>>>>> "Dan" == Dan Sugalski <[EMAIL PROTECTED]>
>>>>> wrote the following on Wed, 02 Aug 2000 16:13:50 -0400

  Dan> I'd also like to teach PI (or whatever it becomes) to produce joined
  Dan> opcodes. Ilya did some profiling of perl 5 code, and there were a
  Dan> bunch of common op pairs. Joining them together (thus saving both
  Dan> dispatch time *and* op setup/teardown) would get us a small win, but
  Dan> then saving 5 cycles an op on ops that take 40 cycles is nothing to
  Dan> sneeze at...

12.5% improvement is *never* anything to sneeze at :-).

Even if the improvement is only 2 cycles per 40, if the underlying op(s)
account for a reasonable percentage of the overall run-time then the
difference will be felt. Shaving 10 cycles off an op that runs 100 times is
less effective than shaving 2 cycles off of one that runs 1000.

Randy
--
"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
Randy J. Ray            | "NSA doesn't need a key to compromise security in
[EMAIL PROTECTED]        | Windows. Programs like Back Orifice can do it without
415-777-9810 x246       | any keys." --Bruce Schneier on the MS/NSA controversy

Reply via email to