In <··········@coils.cims.nyu.edu> ········@coils.cims.nyu.edu (Mark McConnell) writes:
>I have been doing some timing tests to compare various languages, and
>I thought people would like to see the results.
Excellent.
I have found CMUCL to be very efficient on Suns. I wrote a neural net
benchmark in lisp and gcc, and they ran in almost the same time.
(Details were posted earlier). (The new version 17 is much beter than
the previous one.)
I am concerned that any of the tests, especially the CMUCL one, do ANY
garbage collection whatsoever on numeric code like this. My neural
net benchmark did not, and I wonder what is happening.
Note that CMUCL has an advantage of AKCL in not using C, which means
that it can utilize the Sparcks 30 bit fixnum arithmetic with 2 tag
bits.
Also, a good way to make CMUCL fast is to include declaraions that
tell it that a group of functions are to be compiled as a unit, I
forget how its done.
Anthony
--
Anthony Berglas
Rm 312a, Computer Science, Uni of Qld, 4072, Australia.
Uni Ph +61 7 365 4184, Home 391 7727, Fax 365 1999