Compile-time Performance Prediction with

A procedure is described to automatically compile symbolic performance predictions in the course of program translation. It is also shown that a lower bound on the execution time can be predicted which outperforms traditional static estimations at a negligible increase in cost. The method is demonstrated by its application to a parallel LU factorization algorithm on a multiprocessor. 1 Introduction Compile-time performance prediction can provide essential feedback to enable program and machine parameter optimization by both the user and the compiler. In this paper we study the possibility to automatically compile a symbolic execution time prediction given a (parallel) program and machine model. The rationale behind this investigation is that a symbolic expression provides maximum information on the complex interplay of various program and machine parameters involved. Once compiled as a side result of parameterized source code compilation, numerical performance studies may be conducte..