Bonn 2010 – wissenschaftliches Programm
HK 36.5: Poster
Mittwoch, 17. März 2010, 14:00–16:00, HG Aula
Parallel Kalman filter track fit based on vector classes — Ivan Kisel1, Matthias Kretz2, and •Igor Kulakov3,4 — 1GSI Helmholtzzentrum für Schwerionenforschung GmbH — 2Kirchhoff-Institut für Physik, Ruprecht-Karls Universität Heidelberg — 3Goethe-Universität Frankfurt am Main — 4National Taras Shevchenko University of Kyiv, Ukraine
Modern high energy physics experiments have to process terabytes of input data produced in particle collisions. The core of the data reconstruction in high energy physics is the Kalman filter. Therefore, developing the fast Kalman filter algorithm, which uses maximum available power of modern processors, is important, in particular for initial selection of events interesting for the new physics.
One of processors features, which can speed up the algorithm, is a SIMD instruction set, which allows to pack several data items in one register and operate on all of them in one go, thus achieving more operations per clock cycle. Therefore a flexible and useful interface, which uses the SIMD instruction set on different CPU and GPU processors architectures, has been realized as a vector classes library.
The Kalman filter based track fitting algorithm has been implemented with use of the vector classes. Fitting quality tests show good results with the residuals equal to 49 µm and 44 µm for x and y track parameters and relative momentum resolution of 0.7%. The fitting time of 0.053 µs per track has been achieved on Intel Xeon X5550 with 8 cores at 2.6 GHz by using in addition Intel Threading Building Blocks.