Computational time improvement for some shallow water finite volume models applying parallelization and optimized small matrix computations

• M. J. Castro ^[1] ; J. A. García ^[1] ; J. M. González ^[1] ; C. Parés ^[1]
  1. [1] Universidad de Málaga

     Universidad de Málaga

     Málaga, España

     
• Localización: Numerical Mathematics and Advanced Applications: ENUMATH 2005 / coord. por Alfredo Bermúdez de Castro y López-Varela , Mª Dolores Gómez Pedreira , Peregrina Quintela Estevez , Pilar Salgado Rodríguez , 2006, ISBN 978-3-540-34287-8, págs. 288-296
• Idioma: inglés
• DOI: 10.1007/978-3-540-34288-5_22
• Texto completo no disponible (Saber más ...)
• Resumen

  • The goal of this paper is to construct efficient finite volume parallel solvers on non-structured grids for 2d hyperbolic systems of conservation laws with source terms and nonconservative products using SIMD registers. Line method is applied: at every intercell a projected Riemann problem along the normal direction is considered (see [2]). The resulting 2d numerical schemes are explicit and first order accurate. The solver is parallelized following a SIMD approach, by means of SSE ("Streaming SIMD Extensions"), which are present in common processors. A generic C++ wrapper to small matrices libraries that make use of SIMD instructions has been implemented in an efficient way and an application to IPP small matrix library is presented.