optim.coder.infbound

Infinite bound support for code generation

Since R2022b

Syntax

bnd = optim.coder.infbound

bnd = optim.coder.infbound(sz)

bnd = optim.coder.infbound(n1,n2,...,nt)

bnd = optim.coder.infbound(___,typename)

bnd = optim.coder.infbound(___,"like",p)

Description

bnd = optim.coder.infbound creates an infinite bound for use in code generation. bnd represents an infinite bound for code generation targets that do not support Inf.

bnd = optim.coder.infbound(sz) returns an infinite bound array of the same size as the argument sz. If sz is a scalar, the size of the returned array is sz-by-sz.

bnd = optim.coder.infbound(n1,n2,...,nt) returns an infinite bound array of size n1-by-n2-by-...-by-nt for the scalar values n1, n2,…,nt.

bnd = optim.coder.infbound(___,typename) specifies the data type (class) of bnd for any of the previous syntaxes.

bnd = optim.coder.infbound(___,"like",p) specifies that bnd has the same as the data type as the numeric variable p.

Examples

Create Infinite Bounds for Code Generation

Target hardware for code generation does not always accept Inf or –Inf as a bound. In this case, use optim.coder.infbound to represent the infinite bounds.

For example, suppose your problem has upper bounds of [Inf,1,10], meaning x(2) <= 1 and x(3) <= 10, and no upper bound on x(1). The problem also has lower bounds of [0,–Inf,0], meaning x(1) >= 0 and x(3) >= 0, and no lower bound on x(2). You can represent these bounds in several ways.

ub1 = [optim.coder.infbound 1 10]

ub1 = 1×3

   Inf     1    10

lb1 = [0 -optim.coder.infbound 0]

lb1 = 1×3

     0  -Inf     0

% Or
ub = optim.coder.infbound(1,3);
ub(2) = 1;
ub(3) = 10

ub = 1×3

   Inf     1    10

lb = -optim.coder.infbound([1,3]);
lb(1) = 0;
lb(3) = 0

lb = 1×3

     0  -Inf     0

Some target hardware supports only "single" data. Use the typename or "like" syntaxes to generate appropriate bounds.

ub2 = [optim.coder.infbound("single") single(1) single(10)]

ub2 = 1x3 single row vector

   Inf     1    10

s0 = single(0);
lb2 = [s0 -optim.coder.infbound("like",s0) s0]

lb2 = 1x3 single row vector

     0  -Inf     0

These results are produced from running the code in MATLAB®, and show the MATLAB Inf output. When you run optim.coder.infbound for code generation, the resulting bounds are appropriate for the target hardware.

Copyright 2022–2024 The MathWorks, Inc.

Input Arguments

sz — Size of returned Inf array

vector of nonnegative integers

Size of the returned Inf array, specified as a vector of nonnegative integers. Generally, the size of the returned array is sz(1)-by-sz(2)-by-...-by-sz(t), where t is the number of components of sz. However, if sz is a scalar, the size of the returned array is sz-by-sz.

If any entry in sz is zero, the returned array is empty.

Example: [2,3,1,4]

Data Types: double

n — Size of Inf array component

nonnegative integer

Size of an Inf array component, specified as a nonnegative integer. If n is 0, the returned array is empty.

Example: 2

Data Types: double

typename — Output data type

"double" (default) | "single"

Output data type, specified as "double" or "single". If you Disable Support for Nonfinite Numbers (MATLAB Coder) (for example, by setting NonFiniteSupport to false in the associated code generation configuration object), the output entries are realmax(typename).

You cannot specify both typename and "like",p.

Currently, only fmincon supports single-precision code generation.

Data Types: char | string

p — Output data prototype

numeric variable of class "double" (default) | numeric variable of class "single"

Output data prototype, specified as a numeric variable of class "double" or "single". If you Disable Support for Nonfinite Numbers (MATLAB Coder) (for example, by setting NonFiniteSupport to false in the associated code generation configuration object), the output entries are realmax(class(p)).

You cannot specify both typename and "like",p.

Currently, only fmincon supports single-precision code generation.

Data Types: char | string

Extended Capabilities

C/C++ Code Generation

Generate C and C++ code using MATLAB® Coder™.

Code generation for Optimization Toolbox™ solvers is subject to the requirements noted in Code Generation in fmincon Background, Code Generation for quadprog Background, Code Generation in Linear Least Squares: Background, and Code Generation in Nonlinear Least Squares: Background. In particular:

• Solvers support code generation using either the codegen (MATLAB Coder) function or the MATLAB^® Coder™ app. You must have a MATLAB Coder license to generate code.

• For lsqlin, the lb and ub arguments must have the same number of entries as the number of columns in C or must be empty []. Otherwise, the lb and ub arguments must have the same number of entries as the x0 argument or must be empty [].

• Only fmincon can generate code for single-precision hardware. For details, see Code Generation in fmincon Background.

Version History

Introduced in R2022b

R2024a: Inclusion of typename and "like" arguments

To support single-precision code generation, optim.coder.infbound now accepts the typename and "like",p arguments.

Currently, only fmincon supports single-precision code generation.