Stack, register, message based or hybrid for virtual machine?

Hi all,


Just doing a little VM coding for fun, and trying for the fastest
solution. My stack version seems to be faster, about 25% faster than
my register code. I assume it is because of the register lookup
(pointer to pointer), whereas the stack version knows where to look.


I was thinking of some kind of hybrid. Anybody have any thoughts?
I'll post the complete code if interested, just wanted to brush on the
subject first.


Also, It's odd that the first version of ADD is faster than the second:


sp -= 2;
*sp = sp[0] + sp[1];
++pc; sp++;


sp[-2] = sp[-1] + sp[-2];
++pc; --sp;


I guess that just shows that you can't optimize unless you know what
the machine is actually doing.


Regards,
Mike


long codeStack[] = {
    pushi, 3,
    pushi, 2,
    addi,
    tne, iters, (long)&codeStk[2],
    ret
};


long codeRegister[] = {
    loadi, 2, r0,
    addi, r0, 2, r0,
    tne, t0, iters, (long)&codeReg[3],
    ret
};

Return to the comp.compilers page.
Search the comp.compilers archives again.