Re: macOS metal rendering engine in mac port

[Aaron Jensen]

On Sun, May 30, 2021 at 9:03 AM Eli Zaretskii <eliz@gnu.org> wrote:
>
> > From: Aaron Jensen <aaronjensen@gmail.com>
> > Date: Sun, 30 May 2021 08:46:45 -0700
> > Cc: Alan Third <alan@idiocy.org>, emacs-devel@gnu.org
> >
> > On Sun, May 30, 2021 at 2:36 AM Eli Zaretskii <eliz@gnu.org> wrote:
> > >
> > > > From: Aaron Jensen <aaronjensen@gmail.com>
> > > > Date: Sun, 30 May 2021 00:04:43 -0700
> > > > Cc: Alan Third <alan@idiocy.org>, emacs-devel@gnu.org
> > > >
> > > > If I run the test on line 499 of xdisp.c I get:
> > > >
> > > > Native, hash, no line numbers: 3756ms
> > > > Native, hash, line numbers: 4185ms
> > > >
> > > > Native, alist, no line numbers: 4810ms
> > > > Native, alist, line numbers: 7173ms
> > >
> > > In an unoptimized build, "emacs -Q", without native-compilation, I get
> > > 26.39 and 27.22 sec, respectively.
> >
> > My emacs -Q is 6s and 6.32s, respectively.
>
> A similar (and minor) slowdown, in relative terms.

Yes, which, just for my own edification, we can attribute to fewer
faces in the alist ahead of the line number faces, right? Which is
mitigated by the hash patch. I just want to make sure we're not still
trying to chase something down.

> > > > It's not my faces though, it's my header line that's causing that many
> > > > calls, possibly because it's triggering a redraw because the column
> > > > changed? I'm not sure how mode/header lines work when they update.
> > >
> > > You have the column displayed on your header-line?
> >
> > Yes, I do, though maybe that's a bad idea. Does that cause a full
> > redisplay when typing rather than just updating the individual line?
>
> Not a full redisplay, but it does redisplay the header line in
> addition to the point's line, and displaying the header line most
> probably needs more face-merging.

When displaying column number and moving my point to the left, I get
about 1300 face lookups. When not displaying column number, I get 30.
This is with my config, with line numbers on.