Re: [2.6 patch] kill IN_STRING_C

[Adrian Bunk <bunk@stusta.de>]

On Tue, Nov 09, 2004 at 02:58:34PM +0100, Arnd Bergmann wrote:
> > On Mon, Nov 08, 2004 at 02:44:49PM +0100, Andi Kleen wrote:
> > 
> > > > Can you still reproduce this problem?
> > > > If not, I'll suggest to apply the patch below which saves a few kB in 
> > > > lib/string.o .
> > > 
> > > I would prefer to keep it because there is no guarantee in gcc
> > > that it always inlines all string functions unless you pass
> > > -minline-all-stringops. And with that the code would
> > > be bloated much more than the few out of lined fallback
> > > string functions.
> > 
> > If I understand your changelog entry correctly, this wasn't the problem
> > (the asm string functions are "static inline").
> 
> Actually, shouldn't the string functions be "extern inline" then?
> That would mean we use the copy from lib/string.c instead of generating
> a new copy for each file in which gcc decides not to inline the function.
>...

In the kernel, we #define inline to __attribute__((always_inline)) [1]
leaving gcc no room for a decision to not inline it.

> 	Arnd <><

cu
Adrian

[1] only for gcc >= 3.1, but I don't remember problems with older gcc 
    versions

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed