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

bug #1631: fix compilation with ARM NEON and clang, and cleanup the weird pshiftright_and_cast and pcast_and_shiftleft functions.

Browse Source
This commit is contained in:
Gael Guennebaud
2018-11-27 23:45:00 +01:00
parent a1a5fbbd21
commit b131a4db24
8 changed files with 124 additions and 90 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
Eigen/src/Core/arch/NEON/PacketMath.h
View File

@@ -140,8 +140,19 @@ EIGEN_STRONG_INLINE void vst1q_f32(float* to, float32x4_t from) { ::vst1q
EIGEN_STRONG_INLINE void vst1_f32 (float* to, float32x2_t from) { ::vst1_f32 ((float32_t*)to,from); }
#endif
template<> struct unpacket_traits<Packet4f> { typedef float type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
template<> struct unpacket_traits<Packet4i> { typedef int32_t type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
template<> struct unpacket_traits<Packet4f>
{
typedef float type;
typedef Packet4f half;
typedef Packet4i integer_packet;
enum {size=4, alignment=Aligned16};
};
template<> struct unpacket_traits<Packet4i>
{
typedef int32_t type;
typedef Packet4i half;
enum {size=4, alignment=Aligned16};
};
template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float& from) { return vdupq_n_f32(from); }
template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int32_t& from) { return vdupq_n_s32(from); }
@@ -294,6 +305,9 @@ template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, con
}
template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return vbicq_s32(a,b); }
template<int N> EIGEN_STRONG_INLINE Packet4i pshiftright(Packet4i a) { return vshrq_n_s32(a,N); }
template<int N> EIGEN_STRONG_INLINE Packet4i pshiftleft(Packet4i a) { return vshlq_n_s32(a,N); }
template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_f32(from); }
template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int32_t* from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_s32(from); }
@@ -384,20 +398,10 @@ template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a) {
template<> EIGEN_STRONG_INLINE Packet4f pabs(const Packet4f& a) { return vabsq_f32(a); }
template<> EIGEN_STRONG_INLINE Packet4i pabs(const Packet4i& a) { return vabsq_s32(a); }
template<> EIGEN_STRONG_INLINE Packet4f pshiftright_and_cast(Packet4f a, int n) {
return vcvtq_f32_s32(vshrq_n_s32(vreinterpretq_s32_f32(a),n));
}
template<> EIGEN_STRONG_INLINE Packet4f pfrexp<Packet4f>(const Packet4f& a, Packet4f& exponent) {
return pfrexp_float(a,exponent);
}
template<> EIGEN_STRONG_INLINE Packet4f pcast_and_shiftleft<Packet4f>(Packet4f v, int n)
{
Packet4i vi = vcvtq_s32_f32(v);
return vreinterpretq_f32_s32(vshlq_n_s32(vi, n));
}
template<> EIGEN_STRONG_INLINE Packet4f pldexp<Packet4f>(const Packet4f& a, const Packet4f& exponent) {
return pldexp_float(a,exponent);
}

8
Eigen/src/Core/arch/NEON/TypeCasting.h
View File

@@ -41,6 +41,14 @@ template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet4i, Packet4f>(const Packet4i
return vcvtq_f32_s32(a);
}
template<> EIGEN_STRONG_INLINE Packet4i preinterpret<Packet4i,Packet4f>(const Packet4f& a) {
return vreinterpretq_s32_f32(a);
}
template<> EIGEN_STRONG_INLINE Packet4f preinterpret<Packet4f,Packet4i>(const Packet4i& a) {
return vreinterpretq_f32_s32(a);
}
} // end namespace internal
} // end namespace Eigen