Efficiently parallelizing a whole set of meshing tools, as required by an automated mesh adaptati... more Efficiently parallelizing a whole set of meshing tools, as required by an automated mesh adaptation loop, relies strongly on data localization to avoid memory access contention. In this regard, renumbering mesh items through a space filling curve (SFC), like Hilbert or Peano, is of great help and proved to be quite versatile. This paper briefly introduces the Hilbert SFC renumbering technique and illustrates its use with two different approaches to parallelization: an out-of-core method and a shared-memory multi-threaded algorithm.
This paper addresses classical issues that arise when applying anisotropic mesh adaptation to rea... more This paper addresses classical issues that arise when applying anisotropic mesh adaptation to real-life 3D problems as the loss of anisotropy or the necessity to truncate the minimal size when discontinuities are present in the solution. These problematics are due to the complex interaction between the components involved in the adaptive loop: the flow solver, the error estimate and the mesh generator. A solution based on a new continuous mesh framework is proposed to overcome these issues. We show that using this strategy allows an optimal level of anisotropy to be reached and thus enjoy the full benefit of unstructured anisotropic mesh adaptation: optimal distribution of the degrees of freedom, improvement of the ratio accuracy with respect to cpu time, ...
Efficiently parallelizing a whole set of meshing tools, as required by an automated mesh adaptati... more Efficiently parallelizing a whole set of meshing tools, as required by an automated mesh adaptation loop, relies strongly on data localization to avoid memory access contention. In this regard, renumbering mesh items through a space filling curve (SFC), like Hilbert or Peano, is of great help and proved to be quite versatile. This paper briefly introduces the Hilbert SFC renumbering technique and illustrates its use with two different approaches to parallelization: an out-of-core method and a shared-memory multi-threaded algorithm.
This paper addresses classical issues that arise when applying anisotropic mesh adaptation to rea... more This paper addresses classical issues that arise when applying anisotropic mesh adaptation to real-life 3D problems as the loss of anisotropy or the necessity to truncate the minimal size when discontinuities are present in the solution. These problematics are due to the complex interaction between the components involved in the adaptive loop: the flow solver, the error estimate and the mesh generator. A solution based on a new continuous mesh framework is proposed to overcome these issues. We show that using this strategy allows an optimal level of anisotropy to be reached and thus enjoy the full benefit of unstructured anisotropic mesh adaptation: optimal distribution of the degrees of freedom, improvement of the ratio accuracy with respect to cpu time, ...
Uploads
Papers by Adrien Loseille