B.B. Fraguela, D. Andrade, and R. Doallo (Spain)
Modeling and simulation methods, mathematical modeling, memory hierarchies, compiler transformations
Memory hierarchy behavior plays a critical role in the per formance of current computers. Analyzing and improving it is even more important in the case of embedded systems, as a good exploitation of the cache is essential in order to reduce power and energy consumption while speeding up the real-time response these systems are typically required to provide. Traditional techniques like trace-driven simula tions and hardware counters are slow, provide little insight and are inadequate to guide compiler transformations. In this paper we introduce an analytical model that can an alyze statically complete scientific codes characterized by regular access patterns. An extension to model the behav ior of data-dependent conditionals is presented too. This is a new step towards the goal of analyzing complex codes no matter the type of access patterns they exhibit. The exper iments show that the model is very fast and its estimations are very accurate.
Important Links:
Go Back
