Simulating a LAGS processor to consider variable latency on L1 D-Cache

Abstract

Variability is one of the important issues in deep-submicron tecnologies, and the assumption of non-variable, constant la- tencies in the modules of deep-submicron processors can jeopardize their performance. Cache memories have demon- strated their data-dependent latency due to factors like the coupling capacitances or the distance between the port and the required data. In this paper we present, on one hand, a scheme to detect read operation completion on a variable latency cache mem- ory. On the other hand, we present an asynchronous approach to improve processor performance using this feature. Hence, we propose a Locally-Asynchronous Globally-Synchronous (LAGS) superscalar microarchitecture in which read opera- tions on a variable latency L1 data cache are managed through an asynchronous wrapper. In addition, we demonstrate its fea- sibility running SPEC2000 benchmarks on a 64-bit super- scalar processor modeled through an architectural simulator. Simulations show speedups ranging up to 1.44 and averaging 1.22 over a non-variable cache design.

Publication
Proceedings of the 2010 Summer Computer Simulation Conference
J. Manuel Colmenar
J. Manuel Colmenar
Full Professor

My research interests are focused on metaheuristics applied to optimization problems. I have worked on different combinatorial optimization problems applying trajectorial algorithms such us GRASP or VNS. Besides, I am very interested in applications of Grammatical Evolution, specifically in model and prediction domain, as alternative to machine learning approaches.