Author
Keywords
Matlab, C, C++, Mex Files
Review Status
General review by COMISEF Wiki Admin, 03/2009
1. Tutorial Overview
This is the first of five Tutorials that describe how to optimise Matlab programs using C/C++ functions. For an exhaustive look at optimisating Matlab programs see Matlab's Help Guide. The objective of this Tutorial is to simply help you get started with MEX Files with practical examples with a finance theme. In this first Tutorial we give an overview to optimising code using C/C++, and look at some of the reason for and against doing this.
The Titles and links to all Tutorials are listed below.
 An Overview of using C/C++ under Matlab
 Creating a project under Visual C++ Express 2008
 Basic Principles
 Advanced Principles
 Conclusion
2. Introduction
Matlab is known to be computationally efficient, especially when dealing with problems structured as matrices. However there may be some situations where using matrices is not recommended (where it is a unacceptable burden on memory resources), or even impossible (owing to extreme dimensionality of the problem).
One solution is to go for some compiled language. (As opposed tp Matlab, which is interpreted language.) Performance may improve drastically by using compiled functions directly under Matlab. As a matter of fact a lot of default builtin Matlab functions are compiled to improve performance.
3. Advantages of compiled languages
Some Advantages of using compiled languages under Matlab include:
 Compiled functions (improved efficiency)
 Lowlevel control of memory
 Control of multithread coding
 Interaction with Matlab (it is possible to call standard m routines in your C/C++ code, but it is quite slow)
 Extreme flexibility in the fact that it is possible to use either Matlab features or C/C++ features
 Control of I/O
 Possibility to use external libraries
4. The Downside
The flexibility offered by compiled languages may be very powerful when used carefully. However all the difficulties following naturally from using complex languages arise and therefore:
 Do not go down the C/C++ route unless you understand how to use it properly.
What is better? Using exisiting Matlab code that takes 10 days to run but outputs the expected answer. Or spend 10 full days writing and debugging C/C++ code to get an incorrect answer.
 Do not try to optimize if it is not worth it.
As already mentioned, only use C/C++ unless if it is really necessary. For example, when there is a clear evidence after some diagnostic tests that it will alleviates/solves a performance bottneck. Remember that Matlab is very efficient in most cases, especially for applying matrix operations. All aspects of programming are usually already optimized (like the number of functions calls, memory allocation […]), so do attempt to optimize Matlab functions unless you are pretty confident that you can do better.
 C/C++ will not make your algorithm better.
NPcomplex problems under Matlab are still NPcomplex under C/C++, and usually it is even more like NP+ bug.
 Do not use C/C++ if you are only interested in relative performance.
5. Recommendation
In my opinion a safe way to use C/C++ functions is to focus on specific performance bottlenecks in your code. As a simple recipe:
 First implement your algorithm in Matlab M files
 Check that outputs are correct (or at least coherent) before introducing some other potential sources of errors.
 Use tic and toc Matlab functions (or any profiler you like) to analyse where exactly your implementation is inefficient. This will really help you to understand if compiled methods are required.
 Rewrite the small pieces of slow code in C/C++.
Sometimes, you may have to rewrite all your code,but remember that a lot of libraries (like Boost) can help you in your transition to C/C++.
Internal Links
Concepts 
Tutorials 
Tips 
Related Articles 

External links
References 
Weblinks 