Re: [patch/RFA] multiarch INSTRUCTION_NULLIFIED

[Randolph Chung <randolph@tausq.org>]

> does anybody know how ia64 works in this area? if you have an
> if-converted series of instruction and we single step through the
> region, what happens?

it's a bit hard to tell what is happening, because the ia64-gdb seems to
be rather broken.... but it seems to have problems stepping through code
like this:

int foo(int a)
{
        if (a > 10) {
                x = 1;
                y = 2;
                z = 3;
        } else {
                x = 11;
                y = 12;
                z = 13;
        }

        return x + y + z;
}

with a bit of optimization, gcc turns this into:

        .loc 1 5 0
        cmp4.ge p6, p7 = 10, r32
        .loc 1 4 0
        addl r8 = @gprel(x#), gp
        addl r16 = @gprel(y#), gp
        .mmi
        addl r15 = @gprel(z#), gp
        ;;
        .loc 1 6 0
        (p7) addl r8 = @gprel(x#), gp
        (p7) addl r14 = 1, r0
        .mii
        .loc 1 10 0
        (p6) addl r14 = 11, r0
        .loc 1 7 0
        (p7) addl r16 = @gprel(y#), gp
        ;;
        .loc 1 8 0
        (p7) addl r15 = @gprel(z#), gp
        .mii
        .loc 1 10 0
        (p6) st4 [r8] = r14
        [...]

gdb gets very confused about this :)

randolph
-- 
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/