Paralelní simulace biomakromolekul metodou DFTB

Abstract

Cilem teto prace je akcelerace DFTB metody pouzivane k simulaci biomakromolekul. Casove nejnarocnejsi casti simulace jsou identifikovany a akcelerovany s pouzitim internich moznosti puvodniho reseni, paralelizace pomoci OpenMP a moznosti pouzit knihovnu MAGMA, ktera vyuziva hybridnich CPU/GPU algoritmu ke zrychleni operaci linearni algebry. Jednotlive metody akcelerace jsou popsany a otestovany na systemech vod ruznych velikosti. Vysledky testu, prokazujici zrychleni oproti jinym knihovnam implementujicim operace linearni algebry, minimalni dopad paralelizace kodu a problemy s pouzitim interni akcelerace, jsou prezentovany v zaveru prace.

The purpose of this work is to accelerate DFTB method used for biomacromolecule simulation by paralellization. The most time consuming parts of simulation are identified and parallelized using internal options and CPU and GPU acceleration of used linear algebra routines and the code itself. Possibility to use MAGMA - hybrid GPU/CPU linear algebra library and OpenMP parallelization of the code were implemented. All of those methods are described and tested on water systems of different size. Test results of MAGMA acceleration, parallelization and problems with internal optimizations are presented at the end of the work.