The Prague Stringology Conference 2009

M. Oğuzhan Külekci

Filter Based Fast Matching of Long Patterns by Using SIMD Instructions

Abstract:
SIMD instructions exist in many recent microprocessors supporting parallel execution of some operations on multiple data simultaneously via a set of special instructions working on limited number of special registers. Although the usage of SIMD is explored deeply in multimedia processing, implementation of encryption/decryption algorithms, and on some scientific calculations, it has not been much addressed in pattern matching. This study introduces a filter based exact pattern matching algorithm for searching long strings benefiting from SIMD instructions of Intel's SSE (streaming SIMD extensions) technology. The proposed algorithm has worst, best, and average time complexities of O(nm), O(n/m), and O(n/m + nm/216) respectively, while searching an m bytes pattern on a text of n bytes. Experiments on small, medium, and large alphabet text files are conducted to compare the performance of the new algorithm with other alternatives, which are known to be very fast on long string search operations. In all cases the proposed algorithm is the clear winner on the average. When compared with the nearest successor, the matching speed is improved in orders of magnitude on small alphabet sequences. The performance is 40 % better on medium alphabets, and 50 % on natural language text.

Download article: Article in PostScript Article in PDF BibTeX Reference
 PostScript   PDF   BibTeX reference 
Download presentation: Presentation