devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add numext::fma and missing pmadd implementations.

Browse Source
This commit is contained in:
Antonio Sánchez
2025-03-23 01:05:53 +00:00
parent 754bd24f5e
commit d935916ac6
11 changed files with 319 additions and 105 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
Eigen/src/Core/MathFunctions.h
View File

@@ -945,6 +945,38 @@ struct nearest_integer_impl<Scalar, true> {
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar run_trunc(const Scalar& x) { return x; }
};
// Default implementation.
template <typename Scalar, typename Enable = void>
struct fma_impl {
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar run(const Scalar& a, const Scalar& b, const Scalar& c) {
return a * b + c;
}
};
// ADL version if it exists.
template <typename T>
struct fma_impl<
T,
std::enable_if_t<std::is_same<T, decltype(fma(std::declval<T>(), std::declval<T>(), std::declval<T>()))>::value>> {
static T run(const T& a, const T& b, const T& c) { return fma(a, b, c); }
};
#if defined(EIGEN_GPUCC)
template <>
struct fma_impl<float, void> {
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float run(const float& a, const float& b, const float& c) {
return ::fmaf(a, b, c);
}
};
template <>
struct fma_impl<double, void> {
static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double run(const double& a, const double& b, const double& c) {
return ::fma(a, b, c);
}
};
#endif
} // end namespace internal
/****************************************************************************
@@ -1852,6 +1884,15 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar arithmetic_shift_right(const Scalar
return bit_cast<Scalar, SignedScalar>(bit_cast<SignedScalar, Scalar>(a) >> n);
}
// Use std::fma if available.
using std::fma;
// Otherwise, rely on template implementation.
template <typename Scalar>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar fma(const Scalar& x, const Scalar& y, const Scalar& z) {
return internal::fma_impl<Scalar>::run(x, y, z);
}
} // end namespace numext
namespace internal {