[avr-libc-dev] Problem with _delay_ms()

[Curtis Maloney]

Now, I've been working so far on the assumption that avr-libc is developed 
solely for use with GCC.  So it's not a problem to use GCC-sisms in the headers.

I today ran into a problem with _delay_ms(), where it brought in the floating 
point code.  This I found odd, because my project had used _delay_ms() elsewhere 
without this effect.

I traced it down to this:
	The existing code only called _delay_ms() once, so gcc inlined it, and the 
optimisers (even -or perhaps especially- at -Os) cleaned up the float to a 
constant int.

	The new code called it twice, with different values.  So the inline-able 
function was not inlined, and thus the constant wasn't cleaned up.

I fixed this by giving the function declaration 
__attribute__((always_inline))... and this had cut some 0x600+ bytes off my 
program image size...

I suppose I should submit this to the bug tracker?

(gcc 4.0.1, binutils 2.16.1, avr-libc 1.2.5)

--
Curtis Maloney