Page 3 of 4

Re: about LISP performance

Posted: Mon Nov 04, 2013 12:02 am
by kjaer
OpenGenera on Linux is the same as Alpha, except with the VLM hacked to run on x86-64 linux instead of Alpha OSF. My smallest Alpha is a DS20e (that still weighs 100+ lbs.), half the PCI cards in it aren't supported in DU4, and I was never able to get OG2 to run on Tru64 v5. Frankly, OG2 in a Linux VM makes anything else really difficult to recommend.

Re: about LISP performance

Posted: Tue Nov 12, 2013 2:01 pm
by surrealdeal
I've read that ( in at least xemacs ), lisp code is compiled to byte-code, and because emacs-lisp is a smaller language than common lisp, they can get away with hacks such as getting rid of the first few ascii bits, and using smaller space per opcode. ( which the documentation claims makes this a much faster way to execute code ( according to jwz ).

It's kind-of how risc can be faster than intel opcodes because the archetecture is designed to do one thing quickly.

Re: about LISP performance

Posted: Thu Nov 21, 2013 9:18 am
by kjaer
geo wrote:mmm how about the Genera under Alpha? i think that was the latest setup from Symbolics right? do you have such setup? :)


n=50000000 with OpenGenera 2 on Alpha DS20e (2x EV67 667 MHz)

Interpreted = 3752
Compiled = 6

Re: about LISP performance

Posted: Thu Nov 21, 2013 6:20 pm
by geo
kjaer wrote:n=50000000 with OpenGenera 2 on Alpha DS20e (2x EV67 667 MHz)

Interpreted = 3752
Compiled = 6


thanks kjaer!! this will be added to my little notebook :)

wow! you waited an hour for the interpreter :shock: :shock:

ok will find some time to put my result here soon ;)

hmm was wondering how a standalone Lisp machine would perform this..

Re: about LISP performance

Posted: Fri Nov 22, 2013 10:40 am
by kjaer
geo wrote:hmm was wondering how a standalone Lisp machine would perform this..


I'm going to pull out my 3620 in the next couple days and start it as a project. I'll let you know eventually.

Re: about LISP performance

Posted: Mon Nov 25, 2013 6:29 pm
by geo
kjaer wrote:I'm going to pull out my 3620 in the next couple days and start it as a project.

:shock: :shock: :shock: you have a great collection up there kjaer! how i wish i'm near there to see it live :(

kjaer wrote:I'll let you know eventually.
thanks in advance!! :)

Re: about LISP performance

Posted: Sun Jan 05, 2014 7:38 pm
by hamei

Re: about LISP performance

Posted: Mon Jan 06, 2014 2:08 am
by robespierre
that guy is a zealot in more ways than one.

geo: your benchmark is a bit limited because it only tests the speed of incrementing, testing, and branching. in real programs memory architecture and function calling are probably more important. back in the medieval period, Dick Gabriel did some work on benchmarking systems and wrote a book about it.
http://www.dreamsongs.com/Files/Timrep.pdf

these days it would be better to have more up-to-date benchmarks, like I suppose, RSA crypto or *ML parsing.

Re: about LISP performance

Posted: Mon Jan 06, 2014 5:35 pm
by hamei
robespierre wrote:that guy is a zealot in more ways than one.

I think the word you are looking for is 'crackpot' But anyone who dislikes OCAML can't be all bad ...

.. more up-to-date benchmarks, like I suppose, NSA de- crypting or email parsing.

fixed that for you :P

Re: about LISP performance

Posted: Mon Jan 06, 2014 7:57 pm
by geo
hamei wrote:for you, geo :P

http://www.podval.org/~sds/tool.html
thanks ham!! will check this out later on our break ;) what's up there? CNY is fast approaching so kinda busy here.. will head home soon yahoo!! no more cold places this CNY ;)

robespierre wrote:geo: your benchmark is a bit limited because it only tests the speed of incrementing
hahaha i know robes but i just use this as a quick comparison for different platforms :) i just got this test from one article i read.. and yes i did read the Timrep.pdf, one of my fave reference on different Lisp implementations..

yeah.. i even thought maybe Lisp could be used on bitcoin mining? not sure if i understood it right.. oh! i already got Kogge's book from Amazon, i cant stop reading hehe but now im kinda busy with BLE stuff.. get back with SGI and LISPM soon, cheers!!

Re: about LISP performance

Posted: Mon Jan 06, 2014 8:00 pm
by geo
hamei wrote:Quote:
.. more up-to-date benchmarks, like I suppose, NSA de- crypting or email parsing.

fixed that for you
:shock: :shock: :lol: :lol: :lol: nice one ham! that's why our internet is crawling :twisted: :twisted:

Re: about LISP performance

Posted: Tue Jan 07, 2014 2:24 am
by robespierre
No, I meant a zealot, he thinks that the campus campaigns to divest from Israeli stocks are an NSDAP-like attempt to drive Jews out of colleges.

geo: I haven't read that book, let me know what you think of it.

Re: about LISP performance

Posted: Tue Jan 07, 2014 2:57 am
by hamei
robespierre wrote:No, I meant a zealot

I'll take your word for it ...
zealot.jpg
zealot.jpg (12.84 KiB) Viewed 697 times

He does dislike OCAML tho, so there must be a glimmer of sanity left in that broken pate ... just your average IT kinda guy, I guess :D

Re: about LISP performance

Posted: Tue Jan 07, 2014 6:54 am
by geo
robespierre wrote:No, I meant a zealot

hmm i thought you were preferring the foot soldier of the Protoss race from the StarCraft game :lol: :lol: :lol:

robespierre wrote:let me know what you think of it.

will do ;)

hamei wrote:I'll take your word for it ...

hahaha where did the "soviet union paradise and american cops shooting" went? hehehe

Re: about LISP performance

Posted: Thu Feb 27, 2014 10:46 pm
by kjaer
No new benchmark results (though I thought about installing Macintosh Common Lisp on the IIfx and comparing that to the MacIvory II. maybe in the spring.) but I have been jerking around with bringing up color on the NuVista+. There is some stub support for it in Genera 8.3, but it's not complete enough to actually work. So I hacked a patch together, borrowing the sync programs from the NuVista source. I think the timings are maybe not quite correct for the NuVista+ because I've got no color in any of the video modes, and the "hires" (31kHz) mode needs some pretty serious sync processing before any display will play ball. It'd be nice to find a copy of the Mac driver disks so I could extract the timing values from the "official" NuVista+ sync programs, and see if it makes any difference in the displays output from Genera.

But it does work. I haven't looked into any of the S-Graphics packages yet, though the color demos are (however) mildly amusing.