[LLVMdev] On semantics of add instruction - nsw,nuw flags

[Andrew Trick]

> On Jul 23, 2014, at 7:18 AM, Rekha R <rekharamapai at nitc.ac.in> wrote:
> 
> Hi Tim,
> If:
> It means that the person who wrote the IR has guaranteed that there's
> no overflow (by some means) so LLVM can assume it during optimisation.
> then
> In both cases the add with nsw can be removed in favour of the one
> without. Order is completely irrelevant for normal LLVM arithmetic
> instructions.
> 
> why can't we retain add nsw and remove add (in the example Cases)? Because there is a guarantee overflow wouldn't occur.

You can’t do this because the add’s have a different set of uses. This is valid IR modulo any typos I make:

%safeidx = add nsw i32 %a, %b
%idx = add i32 %a, %b
%cmp = cmp <some condition that guarantees no overflow>
br i1 %cmp, label %maynotoverflow, label %mayoverflow

maynotoverflow:
  %p = getelementptr inbounds i32* %o, i32 %idx
  load i32* %p

mayoverflow:
  sext i32 %idx to i64
—

That’s why we need the concept of a “poison” value.

-Andy

> 
> Thanks,
> -- 
> Rekha
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev