[Mspgcc-users] More poor optimization

[msp]

(By the way, is this getting anywhere?  The mailing list archive doesn't
show anything more recent than May 20...)


Just a few more notes about sloppy code generation.

extern unsigned char P1OUT, P2OUT;
extern unsigned int BUF;

void
foo(void)
{
        P1OUT = (unsigned char)BUF;
        P2OUT = (unsigned char)(BUF>>8);
}

compiles to:
foo:
        mov.b   &REG, &P1OUT
        mov     &REG, r12 
        swpb    r12
        mov.b   r12, &P2OUT
        ret

Two improvements that chould be made:
1) As long as you're copying REG to r12, why not do:
        mov     &REG, r12 
        mov.b   r12, &P1OUT
        swpb    r12
        mov.b   r12, &P2OUT
        ret

2) How about the shorter and faster version:
        mov.b   &REG, &P1OUT
        mov.b   &REG+1, &P2OUT
        ret

mspgcc doesn't appaear to know how to read only the high byte of an integer.
(gcc-3.2.3 on x86 can do it).


Second, gcc sometimes creates unnecessary temporaries.

unsigned baz(unsigned x, unsigned y)
{
        return x + y;
}

Compiles to the desired code:
baz:
        add     r14, r15
        ret

But if I swap x + y to y + x, it gets a lot worse:
baz:
        mov     r15, r12 
        mov     r14, r15 
        add     r12, r15
        ret



Thanks for the great tool and apologies for complaining so much!