Re: Suboptimal inline heuristics due to non-code sections

[Nadav Amit]

Nadav Amit <namit@xxxxxxxxxx> wrote:

> Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> 
>> On Tue, May 1, 2018 at 6:40 AM Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
>> 
>>> But if I remove the section completely by removing the
>>> pushsection/popsection, then copy_overflow() gets inlined.
>> 
>>> So GCC's inlining decisions are somehow influenced by the existence of
>>> some random empty section.  This definitely seems like a GCC bug to me.
>> 
>> I think gcc uses the size of the string to approximate the size of an
>> inline asm.
>> 
>> So I don't think it's the "empty section" that makes gcc do this, I think
>> it's literally "our inline asms _look_ bigâ.
> 
> I didnât think about that.
> 
> Playing with the code a bit more, it seems that it is actually related to
> the number of ânew-linesâ in the inline assembly. Removing 4 new-lines from
> _BUG_FLAGS (those that can be removed without breaking assembly) eliminated
> most of the non-inlined versions of copy_overflow().
> 
> Would it be reasonable to remove new-lines in such cases?

My bad. Itâs not the new-line. Let me do some more digging.

Nadav