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

add SSE4 support, start with integer multiplication

Browse Source
This commit is contained in:
Benoit Jacob
2009-11-24 15:12:43 -05:00
parent abdb2a2bd5
commit 684d76eba3
4 changed files with 49 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
Eigen/Core
View File

@@ -67,6 +67,12 @@
#ifdef __SSSE3__
#include <tmmintrin.h>
#endif
#ifdef __SSE4_1__
#include <smmintrin.h>
#endif
#ifdef __SSE4_2__
#include <nmmintrin.h>
#endif
#elif defined __ALTIVEC__
#define EIGEN_VECTORIZE
#define EIGEN_VECTORIZE_ALTIVEC

4
Eigen/src/Core/arch/SSE/PacketMath.h
View File

@@ -118,6 +118,9 @@ template<> EIGEN_STRONG_INLINE Packet4f ei_pmul<Packet4f>(const Packet4f& a, con
template<> EIGEN_STRONG_INLINE Packet2d ei_pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_mul_pd(a,b); }
template<> EIGEN_STRONG_INLINE Packet4i ei_pmul<Packet4i>(const Packet4i& a, const Packet4i& b)
{
#ifdef __SSE4_1__
return _mm_mullo_epi32(a,b);
#else
// this version is slightly faster than 4 scalar products
return ei_vec4i_swizzle1(
ei_vec4i_swizzle2(
@@ -126,6 +129,7 @@ template<> EIGEN_STRONG_INLINE Packet4i ei_pmul<Packet4i>(const Packet4i& a, con
ei_vec4i_swizzle1(b,1,0,3,2)),
0,2,0,2),
0,2,1,3);
#endif
}
template<> EIGEN_STRONG_INLINE Packet4f ei_pdiv<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_div_ps(a,b); }