ParaStats

Parallel computation of statistical values

When working with data sets created from natural processes, it is often important to get statistical values for these data-sets. This assignment implements the calculation of statistical values over large data sets in parallel. The algorithms are implemented in OpenCL (Open Computing Language), a (low level) language based on C (C99) and an API for implementing high performance and parallel computing code. OpenCL's goal is to be cross platform, in particular for the programming of acceleration units, such as graphics cards (GPUs) or FPGAs. Specifically we have implemented and accelerate the calculation of statistical values for large input data sets.

Authors

• Blair Cox (bcox280)
• Logan Small (lsma604)
• Michael Kemp (mkem114)

License

• MIT

Running the executable

You will need to install a runtime, and implementation of OpenCL:

OSX (not tested)

• OpenCL runtimes are provided by default by Apple

Linux (Debian-based and Arch-based tested)

• You'll need a loader; ocl-icd-libopencl1 on Debian-based systems, libOpenCL1 on OpenSuse-based systems, ocl-icd for Arch-based systems
• NVIDIA devices, install nvidia-XXX drivers for Debian-based systems, opencl-nvidia for Arch-based systems
• AMD devices, (untested), install opencl-mesa on Arch-based systems
• Intel devices, intel-opencl-runtime and beignet on Arch-based systems
• To check which OpenCL device implementations are currently on your system, ls /etc/OpenCL/vendors (you'll need at least 1 for running the OpenCL implementation)

Windows (tested)

• Install the relevant OpenCL runtimes/drivers for your devices from their websites
• The provided executable (under releases) works on UG4 windows installation, you'll need to source the OpenCL.dll (from your device driver), or find one online.

Why won't it run?

• Make sure you have the src/kernels/ folder, this contains the critical OpenCL kernel(s)
• Make sure the executables are in the bin/ folder
• Make sure the sample data files (if you're using them) are in the data/ folder

Compiling

OSX (not tested)

• Apple does not provide OpenCL C++ header files, you'll have to download one and include it the compilation process
• Run the universal compile commands

Linux (cinnamon, arch tested)

• To compile OpenCL you need the -lOpenCL library, install ocl-icd-opencl-dev on Debian-based, ocl-icd on Arch-based and OpenSuse-based
• You will also need the headers/bindings, install opencl-headers
• If it can't find the library, you may need a symbolic link: e.g. ln -s /usr/lib/x86_64-linux-gnu/libOpenCL.so.1 /usr/local/lib/libOpenCL.so in Debian-based systems or ln -s /usr/lib64/libOpenCL.so.1 /usr/lib64/libOpenCL.so in OpenSuse
• Run the universal compile commands

Windows (tested)

• Installation of an OpenCL SDK should provide the header files (the Intel SDK is what we used and was least painful)
• For compiling we installed MinGW-w64 (as recommended by JetBrains) and Visual Studio C++ Compiler is not tested
• To compile either install CLion and it will manage the process for you or run the universal compile commands in the MinGW-w64 environment

Universal compile commands

cmake .
make SOFTENG-751-Assignment-ParaStats
make SOFTENG-751-Assignment-ParaStats-Test