Hi everybody,I need your expertise... I will start writing a Scientific program (a numerical simulator for Differential Algebraic Equation system). The default language most people in Power Engineering use is Fortran. We currently have a very fast software developed in fortran. The software, though, is getting so big with so many features that is very hard to maintain (a lot of code is duplicated with minor changes all over the program) in its current form. Also, a lot of the features that will be implemented in the future involve threading, running same algorithms on different data sets concurenlty, running on cluster etc. I want to rewrite the software in an object oriented form in order to make it more maintainable, expandable and reusable. Also, I'm interested in trying Intel TBB for the threading needs and a GUI library for setting up an basic interface for the program. My concern is: What about the speed of the program? Is C++ able to compare with Fortran? What Mathematical libraries should I use? The current fortran implementation uses: -elemental mathematical expressions (pow, exp, mult etc) -dgetrf and dgetrs from BLAS95 (MKL library) are called million times during a simulation with matrices ranging 10-30 elements -a sparse solver (ma37 - HSL mathematical software library) is called several times on a big (30000+ element matrice) I currently use the newest version of Intel Parallel Studio 2011 with MKL. Also, my university owns the Intel cluster toolkit. Many people advocate against the performance of C++ for scientific reasons (use of temporary copies etc). I also found several libraries (blitz++, goto, a++, atlas, eigen ...) promising near-fortran peformance. Does anyone have experience with this kind of numerically intensive implementations using C++? If I use the tools (pow, exp, matrice mult, blas functions, sparse solvers etc) offered by MKL will I get good performance? I am not looking for a definitive answer, since I know it's a difficult question. Any kind of help or insight is welcome. Petros
For more complete information about compiler optimizations, see our Optimization Notice.