Skip to content

contrib/math: preserve negative zero in Expm1 #3273

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
EylonKrause wants to merge 1 commit into
base: master
Choose a base branch
from
+24 −1
Open

contrib/math: preserve negative zero in Expm1 #3273

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 4 additions & 1 deletion hwy/contrib/math/math-inl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2660,7 +2660,10 @@ HWY_INLINE V Expm1(const D d, V x) {

// Reduce, approximate, and then reconstruct.
const V y = impl.ExpPoly(d, impl.ExpReduce(d, x, q));
const V z = IfThenElse(Lt(Abs(x), kLn2Over2), y,
// Reapply the sign of x on the polynomial path so a negative-zero input is
// preserved: expm1(-0) = -0 (C11 F.10.3.3), which the polynomial drops when
// it forms 0.5 * (+0) + (-0) = +0.
const V z = IfThenElse(Lt(Abs(x), kLn2Over2), Or(y, And(x, kNegZero)),

@jan-wassenberg jan-wassenberg Aug 13, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks again for fixing! Rather than Or+And, please use CopySign(y, x) - this is more efficient on some platforms.

@jan-wassenberg jan-wassenberg Aug 13, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Once again thanks for fixing! Rather than Or+And, please use CopySign(y, x) - this is more efficient on some platforms.

Sub(impl.LoadExpShortRange(d, Add(y, kOne), q), kOne));
return IfThenElse(Lt(x, kLowerBound), kNegOne, z);
}
Expand Down
20 changes: 20 additions & 0 deletions hwy/contrib/math/math_test.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -191,6 +191,25 @@ HWY_NOINLINE void TestAllPow() {
ForFloat3264Types(ForPartialVectors<TestPow>());
}

// expm1(+/-0) must preserve the sign of zero (C11 F.10.3.3). The ULP comparator
// used by the other math tests treats +0 == -0, so check the bit pattern.
struct TestExpm1SignedZero {
template <class T, class D>
HWY_NOINLINE void operator()(T, D d) {
using TU = MakeUnsigned<T>;
const T pos0 = ConvertScalarTo<T>(0.0);
const T neg0 = ConvertScalarTo<T>(-0.0);
const T got_pos = GetLane(CallExpm1(d, Set(d, pos0)));
const T got_neg = GetLane(CallExpm1(d, Set(d, neg0)));
HWY_ASSERT_EQ(BitCastScalar<TU>(pos0), BitCastScalar<TU>(got_pos));
HWY_ASSERT_EQ(BitCastScalar<TU>(neg0), BitCastScalar<TU>(got_neg));
}
};

HWY_NOINLINE void TestAllExpm1SignedZero() {
ForFloat3264Types(ForPartialVectors<TestExpm1SignedZero>());
}

} // namespace
// NOLINTNEXTLINE(google-readability-namespace-comments)
} // namespace HWY_NAMESPACE
Expand All @@ -213,6 +232,7 @@ HWY_EXPORT_AND_TEST_P(HwyMathTest, TestAllCbrt);
HWY_EXPORT_AND_TEST_P(HwyMathTest, TestAllTgamma);
HWY_EXPORT_AND_TEST_P(HwyMathTest, TestAllLogGamma);
HWY_EXPORT_AND_TEST_P(HwyMathTest, TestAllPow);
HWY_EXPORT_AND_TEST_P(HwyMathTest, TestAllExpm1SignedZero);
HWY_AFTER_TEST();
} // namespace
} // namespace hwy
Expand Down
Loading