mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-03-07 18:27:40 +08:00
Code Issues Packages Projects Releases Wiki Activity

Avoid global variables with static constructors in NEON/Complex.h

Browse Source
This commit is contained in:
Benoit Jacob 2016-06-28 11:12:49 -04:00
parent 1a9f92e781
commit 38fb606052
2 changed files with 15 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
Eigen/src/Core/arch/AltiVec/Complex.h
View File

@ -14,7 +14,7 @@ namespace Eigen {
namespace internal {
static Packet4ui p4ui_CONJ_XOR = vec_mergeh((Packet4ui)p4i_ZERO, (Packet4ui)p4f_ZERO_);//{ 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
static Packet4ui p4ui_CONJ_XOR() = vec_mergeh((Packet4ui)p4i_ZERO, (Packet4ui)p4f_ZERO_);//{ 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
#ifdef _BIG_ENDIAN
static Packet2ul p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2d_ZERO_, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
static Packet2ul p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO, (Packet4ui) p2d_ZERO_, 8);//{ 0x8000000000000000, 0x0000000000000000 };
@ -86,7 +86,7 @@ template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf
template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_add(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_sub(a.v,b.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(a.v)); }
template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a) { return Packet2cf((Packet4f)vec_xor((Packet4ui)a.v, p4ui_CONJ_XOR)); }
template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a) { return Packet2cf((Packet4f)vec_xor((Packet4ui)a.v, p4ui_CONJ_XOR())); }
template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
@ -100,7 +100,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, con
v1 = vec_madd(v1, b.v, p4f_ZERO);
// multiply a_im * b and get the conjugate result
v2 = vec_madd(v2, b.v, p4f_ZERO);
v2 = (Packet4f) vec_xor((Packet4ui)v2, p4ui_CONJ_XOR);
v2 = (Packet4f) vec_xor((Packet4ui)v2, p4ui_CONJ_XOR());
// permute back to a proper order
v2 = vec_perm(v2, v2, p16uc_COMPLEX32_REV);

18
Eigen/src/Core/arch/NEON/Complex.h
View File

@ -14,9 +14,15 @@ namespace Eigen {
namespace internal {
const uint32_t conj_XOR_DATA[] = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
static uint32x4_t p4ui_CONJ_XOR = vld1q_u32( conj_XOR_DATA );
static uint32x2_t p2ui_CONJ_XOR = vld1_u32( conj_XOR_DATA );
inline uint32x4_t p4ui_CONJ_XOR() {
static const uint32_t conj_XOR_DATA[] = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
return vld1q_u32( conj_XOR_DATA );
}
inline uint32x2_t p2ui_CONJ_XOR() {
static const uint32_t conj_XOR_DATA[] = { 0x00000000, 0x80000000 };
return vld1_u32( conj_XOR_DATA );
}
//---------- float ----------
struct Packet2cf
@ -65,7 +71,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Pa
template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
{
Packet4ui b = vreinterpretq_u32_f32(a.v);
return Packet2cf(vreinterpretq_f32_u32(veorq_u32(b, p4ui_CONJ_XOR)));
return Packet2cf(vreinterpretq_f32_u32(veorq_u32(b, p4ui_CONJ_XOR())));
}
template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
@ -81,7 +87,7 @@ template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, con
// Multiply the imag a with b
v2 = vmulq_f32(v2, b.v);
// Conjugate v2
v2 = vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(v2), p4ui_CONJ_XOR));
v2 = vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(v2), p4ui_CONJ_XOR()));
// Swap real/imag elements in v2.
v2 = vrev64q_f32(v2);
// Add and return the result
@ -196,7 +202,7 @@ template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const P
// Multiply the imag a with b
v2 = vmul_f32(v2, a2);
// Conjugate v2
v2 = vreinterpret_f32_u32(veor_u32(vreinterpret_u32_f32(v2), 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