Portal de Programas de Pós-Graduação (UFRN)

SIGAA - Sistema Integrado de Gestão de Atividades Acadêmicas

PPgSC/UFRN PROGRAMA DE PÓS-GRADUAÇÃO EM SISTEMAS E COMPUTAÇÃO ADMINISTRAÇÃO DO CCET Téléphone/Extension: (84)3342-2225/115 E-mail: ppgsc@ppgsc.ufrn.br https://posgraduacao.ufrn.br/ppgsc

Banca de QUALIFICAÇÃO: HADLEY MAGNO DA COSTA SIQUEIRA

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
DISCENTE : HADLEY MAGNO DA COSTA SIQUEIRA
DATA : 14/12/2018
HORA: 14:00
LOCAL: Auditorio I - DIMAp
TÍTULO:

CGRA implementation targeting PRET machines.

PALAVRAS-CHAVES:

Precision Timed Machines, Coarse Grain Reconfigurable Array, Real Time

PÁGINAS: 70
GRANDE ÁREA: Ciências Exatas e da Terra
ÁREA: Ciência da Computação
SUBÁREA: Sistemas de Computação
ESPECIALIDADE: Hardware
RESUMO:

Precision Timed Machines (PRETs) are processors targeting real-time systems such as cyber-physical ones, for instance. These

processors aim to provide time predictability and repeatability without a severe degradation in performance. To achieve such

goals, state-of-the-art PRET processors use micro architectural designs targeting exploration

of thread-level parallelism. Although such strategies achieve good performance in many cases, in

others it leads to severe processor subutilization. In order to minimize those issues and provide

a better overall performance, this work proposes a Coarse Grain Reconfigurable Array (CGRA): an array

of processing elements that the processor can use to accelerate parts of software. The proposed CGRA

is coupled to a PRET processor and allows exploration of Instruction Level Parallelism (ILP). In order to

couple such CGRA to a PRET processor, the CGRA must maintain intact the timing properties of its processor

and to do this, the proposed CGRA uses an asynchronous communication among processing elements that

allows for single-cycle communication. The communication configuration is statically generated allowing

for tools to reason about time predictability of the proposed architecture. A RTL model of the proposed architecture

was described in SystemC and experiments show gains of performance of 2x and 3x times

MEMBROS DA BANCA:
Presidente - 1694485 - MARCIO EDUARDO KREUTZ
Interno - 1882699 - MONICA MAGALHAES PEREIRA
Externo à Instituição - IVAN SARAIVA SILVA - UFPI

Notícia cadastrada em: 04/12/2018 12:12