Beta 1


Title Implementation and evaluation of fast computational methods for high-resolution ODF problems on multi-core and many-core systems
Author Høstergaard, Martin
Gade-Nielsen, Nicolai Fog (Scientific Computing, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Supervisor Dammann, Bernd (Scientific Computing, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Schmidt, Søren (Materials Research Division, Risø National Laboratory for Sustainable Energy, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Institution Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
Thesis level Master's thesis
Year 2010
Abstract In this master thesis a toolbox in C for high resolution ODF reconstruction is developed, primarily for use on multi- / many-core architectures. The thesis consists of two major parts; implementing fast and precise ray tracing algorithms and fast and highly parallel iterative linear solvers for sparse systems. The ray tracing is used for generating the coefficient matrix in a linear system and it is therefore crucial that it is fast and precise. Several CPU ray tracing algorithms have been implemented and compared. This has resulted in a customized ray tracing algorithm best suited for ODF reconstructions and enables us to use GPUs for ray tracing. The iterative solvers are used for solving the linear system arising from the inverse ODF problem. We have implemented four different iterative solvers, namely Component Averaging (CAV), Conjugate Gradient Least Squares (CGLS), LSQR and Kaczmarz (ART). The performance of the solvers is studied and compared using phantom models. All solvers are also implemented as parallel versions on CPUs using OpenMP, with satisfying speedup rates. For the CGLS solver, the main parts are offloaded to GPUs and outperforms the regular CPU versions. Furthermore, we have implemented other techniques including two types of hierarchical solving, where one of the techniques has shown to be very effective. Support for reading data measurements is also implemented, which enables the application to be used for real ODF reconstructions. This functionality has been successfully tested by performing reconstructions with simulated data in the real format.
Imprint Technical University of Denmark (DTU) : Kgs. Lyngby, Denmark
Series IMM-M.Sc.-2010-85
Keywords Materialer og energilagring; Materials and energy storage; uden tema
Fulltext
Original PDF ep10_85_net.pdf (4.29 MB)
Admin Creation date: 2010-10-12    Update date: 2010-11-15    Source: dtu    ID: 267754    Original MXD