Subj : Re: BC5.02 STL question
To   : borland.public.cpp.borlandcpp
From : Ed Mulroy [TeamB]
Date : Mon May 31 2004 08:29 pm

Of course it also depends upon the processor instruction set and how
its pipelines and cache work.

Assuming the variable is in memory and not in a register
  ++i
  inc [ebx]

  i++
  mov  eax,[ebx]
  inc   eax
  mov  [ebx],eax

It has become common for people to use the wrong operator.  I guess
that is why the architects of the compiler decided to detect that a
post-incremented integer isn't really used for its original value and
optimize it to pre-increment.

With most Pentium class processors that isn't too bad.  However if
you're doing something like I did in a small program not long ago,
that difference is multiplied by 10 million per term and there were
quite a few terms.  In a more normal loop you would never notice the
difference in overhead.

In the situation where 'i' is a class item or '++'  is an overloaded
operator the differences can be great.

..  Ed

> Sebastian Ledesma wrote in message
> news:40bb810f$1@newsgroups.borland.com...
>
> I used to think that modern compiler where more smart
> and: (always assuming integer math)
>
> ++i:
> -load, increment, use and store.
>
> i++
> - load, use, increment and store.
>
> By using CPU register optimizations should be no
> difference (but not really sure about that).

.