The Moving Least Squares Method in Mesh Deformation - Implementation in CUDA/C++ & Performance analysis

Κατσαβριάς, Ευάγγελος; Katsavrias, Efangelos

dc.contributor.author	Κατσαβριάς, Ευάγγελος	el
dc.contributor.author	Katsavrias, Efangelos	en
dc.date.accessioned	2019-05-20T08:54:26Z
dc.date.available	2019-05-20T08:54:26Z
dc.date.issued	2019-05-20
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/48784
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.9207
dc.description	Εθνικό Μετσόβιο Πολυτεχνείο--Μεταπτυχιακή Εργασία. Διεπιστημονικό-Διατμηματικό Πρόγραμμα Μεταπτυχιακών Σπουδών (Δ.Π.Μ.Σ.) “Υπολογιστική Μηχανική”	el
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc-nd/3.0/gr/	*
dc.subject	Παραμόρφωση πλεγμάτων	el
dc.subject	Κινούμενων ελαχίστων τετραγώνων	el
dc.subject	Παράλληλος προγραμματισμός	el
dc.subject	Κάρτες γραφικών	el
dc.subject	Αλγόριθμοι	el
dc.subject	Mesh deformation	en
dc.subject	Moving Least Squares	en
dc.subject	Parallel programming	el
dc.subject	GPUs	el
dc.subject	Algorithms	el
dc.title	The Moving Least Squares Method in Mesh Deformation - Implementation in CUDA/C++ & Performance analysis	en
heal.type	masterThesis
heal.classification	Υπολογιστική μηχανική	el
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2019-02-28
heal.abstract	There are plenty applications in computational mechanics, e.g. in fluid flows around bodies, contact solid mechanics, fluid-structure interaction analysis, etc., where the mesh deformation is part of the process. These problems are quite computationally intensive and very often solved on GPGPUs. Therefore, a mesh deformation method that would perform well on the GPGPU parallel environment, is something desirable to reduce the overall solution time of the problem. Such a mesh deformation method seems to be the Moving Least Squares (MLS) method, which exhibits full data independence in its coarse granularity level of computations, and a moderate parallelization potential in its finer granularity of computations. Prior work on the MLS method for mesh deformation, see Τουρής[2016], had exposed the effectiveness and the good performance of the method in mesh deformation. In this work, a parallel execution of the Moving Least Squares (MLS) method is developed, on General Purpose GPUs (GPGPUs). Initially, the MLS method’s computations are analyzed to expose the potential parallelism with the CUDA compute execution model. Further, CUDA algorithms are proposed to execute the computations efficiently and gain an optimal speedup. Additionally, the MLS method is enriched with the more general rational weighting functions (i.e. inverse distance weighting functions), which allow the utilization of low polynomial degree MLS interpolation, preserving the good quality results for low-medium degree of mesh deformations, see Witteveen and Bijl[2009]. The low polynomial MLS interpolation is an interesting case, because the computations are substantially reduced and the parallelization potential is increased. The results of the developed algorithms expose good compute performance with very high speedups for low polynomial degree MLS interpolations, e.g. up to x20 in the case of the zero degree MLS method (known also as Inverse distance weighting method). The speedups for higher polynomial degrees are moderate, i.e. x5, but the utilized hardware comprised of a very strong CPU and a GPGPU with poor compute resources.	en
heal.advisorName	Γιαννάκογλου, Κυριάκος	el
heal.committeeMemberName	Μπουντουβής, Ανδρέας	el
heal.committeeMemberName	Γιαννάκογλου, Κυριάκος	el
heal.committeeMemberName	Κυρανούδης, Χρήστος	el
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Χημικών Μηχανικών	el
heal.academicPublisherID	ntua
heal.numberOfPages	142 σ.	el
heal.fullTextAvailability	true