Re: wanted: opinions on a benchmark.

[MOON at MIT-MC (David A. Moon)]

    Date:  5 MAR 1981 1835-EST
    From: GJC at MIT-MC (George J. Carrette)

    p.s. On the Lispm the timing difference is much greater,
    which is disturbing.

This is not a very good benchmark.  It gets a division by zero error.
On the Lisp machine it mostly tests the speed of RANDOM (e.g. the ALAN-CLIPPER
test spends 79% of its time in the function RANDOM).  Having fixed those
problems, I find that ALAN-CLIPPER takes 190 microseconds on MC and 340
microseconds on the Lisp machine, while CLIPPER-CALL takes 315 microseconds
on MC and 1250 microseconds on the Lisp machine.  The ratio is 1.6 on MC
and 3.8 on the Lisp machine.  As best as I could measure the twice as large
ratio on the Lisp machine is because on MC function calling and array referencing
are similar in speed to arithmetic (maybe twice as slow), while on the Lisp
machine they are much slower because they do a lot more checking.  Function calling
is definitely too slow on the Lisp machine--I wish that weren't so.
The CLIPPER-CALL also looks like it should be doing bignum arithmetic
due to fixnum overflow in all those multiplications it does, but I
couldn't find any evidence that it actually did get into any bignums.