Path: utzoo!mnetor!uunet!husc6!mailrus!nrl-cmf!ames!pasteur!ucbvax!XEROX.COM!brantly.henr From: brantly.henr@XEROX.COM Newsgroups: comp.sys.atari.st Subject: Re: Lisp/Prolog for the 1040 Message-ID: <880328-084222-3314@Xerox> Date: 28 Mar 88 16:41:00 GMT Sender: daemon@ucbvax.BERKELEY.EDU Organization: The Internet Lines: 71 In a msg dated 23 Mar 88 15:43:06 GMT Sandra Loosemore writes: >It's been a while since I read this review, but I seem to > remember the discrepancy being more like 20X, and the >reason for it is that the person who did the timing was >comparing *compiled* Lisp on the Amiga to *interpreted* >Lisp on the ST. And yes, Lisp interpreters are typically >at least 20 times slower than compiled Lisp, and Cambridge >Lisp does offer a compiler for the ST. I agree with Sandra that *compiled* Lisp runs much faster than when running *interpreted*. Depending on the implementation & the code being compiled, speeds greatly exceeding 20X can sometimes be achieved (compiled vs interpreted). However I must disagree on the content of the article in question. The speed difference I mentioned was indeed close to 2X (Amiga vs ST), and no where in the article did it mention that the benchmarks were compiled vs. interpreted. (had they been compiled vs. interpreted, what would have been the purpose of "Benchmark comparisons"?). Below is the content of the article in question, at least that portion pretaining to Amiga/ST speeds: ------ From AI EXPERT, Feb. 1987 issue, pg 69..... Aesthetic differences aside, the maturity of Amiga Cambridge Lisp is evident in the benchmarks, Lisp on the Amiga runs significantly faster than Lisp on the Atari ST, in spite of the latter's slightly faster clock speed. In fact, some LISP interpreters on a stock IBM PC seem to run faster than Cambridge LISP interpreter on the Atari ST. We ran some of the Gabriel benchmarks using the interpreter on the Atari 520 ST with 1MB of memory. The TAK benchmark took 730 seconds to execute; TAKL 4,014 seconds; DIV2(iterative) 2,582 seconds; DIV2 (recursive) 1,656 seconds; and DERIV 1,652 seconds. On the Amiga 1000 with 512K RAM, the TAK benchmark took 365.16 seconds to execute; STAK 394.18; TAKL 3027.1 seconds; DIV2(iterative) 1046 seconds; DIV2(recursive)766.56 seconds; and DERIV 732.38 seconds. ------- test Atari Amiga Amiga faster by TAK 730 365.16 1.999X TAKL 4,014 3027.1 1.326X DIV2(iterative) 2,582 1046 2.468X DIV2 (recursive) 1,656 766.56 2.160X DERIV 1,652 732.38 2.255X ------- Sandra must agree that if they were indeed comparisons "compiled vs. interpreted", the ratios should have been much higher. I don't think that this is a comparison of the Amiga vs. the Atari ST, but only a comparison of an Amiga implementation vs. an Atari ST implementation of Cambridge Lisp. Metacompco did a better job on the Amiga implementation (speed-wise that is, the Atari ST version has much better graphics support). Dennis..... Xerox R&D Rochester, NY Brantly.Henr@Xerox.Com