home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!stanford.edu!snorkelwacker.mit.edu!ai-lab!zurich.ai.mit.edu!feeley
- From: feeley@zurich.ai.mit.edu (Marc Feeley)
- Newsgroups: comp.lang.scheme
- Subject: Re: Are interpreters now as fast as compiled code used to be?
- Date: 14 Dec 92 21:57:01
- Organization: M.I.T. Artificial Intelligence Lab.
- Lines: 45
- Message-ID: <FEELEY.92Dec14215701@zohar.ai.mit.edu>
- References: <4051@mitech.com>
- NNTP-Posting-Host: zohar.ai.mit.edu
- In-reply-to: gjc@mitech.com's message of 14 Dec 92 12:14:23 GMT
-
- In article <4051@mitech.com> gjc@mitech.com (George J. Carrette) writes:
-
- Path: ai-lab!snorkelwacker.mit.edu!stanford.edu!agate!spool.mu.edu!uunet!mitech!gjc
- From: gjc@mitech.com (George J. Carrette)
- Newsgroups: comp.lang.scheme
- Date: 14 Dec 92 12:14:23 GMT
- Organization: Mitech Corporation, Concord MA
- Lines: 19
-
- Informal survey, how fast is your Scheme interpreter?
-
- The question is: With the latest wave of stock hardware,
- have simple interpreters gotten as fast as
- compilers were only a couple years ago?
-
- (define (standard-fib x)
- (if (< x 2)
- x
- (+ (standard-fib (- x 1))
- (standard-fib (- x 2)))))
-
- (standard-fib 20) => 6765 ; in how many seconds?
-
- In SIOD 2.9 (this is a s-exp interpreter using lists for environment rep)
- we get a result in 0.42 seconds on the low-end VMS Alpha machine.
- (1.3 seconds on Solbourne 5E/900)
-
- -gjc
-
- Well here are the numbers for Gambit running on a lowly 68020 based SUN3
- (which has been around for almost a decade):
-
- 3.100 secs for interpreted code
- .600 secs for compiled code with no declarations
- .040 secs for compiled code with fixnum declarations
-
- There is almost a factor of 100 in performance between interpreted
- code and compiled code with declarations. If you assume the
- hardware's performance increases by a factor of 3 every year (which is
- probably an underestimate right now), it shouldn't take more than 4 or
- 5 years for the switchover to happen.
-
- By the way, does SIOD implement the whole numeric tower?
-
- Marc
-