Linus Torvalds (torvalds.delete@this.linux-foundation.org) on July 12, 2013 9:16 am wrote:
> Mark Roulo (nothanks.delete@this.xxx.com) on July 11, 2013 4:19 pm wrote:
> > Michael S (already5chosen.delete@this.yahoo.com) on July 11, 2013 3:59 pm wrote:
> > > the discussion was about absence of AVX in Silvermont
> >
> > I'm hijacking it a bit because Linus claimed that AVX is "esoteric"
>
> I also very much talked about "engineering tradeoffs".
>
> In a big core, you can go wild. I mentioned in the whole original AVX thread (where I said that FP
> performance doesn't actually matter for most loads once you've gotten at least *some* kind of reasonable
> baseline) that in a big core, once you've done all the work for the things that matter (integer and
> memory), and if you have the resources, you might as well splurge them on the stuff that doesn't really
> matter but makes you look good in benchmarks and makes a couple of people happy.
>
> So go wild in the big cores. As long as AVX doesn't slow down anything
> else, who cares about the extra effort and transistors? Nobody.
>
> But I still don't think that AVX makes sense in a small chip. The engineering trade-offs just aren't there.

The same obviously is true for various optional ARM instructions.