From 49bec952389fbfbf287b95fdaef5d3e612ed396b Mon Sep 17 00:00:00 2001 From: Martin Kronbichler Date: Wed, 22 May 2019 18:55:09 +0200 Subject: [PATCH] Improve announce text --- 9.1/announce | 62 +++++++++++++++++++++++++++++----------------------- 1 file changed, 35 insertions(+), 27 deletions(-) diff --git a/9.1/announce b/9.1/announce index 6798cf5..9351617 100644 --- a/9.1/announce +++ b/9.1/announce @@ -6,9 +6,16 @@ available for free under an Open Source license from the deal.II homepage at The major changes of this release are: - - Improved support for automatic and symbolic differentiation - - EXPAND + - The support for automatic and symbolic differentiation within deal.II has + been completely overhauled and significantly extended. The automatic + differentiation facilities integrate a broad spectrum of functionality, + including support in the evaluation of finite element shape functions + (FEValues) and associated point and tensor data structures. Functionality + from both the ADOL-C as well as Sacado library in various modes is + included. For symbolic differentiation, new wrappers to the + high-performance SymEngine library allow an alternative approach + addressing the syntax levels of expressions to compute derivatives, rather + than the algorithmic approach underlying automatic differentiations. - Full support for hp adaptivity in parallel computations. deal.II has had support for hp-adaptive methods since around 2005, but not for both @@ -16,9 +23,10 @@ The major changes of this release are: number of algorithmic issues were addressed in this release. Details can be found in the release paper. - - Interfaces to the HDF5 file format and libraries - - EXPAND + - A new interfaces to the HDF5 interface has been added in order to leverage + its capabilities in terms of high-performance I/O for large amount of + data. The HDF5 interfaces in deal.II support read and write operations + both in serial and with MPI. - GPU support was significantly extended for the current release: Added features include preconditioners for CUDAWrappers::SparseMatrix @@ -26,29 +34,27 @@ The major changes of this release are: constraints in the matrix-free framework (and thus also support for adaptively refined meshes). - - Parallel geometric multigrid (GMG) improvements - - EXPAND + - Four new tutorial programs have been added, step-61 demonstrating an + implementation of the so-called weak Galerkin method, step-62 solving the + elastic wave equation with perfectly matched layers to calculate the + resonance frequency and bandgap of photonic crystals, step-63 on block and + point smoothers for geometric multigrid in the context of + convection-diffusion problems, and step-64 implementing a Helmholtz solver + with GPU using matrix-free methods. Furthermore, a new code gallery + program solving a Bayesian inverse problem has been added. - - Four new tutorial programs (step-61, step-62, step-63, step-64) - and a new code gallery program - - EXPAND - - - The release contains bug fixes and performance improvements of the + - The release contains performance improvements and bug fixes of the matrix-free framework and related geometric multigrid solvers. In - particular, the implementation of the Chebyshev iteration, an often - used smoother in the matrix-free context, has been revised to reduce - vector accesses. This speeds up matrix-free multigrid solvers by up to - 10-15% on geometries with affine (parallelogram and parallelpiped) - cells, and up to 5% on geometries with cells bounded by curved edges - and faces. + particular, the implementation of the Chebyshev iteration, an often used + smoother in the matrix-free context, has been revised to reduce vector + accesses. Altogether, matrix-free multigrid solvers run up to 15% faster + than in the previous version. - Various variants of geometric multigrid solvers and matrix-free implementations were run on up to 304,128 MPI ranks during the acceptance - phase of the SuperMUC-NG supercomputer, verifying the scalability of - our implementations. Some performance bottlenecks on distributed - triangulation with more than 100k ranks have been fixed. + phase of the SuperMUC-NG supercomputer, verifying the scalability of our + implementations to this scale. Some geometric multigrid data structures + were revised to avoid bottlenecks showing up with more than 100k ranks. - The FE_BernardiRaugel class implements the non-standard Bernardi-Raugel element that can be used to construct a stable @@ -78,9 +84,11 @@ The major changes of this release are: solutions, or any other custom computation of the error, given via std::function objects. - - A new interface to \texttt{boost::geometry::index::rtree} has been added. - - EXPAND + - A new interface to \texttt{boost::geometry::index::rtree} has been + added. The rtree provides a data structure for handling bounding boxes + that cover the whole triangulation, and allow to identify e.g. the cell a + certain point is located in logarithmic time, or queries regarding the + cell owners of far-field cells in a distributed setup. - More than 200 other features and bugfixes. -- 2.39.5