Field results with the APL compiler

Abstract

The STSC compiler is now in use at customer sites. We can finally answer the question of whether the compiler is effective at reducing CPU time for real applications, and whether it is cost effective to invest the extra effort to speed up applications with the compiler. We have found from numerous case studies that (1) it takes an APL programmer about 3 days to learn about the compiler and begin to use it effectively; (2) customers have speeded up whole applications by factors of 1.2 to 9; (3) usually about one-tenth of an application needs to be compiled to speed it up; (4) the extra effort to analyze an application and compile parts of it can be as low as 1% of the time required to develop it; (5) we have documented cases where the purchase price of the compiler is cost-justified in two months of use.

Reducing computational complexity with array predicates

“…[Ber97a,Wie86,Jor79] Introduction of array predicates into the underlying array structures of an interpreter would increase these overheads, because every primitive in the interpreter would have to create, propagate, or destroy predicate values, even if there was no chance of the predicate ever being put to effective use. If, for instance, predicates increase the cost of dispatching a primitive by 5%, but they can only be exploited in 1% of the primitives executed, then the net effect of their introduction will be to slow down most applications, because almost all applications have execution time profiles that closely match the number of primitives executed.…”