Safari wrote:
> 
> did compiler used to generate win binary generate SSE2 code or
> does windows version include hand-coded asm optimizations?

I just heard that intel has provided some p4 optimisation of the noise 
function for windows compilation. 

> 
> at least with ICC 6 only one out of 10000000 executed instructions
> was MMX one when I tested it some time ago (icc -tpp6 -xM).
> 

Yes, the problem is that mmx works for 32-bit FP only. So it can be used 
for colour stuff only. SSE would allow to operate on a vector of 4 32-bit 
floats simultaniously. Now most colour operation are on 3 components only, 
so icc will not vectorize them. Additionally it you have to use explicitly 
loops to make icc use it.

> unfortunately I can't test SSE2 stuff.
> 

SSE2 allows operations on vectors of 2 64-bit FP. This could be helpful for 
all the vector operations. But again it needs the use of loops e.g. in 
vector.h to make icc use it. Unfortunatly I can't test it either.

- Micha

-- 
http://objects.povworld.org - the POV-Ray Objects Collection

Post a reply to this message