From: Denis Davydov Date: Tue, 10 Apr 2018 06:50:30 +0000 (+0200) Subject: fill-in Lapack/ScaLAPACK X-Git-Url: https://gitweb.dealii.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=3492388ca0a03b322d7721466c79e223a963ee98;p=release-papers.git fill-in Lapack/ScaLAPACK --- diff --git a/.gitignore b/.gitignore index a2c275c..aaaaf59 100644 --- a/.gitignore +++ b/.gitignore @@ -7,3 +7,6 @@ *.fdb_latexmk *.fls *~ +*.dvi +*.ps +*.synctex.gz diff --git a/9.0/paper.tex b/9.0/paper.tex index 9c954fa..ac87a2e 100644 --- a/9.0/paper.tex +++ b/9.0/paper.tex @@ -185,14 +185,30 @@ The major changes of this release are: \texttt{IterativeInverse}, and \texttt{PointerMatrix}. \item - Significant extension of matrix-free capabilities, including support for face integrals in discontinuous Galerkin schemes, a new Hermite-like polynomial basis adapted to face integrals that involve derivatives of shape functions, and several performance enhancements + Significant extension of matrix-free capabilities, including support for face + integrals in discontinuous Galerkin schemes, a new Hermite-like polynomial + basis adapted to face integrals that involve derivatives of shape functions, + and several performance enhancements. \item Support for GPU \item - Increased support for dense solvers (LAPACK changes and new ScaLAPACK - interface). + In this version we add template wrappers for ScaLAPACK -- parallel dense linear algebra + library with block-cyclic matrix distribution over 2D process grid. + The functionality and interface is similar to the LAPACK wrappers, + matrix-matrix multiplication, Cholesky and LU factorizations, + eigensolvers, SVD, pseudoinverse, save/load in HDF5 format and + other functionality is implemented. + A user can combine the new linear algebra with other MPI-parallel + algorithms in \dealii{} without the need to worry about MPI communicator + being divisible exactly into 2D process grid. The processes which are not part of + the grid are labeled inactive and ScaLAPACK routines will not be called on them. + % + Additionally, we improved support of LAPACK. We added methods to perform + rank-1 update/downdate, calculate Cholesky factorization, trace and determinant, + as well as estimate the reciprocal condition number. + We also now support configuration with 64-bit BLAS. \item A dedicated sundials module has been created to provide support for the @@ -203,7 +219,7 @@ The major changes of this release are: equations systems in the form $F(t,y,y')=0$. KINSOL is solver for nonlinear algebraic systems. -\item +\item A dedicated optimization/rol module has been created to provide support for the Rapid Optimization Library (ROL). ROL is a package for large-scale optimization.