The C++ library wrapper option allows you to create a shared library from an arbitrary set of MATLAB® files. MATLAB Compiler SDK™ generates a wrapper file and a header file. The header file contains all of the entry points for all of the compiled MATLAB functions.
This example rewrites the C shared library example using C++.
The procedure for creating a C++ shared library from MATLAB files
is identical to the procedure for creating a C shared library, except
you use the cpplib wrapper. Enter the following
command on a single line:
mcc -W cpplib:libmatrixp -T link:lib addmatrix.m multiplymatrix.m eigmatrix.m -v
The -W cpplib:<libname> option tells MATLAB
Compiler SDK to
generate a function wrapper for a shared library and call it <libname>.
The -T link:lib option specifies the target output
as a shared library. Note the directory where the product puts the
shared library because you will need it later.
Due to name mangling in C++, you must compile your driver application with the same version of your third-party compiler that you use to compile your C++ shared library.
In the C++ version of the matrixdriver application matrixdriver.cpp,
arrays are represented by objects of the class mwArray.
Every mwArray class object contains a pointer to
a MATLAB array structure. For this reason, the attributes of
an mwArray object are a superset of the attributes
of a MATLAB array. Every MATLAB array contains information
about the size and shape of the array (i.e., the number of rows, columns,
and pages) and either one or two arrays of data. The first array stores
the real part of the array data and the second array stores the imaginary
part. For arrays with no imaginary part, the second array is not present.
The data in the array is arranged in column-major, rather than row-major,
order.
Avoid issuing cd commands from the driver
application prior to calling mclInitializeApplication.
Failure to do so can cause a failure in MATLAB Runtime initialization.
For information about how MATLAB Compiler SDK uses a proxy layer for the libraries that an application must link, see Understand the mclmcrrt Proxy Layer.
To compile the matrixdriver.cpp driver code,
you use your C++ compiler. By executing the following mbuild command
that corresponds to your development platform, you will use your C++
compiler to compile the code.
mbuild matrixdriver.cpp libmatrixp.lib (Windows) mbuild matrixdriver.cpp -L. -lmatrixp -I. (UNIX)
This command assumes that the shared library and the corresponding header file are in the current working directory.
On Windows®, if this is not the case, specify the full path
to libmatrixp.lib, and use a -I option
to specify the directory containing the header file.
On UNIX®, if this is not the case, replace the “.”
(dot) following the -L and -I options
with the name of the directory that contains these files, respectively.
There are two main differences to note when using a C++ shared library:
Interface functions use the mwArray type
to pass arguments, rather than the mxArray type
used with C shared libraries.
C++ exceptions are used to report errors to the caller.
Therefore, all calls must be wrapped in a try-catch block.
The C++ shared library target generates two sets of interfaces for each MATLAB function. For more information, see Functions Generated from MATLAB Files. The generic signature of the exported C++ functions is as follows:
MATLAB Functions with No Return Values
bool MW_CALL_CONV <function-name>(<const_mwArray_references>);
MATLAB Functions with at Least One Return Value
bool MW_CALL_CONV <function-name>(int <number_of_return_values>,
<mwArray_references>, <const_mwArray_references>);
In this case, const_mwArray_references represents
a comma-separated list of references of type const mwArray& and mwArray_references represents
a comma-separated list of references of type mwArray&.
For example, in the libmatrix library, the C++
interface to the addmatrix MATLAB function
is generated as:
void addmatrix(int nargout, mwArray& a, const mwArray& a1,
const mwArray& a2);where a is an output parameter and a1 and a2 are
input parameters.
Input arguments passed to the MATLAB function via varargin must
be passed via a single mwArray that is a cell array.
Each element in the cell array must constitute an input argument.
Output arguments retrieved from the MATLAB function via varargout must
be retrieved via a single mwArray that is a cell
array. Each element in the cell array will constitute an output argument.
The number of elements in the cell array will be equal to number_of_return_values -
the number of named output parameters. Also note that,
If the MATLAB function takes a varargin argument,
the C++ function must be passed an mwArray corresponding
to that varargin, even if the mwArray is
empty.
If the MATLAB function takes a varargout argument,
the C++ function must be passed an mwArray corresponding
to that varargin, even if number_of_return_values is
set to the number of named output arguments, which means meaning that varargout will
be empty.
The varargout argument needs to
follow any named output arguments and precede any input arguments.
The varargin argument needs to
be the last argument.
C++ interface functions handle errors during execution by throwing
a C++ exception. Use the mwException class for
this purpose. Your application can catch mwExceptions and
query the what() method to get the error message.
To correctly handle errors when calling the C++ interface functions,
wrap each call inside a try-catch block.
try
{
...
(call function)
...
}
catch (const mwException& e)
{
...
(handle error)
...
}
The matrixdriver.cpp application
illustrates the typical way to handle errors when calling the C++
interface functions.
The MATLAB Compiler SDK C/C++ API includes static factory methods for working with sparse arrays.
For a complete list of the methods, see C++ Utility Classes.