tenny@ootool.dec.com (Dave Tenny) (02/04/91)
Some time ago I posted a question or two regarding Allegro Common Lisp performance. My question centered on whether or not ACL had a CPU bottle neck, as well as the usual memory requirements. The only reply I received was a fairly useless and ill-considered reply by -=EPS=- saying, in essense, "you fool, of course it won't run in 8 megs". [To give Mr. -=EPS=- credit however, he *usually* posts useful and constructive comments to this conference, and I appreciate it]. Having spent several years implementing several releases of Vax Lisp internals, I'm well aware of Lisp needs for memory. So, disregarding such useful tidbits, I have done two things. I have spoken to people at Franz, Inc, and I have obtained a memory upgrade (but NOT for the purpose of running lisp). For those who are interested, here are my conclusions about ACL utilization of CPU. From an objective standpoint, ACL on the 68030 compares very favorably with Lisps on other platforms. However, from my subjective standpoint of wanting to use ACL on my cube, I find that it is bottlenecked by CPU speed on the 68030 NeXT, and that this is the major problem with using it for production Lisp code. I'm told that ACL will inline in some instances when you issue inline declarations, but I had trouble making this happen, so don't count on it. Now, what I'd like to know is how ACL 3.0 performs on the 68040 NeXT machines? And what I'm especially curious about is if anybody has used the ACL 4.0 version with native CLOS support on the 68040? Any useful tidbits on this are appreciated. I have my board upgrade on order, so I'll be able to answer the first question eventually. But I'm reluctant to spend $375 for the 4.0 ACL upgrade without some idea of what they've managed to do both for performance and robustness of the native CLOS support. Dave