VectorAPI: SubAll intrinsics for byte, short, float and double

[Paul Sandoz]

> On May 2, 2018, at 08:59, Halimi, Jean-Philippe <jean-philippe.halimi at intel.com> wrote:
> 
> Thanks for your feedback Paul. I agree that some literature might help us leverage data parallelism especially in floating point cases. I like your suggestion and I believe we should keep it for future work. We could definitely think of a "fast-math" approach that would allow trading FP precision for speed.
> 

And/or trading consistency for speed.


> After discussing with Sandhya I believe that we should move forward with the patch as is and keep this idea for future work in order to remain on par with our initial schedule.

Yes, no objections,
Paul.