Scalarizing Fortran 90 Array Syntax

Array syntax, existed in many languages, adds expressive power by allowing operations on and assignments to the array sections. When compiling to a uniprocessor machine, the array statement must be converted into a loop that maintains the correct semantics, by a process called scalarization. Scalarization presents a significant technical problem because an array assignment needs to be implemented as if all inputs are fetched before any outputs are stored. Since a loop intermixes loads and stores, compiler often allocates an array to temporarily hold the result. Because these extra temporary arrays can cause performance problems in cache, many techniques have been developed to avoid their use or minimize their size. In this paper, we present two new strategies– loop alignment and loop skewing–to address this problem. Our experiments with loop alignment show that it is extremely effective in improving memory hierarchy performance of Fortran 90 array code on a scalar uniprocessor.