see Example: Defining a Variable for
Multiple Execution Paths on page 4-4).
The data that you assign to a variable can be a scalar, matrix, or structure. If
your variable is a structure, define the properties of each field explicitly (see
Example: Defining All Fields in a Structure on page 4-5).
4-3
Initializing the new variable to the value of the assigned data sometimes
results in redundant copies in the generated code. To avoid redundant
copies, you can define variables without initializing their values by using the
coder.nullcopy construct as described in “Eliminating Redundant Copies of
Variables in Generated Code” on page 4-8.
When you define variables, they are local by default; they do not persist
between function calls. To make variables persistent, see “Defining and
Initializing Persistent Variables” on page 4-10.
Example: Defining a Variable for Multiple Execution Paths
Consider the following MATLAB code:
...
if c > 0
x = 11;
end
% Later in your code ...
if c > 0
use(x);
end
...
Here, x is assigned only if c > 0 and used only when c > 0. This code
works in MATLAB, but generates a compilation error during code generation
because it detects that x is undefined on some execution paths (when c <= 0),.
To make this code suitable for code generation, define x before using it:
x = 0;
...
if c > 0
x = 11;
end
% Later in your code ...
if c > 0
use(x);
end
...
4-4
Best Practices for Defining Variables for C/C++ Code Generation
Example: Defining All Fields in a Structure
Consider the following MATLAB code:
...
if c > 0
s.a = 11;
disp(s);
else
s.a = 12;
s.b = 12;
end
% Try to use s
use(s);
...
Here, the first part of the if statement uses only the field a, and the else
clause uses fields a and b. This code works in MATLAB, but generates a
compilation error during C/C++ code generation because it detects a structure
type mismatch. To prevent this error, do not add fields to a structure after
you perform certain operations on the structure. For more information, see
Chapter 6, “Code Generation for MATLAB Structures”.
To make this code suitable for C/C++ code generation, define all fields of
s before using it.
...
% Define all fields in structure s
s = struct( a ,0, b , 0);
if c > 0
s.a = 11;
disp(s);
else
s.a = 12;
s.b = 12;
end
% Use s
use(s);
...
4-5
Use Caution When Reassigning Variables

In general, you should adhere to the "one variable/one type" rule for C/C++
code generation; that is, each variable must have a specific class, size and
complexity. Generally, if you reassign variable properties after the initial
assignment, you get a compilation error during code generation, but there are
exceptions, as described in “When You Can Reassign Variable Properties for
C/C++ Code Generation” on page 4-7.
Use Type Cast Operators in Variable Definitions

By default, constants are of type double. To define variables of other types,
you can use type cast operators in variable definitions. For example, the
following code defines variable y as an integer:
...
x = 15; % x is of type double by default.
y = uint8(x); % z has the value of x, but cast to uint8.
...
Define Matrices Before Assigning Indexed Variables

When generating C/C++ code from MATLAB, you cannot grow a variable by
writing into an element beyond its current size. Such indexing operations
produce run-time errors. You must define the matrix first before assigning
values to any of its elements.
For example, the following initial assignment is not allowed for code
generation:
g(3,2) = 14.6; % Not allowed for creating g

% OK for assigning value once created
For more information about indexing matrices, see “Limitations on Matrix

Indexing Operations for Code Generation” on page 8-40.
4-6
When You Can Reassign Variable Properties for C/C++ Code Generation
When You Can Reassign Variable Properties for C/C++

Code Generation
There are certain variables that you can reassign after the initial assignment
with a value of different class, size, or complexity:
Dynamically sized variables
A variable can hold values that have the same class and complexity but
different sizes. If the size of the initial assignment is not constant, the
variable is dynamically sized in generated code. For more information, see
“How Working with Variable-Size Data Is Different for Code Generation”
on page 8-3.
Variables reused in the code for different purposes
You can reassign the type (class, size, and complexity) of a variable after
the initial assignment if each occurrence of the variable can have only one
type. In this case, the variable is renamed in the generated code to create
multiple independent variables. For more information, see “Reusing the Same
Variable with Different Properties” on page 4-11.
4-7
Eliminating Redundant Copies of Variables in Generated

Code
In this section...
“When Redundant Copies Occur” on page 4-8
“How to Eliminate Redundant Copies by Defining Uninitialized Variables”
on page 4-8
“Defining Uninitialized Variables” on page 4-9
When Redundant Copies Occur

During C/C++ code generation, MATLAB checks for statements that attempt
to access uninitialized memory. If it detects execution paths where a variable
is used but is potentially not defined, it generates a compile-time error. To
prevent these errors, define all variables by assignment before using them in
operations or returning them as function outputs.
Note, however, that variable assignments not only copy the properties of the
assigned data to the new variable, but also initialize the new variable to the
assigned value. This forced initialization sometimes results in redundant
copies in C/C++ code. To eliminate redundant copies, define uninitialized
variables by using the coder.nullcopy function, as described in “How to
Eliminate Redundant Copies by Defining Uninitialized Variables” on page
4-8..
How to Eliminate Redundant Copies by Defining

Uninitialized Variables
1 Define the variable with coder.nullcopy.
2 Initialize the variable before reading it.
When the uninitialized variable is an array, you must initialize all of its
elements before passing the array as an input to a function or operator
— even if the function or operator does not read from the uninitialized
portion of the array.
4-8
Eliminating Redundant Copies of Variables in Generated Code
What happens if you access uninitialized data?
Uninitialized memory contains arbitrary values. Therefore, accessing

uninitialized data may lead to segmentation violations or nondeterministic
program behavior (different runs of the same program may yield
inconsistent results).
Defining Uninitialized Variables

In the following code, the assignment statement X = zeros(1,N) not only
defines X to be a 1-by-5 vector of real doubles, but also initializes each element
of X to zero.
function X = fcn %#codegen
N = 5;
X = zeros(1,N);
for i = 1:N
if mod(i,2) == 0
X(i) = i;
else
X(i) = 0;
end
end
This forced initialization creates an extra copy in the generated code. To

eliminate this overhead, use coder.nullcopy in the definition of X:
function X = fcn2 %#codegen
N = 5;
X = coder.nullcopy(zeros(1,N));
for i = 1:N
if mod(i,2) == 0
X(i) = i;
else
X(i) = 0;
end
end
4-9
Defining and Initializing Persistent Variables

Persistent variables are local to the function in which they are defined,
but they retain their values in memory between calls to the function. To
define persistent variables for C/C++ code generation, use the persistent
statement, as in this example:
persistent PROD_X;
The definition should appear at the top of the function body, after the
header and comments, but before the first use of the variable. During code
generation, the value of the persistent variable is initialized to an empty
matrix by default. You can assign your own value after the definition by using
the isempty statement, as in this example:
function findProduct(inputvalue) %#codegen

persistent PROD_X
if isempty(PROD_X)
PROD_X = 1;
end
PROD_X = PROD_X * inputvalue;
end
For more information, see Persistent Variables in the MATLAB Programming

Fundamentals documentation.
4-10
Reusing the Same Variable with Different Properties

In this section...
“When You Can Reuse the Same Variable with Different Properties” on
page 4-11
“When You Cannot Reuse Variables” on page 4-12
“Limitations of Variable Reuse” on page 4-14
When You Can Reuse the Same Variable with

Different Properties
You can reuse (reassign) an input, output, or local variable with different
class, size, or complexity if MATLAB can unambiguously determine the
properties of each occurrence of this variable during C/C++ code generation.
If so, MATLAB creates separate uniquely named local variables in the
generated code. You can view these renamed variables in the code generation
report (see “Viewing Variables in Your MATLAB Code” in the MATLAB
Coder documentation.
A common example of variable reuse is in if-elseif-else or switch-case

statements. For example, the following function example1 first uses the
variable t in an if statement, where it holds a scalar double, then reuses t
outside the if statement to hold a vector of doubles.
function y = example1(u) %#codegen

if all(all(u>0))
% First, t is used to hold a scalar double value
t = mean(mean(u)) / numel(u);
u = u - t;
end
% t is reused to hold a vector of doubles
t = find(u > 0);
y = sum(u(t(2:end-1)));
To compile this example and see how MATLAB renames the reused variable t,
see Variable Reuse in an if Statement on page 4-12.
4-11
When You Cannot Reuse Variables

You cannot reuse (reassign) variables if it is not possible to determine the
class, size, and complexity of an occurrence of a variable unambiguously
during code generation. In this case, variables cannot be renamed and a
compilation error occurs.
For example, the following example2 function assigns a fixed-point value to

x in the if statement and reuses x to store a matrix of doubles in the else
clause. It then uses x after the if-else statement. This function generates a
compilation error because after the if-else statement, variable x can have
different properties depending on which if-else clause executes.
function y = example2(use_fixpoint, data) %#codegen

if use_fixpoint
% x is fixed-point
x = fi(data, 1, 12, 3);
else
% x is a matrix of doubles
x = data;
end
% When x is reused here, it is not possible to determine its
% class, size, and complexity
t = sum(sum(x));
y = t > 0;
end
Variable Reuse in an if Statement
To see how MATLAB renames a reused variable t:
1 Create a MATLAB file example1.m containing the following code.
function y = example1(u) %#codegen

if all(all(u>0))
% First, t is used to hold a scalar double value
t = mean(mean(u)) / numel(u);
u = u - t;
end
% t is reused to hold a vector of doubles
t = find(u > 0);
4-12
y = sum(u(t(2:end-1)));
end
2 Compile example1.
For example, to generate a MEX function, enter:
codegen -o example1x -report example1.m -args {ones(5,5)}
Note codegen requires a MATLAB Coder license.
codegen reports that the compilation was successful. It generates a MEX

function, example1x in the current folder, and provides a link to the code
generation report.
3 Open the code generation report.
4 In the MATLAB code pane of the code generation report, place your pointer
over the variable t inside the if statement.
The code generation report highlights both instances of t in the if

statement because they share the same class, size, and complexity. It
displays the data type information for t at this point in the code. Here,
t is a scalar double.
5 In the MATLAB code pane of the report, place your pointer over the
variable t outside the for-loop.
4-13
This time, the report highlights both instances of t outside the if

statement. The report indicates that t might hold up to 25 doubles. The size
of t is :25, that is, a column vector containing a maximum of 25 doubles.
6 Click the Variables tab to view the list of variables used in example1.
The report displays a list of all the variables in example1. There are two
uniquely named local variables t>1 and t>2.
7 In the list of variables, place your pointer over t>1.
The code generation report highlights both instances of t in the if

statement.
8 In the list of variables, place your pointer over t>2
The code generation report highlights both instances of t outside the if

statement.
Limitations of Variable Reuse

The following variables cannot be renamed in generated code:
• Persistent variables.
• Global variables.
• Variables passed to C code using coder.ref, coder.rref, coder.wref.
• Variables whose size is set using coder.varsize.
• Variables whose names are controlled using coder.cstructname.
• The index variable of a for-loop when it is used inside the loop body.
4-14
• The block outputs of a MATLAB Function block in a Simulink model.

• Chart-owned variables of a MATLAB function in a Stateflow chart.
4-15
Supported Variable Types

You can use the following data types for C/C++ code generation from
MATLAB:
Type Description
char Character array (string)
complex Complex data. Cast function takes real and imaginary
components
double Double-precision floating point
int8, int16, int32 Signed integer
logical Boolean true or false
single Single-precision floating point
struct Structure (see Chapter 6, “Code Generation for
MATLAB Structures”)
uint8, uint16, Unsigned integer
uint32
Fixed-point See “Code Acceleration and Code Generation
from MATLAB for Fixed-Point Algorithms” in the
Fixed-Point Toolbox User’s Guide documentation.
4-16
5
Defining Data for Code

Generation
• “How Working with Data is Different for Code Generation” on page 5-2
• “Code Generation for Complex Data” on page 5-4
• “Code Generation for Characters” on page 5-6
5 Defining Data for Code Generation
How Working with Data is Different for Code Generation

To generate efficient standalone code, you must use the following types and
classes of data differently than you normally would when running your code
in the MATLAB environment:
Data What’s Different More Information

Complex numbers • Complexity of “Code Generation for
variables must be set Complex Data” on page
at time of assignment 5-4
and before first use
• Expressions
containing a complex
number or variable
always evaluate to a
complex result, even
if the result is zero
Characters Restricted to 8 bits of “Code Generation for
precision Characters” on page 5-6
Enumerated data • Supports Chapter 7, “Code
integer-based Generation for
enumerated types Enumerated Data”
only
• Restricted use in
switch statements
and for-loops
Function handles • Function handles Chapter 9, “Code
must be scalar values Generation for Function
Handles”
• Same bound variable
cannot reference
different function
handles
• Cannot pass function
handles to or from
primary or extrinsic
functions
5-2
How Working with Data is Different for Code Generation
Data What’s Different More Information
• Cannot view function

handles from the
debugger
5-3
Code Generation for Complex Data

In this section...
“Restrictions When Defining Complex Variables” on page 5-4
“Expressions Containing Complex Operands Yield Complex Results” on
page 5-5
Restrictions When Defining Complex Variables

For code generation, you must set the complexity of variables at the time of
assignment, either by assigning a complex constant or using the complex
function, as in these examples:
x = 5 + 6i; % x is a complex number by assignment.

y = 7 + 8j; % y is a complex number by assignment.
x = complex(5,6); % x is the complex number 5 + 6i.
Once you set the type and size of a variable, you cannot cast it to another
type or size. In the following example, the variable x is defined as complex
and stays complex:
x = 1 + 2i; % Defines x as a complex variable.

y = int16(x); % Real and imaginary parts of y are int16.
x = 3; % x now has the value 3 + 0i.
Mismatches can also occur when you assign a real operand the complex result
of an operation:
z = 3; % Sets type of z to double (real)

z = 3 + 2i; % ERROR: cannot recast z to complex
As a workaround, set the complexity of the operand to match the result

of the operation:
m = complex(3); % Sets m to complex variable of value 3 + 0i

m = 5 + 6.7i; % Assigns a complex result to a complex number
5-4
Code Generation for Complex Data
Expressions Containing Complex Operands Yield

Complex Results
In general, expressions that contain one or more complex operands always
produce a complex result in generated code, even if the value of the result is
zero. Consider the following example:
x = 2 + 3i;
y = 2 - 3i;
z = x + y; % z is 4 + 0i.
In MATLAB, this code generates the real result z = 4. However, during

code generation, the types for x and y are known, but their values are not.
Because either or both operands in this expression are complex, z is defined
as a complex variable requiring storage for both a real and an imaginary
part. This means that z equals the complex result 4 + 0i in generated code,
not 4 as in MATLAB code.
There are two exceptions to this behavior:
• Functions that take complex arguments, but produce real results
y = real(x); % y is the real part of the complex number x.

y = imag(x); % y is the real-valued imaginary part of x.
y = isreal(x); % y is false (0) for a complex number x.
• Functions that take real arguments, but produce complex results:
z = complex(x,y); % z is a complex number for a real x and y.
5-5
Code Generation for Characters

The complete set of Unicode® characters is not supported for code generation.
Characters are restricted to 8 bits of precision in generated code. Because
many mathematical operations require more than 8 bits of precision, it is
recommended that you do not perform arithmetic with characters if you
intend to generate code from your MATLAB algorithm.
5-6
6
Code Generation for

MATLAB Structures
• “How Working with Structures Is Different for Code Generation” on page

6-2
• “Structure Operations Allowed for Code Generation” on page 6-3
• “Defining Scalar Structures for Code Generation” on page 6-4
• “Defining Arrays of Structures for Code Generation” on page 6-7
• “Making Structures Persistent” on page 6-9
• “Indexing Substructures and Fields” on page 6-10
• “Assigning Values to Structures and Fields” on page 6-12
• “Passing Large Structures as Input Parameters” on page 6-13
6 Code Generation for MATLAB® Structures
How Working with Structures Is Different for Code

Generation
To generate efficient standalone code for structures, you must define and use
structures differently than you normally would when running your code in
the MATLAB environment:
What’s Different More Information

Use a restricted set of operations. “Structure Operations Allowed for
Code Generation” on page 6-3
Observe restrictions on properties “Defining Scalar Structures for Code
and values of scalar structures. Generation” on page 6-4
Make structures uniform in arrays. “Defining Arrays of Structures for
Code Generation” on page 6-7
Reference structure fields “Indexing Substructures and Fields”
individually during indexing. on page 6-10
Avoid type mismatch when assigning “Assigning Values to Structures and
values to structures and fields. Fields” on page 6-12
For an introduction to working with structures in MATLAB, see “Structures”

in the MATLAB Getting Started documentation.
6-2
Structure Operations Allowed for Code Generation
Structure Operations Allowed for Code Generation

To generate efficient standalone code for MATLAB structures, you are
restricted to the following operations:
• Define structures as local and persistent variables by assignment and

using the struct function
• Index structure fields using dot notation
• Define primary function inputs as structures
• Pass structures to subfunctions
6-3
Defining Scalar Structures for Code Generation

In this section...
“Restrictions When Using struct” on page 6-4
“Restrictions When Defining Scalar Structures by Assignment” on page 6-4
“Adding Fields in Consistent Order on Each Control Flow Path” on page 6-4
“Restriction on Adding New Fields After First Use” on page 6-5
Restrictions When Using struct

When you use the struct function to create scalar structures for code
generation, the following restrictions apply:
• Field arguments must be scalar values.

• You cannot create structures of cell arrays.
Restrictions When Defining Scalar Structures by

Assignment
When you define a scalar structure by assigning a variable to a preexisting
structure, you do not need to define the variable before the assignment.
However, if you already defined that variable, it must have the same class,
size, and complexity as the structure you assign to it. In the following
example, p is defined as a structure that has the same properties as the
predefined structure S:
...
S = struct('a', 0, 'b', 1, 'c', 2);
p = S;
...
Adding Fields in Consistent Order on Each Control

Flow Path
When you create a structure, you must add fields in the same order on each
control flow path. For example, the following code generates a compiler
error because it adds the fields of structure x in a different order in each
if statement clause:
6-4
Defining Scalar Structures for Code Generation
function y = fcn(u) %#codegen

if u > 0
x.a = 10;
x.b = 20;
else
x.b = 30; % Generates an error (on variable x)
x.a = 40;
end
y = x.a + x.b;
In this example, the assignment to x.a comes before x.b in the first if
statement clause, but the assignments appear in reverse order in the else
clause. Here is the corrected code:

if u > 0
x.a = 10;
x.b = 20;
else
x.a = 40;
x.b = 30;
end
y = x.a + x.b;
Restriction on Adding New Fields After First Use

You cannot add fields to a structure after you perform any of the following
operations on the structure:
• Reading from the structure

• Indexing into the structure array
• Passing the structure to a function
For example, consider this code:
...
x.c = 10; % Defines structure and creates field c
y = x; % Reads from structure
x.d = 20; % Generates an error
...
6-5
In this example, the attempt to add a new field d after reading from structure
x generates an error.
This restriction extends across the structure hierarchy. For example, you
cannot add a field to a structure after operating on one of its fields or nested
structures, as in this example:
x.c = 10;
y = x.c;
x.d = 20; % Generates an error
In this example, the attempt to add a new field d to structure x after reading
from the structure’s field c generates an error.
6-6
Defining Arrays of Structures for Code Generation
Defining Arrays of Structures for Code Generation

In this section...
“Ensuring Consistency of Fields” on page 6-7
“Using repmat to Define an Array of Structures with Consistent Field
Properties” on page 6-7
“Defining an Array of Structures Using Concatenation” on page 6-8
Ensuring Consistency of Fields

When you create an array of MATLAB structures with the intent of
generating code, you must be sure that each structure field in the array has
the same size, type, and complexity.
Using repmat to Define an Array of Structures with

Consistent Field Properties
You can create an array of structures from a scalar structure by using the
MATLAB repmat function, which replicates and tiles an existing scalar
structure:
1 Create a scalar structure, as described in “Defining Scalar Structures for

Code Generation” on page 6-4.
2 Call repmat, passing the scalar structure and the dimensions of the array.
3 Assign values to each structure using standard array indexing and

structure dot notation.
For example, the following code creates X, a 1-by-3 array of scalar structures.
Each element of the array is defined by the structure s, which has two fields,
a and b:
...
s.a = 0;
s.b = 0;
X = repmat(s,1,3);
X(1).a = 1;
6-7
X(2).a = 2;
X(3).a = 3;
X(1).b = 4;
X(2).b = 5;
X(3).b = 6;
...
Defining an Array of Structures Using Concatenation

To create a small array of structures, you can use the concatenation operator,
square brackets ( [ ] ), to join one or more structures into an array (see
“Concatenating Matrices” in the MATLAB Mathematics documentation). For
code generation, all the structures that you concatenate must have the same
size, class, and complexity.
For example, the following code uses concatenation and a subfunction to

create the elements of a 1-by-3 structure array:
...
W = [ sab(1,2) sab(2,3) sab(4,5) ];
function s = sab(a,b)
s.a = a;
s.b = b;
...
6-8
Making Structures Persistent
Making Structures Persistent

To make structures persist, you define them to be persistent variables and
initialize them with the isempty statement, as described in “Defining and
Initializing Persistent Variables” on page 4-10.
For example, the following function defines structure X to be persistent and

initializes its fields a and b:
function f(u) %#codegen

persistent X
if isempty(X)
X.a = 1;
X.b = 2;
end
6-9
Indexing Substructures and Fields

Use these guidelines when indexing substructures and fields for code
generation:
Reference substructure field values individually using dot notation
For example, the following MATLAB code uses dot notation to index fields
and substructures:
...
substruct1.a1 = 15.2;
substruct1.a2 = int8([1 2;3 4]);
mystruct = struct('ele1',20.5,'ele2',single(100),
'ele3',substruct1);
substruct2 = mystruct;
substruct2.ele3.a2 = 2*(substruct1.a2);
...
The generated code indexes elements of the structures in this example by

resolving symbols as follows:
Dot Notation Symbol Resolution

substruct1.a1 Field a1 of local structure substruct1
substruct2.ele3.a1 Value of field a1 of field ele3, a substructure of local structure
substruct2
substruct2.ele3.a2(1,1) Value in row 1, column 1 of field a2 of field ele3, a substructure
of local structure substruct2
Reference field values individually in structure arrays
To reference the value of a field in a structure array, you must index into the
array to the structure of interest and then reference that structure’s field
individually using dot notation, as in this example:
...
6-10
Indexing Substructures and Fields
y = X(1).a % Extracts the value of field a

% of the first structure in array X
...
To reference all the values of a particular field for each structure in an array,
use this notation in a for loop, as in this example:
...
s.a = 0;
s.b = 0;
X = repmat(s,1,5);
for i = 1:5
X(i).a = i;
X(i).b = i+1;
end
This example uses the repmat function to define an array of structures, each
with two fields a and b as defined by s. See “Defining Arrays of Structures for
Code Generation” on page 6-7 for more information.
Do not reference fields dynamically
You cannot reference fields in a structure by using dynamic names, which

express the field as a variable expression that MATLAB evaluates at run time
(see “Dynamic Field Names” in the MATLAB Getting Started Guide).
6-11
Assigning Values to Structures and Fields

Use these guidelines when assigning values to a structure, substructure,
or field for code generation:
Field properties must be consistent across structure-to-structure

assignments
If: Then:
Assigning one structure to another Define each structure with the same
structure. number, type, and size of fields.
Assigning one structure to a Define the structure with the same
substructure of a different structure number, type, and size of fields as
and vice versa. the substructure.
Assigning an element of one The elements must have the same
structure to an element of another type and size.
structure.
Do not use field values as constants
The values stored in the fields of a structure are not treated as constant values
in generated code. Therefore, you cannot use field values to set the size or
class of other data. For example, the following code generates a compiler error:
...
Y.a = 3;
X = zeros(Y.a); % Generates an error
In this example, even though you set field a of structure Y to the value 3, Y.a
is not a constant in generated code and, therefore, it is not a valid argument
to pass to the function zeros.
Do not assign mxArrays to structures
You cannot assign mxArrays to structure elements; convert mxArrays to

known types before code generation (see “Working with mxArrays” on page
11-16).
6-12
Passing Large Structures as Input Parameters
Passing Large Structures as Input Parameters

If you generate a MEX function for a MATLAB function that takes a large
structure as an input parameter, for example, a structure containing fields
that are matrices, the MEX function might fail to load. This load failure
occurs because, when you generate a MEX function from a MATLAB function
that has input parameters, the code generation software allocates memory for
these input parameters on the stack. To avoid this issue, pass the structure
by reference to the MATLAB function. For example, if the original function
signature is:
y = foo(a, S)
where S is the structure input, rewrite the function to:
[y, S] = foo(a, S)
6-13
6-14
7
Code Generation for

Enumerated Data
• “How Working with Enumerated Data Is Different for Code Generation”

on page 7-2
• “Enumerated Types Supported for Code Generation” on page 7-3
• “When to Use Enumerated Data for Code Generation” on page 7-6
• “Workflows for Using Enumerated Data for Code Generation” on page 7-7
• “How to Define Enumerated Data for Code Generation” on page 7-9
• “How to Instantiate Enumerated Types for Code Generation” on page 7-11
• “How to Generate Code for Enumerated Data” on page 7-12
• “Defining and Using Enumerated Types for Code Generation” on page 7-13
• “Operations on Enumerated Data Allowed for Code Generation” on page
7-15
• “Using Enumerated Data in Control Flow Statements” on page 7-18
• “Restrictions on Use of Enumerated Data in for-Loops” on page 7-24
• “Toolbox Functions That Support Enumerated Types for Code Generation”
on page 7-25
7 Code Generation for Enumerated Data
How Working with Enumerated Data Is Different for Code

Generation
To generate efficient standalone code for enumerated data, you must define
and use enumerated types differently than you normally would when running
your code in the MATLAB environment:
What’s Different More Information

Supports integer-based enumerated “Enumerated Types Supported for
types only Code Generation” on page 7-3
Name of each enumerated data type “Naming Enumerated Types for
must be unique Code Generation” on page 7-10
Each enumerated data type must “How to Define Enumerated Data
be defined in a separate file on the for Code Generation” on page 7-9
MATLAB path and “How to Generate Code for
Enumerated Data” on page 7-12
Restricted set of operations “Operations on Enumerated Data
Allowed for Code Generation” on
page 7-15
Restricted use in for-loops “Restrictions on Use of Enumerated
Data in for-Loops” on page 7-24
See Also
• “Defining and Organizing Classes” in the MATLAB Object-Oriented
Programming documentation for more information about defining
MATLAB classes
• “Enumerations and Modeling” for more information about enumerated
types based on Simulink.IntEnumType
7-2
Enumerated Types Supported for Code Generation

In this section...
“Enumerated Type Based on int32” on page 7-3
“Enumerated Type Based on Simulink.IntEnumType” on page 7-4
Enumerated Type Based on int32

This enumerated data type is based on the built-in type int32. Use this
enumerated type when generating code from MATLAB algorithms.
Syntax
classdef(Enumeration) type_name < int32
Example
classdef(Enumeration) PrimaryColors < int32

enumeration
Red(1),
Blue(2),
Yellow(4)
end
end
In this example, the statement classdef(Enumeration) PrimaryColors

< int32 means that the enumerated type PrimaryColors is based on the
built-in type int32. As such, PrimaryColors inherits the characteristics
of the int32 type, as well as defining its own unique characteristics. For
example, PrimaryColors is restricted to three enumerated values:
Enumerated Value Enumerated Name Underlying Numeric

Value
Red(1) Red 1
Blue(2) Blue 2
Yellow(4) Yellow 4
7-3
How to Use
Define enumerated data in MATLAB code and compile the source file. For
example, to generate C/C++ code from your MATLAB source, you can use
codegen, as described in “Workflow for Generating Code for Enumerated
Data from MATLAB Algorithms” on page 7-7.
Note codegen requires a MATLAB Coder license.
Enumerated Type Based on Simulink.IntEnumType

This enumerated data type is based on the built-in type
Simulink.IntEnumType, which is available with a Simulink
license. Use this enumerated type when exchanging enumerated data with
Simulink blocks and Stateflow charts.
Syntax
classdef(Enumeration) type_name < Simulink.IntEnumType
Example
classdef(Enumeration) myMode < Simulink.IntEnumType
enumeration
OFF(0)
ON(1)
end
end
How to Use
Here are the basic guidelines for using enumerated data based on
Simulink.IntEnumType:
7-4
Application What to Do
When exchanging Define enumerated data in MATLAB
enumerated data with Function blocks in Simulink models.
Simulink blocks Requires Simulink software.
When exchanging Define enumerated data in MATLAB
enumerated data with functions in Stateflow charts. Requires
Stateflow charts Simulink and Stateflow software.
For more information, see:
• “Using Enumerated Data in MATLAB Function Blocks” in the Simulink

documentation
• “Using Enumerated Data in Stateflow Charts” in the Stateflow
documentation
7-5
When to Use Enumerated Data for Code Generation

You can use enumerated types to represent program states and to control
program logic, especially when you need to restrict data to a finite set of
values and refer to these values by name. Even though you can sometimes
achieve these goals by using integers or strings, enumerated types offer the
following advantages:
• Provide more readable code than integers

• Allow more robust error checking than integers or strings
For example, if you mistype the name of an element in the enumerated
type, you get a compile-time error that the element does not belong to the
set of allowable values.
• Produce more efficient code than strings
For example, comparisons of enumerated values execute faster than
comparisons of strings.
7-6
Workflows for Using Enumerated Data for Code Generation
Workflows for Using Enumerated Data for Code

Generation
In this section...
“Workflow for Generating Code for Enumerated Data from MATLAB
Algorithms” on page 7-7
“Workflow for Generating Code for Enumerated Data from MATLAB
Function Blocks” on page 7-8
Workflow for Generating Code for Enumerated Data

from MATLAB Algorithms
Step Action How?
Define an enumerated data type See “How to Define Enumerated
1 that inherits from int32. Data for Code Generation” on
page 7-9.
Instantiate the enumerated type See “How to Instantiate
2 in your MATLAB algorithm. Enumerated Types for Code
Generation” on page 7-11.
Compile the function with See “How to Generate Code for
3 codegen. Enumerated Data” on page 7-12.
This workflow requires a MATLAB Coder license.
7-7
Workflow for Generating Code for Enumerated Data

from MATLAB Function Blocks
Step Action How?
Define an enumerated data See “How to Define Enumerated
1 type that inherits from Data Types for MATLAB
Simulink.IntEnumType. Function Blocks” in the Simulink
documentation.
Add the enumerated data to your See “How to Add Enumerated
2 MATLAB Function block. Data to MATLAB Function
Blocks” in the Simulink
documentation.
Instantiate the enumerated type See “How to Instantiate
3 in your MATLAB Function block. Enumerated Data in MATLAB
Function Blocks” in the Simulink
documentation.
Simulate and/or generate code. See “Enumerations” in the
4 Simulink Coder documentation.
This workflow requires the following licenses:
• Simulink (for simulation)

• MATLAB Coder and Simulink Coder (for code generation)
7-8
How to Define Enumerated Data for Code Generation
How to Define Enumerated Data for Code Generation

Follow these to define enumerated data for code generation from MATLAB
algorithms:
1 Create a class definition file.
In the MATLAB Command Window, select File > New > Class.
2 Enter the class definition as follows:
classdef(Enumeration) EnumTypeName < int32
For example, the following code defines an enumerated type called sysMode:
classdef(Enumeration) sysMode < int32

...
end
EnumTypeName is a case-sensitive string that must be unique among data

type names and workspace variable names. It must inherit from the
built-in type int32.
3 Define enumerated values in an enumeration section as follows:
classdef(Enumeration) EnumTypeName < int32

enumeration
EnumName(N)
...
end
end
For example, the following code defines a set of two values for enumerated
type sysMode:

enumeration
OFF(0)
ON(1)
end
end
7-9
An enumerated type can define any number of values. Each enumerated

value consists of a string EnumName and an underlying integer N. Each
EnumName must be unique within its type, but can also appear in other
enumerated types. The underlying integers need not be either consecutive
or ordered, nor must they be unique within the type or across types.
4 Save the file on the MATLAB path.
The name of the file must match the name of the enumerated data type.
The match is case sensitive.
To add a folder to the MATLAB search path, type addpath pathname

at the MATLAB command prompt. For more information, see “Using
the MATLAB Search Path”, addpath, and savepath in the MATLAB
documentation.
For examples of enumerated data type definitions, see “Class Definition:

sysMode” on page 7-13 and “Class Definition: LEDcolor” on page 7-14.
Naming Enumerated Types for Code Generation

You must use a unique name for each enumerated data type. The name of an
enumerated data type cannot match the name of a toolbox function supported
for code generation, or another data type or a variable in the MATLAB base
workspace. Otherwise, a name conflict occurs.
For example, you cannot name an enumerated data type mode because
MATLAB for code generation provides a toolbox function of the same name.
For a list of toolbox functions supported for code generation, see Chapter 2,
“Functions Supported for Code Generation”.
7-10
How to Instantiate Enumerated Types for Code Generation
How to Instantiate Enumerated Types for Code Generation

To instantiate an enumerated type for code generation from MATLAB
algorithms, use dot notation to specify ClassName.EnumName. For example,
the following function displayState instantiates the enumerated types
sysMode and LEDcolor from “Defining and Using Enumerated Types for Code
Generation” on page 7-13. The dot notation appears highlighted in the code.
function led = displayState(state)

%#codegen
if state == sysMode.ON
led = LEDcolor.GREEN;
else
led = LEDcolor.RED;
end
7-11
How to Generate Code for Enumerated Data

Use the command codegen to generate MEX, C, or C++ code from the
MATLAB algorithm that contains the enumerated data (requires a MATLAB
Coder license). Each enumerated data type must be defined on the MATLAB
path in a separate file as a class derived from the built-in type int32. See
“How to Define Enumerated Data for Code Generation” on page 7-9.
If your function has inputs, you must specify the properties of these inputs
to codegen. For an enumerated data input, use the -args option to pass
one of its allowable values as a sample value. For example, the following
codegen command specifies that the function displayState takes one input
of enumerated data type sysMode.
codegen displayState -args {sysMode.ON}
After executing this command, codegen generates a platform-specific MEX

function that you can test in MATLAB. For example, to test displayState,
type the following command:
displayState(sysMode.OFF)
You should get the following result:
ans =
RED
See Also
• MATLAB Coder documentation to learn more about codegen
• “Defining and Using Enumerated Types for Code Generation” on page 7-13
for a description of the example function displayState and its enumerated
type definitions
7-12
Defining and Using Enumerated Types for Code Generation
Defining and Using Enumerated Types for Code Generation

In this section...
“About the Example” on page 7-13
“Class Definition: sysMode” on page 7-13
“Class Definition: LEDcolor” on page 7-14
“Function: displayState” on page 7-14
About the Example

The following example appears throughout this section to illustrate how
to define and use enumerated types for code generation. The function,
displayState uses enumerated data to represent the modes of a device that
controls the colors of an LED display.
Before using enumerated data, you must define your enumerated data types
as MATLAB classes that inherit from the built-in type int32. Each class
definition must reside in a separate file on the MATLAB path. This example
uses two enumerated types: sysMode to represent the set of allowable modes
and LEDcolor to represent the set of allowable colors.
See Also
• “Workflows for Using Enumerated Data for Code Generation” on page 7-7
• “How to Instantiate Enumerated Types for Code Generation” on page 7-11
• “How to Generate Code for Enumerated Data” on page 7-12
Class Definition: sysMode

Here is the class definition of the sysMode enumerated data type:

enumeration
OFF(0)
ON(1)
7-13
end
end
This definition must reside on the MATLAB path in a file with the same
name as the class, sysMode.m.
Class Definition: LEDcolor

Here is the class definition of the LEDcolor enumerated data type:
classdef(Enumeration) LEDcolor < int32

enumeration
GREEN(1),
RED(2),
end
end
This definition must reside on the MATLAB path in a file called LEDcolor.m.
Function: displayState
The following function displayState uses enumerated data to activate an
LED display, based on the state of a device. It lights a green LED display to
indicate the ON state and lights a red LED display to indicate the OFF state.

%#codegen
else
led = LEDcolor.RED;
end
This function models a simple control.
The compiler directive %#codegen indicates that you intend to generate

code from the MATLAB algorithm. See “Adding the Compilation Directive
%#codegen” on page 11-8.
7-14
Operations on Enumerated Data Allowed for Code Generation
Operations on Enumerated Data Allowed for Code

Generation
To generate efficient standalone code for enumerated data, you are restricted
to the following operations. The examples are based on the definitions of
the enumeration type LEDcolor described in “Class Definition: LEDcolor”
on page 7-14.
Assignment Operator, =
Example Result
xon = LEDcolor.GREEN xon =

xoff = LEDcolor.RED
GREEN
xoff =
RED
Relational Operators, < > <= >= == ~=

Example Result
xon == xoff ans =
xon <= xoff ans =
xon > xoff ans =
7-15
Cast Operation
Example Result
double(LEDcolor.RED) ans =
z = 2 z =
y = LEDcolor(z)
2
y =
RED
Indexing Operation
Example Result
m = [1 2] m =
n = LEDcolor(m)
p = n(LEDcolor.GREEN) 1 2
n =
GREEN RED
p =
GREEN
7-16
Operations on Enumerated Data Allowed for Code Generation
Control Flow Statements: if, switch, while

Statement Example Executable
Example
if “Using the if
if state == sysMode.ON Statement on
led = LEDcolor.GREEN; Enumerated Data
else Types” on page
led = LEDcolor.RED; 7-18
end
switch “Using the switch
switch button Statement on
case VCRButton.Stop Enumerated Data
state = VCRState.Stop; Types” on page
case VCRButton.PlayOrPause 7-19
state = VCRState.Play;
case VCRButton.Next
state = VCRState.Forward;
case VCRButton.Previous
state = VCRState.Rewind;
otherwise
state = VCRState.Stop;
end
while “Using the while
while state ~= State.Ready Statement on
switch state Enumerated Data
case State.Standby Types” on page
initialize(); 7-22
state = State.Boot;
case State.Boot
boot();
state = State.Ready;
end
end
7-17
Using Enumerated Data in Control Flow Statements

The following control statements work with enumerated operands in
generated code. However, there are restrictions (see “Restrictions on Use of
Enumerated Data in for-Loops” on page 7-24).
Using the if Statement on Enumerated Data Types

This example is based on the definition of the enumeration types LEDcolor
and sysMode. The function displayState uses these enumerated data types
to activate an LED display.
Class Definition: sysMode

enumeration
OFF(0)
ON(1)
end
end
name as the class, sysMode.m.
Class Definition: LEDcolor
classdef(Enumeration) LEDcolor < int32

enumeration
GREEN(1),
RED(2),
end
end
This definition must reside on the MATLAB path in a file called LEDcolor.m.
MATLAB Function: displayState

This function uses enumerated data to activate an LED display, based on the
state of a device. It lights a green LED display to indicate the ON state and
lights a red LED display to indicate the OFF state.
7-18

%#codegen
else
led = LEDcolor.RED;
end
Build and Test a MEX Function for displayState
1 Generate a MEX function for displayState. Use the -args option to pass
one of the allowable values for the enumerated data input as a sample
value.
codegen displayState -args {sysMode.ON}
2 Test the function. For example,
displayState(sysMode.OFF)
ans =
RED
Using the switch Statement on Enumerated Data

Types
This example is based on the definition of the enumeration types VCRState
and VCRButton. The function VCR uses these enumerated data types to set
the state of the VCR.
Class Definition: VCRState

classdef(Enumeration) VCRState < int32
enumeration
Stop(0),
Pause(1),
7-19
Play(2),
Forward(3),
Rewind(4)
end
end
name as the class, VCRState.m.
Class Definition: VCRButton
classdef(Enumeration) VCRButton < int32

enumeration
Stop(1),
PlayOrPause(2),
Next(3),
Previous(4)
end
end
This definition must reside on the MATLAB path in a file with the same name
as the class, VCRButton.m.
MATLAB Function: VCR

This function uses enumerated data to set the state of a VCR, based on the
initial state of the VCR and the state of the VCR button.
function s = VCR(button)
%#codegen
persistent state
if isempty(state)
state = VCRState.Stop;
end
switch state
case {VCRState.Stop, VCRState.Forward, VCRState.Rewind}
state = handleDefault(button);
7-20
case VCRState.Play
switch button
case VCRButton.PlayOrPause, state = VCRState.Pause;
otherwise, state = handleDefault(button);
end
case VCRState.Pause
switch button
case VCRButton.PlayOrPause, state = VCRState.Play;
otherwise, state = handleDefault(button);
end
end
s = state;
function state = handleDefault(button)

switch button
case VCRButton.Stop, state = VCRState.Stop;
case VCRButton.PlayOrPause, state = VCRState.Play;
case VCRButton.Next, state = VCRState.Forward;
case VCRButton.Previous, state = VCRState.Rewind;
otherwise, state = VCRState.Stop;
end
Build and Test a MEX Function for VCR
1 Generate a MEX function for VCR. Use the -args option to pass one of the
allowable values for the enumerated data input as a sample value.
codegen -args {VCRButton.Stop} VCR
s = VCR(VCRButton.Stop)
s =
Stop
7-21
Using the while Statement on Enumerated Data Types

This example is based on the definition of the enumeration type State. The
function Setup uses this enumerated data type to set the state of a device.
Class Definition: State

classdef(Enumeration) State < int32
enumeration
Standby(0),
Boot(1),
Ready(2)
end
end
name as the class, State.m.
MATLAB Function: Setup

The following function Setup uses enumerated data to set the state of a device.
function s = Setup(initState)
%#codegen
state = initState;
if isempty(state)
state = State.Standby;
end
while state ~= State.Ready

switch state
case State.Standby
initialize();
state = State.Boot;
case State.Boot
boot();
state = State.Ready;
end
end
7-22
s = state;
function initialize()
% Perform initialization.
function boot()
% Boot the device.
Build and Test a MEX Executable for Setup
1 Generate a MEX executable for Setup. Use the -args option to pass one of
the allowable values for the enumerated data input as a sample value.
codegen Setup -args {State.Standby}
s = Setup(State.Standby)
s =
Ready
7-23
Restrictions on Use of Enumerated Data in for-Loops

Do not use enumerated data as the loop counter variable in for-
loops
To iterate over a range of enumerated data with consecutive values, you can
cast the enumerated data to int32 in the loop counter.
For example, suppose you define an enumerated type ColorCodes as follows:
classdef(Enumeration) ColorCodes < int32

enumeration
Red(1),
Blue(2),
Green(3)
Yellow(4)
Purple(5)
end
end
Because the enumerated values are consecutive, you can use ColorCodes
data in a for-loop like this:
...
for i = int32(ColorCodes.Red):int32(ColorCodes.Purple)
c = ColorCodes(i);
...
end
7-24
Toolbox Functions That Support Enumerated Types for Code Generation
Toolbox Functions That Support Enumerated Types for

Code Generation
The following MATLAB toolbox functions support enumerated types for code
generation:
• cast
• cat
• circshift
• flipdim
• fliplr
• flipud
• histc
• ipermute
• isequal
• isequalwithequalnans
• isfinite
• isinf
• isnan
• issorted
• length
• permute
• repmat
• reshape
• rot90
• shiftdim
• sort
• sortrows
7-25
• squeeze
7-26
8
Code Generation for

Variable-Size Data
• “What Is Variable-Size Data?” on page 8-2

• “How Working with Variable-Size Data Is Different for Code Generation”
on page 8-3
• “Bounded Versus Unbounded Variable-Size Data” on page 8-4
• “When to Use Dynamic Allocation for Variable-Size Data” on page 8-5
• “How to Generate Code for MATLAB Functions with Variable-Size Data”
on page 8-6
• “Tutorial: Generating MEX Code for a MATLAB Function That Expands a
Vector in a Loop” on page 8-9
• “Variable-Size Data in Code Generation Reports” on page 8-18
• “Defining Variable-Size Data for Code Generation” on page 8-20
• “Specifying Upper Bounds for Variable-Size Data” on page 8-27
• “C Code Interface for Unbounded Arrays and Structure Fields” on page 8-30
• “Troubleshooting Issues with Variable-Size Data” on page 8-33
• “Limitations with Variable-Size Support for Code Generation” on page 8-37
• “Restrictions on Variable Sizing in Toolbox Functions Supported for Code
Generation” on page 8-42
8 Code Generation for Variable-Size Data
What Is Variable-Size Data?

Variable-size data is data whose size can change at run time. By contrast,
fixed-size data is data whose size is known and locked at compile time and,
therefore, cannot change at run time.
For example, in the following MATLAB function nway, B is a variable-size

array; its length is not known at compile time.
function B = nway(A,n)
% Compute average of every N elements of A and put them in B.
if ((mod(numel(A),n) == 0) && (n>=1 && n<=numel(A)))
B = ones(1,numel(A)/n);
k = 1;
for i = 1 : numel(A)/n
B(i) = mean(A(k + (0:n-1)));
k = k + n;
end
else
error('n <= 0 or does not divide number of elements evenly');
end
8-2
How Working with Variable-Size Data Is Different for Code Generation
How Working with Variable-Size Data Is Different for

Code Generation
In the MATLAB language, all data can vary in size. By contrast, the
semantics of generated code constrains the class, complexity, and shape of
every expression, variable, and structure field. Therefore, for code generation,
you must use each variable consistently. Each variable must:
• Be either complex or real (determined at first assignment)

• Have a consistent shape
For fixed-size data, the shape is the same as the size returned in MATLAB.
For example, if size(A) == [4 5], the shape of variable A is 4 x 5.
For variable-size data, the shape can be abstract. That is, one or more
dimensions can be unknown (such as 4 x ? or ? x ?).
By default, the compiler detects code logic that attempts to change these fixed
attributes after initial assignments, and flags these occurrences as errors
during code generation. However, you can override this behavior by defining
variables or structure fields as variable-size data. You can then generate
standalone code for bounded and unbounded variable-size data.
See Also
• “Bounded Versus Unbounded Variable-Size Data” on page 8-4
• “Related Products That Support Code Generation from MATLAB” on page
1-12
8-3
Bounded Versus Unbounded Variable-Size Data

You can generate code for bounded and unbounded variable-size data.
Bounded variable-size data has fixed upper bounds; this data can be allocated
statically on the stack or dynamically on the heap. Unbounded variable-size
data does not have fixed upper bounds; this data must be allocated on the
heap.
By default during code generation, the compiler allocates memory on the stack
for variable-size data and performs a sophisticated analysis to determine or
calculate upper bounds. If you use unbounded data, you must enable dynamic
memory allocation so the compiler:
• Does not check for upper bounds

• Allocates memory on the heap instead of the stack
There are performance trade-offs between allocating variable-size data on the

heap versus the stack. For guidelines, see “When to Use Dynamic Allocation
for Variable-Size Data” on page 8-5.
8-4
When to Use Dynamic Allocation for Variable-Size Data
When to Use Dynamic Allocation for Variable-Size Data

When you enable dynamic memory allocation, all of your variable-size data
is allocated on the heap, whether bounded or unbounded. For example, you
cannot selectively allocate bounded data on the stack and unbounded data
on the heap. Therefore, follow these guidelines for when to use dynamic
allocation:
If: Then: Rationale

Your code uses Enable dynamic Unbounded data
both bounded and memory allocation must be dynamically
unbounded data allocated.
Analysis during code Either specify upper You cannot use static
generation fails to bounds explicitly memory allocation for
determine upper or enable dynamic unbounded data.
bounds or calculates memory allocation
upper bounds that are
not precise enough for
your application
You use large, Enable dynamic Prevent stack overflow.
variable-size data sets memory allocation
whose upper bounds
could grow too large for
the stack
You use small Disable dynamic Dynamic memory
variable-size data sets memory allocation allocation is an
or data that does not expensive operation;
change size at run time the performance cost
may be too high for
small data sets.
See Also
• “Specifying Upper Bounds for Variable-Size Data” on page 8-27.
8-5
How to Generate Code for MATLAB Functions with

Variable-Size Data
Here is a basic workflow that recommends first generating MEX code for
verifying the generated code and then generating standalone code after you
are satisfied with the result of the prototype. Code generation requires a
MATLAB Coder license.
To work through these steps with a simple example, see “Tutorial: Generating
MEX Code for a MATLAB Function That Expands a Vector in a Loop” on
page 8-9
1 In the MATLAB Editor, add the compilation directive %#codegen at the

top of your function.
This directive:
• Indicates that you intend to generate code for the MATLAB algorithm
• Turns on checking in the MATLAB Code Analyzer to detect potential
errors during code generation
2 Address issues detected by the Code Analyzer.
In some cases, the MATLAB Code Analyzer warns you when your code
assigns data a fixed size but later grows the data, such as by assignment
or concatenation in a loop. If that data is supposed to vary in size at run
time, you can ignore these warnings.
3 Generate a MEX function using codegen to verify the generated code

(requires a MATLAB Coder license). Use the following command-line
options:
• -args {coder.typeof...} if you have variable-size inputs
• -report to generate a compilation report
For example:
codegen -report myFcn -args {coder.typeof(0,[2 4],1)}
8-6
How to Generate Code for MATLAB® Functions with Variable-Size Data
This command uses coder.typeof to specify one variable-size input for

function myFcn. The first argument, 0, indicates the input data type
(double) and complexity (real). The second argument, [2 4], indicates the
size, a matrix with two dimensions. The third argument, 1, indicates that
the matrix is variable-size. The upper bound is 2 for the first dimension
and 4 for the second dimension.
Note During compilation, codegen detects variables and structure fields

that change size after you define them, and reports these occurrences as
errors. In addition, codegen performs a run-time check to generate errors
when data exceeds upper bounds.
4 Fix size mismatch errors:
Cause: How To Fix: For More

Information:
You try to change the Define the data to be See “Diagnosing and
size of data after its variable sized. Fixing Size Mismatch
size has been locked. Errors” on page 8-33
Fix upper bounds errors

Information:
MATLAB cannot Specify an upper See “Specifying
determine or bound. Upper Bounds for
compute the upper Variable-Size Data”
bound on page 8-27 and
“Diagnosing and Fixing
Errors in Detecting
Upper Bounds” on page
8-35
MATLAB attempts If the data is See “Enabling and
to compute an upper unbounded, enable Disabling Dynamic
8-7

Information:
bound for unbounded dynamic memory Memory Allocation” in

variable-size data. allocation. the MATLAB Coder
documentation.
5 Generate C/C++ code using the codegen function (requires a MATLAB

Coder license).
8-8
Tutorial: Generating MEX Code for a MATLAB® Function That Expands a Vector in a Loop
Tutorial: Generating MEX Code for a MATLAB Function

That Expands a Vector in a Loop
In this section...
“About the MATLAB Function emldemo_uniquetol” on page 8-9
“Step 1: Add Compilation Directive for Code Generation” on page 8-10
“Step 2: Address Issues Detected by the Code Analyzer” on page 8-10
“Step 3: Generate MEX Code” on page 8-10
“Step 4: Fix the Size Mismatch Error” on page 8-12
“Step 5: Fix the Upper Bounds Error” on page 8-14
“Step 6: Generate C/C++ Code” on page 8-16
About the MATLAB Function emldemo_uniquetol

This tutorial uses the function emldemo_uniquetol. This function returns
in vector B a version of input vector A, where the elements are unique to
within tolerance tol of each other. In vector B, abs(B(i) - B(j)) > tol for all
i and j. Initially, assume input vector A can store up to 100 elements. In a
later exercise, you will enable dynamic memory allocation for an unbounded
input vector.
function B = emldemo_uniquetol(A, tol)

A = sort(A);
B = A(1);
k = 1;
for i = 2:length(A)
if abs(A(k) - A(i)) > tol
B = [B A(i)];
k = i;
end
end
Work Through the Tutorial or Use the Supplied Solution

To work through the tutorial, start at “Step 1: Add Compilation Directive for
Code Generation” on page 8-10. To open the supplied solution:
8-9
1 Change to the matlabroot/toolbox/stateflow/sfdemos folder:
cd ([matlabroot, '/toolbox/stateflow/sfdemos'])
2 Open emldemo_uniquetol.m.
Step 1: Add Compilation Directive for Code

Generation
Add the %#codegen compilation directive at the top of the function:
function B = emldemo_uniquetol(A, tol) %#codegen

A = sort(A);
B = A(1);
k = 1;
for i = 2:length(A)
B = [B A(i)];
k = i;
end
end
Step 2: Address Issues Detected by the Code

Analyzer
The Code Analyzer detects that variable B might change size in the for-loop.
It issues this warning:
The variable 'B' appears to change size on every loop iteration.

Consider preallocating for speed.
In this function, vector B should expand in size as it adds values from vector A.
Therefore, you can ignore this warning.
Step 3: Generate MEX Code

To generate MEX code, use the codegen function (requires a MATLAB Coder
license).
1 Generate a MEX function for emldemo_uniquetol:
8-10
codegen emldemo_uniquetol -args {coder.typeof(0,[1 100],1),0} -report
What do these command-line options mean?
The -args option specifies the class, complexity, and size of each input to
function emldemo_uniquetol:
• The first argument, coder.typeof, defines a variable-size input. The
expression coder.typeof(0,[1 100],1) defines input A as a real double
vector with a fixed upper bound. Its first dimension is fixed at 1 and its
second dimension can vary in size up to 100 elements.
For more information, see “Specifying Variable-Size Inputs at the
Command Line” in the MATLAB Coder documentation.
• The second argument, 0, defines input tol as a real double scalar.
The -report option instructs codegen to generate a compilation report,

even if no errors or warnings occur.
For more information, see the codegen reference page in the MATLAB
Coder documentation.
Executing this command generates a compiler error:
??? Size mismatch (size [1 x 1] ~= size [1 x 2]).

The size to the left is the size
of the left-hand side of the assignment.
8-11
2 Open the error report and select the Variables tab.
The error indicates a size mismatch between the left-hand side and right-hand
side of the assignment statement B = [B A(i)];. The assignment B =
A(1) establishes the size of B as a fixed-size scalar (1 x 1). Therefore, the
concatenation of [B A(i)] creates a 1 x 2 vector.
Step 4: Fix the Size Mismatch Error

To fix this error, define B to be a variable-size vector.
1 Add this statement to the emldemo_uniquetol function:
coder.varsize('B');
It should appear before B is used (read). For example:

A = sort(A);
8-12
coder.varsize('B');
B = A(1);
k = 1;
for i = 2:length(A)
B = [B A(i)];
k = i;
end
end
The function coder.varsize defines every instance of B in

emldemo_uniquetol to be variable sized.
2 Generate another code generation report using this command:
This time, the command generates a different error:
??? Computed maximum size is not bounded.

Static memory allocation requires all sizes to be bounded.
The computed size is [1 x :?].
This error may be reported due to a limitation
of the underlying analysis. Please consider enabling
dynamic memory allocation to allow unbounded sizes.
3 Open the error report and select the Variables tab.
8-13
This error occurs because codegen cannot determine an upper bound for B .
Step 5: Fix the Upper Bounds Error

There are two ways to fix this error:
• “Specify Upper Bounds for Variable B” on page 8-14

• “Enable Dynamic Memory Allocation for an Unbounded Variable B” on
page 8-15
Specify Upper Bounds for Variable B

Choose this method if you want to enforce an upper bound for B. In this
exercise, you constrain B to the same upper bound as A.
1 Add a second argument to coder.varsize:
coder.varsize('B', [1 100]);
The argument [1 100] specifies that B is a vector with its first dimension
fixed at size 1 and the second dimension variable to an upper bound of
8-14
100. The value of 100 matches the upper bound of variable-size vector A.
Based on the algorithm, output B is at most as large as input A. By default,
dimensions of 1 are fixed size.
Here is the modified code:

A = sort(A);
B = A(1);
k = 1;
for i = 2:length(A)
B = [B A(i)];
k = i;
end
end
2 Compile the function again:
codegen should compile successfully and generate a MEX function for

emldemo_uniquetol.
This exercise presents one way to specify an upper bound. To learn about other
methods, see “Specifying Upper Bounds for Variable-Size Data” on page 8-27
Enable Dynamic Memory Allocation for an Unbounded

Variable B
Choose this method if you do not know the upper bound for B or do not need
to enforce an upper bound. In this exercise, you will also remove the upper
bound for input A.
1 Enable the DynamicMemoryAllocation configuration option for MEX code

generation:
8-15
cfg = coder.config;
cfg.DynamicMemoryAllocation = 'AllVariableSizeArrays';
2 Compile the function again, adding a -config argument:
codegen emldemo_uniquetol -args {coder.typeof(0,[1 100],1),0} -report -config cfg
Adding -config cfg applies the configuration setting that enables

dynamic memory allocation.
codegen should compile successfully and generate a MEX function for

emldemo_uniquetol.
3 Specify the second dimension of input A as unbounded:
codegen -config cfg -report emldemo_uniquetol -args {coder.typeof(0,[1 Inf]),0}
If you do not know the upper bounds of an input, it is good coding practice
to specify the input as unbounded instead of giving it an arbitrary upper
bound. In this codegen command, the size of the second dimension of
input A is Inf. When you specify the size of a dimension as Inf in a
coder.typeof statement, codegen treats the dimension as unbounded.
You can use Inf only with dynamic allocation.
See Also
• codegen in the MATLAB Coder documentation.
• “Primary Function Input Specification” in the MATLAB Coder
documentation.
Step 6: Generate C/C++ Code

If you have a MATLAB Coder license, you can generate C/C++ code for
variable-size data in the example function. For example, to generate a
C library for unbounded data in the emldemo_uniquetol function using
dynamic memory allocation:
8-16
1 Enable the DynamicMemoryAllocation configuration option for C library

generation:
cfg=coder.config('lib');
cfg.DynamicMemoryAllocation = 'AllVariableSizeArrays';
2 Issue this command:
codegen -config cfg -report emldemo_uniquetol -args {coder.typeof(0,[1 Inf]),0}
In the generated code, MATLAB represents data with unknown upper bounds
as a data type called emxArray. MATLAB provides utility functions for
creating and interacting with emxArrays in your generated code.
See Also
• “C Code Interface for Unbounded Arrays and Structure Fields” on page 8-30
• codegen in the MATLAB Coder documentation.
8-17
Variable-Size Data in Code Generation Reports

In this section...
“What Reports Tell You About Size” on page 8-18
“How Size Appears in Code Generation Reports” on page 8-19
“How to Generate a Code Generation Report” on page 8-19
What Reports Tell You About Size

Code generation reports:
• Differentiate fixed-size from variable-size data

• Identify variable-size data with unknown upper bounds
• Identify variable-size data with fixed dimensions
If you define a variable-size array and then subsequently fix the dimensions
of this array in the code, the report appends * to the size of the variable. In
the generated C code, this variable appears as a variable-size array, but
the size of its dimensions does not change during execution.
• Provide guidance on how to fix size mismatch and upper bounds errors.
8-18
Variable-Size Data in Code Generation Reports
How Size Appears in Code Generation Reports
* means that you declared y as variable size,

but subsequently fixed its dimensions
How to Generate a Code Generation Report

Add the -report option to your codegen command (requires a MATLAB
Coder license).
8-19
Defining Variable-Size Data for Code Generation

In this section...
“When to Define Variable-Size Data Explicitly” on page 8-20
“Using a Matrix Constructor with Nonconstant Dimensions” on page 8-21
“Inferring Variable Size from Multiple Assignments” on page 8-21
“Defining Variable-Size Data Explicitly Using coder.varsize” on page 8-23
When to Define Variable-Size Data Explicitly

For code generation, you must assign variables to have a specific class,
size, and complexity before using them in operations or returning them as
outputs. Generally, you cannot reassign variable properties after the initial
assignment. Therefore, attempts to grow a variable or structure field after
assigning it a fixed size might cause a compilation error. In these cases, you
must explicitly define the data as variable sized using one of these methods:
Method See
Assign the data from a variable-size “Using a Matrix Constructor with
matrix constructor such as Nonconstant Dimensions” on page
• ones 8-21
• zeros
• repmat
Assign multiple, constant sizes “Inferring Variable Size from
to the same variable before using Multiple Assignments” on page 8-21
(reading) the variable.
Define all instances of a variable to “Defining Variable-Size Data
be variable sized Explicitly Using coder.varsize” on
page 8-23
8-20
Using a Matrix Constructor with Nonconstant

Dimensions
You can define a variable-size matrix by using a constructor with nonconstant
dimensions. For example:
function y = var_by_assign(u) %#codegen

if (u > 0)
y = ones(3,u);
else
y = zeros(3,1);
end
If you are not using dynamic memory allocation, you must also add an assert
statement to provide upper bounds for the dimensions. For example:
function y = var_by_assign(u) %#codegen

assert (u < 20);
if (u > 0)
y = ones(3,u);
else
y = zeros(3,1);
end
Inferring Variable Size from Multiple Assignments

You can define variable-size data by assigning multiple, constant sizes to the
same variable before you use (read) the variable in your code. When MATLAB
uses static allocation on the stack for code generation, it infers the upper
bounds from the largest size specified for each dimension. When you assign
the same size to a given dimension across all assignments, MATLAB assumes
that the dimension is fixed at that size. The assignments can specify different
shapes as well as sizes.
When you enable dynamic memory allocation on the heap, MATLAB does not
check for upper bounds; it assumes all variable-size data is unbounded.
Example: Inferring Upper Bounds from Multiple Definitions

with Different Shapes
function y = var_by_multiassign(u) %#codegen
8-21
if (u > 0)
y = ones(3,4,5);
else
y = zeros(3,1);
end
When static allocation is used, this function infers that y is a matrix with
three dimensions, where:
• First dimension is fixed at size 3

• Second dimension is variable with an upper bound of 4
• Third dimension is variable with an upper bound of 5
The code generation report represents the size of matrix y like this:
When dynamic allocation is used, the function analyzes the dimensions of

y differently:

• Second and third dimensions are unbounded
In this case, the code generation report represents the size of matrix y like
this:
8-22
Defining Variable-Size Data Explicitly Using

coder.varsize
Use the function coder.varsize to define one or more variables or structure
fields as variable-size data. Optionally, you can also specify which dimensions
vary along with their upper bounds (see “Specifying Which Dimensions Vary”
on page 8-23). For example:
• Define B as a variable-size 2-by-2 matrix, where each dimension has an

upper bound of 64:
• Define B as a variable-size matrix:
coder.varsize('B');
When you supply only the first argument, coder.varsize assumes all
dimensions of B can vary and that the upper bound is size(B).
For more information, see the coder.varsize reference page.
Specifying Which Dimensions Vary

You can use the function coder.varsize to specify which dimensions vary.
For example, the following statement defines B as a row vector whose first
dimension is fixed at 2, but whose second dimension can grow to an upper
bound of 16:
coder.varsize('B', [2, 16], [0 1])
The third argument specifies which dimensions vary. This argument must be
a logical vector or a double vector containing only zeros and ones. Dimensions
that correspond to zeros or false have fixed size; dimensions that correspond
to ones or true vary in size. coder.varsize usually treats dimensions of size
1 as fixed (see “Defining Variable-Size Matrices with Singleton Dimensions”
on page 8-24).
For more information about the syntax, see the coder.varsize reference
page.
8-23
Allowing a Variable to Grow After Defining Fixed Dimensions

Function var_by_if defines matrix Y with fixed 2-by-2 dimensions before first
use (where the statement Y = Y + u reads from Y). However, coder.varsize
defines Y as a variable-size matrix, allowing it to change size based on decision
logic in the else clause:
function Y = var_by_if(u) %#codegen

if (u > 0)
Y = zeros(2,2);
coder.varsize('Y');
if (u < 10)
Y = Y + u;
end
else
Y = zeros(5,5);
end
Without coder.varsize, MATLAB infers Y to be a fixed-size, 2-by-2 matrix

and generates a size mismatch error during code generation.
Defining Variable-Size Matrices with Singleton Dimensions

A singleton dimension is any dimension for which size(A,dim) = 1. Singleton
dimensions are fixed in size when:
• You specify a dimension with an upper bound of 1 in coder.varsize

expressions.
For example, in this function, Y behaves like a vector with one variable-size
dimension:
function Y = dim_singleton(u) %#codegen

Y = [1 2];
coder.varsize('Y', [1 10]);
if (u > 0)
Y = [Y 3];
else
Y = [Y u];
end
8-24
• You initialize variable-size data with singleton dimensions using matrix

constructor expressions or matrix functions.
For example, in this function, both X and Y behave like vectors where only
their second dimensions are variable sized:
function [X,Y] = dim_singleton_vects(u) %#codegen

Y = ones(1,3);
X = [1 4];
coder.varsize('Y','X');
if (u > 0)
Y = [Y u];
else
X = [X u];
end
You can override this behavior by using coder.varsize to specify explicitly

that singleton dimensions vary. For example:
function Y = dim_singleton_vary(u) %#codegen

Y = [1 2];
coder.varsize('Y', [1 10], [1 1]);
if (u > 0)
Y = [Y Y+u];
else
Y = [Y Y*u];
end
In this example, the third argument of coder.varsize is a vector of ones,

indicating that each dimension of Y varies in size. For more information, see
the coder.varsize reference page.
Defining Variable-Size Structure Fields

To define structure fields as variable-size arrays, use colon (:) as the index
expression. The colon (:) indicates that all elements of the array are variable
sized. For example:
function y=struct_example() %#codegen
d = struct('values', zeros(1,0), 'color', 0);
8-25
data = repmat(d, [3 3]);

coder.varsize('data(:).values');
for i = 1:numel(data)
data(i).color = rand-0.5;
data(i).values = 1:i;
end
y = 0;
for i = 1:numel(data)
if data(i).color > 0
y = y + sum(data(i).values);
end;
end
The expression coder.varsize('data(:).values') defines the field values

inside each element of matrix data to be variable sized.
Here are other examples:
• coder.varsize('data.A(:).B')
In this example, data is a scalar variable that contains matrix A. Each
element of matrix A contains a variable-size field B.
• coder.varsize('data(:).A(:).B')
This expression defines field B inside each element of matrix A inside each
element of matrix data to be variable sized.
8-26
Specifying Upper Bounds for Variable-Size Data

In this section...
“When to Specify Upper Bounds for Variable-Size Data” on page 8-27
“Specifying Upper Bounds on the Command Line for Variable-Size Inputs”
on page 8-27
“Specifying Unknown Upper Bounds for Variable-Size Inputs” on page 8-27
“Specifying Upper Bounds for Local Variable-Size Data” on page 8-28
When to Specify Upper Bounds for Variable-Size

Data
When using static allocation on the stack during code generation, MATLAB
must be able to determine upper bounds for variable-size data. Specify the
upper bounds explicitly for variable-size data from external sources, such
as inputs and outputs.
Specifying Upper Bounds on the Command Line for

Variable-Size Inputs
Use the coder.typeof construct with the -args option on the codegen
command line (requires a MATLAB Coder license). For example:
codegen foo -args {coder.typeof(double(0),[3 100],1)}
This command specifies that the input to function foo is a matrix of real
doubles with two variable dimensions. The upper bound for the first
dimension is 3; the upper bound for the second dimension is 100. For a
detailed explanation of this syntax, see coder.typeof in the MATLAB Coder
documentation.
Specifying Unknown Upper Bounds for Variable-Size

Inputs
If you use dynamic memory allocation, you can specify that you don’t know
the upper bounds of inputs. To specify an unknown upper bound, use the
infinity constant Inf in place of a numeric value. For example:
8-27
codegen foo -args {coder.typeof(double(0), [1 Inf])}
In this example, the input to function foo is a vector of real doubles without
an upper bound.
Specifying Upper Bounds for Local Variable-Size Data

When using static allocation, MATLAB uses a sophisticated analysis to
calculate the upper bounds of local data at compile time. However, when the
analysis fails to detect an upper bound or calculates an upper bound that is
not precise enough for your application, you need to specify upper bounds
explicitly for local variables.
You do not need to specify upper bounds when using dynamic allocation on
the heap. In this case, MATLAB assumes all variable-size data is unbounded
and does not attempt to determine upper bounds.
• “Constraining the Value of a Variable That Specifies Dimensions of

Variable-Size Data” on page 8-28
• “Specifying the Upper Bounds for All Instances of a Local Variable” on
page 8-29
Constraining the Value of a Variable That Specifies Dimensions

of Variable-Size Data
Use the assert function with relational operators to constrain the value of
variables that specify the dimensions of variable-size data. For example:
function y = dim_need_bound(n) %#codegen

assert (n <= 5);
L= ones(n,n);
M = zeros(n,n);
M = [L; M];
y = M;
This assert statement constrains input n to a maximum size of 5, defining L

and M as variable-sized matrices with upper bounds of 5 for each dimension.
8-28
Specifying the Upper Bounds for All Instances of a Local

Variable
Use the coder.varsize function to specify the upper bounds for all instances
of a local variable in a function. For example:
function Y = example_bounds1(u) %#codegen

Y = [1 2 3 4 5];
coder.varsize('Y', [1 10]);
if (u > 0)
Y = [Y Y+u];
else
Y = [Y Y*u];
end
The second argument of coder.varsize specifies the upper bound for each
instance of the variable specified in the first argument. In this example, the
argument [1 10] indicates that for every instance of Y:

• Second dimension can grow to an upper bound of 10
By default, coder.varsize assumes dimensions of 1 are fixed size. For more

information, see the coder.varsize reference page.
8-29
C Code Interface for Unbounded Arrays and Structure

Fields
In this section...
“emxArray: Representation of Data with Unknown Upper Bounds” on page
8-30
“Utility Functions for Creating emxArray Data Structures” on page 8-31
emxArray: Representation of Data with Unknown

Upper Bounds
In generated code, MATLAB represents data with unknown upper bounds
as a structure type called emxArray. An embeddable version of the MATLAB
mxArray, the emxArray is a family of data types, specialized for all base types.
emxArray Structure Definition

typedef struct emxArray_<baseTypeName>
{
<baseTypeName> *data;
int32_T *size;
int32_T allocated;
int32_T numDimensions;
boolean_T canFreeData;
} emxArray_<baseTypeName>;
For example, here’s the definition for an emxArray of base type real_T with
unknown upper bounds:
typedef struct emxArray_real_T

{
real_T *data;
int32_T *size;
int32_T allocated;
int32_T numDimensions;
boolean_T canFreeData;
} emxArray_real_T;
8-30
C Code Interface for Unbounded Arrays and Structure Fields
To define two variables, in1 and in2, of this type, use this statement:
emxArray_real_T *in1, *in2;
emxArray Structure Fields
Field Description
*data Pointer to data of type <baseTypeName>
*size Pointer to first element of size vector. Length
of the vector equals the number of dimensions.
allocatedSize Number of elements currently allocated for the
array. If the size changes, MATLAB reallocates
memory based on the new size.
numDimensions Number of dimensions of the size vector, that
is, the number of dimensions you can access
without crossing into unallocated or unused
memory
canFreeData Boolean flag indicating how to deallocate
memory:
• true – MATLAB deallocates memory
automatically
• false – Calling program determines when
to deallocate memory
Utility Functions for Creating emxArray Data

Structures
To create and interact with emxArrays in your generated code, MATLAB
provides a set of utility functions, along with a header file. To call these
functions in your main C function, include the generated header file.
8-31
Function Arguments Description

emxArray_<baseTypeName> *data Creates a new
*emxCreateWrapper_<baseTypeName> num_rows 2-dimensional
(...) num_cols emxArray, but does not
allocate it on the heap.
Instead uses memory
provided by the user
and sets canFreeData
to false so it never
inadvertently free user
memory, such as the
stack.
emxArray_<baseTypeName> *data Same as
*emxCreateWrapperND_<baseTypeName> numDimensions emxCreateWrapper,
(...) *size except it creates a
new N-dimensional
emxArray.
emxArray_<baseTypeName> num_rows Creates a new
*emxCreate_<baseTypeName> (...) num_cols two-dimensional
emxArray on the heap,
initialized to zero. All
data elements have the
data type specified by
baseTypeName. .
emxArray_<baseTypeName> numDimensions Same as emxCreate,
*emxCreateND_<baseTypeName> (...) *size except it creates a
new N-dimensional
emxArray on the heap.
emxArray_<baseTypeName> *emxArray Frees dynamic
*emxDestroyArray_<baseTypeName> memory allocated
(...) and by *emxCreate
*emxCreateND
functions.
8-32
Troubleshooting Issues with Variable-Size Data

In this section...
“Diagnosing and Fixing Size Mismatch Errors” on page 8-33
“Diagnosing and Fixing Errors in Detecting Upper Bounds” on page 8-35
Diagnosing and Fixing Size Mismatch Errors

Check your code for these issues:
Assigning Variable-Size Matrices to Fixed-Size Matrices
You cannot assign variable-size matrices to fixed-size matrices in generated

code. Consider this example:
function Y = example_mismatch1(n) %#codegen

assert(n<10);
B = ones(n,n);
A = magic(3);
A(1) = mean(A(:));
if (n == 3)
A = B;
end
Y = A;
Compiling this function produces this error:
??? Dimension 1 is fixed on the left-hand side

but varies on the right ...
There are several ways to fix this error:
• Allow matrix A to grow by adding the coder.varsize construct:
function Y = example_mismatch1_fix1(n) %#codegen

coder.varsize('A');
assert(n<10);
B = ones(n,n);
A = magic(3);
8-33
A(1) = mean(A(:));
if (n == 3)
A = B;
end
Y = A;
• Explicitly restrict the size of matrix B to 3-by-3 by modifying the assert

statement:

coder.varsize('A');
assert(n==3)
B = ones(n,n);
A = magic(3);
A(1) = mean(A(:));
if (n == 3)
A = B;
end
Y = A;
• Use explicit indexing to make B the same size as A:

assert(n<10);
B = ones(n,n);
A = magic(3);
A(1) = mean(A(:));
if (n == 3)
A = B(1:3, 1:3);
end
Y = A;
Empty Matrix Reshaped to Match Variable-Size Specification
If you assign an empty matrix [] to variable-size data, MATLAB might

silently reshape the data in generated code to match a coder.varsize
specification. For example:
function Y = test(u) %#codegen

Y = [];
coder.varsize(`Y', [1 10]);
8-34
If u < 0
Y = [Y u];
end
In this example, coder.varsize defines Y as a column vector of up to 10

elements, so its first dimension is fixed at size 1. The statement Y = []
designates the first dimension of Y as 0, creating a mismatch. The right
hand side of the assignment is an empty matrix and the left hand side is a
variable-size vector. In this case, MATLAB reshapes the empty matrix Y =
[] in generated code to Y = zeros(1,0) so it matches the coder.varsize
specification.
Performing Binary Operations on Fixed and Variable-Size Operands
You cannot perform binary operations on operands of different sizes.

Operands have different sizes if one has fixed dimensions and the other has
variable dimensions. For example:
function z = mismatch_operands(n) %#codegen

assert(n>=3 && n<10);
x = ones(n,n);
y = magic(3);
z = x + y;
When you compile this function, you get an error because y has fixed
dimensions (3 x 3), but x has variable dimensions. Fix this problem by using
explicit indexing to make x the same size as y:
function z = mismatch_operands_fix(n) %#codegen

assert(n>=3 && n<10);
x = ones(n,n);
y = magic(3);
z = x(1:3,1:3) + y;
Diagnosing and Fixing Errors in Detecting Upper

Bounds
Check your code for these issues:
8-35
Using Nonconstant Dimensions in a Matrix Constructor
You can define variable-size data by assigning a variable to a matrix with

nonconstant dimensions. For example:
function y = dims_vary(u) %#codegen

if (u > 0)
y = ones(3,u);
else
y = zeros(3,1);
end
However, compiling this function generates an error because you did not
specify an upper bound for u. There are several ways to fix the problem:
• Enable dynamic memory allocation and recompile. During code generation,

MATLAB does not check for upper bounds when it uses dynamic memory
allocation for variable-size data.
• If you do not want to use dynamic memory allocation, add an assert
statement before the first use of u:
function y = dims_vary_fix(u) %#codegen

assert (u < 20);
if (u > 0)
y = ones(3,u);
else
y = zeros(3,1);
end
8-36
Limitations with Variable-Size Support for Code Generation
Limitations with Variable-Size Support for Code

Generation
In this section...
“Limitation on Scalar Expansion” on page 8-37
“Incompatibility with MATLAB in Determining Size of Variable-Size N-D
Arrays” on page 8-38
“Limitation on Vector-Vector Indexing” on page 8-39
“Limitations on Matrix Indexing Operations for Code Generation” on page
8-40
“Dynamic Memory Allocation Not Supported for MATLAB Function Blocks”
on page 8-41
“Limitation on Mixing Stack and Heap Allocation” on page 8-41
Limitation on Scalar Expansion

Scalar expansion is a method of converting scalar data to match the
dimensions of vector or matrix data. During code generation, the standard
MATLAB scalar expansion rules apply except when adding two variable-size
expressions. In this case, both operands must be the same size. MATLAB
does not perform scalar expansion even if one of the variable-size expressions
is scalar. Instead, it generates a size mismatch error at run time.
For example, the following code applies the standard MATLAB scalar
expansion rules:
function y = scalar_exp_test()%#codegen
A = zeros(2,2);
coder.varsize('A');
B = 1;
y = A + B;
It determines that B is scalar and adds it to the variable-size matrix A to

produce this result:
ans =
8-37
1 1
1 1
However, suppose B is also variable sized:
function y = scalar_exp_test_err1()%#codegen
A = zeros(2,2);
coder.varsize('A','B');
B = 1;
y = A + B;
In this case, the coder.varsize statement obscures the fact that B is scalar.
The function compiles successfully, but generates a run-time error:
??? Sizes mismatch: [2][2] ~= [1][1]
Workaround
To fix the problem, use indexing to force B to be a scalar value:
function y = scalar_exp_test_fix()%#codegen
A = zeros(2,2);
coder.varsize('A','B');
B = 1;
y = A + B(1);
Incompatibility with MATLAB in Determining Size of

Variable-Size N-D Arrays
For variable-size N-D arrays, the size function can return a different result in
generated code than in MATLAB. In generated code, size(A) always returns
a fixed-length output because it does not drop trailing singleton dimensions
of variable-size N-D arrays. By contrast, size(A) in MATLAB returns a
variable-length output because it drops trailing singleton dimensions.
For example, if the shape of array A is :?x:?x:? and size(A,3)==1, size(A)

returns:
• Three-element vector in generated code

• Two-element vector in MATLAB code
8-38
Workarounds
If your application requires generated code to return the same size of
variable-size N-D arrays as MATLAB code, consider one of these workarounds:
• Use the two-argument form of size.

For example, size(A,n) always returns the same answer in generated code
and MATLAB code.
• Rewrite size(A):
B = size(A);
X = B(1:ndims(A));
This version returns X with a variable-length output. However, you cannot

pass a variable-size X to matrix constructors such as zeros that require a
fixed-size argument.
Limitation on Vector-Vector Indexing

In vector-vector indexing, you use one vector as an index into another vector.
When either vector is variable sized, you might get a run-time error during
code generation. Consider the index expression A(B). The general rule for
indexing is that size(A(B)) == size(B). However, when both A and B are
vectors, MATLAB applies a special rule: use the orientation of A as the
orientation of the output. For example, if size(A) == [1 5] and size(B) ==
[3 1], then size(A(B)) == [1 3].
In this situation, if MATLAB detects that both A and B are vectors at compile
time, it applies the special rule and gives the same result as MATLAB.
However, if either A or B is a variable-size matrix (has shape ?x?) at compile
time, MATLAB applies only the general indexing rule. Then, if both A and B
become vectors at run time, MATLAB reports a run-time error in simulation.
Workaround
Force your data to be a vector by using the colon operator for indexing:
A(B(:)). For example, suppose your code intentionally toggles between
vectors and regular matrices at run time. You can do an explicit check to
provide deterministic behavior:
...
8-39
if isvector(A) && isvector(B)

C = A(:);
D = C(B(:));
else
D = A(B);
end
...
The indexing in the first branch specifies that C and B(:) are compile-time
vectors. As a result, MATLAB applies the standard vector-vector indexing
rule during code generation.
Limitations on Matrix Indexing Operations for Code

Generation
The following limitations apply to matrix indexing operations for code
generation:
• M(i:j) where i and j change in a loop

During code generation, memory is never dynamically allocated for the size
of the expressions that change as the program executes. To implement this
behavior, use for loops as shown in the following example:
...
M = ones(10,10);
for i=1:10
for j = i:10
M(i,j) = 2 * M(i,j);
end
end
...
Note The matrix M must be defined before entering the loop, as shown in
the highlighted code.
• M(i:i+k) where i is unknown but k is known
8-40
In this case, since i — and therefore i+k — are not known, memory cannot
be allocated for the numerical result. However, memory can be allocated
for the following workaround:
M(i + (0:k))
In this case, an unknown scalar value i is added to each element of the

known index vector 0...k. This means that memory for k+1 elements
of M is allocated.
• Initialization of the following style:
for i = 1:10
M(i) = 5;
end
In this case, the size of M changes as the loop is executed.
Dynamic Memory Allocation Not Supported for

MATLAB Function Blocks
You cannot use dynamic memory allocation for variable-size data in MATLAB
Function blocks. Use bounded instead of unbounded variable-size data.
Limitation on Mixing Stack and Heap Allocation

When you enable dynamic memory allocation, all of your variable-size data
is allocated on the heap, whether bounded or unbounded. For example, you
cannot selectively allocate bounded data on the stack and unbounded data
on the heap.
8-41
Restrictions on Variable Sizing in Toolbox Functions

Supported for Code Generation
In this section...
“Common Restrictions” on page 8-42
“Toolbox Functions with Variable Sizing Restrictions” on page 8-43
Common Restrictions
The following common restrictions apply to multiple toolbox functions
supported for code generation. To determine which of these restrictions apply
to specific library functions, see the table in “Toolbox Functions with Variable
Sizing Restrictions” on page 8-43.
Variable-length vector restriction

Inputs to the library function must be variable-length vectors or fixed-size
vectors. A variable-length vector is a variable-size array that has the shape
1x:n or :nx1 (one dimension is variable sized and the other is fixed at size 1).
Other shapes are not permitted, even if they are vectors at run time.
Automatic dimension restriction

When the function selects the working dimension automatically, it bases the
selection on the upper bounds for the dimension sizes. In the case of the sum
function, sum(X) selects its working dimension automatically, while sum(X,
dim) uses dim as the explicit working dimension.
For example, if X is a variable-size matrix with dimensions 1x:3x:5, sum(x)

behaves like sum(X,2) in generated code. In MATLAB, it behaves like
sum(X,2) provided size(X,2) is not 1. In MATLAB, when size(X,2) is 1,
sum(X) behaves like sum(X,3) . Consequently, you get a run-time error if an
automatically selected working dimension assumes a length of 1 at run time.
To avoid the issue, specify the intended working dimension explicitly as

a constant value.
8-42
Restrictions on Variable Sizing in Toolbox Functions Supported for Code Generation
Array-to-vector restriction
The function issues an error when a variable-size array that is not a
variable-length vector assumes the shape of a vector at run time. To avoid
the issue, specify the input explicitly as a variable-length vector instead of
a variable-size array.
Array-to-scalar restriction
The function issues an error if a variable-size array assumes a scalar value at
run time. To avoid this issue, specify all scalars as fixed size.
Toolbox Functions with Variable Sizing Restrictions

Function Restrictions with Variable-Size Data
all
• See “Automatic dimension restriction” on page
8-42.
• An error occurs if you pass the first argument a
variable-size matrix that is 0-by-0 at run time.
any
8-42.
• An error occurs if you pass the first argument a
variable-size matrix that is 0-by-0 at run time.
bsxfun
• Dimensions expand only where one input array
or the other has a fixed length of 1.
cat
• Dimension argument must be a constant.
• An error occurs if variable-size inputs are
empty at run time.
8-43

conv
• See “Variable-length vector restriction” on page
8-42.
• Input vectors must have the same orientation,
either both row vectors or both column vectors.
cov
• For cov(X), see“Array-to-vector restriction” on
page 8-43.
cross
• Variable-size array inputs that become vectors
at run time must have the same orientation.
deconv
• For both arguments, see“Variable-length vector
restriction” on page 8-42.
detrend
• For first argument for row vectors only, see
“Array-to-vector restriction” on page 8-43 .
diag
• See “Array-to-vector restriction” on page 8-43 .
diff
8-42.
• Length of the working dimension must be
greater than the difference order input when
the input is variable sized. For example, if the
input is a variable-size matrix that is 3-by-5 at
run time, diff(x,2,1) works but diff(x,5,1)
generates a run-time error.
fft
8-42.
8-44

filter
• For first and second arguments, see
“Variable-length vector restriction” on page
8-42.
8-42.
hist
• For second argument, see “Variable-length
vector restriction” on page 8-42.
• For second input argument, see“Array-to-scalar
histc
8-42.
ifft
8-42.
ind2sub
• First input (the size vector input) must be fixed
size.
interp1
• For the Y input and xi input, see“Array-to-vector
• Y input can become a column vector dynamically.
• A run-time error occurs if Y input is not a
variable-length vector and becomes a row vector
at run time.
ipermute
• Order input must be fixed size.
issorted
• For optional rows input, see “Variable-length
vector restriction” on page 8-42.
8-45

magic
• Argument must be a constant.
• Output can be fixed-size matrices only.
max
8-42.
mean
8-42.
• An error occurs if you pass as the first argument
a variable-size matrix that is 0-by-0 at run time.
median
8-42.
min
8-42.
mode
8-42.
8-46

mtimes
• When an input is variable sized, MATLAB
determines whether to generate code for a
general matrix*matrix multiplication or a
scalar*matrix multiplication, based on whether
one of the arguments is a fixed-size scalar. If
neither argument is a fixed-size scalar, the
inner dimensions of the two arguments must
agree even if a variable-size matrix input
happens to be a scalar at run time.
nchoosek
• Inputs must be fixed sized.
• Second input must be a constant for static
allocation. If you enable dynamic allocation,
second input can be a variable.
• You cannot create a variable-size array by
passing in a variable k unless you enable
dynamic allocation.
permute
• Order input must be fixed size.
planerot
• Input must be a fixed-size, two-element column
vector. It cannot be a variable-size array that
takes on the size 2-by-1 at run time.
poly
8-42.
polyfit
• For first and second arguments, see
“Variable-length vector restriction” on page
8-42.
8-47

prod
8-42.
rand
• For an upper-bounded variable N, rand(1,N)
produces a variable-length vector of 1x:M where
M is the upper bound on N.
• For an upper-bounded variable N, rand([1,N])
may produce a variable-length vector of :1x:M
where M is the upper bound on N.
randn
• For an upper-bounded variable N, randn(1,N)
produces a variable-length vector of 1x:M where
M is the upper bound on N.
• For an upper-bounded variable N, randn([1,N])
may produce a variable-length vector of :1x:M
where M is the upper bound on N.
reshape
• When the input is a variable-size empty array,
the maximum dimension size of the output
array (also empty) cannot be larger than that
of the input.
roots
8-42.
8-48

shiftdim
• If you do not supply the second argument, the
number of shifts is determined at compilation
time by the upper bounds of the dimension
sizes. Consequently, at run time the number of
shifts is always constant.
• An error occurs if the dimension that is shifted
to the first dimension has length 1 at run
time. To avoid the error, supply the number of
shifts as the second input argument (must be a
constant).
• First input argument must always have the
same number of dimensions when you supply a
positive number of shifts.
std
8-42.
• An error occurs if you pass a variable-size
matrix with 0-by-0 dimensions at run time.
sub2ind
• First input (the size vector input) must be fixed
size.
sum
8-42.
trapz
8-42.
8-49

typecast
8-42 on first argument.
var
8-42.
• An error occurs if you pass a variable-size
matrix with 0-by-0 dimensions at run time.
8-50
9
Code Generation for

Function Handles
• “How Working with Function Handles is Different for Code Generation”

on page 9-2
• “Example: Defining and Passing Function Handles for Code Generation”
on page 9-3
• “Limitations with Function Handles for Code Generation” on page 9-6
9 Code Generation for Function Handles
How Working with Function Handles is Different for Code

Generation
You can use function handles to invoke functions indirectly and parameterize
operations that you repeat frequently (see “Function Handles” in the
MATLAB Programming Fundamentals documentation). You can perform the
following operations with function handles:
• Define handles that reference user-defined functions and built-in functions

supported for code generation (see Chapter 2, “Functions Supported for
Code Generation”)
Note You cannot define handles that reference extrinsic MATLAB

functions (see “Calling MATLAB Functions” on page 11-11).
• Define function handles as scalar values

• Pass function handles as arguments to other functions (excluding extrinsic
functions)
To generate efficient standalone code for enumerated data, you are restricted
to using a subset of the operations you can perform with function handles
in MATLAB, as described in “Limitations with Function Handles for Code
Generation” on page 9-6
9-2
Example: Defining and Passing Function Handles for Code Generation
Example: Defining and Passing Function Handles for Code

Generation
The following code example shows how to define and call function handles for
code generation. You can copy the example to a MATLAB Function block
in Simulink or MATLAB function in Stateflow. To convert this function to
a MEX function using codegen, uncomment the two calls to the assert
function, highlighted below:
function addval(m)
%#codegen
% Define class and size of primary input m

% Uncomment next two lines to build MEX function with codegen
% assert(isa(m,'double'));
% assert(all (size(m) == [3 3]));
% Define MATLAB function disp to be extrinsic

disp(m);
% Pass function handle to addone

% to add one to each element of m
m = map(@addone, m);
disp(m);
% Pass function handle to addtwo

% to add two to each element of m
m = map(@addtwo, m);
disp(m);
function y = map(f,m)
y = m;
for i = 1:numel(y)
y(i) = f(y(i));
end
function y = addone(u)
9-3
y = u + 1;
function y = addtwo(u)
y = u + 2;
This code passes function handles @addone and @addtwo to the function map
which increments each element of the matrix m by the amount prescribed
by the referenced function. Note that map stores the function handle in the
input variable f and then uses f to invoke the function — in this case addone
first and then addtwo.
If you have MATLAB Coder, you can use the function codegen to convert the
function addval to a MEX executable that you can run in MATLAB. Follow
these steps:
1 At the MATLAB command prompt, issue this command:
codegen addval
2 Define and initialize a 3-by-3 matrix by typing a command like this at

the MATLAB prompt:
m = zeros(3)
3 Execute the function by typing this command:
addval(m)
You should see the following result:
0 0 0
0 0 0
0 0 0
1 1 1
1 1 1
1 1 1
3 3 3
3 3 3
3 3 3
9-4
Example: Defining and Passing Function Handles for Code Generation
For more information, see “Generating MEX Functions from MATLAB Code
at the Command Line” in the MATLAB Coder documentation.
9-5
Limitations with Function Handles for Code Generation

Function handles must be scalar values.
You cannot store function handles in matrices or structures.
You cannot use the same bound variable to reference different

function handles.
After you bind a variable to a specific function, you cannot use the same
variable to reference two different function handles, as in this example
%Incorrect code
...
x = @plus;
x = @minus;
...
This code produces a compilation error.
You cannot pass function handles to or from extrinsic functions.
You cannot pass function handles to or from feval and other extrinsic
MATLAB functions. For more information, see “Declaring MATLAB
Functions as Extrinsic Functions” on page 11-11
You cannot pass function handles to or from primary functions.
You cannot pass function handles as inputs to or outputs from primary

functions. For example, consider this function:
function x = plotFcn(fhandle, data)
assert(isa(fhandle,'function_handle') && isa(data,'double'));
coder.extrinsic('plot');
plot(data, fhandle(data));
x = fhandle(data);
9-6
Limitations with Function Handles for Code Generation
In this example, the function plotFcn receives a function handle and its
data as primary inputs. plotFcn attempts to call the function referenced by
the fhandle with the input data and plot the results. However, this code
generates a compilation error, indicating that the function isa does not
recognize 'function_handle' as a class name when called inside a MATLAB
function to specify properties of primary inputs.
You cannot view function handles from the debugger
You cannot display or watch function handles from the debugger. They
appear as empty matrices.
9-7
9-8
10
Defining Functions for Code

Generation
• “Specifying Variable Numbers of Arguments” on page 10-2

• “Supported Index Expressions” on page 10-3
• “Using varargin and varargout in for-Loops” on page 10-4
• “Implementing Wrapper Functions with varargin and varargout” on page
10-7
• “Passing Property/Value Pairs with varargin” on page 10-8
• “Rules for Using Variable Length Argument Lists for Code Generation” on
page 10-10
10 Defining Functions for Code Generation
Specifying Variable Numbers of Arguments

You can use varargin and varargout for passing and returning variable
numbers of parameters to MATLAB functions called from a top-level function.
Common applications of varargin and varargout for code generation include:
• Using for-loops to apply operations to a variable number of arguments

• Implementing wrapper functions that accept any number of inputs and
pass them to another function
• Passing variable numbers of property/value pairs as arguments to a
function
Code generation relies on loop unrolling to produce simple and efficient code
for varargin and varargout. This technique permits most common uses of
varargin and varargout, but not all (see “Rules for Using Variable Length
Argument Lists for Code Generation” on page 10-10). This following sections
explain how to code effectively using these constructs.
For more information about using varargin and varargout in MATLAB

functions, see Passing Variable Numbers of Arguments in the MATLAB
Programming Fundamentals documentation.
10-2
Supported Index Expressions
Supported Index Expressions

In MATLAB, varargin and varargout are cell arrays. Generated code does
not support cell arrays, but does allow you to use the most common syntax
— curly braces {} — for indexing into varargin and varargout arrays, as
in this example:
%#codegen
function [x,y,z] = fcn(a,b,c)
[x,y,z] = subfcn(a,b,c);
function varargout = subfcn(varargin)

for i = 1:length(varargin)
varargout{i} = varargin{i};
end
You can use the following index expressions. The exp arguments must be
constant expressions or depend on a loop index variable.
Expression Description
varargin varargin{exp} Read the value of element
(read only) exp
varargin{exp1: exp2} Read the values of elements
exp1 through exp2
varargin{:} Read the values of all
elements
varargout varargout{exp} Read or write the value of
(read and write) element exp
Note The use of () is not supported for indexing into varargin and
varargout arrays.
10-3
Using varargin and varargout in for-Loops

You can use varargin and varargout in for-loops to apply operations to
a variable number of arguments. To index into varargin and varargout
arrays in generated code, the value of the loop index variable must be known
at compile time. Therefore, during code generation, the compiler attempts
to automatically unroll these for-loops. Unrolling eliminates the loop logic
by creating a separate copy of the loop body in the generated code for each
iteration. Within each iteration, the loop index variable becomes a constant.
For example, the following function automatically unrolls its for-loop in the
generated code:
%#codegen
function [cmlen,cmwth,cmhgt] = conv_2_metric(inlen,inwth,inhgt)
[cmlen,cmwth,cmhgt] = inch_2_cm(inlen,inwth,inhgt);
function varargout = inch_2_cm(varargin)

varargout{i} = varargin{i} * 2.54;
end
When to Force Loop Unrolling

To automatically unroll for-loops containing varargin and varargout
expressions, the relationship between the loop index expression and the index
variable must be determined at compile time.
In the following example, the function fcn cannot detect a logical relationship
between the index expression j and the index variable i:
%#codegen

j = 0;
j = j+1;
varargout{j} = varargin{j};
10-4
Using varargin and varargout in for-Loops
end
As a result, the function does not unroll the loop and generates a compilation
error:
Nonconstant expression or empty matrix.

This expression must be constant because
its value determines the size or class of some expression.
To correct the problem, you can force loop unrolling by wrapping the loop
header in the function coder.unroll, as follows:
%#codegen

j = 0;
for i = coder.unroll(1:length(varargin))
j = j + 1;
varargout{j} = varargin{j};
end;
For more information, see coder.unroll in the Code Generation from

MATLAB reference documentation.
Example: Using Variable Numbers of Arguments in

a for-Loop
The following example multiplies a variable number of input dimensions in
inches by 2.54 to convert them to centimeters:
%#codegen
function [cmlen,cmwth,cmhgt] = conv_2_metric(inlen,inwth,inhgt)
[cmlen,cmwth,cmhgt] = inch_2_cm(inlen,inwth,inhgt);

end
10-5
Key Points About the Example
• varargin and varargout appear in the subfunction inch_2_cm, not in the

top-level function conv_2_metric.
• The index into varargin and varargout is a for-loop variable
For more information, see “Rules for Using Variable Length Argument Lists
for Code Generation” on page 10-10.
10-6
Implementing Wrapper Functions with varargin and varargout
Implementing Wrapper Functions with varargin and

varargout
You can use varargin and varargout to write wrapper functions that accept
any number of inputs and pass them directly to another function.
Example: Passing Variable Numbers of Arguments

from One Function to Another
The following example passes a variable number of inputs to different
optimization functions, based on a specified input method:
%#codegen
function answer = fcn(method,a,b,c)
answer = optimize(method,a,b,c);
function answer = optimize(method,varargin)

if strcmp(method,'simple')
answer = simple_optimization(varargin{:});
else
answer = complex_optimization(varargin{:});
end
...
Key Points About the Example
• You can use {:} to read all elements of varargin and pass them to another
function.
• You can mix variable and fixed numbers of arguments.
For more information, see “Rules for Using Variable Length Argument Lists
for Code Generation” on page 10-10.
10-7
Passing Property/Value Pairs with varargin

You can use varargin to pass property/value pairs in functions. However,
for code generation, you must take precautions to avoid type mismatch errors
when evaluating varargin array elements in a for-loop:
If Do This:
You assign varargin array elements Verify that for all pairs, the size,
to local variables in the for-loop type, and complexity are the same
for each property and the same for
each value
Properties or values have different Do not assign varargin array
sizes, types, or complexity elements to local variables in a
for-loop; reference the elements
directly
For example, in the following function test1, the sizes of the property strings
and numeric values are not the same in each pair:
%#codegen
function test1
v = create_value('size', 18, 'rgb', [240 9 44]);
end
function v = create_value(varargin)
v = new_value();
for i = 1 : 2 : length(varargin)
name = varargin{i};
value = varargin{i+1};
switch name
case 'size'
v = set_size(v, value);
case 'rgb'
v = set_color(v, value);
otherwise
end
end
end
10-8
Passing Property/Value Pairs with varargin
...
Generated code determines the size, type, and complexity of a local variable
based on its first assignment. In this example, the first assignments occur
in the first iteration of the for-loop:
• Defines local variable name with size equal to 4

• Defines local variable value with a size of scalar
However, in the second iteration, the size of the property string changes to
3 and the size of the numeric value changes to a vector, resulting in a type
mismatch error. To avoid such errors, reference varargin array values
directly, not through local variables, as highlighted in this code:
%#codegen
function test1
v = create_value('size', 18, 'rgb', [240 9 44]);
end
function v = create_value(varargin)
v = new_value();
for i = 1 : 2 : length(varargin)
switch varargin{i}
case 'size'
v = set_size(v, varargin{i+1});
case 'rgb'
v = set_color(v, varargin{i+1});
otherwise
end
end
end
...
10-9
Rules for Using Variable Length Argument Lists for Code

Generation
Do not use varargin or varargout in top-level functions
You cannot use varargin or varargout as arguments to top-level functions.

A top-level function is:
• The function called by Simulink in MATLAB Function block or by Stateflow

in a MATLAB function.
• The function that you provide on the command line to codegen
For example, the following code generates compilation errors:
%#codegen
end
To fix the problem, write a top-level function that specifies a fixed number
of inputs and outputs and then call inch_2_cm as an external function or
subfunction, as in this example:
%#codegen
function [cmL, cmW, cmH] = conv_2_metric(inL, inW, inH)
[cmL, cmW, cmH] = inch_2_cm(inL, inW, inH);

end
Use curly braces {} to index into the argument list
For code generation, you can use curly braces {}, but not parentheses (),
to index into varargin and varargout arrays. For more information, see
“Supported Index Expressions” on page 10-3.
10-10
Rules for Using Variable Length Argument Lists for Code Generation
Verify that indices can be computed at compile time
If you use an expression to index into varargin or varargout, make sure that
the value of the expression can be computed at compile time. For examples,
see “Using varargin and varargout in for-Loops” on page 10-4.
Do not write to varargin
Generated code treats varargin as a read-only variable. If you want to write

to any of the input arguments, copy the values into a local variable.
10-11
10-12
11
Calling Functions for Code

Generation
• “How MATLAB Resolves Function Calls in Generated Code” on page 11-2

• “How MATLAB Resolves File Types on the Path for Code Generation” on
page 11-6
• “Adding the Compilation Directive %#codegen” on page 11-8
• “Calling Subfunctions” on page 11-9
• “Calling Supported Toolbox Functions” on page 11-10
• “Calling MATLAB Functions” on page 11-11
11 Calling Functions for Code Generation
How MATLAB Resolves Function Calls in Generated Code

From a MATLAB function, you can call subfunctions, supported toolbox
functions, and other MATLAB functions. MATLAB resolves function names
for code generation as follows:
11-2
How MATLAB® Resolves Function Calls in Generated Code
Start
Dispatch to Function
Yes on Yes
MATLAB Extrinsic
for execution MATLAB function?
at runtime path?
No No
Yes
Subfunction?
No
Function Yes Yes

on the code Suitable
generation for code
path? generation?
No No
Generate
C code
Function
on Yes
MATLAB
path?
No
Generate error
11-3
Key Points About Resolving Function Calls

The diagram illustrates key points about how MATLAB resolves function
calls for code generation:
• Searches two paths, the code generation path and the MATLAB path
See “Compile Path Search Order” on page 11-4.
• Attempts to compile all functions unless you explicitly declare them to
be extrinsic
An extrinsic function is a function on the MATLAB path that the compiler
dispatches to MATLAB software for execution because the target language
does not support the function. MATLAB does not generate code for extrinsic
functions. You declare functions to be extrinsic by using the construct
coder.extrinsic, as described in “Declaring MATLAB Functions as
Extrinsic Functions” on page 11-11.
• Resolves file type based on precedence rules described in “How MATLAB
Resolves File Types on the Path for Code Generation” on page 11-6
Compile Path Search Order

During code generation, function calls are resolved on two paths:
1 Code generation path
MATLAB searches this path first during code generation. The code
generation path contains the toolbox functions supported for code
generation.
2 MATLAB path
If the function is not on the code generation path, MATLAB searches this
path.
MATLAB applies the same dispatcher rules when searching each path (see
“Determining Which Function Gets Called” in the MATLAB Programming
Fundamentals documentation).
11-4
How MATLAB® Resolves Function Calls in Generated Code
When to Use the Code Generation Path

Use the code generation path to override a MATLAB function with a
customized version. A file on the code generation path always shadows a file
of the same name on the MATLAB path.
11-5
How MATLAB Resolves File Types on the Path for Code

Generation
MATLAB uses the following precedence rules for code generation:
11-6
How MATLAB® Resolves File Types on the Path for Code Generation
Start
M-file and Yes

MEX-file in same
directory?
No
Yes
MEX-file?
No
Yes
Generate MDL-file? Compile
error M-file
No
Yes
P-file?
No
No Yes
M-file?
11-7
Adding the Compilation Directive %#codegen

Add the %#codegen directive (or pragma) to your function to indicate that you
intend to generate code for the MATLAB algorithm. Adding this directive
instructs the MATLAB code analyzer to help you diagnose and correct
violations that would result in errors during code generation.
11-8
Calling Subfunctions
Calling Subfunctions
Subfunctions are functions defined in the body of MATLAB function. They
work the same way for code generation as they do when executing your
algorithm in the MATLAB environment.
The following example illustrates how to define and call a subfunction mean:
function [mean, stdev] = stats(vals)

%#codegen
% Calculates a statistical mean and a standard

% deviation for the values in vals.
coder.extrinsic('plot');
len = length(vals);
mean = avg(vals, len);
stdev = sqrt(sum(((vals-avg(vals,len)).^2))/len;
plot(vals,'-+');
function mean = avg(array,size)

mean = sum(array)/size;
See “Subfunctions” in the MATLAB Programming Fundamentals

documentation for more information.
11-9
Calling Supported Toolbox Functions

You can call toolbox functions directly if they are supported for code
generation. For a list of supported functions, see “Functions Supported for
Code Generation — Alphabetical List” on page 2-3.
11-10
Calling MATLAB® Functions
Calling MATLAB Functions

MATLAB attempts to generate code for all functions unless you explicitly
declare them to be extrinsic (see “How MATLAB Resolves Function Calls in
Generated Code” on page 11-2). Extrinsic functions are not compiled, but
instead executed in MATLAB during simulation (see “How MATLAB Resolves
Extrinsic Functions During Simulation” on page 11-15).
There are two ways to declare a function to be extrinsic:
• Use the coder.extrinsic construct in main functions or subfunctions (see

“Declaring MATLAB Functions as Extrinsic Functions” on page 11-11).
• Call the function indirectly using feval (see “Calling MATLAB Functions
Using feval” on page 11-15).
Declaring MATLAB Functions as Extrinsic Functions

To declare a MATLAB function to be extrinsic, add the coder.extrinsic
construct at the top of the main function or a subfunction:
coder.extrinsic('function_name_1', ... , 'function_name_n');
Example: Declaring Extrinsic Functions

The following code declares the MATLAB patch and axis functions extrinsic
in the subfunction create_plot:
function c = pythagoras(a,b,color) %#codegen

% Calculates the hypotenuse of a right triangle
% and displays the triangle.
c = sqrt(a^2 + b^2);
create_plot(a, b, color);
function create_plot(a, b, color)

%Declare patch and axis as extrinsic
coder.extrinsic('patch', 'axis');
11-11
x = [0;a;a];
y = [0;0;b];
patch(x, y, color);
axis('equal');
By declaring these functions extrinsic, you instruct the compiler not to

generate code for patch and axis, but instead dispatch them to MATLAB
for execution.
To test the function, follow these steps:
1 Convert pythagoras to a MEX function by executing this command at

the MATLAB prompt:
codegen -args {1, 1, [.3 .3 .3]} pythagoras
2 Run the MEX function by executing this command:
pythagoras(3, 4, [1.0 0.0 0.0]);
MATLAB displays a plot of the right triangle as a red patch object:
11-12
When to Use the coder.extrinsic Construct

Use the coder.extrinsic construct to:
• Call MATLAB functions that produce no output — such as plot and

patch — for visualizing results during simulation, without generating
unnecessary code (see “How MATLAB Resolves Extrinsic Functions During
Simulation” on page 11-15).
• Make your code self-documenting and easier to debug. You can scan the
source code for coder.extrinsic statements to isolate calls to MATLAB
functions, which can potentially create and propagate mxArrays (see
“Working with mxArrays” on page 11-16).
11-13
• Save typing. With one coder.extrinsic statement, each subsequent

function call is extrinsic, as long as the call and the statement are in the
same scope (see “Scope of Extrinsic Function Declarations” on page 11-14).
• Declare the MATLAB function(s) extrinsic throughout the calling function
scope (see “Scope of Extrinsic Function Declarations” on page 11-14). To
narrow the scope, use feval (see “Calling MATLAB Functions Using feval”
on page 11-15).
Rules for Extrinsic Function Declarations

Observe the following rules when declaring functions extrinsic for code
generation:
• Declare the function extrinsic before you call it.

• Do not use the extrinsic declaration in conditional statements.
Scope of Extrinsic Function Declarations

The coder.extrinsic construct has function scope. For example, consider
the following code:
function y = foo %#codegen

coder.extrinsic('rat','min');
[N D] = rat(pi);
y = 0;
y = min(N, D);
In this example, rat and min as treated as extrinsic every time they are
called in the main function foo. There are two ways to narrow the scope of
an extrinsic declaration inside the main function:
• Declare the MATLAB function extrinsic in a subfunction, as in this

example:
coder.extrinsic('rat');
[N D] = rat(pi);
y = 0;
y = mymin(N, D);
11-14
function y = mymin(a,b)
coder.extrinsic('min');
y = min(a,b);
Here, the function rat is extrinsic every time it is called inside the main
function foo, but the function min is extrinsic only when called inside the
subfunction mymin.
• Call the MATLAB function using feval, as described in “Calling MATLAB
Functions Using feval” on page 11-15.
Calling MATLAB Functions Using feval

The function feval is automatically interpreted as an extrinsic function
during code generation. Therefore, you can use feval to conveniently call
functions that you want to execute in the MATLAB environment, rather than
compiled to generated code.
Consider the following example:
function y = foo
[N D] = rat(pi);
y = 0;
y = feval('min', N, D);
Because feval is extrinsic, the statement feval('min', N, D) is evaluated

by MATLAB — not compiled — which has the same effect as declaring the
function min extrinsic for just this one call. By contrast, the function rat is
extrinsic throughout the function foo.
How MATLAB Resolves Extrinsic Functions During

Simulation
MATLAB resolves calls to extrinsic functions — functions that do not support
code generation — as follows:
11-15
During simulation, MATLAB generates code for the call to an extrinsic

function, but does not generate the function’s internal code. Therefore, you
can run the simulation only on platforms where you install MATLAB software.
During code generation, MATLAB attempts to determine whether the

extrinsic function affects the output of the function in which it is called — for
example by returning mxArrays to an output variable (see “Working with
mxArrays” on page 11-16). If there is no effect on output, MATLAB proceeds
with code generation, but excludes the extrinsic function from the generated
code. Otherwise, MATLAB issues a compiler error.
Working with mxArrays

The output of an extrinsic function is an mxArray — also called a MATLAB
array. The only valid operations for mxArrays are:
• Storing mxArrays in variables

• Passing mxArrays to functions and returning them from functions
• Converting mxArrays to known types at run time
To use mxArrays returned by extrinsic functions in other operations, you must

first convert them to known types, as described in “Converting mxArrays to
Known Types” on page 11-17.
11-16
Converting mxArrays to Known Types

To convert anmxArray to a known type, assign the mxArray to a variable
whose type is defined. At run time, the mxArray is converted to the type of the
variable assigned to it. However, if the data in the mxArray is not consistent
with the type of the variable, you get a run-time error.
For example, consider this code:

[N D] = rat(pi);
y = min(N, D);
Here, the top-level function foo calls the extrinsic MATLAB function rat,
which returns two mxArrays representing the numerator N and denominator
D of the rational fraction approximation of pi. Although you can pass these
mxArrays to another MATLAB function — in this case, min — you cannot
assign the mxArray returned by min to the output y.
If you run this function foo in a MATLAB Function block in a Simulink

model, the code generates the following error during simulation:
Function output 'y' cannot be of MATLAB type.
To correct this problem, define y to be the type and size of the value that you
expect min to return — in this case, a scalar double — as follows:

[N D] = rat(pi);
y = 0; % Define y as a scalar of type double
y = min(N,D);
Restrictions on Extrinsic Functions for Code

Generation
The full MATLAB run-time environment is not supported during code
generation. Therefore, the following restrictions apply when calling MATLAB
functions extrinsically:
11-17
• MATLAB functions that inspect the caller or write to the caller’s workspace
do not work during code generation. Such functions include:
- dbstack
- evalin
- assignin
• The MATLAB debugger cannot inspect variables defined in extrinsic
functions.
• Functions in generated code may produce unpredictable results if your
extrinsic function performs any of the following actions at run time:
- Change folders
- Change the MATLAB path
- Delete or add MATLAB files
- Change warning states
- Change MATLAB preferences
- Change Simulink parameters
Limit on Function Arguments

You can call functions with up to 64 inputs and 64 outputs.
11-18
12
Generating Efficient and

Reusable Code
• “Generating Efficient Code” on page 12-2

• “Generating Reusable Code” on page 12-4
12 Generating Efficient and Reusable Code
Generating Efficient Code

In this section...
“Unrolling for-Loops” on page 12-2
“Inlining Functions” on page 12-2
“Eliminating Redundant Copies of Function Inputs” on page 12-2
Unrolling for-Loops
Unrolling for-loops eliminates the loop logic by creating a separate copy of the
loop body in the generated code for each iteration. Within each iteration, the
loop index variable becomes a constant. By unrolling short loops with known
bounds at compile time, MATLAB generates highly optimized code with no
branches.
You can also force loop unrolling for individual functions by wrapping
the loop header in an coder.unroll function. For more information, see
coder.unroll in the Code Generation from MATLAB Function Reference.
Inlining Functions
MATLAB uses internal heuristics to determine whether or not to inline
functions in the generated code. You can use the coder.inline directive to
fine-tune these heuristics for individual functions. See coder.inline in the
Code Generation from MATLAB Function Reference.
Eliminating Redundant Copies of Function Inputs

You can reduce the number of copies in your generated code by writing
functions that use the same variable as both an input and an output. For
example:
function A = foo( A, B ) %#codegen

A = A * B;
end
This coding practice uses a reference parameter optimization. When a

variable acts as both input and output, MATLAB passes the variable by
12-2
Generating Efficient Code
reference in the generated code instead of redundantly copying the input to a

temporary variable. For example, input A above is passed by reference in the
generated code because it also acts as an output for function foo:
...
/* Function Definitions */
void foo(real_T *A, real_T B)
{
*A *= B;
}
...
The reference parameter optimization reduces memory usage and improves

run-time performance, especially when the variable passed by reference is
a large data structure. To achieve these benefits at the call site, call the
function with the same variable as both input and output.
By contrast, suppose you rewrite function foo without using this optimization:
function y = foo2( A, B ) %#codegen

y = A * B;
end
In this case, MATLAB generates code that passes the inputs by value and
returns the value of the output:
...
/* Function Definitions */
real_T foo2(real_T A, real_T B)
{
return A * B;
}
...
12-3
12 Generating Efficient and Reusable Code
Generating Reusable Code

With MATLAB, you can generate reusable code in the following ways:
• Write reusable functions using standard MATLAB function file names

which you can call from different locations, for example, in a Simulink
model or MATLAB function library.
• Compile external functions on the MATLAB path and integrate them into
generated C code for embedded targets.
See “How MATLAB Resolves Function Calls in Generated Code” on page 11-2.
Common applications include:
• Overriding generated library function with a custom implementation

• Implementing a reusable library on top of standard library functions that
can be used with Simulink
• Swapping between different implementations of the same function
12-4
A
Examples
Use this list to find examples in the documentation.

A Examples
Data Management
Example: Defining a Variable for Multiple Execution Paths on page 4-4
Example: Defining All Fields in a Structure on page 4-5
“Defining Uninitialized Variables” on page 4-9
Variable Reuse in an if Statement on page 4-12
Code Generation for Structures

“Adding Fields in Consistent Order on Each Control Flow Path” on page 6-4
“Using repmat to Define an Array of Structures with Consistent Field
Properties” on page 6-7
“Defining an Array of Structures Using Concatenation” on page 6-8
“Making Structures Persistent” on page 6-9
Code Generation for Enumerated Data

“Defining and Using Enumerated Types for Code Generation” on page 7-13
“Using the if Statement on Enumerated Data Types” on page 7-18
“Using the switch Statement on Enumerated Data Types” on page 7-19
“Using the while Statement on Enumerated Data Types” on page 7-22
Code Generation for Variable-Size Data

“Tutorial: Generating MEX Code for a MATLAB Function That Expands a
Vector in a Loop” on page 8-9
“Example: Inferring Upper Bounds from Multiple Definitions with
Different Shapes” on page 8-21
“Constraining the Value of a Variable That Specifies Dimensions of
Variable-Size Data” on page 8-28
“Specifying the Upper Bounds for All Instances of a Local Variable” on
page 8-29
A-2

“Example: Defining and Passing Function Handles for Code Generation”
on page 9-3
Using Variable-Length Argument Lists

“Example: Using Variable Numbers of Arguments in a for-Loop” on page
10-5
“Example: Passing Variable Numbers of Arguments from One Function to
Another” on page 10-7
Optimizing Generated Code

“Eliminating Redundant Copies of Function Inputs” on page 12-2
A-3
A Examples
A-4
Index
A
Index resolving extrinsic function calls in generated
arguments code 11-16
limit on number for code generation from rules for defining uninitialized variables 4-8
MATLAB 11-18 setting properties of indexed variables 4-6
supported toolbox functions 11-10
unrolling for-loops 12-2
C using type cast operators in variable
C/C++ code generation for supported definitions 4-6
functions 2-1 variables, complex 5-4
code generation from MATLAB working with mxArrays 11-16
best practices for working with variables 4-3 coder.extrinsic 11-11
calling MATLAB functions 11-11 coder.nullcopy
calling MATLAB functions using feval 11-15 uninitialized variables 4-8
calling subfunctions 11-9 communications system toolbox System objects
characters 5-6 supported for code generation from
communications system toolbox System MATLAB 3-7
objects 3-7 computer vision system toolbox System objects
compilation directive %#codegen 11-8 supported for code generation from
computer vision system toolbox System MATLAB 3-3
objects 3-3
converting mxArrays to known types 11-17
D
declaring MATLAB functions as extrinsic
functions 11-11 defining uninitialized variables
defining persistent variables 4-10 rules 4-8
defining variables 4-2 defining variables
defining variables by assignment 4-3 for C/C++ code generation 4-3
dsp system toolbox System objects 3-13 dsp system toolbox System objects
eliminating redundant copies of function supported for code generation from
inputs 12-2 MATLAB 3-13
eliminating redundant copies of uninitialized
variables 4-8 E
function handles 9-1
eliminating redundant copies of function
generating efficient code 12-2
inputs 12-2
how it resolves function calls 11-2
extrinsic functions 11-11
initializing persistent variables 4-10
inlining functions 12-2
limit on number of function arguments 11-18 F
pragma 11-8 function handles
resolving extrinsic function calls during for code generation from MATLAB 9-1
simulation 11-15 functions
Index-1
Index
limit on number of arguments for code string functions 2-90

generation 11-18 structure functions 2-91
Functions supported for C/C++ code trigonometric functions 2-91
generation 2-1 Functions supported for MEX and C/C++ code
alphabetical list 2-3 generation
arithmetic operator functions 2-60 categorized list 2-59
bit-wise operation functions 2-61
casting functions 2-61
I
Communications System Toolbox
functions 2-62 indexed variables
complex number functions 2-62 setting properties for code generation from
Computer Vision System Toolbox MATLAB 4-6
functions 2-63 initialization
data type functions 2-64 persistent variables 4-10
derivative and integral functions 2-64
discrete math functions 2-64 M
error handling functions 2-65
MATLAB
exponential functions 2-65
features not supported for code
filtering and convolution functions 2-66
generation 1-11
Fixed-Point Toolbox functions 2-66
MATLAB for code generation
histogram functions 2-75
variable types 4-16
Image Processing Toolbox functions 2-75
MATLAB functions
input and output functions 2-76
and generating code for mxArrays 11-16
interpolation and computational geometry
mxArrays
functions 2-76
converting to known types 11-17
linear algebra functions 2-77
for code generation from MATLAB 11-16
logical operator functions 2-77
MATLAB Compiler functions 2-78
matrix/array functions 2-78 P
nonlinear numerical methods 2-82 persistent variables
polynomial functions 2-82 defining for code generation from
relational operator functions 2-82 MATLAB 4-10
rounding and remainder functions 2-83 initializing for code generation from
set functions 2-83 MATLAB 4-10
signal processing functions 2-84
Signal Processing Toolbox functions 2-84
special value functions 2-89 S
specialized math functions 2-89 signal processing functions
statistical functions 2-90 for C/C++ code generation 2-84
Index-2
Index
T variables
type cast operators eliminating redundant copies in C/C++ code
using in variable definitions 4-6 generated from MATLAB 4-8
Variables
defining by assignment for code generation
U from MATLAB 4-3
uninitialized variables defining for code generation from
eliminating redundant copies in generated MATLAB 4-2
code 4-8
V
variable types supported for code generation
from MATLAB 4-16
Index-3

User's Guide: Code Generation From MATLAB

Uploaded by

User's Guide: Code Generation From MATLAB

Uploaded by

Code Generation from MATLAB®

suggest@mathworks.com Product enhancement suggestions

The MathWorks, Inc.

About Code Generation from MATLAB

Prerequisites for Code Generation from MATLAB . . . . 1-3

Preparing MATLAB Code for C/C++ and MEX Code

Expected Differences in Behavior After Compiling Your

MATLAB Language Features Supported for Code

MATLAB Language Features Not Supported for Code

Related Products That Support Code Generation from

Functions Supported for Code Generation —

Functions Supported for Code Generation —

System Objects Supported for Code Generation

Computer Vision System Toolbox System Objects .... 3-3

Communications System Toolbox System Objects .... 3-7

DSP System Toolbox System Objects . . . . . . . . . . . . . . . . 3-13

Defining MATLAB Variables for C/C++ Code

Best Practices for Defining Variables for C/C++ Code

When You Can Reassign Variable Properties for C/C++

Eliminating Redundant Copies of Variables in

Defining and Initializing Persistent Variables . . . . . . . . 4-10

Reusing the Same Variable with Different

Supported Variable Types . . . . . . . . . . . . . . . . . . . . . . . . . . 4-16

Defining Data for Code Generation

Code Generation for Complex Data . . . . . . . . . . . . . . . . . 5-4

Code Generation for Characters . . . . . . . . . . . . . . . . . . . . 5-6

Code Generation for MATLAB Structures

Defining Scalar Structures for Code Generation . . . . . . 6-4

Defining Arrays of Structures for Code Generation . . . 6-7

Making Structures Persistent . . . . . . . . . . . . . . . . . . . . . . . 6-9

Indexing Substructures and Fields . . . . . . . . . . . . . . . . . . 6-10

Assigning Values to Structures and Fields . . . . . . . . . . . 6-12

Passing Large Structures as Input Parameters . . . . . . . 6-13

Code Generation for Enumerated Data

Enumerated Types Supported for Code Generation . . . 7-3

When to Use Enumerated Data for Code Generation . . 7-6

How to Define Enumerated Data for Code

How to Instantiate Enumerated Types for Code

How to Generate Code for Enumerated Data . . . . . . . . . 7-12

Defining and Using Enumerated Types for Code

Operations on Enumerated Data Allowed for Code

Using Enumerated Data in Control Flow Statements . . 7-18

Restrictions on Use of Enumerated Data in

Code Generation for Variable-Size Data

How Working with Variable-Size Data Is Different for

Bounded Versus Unbounded Variable-Size Data . . . . . . 8-4

When to Use Dynamic Allocation for Variable-Size

How to Generate Code for MATLAB Functions with

Tutorial: Generating MEX Code for a MATLAB

Variable-Size Data in Code Generation Reports . . . . . . 8-18

Defining Variable-Size Data for Code Generation . . . . . 8-20

Specifying Upper Bounds for Variable-Size Data . . . . . 8-27

C Code Interface for Unbounded Arrays and Structure

Troubleshooting Issues with Variable-Size Data . . . . . . 8-33

Limitations with Variable-Size Support for Code

Restrictions on Variable Sizing in Toolbox Functions

Code Generation for Function Handles

Example: Defining and Passing Function Handles for

Limitations with Function Handles for Code

Defining Functions for Code Generation

Supported Index Expressions . . . . . . . . . . . . . . . . . . . . . . . 10-3

Using varargin and varargout in for-Loops . . . . . . . . . . 10-4

Implementing Wrapper Functions with varargin and

Passing Property/Value Pairs with varargin . . . . . . . . . 10-8

Calling Functions for Code Generation

How MATLAB Resolves File Types on the Path for Code

Adding the Compilation Directive %#codegen . . . . . . . . 11-8

Calling Subfunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11-9

Calling Supported Toolbox Functions . . . . . . . . . . . . . . . 11-10

Calling MATLAB Functions . . . . . . . . . . . . . . . . . . . . . . . . . 11-11

Generating Efficient and Reusable Code

Generating Reusable Code . . . . . . . . . . . . . . . . . . . . . . . . . 12-4

Code Generation for Structures . . . . . . . . . . . . . . . . . . . . . A-2

Code Generation for Enumerated Data . . . . . . . . . . . . . . A-2

Code Generation for Variable-Size Data . . . . . . . . . . . . . A-2

Code Generation for Function Handles . . . . . . . . . . . . . . A-3