Vectorization performance on "all zeros" vectors

[Luca_A_]

Hi all,

I have a quick question about (I think) the ALU implementation of the cores.
Does it make any difference in terms of performance if in a vector multiplication one vector is composed by all zero values?
The question arises from my attempt to implement a work stealing algorithm for a dense matrix multiplication.
I made a few tests and it seems that there's no difference at all, but maybe I have to specify something?

Every suggestion is always welcome.

Regards,
Luca

[TaylorIoTKidd]

Luca,

I talked to some engineers closer to the hardware than I. They do not believe there are any HW optimizations to address such special cases.

Just from a silicon footprint standpoint, this makes sense to me. (Special cases generally need more silicon to maintain performance.) KNL may be different (though I don't know).

Regards
---
Taylor