Implement new 1 packet x 8 gebp kernel

This commit is contained in:
Gael Guennebaud
2014-03-26 18:53:00 +01:00
parent b286a1e75c
commit bc401eb6fa
3 changed files with 297 additions and 436 deletions

View File

@@ -397,7 +397,7 @@ template<> EIGEN_STRONG_INLINE void
pbroadcast4<Packet4f>(const float *a,
Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3)
{
a3 = pload<Packet4f>(a);
a3 = ploadu<Packet4f>(a);
a0 = vec4f_swizzle1(a3, 0,0,0,0);
a1 = vec4f_swizzle1(a3, 1,1,1,1);
a2 = vec4f_swizzle1(a3, 2,2,2,2);
@@ -413,10 +413,10 @@ pbroadcast4<Packet2d>(const double *a,
a2 = _mm_loaddup_pd(a+2);
a3 = _mm_loaddup_pd(a+3);
#else
a1 = pload<Packet2d>(a);
a1 = ploadu<Packet2d>(a);
a0 = vec2d_swizzle1(a1, 0,0);
a1 = vec2d_swizzle1(a1, 1,1);
a3 = pload<Packet2d>(a+2);
a3 = ploadu<Packet2d>(a+2);
a2 = vec2d_swizzle1(a3, 0,0);
a3 = vec2d_swizzle1(a3, 1,1);
#endif