Integrated compilation and scalability analysis for parallel systems

Despite the performance potential of parallel systems, several factors have hindered their widespread adoption. Of these, performance variability is among the most significant. Data parallel languages, which facilitate the programming of those systems, increase the semantic distance between the program´s source code and its observable performance, thus aggravating the optimization problem. In this paper, we present a new methodology to automatically predict the performance scalability of data parallel applications on multicomputers. Our technique represents the execution time of a program as a symbolic expression that includes the number of processors (P), problem size (N), and other system-dependent parameters. This methodology is strongly based on information collected at compile time. By extending an existing data parallel compiler (Fortran D95), we derive during compilation, a symbolic cost model that represents the expected cost of each high-level code section and, inductively, of the complete program