@MastersThesis{SilvaJúnior:2015:PoEfTr,
               author = "Silva J{\'u}nior, Manoel Baptista da",
                title = "Portabilidade com efici{\^e}ncia de trechos da din{\^a}mica do 
                         modelo BRAMS entre arquiteturas multi-core e many-core",
               school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
                 year = "2015",
              address = "S{\~a}o Jos{\'e} dos Campos",
                month = "2015-05-29",
             keywords = "modelo meteorolo\́,,gico, processamento paralelo, GPU, 
                         OpenMP, OpenACC, meteorological models, parallel processing.",
             abstract = "O aumento cont{\'{\i}}nuo da resolu{\c{c}}{\~a}o espacial e 
                         temporal dos modelos meteorol{\'o}gicos demanda cada vez mais 
                         velocidade e capacidade de processamento. Executar esses modelos 
                         em tempo h{\'a}bil requer o uso de supercomputadores com centenas 
                         ou milhares de n{\'o}s. Atualmente estes modelos s{\~a}o 
                         executados em produ{\c{c}}{\~a}o no CPTEC em um supercomputador 
                         com n{\'o}s compostos por CPUs com dezenas de n{\'u}cleos 
                         (multi-core). Gera{\c{c}}{\~o}es mais recentes de 
                         supercomputadores apresentam n{\'o}s com CPUs acopladas a 
                         aceleradores de processamento, tipicamente placas gr{\'a}ficas 
                         (GPGPUs), compostas de centenas de n{\'u}cleos (many-core). 
                         Alterar o c{\'o}digo do modelo de forma a usar com alguma 
                         efici{\^e}ncia n{\'o}s com ou sem placas gr{\'a}ficas 
                         (c{\'o}digo port{\'a}til) {\'e} um desafio. A interface de 
                         programa{\c{c}}{\~a}o OpenMP {\'e} o padr{\~a}o estabelecido 
                         h{\'a} d{\'e}cadas para explorar eficientemente as arquiteturas 
                         multi-core. Uma nova interface de programa{\c{c}}{\~a}o, o 
                         OpenACC, foi recentemente proposta para explorar as arquiteturas 
                         many-core. Ambas interfaces s{\~a}o semelhantes, baseadas em 
                         diretivas de paraleliza{\c{c}}{\~a}o para execu{\c{c}}{\~a}o 
                         concorrente de threads. Este trabalho demonstra que {\'e} 
                         poss{\'{\i}}vel escrever um {\'u}nico c{\'o}digo paralelizado 
                         com as duas interfaces que apresente efici{\^e}ncia 
                         aceit{\'a}vel, de forma a poder ser executado num n{\'o} com 
                         arquitetura multi-core ou ent{\~a}o em um n{\'o} com arquitetura 
                         many-core. O c{\'o}digo escolhido como estudo de caso {\'e} a 
                         advec{\c{c}}{\~a}o de escalares, um trecho da din{\^a}mica do 
                         modelo meteorol{\'o}gico regional BRAMS (Brazilian Regional 
                         Atmospheric Modelling System). ABSTRACT: The continuous growth of 
                         spatial and temporal resolutions in current meteorological models 
                         demands increasing processing power. The prompt execution of these 
                         models requires the use of supercomputers with hundreds or 
                         thousands of nodes. Currently, these models are executed at the 
                         operational environment of CPTEC on a supercomputer composed of 
                         nodes with CPUs with tens of cores (multi-core). Newer 
                         supercomputer generations have nodes with CPUs coupled to 
                         processing accelerators, typically graphics cards (GPGPUs), 
                         containing hundreds of cores (many-core). The rewriting of the 
                         model codes in order to use such nodes efficiently, with or 
                         without graphics cards (portable code), represents a challenge. 
                         The OpenMP programming interface proposed decades ago is a 
                         standard for decades to efficiently exploit multi-core 
                         architectures. A new programming interface, OpenACC, proposed 
                         decades ago is the many-core architectures. These two programming 
                         interfaces are similar, since they are based on parallelization 
                         directives for the concurrent execution of threads. This work 
                         shows the feasibility of writing a single code imbedding both 
                         interfaces and presenting acceptable efficiency. When executed on 
                         nodes with multi-core or many-core architecture. The code chosen 
                         as a case study is the advection of scalars, a part of the 
                         dynamics of the regional meteorological model BRAMS (Brazilian 
                         Regional Atmospheric Modeling System).",
            committee = "Velho, Haroldo Fraga de Campos (presidente) and Stephany, Stephan 
                         (orientador) and Carvalho, Solon Ven{\^a}ncio de and Fazenda, 
                         {\'A}lvaro Luiz and Panetta, Jairo",
           copyholder = "SID/SCD",
         englishtitle = "Portability with efficiency in a part of dynamics of the model 
                         BRAMS between multi-core and many-core architectures",
             language = "pt",
                pages = "90",
                  ibi = "8JMKD3MGP3W34P/3JDAC42",
                  url = "http://urlib.net/ibi/8JMKD3MGP3W34P/3JDAC42",
           targetfile = "publicacao.pdf",
        urlaccessdate = "28 abr. 2024"
}