🚀 go-pugleaf

jarreau@vuse.vanderbilt.edu (Michael P. Jarreau) writes:

>The speed increase comes from the fact that there is a single machine
>instruction which is something to the effect of "decrement register
>and jump to address xxx IF NOT ZERO" thus costing one machine cycle.

But unless you have a zero-instruction body, the extra few cycles saved
by changing increments to decrements can't be that much of a total percentage
change in CPU time (and if it really were, wouldn't this be an easy
optimization for a good optimizing compiler?).

>Other speed optimizations can be made by avoiding comparisons which 
>contain expressions.  These expressions are re-evaluated at each
>iteration of a loop without regard to whether any of the concerned
>variables have changed.

Again this seems like something that any decent optimizing compiler could
take care of. Moving loop independent expressions outside of a loop has 
been common practice for several years now. Which compiler are you using
and do you have the optimizer enabled?

>I expect that this holds true with other processors, I just don't have
>any recent experience on them.

The above type of micro-optimizations used to be common practice when 
writing effiecent C programs, but with the advent of optimizing compilers
it is usually best for the programmer to spend their time with macro-
optimizations such as better algorithms and leave the low-level type
of operations to the compiler writers.

If you are finding that you MUST micro-optimize your C code to get the
performance you desire, then I would take it up as a problem with your
C compiler vendor.
--
---
Rodney Radford          || Computer Graphics/Imaging
sasrer@unx.sas.com      || SAS Institute, Inc.
(919) 677-8000 x7703    || Cary, NC  27513