Abstract
The recent years have seen an increasing demand in fast sequence align- ments, fuelled by a rapidly growing amount of sequence data. Meanwhile, with growing computing power optimal sequence alignment algorithms came back into the focus of these analyses. This work presents a new library for fast database searches based on optimal sequence alignments. It performs database searches accelerated on multiple threads and single instruction multiple data (SIMD) opera- tions. The library implements Rognes approach for accelerating database searches, while being designed for extensibility. A modular structure al- lows for an easy integration of new and improved algorithms. Addition- ally, the application programmable interface (API) of the library is designed for easy use and flexibility, allowing an extensive configuration of the com- putations. Besides the modular structure, the key features are the database searches based on SIMD instructions. These are optimised for the widely used streaming SIMD extensions (SSE) and the more recent advanced vector ex- tensions (AVX) implementing twice as wide registers. The focus of this thesis is the evaluation of the performance of libssa and of the optimised implementation of the database searches, with emphasis on the benefits of the computation on AVX over SSE. It presents that AVX significantly improves the performance by up to 1.83 times over SSE.