Cineralismo em arquitecturas paralelas de uso geral

This dissertation investigates which parallel algorithms for general-purpose concurrent machines are the best candidates likely to enable real-time rates in the synthesis of realistic images (cinerealism). This work builds upon the sorting classification of parallel rendering proposed, initially, by the University of North Carolina at Chapel Hill. Parallel rendering algorithms are distinguished by the data type that is redistributed in the graphics pipeline in order to sort from object-space to image-space: Sort-First class redistributes World-Coordinates primitives, Sort-Middle class redistributes Screen-Coordinates primitives and Sort-Last class redistributes pixels (or samples). Communication and redundancy have been modeled, and a framework to reason about load balance in parallel rendering has been proposed. The former results in a model of specific tradeoffs between Sort-First, Sort-Middle and Sort-Last; the later provides an umbrella under which is discussed and compared load balancing results to date. From the point of view of communication and redundancy, there isn't a clear "best algorithm". The algorithms were developed on two defferent general-purpose parallel systems: the MPP/MP-2 and the MultiCluster-2. The MPP/MP-" is a massively data parallel (SIMD programming paradigm) processing system marketed by MasPar Computer Corporation. The MultiCluster-2 is a distributed-memory Transputer-based multiprocessor (MIMD programming paradigm) marketed by the Parsytec GmbHAvailable from Fundacao para a Ciencia e a Tecnologia, Servico de Informacao e Documentacao, Av. D. Carlos I, 126, 1200 Lisboa / FCT - Fundação para o Ciência e a TecnologiaSIGLEPTPortuga