32bit and 64bit Matlab mysterious discrepancy

Hello,

I have come across a very strange behaviour: while calling a shared library from Matlab, it seems that I cannot pass floating point values correctly when running Matlab 64bit. Matlab 32bit works fine. Here an example to illustrate what I mean:

The shared library function:

extern __declspec(dllexport) float myTest(float b)
{
  return(b); // just send back the received value
}

The prototype file:

..
fcns.name{fcnNum}='myTest';
fcns.calltype{fcnNum}='cdecl';
fcns.LHS{fcnNum}='single';
fcns.RHS{fcnNum}={'single'};
fcnNum=fcnNum+1;
..

From the Matlab command line, I type:

loadlibrary('mySharedLibrary',@sharedLibraryPrototypeFile);
calllib('mySharedLibrary','myTest',1.234)

I have 2 versions of my shared library: one compiled in 32bit, the other one compiled in 64bit.

- From Matlab32, the 32bit library is loaded and above function called. The return value is always the sent value (in above example, "1.234")

- From Matlab64, the 64bit library is loaded and above function called. The return values is never the sent value (in above example, the return value is "-3.6910e+05"). It is as if the 64bit Matlab was allergic sending float32 (single) values to my shared library. Receiveing float32 works fine (i.e. if I hard code "return(1.234f)" in my shared library).

If in above functions I replace the float argument with an integer argument (int32), both Matlab work as expected.

Does anyone have a clue what mysterious thing is going on?

Cheers