The need for optimization standards

Abstract

The premise of this paper is that in writing programs for military systems, programing style is often strongly affected by a compiler's optimization behavior. For example, in the SAM-D project (see [1]), early deliveries of a JOVLAL J3 compiler did not optimize common sub-expressions. SAM-D programmers asserted that the programs written to compensate for this lack of optimization were less readable, understandable, and maintainable than those written later for an optimizing version of the compiler. They asserted, moreover, that in their opinion, the most significant effect of optimization was to permit more readable programs to be written. This assertion has been further studied and verified in [1]. Similar findings have been noted in [2], [3], and [4]. Consequently, the principal point I wish to make is this:

• HOL Standards must address the effect of optimizations on programming style — Unless compiler optimization behavior is standardized across implementations, the impact on coding, training, and maintenance will be similar to that of programming in different dialects of the "same" language, because programmers will adapt their programming style to conform to the optimization behavior of different compiler implementations.

In the remainder of the paper, I will:

• present some examples supporting this conclusion by illustrating the style of optimization standards that should be applied to the DoD Common Language;

• discuss briefly the interaction between language design decisions and optimization standards;

• note some possible effects of optimization standards on compiler procurement procedures; and

• conclude with a brief analysis of the impact of these ideas on Common Language requirements as expressed in the TINMAN [5] document.

10.21-13.1 September 1976

Prepared for presentation at the Workshop on the DoD Common Language, Cornell University, September 30 – October 1, 1976

Prepared under U.S. Army Contract DAAB07-75-C-0373