After profiling i've found the source of problem. When I convert 32 bit value to string, i use integer division : a= a/base. (base = 8,10,16).

So, as I now, it's naturaly, because universal division function is very slow. For optimization developers usually use (in such cases) "division by constatnt" alghorithms. For example, if i need alghorithm a=a/8, I just write a>>=3; And that's all!

But when I use 10 base system I need a = a/10 alghorithm.
On PC there is standart agorithm (a = a*(2^32/base), in high 4 bytes we recive a*2/base). But this trick is possible only on 32bit architecture.

QUESTION: Does anybody know algorithm for 32 bit value division by constatnt in the context of 8 bit architecture?
I tryed to study Keil's sprintf version but it's a 'strong meat' for me...