maybe a floating point issue?

[czylabsonasa]

>
> Not sure why this happens with GDC on -O2, I think LDC would 
> refrain from changing results by default.

Timon, good finding! Then, with D, on the kattis site we should 
begin with the second step :-) I'm not happy about it, bcos there 
are plenty of interesting geometry/probability tasks on that 
site, where floating point numbers and (naturally) higher dim. 
arrays involved,
instead of focusing the problem, we must fight with 
language/compiler.