CEED Annual Meeting
August 9-11, 2022
University of Illinois + Virtual
Overview
CEED will hold its sixth annual meeting August 9-11, 2022 in a hybrid format: in-person at the Siebel Center for Computer Science on the UIUC campus in Urbana and virtually using ECP Zoom for videoconferencing and Slack for side discussions. The goal of the meeting is to report on the progress in the center, deepen existing and establish new connections with ECP hardware vendors, ECP software technologies projects and other collaborators, plan project activities and brainstorm/work as a group to make technical progress. In addition to gathering together many of the CEED researchers, the meeting will include representatives of the ECP management, hardware vendors, software technology and other interested projects.
Meeting format
The meeting will include the following elements:
- Project review and updates from the CEED team
- Contributed talks from AD, ST, vendors and external partners
- Technical discussions in small breakout groups
In addition to the in-person meeting, we will provide the following options for remote participants:
- Live presentations and wrap-up will be on ECP Zoom (links to be posted the week of the meeting).
- Side discussions and breakout sessions will be on the meeting Slack space -- Please join in advance.
Meeting Agenda
The meeting activities will take place 9am-4:30pm central time.
Tuesday, August 9
| Time (CDT) | Activity |
|---|---|
| 9:00-9:20 | Welcome & CEED Overview Tzanio Kolev (LLNL) |
| 9:20-9:40 | Finite Element Thrust & MFEM Update Veselin Dobrev (LLNL) This talk will provide a summary of the Finite Element thrust activities during the last year of the CEED project with a main focus on new developments in the MFEM software library. |
| 9:40-10:00 | nekRS Update Paul Fischer (UIUC) We give an overview of some of the developments in Nek5000/RS for CEED/ECP. |
| 10:00-10:20 | libParanumal: Matrix core acceleration Tim Warburton (VTech) We revisit the CEED bake-off problems on the latest GPU from major manufacturers. We will illustrate the importance of exploiting intrinsic matrix instructions to reach the device memory bandwidth roofline. |
| 10:20-10:40 | Coffee Break • side discussions on Slack • propose breakout topics |
| 10:40-11:00 | Software Thrust & libCEED Update Jed Brown (Boulder) |
| 11:00-11:20 | PETSc GPU Support Update Junchao Zhang (ANL) + Jacob Faibussowitsch (UIUC) This talk includes two parts. In the first part, we give an update of the current PETSc GPU support status, and some new features we recently added, including matrix coordinate (COO) assembly on device, and PETSc/Kokkos interface helping users to write single-sourced code. In the second part, we describe recent updates to the library backend regarding GPUs where we introduce new library objects PetscDevice and PetscDeviceContext. We end by showcasing an experimental new Vec API to allow for safe, transparent asynchronous compute. |
| 11:20-11:40 | Multigrid smoothers: Chebyshev acceleration and symmetric Schwarz smoothers for SEM James Lottes (Google) |
| 11:40-12:00 | Solver developments in nekRS Malachi Philips (UIUC) Discuss new developments in nekRS |
| 12:00-1:30 | Lunch Break + Group Photo |
| 1:30-2:40 | Technical Discussions Group discussion on topics such as: • CPUs vs GPUs, AMD and Intel GPUs, ARM processors • preconditioning, strong scaling, meshing, visualization, etc. |
| 2:40-3:00 | Coffee Break • side discussions on Slack • propose breakout topics |
| 3:00-3:20 | Applications and ModSim: How the two are intertwined Ian Karlin (NVIDIA) Applications and simplified versions of them are crucial to influencing future hardware. However, the challenges of simulation tools mean that often key performance features are missed. These gaps and how they are used in benchmarking of future systems can mean that what users want from future machines and what is procured can be significantly different. |
| 3:20-4:30 | Technical Discussions Group discussion on topics such as: • CPUs vs GPUs, AMD and Intel GPUs, ARM processors • preconditioning, strong scaling, meshing, visualization, etc. |
| 4:30 | Day 1 Wrap-up |
| 6:30-8:30 | Meeting Dinner Silver Creek, Urbana |
Wednesday, August 10
| Time (CDT) | Activity |
|---|---|
| 9:00-9:20 | Applications Thrust & Nek Update Misun Min (ANL) |
| 9:20-9:40 | ExaSMR updates: Toward full system simulations Elia Merzari (PSU) |
| 9:40-10:00 | Post-Modern GMRES+AMG with ILU Smoothers on GPU Stephen Thomas (AMD) We will present a new post-modern low-synch GMRES formulation (from last CEED meeting) with a formal proof of backward stability and O(eps) orthogonality for the Krylov vectors. The basic idea is to project the properly normalized Krylov vectors onto their orthogonal complement with a multiplicative Gauss-Seidel projection step. PM-GMRES is then preconditioned with Hypre's C-AMG using ILU(0) and ILUT Schur complement smoothers. The ROCm sparse ILU(0) set-up time on MI250X GPU is 10x faster than cuSparse on the NVIDIA V100. We will present the implementation of Chow and Patel's ILU(0) with fixed-point iteration leading to a further reduction of 2x to 5x in set-up times. These solvers combine to accelerate the Combustion-PeleLM and MFIX-Exa models by 5x on Crusher with iterative (Jacobi-Neuman) triangular solvers replacing the direct csrTRSV in ROCsparse. These algorithms lend themselves naturally to GPU acceleration as they rely on matrix polynomials. Further acceleration with RAS and ORAS is anticipated for such cut-cell and imbedded boundary methods, work in collaboration with Professor Luke Olson. |
| 10:00-10:20 | Mimetic Galerkin Differences Christopher Eldred (SNL) Mimetic Galerkin Differences are a type of compatible Galerkin discretization method, with some interesting properties compared to the more mainstream compatible finite element methods (FEEC). This talk will discuss these methods, their properties and compare them with other compatible Galerkin methods, including FEEC and compatible isogeometric methods. |
| 10:20-10:40 | Coffee Break • side discussions on Slack |
| 10:40-11:00 | Hardware Thrust & MAGMA Update Stanimire Tomov + Natalie Beams (UTK) |
| 11:00-11:20 | Exploiting Task Graphs for Fun and Strong Scaling Freddie Witherden (Texas A&M) In this talk we will describe how task graphs can be used to improve strong scaling of high-order FEM codes. Specifically, our focus will be on the recently-introduced graph APIs in CUDA and HIP and the steps required to port over an existing code to them. Interactions between task graphs and MPI will also be discussed, along with how such APIs can be emulated with OpenCL. |
| 11:20-11:40 | Efficient high-order finite element computations with a focus on CPUs via the deal.II library Martin Kronbichler (Augsburg) |
| 11:40-12:00 | Conforming Mesh Adaptation Cameron Smith + Morteza Hakimi (RPI) |
| 12:00-1:30 | Lunch Break |
| 1:30-1:50 | El Capitan Update David Richards (LLNL) |
| 1:50-3:00 | Technical Discussions Group discussion on topics such as: • CPUs vs GPUs, AMD and Intel GPUs, ARM processors • preconditioning, strong scaling, meshing, visualization, etc. |
| 3:00-3:20 | Coffee Break • side discussions on Slack |
| 3:20-4:30 | Technical Discussions Group discussion on topics such as: • CPUs vs GPUs, AMD and Intel GPUs, ARM processors • preconditioning, strong scaling, meshing, visualization, etc. |
| 4:30 | Day 2 Wrap-up |
Thursday, August 11
| Time (CDT) | Activity |
|---|---|
| 9:00-10:20 | Internal planning • Future milestones • Future papers • post-ECP |
| 10:20-10:40 | Coffee Break |
| 10:40-12:00 | Meetings and discussions • Collaborations • Hackathon |
Participants
- See the meeting Slack space
Lodging Options
The meeting will be held at the Siebel Center for Computer Science on the UIUC campus in Urbana. There are a number of hotels in a walking distance from the building, for example Hampton Inn, 1200 W. University Ave.
See the map below for more details.
Questions?
For questions, please contact the meeting organizers at ceed-meeting@llnl.gov.