Parallel Implementation of db6 Wavelet Transform

Abstract

This work describes a data-level parallelizationstrategy to accelerate the discrete Wavelet transform, which wasimplemented and compared in two multi-threaded architectures,both with shared memory. The first considered architecturewas a multi-core server and the second one was a graphicprocessing unit. Comparisons were based on performance metrics(i.e. execution time, speeedup, efficiency, and cost) for fivedecomposition levels of the DWT Daubechies db6 over randomarrays of length 103, 104, 105, 106, 107, 108, and 109. Executiontimes in our proposed GPU strategy were around 1.2 × 10−5seconds, compare to 3501 × 10−5 seconds of the sequentialimplementation. On the other hand, the maximum achievablespeedup and efficiency was reached by our proposed multi-corestrategy for a number of assigned threads equal to 32.