Phase Detection and Analysis among Multiple Program Inputs

  • Rafael Soares UNICAMP
  • Luis Antonioli UNICAMP
  • Emilio Francesquini UNICAMP / UFABC
  • Rodolfo Azevedo UNICAMP


Phase analysis has been shown to be an efficient technique to decrease the time needed to execute detailed micro-architectural simulations. The SimPoint method selects small but representative portions of code from each execution phase to extrapolate performance and behavior results with high accuracy and fast execution time when compared to a complete execution of the code. However, the current SimPoint technique is limited to a single program and a single input. In this work we detail an analysis technique which is able to determine phase equivalence among multiple inputs for the same program and, consequently, avoid the redundant execution of the phases. We evaluate our proposal using SPECint 2006 with multiple inputs and show that our technique, while maintaining the precision of the original approach, reduces in 32% the number of SimPoints, on average, thus also improving performance proportionally.
Palavras-chave: Measurement, Proposals, Computational modeling, Space exploration, Benchmark testing, Microarchitecture, Hardware, Phase Analysis, Sampled simulation
Como Citar

Selecione um Formato
SOARES, Rafael; ANTONIOLI, Luis; FRANCESQUINI, Emilio; AZEVEDO, Rodolfo. Phase Detection and Analysis among Multiple Program Inputs. In: SIMPÓSIO EM SISTEMAS COMPUTACIONAIS DE ALTO DESEMPENHO (WSCAD), 19. , 2018, São Paulo. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 2018 . p. 155-161.