A predictive performance model for superscalar processors

Designing and optimizing high performance micropro-cessors is an increasingly difcult task due to the size and complexity of the processor design space, high cost of de-tailed simulation and several constraints that a processor design must satisfy. In this paper, we propose the use of em-pirical non-linear modeling techniques to assist processor architects in making design decisions and resolving com-plex trade-offs. We propose a procedure for building accu-rate non-linear models that consists of the following steps: (i) selection of a small set of representative design points spread across processor design space using latin hypercube sampling, (ii) obtaining performance measures at the se-lected design points using detailed simulation, (iii) building non-linear models for performance using the function ap-proximation capabilities of radial basis function networks, and (iv) validating the models using an independently and randomly generated set of design points. We evaluate our model building procedure by constructing non-linear per-formance models for programs from the SPEC CPU2000 benchmark suite with a microarchitectural design space that consists of 9 key parameters. Our results show that the models, built using a relatively small number of simula-tions, achieve high prediction accuracy (only 2.8 % error in CPI estimates on average) across a large processor design space. Our models can potentially replace detailed simula-tion for common tasks such as the analysis of key microar-chitectural trends or searches for optimal processor design points. 1