Sam Steingold <···@usa.net> writes:
>I just upgraded to Linux (from w32, where I had only CLISP to chose
>from) and I am trying to choose the best CL available.
>GCL compilation failed, and Allegro 4.3 doesn't support :key in
>#'reduce, so I was left with CLISP and CMU CL.
>I took a part of my code and compiled it under both CMU CL and CLISP and
>timed the execution. The code basically does lots of number crunching,
>plus a moderate amount of consing while reading initial data (lists of
>length ~ 5,000). The results were quite surprising (to me, and to Bruno,
>who, when telling me to post, said that CMUCL will be the best, ACL the
>second and CLISP the last :-)
The results below show a lot of consing, so I assume you're using
bignums, either intentionally or because you didn't bother to declare
the numeric types.
CLISP is in fact excellent in bignums, while the CMUCL compiler
doesn't really optimize anything about bignums or undeclared number
types in general.
If my assumptions are right and you don't really need bignums, you can
dramatically improve CMUCL's performance by declaring your number
variable to a proper type. The compiler is very helpful in pointing
out possible improvements. Savety vs. speed optimization settings will
also have a big effect.
Also, it may help to put all your related number definitions
(functions and data) into one big source file as long as you're not
shure what definitions are seem where and when while compiling.
If you still have problems, please post the code, including the
statements you use for compiling and loading.
Martin
--
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
···············@wavehh.hanse.de http://www.cons.org/cracauer/
BSD User Group Hamburg/Germany http://www.bsdhh.org/