aboutsummaryrefslogtreecommitdiffhomepage
path: root/Eigen/src/Core/arch/NEON/Complex.h
diff options
context:
space:
mode:
Diffstat (limited to 'Eigen/src/Core/arch/NEON/Complex.h')
-rw-r--r--Eigen/src/Core/arch/NEON/Complex.h92
1 files changed, 48 insertions, 44 deletions
diff --git a/Eigen/src/Core/arch/NEON/Complex.h b/Eigen/src/Core/arch/NEON/Complex.h
index 9678040e7..d1943ba3b 100644
--- a/Eigen/src/Core/arch/NEON/Complex.h
+++ b/Eigen/src/Core/arch/NEON/Complex.h
@@ -22,11 +22,13 @@
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
-#ifndef EIGEN_COMPLEX_ALTIVEC_H
-#define EIGEN_COMPLEX_ALTIVEC_H
+#ifndef EIGEN_COMPLEX_NEON_H
+#define EIGEN_COMPLEX_NEON_H
-static uint32x4_t ei_p4ui_CONJ_XOR = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
-static uint32x2_t ei_p2ui_CONJ_XOR = { 0x00000000, 0x80000000 };
+namespace internal {
+
+static uint32x4_t p4ui_CONJ_XOR = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
+static uint32x2_t p2ui_CONJ_XOR = { 0x00000000, 0x80000000 };
//---------- float ----------
struct Packet2cf
@@ -36,7 +38,7 @@ struct Packet2cf
Packet4f v;
};
-template<> struct ei_packet_traits<std::complex<float> > : ei_default_packet_traits
+template<> struct packet_traits<std::complex<float> > : default_packet_traits
{
typedef Packet2cf type;
enum {
@@ -56,9 +58,9 @@ template<> struct ei_packet_traits<std::complex<float> > : ei_default_packet_tr
};
};
-template<> struct ei_unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2}; };
+template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2}; };
-template<> EIGEN_STRONG_INLINE Packet2cf ei_pset1<Packet2cf>(const std::complex<float>& from)
+template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>& from)
{
float32x2_t r64;
r64 = vld1_f32((float *)&from);
@@ -66,15 +68,15 @@ template<> EIGEN_STRONG_INLINE Packet2cf ei_pset1<Packet2cf>(const std::complex<
return Packet2cf(vcombine_f32(r64, r64));
}
-template<> EIGEN_STRONG_INLINE Packet2cf ei_padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(ei_padd<Packet4f>(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf ei_psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(ei_psub<Packet4f>(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf ei_pnegate(const Packet2cf& a) { return Packet2cf(ei_pnegate<Packet4f>(a.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf ei_pconj(const Packet2cf& a)
+template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(padd<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(psub<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate<Packet4f>(a.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
{
- return Packet2cf(vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(a.v), ei_p4ui_CONJ_XOR)));
+ return Packet2cf(vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(a.v), p4ui_CONJ_XOR)));
}
-template<> EIGEN_STRONG_INLINE Packet2cf ei_pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
Packet4f v1, v2;
float32x2_t a_lo, a_hi;
@@ -88,7 +90,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf ei_pmul<Packet2cf>(const Packet2cf& a,
// Multiply the imag a with b
v2 = vmulq_f32(v2, b.v);
// Conjugate v2
- v2 = vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(v2), ei_p4ui_CONJ_XOR));
+ v2 = vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(v2), p4ui_CONJ_XOR));
// Swap real/imag elements in v2.
a_lo = vrev64_f32(vget_low_f32(v2));
a_hi = vrev64_f32(vget_high_f32(v2));
@@ -97,39 +99,39 @@ template<> EIGEN_STRONG_INLINE Packet2cf ei_pmul<Packet2cf>(const Packet2cf& a,
return Packet2cf(vaddq_f32(v1, v2));
}
-template<> EIGEN_STRONG_INLINE Packet2cf ei_pand <Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+template<> EIGEN_STRONG_INLINE Packet2cf pand <Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
return Packet2cf(vreinterpretq_f32_u32(vorrq_u32(vreinterpretq_u32_f32(a.v),vreinterpretq_u32_f32(b.v))));
}
-template<> EIGEN_STRONG_INLINE Packet2cf ei_por <Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+template<> EIGEN_STRONG_INLINE Packet2cf por <Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
return Packet2cf(vreinterpretq_f32_u32(vorrq_u32(vreinterpretq_u32_f32(a.v),vreinterpretq_u32_f32(b.v))));
}
-template<> EIGEN_STRONG_INLINE Packet2cf ei_pxor <Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+template<> EIGEN_STRONG_INLINE Packet2cf pxor <Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
return Packet2cf(vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(a.v),vreinterpretq_u32_f32(b.v))));
}
-template<> EIGEN_STRONG_INLINE Packet2cf ei_pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
return Packet2cf(vreinterpretq_f32_u32(vbicq_u32(vreinterpretq_u32_f32(a.v),vreinterpretq_u32_f32(b.v))));
}
-template<> EIGEN_STRONG_INLINE Packet2cf ei_pload <std::complex<float> >(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(ei_pload((const float*)from)); }
-template<> EIGEN_STRONG_INLINE Packet2cf ei_ploadu<std::complex<float> >(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ei_ploadu((const float*)from)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pload <std::complex<float> >(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload((const float*)from)); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploadu<std::complex<float> >(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu((const float*)from)); }
-template<> EIGEN_STRONG_INLINE void ei_pstore <std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE ei_pstore((float*)to, from.v); }
-template<> EIGEN_STRONG_INLINE void ei_pstoreu<std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE ei_pstoreu((float*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> * to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
-template<> EIGEN_STRONG_INLINE void ei_prefetch<std::complex<float> >(const std::complex<float> * addr) { __pld((float *)addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> * addr) { __pld((float *)addr); }
-template<> EIGEN_STRONG_INLINE std::complex<float> ei_pfirst<Packet2cf>(const Packet2cf& a)
+template<> EIGEN_STRONG_INLINE std::complex<float> pfirst<Packet2cf>(const Packet2cf& a)
{
std::complex<float> EIGEN_ALIGN16 x[2];
vst1q_f32((float *)x, a.v);
return x[0];
}
-template<> EIGEN_STRONG_INLINE Packet2cf ei_preverse(const Packet2cf& a)
+template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
{
float32x2_t a_lo, a_hi;
Packet4f a_r128;
@@ -141,12 +143,12 @@ template<> EIGEN_STRONG_INLINE Packet2cf ei_preverse(const Packet2cf& a)
return Packet2cf(a_r128);
}
-EIGEN_STRONG_INLINE Packet2cf ei_pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
+EIGEN_STRONG_INLINE Packet2cf pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
{
return Packet2cf(vrev64q_f32(a.v));
}
-template<> EIGEN_STRONG_INLINE std::complex<float> ei_predux<Packet2cf>(const Packet2cf& a)
+template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
{
float32x2_t a1, a2;
std::complex<float> s;
@@ -159,7 +161,7 @@ template<> EIGEN_STRONG_INLINE std::complex<float> ei_predux<Packet2cf>(const Pa
return s;
}
-template<> EIGEN_STRONG_INLINE Packet2cf ei_preduxp<Packet2cf>(const Packet2cf* vecs)
+template<> EIGEN_STRONG_INLINE Packet2cf preduxp<Packet2cf>(const Packet2cf* vecs)
{
Packet4f sum1, sum2, sum;
@@ -171,7 +173,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf ei_preduxp<Packet2cf>(const Packet2cf*
return Packet2cf(sum);
}
-template<> EIGEN_STRONG_INLINE std::complex<float> ei_predux_mul<Packet2cf>(const Packet2cf& a)
+template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
{
float32x2_t a1, a2, v1, v2, prod;
std::complex<float> s;
@@ -187,7 +189,7 @@ template<> EIGEN_STRONG_INLINE std::complex<float> ei_predux_mul<Packet2cf>(cons
// Multiply the imag a with b
v2 = vmul_f32(v2, a2);
// Conjugate v2
- v2 = vreinterpret_f32_u32(veor_u32(vreinterpret_u32_f32(v2), ei_p2ui_CONJ_XOR));
+ v2 = vreinterpret_f32_u32(veor_u32(vreinterpret_u32_f32(v2), p2ui_CONJ_XOR));
// Swap real/imag elements in v2.
v2 = vrev64_f32(v2);
// Add v1, v2
@@ -199,7 +201,7 @@ template<> EIGEN_STRONG_INLINE std::complex<float> ei_predux_mul<Packet2cf>(cons
}
template<int Offset>
-struct ei_palign_impl<Offset,Packet2cf>
+struct palign_impl<Offset,Packet2cf>
{
EIGEN_STRONG_INLINE static void run(Packet2cf& first, const Packet2cf& second)
{
@@ -210,43 +212,43 @@ struct ei_palign_impl<Offset,Packet2cf>
}
};
-template<> struct ei_conj_helper<Packet2cf, Packet2cf, false,true>
+template<> struct conj_helper<Packet2cf, Packet2cf, false,true>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
- { return ei_padd(pmul(x,y),c); }
+ { return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
- return ei_pmul(a, ei_pconj(b));
+ return pmul(a, pconj(b));
}
};
-template<> struct ei_conj_helper<Packet2cf, Packet2cf, true,false>
+template<> struct conj_helper<Packet2cf, Packet2cf, true,false>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
- { return ei_padd(pmul(x,y),c); }
+ { return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
- return ei_pmul(ei_pconj(a), b);
+ return pmul(pconj(a), b);
}
};
-template<> struct ei_conj_helper<Packet2cf, Packet2cf, true,true>
+template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
{
EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
- { return ei_padd(pmul(x,y),c); }
+ { return padd(pmul(x,y),c); }
EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
{
- return ei_pconj(ei_pmul(a, b));
+ return pconj(pmul(a, b));
}
};
-template<> EIGEN_STRONG_INLINE Packet2cf ei_pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
// TODO optimize it for AltiVec
- Packet2cf res = ei_conj_helper<Packet2cf,Packet2cf,false,true>().pmul(a,b);
+ Packet2cf res = conj_helper<Packet2cf,Packet2cf,false,true>().pmul(a,b);
Packet4f s, rev_s;
float32x2_t a_lo, a_hi;
@@ -256,7 +258,9 @@ template<> EIGEN_STRONG_INLINE Packet2cf ei_pdiv<Packet2cf>(const Packet2cf& a,
a_hi = vrev64_f32(vget_high_f32(s));
rev_s = vcombine_f32(a_lo, a_hi);
- return Packet2cf(ei_pdiv(res.v, vaddq_f32(s,rev_s)));
+ return Packet2cf(pdiv(res.v, vaddq_f32(s,rev_s)));
}
-#endif // EIGEN_COMPLEX_ALTIVEC_H
+} // end namespace internal
+
+#endif // EIGEN_COMPLEX_NEON_H
generated by cgit v1.2.3 (git 2.39.1) at 2024-05-27 12:20:09 +0000