Simulating a simple neural network on branch prediction - DOI: 10.4025/actascitechnol.v25i2.2184

• Autor(a) principal: Ribas, Vinícius Marra
• Data de Publicação: 2008
• Outros Autores: Figueiredo, Maurício Fernandes, Gonçalves, Ronaldo Augusto de Lara
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Acta scientiarum. Technology (Online)
• Texto Completo: http://www.periodicos.uem.br/ojs/index.php/ActaSciTechnol/article/view/2184
• topic: processadores superescalares; previsão de desvios; perceptron; 3.00.00.00-9 Engenharias

This work evaluates the usability of perceptron and its efficiency on branch prediction in superscalar architectures, using two simulators. Firstly, we simulated the use of perceptron on the classification of colored points on the Cartesian plane. The classifier showed that perceptron could be used on branch prediction, because the action of classifying is similar to the action of predicting. Based on this previous analysis, we also simulated the use of perceptron on branch prediction using branch traces automatically generated. The predictor presented satisfactory results. In all cases, we concluded that simulation is a good strategy to measure performance of branch predictor based on perceptron before its implementation in hardware.