From 78a75329ddc0d9ab98ede1c8686a82c71064fe33 Mon Sep 17 00:00:00 2001 From: Wolfgang Bangerth Date: Mon, 14 May 2018 22:21:41 +0800 Subject: [PATCH] Some last edits. --- 9.0/paper.tex | 32 ++++++++++++++++++-------------- 1 file changed, 18 insertions(+), 14 deletions(-) diff --git a/9.0/paper.tex b/9.0/paper.tex index 343843c..556e6c3 100644 --- a/9.0/paper.tex +++ b/9.0/paper.tex @@ -229,16 +229,17 @@ are mainly used in Boundary Element Methods. \texttt{ParameterAcceptor::initialize}, greatly simplifying dealing with parameters in user codes. -\item New caching mechanism for expensive grid computations: we introduced a new +\item A new caching mechanism for expensive grid computations: this version introduces a new class \texttt{GridTools::Cache} that caches computationally intensive information about a \texttt{Triangulation}. This class allows the user to query some of the data structures constructed using functions in the \texttt{GridTools} namespace. This data is then computed only once, and -then cached inside this class for faster access whenever the triangulation has +cached inside this class for faster access whenever the triangulation has not changed. The cache is marked for update by the \texttt{Triangulation} itself -using signals. +using signals so that data is properly invalidated upon mesh +refinement and coarsening. Some of the methods in \texttt{GridTools} already use this cache to speed up repeated calls to the same expensive methods. @@ -348,7 +349,8 @@ for nonlinear viscosity models in fluid flow. demonstrates this technique based on the Trilinos Sacado \cite{Bartlett2006a} package, but the functionality was not available pervasively throughout deal.II. -This has changed with release 9.0 where support is given for differentiation +This has changed with release 9.0 where support for differentiation is +now available using a selection of ``white-listed'' libraries (namely ADOL-C \citep{Griewank1996a} and Sacado) and a subset of their supported number types. Currently, we offer support for the following cases: @@ -361,13 +363,13 @@ we offer support for the following cases: \item Sacado nested reverse and dynamic forward (twice differentiable). \end{itemize} In practice, this support means that these ADOL-C and Sacado data -types can be used in the \texttt{FEValues}, \texttt{FEValuesViews}, +types can be used as the underlying ``scalar'' in the \texttt{FEValues}, \texttt{FEValuesViews}, \texttt{Tensor}, \texttt{SymmetricTensor}, and related classes that are generally used to assemble linear systems and right hand sides. Given the updated capabilities of the library, there is now a dedicated module that presents the AD compatibility and capabilities of the \dealii libraries. -Furthermore, the use of Sacado is now demonstrated in a much more simplified +Furthermore, the use of Sacado is demonstrated in a much more simplified and transparent manner in a modernized version of an existing ``code gallery'' example \cite{Pelteret2016a}. @@ -449,7 +451,7 @@ release, we have written several new interfaces as discussed in the following. \dealii now has interfaces to SUNDIALS's ARKode, IDA, and KINSOL sub-packages. ARKode is a solver library that provides adaptive-step time integration. IDA is a package for the solution of differential-algebraic - equations systems in the form $F(t,y,y')=0$. KINSOL is solver for nonlinear + equations systems in the form $F(t,y,y')=0$. KINSOL is a solver for nonlinear algebraic systems. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% @@ -475,7 +477,7 @@ more reliable. These changes include some minor incompatibilities: all \texttt{Mapping::clone()}) now return \texttt{std::unique\_ptr}s instead of C-style raw pointers. Indeed, nearly all interfaces throughout the library that return a pointer now return either a \texttt{std::shared\_ptr} or a \texttt{std::unique\_ptr}, -thereby clarifying object ownership responsibilities and avoiding memory leaks. +thereby clarifying object ownership and avoiding memory leaks. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% @@ -487,18 +489,20 @@ GPUs has been increasing during the last few years. This release of \dealii adds support for GPUs both for matrix-based and matrix-free applications. For matrix-based applications, we rely on -cuSPARSE\cite{cusparse} and cuSOLVER\cite{cusolver} for operations on sparse +cuSPARSE \cite{cusparse} and cuSOLVER \cite{cusolver} for operations on sparse matrices such as matrix-vector multiplication and for direct solvers. We have introduced a new type of sparse matrix, \texttt{CUDAWrappers::SparseMatrix}, which moves onto the device a \dealii \texttt{SparseMatrix} and changes the format of the underlying data to the appropriate CSR format used by cuSPARSE. We also have added wrappers for Cholesky and LU factorizations provided by -cuSOLVER. In practice, a user would assemble the matrix associated to the system -on the host and then move the matrix to the device. At this point, the system +cuSOLVER. In practice, a user would assemble the system matrix and +right hand side vector +on the host and then move them to the device. At this point, the +linear system would be solved on the device and the solution would be moved back to the host. We also have some support for matrix-free computation on a GPU. For now, the -evaluation of the operator is limited to mesh without hanging-nodes. +evaluation of the operator is limited to meshes without hanging nodes. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% @@ -588,7 +592,7 @@ with 12 cores in about a minute. The tutorial program \texttt{step-60} shows how to perform computations on non-matching grids, and it presents advanced manipulation of -\texttt{ParameterHandler} objects, exploiting the new \texttt{ParameterAccessor} +\texttt{ParameterHandler} objects using the new \texttt{ParameterAccessor} and \texttt{ParameterAccessorProxy} classes. \texttt{step-60} solves a Poisson problem on a domain $\Omega$, subject to equality constraints defined on an embedded domain $\Gamma$. The embedded domain can be of co-dimension one or @@ -596,7 +600,7 @@ co-dimension zero, and its definition is independent with respect to $\Omega$. In order to enforce correctly the constraints, a non-matching coupling matrix needs to be constructed. This is achieved using the new \texttt{NonMatching::create\_coupling\_sparsity\_pattern} and -\texttt{NonMatching::create\_coupling\_mass\_matrix}, that exploit some new +\texttt{NonMatching::create\_coupling\_mass\_matrix} functions that exploit new functionality in the \texttt{GridTools} namespace. \dealii has a separate ``code gallery'' that -- 2.39.5