Re: Use compiler intrinsics for bit ops in hash

On Thu, Feb 27, 2020 at 02:41:49PM +0800, John Naylor wrote:
> On Thu, Feb 27, 2020 at 1:56 PM David Fetter <david(at)fetter(dot)org> wrote:
> > [v6 set]
>
> Hi David,
>
> In 0002, the pg_bitutils functions have a test (input > 0), and the
> new callers ceil_log2_* and next_power_of_2_* have asserts. That seems
> backward to me.

To me, too, now that you mention it. My thinking was a little fuzzed
by trying to accommodate platforms with intrinsics where clz is
defined for 0 inputs.

> I imagine some callers of bitutils will already know the value > 0,
> and it's probably good to keep that branch out of the lowest level
> functions. What do you think?

I don't know quite how smart compilers and CPUs are these days, so
it's unclear to me how often that branch would actually happen.

Anyhow, I'll get a revised patch set out later today.

Best,
David.
--
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778

Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate