Publications

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Sierra/SD is an engineering structural dynamics code that provides Sandia and other customers a tool to model structural and acoustic physics on large complex physical systems using massively parallel processing. This report provides a detailed overview on Sierra/SD’s most recent physics package: coupled electro-mechanical physics. This capability uses the finite element method to model coupled electro-mechanical physics exhibited by piezoelectric materials. This report provides an applications overview, theory overview, and verification examples demonstrating the electro-mechanical physics modeling capabilities of Sierra/SD.

Abstract not provided.

Abstract not provided.

Abstract not provided.

The “how to” document guides the user through complicated aspects of software usage. It should supplement both the User’s manual and the Theory document, by providing examples and detailed discussion that reduce learning time for complex set ups. These documents are intended to be used together. We will not formally list all parameters for an input here – see the User’s manual for this. All the examples in the “How To” document are part of the Sierra/SD test suite, and each will run with no modification. The nature of this document casts together a number of rather unrelated procedures. Grouping them is difficult. Please try to use the table of contents and the index as a guide in finding the analyses of interest.

The explosion of both sensors and GPS-enabled devices has resulted in position/time data being the next big frontier for data analytics. However, many of the problems associated with large numbers of trajectories do not necessarily have an analog with many of the historic big-data applications such as text and image analysis. Modern trajectory analytics exploits much of the cutting-edge research in machine-learning, statistics, computational geometry and other disciplines. We will show that for doing trajectory analytics at scale, it is necessary to fundamentally change the way the information is represented through a feature-vector approach. We then demonstrate the ability to solve large trajectory analytics problems using this representation.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

This report focuses on the two primary goals set forth in Sandia’s TAFI effort, referred to here under the name Kebab. The first goal is to overlay a trajectory onto a large database of historical trajectories, all with very different sampling rates than the original track. We demonstrate a fast method to accomplish this, even for databases that hold over a million tracks. The second goal is to then demonstrate that these matched historical trajectories can be used to make predictions about unknown qualities associated with the original trajectory. As part of this work, we also examine the problem of defining the qualities of a trajectory in a reproducible way.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Recently, Graph Neural Networks (GNNs) have received a lot of interest because of their success in learning representations from graph structured data. However, GNNs exhibit different compute and memory characteristics compared to traditional Deep Neural Networks (DNNs). Graph convolutions require feature aggregations from neighboring nodes (known as the aggregation phase), which leads to highly irregular data accesses. GNNs also have a very regular compute phase that can be broken down to matrix multiplications (known as the combination phase). All recently proposed GNN accelerators utilize different dataflows and microarchitecture optimizations for these two phases. Different communication strategies between the two phases have been also used. However, as more custom GNN accelerators are proposed, the harder it is to qualitatively classify them and quantitatively contrast them. In this work, we present a taxonomy to describe several diverse dataflows for running GNN inference on accelerators. This provides a structured way to describe and compare the design-space of GNN accelerators.

Abstract not provided.

In this work, we present modal-based methods for model calibration in structural dynamics, and address several key challenges in the solution of gradient-based optimization problems with eigenvalues and eigenvectors, including the solution of singular Helmholtz problems encountered in sensitivity calculations, non-differentiable objective functions caused by mode swapping during optimization, and cases with repeated eigenvalues. Unlike previous literature that relied on direct solution of the eigenvector adjoint equations, we present a parallel iterative domain decomposition strategy (Adjoint Computation via Modal Superposition with Truncation Augmentation) for the solution of the singular Helmholtz problems. For problems with repeated eigenvalues we present a novel Mode Separation via Projection algorithm, and in order to address mode swapping between inverse iterations we present a novel Injective mode ordering metric. We present the implementation of these methods in a massively parallel finite element framework with the ability to use measured modal data to extract unknown structural model parameters from large complex problems. A series of increasingly complex numerical examples are presented that demonstrate the implementation and performance of the methods in a massively parallel finite element framework [7,5], using gradient-based optimization techniques in the Rapid Optimization Library (ROL) [21].

MLIR (Multi-Level Intermediate Representation), is an extensible compiler framework that supports high-level data structures and operation constructs. These higher-level code representations are particularly applicable to the artificial intelligence and machine learning (AI/ML) domain, allowing developers to more easily support upcoming heterogeneous AI/ML accelerators and develop flexible domain specific compilers/frameworks with higher-level intermediate representations (IRs) and advanced compiler optimizations. The result of using MLIR within the LLVM compiler framework is expected to yield significant improvement in the quality of generated machine code, which in turn will result in improved performance and hardware efficiency

Abstract not provided.

Over the past decade as Moore's Law has slowed, the need for new forms of computation that can provide sustainable performance improvements has risen. A new method, called in situ computing, has shown great potential to accelerate matrix vector multiplication (MVM), an important kernel for a diverse range of applications from neural networks to scientific computing. Existing in situ accelerators for scientific computing, however, have a significant limitation: These accelerators provide no acceleration for preconditioning-A key bottleneck in linear solvers and in scientific computing workflows. This paper enables in situ acceleration for state-of-The-Art linear solvers by demonstrating how to use a new in situ matrix inversion accelerator for analog preconditioning. As existing techniques that enable high precision and scalability for in situ MVM are inapplicable to in situ matrix inversion, new techniques to compensate for circuit non-idealities are proposed. Additionally, a new approach to bit slicing that enables splitting operands across multiple devices without external digital logic is proposed. For scalability, this paper demonstrates how in situ matrix inversion kernels can work in tandem with existing domain decomposition techniques to accelerate the solutions of arbitrarily large linear systems. The analog kernel can be directly integrated into existing preconditioning workflows, leveraging several well-optimized numerical linear algebra tools to improve the behavior of the circuit. The result is an analog preconditioner that is more effective (up to 50% fewer iterations) than the widely used incomplete LU factorization preconditioner, ILU(0), while also reducing the energy and execution time of each approximate solve operation by 1025x and 105x respectively.

Over the past decade as Moore's Law has slowed, the need for new forms of computation that can provide sustainable performance improvements has risen. A new method, called in situ computing, has shown great potential to accelerate matrix vector multiplication (MVM), an important kernel for a diverse range of applications from neural networks to scientific computing. Existing in situ accelerators for scientific computing, however, have a significant limitation: These accelerators provide no acceleration for preconditioning-A key bottleneck in linear solvers and in scientific computing workflows. This paper enables in situ acceleration for state-of-The-Art linear solvers by demonstrating how to use a new in situ matrix inversion accelerator for analog preconditioning. As existing techniques that enable high precision and scalability for in situ MVM are inapplicable to in situ matrix inversion, new techniques to compensate for circuit non-idealities are proposed. Additionally, a new approach to bit slicing that enables splitting operands across multiple devices without external digital logic is proposed. For scalability, this paper demonstrates how in situ matrix inversion kernels can work in tandem with existing domain decomposition techniques to accelerate the solutions of arbitrarily large linear systems. The analog kernel can be directly integrated into existing preconditioning workflows, leveraging several well-optimized numerical linear algebra tools to improve the behavior of the circuit. The result is an analog preconditioner that is more effective (up to 50% fewer iterations) than the widely used incomplete LU factorization preconditioner, ILU(0), while also reducing the energy and execution time of each approximate solve operation by 1025x and 105x respectively.

A key challenge to nonlocal models is the analytical complexity of deriving them from first principles, and frequently their use is justified a posteriori. In this work we extract nonlocal models from data, circumventing these challenges and providing data-driven justification for the resulting model form. Extracting data-driven surrogates is a major challenge for machine learning (ML) approaches, due to nonlinearities and lack of convexity — it is particularly challenging to extract surrogates which are provably well-posed and numerically stable. Our scheme not only yields a convex optimization problem, but also allows extraction of nonlocal models whose kernels may be partially negative while maintaining well-posedness even in small-data regimes. To achieve this, based on established nonlocal theory, we embed in our algorithm sufficient conditions on the non-positive part of the kernel that guarantee well-posedness of the learnt operator. These conditions are imposed as inequality constraints to meet the requisite conditions of the nonlocal theory. We demonstrate this workflow for a range of applications, including reproduction of manufactured nonlocal kernels; numerical homogenization of Darcy flow associated with a heterogeneous periodic microstructure; nonlocal approximation to high-order local transport phenomena; and approximation of globally supported fractional diffusion operators by truncated kernels.

1 The exponential growth of data has driven technology providers to develop new protocols, such as cache coherent interconnects and memory semantic fabrics, to help users and facilities leverage advances in memory technologies to satisfy these growing memory and storage demands. Using these new protocols, fabric-Attached memories (FAM) can be directly attached to a system interconnect and be easily integrated with a variety of processing elements (PEs). Moreover, systems that support FAM can be smoothly upgraded and allow multiple PEs to share the FAM memory pools using well-defined protocols. The sharing of FAM between PEs allows efficient data sharing, improves memory utilization, reduces cost by allowing flexible integration of different PEs and memory modules from several vendors, and makes it easier to upgrade the system. One promising use-case for FAMs is in High-Performance Compute (HPC) systems, where the underutilization of memory is a major challenge. However, adopting FAMs in HPC systems brings new challenges. In addition to cost, flexibility, and efficiency, one particular problem that requires rethinking is virtual memory support for security and performance. To address these challenges, this paper presents decoupled access control and address translation (DeACT), a novel virtual memory implementation that supports HPC systems equipped with FAM. Compared to the state-of-The-Art two-level translation approach, DeACT achieves speedup of up to 4.59x (1.8x on average) without compromising security.1Part of this work was done when Vamsee was working under the supervision of Amro Awad at UCF. Amro Awad is now with the ECE Department at NC State.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Digital engineering strategies typically assume that digital engineering models interoperate seamlessly across the multiple different engineering modeling software applications involved, such as model- based systems engineering (MBSE), mechanical computer-aided design (MCAD), electrical computer-aided design (ECAD), and other engineering modeling applications. The presumption is that the data schema in these modeling software applications are structured in the familiar flat- tabular schema like any other software application. Engineering domain-specific applications (e.g., systems, mechanical, electrical, simulation) are typically designed to solve domain-specific problems, necessarily excluding explicit representations of non-domain information to help the engineer focus on the domain problems (system definition, design, simulation). Such exclusions become problematic in inter-domain information exchange. The obvious assumptions of one domain might not be so obvious to experts in another domain. Ambiguity in domain-specific language can erode the ability to enable different domain modeling applications to interoperate, unless the underlying language is understood and used as the basis for translation from one application to another. The engineering modeling software application industry has struggled for decades to enable these applications to interoperate. Industry standards have been developed, but they have not unified the industry. Why is this? The authors assert that the industry has relied on traditional database integration methods. The basic issue prohibiting successful application integration then is that traditional database-driven integration does not consider the distinct languages of each domain. An engineering models meaning is expressed through the underlying language of that engineering domain. In essence, traditional integration methods do not retain the semantic context (meaning) of the model. The basis of this research stems from the widely held assumption that systems engineering models are (or can be) structured according to the underlying semantic ontology of the model. This assumption can be imagined from two thoughts. 1) Digital systems engineering models are often represented using graph theory (the graph of a complex systems model can contain millions of nodes and edges). When examining the nodes one at a time and following the outbound edges of each node one by one, one can end up with rudimentary statements about the model (i.e., node A relates to node B), as in a semantic graph. 2) Likewise, from the study of natural languages, a sentence can be structured into unambiguous triples of subject-predicate-object within formal and highly expressive semantic ontologies. The rudimentary statements about a systems model discerned with graph theory closely mimic the triples used in the ontologies that try to structure natural languages. In other words, a systems models semantic graph can be (or is) structured into an ontology. Additionally, it is well established in industry that through natural language processing (NLP), which provides the means to create language structures, that computers can interpret ontological graphs. Therefore, the authors hypothesized that if the integrity of the underlying semantic structure of a systems model is retained, the contextual meaning of the model is retained. By structuring system models into the triples of the underlying ontology during the transformation from one MBSE application to another, the authors have provided a proof of the concept that the meaning of a system model can be retained during transformation. The authors assert that this is the missing ingredient in effective systems model-to-model interoperability. ACKNOWLEDGEMENTS The authors would like to thank the FY19 Model Interoperability team members who provided a solid foundation for the FY20 team to leverage: John McCloud, for the work he did to guide us toward the right use of technology that will appropriately discover and manipulate ontologies. Carlos Tafoya, for the work he did to develop an application programming interface (API)/Adapter that would export ontology-based data from GENESYS. Peter Chandler, for the work he did to architect our overall integration solution, with an eye toward the future that would influence a large-scale federated production-level systems engineering digital model ecosystem.

Abstract not provided.

Abstract not provided.

Abstract not provided.