Re: [avr-gcc-list] GCC does useless 16-bit operations

[Dave Hylands]

Hi Daniel,

> But
>
>   uint8_t temp = (tx_producer_index + (uint8_t) 1) % 64;

64 is an int, which is 16 bits, so the whole expression is promoted to int.

Using 4.1.1, the following source file:

#include <stdint.h>

uint8_t func( uint8_t x )
{
   return (uint8_t)( x + (uint8_t)1 ) % (uint8_t)64;
}

compiles to this:

func:
       subi r24,lo8(-(1))
       clr r25
       andi r24,lo8(63)
       andi r25,hi8(63)
       ret

which is about as optimal as it can be since the return value is still
promoted to an int (this was compiled with the -Os flag).

If you really want 8 bit ints, use the -mint8 and then it compiles down to:

func:
       subi r24,lo8(-(1))
       andi r24,lo8(63)
       ret

Aside: It seems that stdint.h is incompatible with the -mint8 flag.

--
Dave Hylands
Vancouver, BC, Canada
http://www.DaveHylands.com/