Cette thèse étudie l'application de méthodes de parallélisation en temps pour la simulation numérique directe des écoulements turbulents. Après une étude préliminaire, on choisit de se focaliser sur l'algorithme Parareal avec grossissement spatial. Le comportement de l'algorithme est étudié en premier lieu sur l'équation d'advection, comme simplification des équations de Navier-Stokes, par une analyse de Fourier et une série d'expériences numériques, afin d'en cerner les mécanismes et paramètres dimensionnants. L'algorithme est ensuite étudié dans un contexte HPC, à l'aide du code de simulation massivement parallèle Hybrid. Deux situations d'écoulements turbulents tridimensionnels sont à l'étude: la décroissance d'une turbulence homogène isotrope et l'écoulement de canal turbulent. Ce travail propose une première mesure de l'efficacité de la parallélisation combinée espace-temps, ainsi qu'une évaluation précise de la capacité de l'algorithme à représenter les propriétés physiques de la turbulence. / This thesis aims at studying the application of time-parallel integration methods for the Direct Numerical Simulation of turbulent flows. After a preliminary study, we choose to focus on the Parareal algorithm with spatial coarsening. The behavior of the algorithm is first studied on the advection equation, as a simplified model for the Navier-Stokes equations, using a Fourier analysis and numerical experiments, to understand its mechanisms and identify the relevant parameters. The algorithm is then studied in a HPC context, using the massively parallel CFD simulation code Hybrid. Two tri-dimensional turbulent flow problems are investigated : the decay of an Homogeneous Isotropic Turbulence and the Turbulent Channel Flow. This work offers a first evaluation of combined space-time parallel efficiency, and analyse the algorithm’s abilities to correctly reproduce the physical properties of turbulence.

La simulation interactive du corps humain est un problème crucial en informatique médicale. Les approches sont multiples pour arriver à cet objectif. Diminuer le temps de calcul est le leitmotiv d'un grand nombre de travaux ces dernières années. Pour les recherches qui utilisent des modèles physiques inspirés de la Mécanique des Milieux Continus pour la simulation des objets déformables, ce sont principalement les forces internes et leurs dérivées qui font l'objet d'études pour l'amélioration des performances au niveau du temps de calcul. Nous avons choisi de développer la Méthode des Masses-Tenseurs, modèle physique souvent utilisé pour son bon compromis temps de calcul — précision. Notre première contribution est l'utilisation du calcul formel pour la génération des équations des forces internes et de leurs dérivées. Notre deuxième contribution est la parallélisation de ce modèle physique en calculant les équations générées sur le GPU. Notre troisième contribution est l'extension de ce modèle physique à d'autres types d'éléments : triangle, quadrangle, hexaèdre, prisme et pyramide. Tenir compte des déformations pour utiliser la loi de comportement la plus efficace en temps de calcul lorsque c'est possible, est une stratégie que nous avons mis en place. Dans la même idée, nous prenons en compte la géométrie du modèle à simuler pour utiliser des éléments plus complexes mais en nombre réduit. Pour utiliser ces stratégies, nous avons développé et utilisé des modèles mixtes en loi de comportement et en type d'éléments. Nos travaux se placent dans le contexte du projet ETOILE pour le développement d'un modèle biomécanique du système respiratoire / Interactive simulation of the human body is a crucial issue in medical computer sciences. There are many approaches to reach this goal. Reducing the computation time is the leitmotiv of a large number of efforts in recent years. For researches which use physical models derived from continuum mechanics for the simulation of deformable objects, it is primarily the internal forces and their derivatives which are the subject of study for improving computation time. We chose to develop the Tensor Mass Method, a physical model often used for its good computation time vs accuracy trade-off. Our first contribution is the use of computer algebra to generate the internal forces and their derivatives. Our second contribution is the parallelization of this physical model by computing the generated equations on the GPU. Our third contribution is an extension of this physical model to other elements : triangle, quandrangle , hexahedron, prism and pyramid. Considering deformations to use the most effective constitutive law in terms of computation time whenever possible is a good strategy that we started developing. In the same idea, we take the geometry of the simulated model into account to introduce more complex elements, albeit in reduced numbers. To use these strategies, we have developed mixed models in constitutive laws and elements. Our research was performed in the framework of the ETOILE project, to develop a biomechanical model of the respiratory system

The progression of technical development has yielded practicable camera systems for the acquisition of so called depth maps, images with depth information. Images and movies with depth information open the door for new types of applications in the area of computer graphics and vision. That implies that they will need to be processed in all increasing volumes. Increased depth image processing puts forth the demand for a standardized data format for the exchange of image data with depth information, both still and animated. Software to convert acquired depth data to such videoformats is highly necessary. This diploma thesis sheds light on many of the issues that come with this new task group. It spans from data acquisition over readily available software for the data encoding to possible future applications. Further, a software architecture fulfilling all of the mentioned demands is presented. The encoder is comprised of a collection of UNIX programs that generate MPEG Z/Alpha, an MPEG2 based video format. MPEG Z/Alpha contains beside MPEG2's standard data streams one extra data stream to store image depth information (and transparency). The decoder suite, called TexMPEG, is a C library for the in-memory decompression of MPEG Z/Alpha. Much effort has been put into video decoder parallelization, and TexMPEG is now capable of decoding multiple video streams, not only in parallel internally, but also with inherent frame synchronization between parallely decoded MPEG videos.

Scientific applications that operate on large data sets require huge amount of computation power and memory. These applications are typically run on High Performance Computing (HPC) systems that consist of multiple compute nodes, connected over an network interconnect such as InfiniBand. Each compute node has its own memory and does not share the address space with other nodes. A significant amount of work has been done in past two decades on parallelizing for distributed-memory architectures. A majority of this work was done in developing compiler technologies such as high performance Fortran (HPF) and partitioned global address space (PGAS). However, several steps involved in achieving good performance remained manual. Hence, the approach currently used to obtain the best performance is to rely on highly tuned libraries such as ScaLAPACK. The objective of this work is to improve automatic compiler and runtime support for distributed-memory clusters for regular programs. Regular programs typically use arrays as their main data structure and array accesses are affine functions of outer loop indices and program parameters. A lot of scientific applications such as linear-algebra kernels, stencils, partial differential equation solvers, data-mining applications and dynamic programming codes fall in this category. In this work, we propose techniques for finding computation mapping and data allocation when compiling regular programs for distributed-memory clusters. Techniques for transformation and detection of parallelism, relying on the polyhedral framework already exist. We propose automatic techniques to determine computation placements for identified parallelism and allocation of data. We model the problem of finding good computation placement as a graph partitioning problem with the constraints to minimize both communication volume and load imbalance for entire program. We show that our approach for computation mapping is more effective than those that can be developed using vendor-supplied libraries. Our approach for data allocation is driven by tiling of data spaces along with a compiler assisted runtime scheme to allocate and deallocate tiles on-demand and reuse them. Experimental results on some sequences of BLAS calls demonstrate a mean speedup of 1.82× over versions written with ScaLAPACK. Besides enabling weak scaling for distributed memory, data tiling also improves locality for shared-memory parallelization. Experimental results on a 32-core shared-memory SMP system shows a mean speedup of 2.67× over code that is not data tiled.

No description available.

Les travaux présentés dans cette thèse proposent une méthode de génération automatique de tests à partir de modèles.Cette méthode emploie deux langages de modélisations UML4MBT et OCL4MBT qui ont été spécifiquement dérivées d’ UML et OCL pour la génération de tests. Ainsi les comportements, la structure et l’état initial du système sont décrits au travers des diagrammes de classes, d’objets et d’états-transitions.Pour générer des tests, l’évolution du modèle est représente sous la forme d’un système de transitions. Ainsi la construction de tests est équivalente à la découverte de séquences de transitions qui relient l’´état initial du système à des états validant les cibles de test.Ces séquences sont obtenues par la résolution de scénarios d’animations par des prouveurs SMT et solveurs CSP. Pour créer ces scénarios, des méta-modèles UML4MBT et CSP4MBT regroupant formules logiques et notions liées aux tests ont été établies pour chacun des outils.Afin d’optimiser les temps de générations, des stratégies ont été développé pour sélectionner et hiérarchiser les scénarios à résoudre. Ces stratégies s’appuient sur la parallélisation, les propriétés des solveurs et des prouveurs et les caractéristiques de nos encodages pour optimiser les performances. 5 stratégies emploient uniquement un prouveur et 2 stratégies reposent sur une collaboration du prouveur avec un solveur.Finalement l’intérêt de cette nouvelle méthode à été validée sur des cas d’études grâce à l’implémentation réalisée. / This thesis describes an automatic test generation process from models.This process uses two modelling languages, UML4MBT and OCL4MBT, created specificallyfor tests generation. Theses languages are derived from UML and OCL. Therefore the behaviours,the structure and the initial state of the system are described by the class diagram, the objectdiagram and the state-chart.To generate tests, the evolution of the model is encoded with a transition system. Consequently,to construct a test is to find transition sequences that rely the initial state of the system to thestates described by the test targets.The sequence are obtained by the resolution of animation scenarios. This resolution is executedby SMT provers and CSP solvers. To create the scenario, two dedicated meta-models, UML4MBTand CSP4MBT have been established. Theses meta-models associate first order logic formulas withthe test notions.7 strategies have been developed to improve the tests generation time. A strategy is responsiblefor the selection and the prioritization of the scenarios. A strategy is built upon the properties ofthe solvers and provers and the specification of our encoding process. Moreover the process canalso be paralleled to get better performance. 5 strategies employ only a prover and 2 make theprover collaborate with a solver.Finally the interest of this process has been evaluated through a list of benchmark on variouscases studies.

Dans cette thèse, nous présentons notre travail de recherche dans le domaine du calcul haute performance en mécanique des fluides. Avec la demande croissante de simulations à haute résolution, il est devenu important de développer des solveurs numériques pouvant tirer parti des architectures récentes comprenant des processeurs multi-cœurs et des accélérateurs. Nous nous proposons dans cette thèse de développer un solveur efficace pour la résolution sur architectures hétérogènes CPU/GPU des équations de Navier-Stokes (NS) relatives aux écoulements 3D de fluides incompressibles.Tout d'abord nous présentons un aperçu de la mécanique des fluides avec les équations de NS pour fluides incompressibles et nous présentons les méthodes numériques existantes. Nous décrivons ensuite le modèle mathématique, et la méthode numérique choisie qui repose sur une technique de prédiction-projection incrémentale.Nous obtenons une distribution équilibrée de la charge de calcul en utilisant une méthode de décomposition de domaines. Une parallélisation à deux niveaux combinée avec de la vectorisation SIMD est utilisée dans notre implémentation pour exploiter au mieux les capacités des machines multi-cœurs. Des expérimentations numériques sur différentes architectures parallèles montrent que notre solveur NS obtient des performances satisfaisantes et un bon passage à l'échelle.Pour améliorer encore la performance de notre solveur NS, nous intégrons le calcul sur GPU pour accélérer les tâches les plus coûteuses en temps de calcul. Le solveur qui en résulte peut être configuré et exécuté sur diverses architectures hétérogènes en spécifiant le nombre de processus MPI, de threads, et de GPUs.Nous incluons également dans ce manuscrit des résultats de simulations numériques pour des benchmarks conçus à partir de cas tests physiques réels. Les résultats obtenus par notre solveur sont comparés avec des résultats de référence. Notre solveur a vocation à être intégré dans une future bibliothèque de mécanique des fluides pour le calcul sur architectures parallèles CPU/GPU. / In this PhD thesis, we present our research in the domain of high performance software for computational fluid dynamics (CFD). With the increasing demand of high-resolution simulations, there is a need of numerical solvers that can fully take advantage of current manycore accelerated parallel architectures. In this thesis we focus more specifically on developing an efficient parallel solver for 3D incompressible Navier-Stokes (NS) equations on heterogeneous CPU/GPU architectures. We first present an overview of the CFD domain along with the NS equations for incompressible fluid flows and existing numerical methods. We describe the mathematical model and the numerical method that we chose, based on an incremental prediction-projection method.A balanced distribution of the computational workload is obtained by using a domain decomposition method. A two-level parallelization combined with SIMD vectorization is used in our implementation to take advantage of the current distributed multicore machines. Numerical experiments on various parallel architectures show that this solver provides satisfying performance and good scalability.In order to further improve the performance of the NS solver, we integrate GPU computing to accelerate the most time-consuming tasks. The resulting solver can be configured for running on various heterogeneous architectures by specifying explicitly the numbers of MPI processes, threads and GPUs. This thesis manuscript also includes simulation results for two benchmarks designed from real physical cases. The computed solutions are compared with existing reference results. The code developed in this work will be the base for a future CFD library for parallel CPU/GPU computations.

This thesis presents a case of study of the development and performance analysis of a surface grading application with real-time compliance. We address the issue of spatial and temporal uniformity in the acquisition system. In a surface grading application it is crucial to ensure the uniform response of the system through time and space. All the results presented for surface grading were obtained using real data from the ceramic tile industry. The VxC TSG database is public and can be accessed at www.disca.upv.es/vision/vxctsg. We present a method based on soft colour-texture descriptors computed in perceptually uniform colour spaces. The method is parameterized and the involved factors are studied using two statistical procedures; experimental design and logistic regression. Although it is not a new theoretical contribution, we have found and demonstrate that a simple set of global colour and texture statistics, together with well-known classifiers, are powerful enough to meet stringent factory requirements for real-time and performance. Also the method is compared with two other approaches from the scientific literature; Colour Histograms and Centile-LBP. Finally, we explore the method's capacity for on-line inspection in a study of real-time compliance and parallelization. / López García, F. (2005). Real-time surface grading of ceramic tiles [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/1909 / Palancia

This diploma thesis deals with processes of electrical circuits calculations in the last years' worldwide standards like Dymola, MATLAB, Maple or SPICE applications. Circuits calculations are linked with methods for solving linear differential equations, used in this work also by verification of functionality of designed models for CMOS inverter, CMOS NAND, CMOS NOR. Numerical integration method in combination with Taylor series is a suitable method also for parallel calculations of CMOS VLSI circuits. CMOS circuits simulation was implemented with this method in applications in MATLAB language, solving circuits, represented by differential equations. Functionality of the applications was verified by some real examples. Significant acceleration of calculations using Taylor series compared to other methods is an important factor in choosing methods used in circuit simulations.

This master's thesis addresses processing of flow monitoring records from a point of view of an IPFIX collector. It analysis the current solution of the modular collector, which went through considerable historical development, and focuses on revealing its strengths and weaknesses. Based on acquired knowledge, a new collector is designed. The new solution, which significantly modifies individual components for processing of flow records, focuses on high throughput and adds missing functionalities. The document also compares performance of both generations and the new collector clearly dominates.