[llvm] b38d897 - [ConstantRange] binaryXor(): special-case binary complement case - the result is precise

[Nikita Popov via llvm-commits]

On Tue, Sep 22, 2020 at 8:52 PM Roman Lebedev <lebedev.ri at gmail.com> wrote:

> On Tue, Sep 22, 2020 at 9:47 PM Nikita Popov <nikita.ppv at gmail.com> wrote:
> >
> > On Tue, Sep 22, 2020 at 8:38 PM Roman Lebedev via llvm-commits <
> llvm-commits at lists.llvm.org> wrote:
> >>
> >>
> >> Author: Roman Lebedev
> >> Date: 2020-09-22T21:37:29+03:00
> >> New Revision: b38d897e802664034c7e6e4654328256ed370a61
> >>
> >> URL:
> https://github.com/llvm/llvm-project/commit/b38d897e802664034c7e6e4654328256ed370a61
> >> DIFF:
> https://github.com/llvm/llvm-project/commit/b38d897e802664034c7e6e4654328256ed370a61.diff
> >>
> >> LOG: [ConstantRange] binaryXor(): special-case binary complement case -
> the result is precise
> >>
> >> Use the fact that `~X` is equivalent to `-1 - X`, which gives us
> >> fully-precise answer, and we only need to special-handle the wrapped
> case.
> >>
> >> This fires ~16k times for vanilla llvm test-suite + RawSpeed.
> >>
> >> Added:
> >>
> >>
> >> Modified:
> >>     llvm/include/llvm/IR/ConstantRange.h
> >>     llvm/lib/IR/ConstantRange.cpp
> >>     llvm/unittests/IR/ConstantRangeTest.cpp
> >>
> >> Removed:
> >>
> >>
> >>
> >>
> ################################################################################
> >> diff  --git a/llvm/include/llvm/IR/ConstantRange.h
> b/llvm/include/llvm/IR/ConstantRange.h
> >> index 318532b24e83..494a14a10cdb 100644
> >> --- a/llvm/include/llvm/IR/ConstantRange.h
> >> +++ b/llvm/include/llvm/IR/ConstantRange.h
> >> @@ -409,6 +409,11 @@ class LLVM_NODISCARD ConstantRange {
> >>    /// value in \p Other.
> >>    ConstantRange srem(const ConstantRange &Other) const;
> >>
> >> +  /// Return a new range representing the possible values resulting
> from
> >> +  /// a binary-xor of a value in this range by an all-one value,
> >> +  /// aka bitwise complement operation.
> >> +  ConstantRange binaryNot() const;
> >> +
> >>    /// Return a new range representing the possible values resulting
> >>    /// from a binary-and of a value in this range by a value in \p
> Other.
> >>    ConstantRange binaryAnd(const ConstantRange &Other) const;
> >>
> >> diff  --git a/llvm/lib/IR/ConstantRange.cpp
> b/llvm/lib/IR/ConstantRange.cpp
> >> index 396c39b5b3a8..7b8dd66b993c 100644
> >> --- a/llvm/lib/IR/ConstantRange.cpp
> >> +++ b/llvm/lib/IR/ConstantRange.cpp
> >> @@ -1240,6 +1240,16 @@ ConstantRange ConstantRange::srem(const
> ConstantRange &RHS) const {
> >>    return ConstantRange(std::move(Lower), std::move(Upper));
> >>  }
> >>
> >> +ConstantRange ConstantRange::binaryNot() const {
> >> +  if (isEmptySet())
> >> +    return getEmpty();
> >> +
> >> +  if (isWrappedSet())
> >> +    return getFull();
> >
> >
> > Probably don't need those checks, as you're forwarding to the sub()
> implementation anyway?
> It is needed if we want to give the right answer.
>

Would you mind elaborating on that? It's not immediately obvious to me why
the X ^ -1 to -1 - X transform would not hold for wrapped ranges.

Possibly it's just a testing issue? TestUnsignedUnaryOpExhaustive tests for
the unsigned envelope, so it will report false positives if the result is a
(correct) wrapped range. Unfortunately, I don't think we have a helper for
the general case right now.

Nikita
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200922/ddd12ee6/attachment.html>