benchmarking

[Bob Boyer <boyer@rascal.ics.utexas.edu>]

I liked your remark about proclamation in ACKL:

  it doesn't help much if people don't use it.  

How true.  I'll suspect that not very many users of AKCL understand
this issue.  Perhaps the AKCL compiler should print warnings of
inefficiency when SPEED=3, SAFETY=0, and no proclamation has been
made.

There is a related issue about fixnum arithmetic and array accesses.
Suitably declared AKCL fixnum arithmetic seems to go at native C
speeds, as you can see from disassemble.  Suitably declared ACKL array
accesses and changes seem to go within a factor of 2 of C.  But
improperly declared stuff can go 80 or so times slower.  So, you can
get quite good results, but it takes some patience and a little skill.
Normally I have to run disassemble and look at the C a few times
before I get the declarations right.

Bob