Re: Just how fast is LISP?

Raul Deluth Miller-Rockwell's message of Thu, 21 Nov 91 01:30:38 says:
> I used to have an MIT-LCS Tech Report talking about a MACLISP compiler
> which would generate the same code for numeric applications as a
> FORTRAN compiler. Of course, you did have to do FORTRAN-like things
> (declare floating point variables, for instance).
>
> While one may argue about the value of this approach, it is certainly
> true that identical object code will run the same speed, no matter
> what the source language.


Perhaps what is being referred to is an old article by
Richard Fateman. Here's the reference:




Richard Fateman, "Reply to an Editorial", ACM SIGSAM Bulletin, vol. 25,
March 1973, pp. 9-11.


This is (obviously) a response to an earlier article and has examples of a
program that was written in FORTRAN, then transcribed into Lisp (the paper
presents no further information on this transcription) and the execution
times compared on a DEC PDP-10. As you might have guessed, the Lisp
version ran faster (1.81 secs vs. 2.22 secs). A recursive Lisp version of
the same program ran in 2.20 seconds. The author then says:


"The point we wish to make is that compiled properly, LISP may be as
efficient a vehicle for conveying algorithms, even numerical ones, as any
other higher-level language, e.g. FORTRAN. An examination of the machine
code produced by the two compilations above shows that the inner-loop
arithmetic compilations are virtually identical, but that LISP subroutine
calls are less expensive."


As I heard it, DEC felt so embarassed by this that they modified the
FORTRAN compiler to generate better code. In any case, this report is 18
years old and the information there may have little relevance to the
current state-of-the-art.


Joe Varghese
Mentor Graphics Corporation
joe_varghese@mentorg.com OR varghese@mentor.com
--

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