* Please be aware that __always_inline doesn't mean "always inline"!
@ 2012-09-26 23:20 Daniel Santos
2012-09-26 23:50 ` Andrew Morton
0 siblings, 1 reply; 3+ messages in thread
From: Daniel Santos @ 2012-09-26 23:20 UTC (permalink / raw)
To: LKML, linux-mm, torvalds
I've noticed that there's a lot of misperception about the meaning of
the __always_inline, or more specifically,
__attribute__((always_inline)), which does not actually cause the
function to always be inlined. Rather, it *allows* gcc to inline the
function, even when compiling without optimizations. Here is the
description of the attribute from gcc's docs
(http://gcc.gnu.org/onlinedocs/gcc-4.7.2/gcc/Function-Attributes.html)
always_inline
Generally, functions are not inlined unless optimization is specified.
For functions declared inline, this attribute inlines the function even
if no optimization level was specified.
This would even appear to imply that such functions aren't even marked
as "inline" (something I wasn't aware of until today). The only
mechanism I'm currently aware of to force gcc to inline a function is
the flatten attribute (see https://lkml.org/lkml/2012/9/25/643) which
works backwards, you declare it on the calling function, and it forces
gcc to inline all functions (marked as inline) that it calls.
Daniel
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: Please be aware that __always_inline doesn't mean "always inline"!
2012-09-26 23:20 Please be aware that __always_inline doesn't mean "always inline"! Daniel Santos
@ 2012-09-26 23:50 ` Andrew Morton
2012-09-27 0:03 ` Daniel Santos
0 siblings, 1 reply; 3+ messages in thread
From: Andrew Morton @ 2012-09-26 23:50 UTC (permalink / raw)
To: Daniel Santos; +Cc: Daniel Santos, LKML, linux-mm, torvalds
On Wed, 26 Sep 2012 18:20:44 -0500
Daniel Santos <danielfsantos@att.net> wrote:
> I've noticed that there's a lot of misperception about the meaning of
> the __always_inline, or more specifically,
> __attribute__((always_inline)), which does not actually cause the
> function to always be inlined. Rather, it *allows* gcc to inline the
> function, even when compiling without optimizations. Here is the
> description of the attribute from gcc's docs
> (http://gcc.gnu.org/onlinedocs/gcc-4.7.2/gcc/Function-Attributes.html)
>
> always_inline
> Generally, functions are not inlined unless optimization is specified.
> For functions declared inline, this attribute inlines the function even
> if no optimization level was specified.
>
> This would even appear to imply that such functions aren't even marked
> as "inline" (something I wasn't aware of until today). The only
> mechanism I'm currently aware of to force gcc to inline a function is
> the flatten attribute (see https://lkml.org/lkml/2012/9/25/643) which
> works backwards, you declare it on the calling function, and it forces
> gcc to inline all functions (marked as inline) that it calls.
As I mentioned in the other thread, the __always_inline's in fs/namei.c
(at least) are doing exactly what we want them to do, so some more
investigation is needed here?
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: Please be aware that __always_inline doesn't mean "always inline"!
2012-09-26 23:50 ` Andrew Morton
@ 2012-09-27 0:03 ` Daniel Santos
0 siblings, 0 replies; 3+ messages in thread
From: Daniel Santos @ 2012-09-27 0:03 UTC (permalink / raw)
To: Andrew Morton; +Cc: Daniel Santos, LKML, linux-mm, torvalds
On 09/26/2012 06:50 PM, Andrew Morton wrote:
> As I mentioned in the other thread, the __always_inline's in fs/namei.c
> (at least) are doing exactly what we want them to do, so some more
> investigation is needed here?
Yes, definitely. When I did some tests on it (to confirm the behavior) a
few months ago, it did behave as advertised. Sounds like this definitely
needs more research. Thanks Andrew.
Daniel
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2012-09-27 0:03 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2012-09-26 23:20 Please be aware that __always_inline doesn't mean "always inline"! Daniel Santos
2012-09-26 23:50 ` Andrew Morton
2012-09-27 0:03 ` Daniel Santos
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).