topic Ok, thanks for clearing that in Intel® Embree Ray Tracing Kernels

ssef constructor question

P_V__Hariprasad — Sat, 26 Oct 2013 13:00:57 GMT

Sory, don't undestand much SSE, but why this constructor was commented:

__forceinline ssef ( float a ) : m128(_mm_set1_ps(a)) {}

and now this one is used:

__forceinline ssef ( const float& a ) : m128(_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(_mm_load_ss(&a)), _MM_SHUFFLE(0, 0, 0, 0)))) {}

The other constructors are still using _mm_set_ps().

I think the commented out

SvenW_Intel — Tue, 29 Oct 2013 13:31:19 GMT

I think the commented out version should be the better choice, as it does not enforce reading the float from memory.

Ok, thanks for clearing that

P_V__Hariprasad — Tue, 29 Oct 2013 15:26:20 GMT

Ok, thanks for clearing that out. The actual problem I had is that I must be able to pass also doubles there. And since I'm a SSE noob, I really don't like such exotic code there :). Will restore the old constructor with clear conscious now :).