A5077929064- Parallel Computing and Optimization Techniques
- Distributed and Parallel Computing Systems
- Advanced Numerical Methods in Computational Mathematics
- Electromagnetic Simulation and Numerical Methods
- Matrix Theory and Algorithms
- Computational Fluid Dynamics and Aerodynamics
- Electromagnetic Scattering and Analysis
- Neural Networks and Applications
- Embedded Systems Design Techniques
- Interconnection Networks and Systems
- Numerical methods in engineering
- Advanced Data Storage Technologies
- Numerical methods for differential equations
- Model Reduction and Neural Networks
- Computer Graphics and Visualization Techniques
- Advanced Numerical Analysis Techniques
- Electric Motor Design and Analysis
- Advanced Mathematical Modeling in Engineering
- Induction Heating and Inverter Technology
- Gas Dynamics and Kinetic Theory
- Industrial Vision Systems and Defect Detection
- Cloud Computing and Resource Management
- Scientific Computing and Data Management
- Survey Sampling and Estimation Techniques
- Magnetic Properties and Applications
Sandia National Laboratories California
2014-2025
Sandia National Laboratories
2009-2024
United States Department of Energy
2014
William & Mary
2000
A template-based generic programming approach was presented in Part I of this series papers [Sci. Program. 20 (2012), 197–219] that separates the development effort a physical model from computing additional quantities, such as derivatives, needed for embedded analysis algorithms. In paper, we describe implementation details using simulation and partial differential equations (PDEs). We detail several hurdles have encountered, some software infrastructure developed to overcome them. end with...
Trilinos is an object-oriented software framework for the solution of large-scale, complex multi-physics engineering and scientific problems. While was originally designed scalable solutions large problems, fidelity needed by many simulations significantly greater than what one could have envisioned two decades ago. When problem sizes exceed a billion elements even applications solver stacks require complete revision. The second-generation employs C++ templates in order to solve arbitrarily...
We consider the sequence of sparse matrix-matrix multiplications performed during setup phase algebraic multigrid. In particular, we show that most commonly used parallel algorithm is often not communication-efficient one for all involved. By using an alternative algorithm, communication costs are reduced (in theory and practice), demonstrate performance benefit both model (structured) more realistic unstructured problems on large-scale distributed-memory systems. Our theoretical analysis...
With the rise in popularity of compatible finite element, difference, and volume discretizations for time domain eddy current equations, there has been a corresponding need fast solvers resulting linear algebraic systems. However, traits that make preferred choice Maxwell's equations also render these systems essentially intractable by truly black-box techniques. We propose an reformulation discrete along with new multigrid (AMG) technique this reformulated problem. The process takes...
We develop and analyze least-squares finite element methods for two complementary div-curl elliptic boundary value problems. The first one prescribes the tangential component of vector field on is solved using curl-conforming elements. second problem specifies normal handled by div-conforming prove that both formulations are norm-equivalent with respect to suitable discrete norms, yield optimal asymptotic error estimates, give rise algebraic systems can be efficient multigrid methods....
Trilinos is an object-oriented software framework for the solution of large-scale, complex multi-physics engineering and scientific problems. While original version was designed highly scalable solutions large problems, need increasingly higher fidelity simulations has pushed problem sizes beyond what could have been envisioned two decades ago. When exceed a billion elements even applications solver stacks require complete revision. The next-generation employs C++ templates in order to solve...
Sparse matrix-matrix multiplication is a critical kernel for several scientific computing applications, especially the setup phase of algebraic multigrid. The MPI+X programming model, which growing in popularity, requires that such kernels be implemented way exploits on-node parallelism. We present single-pass OpenMP variant Gustavson's sparse matrix algorithm designed architectures (e.g. CPU or Intel Xeon Phi) with reasonably large memory and modest thread counts (tens threads, not...
Abstract Four adaptations of the smoothed aggregation algebraic multigrid (SA‐AMG) method are proposed with an eye toward improving convergence and robustness solver in situations when discretization matrix contains many weak connections. These connections can cause higher than expected levels fill‐in within coarse matrices also give rise to suboptimal smoothing prolongator phase. drawbacks due relatively small size some diagonal entries filtered that one obtains after dropping The new...
Centered on modern C++ and the SYCL standard for heterogeneous programming, Data Parallel (dpc++) Intel's oneAPI software ecosystem aim to lower barrier entry use of accelerators like FPGAs in diverse applications. In this work, we consider usage scientific computing, particular with a multigrid solver, MueLu. We report early experiences implementing kernels solver DPC++ execution Stratix 10 FPGAs, evaluate several algorithmic design implementation choices. These choices not only impact...
As a rule, Top 500 class supercomputers are extensively benchmarked as part of their acceptance testing process. However, barring publicly posted LINPACK / HPCG results, most benchmark results often inaccessible outside the hosting institution. Moreover, these higher level benchmarks do not provide easy answers to common questions such "What is realizable memory bandwidth?" or launch latency on accelerator?" To partially address issues, we executed selected single-node micro-benchmarks —...