On 06/05/2018 10:02 PM, Kyrill Tkachov wrote:
Adding some folks who know more about other CPUs as well.
Are you okay with enabling these instructions in AArch64?
If you could give this a spin on some benchmarks you
care about on your platforms it would be really useful data.
Sameera had written something similar (at least in terms of the result,
I don't remember if the approach was the same) and saw the same results
as you did; non-significant changes to performance for CPU2017 because
of which she did not submit it upstream.
That said, I think it is OK to have this upstream. apinski's suggestion
to add a tuning flag makes sense too; it will make testing and tuning
easier for us if we need to in future.
Siddhesh