Re: The Emacs master is much slower than the emacs-27 branch.

[Gregory Heytings]

> > > (3) I'm yet unsure about -O0 builds.  Are they (1)- or (2)-like?
> >
> > Like (2), at least with GCC on 64-bit GNU/Linux.  Apparently Eli does 
> > not see this on MS-Windows.
>
> -O0 does not observe the "inline" declaration nor does consider "static" 
> functions for inlining, so -O0 should be comparable among versions, 
> since the "extern inline" -> "static" -> "static inline" changes are not 
> considered by -O0.

I stand corrected, I mixed things up.  The benchmark with GCC 10.2 on 
64-bit GNU/Linux gives:

    |   A  |   B  |   C
-Og | ~21s | ~67s | ~25s
-O0 | ~59s | ~63s | ~75s
-O1 | ~17s | ~26s | ~21s
-O2 | ~16s | ~18s | ~20s
-O3 | ~16s | ~17s | ~19s

A = commit 507d5548349540dbde67d3e535a4607fd2207c49 (last with 'extern inline')
B = emacs-27 ('static')
C = master ('static inline')

(Note that the difference between A and C is in part due to the difference 
of xdisp.c file length: 33842 for A, 35661 for C.)

In short: the '#define EXTERN_INLINE' changes only affect builds with -Og.