In today’s world of high tech manufacturing and computer-aided design simulations of models is at theheart of the whole manufacturing process. Trying to represent and study the variables of real worldmodels using simulation computer programs can turn out to be a very expensive and time consumingtask. On the other hand advancements in modern multi-core CPUs and general purpose GPUs promiseremarkable computational power. Properly utilizing this computational power can provide reduced simulation time. To this end modernmodeling environments provide different optimization and parallelization options to take advantage ofthe available computational power. Some of these parallelization approaches are based onautomatically extracting parallelism with the help of a compiler. Another approach is to provide themodel programmers with the necessary language constructs to express any potential parallelism intheir models. This second approach is taken in this thesis work. The OpenModelica modeling and simulation environment for the Modelica language has beenextended with new language constructs for explicitly stating parallelism in algorithms. This slightlyextended algorithmic subset of Modelica is called ParModelica. The new extensions allow modelswritten in ParModelica to be translated to optimized OpenCL code which can take advantage of thecomputational power of available Multi-core CPUs and general purpose GPUs.

Current trends in processor design have migrated toward chip multiprocessors (CMPs). CMPs are designed to exploit both instruction-level parallelism (ILP) within processors and thread-level parallelism (TLP) within and across processors. However, the conventional design of current CMPs is forced to make a choice between high single-thread performance and high peak throughput. This inability to adjust to varying levels of ILP and TLP results in processor inefficiency. To cope with the dilemma of designing CMPs confronted by the processor designers, this dissertation proposed the hyperscalar concept for current multi-core designs. The hyperscalar concept enables the multi-core architectures to dynamically group many scalar in-order cores as a superscalar processor to accelerate a sequential thread. The reconfigure feature of hyperscalar architecture contributes to the high flexibility in adapting different types of applications, providing high single-thread performance when thread level parallelism (TLP) is low and high throughput when TLP is high. Based on the hyperscalar concept, this dissertation first proposed a hyperscalar dual-core architecture. It can play three different roles (a 2-issue statically scheduled superscalar processor, a homogeneous dual-core processor, or a standalone single-core processor). An Instruction-dependency Analyzer (IA) that connects two scalar in-order cores is designed to handle the role switching. The design of IA makes it possible for the two cores to work together like a 2-issue statically scheduled superscalar processor. The IA dispatches instructions with data dependencies to the same core so that the data dependencies can be resolved by existing forwarding paths in the core. Simulation results show that when the proposed architecture works in a statically scheduled superscalar manner, it achieves a 30.3% higher instructions per cycle (IPC) than the traditional five-stage pipelined core based on 35 benchmarks from the MiBench suite. The increases in area and power for extending a homogeneous dual-core processor to a hyperscalar dual-core processor are only 1.8% and 1.75%, respectively, using 90nm CMOS technology. On top of that, this dissertation further extended the hyperscalar dual-core architecture to hyperscalar multi-core architecture capable of flexibly providing high throughput for uniform parallel application as well as high performance for more general workloads. It can dynamically unite many scalar cores as a larger OOO superscalar processor to accelerate a thread. To accomplish this, the Virtual Shared Register File (VSRF) concept was proposed to help the instructions of a thread in different cores can logically face a uniform set of register file. Simulation results show that the 2, 4, 8, 16, and 32-core-united configurations of the hyperscalar multi-core architecture archive 95%, 84%, 82%, 85%, and 90% of the performance of the monolithic 2, 4,8, 16, and 32-issue OOO superscalar processors based the SPEC2000 benchmarks. Finally, this dissertation proposed a new technology, called multi-streaming SIMD, applicable for hyperscalar architecture to efficiently exploit data-level parallelism (DLP). The multi-streaming SIMD technology enables current multimedia extensions to simultaneously manipulate multiple data streams. Simulation results show that when a multi-streaming SIMD computing engine has four 4-register multimedia operation storage units, it provides a factor of 3.3x to 5.5x performance enhancement for traditional MMX extensions on twelve multimedia kernels. After exploring the above research topics discussed in this dissertation, a promising architecture for future multi-core designs was realized.

Se ha determinado, en una arquitectura multi-Core SMP, el lugar donde incorporar el PP o el HPP sin alterar el ISA del resto de los core. Se ha obtenido una familia de procesadores que ejecutan los algoritmos de Petri para dar solución a sistemas reactivos y concurrentes, con una sólida verificación formal que permite la programación directa de los procesadores. Para esto, se ha construido el hardware de un PP y un HPP, con un IP-Core en una FPGA, integrado a un sistema multi-Core SMP, que ejecuta distintos tipo de RdP. Esta familia de procesadores es configurable en distintos aspectos: - Tamaño del procesador (cantidad de plazas y transiciones). - Procesadores con tiempo y procesadores temporales. - Arquitectura heterogénea, que permite distribuir los recursos empleados para instanciar el procesador según se requiera, y obtener un ahorro sustancial. - La posibilidad de configurar el procesador en pos de obtener los requerimientos y minimizar los recursos. Muy valorado en la construcción de sistemas embebidos. En los sistemas con alta necesidad de concurrencia y sincronización, donde se ha evaluado este procesador, las prestaciones han mostrado una importante mejora en el desempeño. El procesador tiene la capacidad de resolver simultáneamente, por conjuntos múltiples disparos, lo que disminuye los tiempos de consulta y decisión, además los programas ejecutados cumplen con los formalismos de las RdP extendidas y sincronizadas, y los resultados de su ejecución son determinísticos. Los tiempos de respuesta para determinar una sincronización son de dos ciclos por consulta (entre la solicitud de un disparo y la respuesta).

Energy consumption by computing systems is rapidly increasing due to the growth of data centers and pervasive computing. In 2006 data center energy usage in the United States reached 61 billion kilowatt-hours (KWh) at an annual cost of 4.5 billion USD [Pl08]. It is projected to reach 100 billion KWh by 2011 at a cost of 7.4 billion USD. The nature of energy usage in these systems provides an opportunity to reduce consumption. Specifically, the power and performance demand of computing systems vary widely in time and across workloads. This has led to the design of dynamically adaptive or power managed systems. At runtime, these systems can be reconfigured to provide optimal performance and power capacity to match workload demand. This causes the system to frequently be over or under provisioned. Similarly, the power demand of the system is difficult to account for. The aggregate power consumption of a system is composed of many heterogeneous systems, each with a unique power consumption characteristic. This research addresses the problem of when to apply dynamic power management in multi-core processors by accounting for and predicting power and performance demand at the core-level. By tracking performance events at the processor core or thread-level, power consumption can be accounted for at each of the major components of the computing system through empirical, power models. This also provides accounting for individual components within a shared resource such as a power plane or top-level cache. This view of the system exposes the fundamental performance and power phase behavior, thus making prediction possible. This dissertation also presents an extensive analysis of complete system power accounting for systems and workloads ranging from servers to desktops and laptops. The analysis leads to the development of a simple, effective prediction scheme for controlling power adaptations. The proposed Periodic Power Phase Predictor (PPPP) identifies patterns of activity in multi-core systems and predicts transitions between activity levels. This predictor is shown to increase performance and reduce power consumption compared to reactive, commercial power management schemes by achieving higher average frequency in active phases and lower average frequency in idle phases. / text

We design the microarchitecture of the Multi-Level Computing Architecture (MLCA), focusing on its Control Processor (CP). The design of the microarchitecture of the CP faces us with both opportunities and challenges that stem from the coarse granularity of the tasks and the large number of inputs and outputs for each task instruction. Thus, we explore changes to standard superscalar microarchitectural techniques. We design the entire CP microarchitecture and implement it on an FPGA using SystemVerilog. We synthesize and evaluate the MLCA system based on a 4-processor shared-memory multiprocessor. The performance of realistic applications shows scalable speedups that are comparable to that of simulation. We believe that our implementation achieves low complexity in terms of FPGA resource usage and operating frequency. In addition, we argue that our design methodology allows the scalability of the CP as the entire system grows.

We design the microarchitecture of the Multi-Level Computing Architecture (MLCA), focusing on its Control Processor (CP). The design of the microarchitecture of the CP faces us with both opportunities and challenges that stem from the coarse granularity of the tasks and the large number of inputs and outputs for each task instruction. Thus, we explore changes to standard superscalar microarchitectural techniques. We design the entire CP microarchitecture and implement it on an FPGA using SystemVerilog. We synthesize and evaluate the MLCA system based on a 4-processor shared-memory multiprocessor. The performance of realistic applications shows scalable speedups that are comparable to that of simulation. We believe that our implementation achieves low complexity in terms of FPGA resource usage and operating frequency. In addition, we argue that our design methodology allows the scalability of the CP as the entire system grows.

No atual contexto de inovações em multi-core, em que as novas tecnologias de integração estão fornecendo um número crescente de transistores por chip, o estudo de técnicas de aumento de vazão de dados é de suma importância para os atuais e futuros processadores multi-core e many-core. Com a contínua demanda por desempenho computacional, as memórias cache vêm sendo largamente adotadas nos diversos tipos de projetos arquiteturais de computadores. Os atuais processadores disponíveis no mercado apontam na direção do uso de memórias cache L2 compartilhadas. No entanto, ainda não está claro quais os ganhos e custos inerentes desses modelos de compartilhamento da memória cache. Assim, nota-se a importância de estudos que abordem os diversos aspectos do compartilhamento de memória cache em processadores com múltiplos núcleos. Portanto, essa dissertação visa avaliar diferentes compartilhamentos de memória cache, modelando e aplicando cargas de trabalho sobre as diferentes organizações, a fim de obter resultados significativos sobre o desempenho e a influência do compartilhamento da memória cache em processadores multi-core. Para isso, foram avaliados diversos compartilhamentos de memória cache, utilizando técnicas tradicionais de aumento de desempenho, como aumento da associatividade, maior tamanho de linha, maior tamanho de memória cache e também aumento no número de níveis de memória cache, investigando a correlação entre essas arquiteturas de memória cache e os diversos tipos de aplicações da carga de trabalho. Os resultados mostram a importância da integração entre os projetos de arquitetura de memória cache e o projeto físico da memória, a fim de obter o melhor equilíbrio entre tempo de acesso à memória cache e redução de faltas de dados. Nota-se nos resultados, dentro do espaço de projeto avaliado, que devido às limitações físicas e de desempenho, as organizações 1Core/L2 e 2Cores/L2, com tamanho total igual a 32 MB (bancos de 2 MB compartilhados), tamanho de linha igual a 128 bytes, representam uma boa escolha de implementação física em sistemas de propósito geral, obtendo um bom desempenho em todas aplicações avaliadas sem grandes sobrecustos de ocupação de área e consumo de energia. Além disso, como conclusão desta dissertação, mostra-se que, para as atuais e futuras tecnologias de integração, as tradicionais técnicas de ganho de desempenho obtidas com modificações na memória cache, como aumento do tamanho das memórias, incremento da associatividade, maiores tamanhos da linha, etc. não devem apresentar ganhos reais de desempenho caso o acréscimo de latência gerado por essas técnicas não seja reduzido, a fim de equilibrar entre a redução na taxa de faltas de dados e o tempo de acesso aos dados. / In the current context of innovations in multi-core processors, where the new integration technologies are providing an increasing number of transistors inside chip, the study of techniques for increasing data throughput has great importance for the current and future multi-core and many-core processors. With the continuous demand for performance, the cache memories have been widely adopted in various types of architectural designs of computers. Nowadays, processors on the market point out for the use of shared L2 cache memory. However, it is not clear the gains and costs of these shared cache memory models. Thus, studies that address different aspects of shared cache memory have great importance in context of multi-core processors. Therefore, this dissertation aims to evaluate different shared cache memory, modeling and applying workloads on different organizations in order to obtain significant results from the performance and the influence of the shared cache memory multi-core processors. Thus, several types of shared cache memory were evaluated using traditional techniques to increase performance, such as increasing the associativity, larger line size, larger cache memory and also the increase on the cache memory hierarchy, investigating the correlation between the cache memory architecture and the workload applications. The results show the importance of integration between cache memory architecture project and memory physical design in order to obtain the best trade-off between cache memory access time and cache misses. According to the results, within evaluations, due to physical limitations and performance, organizations 1Core/L2 and 2Cores/L2 with total cache size equal to 32MB, using banks of 2 MB, line size equal to 128 bytes, represent a good choice for physical implementation in general purpose systems, obtaining a good performance in all evaluated applications without major extra costs of area occupation and power consumption. Furthermore, as a conclusion in this dissertation is shown that, for current and future integration technologies, traditional techniques for performance gain obtained with changes in the cache memory such as, increase of the memory size, increasing the associativity, larger line sizes etc.. should not lead to real performance gains if the additional latency generated by these techniques was not treated, in order to balance between the reduction of cache miss rate and the data access time.

No atual contexto de inovações em multi-core, em que as novas tecnologias de integração estão fornecendo um número crescente de transistores por chip, o estudo de técnicas de aumento de vazão de dados é de suma importância para os atuais e futuros processadores multi-core e many-core. Com a contínua demanda por desempenho computacional, as memórias cache vêm sendo largamente adotadas nos diversos tipos de projetos arquiteturais de computadores. Os atuais processadores disponíveis no mercado apontam na direção do uso de memórias cache L2 compartilhadas. No entanto, ainda não está claro quais os ganhos e custos inerentes desses modelos de compartilhamento da memória cache. Assim, nota-se a importância de estudos que abordem os diversos aspectos do compartilhamento de memória cache em processadores com múltiplos núcleos. Portanto, essa dissertação visa avaliar diferentes compartilhamentos de memória cache, modelando e aplicando cargas de trabalho sobre as diferentes organizações, a fim de obter resultados significativos sobre o desempenho e a influência do compartilhamento da memória cache em processadores multi-core. Para isso, foram avaliados diversos compartilhamentos de memória cache, utilizando técnicas tradicionais de aumento de desempenho, como aumento da associatividade, maior tamanho de linha, maior tamanho de memória cache e também aumento no número de níveis de memória cache, investigando a correlação entre essas arquiteturas de memória cache e os diversos tipos de aplicações da carga de trabalho. Os resultados mostram a importância da integração entre os projetos de arquitetura de memória cache e o projeto físico da memória, a fim de obter o melhor equilíbrio entre tempo de acesso à memória cache e redução de faltas de dados. Nota-se nos resultados, dentro do espaço de projeto avaliado, que devido às limitações físicas e de desempenho, as organizações 1Core/L2 e 2Cores/L2, com tamanho total igual a 32 MB (bancos de 2 MB compartilhados), tamanho de linha igual a 128 bytes, representam uma boa escolha de implementação física em sistemas de propósito geral, obtendo um bom desempenho em todas aplicações avaliadas sem grandes sobrecustos de ocupação de área e consumo de energia. Além disso, como conclusão desta dissertação, mostra-se que, para as atuais e futuras tecnologias de integração, as tradicionais técnicas de ganho de desempenho obtidas com modificações na memória cache, como aumento do tamanho das memórias, incremento da associatividade, maiores tamanhos da linha, etc. não devem apresentar ganhos reais de desempenho caso o acréscimo de latência gerado por essas técnicas não seja reduzido, a fim de equilibrar entre a redução na taxa de faltas de dados e o tempo de acesso aos dados. / In the current context of innovations in multi-core processors, where the new integration technologies are providing an increasing number of transistors inside chip, the study of techniques for increasing data throughput has great importance for the current and future multi-core and many-core processors. With the continuous demand for performance, the cache memories have been widely adopted in various types of architectural designs of computers. Nowadays, processors on the market point out for the use of shared L2 cache memory. However, it is not clear the gains and costs of these shared cache memory models. Thus, studies that address different aspects of shared cache memory have great importance in context of multi-core processors. Therefore, this dissertation aims to evaluate different shared cache memory, modeling and applying workloads on different organizations in order to obtain significant results from the performance and the influence of the shared cache memory multi-core processors. Thus, several types of shared cache memory were evaluated using traditional techniques to increase performance, such as increasing the associativity, larger line size, larger cache memory and also the increase on the cache memory hierarchy, investigating the correlation between the cache memory architecture and the workload applications. The results show the importance of integration between cache memory architecture project and memory physical design in order to obtain the best trade-off between cache memory access time and cache misses. According to the results, within evaluations, due to physical limitations and performance, organizations 1Core/L2 and 2Cores/L2 with total cache size equal to 32MB, using banks of 2 MB, line size equal to 128 bytes, represent a good choice for physical implementation in general purpose systems, obtaining a good performance in all evaluated applications without major extra costs of area occupation and power consumption. Furthermore, as a conclusion in this dissertation is shown that, for current and future integration technologies, traditional techniques for performance gain obtained with changes in the cache memory such as, increase of the memory size, increasing the associativity, larger line sizes etc.. should not lead to real performance gains if the additional latency generated by these techniques was not treated, in order to balance between the reduction of cache miss rate and the data access time.

Nous nous intéressons dans cette thèse aux grandes architectures parallèles hybrides, c'est-à-dire aux architectures parallèles qui sont une combinaison de processeurs généraliste (Intel Xeon par exemple) et de processeurs accélérateur (GPU Nvidia). L'exploitation efficace de ces grappes hybrides pour le calcul haute performance est au cœur de nos travaux. L'hétérogénéité des ressources de calcul au sein des grappes hybrides pose de nombreuses problématiques lorsque l'on souhaite les exploiter efficacement avec de grandes applications scientifiques existantes. Deux principales problématiques ont été traitées. La première concerne le partage des accélérateurs pour les applications MPI et la seconde porte sur la programmation et l'exécution concurrente de code entre CPU et accélérateur. Les architectures hybrides sont très hétérogènes : en fonction des architectures, le ratio entre le nombre d'accélérateurs et le nombre de coeurs CPU est très variable. Ainsi, nous avons tout d'abord proposé une notion de virtualisation d'accélérateur, qui permet de donner l'illusion aux applications qu'elles ont la capacité d'utiliser un nombre d'accélérateurs qui n'est pas lié au nombre d'accélérateurs physiques disponibles dans le matériel. Un modèle d'exécution basé sur un partage des accélérateurs est ainsi mis en place et permet d'exposer aux applications une architecture hybride plus homogène. Nous avons également proposé des extensions aux modèles de programmation basés sur MPI / threads afin de traiter le problème de l'exécution concurrente entre CPU et accélérateurs. Nous avons proposé pour cela un modèle basé sur deux types de threads, les threads CPU et accélérateur, permettant de mettre en place des calculs hybrides exploitant simultanément les CPU et les accélérateurs. Dans ces deux cas, le déploiement et l'exécution du code sur les ressources hybrides est crucial. Nous avons pour cela proposé deux bibliothèques logicielles S_GPU 1 et S_GPU 2 qui ont pour rôle de déployer et d'exécuter les calculs sur le matériel hybride. S_GPU 1 s'occupant de la virtualisation, et S_GPU 2 de l'exploitation concurrente CPU -- accélérateurs. Pour observer le déploiement et l'exécution du code sur des architectures complexes à base de GPU, nous avons intégré des mécanismes de traçage qui permettent d'analyser le déroulement des programmes utilisant nos bibliothèques. La validation de nos propositions a été réalisée sur deux grandes application scientifiques : BigDFT (simulation ab-initio) et SPECFEM3D (simulation d'ondes sismiques). Nous les avons adapté afin qu'elles puissent utiliser S_GPU 1 (pour BigDFT) et S_GPU 2 (pour SPECFEM3D). / We focus on large parallel hybrid architectures based on a combination of general processors (eg Intel Xeon) and accelerators (Nvidia GPU). Using with efficiency these hybrid clusters for high performance computing is central in our work. The heterogeneity of computing resources in hybrid clusters leads to many issues when we want to use large scientific applications on it. Two main issues were addressed in this thesis. The first one concerns the sharing of accelerators for MPI applications and the second one focuses on programming and concurrent execution of application between CPUs and accelerators. Hybrid architectures are very heterogeneous: for each cluster, the ratio between the number of accelerators and the number of CPU cores can be different. Thus, we first propose a concept of accelerator virtualization, which allows applications to view an architecture in which the number of accelerators is not related to the number of physical accelerators. An execution model based on the sharing of accelerators is proposed. We also propose extensions to the programming model based on MPI + threads to address the problem of concurrent execution between CPUs and accelerators. We propose a system based on two types of threads (CPU and accelerator threads) to implement hybrid calculations simultaneously exploiting the CPU and accelerators model. In both cases, the deployment and the execution of code on hybrid resources is critical. Consequently, we propose two software libraries, called S_GPU 1 and S_GPU 2, designed to deploy and perform calculations on the hybrid hardware. S_GPU 1 deals with virtualization and S_GPU 2 allows concurrent operations on CPUs and accelerators. To observe the deployment and the execution of code on complex hybrid architectures, we integrated trace mechanisms for analyzing the progress of the programs using our libraries. The validation of our proposals has been carried out on two large scientific applications: BigDFT (ab-initio simulation) and SPECFEM3D (simulation of seismic waves).

Submitted by Automa??o e Estat?stica (sst@bczm.ufrn.br) on 2017-01-10T12:35:43Z No. of bitstreams: 1 CarlosAvelinoDeBarros_TESE.pdf: 14220676 bytes, checksum: 5528aa72761d295e342d61cf12387c5e (MD5) / Approved for entry into archive by Arlan Eloi Leite Silva (eloihistoriador@yahoo.com.br) on 2017-01-11T12:51:22Z (GMT) No. of bitstreams: 1 CarlosAvelinoDeBarros_TESE.pdf: 14220676 bytes, checksum: 5528aa72761d295e342d61cf12387c5e (MD5) / Made available in DSpace on 2017-01-11T12:51:22Z (GMT). No. of bitstreams: 1 CarlosAvelinoDeBarros_TESE.pdf: 14220676 bytes, checksum: 5528aa72761d295e342d61cf12387c5e (MD5) Previous issue date: 2016-07-13 / O per?odo que durou do surgimento dos microprocessadores at? o in?cio deste s?culo ficou marcado pela expans?o geom?trica da frequ?ncia de opera??o desses dispositivos. Se por um lado isso proporcionou aumentos cont?nuos do desempenho, tamb?m foi respons?vel por um crescimento problem?tico do aquecimento e do consumo de energia. No sentido de atenuar esses problemas, os processadores multi-core passaram a substituir extensivamente os processadores single-core, oferecendo uma alternativa vi?vel para aumentar o desempenho sem o aumento da frequ?ncia. Como uma das formas de se mapear o consumo de energia, apresentamos o desenvolvimento de dois conjuntos de modelos matem?ticos para a representa??o da pot?ncia el?trica dissipada nos processadores. De acordo com considera??es feitas em rela??o a suas parcelas est?tica e din?mica, estabelecemos a pot?ncia total como vari?vel dependente da frequ?ncia de opera??o dos respectivos processadores analisados. Demonstramos, a partir desses modelos matem?ticos, que o consumo relativo de energia dos processadores pode ser associado a medidas de desempenho empregadas em processamento paralelo, como speedup e efici?ncia. Tamb?m utilizamos os modelos para monitorar a influ?ncia de diversos fatores na redu??o do consumo de energia nos processadores multi-core, tais como o percentual da por??o paralela do c?digo, a quantidade de n?cleos a ser empregado de cada vez, a frequ?ncia de trabalho e o pr?prio speedup. Os resultados das an?lises, em simula??es e em hardware, confirmam as previs?es te?ricas e despertam a possibilidade de melhorar o desempenho energ?tico dos processadores multi-core, sobretudo nas condi??es em que os fatores de influ?ncia podem ser flexibilizados. / The period that lasted from the advent of microprocessors until early this century was marked by the geometric expansion of their operating frequency. If on one hand it provided continuous performance increases, on the other hand it was also responsible for a problematic growth in heating and energy consumption. In attempt to mitigate these problems, multi-core processors have been used extensively in replacement for singlecore processors, offering a viable alternative to increase performance without increasing frequency. As a way of mapping energy consumption, we present the development of two sets of mathematical models for the representation of the electrical power dissipated in the processors. According to considerations we made in relation to static and dynamic parts of the power, we established an equation for the total power as a function of the operating frequency of the respective analyzed processors. We demonstrate, based on these mathematical models, that the relative energy consumption of processors can be related to parallel processing performance measures, such as speedup and efficiency. We also use the models to monitor the influence of several factors on the reduction of energy consumption in multi-core processors, such as the percentage of the parallel portion of the code, the number of cores to be used each time, the working frequency and speedup itself. Results of the analyzes confirm the theoretical predictions and alert to the possibility of improving the energy performance of multi-core processors, especially in conditions in which the factors of influence can be made flexible.