Computational cost estimates for parallel shared memory isogeometric multi-frontal solvers

In this paper we present computational cost estimates for parallel shared memory isogeometric multi-frontal solvers. The estimates show that the ideal isogeometric shared memory parallel direct solver scales as $\mathcal{O}( p^2log(N/p))$ for one dimensional problems, $\mathcal{O}(Np^2)$ for two dim...

Descripción completa

Detalles Bibliográficos
Autores: Wozniak, M., Kuznik, K., Paszynski, M., Calo, V.M., Pardo, D.
Tipo de recurso: artículo
Estado:Versión aceptada para publicación
Fecha de publicación:2014
País:España
Institución:Basque Center for Applied Mathematics (BCAM)
Repositorio:BIRD. BCAM's Institutional Repository Data
OAI Identifier:oai:bird.bcamath.org:20.500.11824/82
Acceso en línea:http://hdl.handle.net/20.500.11824/82
Access Level:acceso abierto
Palabra clave:Cost estimating
Polynomial approximation
Two dimensional
Computational costs
Direct solvers
Number of degrees of freedom
Nvidia CUDA
One dimensional problems
Shared-memory parallels
Three-dimensional problems
Two-dimensional problem
One dimensional
Descripción
Sumario:In this paper we present computational cost estimates for parallel shared memory isogeometric multi-frontal solvers. The estimates show that the ideal isogeometric shared memory parallel direct solver scales as $\mathcal{O}( p^2log(N/p))$ for one dimensional problems, $\mathcal{O}(Np^2)$ for two dimensional problems, and $\mathcal{O}(N^{4/3}p^2)$ for three dimensional problems, where $N$ is the number of degrees of freedom, and p is the polynomial order of approximation. The computational costs of the shared memory parallel isogeometric direct solver are compared with those corresponding to the sequential isogeometric direct solver, being the latest equal to $\mathcal{O}(N p^2)$ for the one dimensional case, $\mathcal{O}(N^{1.5}p^3)$ for the two dimensional case, and $\mathcal{O}(N^2p^3)$ for the three dimensional case. The shared memory version significantly reduces both the scalability in terms of $N$ and $p$. Theoretical estimates are compared with numerical experiments performed with linear, quadratic, cubic, quartic, and quintic B-splines, in one and two spatial dimensions.