Are there known reasons for huge perfomance differences in different MATLAB-environments using the same mex-File?

I implemented the Viterbi-Algorithm in a mex-File for use in R2014a and R2009b, both x64. I don't encounter errors, but the performance is drastically different. R2014a sometimes takes more than 500 times as long. How do I debug for that?

EDIT: I compiled the file once, in R2014a using the 'Microsoft Visual C++ 2012 (C)' Compiler on a Windows 7 with the -largeArrayDims option and used it in both versions. I didn't have any errors or warnings with verbose mex-settings.

Example:

I created different versions of the Viterbi-Algorithm and tested their perfomance on both systems. The runtime should increase linearly like in R2009b (left), not quadratically like in R2014a (right) for Option 'tsp'. The only thing I could imagine causing this are different memory allocation mechanisms.

[Titus Edelhofer]

Hi Matthias,

to be sure, I assume you use the same C compiler? Or at least "similar"? And furthermore I assume it's not with debug settings once and optimization settings the other time?

Titus

[Dasharath Gulvady]

There are no such known issues. Are you facing any performance issue different versions of MATLAB with the same MEX file? Can you tell us what the file does? Can you share with us a sample code?

Which operating system are you on?

[James Tursa]

Does "same mex file" mean the exact same file? Or does it mean the same source code but compiled twice, once for R1009b and once for R2014a?