benchmarking

[Robert S. Boyer <boyer@CLI.COM>]

Bill,

In response to a remark that Rob MacLachlan posted to comp.lang.lisp,
I sent him the following enclosed message.  He responded with the
message enclosed after that.

I, too, would like to get a new copy of your Gabriel benchmark
program.  When I tried to run my old copy, the one available by
anonymous ftp from cli.com as the file /pub/kcl/gabriel.tar.Z, under
cmulisp on funware, a Sparc 2, I got a stupid IO error, perhaps
because cmulisp or the Sparc does terminal IO a little differently
than the other Lisps or cpus on which I have run the that benchmark
suite, including AKCL, Lucid, and Allegro on Sun-3s.

I hope that your newest version of gabriel.tar.Z also includes your
new CLTL2 compatible make-declare.lisp, which you produced in response
to the Lisp Kit benchmark that was posted to comp.lang.lisp.

Finally, may I suggest that you include an explicit

  (proclaim (quote (optimize (compilation-speed 0) (safety 0) (speed 3) (debug 0))))

in the makefile so that these tests are always run on compiled code
that is supposedly the fastest for the particular lisp.  At least I
BELIEVE that is how you are supposed to get the fastest speed in all
Common Lisps.  (I hope Jonl will object if that is not exactly the
right declaration to get the compiler that produces the very fastest
code for Lucid.)

It seems to me that it continues to be highly desireable for there to
be a publicly available version of your gabriel.tar.Z that is easy to
run on any Lisp, and I hope that Jonl and Rob will be kind enough to
criticize it to the point that it is unobjectionable to them.

Thanks a lot,

Bob