Re: [avr-gcc-list] Re: Optimisation of bit set/clear in uint32_t

[Dale Whitfield]

Hi David                                         >@2009.04.22_12:48:52_+0200

<snip>

> However, based on your other post in this thread, you are correct in  
> that volatile access is not what you need - it's atomic access (that  
> applies to 8-bit data as well as 32-bit data).
>

I need to make it very clear to all concerned that I am painfully aware
of the issues and differences between atomicity and volatile. I also
want to move the discussion on because this is about optimisation.

I compile this code: (note no volatile, same source, same compiler
options)

uint32_t        status;
static inline void set_status(uint32_t flag, uint8_t set) 
__attribute__((always_inline));

void set_status(uint32_t flag, uint8_t set) {
        if (set) status |= flag;
        else status &= ~flag;
}


If I strip down main() to this it shows the same result.

void main(void) 
{
        set_status(0x00400000UL, 1);
}

Gave:

0000e6f2 <main>:
    e6f2:       80 91 f3 17     lds     r24, 0x17F3
    e6f6:       90 91 f4 17     lds     r25, 0x17F4
    e6fa:       a0 91 f5 17     lds     r26, 0x17F5
    e6fe:       b0 91 f6 17     lds     r27, 0x17F6
    e702:       a0 64           ori     r26, 0x40       ; 64
    e704:       80 93 f3 17     sts     0x17F3, r24
    e708:       90 93 f4 17     sts     0x17F4, r25
    e70c:       a0 93 f5 17     sts     0x17F5, r26
    e710:       b0 93 f6 17     sts     0x17F6, r27
    e714:       08 95           ret

Dale.