[Quantlib-dev] Parallel version of QuantLib (sort of)

[Kakhkhor A. <kab...@gm...>]

Dear All,

I implement a parallel C++ template library similar to QuantLib. The
source code and the manual are available from the project's homepage
http://code.google.com/p/qlp/. The library is called QLP -  short for
quantitative library parallel.

At this time only equity Monte-Carlo engines and several analytical
formulas are fully implemented. Work in progress to finish LMM
engines.

One of the main goals was to achieve superior performance by means of
multi-threading and SSE/SSE2 instructions. Another main goal is to
facilitate the development of optimized parallel engines.

According to the preliminary benchmarks of arithmetic Asian engines,
QLP engines run about 50x(number of cores) times faster than similar
engines from QuantLib. On quad core 2.4 GHz platform the speed was
about 4 times slower than the speed achieved with Tesla C1060 GPU.
Tesla timing was taken from the published paper.

The project is still in pre-release state, but there are quite a few
thing that are fully implemented (more or less). These are mainly low
level building blocks: RNGs, inverse  CDF, Brownian Bridge, etc. High
level "stuff" isn't very stable yet. I would appreciate if you
download the source code, give it a try and then let me know your
opinion and suggestions.

The license is GPL3. However, if if anyone decides to port something
into QuantLib I can submit the source code to the QuantLib project.
The GPL3 is required by GSL, but GSL is used for numerical integration
only. QuantLib has its own quadratures. Replacing GSL shouldn't be a
problem.

Regards,
Kakhkhor Abdijalilov.