On Tue, Oct 10, 2023 at 11:33 AM Robert Haas <robertmhaas@gmail.com> wrote:
> On Mon, Oct 9, 2023 at 4:47 PM Andres Freund <andres@anarazel.de> wrote:
> > I think we might be able to speed some of this up by pre-compute values so we
> > can implement things like bytesleft / UsableBytesInPage with shifts. IIRC we
> > already insist on power-of-two segment sizes, so instead of needing to divide
> > by a runtime value, we should be able to shift by a runtime value (and the
> > modulo should be a mask).
>
> Huh, is there a general technique for this when dividing by a
> non-power-of-two? The segment size is a power of two, as is the page
> size, but UsableBytesIn{Page,Segment} are some random value slightly
> less than a power of two.
BTW in case someone is interested, Hacker's Delight (a book that has
come up on this list a few times before) devotes a couple of chapters
of magical incantations to this topic. Compilers know that magic, and
one thought I had when I first saw this discussion was that we could
specialise the code for the permissible wal segment sizes. But nuking
the variable sized page headers sounds better.