<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>BigDFT | TC</title><link>https://www.cimic.fr/tag/bigdft/</link><atom:link href="https://www.cimic.fr/tag/bigdft/index.xml" rel="self" type="application/rss+xml"/><description>BigDFT</description><generator>Wowchemy (https://wowchemy.com)</generator><language>en-us</language><lastBuildDate>Wed, 21 Oct 2026 16:00:00 +0200</lastBuildDate><image><url>https://www.cimic.fr/media/icon_hud7622ce574d3dba24d69684c2b50c81e_82098_512x512_fill_lanczos_center_3.png</url><title>BigDFT</title><link>https://www.cimic.fr/tag/bigdft/</link></image><item><title>PSolverKokkos</title><link>https://www.cimic.fr/project/psolverkokkos/</link><pubDate>Wed, 21 Oct 2026 16:00:00 +0200</pubDate><guid>https://www.cimic.fr/project/psolverkokkos/</guid><description>&lt;p>&lt;code>psolver&lt;/code> is the library of the &lt;a href="https://www.cimic.fr/project/l_sim-bigdft/">BigDFT&lt;/a> suite that computes the electrostatic potential of a charge density discretized on the wavelet basis. It solves the Poisson equation with a 3D Fast Fourier Transform, and it comes with several backends for different hardware and programming models (pure Fortran/OpenMP, SYCL, CUDA, &amp;hellip;).&lt;/p>
&lt;p>PSolverKokkos is the new Kokkos backend for &lt;code>psolver&lt;/code>. It is packaged as the external library &lt;code>libpsolverkokkos&lt;/code> and offers two execution paths, &lt;code>KOKKOSCPU&lt;/code> and &lt;code>KOKKOSGPU&lt;/code>, that map onto the device backends of Kokkos (Serial, OpenMP and C++ Threads on CPUs, CUDA, HIP and SYCL on GPUs). One implementation covers the whole range of hardware targeted by &lt;code>psolver&lt;/code>. It was developed as a joint project between L_Sim (CEA Grenoble) and the Maison de la Simulation (CEA Saclay) within the CExA initiative.&lt;/p>
&lt;h2 id="kokkos-tea-time">Kokkos Tea Time&lt;/h2>
&lt;p>I presented this work on 21 October 2026 at the &lt;a href="https://cexa-project.org/kokkos-tea-time/2026-10-21-tea-time-kokkos-portability-to-bigdfts/" target="_blank" rel="noopener">Kokkos Tea Time&lt;/a>, a monthly webinar about Kokkos.&lt;/p>
&lt;p>&lt;strong>Bringing Kokkos Performance Portability to BigDFT&amp;rsquo;s Poisson Solver: a Fortran/C++ Interoperability Journey&lt;/strong>&lt;/p>
&lt;p>The talk covers the design of the backend, with a focus on what it takes to embed a modern C++ performance-portability library in a large Fortran code: memory ownership, lifetime of views across the language boundary, runtime selection of the backend and integration in the build system. It also compares the performance with the native backends of &lt;code>psolver&lt;/code>, for example &lt;code>KOKKOSGPU&lt;/code> on CUDA against the hand-written CUDA backend.&lt;/p>
&lt;ul>
&lt;li>&lt;a href="https://www.cimic.fr/uploads/kokkos-tea-time/">Slides&lt;/a> (press &lt;code>p&lt;/code> for the presenter view)&lt;/li>
&lt;li>&lt;a href="https://www.cimic.fr/uploads/kokkos-tea-time/deck.pdf">Slides as a PDF&lt;/a>&lt;/li>
&lt;li>&lt;a href="https://gitlab.com/thibaultcimic/kokkos-tea-time" target="_blank" rel="noopener">Everything needed to rebuild the slides&lt;/a>: sources, figure scripts and the parsed benchmark data&lt;/li>
&lt;/ul></description></item></channel></rss>