Re: Folk Theorem: Assemblers are superior to Compilers

The one "proof" that compiled code can be as "good" as assembled code that
I know about is from the Bliss project back in the mid '70s. The Bliss
compiler for the PDP-11 was optimized for size of generated code, not
speed, because the PDP-11 had only a 64KB address space. The authors did
a study in which a dozen or so DEC PDP-11 assembly language experts were
given a couple of programs each to write. The programs took them a day or
so apiece to write. Some people (I don't know if it was the same people
or CMU grad students) wrote the same programs in Bliss given the same
amount of time. There was some variance, of course, but by and large the
Bliss programs were smaller.


The claim was that since the compiler was fairly good, given a fixed
amount of time a Bliss programmer could spend a larger fraction of the
time optimizing the program than the assembly language programmer could,
because it took less time to get the program correct.


I don't know where I saw this paper, but Wulf was probably a coauthor. He
should at least know where it was published. It has been at least 15
years since I read the paper, so it obviously made an impression on me.


-Ralph Johnson
--

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