Help text for config OPTIMIZE_INLINING

Help text for config OPTIMIZE_INLINING

[Sitsofe Wheeler]

The main help text for the OPTIMIZE_INLINING kconfig option says:
"The gcc 4.x series have a rewritten inlining algorithm and disabling this option will generate a smaller kernel there."

That suggests that to generate a smaller kernel one should set the option to n. However the short text says:
"Allow gcc to uninline functions marked 'inline'"

While suggests that one should set it to y to generate smaller kernels. My own experience says that setting it to y DOES generate a smaller kernel so perhaps the main text should be changed to reflect this?


 Sitsofe | http://sucs.org/~sits/



      

Re: Help text for config OPTIMIZE_INLINING

[Ingo Molnar]

* Sitsofe Wheeler <sitsofe@yahoo.com> wrote:

> The main help text for the OPTIMIZE_INLINING kconfig option says: "The 
> gcc 4.x series have a rewritten inlining algorithm and disabling this 
> option will generate a smaller kernel there."
> 
> That suggests that to generate a smaller kernel one should set the 
> option to n. However the short text says: "Allow gcc to uninline 
> functions marked 'inline'"
> 
> While suggests that one should set it to y to generate smaller 
> kernels. My own experience says that setting it to y DOES generate a 
> smaller kernel so perhaps the main text should be changed to reflect 
> this?

sure, could you send a patch (with lkml Cc:-ed) that does this?

Also, would be nice if you sent with/without vmlinux size numbers for 
the config you are using. Btw., do you have 
CONFIG_CC_OPTIMIZE_FOR_SIZE=y enabled as well? That too shrinks the 
kernel (which speeds it up as well on moder CPUs), quite visibly.

	Ingo

Re: Help text for config OPTIMIZE_INLINING

[Sitsofe Wheeler]

> From: Ingo Molnar <mingo@elte.hu>

> 
> * Sitsofe Wheeler wrote:
>  
> > While [this] suggests that one should set it to y to generate smaller 
> > kernels. My own experience says that setting it to y DOES generate a 
> > smaller kernel so perhaps the main text should be changed to reflect 
> > this?
> 
> sure, could you send a patch (with lkml Cc:-ed) that does this?

Will do (I won't send it via Yahoo mail though because it will be mangled).

> Also, would be nice if you sent with/without vmlinux size numbers for 
> the config you are using. Btw., do you have 

5463127 2008-10-11 19:51 vmlinux.no-opt
5456152 2008-10-11 19:56 vmlinux.opt

So about 6975 bytes smaller. This is a config which is vastly cut down, compiler was gcc version 4.1.2 20070925 (Red Hat 4.1.2-33) on Fedora 8.

> CONFIG_CC_OPTIMIZE_FOR_SIZE=y enabled as well? That too shrinks the 
> kernel (which speeds it up as well on moder CPUs), quite visibly.

In the past I've tried it out but it seemed to make my crude benchmark (glxgears) go slower...