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

Unify Inverse_SSE.h and Inverse_NEON.h into a single generic implementation using PacketMath.

Browse Source
This commit is contained in:
Guoqiang QI
2020-11-17 12:27:01 +00:00
committed by David Tellenbach
parent 8e9cc5b10a
commit 394f564055
8 changed files with 482 additions and 722 deletions
Download Patch File Download Diff File Expand all files Collapse all files

49
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@@ -52,22 +52,59 @@ template<> struct is_arithmetic<__m128d> { enum { value = true }; };
template<> struct is_arithmetic<Packet4i> { enum { value = true }; };
template<> struct is_arithmetic<Packet16b> { enum { value = true }; };
#define EIGEN_SSE_SHUFFLE_MASK(p,q,r,s) ((s)<<6|(r)<<4|(q)<<2|(p))
template<int p, int q, int r, int s>
struct shuffle_mask{
enum { mask = (s)<<6|(r)<<4|(q)<<2|(p) };
};
// TODO: change the implementation of all swizzle* ops from macro to template,
#define vec4f_swizzle1(v,p,q,r,s) \
(_mm_castsi128_ps(_mm_shuffle_epi32( _mm_castps_si128(v), EIGEN_SSE_SHUFFLE_MASK(p,q,r,s))))
Packet4f(_mm_castsi128_ps(_mm_shuffle_epi32( _mm_castps_si128(v), (shuffle_mask<p,q,r,s>::mask))))
#define vec4i_swizzle1(v,p,q,r,s) \
(_mm_shuffle_epi32( v, EIGEN_SSE_SHUFFLE_MASK(p,q,r,s)))
Packet4i(_mm_shuffle_epi32( v, (shuffle_mask<p,q,r,s>::mask)))
#define vec2d_swizzle1(v,p,q) \
(_mm_castsi128_pd(_mm_shuffle_epi32( _mm_castpd_si128(v), EIGEN_SSE_SHUFFLE_MASK(2*p,2*p+1,2*q,2*q+1))))
Packet2d(_mm_castsi128_pd(_mm_shuffle_epi32( _mm_castpd_si128(v), (shuffle_mask<2*p,2*p+1,2*q,2*q+1>::mask))))
#define vec4f_swizzle2(a,b,p,q,r,s) \
(_mm_shuffle_ps( (a), (b), EIGEN_SSE_SHUFFLE_MASK(p,q,r,s)))
Packet4f(_mm_shuffle_ps( (a), (b), (shuffle_mask<p,q,r,s>::mask)))
#define vec4i_swizzle2(a,b,p,q,r,s) \
(_mm_castps_si128( (_mm_shuffle_ps( _mm_castsi128_ps(a), _mm_castsi128_ps(b), EIGEN_SSE_SHUFFLE_MASK(p,q,r,s)))))
Packet4i(_mm_castps_si128( (_mm_shuffle_ps( _mm_castsi128_ps(a), _mm_castsi128_ps(b), (shuffle_mask<p,q,r,s>::mask)))))
EIGEN_STRONG_INLINE Packet4f vec4f_movelh(const Packet4f& a, const Packet4f& b)
{
return Packet4f(_mm_movelh_ps(a,b));
}
EIGEN_STRONG_INLINE Packet4f vec4f_movehl(const Packet4f& a, const Packet4f& b)
{
return Packet4f(_mm_movehl_ps(a,b));
}
EIGEN_STRONG_INLINE Packet4f vec4f_unpacklo(const Packet4f& a, const Packet4f& b)
{
return Packet4f(_mm_unpacklo_ps(a,b));
}
EIGEN_STRONG_INLINE Packet4f vec4f_unpackhi(const Packet4f& a, const Packet4f& b)
{
return Packet4f(_mm_unpackhi_ps(a,b));
}
#define vec4f_duplane(a,p) \
vec4f_swizzle2(a,a,p,p,p,p)
#define vec2d_swizzle2(a,b,mask) \
Packet2d(_mm_shuffle_pd(a,b,mask))
EIGEN_STRONG_INLINE Packet2d vec2d_unpacklo(const Packet2d& a, const Packet2d& b)
{
return Packet2d(_mm_unpacklo_pd(a,b));
}
EIGEN_STRONG_INLINE Packet2d vec2d_unpackhi(const Packet2d& a, const Packet2d& b)
{
return Packet2d(_mm_unpackhi_pd(a,b));
}
#define vec2d_duplane(a,p) \
vec2d_swizzle2(a,a,(p<<1)|p)
#define _EIGEN_DECLARE_CONST_Packet4f(NAME,X) \
const Packet4f p4f_##NAME = pset1<Packet4f>(X)