[LLVMdev] Stack alignment problem

[Vladimir Prus]

On Friday 25 March 2005 07:55, Chris Lattner wrote:

> >  for (MachineFunction::iterator BB = Fn.begin(), E = Fn.end(); BB != E;
> > ++BB) for (MachineBasicBlock::iterator I = BB->begin(); I != BB->end(); )
> > if (I->getOpcode() == FrameSetupOpcode ||
> >          I->getOpcode() == FrameDestroyOpcode) {
> >        .........
> >               HasCalls = true;
> >          ........
> >
> > So, stack is aligned only if there is instruction with FrameSetupOpcode
> > or FrameDestroyOpcode. In X86, it's defined as
> >
> >   def ADJCALLSTACKDOWN : I<0, Pseudo, (ops), "#ADJCALLSTACKDOWN">;
> >   def ADJCALLSTACKUP   : I<0, Pseudo, (ops), "#ADJCALLSTACKUP">;
> >
> > And I'm not quite sure why I need to define and insert something similar.
> > Why can't the above code just check for call instructions, as opposed to
> > "FrameSetup" instructions?
>
> Every target (not just x86) uses these.  Basically they are pseudo
> instructions that mark the begin/end of the call sequence.  On targets
> that support frame pointer elimination, they expand different ways (e.g.
> into a noop if fp-eliminating, or into [e.g.] "sub ESP, 40" if not).
>
> I think it's a good idea for your target to do similar things, for
> consistency if nothing else, but if you really want to, I wouldn't be
> opposed to making that also check for instructions that isCall.

I though about it, and decided that modifying the above code is too much work. 
Especially given that the most reasonable idea to me would be to check only 
for call instructions, and that would mean modifying all targets.
So, I've added those pseudoinstructions and it seems to work.

- Volodya