Re: [avr-gcc-list] Inconsisten behaviour of switch statement

[Joerg Wunsch]

As Schwichtenberg, Knut wrote:

> Is compiled to:
>
> .LM2:
>               lds r24,minute
>               clr r25
>               cpi r24,15
>               cpc r25,__zero_reg__
>               breq .L6
> .LM3:
>               cpi r24,16
>               cpc r25,__zero_reg__
>               brge .L10
>               or r24,r25
>               breq .L6
>               rjmp .L2
> .L10:
>               cpi r24,30
>               cpc r25,__zero_reg__
>               breq .L6
>               sbiw r24,45
>               brne .L2
> .L6:

Yes, but this is technically OK.  The volatile variable is read
exactly once, and then the decision is based on the value cached in
register rr24/25.  This is different from the behaviour you've been
describing in the initial posting (which I'd really consider erroneous
but could not reproduce so far).

> BTW: casting of the constant, the variable or both does not convince
> gcc to use a byte compare.

This is the "missed optimization" Eric was talking about.  Anyway,
GCC 4.1.1 gets you:

.global decide
        .type   decide, @function
decide:
/* prologue: frame size=0 */
/* prologue end (size=0) */
        lds r24,minute
        cpi r24,lo8(15)
        breq .L3
        cpi r24,lo8(16)
        brsh .L4
        tst r24
        breq .L3
        ret
.L4:
        cpi r24,lo8(30)
        breq .L3
        cpi r24,lo8(45)
        brne .L5
.L3:
        rcall dosomething
.L5:
        ret

So it properly detects the high byte doesn't need to be taken care of.

--
cheers, J"org               .-.-.   --... ...--   -.. .  DL8DTL

http://www.sax.de/~joerg/                        NIC: JW11-RIPE
Never trust an operating system you don't have sources for. ;-)