bill@IPSA.ARPA.UUCP (03/25/87)
There was recently some questions about the performance of Cambridge Lisp
on the ST. It's interpreted code was said to be 20X slower than lisps
running on the MAC. To which someone else pointed out that compiled lisp
can be 10 to 100 times faster than interpreted. Anyhow I took the time to
run a couple of benchmarks on Campridge Lisp to see how much faster the
compiled code really is.
The following are a couple of the standard lisp benchmarks. All times shown
are in seconds. The GC time is the same for both compiled and interpreted code.
Interpreted Compiled GC Time
----------- -------- -------
TAK 726 4 0
DIV-ITER 2576 10 10
DIV-REC 1662 12 10
BOYER 10250 64 180
Anyhow, this seems to indicate that the compiler brings about a rather
dramatic seed increase. The times above show increases from 160 to 260
times! I suspect this would make it considerably faster than the lisps
available for other micros. Comparing these to the values shown in the
current issue of Byte for a 8MHZ 286 runing Gold Hill Common Lisp shows
Cambridge Lisp to be about 20% faster.
If we look at the value for BOYER in particular, as it is indicative of lisp
functionality, the numbers for some other lisps are: Gold Hill 77,
Sun 3 Common Lisp 52, Symbolics 10, VAX 750 PSL 43+41. All of these, except
the last, had sufficient memory to avoid garbage collection. All in all,
Cambridge Lisp is fast, its only real lost is that it is limited by the
current 1MB Atari machines. (Something which could well be changed very soon.)
/bill