Author: Tobias Neumann

Programming language: C/C++11

Technologies: SSE2, AVX2, OpenCL, OpenMP

Source: tar, zip

versalignLib is a showcasing project of parallelization technologies comprising implementations of the sequence alignment algorithms Smith-Waterman and Needleman-Wunsch.

It builds a set of libraries, each library containing implementations utilizing a different parallelization technology (OpenMP, SSE2, AVX2, OpenCL). They are provided as shared objects and can be dynamically loaded depending on available resources and can be benchmarked to one another.

For any details on usage and benchmarks of the reference implementation, please refer to the Github wiki.

Building versalignLib requires cmake (>=2.8.11), g++ and an OpenCL implementation.

Typically cmake and g++ should be already available on your machine. If not, please install them.

To activate OpenMP parallelization, your compiler needs to have OpenMP support. Otherwise all Kernels except the OpenCL Kernel will run single-threaded. OpenMP is natively supported by gcc and for newer versions of Clang.

To build the OpenCL library, an OpenCL implementation must be available on your machine.

Mac OS X comes natively with an OpenCL implementation. No need for further setup.

For Linux, versalignLib utilizes the AMD OpenCL™ APP SDK. All you need to to is to make versalignLib aware of the AMD OpenCL™ APP SDK library by setting the following environment variables:

cd versalignLib
export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:$PWD/opencl-AMD-sdk/x86_64/lib"
export OPENCL_VENDOR_PATH="$PWD/opencl-AMD-sdk/x86_64/lib/vendors"

In case the symlink to the libOpenCL.so.1 library is not preserved, recreate it like this:

(cd opencl-AMD-sdk/x86_64/lib ; rm libOpenCL.so ; ln -s libOpenCL.so.1 libOpenCL.so)

cmake automatically checks whether AVX2 build support is available and will only produce the shared object if if detects support.

Caveat 1: You still need to check for AVX2 instruction support during runtime, for more see the versalignLib example implementation!

Caveat 2: OpenMP support for the AVX2 Kernel is currently disabled and needs more bugfixing!

To build versalignLib simply run the following commands:

cd versalignLib
mkdir build
cd build

# Release
cmake -DCMAKE_BUILD_TYPE=Release ..
# Debug
cmake -DCMAKE_BUILD_TYPE=Debug ..

make

CPU:	64 bit SSE2 enabled, (optional) AVX2 enabled
RAM:	Tested on regular systems with minimum 4 GB RAM
OS:	Linux (Ubuntu Server 16.04, Debian 7.7), Mac OSX (10.11)
Software:	cmake (>=2.8.11)