GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-12 03:39:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix typecasting for arm32

Browse Source
This commit is contained in:
Antonio Sánchez 2023-11-23 00:47:50 +00:00 committed by Rasmus Munk Larsen
parent b2cb49e280
commit dd8c71e628
1 changed files with 37 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

41
Eigen/src/Core/arch/NEON/TypeCasting.h
View File

@ -1109,15 +1109,32 @@ template <>
struct type_casting_traits<numext::int64_t, float> { struct type_casting_traits<numext::int64_t, float> {
enum { VectorizedCast = 1, SrcCoeffRatio = 2, TgtCoeffRatio = 1 }; enum { VectorizedCast = 1, SrcCoeffRatio = 2, TgtCoeffRatio = 1 };
}; };
template <> template <>
EIGEN_STRONG_INLINE Packet4f pcast<Packet2l, Packet4f>(const Packet2l& a, const Packet2l& b) { EIGEN_STRONG_INLINE Packet4f pcast<Packet2l, Packet4f>(const Packet2l& a, const Packet2l& b) {
#if EIGEN_ARCH_ARM64
return vcombine_f32(vcvt_f32_f64(vcvtq_f64_s64(a)), vcvt_f32_f64(vcvtq_f64_s64(b))); return vcombine_f32(vcvt_f32_f64(vcvtq_f64_s64(a)), vcvt_f32_f64(vcvtq_f64_s64(b)));
} #else
template <> EIGEN_ALIGN_MAX int64_t lvals[4];
EIGEN_STRONG_INLINE Packet2f pcast<Packet2l, Packet2f>(const Packet2l& a) { pstore(lvals, a);
return vcvt_f32_f64(vcvtq_f64_s64(a)); pstore(lvals + 2, b);
EIGEN_ALIGN_MAX float fvals[4] = {static_cast<float>(lvals[0]), static_cast<float>(lvals[1]),
static_cast<float>(lvals[2]), static_cast<float>(lvals[3])};
return pload<Packet4f>(fvals);
#endif
} }
template <>
EIGEN_STRONG_INLINE Packet2f pcast<Packet2l, Packet2f>(const Packet2l& a) {
#if EIGEN_ARCH_ARM64
return vcvt_f32_f64(vcvtq_f64_s64(a));
#else
EIGEN_ALIGN_MAX int64_t lvals[2];
pstore(lvals, a);
EIGEN_ALIGN_MAX float fvals[2] = {static_cast<float>(lvals[0]), static_cast<float>(lvals[1])};
return pload<Packet2f>(fvals);
#endif
}
template <> template <>
struct type_casting_traits<numext::int64_t, numext::int32_t> { struct type_casting_traits<numext::int64_t, numext::int32_t> {
@ -1233,11 +1250,27 @@ struct type_casting_traits<numext::uint64_t, float> {
}; };
template <> template <>
EIGEN_STRONG_INLINE Packet4f pcast<Packet2ul, Packet4f>(const Packet2ul& a, const Packet2ul& b) { EIGEN_STRONG_INLINE Packet4f pcast<Packet2ul, Packet4f>(const Packet2ul& a, const Packet2ul& b) {
#if EIGEN_ARCH_ARM64
return vcombine_f32(vcvt_f32_f64(vcvtq_f64_u64(a)), vcvt_f32_f64(vcvtq_f64_u64(b))); return vcombine_f32(vcvt_f32_f64(vcvtq_f64_u64(a)), vcvt_f32_f64(vcvtq_f64_u64(b)));
#else
EIGEN_ALIGN_MAX uint64_t uvals[4];
pstore(uvals, a);
pstore(uvals + 2, b);
EIGEN_ALIGN_MAX float fvals[4] = {static_cast<float>(uvals[0]), static_cast<float>(uvals[1]),
static_cast<float>(uvals[2]), static_cast<float>(uvals[3])};
return pload<Packet4f>(fvals);
#endif
} }
template <> template <>
EIGEN_STRONG_INLINE Packet2f pcast<Packet2ul, Packet2f>(const Packet2ul& a) { EIGEN_STRONG_INLINE Packet2f pcast<Packet2ul, Packet2f>(const Packet2ul& a) {
#if EIGEN_ARCH_ARM64
return vcvt_f32_f64(vcvtq_f64_u64(a)); return vcvt_f32_f64(vcvtq_f64_u64(a));
#else
EIGEN_ALIGN_MAX uint64_t uvals[2];
pstore(uvals, a);
EIGEN_ALIGN_MAX float fvals[2] = {static_cast<float>(uvals[0]), static_cast<float>(uvals[1])};
return pload<Packet2f>(fvals);
#endif
} }