Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> On 2019-Feb-15, Tom Lane wrote:
>> Sure, I'll pick it up. I agree it's probably a marginal performance
>> change, but it seems a shame to give up when we were 80% of the way
>> there.
> (BTW, my reading of the articles I cited, as well as my own runs of the
> test programs therein, suggest that in order to get a really good
> performance improvement you need to hand-code calls to the POPCNT
> instruction in assembly rather than rely on the compiler intrinsics. I
> think there is^Wwas almost enough infrastructure to do that. I'm not
> sure that the operations being optimized by these changes are
> interesting enough, in the grand scheme of things.)
Yeah, that last point is a fair one; if the situation were static,
I'm not sure it'd be worth the trouble either. But I think that
once we have these functions we may find more uses for them. IIRC,
there's already one patch in the CF queue that wanted to use
__builtin_clz for something or other.
regards, tom lane