This paper presents program analyses and transformations that discover a general class of auxiliary information for any incremental computation problem. Combining these techniques with previous techniques for caching intermediate results, we obtain a systematic approach that transforms nonincremental programs into efficient incremental programs that use and maintain useful auxiliary information as well as useful intermediate results. The use of auxiliary information allows us to achieve a greater degree of incrementality than otherwise possible. Applications of the approach include strength reduction in optimizing compilers and finite differencing in transformational programming.

An aggregate array computation is a loop that computes accumulated quantities over array elements. Such computations are common in programs that use arrays, and the array elements involved in such computations often overlap, especially across iterations of loops, resulting in signi cant redundancy in the overall computation. This paper presents a method and algorithms that eliminate such overlapping aggregate array redundancies and shows both analytical and experimental performance improvements. The method is based on incrementalization, i.e., updating the values of aggregate array computations from iteration to iteration rather than computing them from scratch in each iteration. This involves maintaining additional information not maintained in the original program. We reduce various analysis problems to solving inequality constraints on loop variables and array subscripts, and we apply results from work on array data dependence analysis. Incrementalizing aggregate array computations produces drastic program speedup compared to previous optimizations. Previous methods for loop optimizations of arrays do not perform incrementalization, and previous techniques for loop incrementalization do not handle arrays.

This paper presents a principled approach for optimizing iterative (or recursive) programs. The approach formulates a loop body as a function f and a change operation \Phi, incrementalizes f with respect to \Phi, and adopts an incrementalized loop body to form a new loop that is more efficient. Three general optimizations are performed as part of the adoption; they systematically handle initializations, termination conditions, and final return values on exits of loops. These optimizations are either omitted, or done in implicit, limited, or ad hoc ways in previous methods. The new approach generalizes classical loop optimization techniques, notably strength reduction, in optimizing compilers, and it unifies and systematizes various optimization strategies in transformational programming. Such principled strength reduction performs drastic program efficiency improvement via incrementalization and appreciably reduces code size via associated optimizations. We give examples where this app...

ion of a function f adds an extra cache parameter to f . Simplification simplifies the definition of f given the added cache parameter. However, as to how the cache parameter should be used in the simplification to provide incrementality, KIDS provides only the observation that distributive laws can often be applied. The Munich CIP project [BMPP89,Par90] has a strategy for finite differencing that captures similar ideas. It first "defines by a suitable embedding a function f 0 ", and then "derives a recursive version of f 0 using generalized unfold/fold strategy", but it provides no special techniques for discovering incrementality. We believe that both works provide only general strategies with no precise procedure to follow and therefore are less automatable than ours. Chapter 4 Caching intermediate results The value of f 0 (x \Phi y) may often be computed faster by using not only the return value of f 0 (x), as discussed in Chapter 3, but also the values of some subcomputation...

We sketch a method for deduction-oriented software and system development The method incorporates formal machine-supported specification and verification as activities in software and systems development. We describe experiences in applying this method. These experiences have been gained by using the LP, the Larch proof assistant, as a tool for a number of small and medium size case studies for the formal development of software and systems. LP is used for the verification of the development steps. These case studies include z quicksort, z the majority vote problem, z code generation by a compiler and its correctness, z an interactive queue and its refinement into a network. The developments range over levels of requirement specifications, designs and abstract implementations. The main issues are questions of a development method and how to make good use of a formal tool like LP in a goal-directed way within the development. We further discuss of the value of advanced specific...

This paper presents program analyses and transformations for strengthening invariants for the purpose of efficient computation. Finding the stronger invariants corresponds to discovering a general class of auxiliary information for any incremental computation problem. Combining the techniques with previous techniques for caching intermediate results, we obtain a systematic approach that transforms non-incremental programs into ecient incremental programs that use and maintain useful auxiliary information as well as useful intermediate results. The use of auxiliary information allows us to achieve a greater degree of incrementality than otherwise possible. Applications of the approach include strength reduction in optimizing compilers and finite differencing in transformational programming.

An aggregate array computation is a loop that computes accumulated quantities over array elements. Such computations are common in programs that use arrays, and the array elements involved in such computations often overlap, especially across iterations of loops, resulting in significant redundancy in the overall computation. This paper presents a method and algorithms that eliminate such overlapping aggregate array redundancies and shows analytical and experimental performance improvements. The method is based on incrementalization, i.e., updating the values of aggregate array computations from iteration to iteration rather than computing them from scratch in each iteration. This involves maintaining additional values not maintained in the original program. We reduce various analysis problems to solving inequality constraints on loop variables and array subscripts, and we apply results from work on array data dependence analysis. For aggregate array computations that have significant redundancy, incrementalization produces drastic speedup compared to previous optimizations; when there is little redundancy, the benefit might be offset by cache effects and other factors. Previous methods for loop optimizations of arrays do not perform incrementalization, and previous techniques for loop incrementalization do not handle arrays. 1

The transformational programming, method of algorithm derivation starts with a formal specification of the result to be achieved (which provides no indication of how the result is to be achieved), plus some informal ideas as to what techniques will be used in the implementation. The formal specification is then transformed into an implementation, by means of correctness-preserving refinement and transformation steps. The informal ideas are used to guide the selection of transformations to apply: since they only guide the selection of valid transformations, the ideas do not themselves have to be formalised. At any stage in the process, sub-specifications can be extracted and transformed separately. The main difference between this approach and the invariant based programming approach (and similar stepwise refinement methods) is that loops can be introduced and manipulated while maintaining program correctness and with no need to derive loop invariants. Another difference is that at every stage in the process we are working with a correct program: there is never any need for a separate “verification” step. These factors help to ensure that the method is capable of scaling up to the dvelopment of large and complex software systems.