Also note that add eax,1 may be faster than inc eax on modern out-of-order architectures since inc does a partial flags update (so it must wait for the previous flag update to complete), see the Intel optimization manual for Core 2 on this.

for-instance..

..given a pointer to buffer and its size the last item is at (pointer + size) – 1

..the previous item in an array is at (index) – 1

Also I dont agree with the conjecture that most loops are from 0 (or 1) to n..

..most loops are simply repeated ‘n’ times .. thus ‘n’ itself can be used as the counter, requiring only 1 register of overhead instead of the 2 registers needed for ‘0 to n’ style loops.

If you are suprised that C compilers can turn your 0 to n style loop into something different, think again. Modern compilers will even turn code using several incrementing/decrementing pointers into code using a single incrementing/decrementing pointer plus an invariant displacement. Same number of registers needed, but only one increment per iteration instead of 2 or more.

A easy way to figure out what is happening is figure out what operations are generally around the dec call.

Such a list makes me wonder just how close to reality they were.

Best Regards…
ngadutrafik 2007