Parallel Random Number Generation

Random number generators have been studied and used for decades, and various kinds of generators have been proposed and improved to fit different types of problems. Better generators fit the problem tightly and utilize the architecture fully. Under current architecture, multiple processor cores enable simultaneous execution of independent computational threads. High-performance computing uses programs with multiple threads. Random number generators are being studied as a source of independent, paralleled, and reliable streams. Parallelization of random number generators is not trivial; different schemes and approaches have been proposed and scrutinized. In my work, correlations of random number streams will be examined from the perspective of computational finance. I extended the support of SPRNG to shared memory, more specifically OpenMP. I implemented some of the generators in SPRNG in GPU, with completely redesigned GPU-oriented data structure and optimizations. In supplemental files, generators of SGLCG, ALFG, and MLFG are put into three separate packages accordingly. These packages are not the final release version.