This dissertation describes detailed performance engineering and optimization of an unstructured computational aerodynamics software system with irregular memory accesses on various multi- and many-core emerging high performance computing scalable architectures, which are expected to be the building blocks of energy-austere exascale systems, and on which algorithmic- and architecture-oriented optimizations are essential for achieving worthy performance. We investigate several state-of-the-practice shared-memory optimization techniques applied to key kernels for the important problem class of unstructured meshes. We illustrate for a broad spectrum of emerging microprocessor architectures as representatives of the compute units in contemporary leading supercomputers, identifying and addressing performance challenges without compromising the floating-point numerics of the original code. While the linear algebraic kernels are bottlenecked by memory bandwidth for even modest numbers of hardware cores sharing a common address space, the edge-based loop kernels, which arise in the control volume discretization of the conservation law residuals and in the formation of the preconditioner for the Jacobian by finite-differencing the conservation law residuals, are compute-intensive and effectively exploit contemporary multi- and many-core processing hardware. We therefore employ low- and high-level algorithmic- and architecture-specific code optimizations and tuning in light of thread- and data-level parallelism, with a focus on strong thread scaling at the node-level. Our approaches are based upon novel multi-level hierarchical workload distribution mechanisms of data across different compute units (from the address space down to the registers) within every hardware core. We analyze the demonstrated aerodynamics application on specific computing architectures to develop certain performance metrics and models to bespeak the upper and lower bounds of the performance. We present significant full application speedup relative to the baseline code, on a succession of many-core processor architectures, i.e., Intel Xeon Phi Knights Corner (5.0x) and Knights Landing (2.9x). In addition, the performance of Knights Landing outperforms, at significantly lower power consumption, Intel Xeon Skylake with nearly twofold speedup. These optimizations are expected to be of value for many other unstructured mesh partial differential equation-based scientific applications as multi- and many- core architecture evolves.

[pt] Técnicas de visualização científica criam imagens na tentativa de revelar estruturas e fenômenos complexos. Técnicas ilustrativas têm sido incorporadas aos sistemas de visualizacão científica para melhorar a expressividade de tais imagens. A visualização de linhas caracteríticas é uma técnica importante para transmitir uma melhor informacão sobre a forma das superfícies. Neste trabalho, propomos combinar visualização volumétrica de malhas não estruturadas com isosuperfícies ilustradas. Isto é feito estendendo um algoritmo de traçado de raio em GPU para incorporar ilustração com linhas de variação extrema da iluminação(photic extremum lines), um tipo de linha característica que captura mudanças bruscas de luminância, revelando formas de um jeito perceptualmente correto. / [en] Scientic visualization techniques create images attempting to reveal complex structures and phenomena. Illustrative techniques have been incorporated to scientic visualization systems in order to improve the expressiveness of such images. The rendering of feature lines is an important technique for better depicting surface shapes and features. In this thesis, we propose to combine volume visualization of unstructured meshes with illustrative isosurfaces. This is accomplished by extending a GPU-based ray-casting algorithm to incorporate illustration with photic extremum lines, a type of feature lines able to capture sudden changes of luminance, conveying shapes in a perceptually correct way.

<p>Computational Fluid Dynamic (CFD) simulations often produce datasets defined over unstructured grids with solid boundaries. Though unstructured grids allow for the flexible representation of this geometry and the refinement of the grid resolution, they suffer from high storage cost, non-trivial spatial queries, and low reconstruction smoothness. Rectilinear grids, in contrast, have a negligible memory footprint and readily support smooth data reconstruction, though with reduced geometric flexibility.</p><p>This thesis is concerned with the creation of accurate reconstruction of large unstructured datasets on rectilinear grids with the capability of representing complex boundaries. We present an efficient method to automatically determine the geometry of a rectilinear grid upon which a low-error data reconstruction can be achieved with a given reconstruction kernel. Using this rectilinear grid, we address the potential ill-posedness of the data fitting problem, as well as the necessary balance between smoothness and accuracy, through a bi-level smoothness regularization. To tackle the computational challenge posed by very large input datasets and high-resolution reconstructions, we propose a block-based approach that allows us to obtain a seamless global approximation solution from a set of independently computed sparse least-squares problems. </p><p></p><p>We endow the approximated rectilinear grid with boundary handling capabilities that allows for accommodating challenging boundaries while supporting high-order reconstruction kernels. Results are presented for several 3D datasets that demonstrate the quality of the visualization results that our reconstruction enables, at a greatly reduced computational and memory cost. Our data representation enjoys all the benefits of conventional rectilinear grids while addressing their fundamental geometric limitations.</p>

Meshes based on triangulation (2D) or tetrahedralization (3D) are widely used in applications such as computer graphics, interpolation, surveying, and terrain modelling. Although the most important use is in numerical methods for the solution of partial differential equations. This solution is use for simulation of complex physical processes, which are described by this equations. The main topic of this thesis is mash generation.

This work examines the feasibility of a novel high-order numerical method, which has been termed Flux Correction. It has been given this name because it "corrects" the flux terms of an established numerical method, cancelling various error terms in the fluxes and making the method higher-order. In this work, this change is made to a traditionally second-order finite volume Galerkin method. To accomplish this, higher-order gradients of solution variables, as well as gradients of the fluxes are introduced to the method. Gradients are computed using lagrange interpolations in a fashion reminiscent of Finite Element techniques. For the Euler Equations, Flux Correction is compared against Flux Reconstruction, a derivative of the high-order Discontinuous Galerkin and Spectral Difference methods, both of which are currently popular areas of research in high-order numerical methods. Flux Correction is found to compare favorably in terms of accuracy, and exceeds expectations for convergence rates. For the full Navier-Stokes Equations, the effect of curved elements on Flux Correction are examined. Flux Correction is found to react negatively to curved elements due to the gradient procedure's poor handling of high-aspect ratio elements.

Cílem mé disertační práce je analyzovat a vyvinout parametrizační metodu pro 2D a 3D tvarové optimalizace v kontextu průmyslového aerodynamického návrhu letounu založeném na CFD simulacích. Aerodynamická tvarová optimalizace je efektivní nástroj, který si klade za cíl snížení nákladů na návrh letounů. Nástroj založený na automatickém hledání optimálního tvaru. Klíčovou částí úspěšného optimalizačního procesu je použití vhodné parametrizační metody, metody schopné garantovat možnost dosažení optimálního tvaru. Parametrizační metody obecně používané v oblasti aerodynamické tvarové optimalizace momentálně nejsou připravený na komplikované průmyslové aplikace vyskytující se u moderních dopravních letounů, které mají šípová zalomená křídla s winglety a motorovými gondolami, přechodové prvky spojující např. trup s křídlem atd.. Existuje tedy potřeba nalezení obecné parametrizační metody, která bude aplikovatelná na širokou škálu různých geometrických tvarů. Free-Form Deformation (FFD[1]) parametrizace může, vzhledem ke svým schopnostem při zacházení s geometrií, být odpovědí na tuto potřebu. Adaptivní parametrizace by se měla být schopna automaticky přizpůsobit danému tvaru tak, aby byly její kontrolní body vhodně rozmístěny. Což umožní dostatečnou kontrolu deformací objektu, která zaručí možnost vytvoření optimálního tvaru objektu a splnění geometrických omezení. Primární aplikací takové parametrizační metody je deformace tvaru objektu. Dalším navrhovaným cílem je modifikace FFD parametrizační metody pro současné deformace tvaru objektu a CFD výpočetní sítě, umožnující velké deformace objektu při zachování kvality výpočetní sítě.

No description available.

The MSU Loci/BLAST CFD code was used to study blast wave interactions with structures for different urban environments. A series of analyses which included single building structures inside of ERDC's Blast Load Simulator (BLS) with different obliquity orientations to the flow direction, two building structures inside the BLS with varying gap distances between the structures, and open air blast simulations with four structure scenarios at different building spacings and different blast orientations were performed. Unstructured and overset grid techniques were used during the modeling process and were compared for consistency with shock physics and computational performance. Results show Loci/BLAST's capability to accurately model blast wave interactions in urban environments for both unstructured and overset grids.

Truncation errors and computational cost are obstacles that still hinder large-scale applications of the Computational Fluid Dynamics method. The discontinuous Galerkin method is one of the high-order schemes utilized extensively in recent years, which is locally conservative, stable, and high-order accurate. Besides that, it can handle complex geometries and irregular meshes with hanging nodes. In this document, the nondimensional compressible Euler equations and Reynolds- Averaged Navier-Stokes equations are discretized by discontinuous Galerkin methods with a two-equations turbulence model on both structured and unstructured meshes. The traditional equation of state for an ideal gas model is substituted by a multispecies thermodynamics model in order to complete the governing equations. An approximate Riemann solver is used for computing the convective flux, and the diffusive flux is approximated with some internal penalty based schemes. The temporal discretization of the partial differential equations is either performed explicitly with the aid of Rung-Kutta methods or with semi-implicit methods. Inspired by the artificial viscosity diffusion based limiter for shock-capturing method, which has been extensively studied, a novel and robust technique based on the introduction of mass diffusion to the species governing equations to guarantee that the species mass fractions remain positive has been thoroughly investigated. This contact-surface-capturing method is conservative and a high order of accuracy can be maintained for the discontinuous Galerkin method. For each time step of the algorithm, any trouble cell is first caught by the contact-surface discontinuity detector. Then some amount of mass diffusions are added to the governing equations to change the gas mixtures and arrive at an equilibrium point satisfying some conditions. The species properties are reasonable without any oscillations. Computations are performed for many steady and unsteady flow problems. For general non-mixing fluid flows, the classical air-helium shock bubble interaction problem is the central test case for the high-order discontinuous Galerkin method with a mass diffusion based limiter chosen. The computed results are compared with experimental, exact, and empirical data to validate the fluid dynamic solver.

No description available.