Re: [Emacs-diffs] trunk r114452: * data.c (POPCOUNT_STATIC_INLINE): New macro, as a hack for popcount.

Re: [Emacs-diffs] trunk r114452: * data.c (POPCOUNT_STATIC_INLINE): New macro, as a hack for popcount.

[Daniel Colascione]

[-- Attachment #1: Type: text/plain, Size: 888 bytes --]

On 9/24/13 7:53 AM, Paul Eggert wrote:
> ------------------------------------------------------------
> revno: 114452
> revision-id: eggert@cs.ucla.edu-20130924145349-tycv5yrn6ucxfihc
> parent: dancol@dancol.org-20130924114853-75gh9xshqeug2ma6
> committer: Paul Eggert <eggert@cs.ucla.edu>
> branch nick: trunk
> timestamp: Tue 2013-09-24 07:53:49 -0700
> message:
>   * data.c (POPCOUNT_STATIC_INLINE): New macro, as a hack for popcount.

We also want to inline size_t_to_host_endian, since it'll be a noop on
most machine people actually use; really, we want to inline everything I
initially marked "static inline".

We should just use these two keywords and not define yet another
domain-specific macro. If we run on a compiler too old to understand
"inline", we can define it out. There's no need for
POPCOUNT_STATIC_INLINE. What problem is it supposed to solve?


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [Emacs-diffs] trunk r114452: * data.c (POPCOUNT_STATIC_INLINE): New macro, as a hack for popcount.

[Paul Eggert]

Daniel Colascione wrote:

> POPCOUNT_STATIC_INLINE. What problem is it supposed to solve?

It's supposed to solve the performance problem you mentioned,
which it does.  It's just a short-term fix until we get
better and more-portable support for popcount in.
(The new code currently doesn't work on OpenBSD, for example.)
Once it's fixed, the other micro-optimization you mentioned
may not even be applicable, not that it has any significant
practical effect even now.