[Rapp-dev] Full SIMD implementation of integral

[Willie Betschart]

Hello RAPP -Dev!

Here's a patch containing a full SIMD implementation of integral. 

I also added SWAR macros for type conversions, mainly because benchmarking 
build didn't went through SWAR build, not sure why this was built.

I added two new macro's, SPLAT_U16 and SPLAT_U32. I have tested this separately 
but got difficulties adding unit tests so I wait with that. Splat is used to 
add the previous state when next buffers are processed.

The integral was faster than generic but still not as efficient as the hybrid 
of hardcoded swar and SIMD. Also I needed ssse3's align, sse2 wasn't that fast.
I added a description in the source /compute/vector/rc_integral.c how it works.

Best wishes
Willie

full_simd_integral.patch
Description: full_simd_integral.patch