Implemented proper support for naked asm using llvm module level asm. Still not 100% complete, but already 1000 times better that what we had before. Don's BignumX86 implementation from Tango (when turned into a standalone unittest) seems to fully work with no changes, and great performance :) Fixed align N; in asm blocks. Fixed inreg parameter passing on x86 for ref/out params. Removed support for lazy initialization of function local static variables, I have no idea why I ever implemented this, it's not in the D spec, and DMD doesn't support it :P Some of the global variable related changes might cause minor regressions, but they should be easily fixable.

int foo(int op)(int a, int b)
{
    version(X86)
    {
    const OP = (op == '+') ? "add" : "sub";
    asm { naked; }
    mixin("asm{"~OP~" EAX, [ESP+4];}");
    asm { ret 4; }
    }
    else static assert(0, "todo");
}

void main()
{
        int i = foo!('+')(2, 4);
        assert(i == 6);
        i = foo!('-')(2, 4);
        assert(i == 2);
}