Iperf : a Framework for Automatic Construction of Performance Prediction Models

Performance prediction models at the source code level are crucial in optimizing compilers, programming environments, and performance debug-ging tools. For each performance prediction task, minimal-cost models are needed that deliver the required accuracy. Current techniques to derive the desired models are ad hoc. This paper discusses a new framework for automatic construction of cost-eeective performance prediction models for diierent target systems at the program source level. A target system consists of a target optimizing compiler, a target operating system, and a target architecture with a multi-level memory hierarchy. Preliminary results for a small computation kernel on a set of twelve target systems indicate the eeectiveness of the proposed framework. 1 Motivation Performance prediction models 1 are needed in the source code optimization process. The ideal model to support an optimization transformation is accurate and cost-eeective. An accurate model ranks the optimization alternatives correctly according to their expected performance beneets on a target system. Since a more detailed performance model is typically more expensive to compute, the ideal performance model is also minimal, i.e., may ignore target system components that do not signiicantly contribute to distinguish the optimization alternatives. Unfortunately , nding the best performance model for an 1 Abbreviated as performance models or models throughout the paper. optimization task is extremely hard, since it may require an in-depth understanding of all components of the target system, including other optimizations performed by the target compiler and their interactions with the features of an advanced operating system and modern computer architecture. In fact, nding a model may be even impossible, since a model with the required accuracy and cost-eeectiveness may not exist. What is needed are tools to support the dii-cult task of nding an adequate performance model. The need for software support for nding accurate and cost-eeective performance models has become even more apparent in the light of recent trends in architecture and language design. For example, the success of Intel's and Hewlett Packard's new IA-64 architecture (Merced) will depend on the ability of optimizing compilers to take advantage of the par-allelism provided by the machine, putting an even greater burden on the compiler to achieve eecient program execution than for current superscalar ar-chitectures. Another new challenge for optimizing compilers are dynamic compilation strategies for languages such as Java. Dynamic optimizations are performed at program execution time and therefore have more severe compile time constraints than static compilation systems. In this …