Implement a generic vectorized version of Smith's algorithms for complex division.

This commit is contained in:
Rasmus Munk Larsen
2021-06-30 15:53:06 -07:00
parent 6035da5283
commit 9312a5bf5c
9 changed files with 46 additions and 114 deletions

View File

@@ -174,14 +174,9 @@ EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
{
// TODO optimize it for SSE3 and 4
Packet2cf res = pmul(a, pconj(b));
__m128 s = _mm_mul_ps(b.v,b.v);
return Packet2cf(_mm_div_ps(res.v,_mm_add_ps(s,vec4f_swizzle1(s, 1, 0, 3, 2))));
return pdiv_complex(a, b);
}
//---------- double ----------
struct Packet1cd
{
@@ -299,10 +294,7 @@ EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
{
// TODO optimize it for SSE3 and 4
Packet1cd res = pmul(a,pconj(b));
__m128d s = _mm_mul_pd(b.v,b.v);
return Packet1cd(_mm_div_pd(res.v, _mm_add_pd(s,_mm_shuffle_pd(s, s, 0x1))));
return pdiv_complex(a, b);
}
EIGEN_STRONG_INLINE Packet1cd pcplxflip/* <Packet1cd> */(const Packet1cd& x)