biep@klipper.UUCP (J. A. "Biep" Durieux) (10/28/85)
Among the many reactions I received upon my request for Lisp machine comparisons, there was a nice list of testresults by Ralph P. Sobek ( mcvax!inria!lasso!ralph ). Because many people might be interested in those results, I post them for him to net.ai. He ran some tests on a variety of LISP workstations while he was at IJCAI-85, in Los Angeles. Here are the results; Fibonacci comparisons(1): ------------------------------------------------------------------------------- Lisp System/Computer Interpreted Compiled Notes ------------------------------------------------------------------------------- Xerox 1132 (Dorado) 10.9 sec. 0.145 sec. (2) Symbolics 3640 10.5 0.158 Lisp/VM (IBM 4381) 12.8 0.45 " " 0.15 (3) Interlisp 370 (IBM 3083) 1.502 0.189 Xerox 1185/1186 42.03 0.375 (2) Xerox 1108 (Dandelion) 41.451 0.405 Hewlett Packard 68020 (Common L) 18.3 < 1. (4),(5) LMI/Zetalisp-Plus 92.78 0.53 Macintosh/Experlisp -- 0.53 (6) Sun-2 (Franz lisp) 28.51 - (4) Sun-2 (Lucid Common Lisp) 28.00 0.58 Texas Instruments Explorer 1 min. < 1. (4) Micro-Vax 2 (Vax Lisp) VMS 181.53 2.54 ---------------- (1) Definition of Fibonacci used: (lambda (n) (cond ((eq n 1) 1) ((eq n 2) 1) (t (iplus (fib (sub1 n)) (fib (idifference n 2)))))) Depending on the different versions of lisp, plus had to be used for iplus, etc. [I suppose, however he did not mention it, that he ran (fib 20) -- Biep] (2) Other definitions of Fiboacci can exist: a) a two-branch 'cond' with an 'or', or b) a selectq definition. The Xerox compiler generates equivalent code for all 3 definitions. (3) with fixed number declaration. (4) Used stop watch. (5) Has optimizing compiler which 4 levels of speed-up: none 8.0 times over interpreted 1 11.0 2 16.8 3 23.7 (6) only compiled. Has incremental compiler. -- Biep. {seismo|decvax|philabs|garfield|okstate}!mcvax!vu44!biep Nous aurions souvent honte de nos plus belles actions si le monde voyait tous les motifs qui les produisent -- Francois de La Rochefoucauld