t

Tutorial

THEORETICAL RESULTS IN COMPILER DESIGN

AND

IMPLEMENTATION

Chairperson:

Speakers:

Meera Blattner, Rice University

Michael Harrison, University of California, Berkeley

Susan Graham, University of California, Berkeley

Kenneth Kennedy, Rice University

The aim of this session is to survey some advances in theoretical computer science and their impact

on the design of progrmmning language compilers. There has been considerable research activity in three

major areas: the analysis of program syntax (parsing), the detection of errors at compile time, and the

optimization of compiled code. The tutorials presented here do not purport to sunmmrize all the work

that has been done; instead, they attempt to capture the flavor of current research in these areas and

demonstrate its practical importance.

On the Parsing of General Context Free Language,

Michael Harrison

A variety of methods are known for parsing

arbitrary context free languages and these are

quickly reviewed. All of the "classical methods"

take time proportional to n 3 where n is the

length of the input. It is possible to reduce

language recognition to boolean matrix multipli-

cation and so to find an alborithm which works in

time n 2"81 or, if arcane techniques are used,

time n 2 log n . A new practical algorithm,

which takes time n3/log n and is on-line, is

introduced. (An on-llne algorithm tells if

al...a i is in the language before ever reading

ai+l. ) These new results clarify older tech-

niques and it is possible to show an intimate

and unsuspected relationship between the Cocke-

Kasami-Younger algorithm and Earley's algorithm.

of a program be determined. The standard tech-

niques of control flow analysis are reviewed and

related to "structured programming". In the

literature on programming, an "accepted" set of

control mechanisms for structured programming has

evolved; this set can be easily and clearly de-

fined using a graph gr-,,,ar. However, many authors

feel that the accepted set is too restrictive. For

example, it excludes multiple exits from loops. A

new graph grammar, which generalizes the

"admissible" control structures to include many de-

sirable extensions to the structured set, is intro-

duced and its usefulness for compiler optimization

is discussed. Optimization methods based on this

grammar are shown to be suitably efficient.

Getting High Quality Error Diagnostics from

Compilers, Susan Graham

A variety of known techniques for handling

compile-tlme detectable errors are surveyed.

These techniques, ranging from ad-hoc to highly

theoretical, are discussed in terms of their prac-

ticality, efficiency and completeness. Evidence

is presented that the most successful methods for

diagnosing and repairing errors involve a

judicious blend of ideas from both theoretical

computer science and software engineering.

Compile-Time Analysis of Pro~rsmControl Flow

Structure, Kenneth Kennedy

Many approaches to the optimization of com-

piled code require that the control flow s~ructure

23