Sledge: In that case I would bet that you forgot to disable debug in your performance tests ...
The sqr of BM is that fast that it is even with enabled sqr still faster than B3d in squared distance comparision normally ... (its actually that fast that all approx implementations with bit shift etc I found on the net are slower than bm / MingW 5.1.X sqr ...)
The sqr of BM is that fast that it is even with enabled sqr still faster than B3d in squared distance comparision normally ... (its actually that fast that all approx implementations with bit shift etc I found on the net are slower than bm / MingW 5.1.X sqr ...)