Shoverand

A Pseudorandom Number Generation library for GPU

ShoveRand provides implementations of both Pseudorandom Number Generators (PRNGs) for CUDA-enabled GPUs and also a framework allowing developers to easily add their own PRNGs to ShoveRand.

Its API is quite similar to what can be encountered when using high-level CPU languages like C++ or Java. Shoverand's main goal is to handle the distribution of stochastic streams automatically without any intervention from the user.

Shoverand also targets PRNG developers: indeed, Shoverand only integrates third-party PRNGs and focuses on unifying their interface. To do so, we integrate compile-time constraints that check whether the algorithm meets our guidelines.

Requirements

• git
• CMake
• CUDA

Setting up Shoverand

Shoverand is mainly based upon header files to be included directly in your source code. Indeed, Shoverand highly relies on the C++ template mechanism to provide its random number generation facilities. However, some parts need to be compiled using the following cmake process.

mkdir build_<build_type>
cd build_<build_type>
cmake -D CUDA_SDK_ROOT_DIR=<path_to_cuda_sdk> -D CMAKE_BUILD_TYPE=<None | Debug | Release | RelWithDebInfo | MinSizeRel>
make

Usage

A simple example of how to use Shoverand can be found in example/montecarlo/montecarlo_pi.cu and test/test-shoverand.cu. When using Shoverand in your own project, do not forget to link against the libshoverand.so library that is built in the lib/ directory.

You can also find an example showing how to integrate a PRNG in shoverand in example/dummy_generator/dummy_generator.hxx.

These examples were presented as a tutorial at the HPCS 2012 conference, in Madrid (http://hpcs2012.cisedu.info/4-program/tutorials ==> T4).