llvm dev - Mar 2012 - [LLVMdev] rematerialization question

I am a bit confused how the rematerialization works.

It seems currently in our backend we get lots of code where some stack 
offset address is calculated, but this address is then spilled to stack, 
and loaded from stack later.

This does not make sense, it would be better to just recalculate the 
address later, ie rematerialize the original stack offset calculation.

But marking some instruction rematerializable means all operands have to 
be always available?

So I cannot make my add operation which is used for the stack offset 
calculations rematerializable because not all operands of all adds are 
always available?


So how can I make llvm to try to rematerialize only those stack offset 
adds, but not all adds?

And also, when I have the stack update code in beginning and at end of 
the functions, these stack+immediate adds/subs also should not be 
rematerialized as the sp has changed?

Hi Heikki,

Le 15/03/2012 15:31, Heikki Kultala a écrit :
> I am a bit confused how the rematerialization works.
>
> It seems currently in our backend we get lots of code where some stack
> offset address is calculated, but this address is then spilled to stack,
> and loaded from stack later.
>
> This does not make sense, it would be better to just recalculate the
> address later, ie rematerialize the original stack offset calculation.
>
> But marking some instruction rematerializable means all operands have to
> be always available?
Yes, the inline spiller will verify that condition.
> So I cannot make my add operation which is used for the stack offset
> calculations rematerializable because not all operands of all adds are
> always available?
Well, you can mark your add operation as being rematerializable in your 
BE but it will not get remat'd if its operands are not available.
>
> So how can I make llvm to try to rematerialize only those stack offset
> adds, but not all adds?
You can force remat for those adds you want to by patching the spiller 
(see InlineSpiller.cpp). I did it for my target (a dirty patch though) 
but I cannot assure that remat will always be possible :(.

Regards,
Ivan
>
> And also, when I have the stack update code in beginning and at end of
> the functions, these stack+immediate adds/subs also should not be
> rematerialized as the sp has changed?
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev