[PATCH][Middle-end]Add a new option to finer control inlining based on function's visibility

[Qing Zhao]

> On Sep 13, 2018, at 2:06 AM, Richard Biener <rguenther@suse.de> wrote:
> 
> On Wed, 12 Sep 2018, Alexander Monakov wrote:
> 
>> On Wed, 12 Sep 2018, Richard Biener wrote:
>>> With LTO "static" is too blurry - would -finline-only-hidden be OK
>>> with you?
>> 
>> This doesn't sound right - in non-pic, everything is hidden, effectively.
>> And the intended use is with Linux kernel, which does not use -fpic.
>> 
>> I agree with LTO this option makes less sense, but I wouldn't expect LTO
>> to be used for livepatching-capable kernels.
> 
> The issue is that with LTO this option probably cannot be reliably
> implemented (well, I guess we could stick 'noinline' attributes
> onto all non-static declared functions…).

currently, I am using 

callee->externally_visible 

i.e.:

  /* Set when function is visible by other units.  */
  unsigned externally_visible : 1;

to distinguish whether the function is static (visible ONLY inside the file) or global (visible by other files)

if externally_visible is TRUE and the option -finline-only-static is specified, disable the inlining.


> 
> Btw, what about
> 
> inline T foo() {}
> 
> in C99?

for this case, foo is not visible outside of the file? right?

the current implementation based on externally_visible can correctly treat this routine as “static” (visible ONLY inside the file), and inline the call to foo.
this seems the correct behavior.


>  Those are not declared static (in fact there may be
> extern T foo () declarations somewhere).  I also think we
> have to continue to inline always-inline functions.

for always-inline functions, if they are NOT static, I think that the option -finline-no-static should NOT inline them for help hot-patching purpose. 

> 
> That is, is it really "-fimplicit-inline-only-static"?  Would
> it make more sense to have a -fno-implicit-inline switch which
> will not inline any function that is not declared inline?

not quite understand this, why adding “implicit”?
the new option -finline-no-static will control inlining of functions with or without “declared inline”.

thanks.

Qing


> That might be already available via -fno-inline-small-functions
> [-fno-inline-functions].
> 
> Richard.