Re: History and evolution of compilers

William Clodius <wclodius@lanl.gov> wrote:
>Wexelblat, R.L. ed, "Proceedings: ACM SIGPLAN History of Programming
>Languages Conference" (Los Angeles, July 1978). 758 pp. Academic Press
>1981. ACM No548780. ACM Order Dept: (800)342-6626. QA76.7 .H56
>
>I believe is out of print...


Note that preprints from that conference appeared as the August 1978
issue of ACM SIGPLAN Notices, so any library which gets the ACM
journals will have at least a first approximation to this.


>There were several implementations of what would currently be
>recognized as higher level languages available at the time FORTRAN I
>appeared. Most of them, however, would be considered closer to
>interpreters than compilers and none of them had the degree of
>sophistication in optimization as the IBM FORTRAN I/II compilers...


Backus and his team sweated quite hard on optimization in the FORTRAN
I compiler. At the time there was intense skepticism about the
usefulness of high-level languages, mostly as the result of
inefficient interpretive implementations (tolerable on machines with
no floating-point hardware, since software floating-point arithmetic
dominated performance there anyway, but not impressive on the newer
hardware). Backus's people thought that FORTRAN might not sell if
using it incurred any significant efficiency penalty, and so their
objective was to generate code that was *better* than the average
hand-coded assembler. And astonishingly enough, in one of the first
real compilers, they often succeeded.


>...Surprizingly, optimization for other Fortran compilers,
>including IBM's Fortran IV compiler, was relatively minimal untill the
>mid to late 60's.


Backus says that the IBM end of this was largely the result of a misguided
attempt to produce a single compiler combining production-grade code with
better diagnostics and debugging facilities (a notorious weak area of the
FORTRAN I/II/III compilers) and faster compile times. He tried, but
failed, to convince the FORTRAN IV developers that separate checkout and
optimizing compilers were needed.
--
| Henry Spencer
| henry@zoo.toronto.edu
[I gather that as soon as they added an option to the Fortran compiler
that let you ask for faster compilation and slower object code,
everyone used it all the time. There's also a legend that the
FREQUENCY statement, which let you give optimization hints about which
branches of an IF would be taken most often was implemented backwards
and nobody noticed. People claimed they needed wonderful
optimization, but for the most part they sure didn't act like
it. Fortran IV did get two compilers on the 360 series, Fortran G which was
fast and generated rotten code, and Fortran H which produced very good
code. -John]






--

Return to the comp.compilers page.
Search the comp.compilers archives again.