From: Daniel Arndt Date: Sat, 2 Mar 2019 16:16:12 +0000 (+0100) Subject: Document the execution space some more X-Git-Tag: v9.1.0-rc1~309^2 X-Git-Url: https://gitweb.dealii.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=92951fa7959b8a6b53d40b61516b304068023bcc;p=dealii.git Document the execution space some more --- diff --git a/include/deal.II/lac/trilinos_tpetra_vector.h b/include/deal.II/lac/trilinos_tpetra_vector.h index 35e31cc92e..69303e2284 100644 --- a/include/deal.II/lac/trilinos_tpetra_vector.h +++ b/include/deal.II/lac/trilinos_tpetra_vector.h @@ -54,11 +54,18 @@ namespace LinearAlgebra * LinearAlgebra::VectorSpaceVector class and requires Trilinos to be * compiled with MPI support. * - * Tpetra uses Kokkos for thread-parallelism. In case Trilinos was - * configured with CUDA support, this class stores the values in unified - * virtual memory space and performs its action on the GPU. In particular, - * there is no need for manually synchronizing memory between host and - * device. + * Tpetra uses Kokkos for thread-parallelism and chooses the execution and + * memory space automatically depending on Kokkos configuration. The + * priority is ranked from highest to lowest: + * - Kokkos::Cuda + * - Kokkos::OpenMP + * - Kokkos::Threads + * - Kokkos::Serial + * + * In case Kokkos was configured with CUDA support, this class stores the + * values in unified virtual memory space and performs its action on the + * GPU. In particular, there is no need for manually synchronizing memory + * between host and device. * * @ingroup TrilinosWrappers * @ingroup Vectors