Aplicações podem apresentar laços perfeitamente aninhados que demandam um alto poder de processamento. Diversas aplicações científicas contêm laços aninhados em suas estruturas. Tais laços podem processar computações heterogêneas. Uma solução para reduzir o tempo de execução desta classe de aplicações é a paralelização destes laços. A heterogeneidade dos tempos de execução de computações presentes nas iterações de laços perfeitamente aninhados demanda uma paralelização adequada visando uma distribuição de carga homogênea entre os recursos computacionais para reduzir a ociosidade de tais recursos. Esta heterogeneidade implica em um número ideal de recursos computacionais a partir do qual, o seu aumento não impactaria no ganho de desempenho, uma vez que, o tempo mínimo possível é o tempo de execução da tarefa que consome o maior tempo de processamento. Neste trabalho é proposta uma metodologia e ferramentas para paralelização de laços perfeitamente aninhados sem dependência de dados e com processamento heterogêneo em sistemas paralelos e distribuídos. A implementação da metodologia proposta em aplicações melhora o desempenho da execução e reduz a ociosidade dos recursos de processamento. Na metodologia proposta, alguns procedimentos são apoiados por ferramentas desenvolvidas para auxiliá-los. O sistema de processamento poderá ser: um computador Multicore, um Cluster real ou virtual alocado na nuvem. Resultados experimentais são apresentados neste trabalho. Tais resultados mostram a viabilidade e eficiência da metodologia proposta. / Applications may have nested perfectly loops that require a high processing power. Various scientific applications contain nested loops in their structures. Such loops can process heterogeneous computations. A solution to reduce the execution time of this class of applications is the parallelization of these loops. The heterogeneity of the execution times of computations present in the iterations of nested perfectly loops demands an adequate parallelization aiming at a homogeneous load distribution among the computational resources to reduce the idleness of such resources. This heterogeneity implies an ideal number of computational resources which, its increase would not impact the performance gain, since the minimum possible time is the execution time of the task that consumes the longest processing time. In this work is proposed a methodology and tools for parallelization of loops perfectly nested with heterogeneous processing in parallel and distributed systems. The implementation of proposed methodology in application improves execution performance and reduce idles of the processing resources. In the methodology proposed, some procedures are supported by tools developed to assist them. The processing system can be: a computer multicore, a cluster real or virtual allocated in cloud. Experimental results are presented in this work. These results show the feasibility and efficiency of the proposed methodology.

Aplicações podem apresentar laços perfeitamente aninhados que demandam um alto poder de processamento. Diversas aplicações científicas contêm laços aninhados em suas estruturas. Tais laços podem processar computações heterogêneas. Uma solução para reduzir o tempo de execução desta classe de aplicações é a paralelização destes laços. A heterogeneidade dos tempos de execução de computações presentes nas iterações de laços perfeitamente aninhados demanda uma paralelização adequada visando uma distribuição de carga homogênea entre os recursos computacionais para reduzir a ociosidade de tais recursos. Esta heterogeneidade implica em um número ideal de recursos computacionais a partir do qual, o seu aumento não impactaria no ganho de desempenho, uma vez que, o tempo mínimo possível é o tempo de execução da tarefa que consome o maior tempo de processamento. Neste trabalho é proposta uma metodologia e ferramentas para paralelização de laços perfeitamente aninhados sem dependência de dados e com processamento heterogêneo em sistemas paralelos e distribuídos. A implementação da metodologia proposta em aplicações melhora o desempenho da execução e reduz a ociosidade dos recursos de processamento. Na metodologia proposta, alguns procedimentos são apoiados por ferramentas desenvolvidas para auxiliá-los. O sistema de processamento poderá ser: um computador Multicore, um Cluster real ou virtual alocado na nuvem. Resultados experimentais são apresentados neste trabalho. Tais resultados mostram a viabilidade e eficiência da metodologia proposta. / Applications may have nested perfectly loops that require a high processing power. Various scientific applications contain nested loops in their structures. Such loops can process heterogeneous computations. A solution to reduce the execution time of this class of applications is the parallelization of these loops. The heterogeneity of the execution times of computations present in the iterations of nested perfectly loops demands an adequate parallelization aiming at a homogeneous load distribution among the computational resources to reduce the idleness of such resources. This heterogeneity implies an ideal number of computational resources which, its increase would not impact the performance gain, since the minimum possible time is the execution time of the task that consumes the longest processing time. In this work is proposed a methodology and tools for parallelization of loops perfectly nested with heterogeneous processing in parallel and distributed systems. The implementation of proposed methodology in application improves execution performance and reduce idles of the processing resources. In the methodology proposed, some procedures are supported by tools developed to assist them. The processing system can be: a computer multicore, a cluster real or virtual allocated in cloud. Experimental results are presented in this work. These results show the feasibility and efficiency of the proposed methodology.

Les applications de flux de données sont des cibles importantes de l’optimisation de programme en raison de leur haute exigence de calcul et la diversité de leurs domaines d’application: communication, systèmes embarqués, multimédia, etc. L’un des problèmes les plus importants et difficiles dans la conception des langages de programmation destinés à ce genre d’applications est comment les ordonnancer à grain fin à fin d’exploiter les ressources disponibles de la machine.Dans cette thèse on propose un "framework" pour l’ordonnancement à grain fin des applications de flux de données et des boucles imbriquées en général. Premièrement on essaye de paralléliser le nombre maximum de boucles en appliquant le pipeline logiciel. Après on merge le prologue et l’épilogue de chaque boucle (phase) parallélisée pour éviter l’augmentation de la taille du code. Ce processus est un pipeline multidimensionnel, quelques occurrences (ou instructions) sont décalées par des iterations de la boucle interne et d’autres occurrences (instructions) par des iterationsde la boucle externe. Les expériences montrent que l’application de cette technique permet l’amélioration des performances, extraction du parallélisme sans augmenter la taille du code, à la fois dans le cas des applications de flux des donnée et des boucles imbriquées en général. / Applications based on streams, ordered sequences of data values, are important targets of program optimization because of their high computational requirements and the diversity of their application domains: communication, embedded systems, multimedia, etc. One of the most important and difficult problems in special purpose stream language design and implementation is how to schedule these applications in a fine-grain way to exploit available machine resources In this thesis we propose a framework for fine-grain scheduling of streaming applications and nested loops in general. First, we try to pipeline steady state phases (inner loops), by finding the repeated kernel pattern, and executing actor occurrences in parallel as much as possible. Then we merge the kernel prolog and epilog of pipelined phases to move them out of the outer loop. Merging the kernel prolog and epilog means that we shift acotor occurrences, or instructions, from one phase iteration to another and from one outer loop iteration to another, a multidimensional shifting. Experimental shows that our framwork can imporove perfomance, prallelism extraction without increasing the code size, in streaming applications and nested loops in general.

Les algorithmes des systèmes temps réels incluent de plus en plus de nids de boucles, qui sont caractérisés par un temps d’exécution important. De ce fait, plusieurs démarches de parallélisme des boucles imbriquées ont été proposées dans l’objectif de réduire leurs temps d’exécution. Ces démarches peuvent être classifiées selon deux niveaux de granularité : le parallélisme au niveau des itérations et le parallélisme au niveau des instructions. Dans le cas du deuxième niveau de granularité, les techniques visent à atteindre un parallélisme total des instructions appartenant à une même itération. Cependant, le parallélisme est contraint par les dépendances des données inter-itérations ce qui implique le décalage des instructions à travers les boucles imbriquées, provocant ainsi une augmentation du code proportionnelle au niveau du parallélisme. Par conséquent, le parallélisme total au niveau des instructions des nids de boucles engendre des implémentations avec des temps d’exécution non-optimaux et des tailles du code importantes. Les travaux de cette thèse s’intéressent à l’amélioration des stratégies de parallélisme des nids de boucles. Une première contribution consiste à proposer une nouvelle technique de parallélisme au niveau des instructions baptisée « retiming multidimensionnel décalé ». Elle vise à ordonnancer les nids de boucles avec une période de cycle minimale, sans atteindre un parallélisme total. Une deuxième contribution consiste à mettre en pratique notre technique dans le contexte de l’implémentation temps réel embarquée des nids de boucles. L’objectif est de respecter la contrainte du temps d’exécution tout en utilisant un code de taille minimale. Dans ce contexte, nous avons proposé une première démarche d’optimisation qui consiste à utiliser notre technique pour déterminer le niveau parallélisme minimal. Par la suite, nous avons décrit une deuxième démarche permettant de combiner les parallélismes au niveau des instructions et au niveau des itérations, en utilisant notre technique et le « loop striping » / The real time implementation algorithms always include nested loops which require important execution times. Thus, several nested loop parallelism techniques have been proposed with the aim of decreasing their execution times. These techniques can be classified in terms of granularity, which are the iteration level parallelism and the instruction level parallelism. In the case of the instruction level parallelism, the techniques aim to achieve a full parallelism. However, the loop carried dependencies implies shifting instructions in both side of nested loops. Consequently, these techniques provide implementations with non-optimal execution times and important code sizes, which represent limiting factors when implemented on embedded real-time systems. In this work, we are interested on enhancing the parallelism strategies of nested loops. The first contribution consists of purposing a novel instruction level parallelism technique, called “delayed multidimensional retiming”. It aims to scheduling the nested loops with the minimal cycle period, without achieving a full parallelism. The second contribution consists of employing the “delayed multidimensional retiming” when providing nested loop implementations on real time embedded systems. The aim is to respect an execution time constraint while using minimal code size. In this context, we proposed a first approach that selects the minimal instruction parallelism level allowing the execution time constraint respect. The second approach employs both instruction level parallelism and iteration level parallelism, by using the “delayed multidimensional retiming” and the “loop striping”