--- In [email protected], "peternilsson42" <peternilsso...@...> wrote:
>
> I have used 8MB implementations where int was still only
> 16-bit. I dare say there are some embedded environments
> where this is still the case.

Yep.

> You could argue that size_t may have performance issues
> because it is often wider than int. But I've yet to encounter
> an implementation where the performance actually warranted
> a smaller/faster type.
> 
> True, on some systems wider types can be fractionally slower,
> but you're not likely to encounter an implementation where
> size_t is particularly slow.

All true, but sometimes when you've got real-time constraints and you've 
optimised everything else, you're left with 2 choices, neither of them 
particularly pleasant:

1. Do stuff like use 8 bit counters instead of 16 bit counters (and rearrange 
the order of fields in structures, don't always make logically static functions 
physically static, avoid post-increment, etc.).
2. Code it in assembler.

Having said that, in the environment I work in (gcc-based front end with 
proprietary CPU), sometimes 16 bits are faster than 8 bits.

Reply via email to