How Do I Use clib to Call a C++ Function with a float Argument?

I am trying to publish a C++ library interface per this documentation: Publish Interface to Shared C++ Library. This process is also described in this blog post by Vivek Bhownani. I'm currently building a simple HelloWorld dll and the process works fine when my function only has integer arguments and return values. As soon as I add a function that takes float arguments, the library interface builds just fine, but calling the function with a float argument returns the following error in Matlab:

User Library threw non-standard exception with message:'Can't convert the Array to this TypedArray'.

This is a simple scalar argument and return value, after reviewing the list of unsupported C++ features I think that this simple library call should work. My library header and source are posted below, can anyone provide some advice? Would love to hear from Vivek!

DLL Header file:

#pragma once

#include "windows.h"

#ifdef TESTDLL_EXPORTS
#define MATHLIBRARY_API __declspec(dllexport)
#else
#define MATHLIBRARY_API __declspec(dllimport)
#endif

// HelloWorld functions
MATHLIBRARY_API unsigned long square_int(unsigned long a);

MATHLIBRARY_API float square_float(const float a);

DLL Source file:

#include "my_test_dll.h"

unsigned long square_int(unsigned long a)
{
    return a * a;
}

float square_float(const float a)
{
    return a * a;
}

The script that generates the Matlab interface:

% Full path to files in the library
productPath = 'C:\git_checkouts\temp\test_dll';
% Header file name
hppFile = 'my_test_dll.h';
% Full path to folder containing all header files
hppPath = productPath;
% Full path to folder containing include files
iPath = hppPath;
% Full path to folder containing library files
libPath = fullfile(productPath,'x64','Release');
% Library file name
libFile = 'test_dll.lib';

myPkg = 'myPackage';

clibgen.generateLibraryDefinition(fullfile(hppPath,hppFile),...
'IncludePath', iPath,... 
'Libraries', fullfile(libPath,libFile),... 
'PackageName', myPkg,...
'Verbose',true)

%% Build the library interface from the just-generated definition file
build(definemyPackage);

%% Add the path to the dll to the system path. This is temporary for this instance of Matlab!
sysPath = getenv('PATH');
setenv('PATH',[sysPath ';' libPath])
getenv('PATH');

%% Finally, add the newly created Matlab interface file to the path
addpath('myPackage');

And finally, calling the int and the float functins from the Matlab command line gives the following:

>> clib.myPackage.square_int(4)

ans =

  uint32

   16

>> clib.myPackage.square_float(4.5)
Error using clib.myPackage.square_float
User Library threw non-standard exception with message:'Can't convert the Array to this TypedArray'.
 

[Vivek]

Hi,

I saw your post on my blog, but I thought I'd address this question right here. Thank you for reporting this behavior. I've made a note of it, and we'll be fixing it soon. As for the record floating point types are supported in the MATLAB C++ interface, beginning in 19a.

Vivek

[Peter Hristov]

Hi, Vivek!

Is there any progress on this issue? I am also getting this error and suspect it is to do with the 'const' modifier in C++.

Answer 1

[SpeedMan2020]

Ok, a bit more tweaking and I've got this issue fixed. The issue was the const modifier on the floating point argument to square_float(), apparently Matlab doesn't like this. Could be a very useful thing to sort out in the next release. When I remove the const modifier and rebuild the dll and the Matlab interface, I get the following results at the command line:

>> clib.myPackage.square_int(4)

ans =

  uint32

   16

>> clib.myPackage.square_float(4.5)

ans =

  single

   20.2500

I hope this helps in case anyone else runs into this problem.