Main Content

Out-of-Process Execution of C++ MEX Functions

MATLAB® can run C++ MEX functions in a separate process. Running a C++ MEX function in a separate process enables you to:

  • Isolate the MATLAB process from crashes in the C++ MEX function.

  • Use some third-party libraries in the C++ MEX function that are not compatible with MATLAB.

  • Conserve memory by running multiple C++ MEX functions in a single process.

  • Attach C++ debuggers to the MEX host process.

  • Interrupt C++ MEX function execution using Ctrl+C

Note

The size of variables passed between C++ and MATLAB is limited to 2 GB when you call a C++ function out-of-process. This limit applies to the data plus supporting information passed between the processes.

How to Run Out of Process

Follow these steps to run your C++ MEX function out of process:

  1. Write your C++ MEX function and build it using the instructions in Build C++ MEX Programs. There are no required code differences between functions written for in-process and out-of-process execution.

  2. Create a MEX host process using the mexhost function.

  3. Use the feval method of the matlab.mex.MexHost object returned by mexhost to run your C++ MEX function in the host process.

When running out of process, C++ MEX functions always execute in the same folder as the MATLAB current folder. Changing the MATLAB current folder after creating the MEX host object results in the same change in the C++ MEX function context.

Running arrayProduct MEX Function Out of Process

The following example runs the arrayProduct C++ MEX function out-of-process. The C++ MEX source code is available in the file arrayProduct.cpp. To use this example C++ MEX function, open the arrayProduct.cpp source file, save it on your MATLAB path, and build the C++ MEX function using the instructions in Build C++ MEX Programs.

After building the C++ MEX function, start the MEX host process using the mexhost function. This function returns an object of the matlab.mex.MexHost class.

Use the feval method to run the arrayProduct C++ MEX function in the host process. Pass the C++ MEX functions name and arguments to feval. Return the results to MATLAB by assigning the output of feval.

mh = mexhost;
result = feval(mh,'arrayProduct',2,[1 2 3 4])
result =
     2     4     6     8

Run arrayProduct in another MEX host process. First create the MEX host process by calling mexhost again and assigning the output to a different variable. Then call feval on the new host.

mh2 = mexhost;
result2 = feval(mh2,'arrayProduct',2,rand(1,10));

You can run other C++ MEX functions in the same process using the same matlab.mex.MexHost object returned by mexhost.

result2 = feval(mh2,'anotherMexFunction',inputs);

Process Life Cycle

The MEX host process life cycle is coupled to the life cycle of the matlab.mex.MexHost object returned by mexhost. Like any handle object, MATLAB invokes the delete method when the object is no longer referenced anywhere. If MATLAB invokes delete or if you call delete on the object explicitly, MATLAB ends the process associated with the MexHost object.

You can end the process explicitly by calling the clear function with any of these options:

  • clear the variable returned by the mexhost function

  • clear classes, clear java, or clear all

Rebuilding a C++ MEX Function

The mex command automatically unloads C++ MEX functions before rebuilding them. Therefore, you do not need to explicitly unload C++ MEX functions from the MEX host process.

You can explicitly unload all C++ MEX functions from a MEX host process by calling clear mex or clear functions. Unload a specific C++ MEX function by calling clear on the function name.

To prevent the unloading of a C++ MEX function, use the mexLock function. To unlock the C++ MEX function, use mexUnlock. These functions set or unset a flag for the C++ MEX function on the host process to control the unloading of C++ MEX functions.

Getting Information About the MEX Host Process

Use the matlab.mex.MexHost object to get information about the MEX host process. These properties provide information about the process and loaded functions.

  • EnvironmentVariables contains names and values of environment variables set for the process.

  • Functions contains the names of all C++ MEX functions loaded into the MEX host process.

  • ProcessName contains the name of the MEX host process, which is MATLABMexHost by default.

  • ProcessIdentifier contains the process identifier.

For more information on these properties, see matlab.mex.MexHost.

Always Run Out of Process

In certain cases, you might always want to run your C++ MEX function out-of-process. For example, running out of process can be necessary if there are third-party library conflicts between MATLAB and the C++ MEX function.

To provide a convenient way to run the C++ MEX function out of process, create a wrapper MATLAB function. The wrapper function creates the MEX host object and runs the C++ MEX function in that process.

For example, suppose that you want to create a wrapper function for the arrayProduct.cpp C++ MEX function that is included in the documentation as an example. Create a MATLAB function with the name arrayProduct.m and put this file in a folder that is on the MATLAB path. Build the C++ MEX function and assign a different name for the compiled MEX file.

The wrapper function creates a matlab.mex.MexHost object that is assigned to a persistent variable so that the process is not destroyed between subsequent calls to the wrapper function. The function uses this object to call the feval method for out-of-process execution of the C++ MEX function. If the MEX host object is not valid, then the function creates a MEX host process.

function result = arrayProduct(scalefactor,inputarray)
    persistent mh
    if ~(isa(mh,'matlab.mex.MexHost') && isvalid(mh))
        mh = mexhost;
    end
    result = feval(mh,"arrayProductMEX",scalefactor,inputarray);
end

To build the C++ MEX function, open the arrayProduct.cpp source file and save it on your MATLAB path. Build the C++ MEX function using the instructions in this topic, Build C++ MEX Programs.

The following command builds MEX file with the root name arrayProductMEX and puts it in the folder with the wrapper function, which is assumed to be MyPathFolder in this example. The mex command creates the folder if it does not exist.

mex -output arrayProductMEX -outdir MyPathFolder arrayProduct.cpp

To use the C++ MEX function, call it from the command line like any function via the wrapper function.

result = arrayProduct(2,[1 2 3 4]);
result
result =

     2     4     6     8

The following code calls the C++ MEX function in a for loop. The first call to arrayProduct creates the MEX host process. Subsequent calls use the same process unless the process is destroyed by, for example, a crash of the C++ MEX function.

for k = 1:5
   results(k,:) = arrayProduct(k,[1 2 3 4]);
end
results
results =
     1     2     3     4
     2     4     6     8
     3     6     9    12
     4     8    12    16
     5    10    15    20

To destroy the MEX host process, clear all functions that define the MEX host process variable (mh in this example) as persistent. In this example, clear the arrayProduct.m function.

clear arrayProduct.m

Calling clear functions causes the destruction of the matlab.mex.MexHost object that is stored in the persistent variable and therefore terminates the MEX host process.

Debugging Out-of-Process MEX Functions

You can attach a C++ debugger to your C++ MEX function and set break points to debug your program. Here are the steps to set up and debug a C++ MEX function out of process.

  • Build C++ MEX source code using the mex command with the -g option to include debugging symbols.

  • Create a host process using the mexhost function. This function returns an object of the matlab.mex.MexHost.

  • Get the process identifier from the matlab.mex.MexHost object ProcessIdentifier property.

  • Use the process identifier to attach a C++ debugger to the process.

  • Insert breakpoints in the source code.

  • Run the C++ MEX function out-of-process using the feval method.

For information on using specific debuggers, see the documentation for those debuggers.

Debugging Using Microsoft Visual Studio

  1. Ensure Visual Studio® is your selected C++ compiler. This example uses Microsoft® Visual Studio 2015.

    cpp = mex.getCompilerConfigurations('C++','Selected');
    cpp.Name
    ans =
    
        'Microsoft Visual C++ 2015'
  2. Build your C++ MEX source code using the mex command with the -g option. This example assumes that you have a C++ MEX source file named myMexFunction.cpp.

    mex -g myMexFunction.cpp
  3. Create a MEX host process and return the MexHost object.

    mh = mexhost;
  4. Start Visual Studio. Do not exit your MATLAB session.

  5. From the Visual Studio Debug menu, select Attach to Process.

    In the Attach to Process dialog box, select the MATLABMexHost process, and click Attach.

  6. Visual Studio loads data, and then displays an empty code pane.

  7. Open your C++ MEX source file by clicking File > Open > File and selecting the file.

  8. Set a breakpoint by right-clicking the desired line of code and clicking Breakpoint > Insert Breakpoint on the context menu.

  9. In MATLAB, run the C++ MEX function out of process using the matlab.mex.MexHost feval method.

    result = feval(mh,'myMexFunction',input1,input2,...)
  10. Use the features provided by the debugger to debug your source code.

Debugging on Linux Systems

On Linux® systems, you can use a debugger such as the GNU® gdb debugger. Follow these steps to use the gdb debugger.

  1. Build your C++ MEX source code using the mex command with the -g option. This example assumes that you have a C++ MEX source file named myMexFunction.cpp.

    mex -g myMexFunction.cpp
  2. Create a MEX host process and return the MexHost object.

    mh = mexhost;
  3. Get the process identifier from the ProcessIdentifier property of the MexHost object. The returned value is a string representing the identifier of the MEX host process. For example,

    mh.ProcessIdentifier
    ans = 
    
        "13892"

    The process identifier is different for each process created.

  4. Attach a debugger to the MEX host process from the Linux terminal. For example, using the GNU gdb debugger, call gdb with the C++ MEX file name and the process identifier that you obtained from the MEX host object in MATLAB:

    gdb myMexFunction -pid=13892
  5. Set break points in the C++ MEX function. For example, using gdb, set the break point in myMexFunction.cpp at line 21:

    break myMexFunction.cpp:21
  6. From MATLAB, run the C++ MEX function out-of-process using the matlab.mex.MexHost feval method.

    result = feval(mh,'myMexFunction',input1,input2,...)

    MATLAB waits for a response from the debugger.

  7. From the Linux terminal, use the features provided by the debugger to debug your source code.

Debugging on Macintosh Systems

On Macintosh systems, use the LLDB debugger.

  1. Build your C++ MEX source code using the mex command with the -g option. This example assumes that you have a C++ MEX source file named myMexFunction.cpp.

    mex -g myMexFunction.cpp
  2. Create a MEX host process and return the MexHost object.

    mh = mexhost;
  3. Get the process identifier from the ProcessIdentifier property of the MexHost object. The returned value is a string representing the identifier of the MEX host process.

    mh.ProcessIdentifier
    ans = 
    
        "13892"

    The process identifier is different for each process created.

  4. Attach a debugger to the MEX host process from the macOS Terminal. Call lldb with the MEX host process identifier that you obtained from the MEX host object in MATLAB. For example, assuming the process identifier is 13892, attach the LLDB debugger to this process:

    lldb -p 13892
  5. Set a break points in your source code. For example, this command sets a break point in myMexFunction.cpp at line number 21.

    breakpoint set -f myMexFunction.cpp -l 21
  6. From MATLAB, run the C++ MEX function out-of-process using the matlab.mex.MexHost feval method.

    result = feval(mh'myMexFunction',input1,input2,...)

    MATLAB waits for a response from the debugger.

  7. Enter C or continue from the macOS Terminal. Program execution stops at breakpoints.

  8. From the macOS Terminal, use the features provided by the debugger to debug your source code.

See Also

|

Related Topics