> Getting the whole story working well, especially for > explicitly vectorized loops, may require new intrinsics > (such as [4]), but I think we can make progress with strip > mining or predication alone. Is now a good time to > investigate this? I'll give it a shot. Roland.