Search NASA⌕ Search

SEARCH · Search NASA

Results for “stencil computation”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 163 records · Page 9

Analysis and algorithms for a regularized Cauchy problem arising from a non-linear elliptic PDE for seismic velocity estimation

In the present work we derive and study a nonlinear elliptic PDE coming from the problem of estimation of sound speed inside the Earth. The physical setting of the PDE allows us to pose only a Cauchy problem, and hence is ill-posed. However we are still able to solve it numerically on a long enough time interval to be of practical use. We used two approaches. The first approach is a finite difference time-marching numerical scheme inspired by the Lax-Friedrichs method. The key features of this scheme is the Lax-Friedrichs averaging and the wide stencil in space. The second approach is a spectral Chebyshev method with truncated series. We show that our schemes work because of (1) the special input corresponding to a positive finite seismic velocity, (2) special initial conditions corresponding to the image rays, (3) the fact that our finite-difference scheme contains small error terms which damp the high harmonics; truncation of the Chebyshev series, and (4) the need to compute the solution only for a short interval of time. We test our numerical scheme on a collection of analytic examples and demonstrate a dramatic improvement in accuracy in the estimation of the sound speed inside the Earth in comparison with the conventional Dix inversion. Our test on the Marmousi example confirms the effectiveness of the proposed approach.

97 MATHEMATICS AND COMPUTING↗

F-ANG+: A 3-D Augmented-Stencil Face-Averaged Nodal-Gradient Cell-Centered Finite-Volume Method for Hypersonic Flows

We describe the extension of a 2-D simplified face-averaged nodal-gradient (F-ANG) method to 3-D and demonstrate that the 3-D simplified F-ANG method is accomplished by augmenting the nodecentered gradient least squares stencil. This augmented stencil F-ANG method is shown to result in advection and diffusion schemes that are stable for hexahedral, prismatic, pyramidal and tetrahedral cells without having to resort to cell-averaged nodal gradients. In addition, we describe the modifications to the augmented stencil required to support the use of wall function boundary conditions. Finally we describe a consistent, face-stencil based multi-dimensional limiter procedure (MLP), and show it to be fully consistent and compatible with the linearity-preserving unstructured- MUSCL (LP-U-MUSCL) scheme for all values of kappa. These methods and schema are implemented in the cell-centered finite-volume code VULCAN-CFD, which is then used to investigate whether the robustness improvements demonstrated in 2-D carry over to 3-D by computing hypersonic flows using mixed-element grids as well as highly adapted tetrahedral grids.

Weighted Least-Squares↗

Accurate modeling of plasma acceleration with arbitrary order pseudo-spectral particle-in-cell methods

Particle in Cell (PIC) simulations are a widely used tool for the investigation of both laser- and beam-driven plasma acceleration. It is a known issue that the beam quality can be artificially degraded by numerical Cherenkov radiation (NCR) resulting primarily from an incorrectly modeled dispersion relation. Pseudo-spectral solvers featuring infinite order stencils can strongly reduce NCR - or even suppress it - and are therefore well suited to correctly model the beam properties. For efficient parallelization of the PIC algorithm, however, localized solvers are inevitable. Arbitrary order pseudo-spectral methods provide this needed locality. Yet, these methods can again be prone to NCR. Here in this paper, we show that acceptably low solver orders are sufficient to correctly model the physics of interest, while allowing for parallel computation by domain decomposition.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

A High-Order Finite Spectral Volume Method for Conservation Laws on Unstructured Grids

A time accurate, high-order, conservative, yet efficient method named Finite Spectral Volume (FSV) is developed for conservation laws on unstructured grids. The concept of a 'spectral volume' is introduced to achieve high-order accuracy in an efficient manner similar to spectral element and multi-domain spectral methods. In addition, each spectral volume is further sub-divided into control volumes (CVs), and cell-averaged data from these control volumes is used to reconstruct a high-order approximation in the spectral volume. Riemann solvers are used to compute the fluxes at spectral volume boundaries. Then cell-averaged state variables in the control volumes are updated independently. Furthermore, TVD (Total Variation Diminishing) and TVB (Total Variation Bounded) limiters are introduced in the FSV method to remove/reduce spurious oscillations near discontinuities. A very desirable feature of the FSV method is that the reconstruction is carried out only once, and analytically, and is the same for all cells of the same type, and that the reconstruction stencil is always non-singular, in contrast to the memory and CPU-intensive reconstruction in a high-order finite volume (FV) method. Discussions are made concerning why the FSV method is significantly more efficient than high-order finite volume and the Discontinuous Galerkin (DG) methods. Fundamental properties of the FSV method are studied and high-order accuracy is demonstrated for several model problems with and without discontinuities.

Wang, Z. J.↗

A multilevel adaptive projection method for unsteady incompressible flow

There are two main requirements for practical simulation of unsteady flow at high Reynolds number: the algorithm must accurately propagate discontinuous flow fields without excessive artificial viscosity, and it must have some adaptive capability to concentrate computational effort where it is most needed. We satisfy the first of these requirements with a second-order Godunov method similar to those used for high-speed flows with shocks, and the second with a grid-based refinement scheme which avoids some of the drawbacks associated with unstructured meshes. These two features of our algorithm place certain constraints on the projection method used to enforce incompressibility. Velocities are cell-based, leading to a Laplacian stencil for the projection which decouples adjacent grid points. We discuss features of the multigrid and multilevel iteration schemes required for solution of the resulting decoupled problem. Variable-density flows require use of a modified projection operator--we have found a multigrid method for this modified projection that successfully handles density jumps of thousands to one. Numerical results are shown for the 2D adaptive and 3D variable-density algorithms.

Howell, Louis H.↗

Performance Optimization and Auto-Tuning

In the broader computational research community, one subject of recent research is the problem of adapting algorithms to make effective use of multi- and many-core processors. Effective use of these architectures, which have complex memory hierarchies with many layers of cache, typically involves a careful examination of how an algorithm moves data through the memory hierarchy. Unfortunately, there is often a non-obvious relationship between algorithmic parameters like blocking strategies, and their impact on memory utilization, and, in turn, the relationship with runtime performance. Auto-tuning is an empirical method used to discover optimal values for tunable algorithmic parameters under such circumstances. The challenge is compounded by the fact that the settings that produce the best performance for a given problem and a given platform may not be the best for a different problem on the same platform, or the same problem on a different platform. The high performance visualization research community has begun to explore and adapt the principles of auto-tuning for the purpose of optimizing codes on modern multi- and many-core processors. This report focuses on how performance optimization studies reveal a dramatic variation in performance for two fundamental visualization algorithms: one based on a stencil operation having structured, uniform memory access, and the other is ray casting volume rendering, which uses unstructured memory access patterns. The two case studies highlighted in this report show the extra effort required to optimize such codes by adjusting the tunable algorithmic parameters can return substantial gains in performance. Additionally, these case studies also explore the potential impact of and the interaction between algorithmic optimizations and tunable algorithmic parameters, along with the potential performance gains resulting from leveraging architecture-specific features.

97 MATHEMATICS AND COMPUTING↗

Application of traditional CFD methods to nonlinear computational aeroacoustics problems

This paper describes an implementation of a high order finite difference technique and its application to the category 2 problems of the ICASE/LaRC Workshop on Computational Aeroacoustics (CAA). Essentially, a popular Computational Fluid Dynamics (CFD) approach (central differencing, Runge-Kutta time integration and artificial dissipation) is modified to handle aeroacoustic problems. The changes include increasing the order of the spatial differencing to sixth order and modifying the artificial dissipation so that it does not significantly contaminate the wave solution. All of the results were obtained from the CM5 located at the Numerical Aerodynamic Simulation Laboratory. lt was coded in CMFortran (very similar to HPF), using programming techniques developed for communication intensive large stencils, and ran very efficiently.

Chyczewski, Thomas S.↗

Implementing Connected Component Labeling as a User Defined Operator for SciDB

We have implemented a flexible User Defined Operator (UDO) for labeling connected components of a binary mask expressed as an array in SciDB, a parallel distributed database management system based on the array data model. This UDO is able to process very large multidimensional arrays by exploiting SciDB's memory management mechanism that efficiently manipulates arrays whose memory requirements far exceed available physical memory. The UDO takes as primary inputs a binary mask array and a binary stencil array that specifies the connectivity of a given cell to its neighbors. The UDO returns an array of the same shape as the input mask array with each foreground cell containing the label of the component it belongs to. By default, dimensions are treated as non-periodic, but the UDO also accepts optional input parameters to specify periodicity in any of the array dimensions. The UDO requires four stages to completely label connected components. In the first stage, labels are computed for each subarray or chunk of the mask array in parallel across SciDB instances using the weighted quick union (WQU) with half-path compression algorithm. In the second stage, labels around chunk boundaries from the first stage are stored in a temporary SciDB array that is then replicated across all SciDB instances. Equivalences are resolved by again applying the WQU algorithm to these boundary labels. In the third stage, relabeling is done for each chunk using the resolved equivalences. In the fourth stage, the resolved labels, which so far are "flattened" coordinates of the original binary mask array, are renamed with sequential integers for legibility. The UDO is demonstrated on a 3-D mask of O(1011) elements, with O(108) foreground cells and O(106) connected components. The operator completes in 19 minutes using 84 SciDB instances.

UDO↗

TEMPI: An Interposed MPI Library with Canonical Representation of MPI Datatypes [Slides]

These points are covered in this presentation: Distributed GPU stencil, non-contiguous data; Equivalence of strided datatypes and minimal representation; GPU communication methods; Deploying on managed systems; Large messages and MPI datatypes; Translation and canonicalization; Automatic model-driven transfer method selection; and Interposed library implementation.

97 MATHEMATICS AND COMPUTING↗

A Two-Dimensional Linear Bicharacteristic Scheme for Electromagnetics

The upwind leapfrog or Linear Bicharacteristic Scheme (LBS) has previously been implemented and demonstrated on one-dimensional electromagnetic wave propagation problems. This memorandum extends the Linear Bicharacteristic Scheme for computational electromagnetics to model lossy dielectric and magnetic materials and perfect electrical conductors in two dimensions. This is accomplished by proper implementation of the LBS for homogeneous lossy dielectric and magnetic media and for perfect electrical conductors. Both the Transverse Electric and Transverse Magnetic polarizations are considered. Computational requirements and a Fourier analysis are also discussed. Heterogeneous media are modeled through implementation of surface boundary conditions and no special extrapolations or interpolations at dielectric material boundaries are required. Results are presented for two-dimensional model problems on uniform grids, and the Finite Difference Time Domain (FDTD) algorithm is chosen as a convenient reference algorithm for comparison. The results demonstrate that the two-dimensional explicit LBS is a dissipation-free, second-order accurate algorithm which uses a smaller stencil than the FDTD algorithm, yet it has less phase velocity error.

Beggs, John H.↗

A priori analysis on deep learning of subgrid-scale parameterizations for Kraichnan turbulence

In the present study, we investigate different data-driven parameterizations for large eddy simulation of two-dimensional turbulence in the a priori settings. These models utilize resolved flow field variables on the coarser grid to estimate the subgrid-scale stresses. We use data-driven closure models based on localized learning that employs a multilayer feedforward artificial neural network with point-to-point mapping and neighboring stencil data mapping, and convolutional neural network fed by data snapshots of the whole domain. The performance of these data-driven closure models is measured through a probability density function and is compared with the dynamic Smagorinksy model (DSM). The quantitative performance is evaluated using the cross-correlation coefficient between the true and predicted stresses. We analyze different frameworks in terms of the amount of training data, selection of input and output features, their characteristics in modeling with accuracy, and training and deployment computational time. We also demonstrate computational gain that can be achieved using the intelligent eddy viscosity model that learns eddy viscosity computed by the DSM instead of subgrid-scale stresses. Here, we detail the hyperparameters optimization of these models using the grid search algorithm.

42 ENGINEERING↗

Efficient implicit LES method for the simulation of turbulent cavitating flows

We present a numerical method for efficient large-eddy simulation of compressible liquid flows with cavitation based on an implicit subgrid-scale model. Phase change and subgrid-scale interface structures are modeled by a homogeneous mixture model that assumes local thermodynamic equilibrium. Unlike previous approaches, emphasis is placed on operating on a small stencil (at most four cells). The truncation error of the discretization is designed to function as a physically consistent subgrid-scale model for turbulence. We formulate a sensor functional that detects shock waves or pseudo-phase boundaries within the homogeneous mixture model for localizing numerical dissipation. In smooth regions of the flow field, a formally non-dissipative central discretization scheme is used in combination with a regularization term to model the effect of unresolved subgrid scales. The new method is validated by computing standard single- and two-phase test-cases. Comparison of results for a turbulent cavitating mixing layer obtained with the new method demonstrates its suitability for the target applications.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

High-order essentially non-oscillatory methods for computational aeroacoustics

The desire to obtain acoustic information from the numerical solution of a nonlinear system of equations is a demanding proposition for a computational algorithm. High-order accuracy is required for the propagation of high-frequency, low-amplitude waves. In addition, it is desirable to highly resolve discontinuities that can develop in the solutions of the Euler or Navier-Stokes equations. The class of essentially non-oscillatory (ENO) shock-capturing schemes has been designed to have both of these properties. The dual capacity of ENO schemes for high-order accuracy and non-oscillatory shock-capturing is achieved through the use of adaptive stenciling, which makes these schemes highly nonlinear. These schemes are briefly described and referenced herein. A fourth-order algorithm is then applied to the solution of an acoustic wave in a quasi-one-dimensional converging-diverging nozzle.

Casper, Jay↗

A Reconstructed Discontinuous Galerkin Method for the Compressible Euler Equations on Arbitrary Grids

A reconstruction-based discontinuous Galerkin (DG) method is presented for the solution of the compressible Euler equations on arbitrary grids. By taking advantage of handily available and yet invaluable information, namely the derivatives, in the context of the discontinuous Galerkin methods, a solution polynomial of one degree higher is reconstructed using a least-squares method. The stencils used in the reconstruction involve only the van Neumann neighborhood (face-neighboring cells) and are compact and consistent with the underlying DG method. The resulting DG method can be regarded as an improvement of a recovery-based DG method in the sense that it shares the same nice features as the recovery-based DG method, such as high accuracy and efficiency, and yet overcomes some of its shortcomings such as a lack of flexibility, compactness, and robustness. The developed DG method is used to compute a variety of flow problems on arbitrary meshes to demonstrate the accuracy and efficiency of the method. The numerical results indicate that this reconstructed DG method is able to obtain a third-order accurate solution at a slightly higher cost than its second-order DG method and provide an increase in performance over the third order DG method in terms of computing time and storage requirement.

97 MATHEMATICS AND COMPUTING↗

Helium: lifting high-performance stencil kernels from stripped x86 binaries to halide DSL code

Highly optimized programs are prone to bit rot, where performance quickly becomes suboptimal in the face of new hardware and compiler techniques. In this paper we show how to automatically lift performance-critical stencil kernels from a stripped x86 binary and generate the corresponding code in the high-level domain-specific language Halide. Using Halide's state-of-the-art optimizations targeting current hardware, we show that new optimized versions of these kernels can replace the originals to rejuvenate the application for newer hardware. The original optimized code for kernels in stripped binaries is nearly impossible to analyze statically. Instead, we rely on dynamic traces to regenerate the kernels. We perform buffer structure reconstruction to identify input, intermediate and output buffer shapes. Here, we abstract from a forest of concrete dependency trees which contain absolute memory addresses to symbolic trees suitable for high-level code generation. This is done by canonicalizing trees, clustering them based on structure, inferring higher-dimensional buffer accesses and finally by solving a set of linear equations based on buffer accesses to lift them up to simple, high-level expressions. Helium can handle highly optimized, complex stencil kernels with input-dependent conditionals. We lift seven kernels from Adobe Photoshop giving a 75 % performance improvement, four kernels from Irfan View, leading to 4.97 x performance, and one stencil from the mini GMG multigrid benchmark netting a 4.25 x improvement in performance. We manually rejuvenated Photoshop by replacing eleven of Photoshop's filters with our lifted implementations, giving 1.12 x speedup without affecting the user experience.

97 MATHEMATICS AND COMPUTING↗

A fourth-order Cartesian grid embeddedboundary method for Poisson’s equation

In this paper, we present a fourth-order algorithm to solve Poisson's equation in two and three dimensions. We use a Cartesian grid, embedded boundary method to resolve complex boundaries. We use a weighted least squares algorithm to solve for our stencils. We use convergence tests to demonstrate accuracy and we show the eigenvalues of the operator to demonstrate stability. We compare accuracy and performance with an established second-order algorithm. We also discuss in depth strategies for retaining higher-order accuracy in the presence of nonsmooth geometries.

97 MATHEMATICS AND COMPUTING↗

Distributed Relaxation for Conservative Discretizations

A multigrid method is defined as having textbook multigrid efficiency (TME) if the solutions to the governing system of equations are attained in a computational work that is a small (less than 10) multiple of the operation count in one target-grid residual evaluation. The way to achieve this efficiency is the distributed relaxation approach. TME solvers employing distributed relaxation have already been demonstrated for nonconservative formulations of high-Reynolds-number viscous incompressible and subsonic compressible flow regimes. The purpose of this paper is to provide foundations for applications of distributed relaxation to conservative discretizations. A direct correspondence between the primitive variable interpolations for calculating fluxes in conservative finite-volume discretizations and stencils of the discretized derivatives in the nonconservative formulation has been established. Based on this correspondence, one can arrive at a conservative discretization which is very efficiently solved with a nonconservative relaxation scheme and this is demonstrated for conservative discretization of the quasi one-dimensional Euler equations. Formulations for both staggered and collocated grid arrangements are considered and extensions of the general procedure to multiple dimensions are discussed.

Diskin, Boris↗

An Angular Method with Position Control for Block Mesh Squareness Improvement

We optimize a target function de ned by angular properties with a position control term for a basic stencil with a block-structured mesh, to improve element squareness in 2D and 3D. Comparison with the condition number method shows that besides a similar mesh quality regarding orthogonality can be achieved as the former does, the new method converges faster and provides a more uniform global mesh spacing in our numerical tests.

97 MATHEMATICS AND COMPUTING↗