[LLVMdev] Conditions that cause Clang refuse inlining a function

[David Blaikie]

On Thu, Sep 4, 2014 at 6:57 AM, David Sela <sela.david at gmail.com> wrote:

> Can these thresholds be customized with flags?
>

Not with any documented flags (ie: guaranteed to be around forever) I don't
think.


> The always_inline attribute is only a hint and I can't know if the
> compiler will actually do the inline or not.
>

always_inline is more than a hint - the compiler makes pretty sure this
happens (even at -O0, etc, etc). About the only thing that can thwart this
is recursion (at which point it'd be impossible to inline). I'm not sure if
we have a warning to tell you about these cases.


>
>
> On Thu, Sep 4, 2014 at 4:08 PM, David Blaikie <dblaikie at gmail.com> wrote:
>
>> There are lots of reasons a compiler will choose not to inline a function
>> (inlining, while usually a win for performance is a time/space tradeoff, so
>> often a compiler will try not to make a function 'too large' by inlining
>> too many functions into it - it will simply so at some threshold)
>>
>> In any case, if the correctness of your program depends on the function
>> being inlined, you can use the always_inline attribute.
>> On Sep 4, 2014 5:46 AM, "David Sela" <sela.david at gmail.com> wrote:
>>
>>> Hi Anton,
>>> Thanks for the answer.
>>>
>>> The link you provided doesn't answer the question. It lists the ways to
>>> overcome the linking error if the inline was not successful.
>>> The only condition that is found there to make the compiler not inline
>>> is "when compiling without optimization".
>>>
>>> I asked the question in the forum because I think that this info is not
>>> documented and most chances that the people that know the code inside-out
>>> will know the answer.
>>>
>>> Thanks,
>>> David
>>>
>>>
>>>
>>> On Thu, Sep 4, 2014 at 3:30 PM, Anton Korobeynikov <
>>> anton at korobeynikov.info> wrote:
>>>
>>>> Hello
>>>>
>>>> First of all, such a question definitely belongs to cfe-dev, not
>>>> llvmdev.
>>>>
>>>> Next, 'inline' is certainly a hint, and the full the answer is a first
>>>> entry here: http://clang.llvm.org/compatibility.html#inline
>>>>
>>>> On Thu, Sep 4, 2014 at 3:42 PM, David Sela <sela.david at gmail.com>
>>>> wrote:
>>>> > Hi,
>>>> >
>>>> > I want to have some functions in my code inline so I use the inline
>>>> keyword:
>>>> >
>>>> > inline void foo() {}
>>>> >
>>>> > On some functions the compiler inlines the function but it fails to
>>>> do so on
>>>> > other functions and thus I get a linkage error:
>>>> >
>>>> > error: undefined reference to 'foo'
>>>> >
>>>> > What are the conditions that make the compiler refuse inline?
>>>> >
>>>> > Thanks,
>>>> > David
>>>> >
>>>> > p.s. I know that there are ways to pass compilation and let the
>>>> compiler
>>>> > decide if to inline like using __attribute__((always_inline)) but I
>>>> must
>>>> > have the function inline.
>>>> >
>>>> > _______________________________________________
>>>> > LLVM Developers mailing list
>>>> > LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>>>> > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>>>> >
>>>>
>>>>
>>>>
>>>> --
>>>> With best regards, Anton Korobeynikov
>>>> Faculty of Mathematics and Mechanics, Saint Petersburg State University
>>>>
>>>
>>>
>>> _______________________________________________
>>> LLVM Developers mailing list
>>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>>>
>>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140904/059ca577/attachment.html>