r218141 - In the Itanium ABI, move stuff to the comdat of variables with static init.

[Richard Smith]

On Thu, Dec 18, 2014 at 5:52 PM, David Majnemer <david.majnemer at gmail.com>
wrote:
>
> On Thu, Dec 18, 2014 at 5:46 PM, Richard Smith <richard at metafoo.co.uk>
> wrote:
>>
>> On Thu, Dec 18, 2014 at 2:30 PM, David Majnemer <david.majnemer at gmail.com
>> > wrote:
>>>
>>> On Wednesday, December 17, 2014, Richard Smith <richard at metafoo.co.uk>
>>> wrote:
>>>
>>>> On Wed, Dec 17, 2014 at 3:52 PM, Rafael Espíndola <
>>>> rafael.espindola at gmail.com> wrote:
>>>>>
>>>>> > Based on talking to David, we think the frontend should be
>>>>> responsible for
>>>>> > picking whether things are ODR or not, and then LLVM optimizations
>>>>> can
>>>>> > continue to be mostly ignorant of comdats. For example, the D5
>>>>> destructor
>>>>> > comdat group functions should all be inlinable.
>>>>>
>>>>> +1 :-)
>>>>>
>>>>
>>>> I think that makes a lot of sense. Does this imply that globals within
>>>> a comdat would need to have some kind of weak linkage?
>>>>
>>>
>>> The only non-weak linkage we allow in COMDATs is external.  The closest
>>> analog is weak_odr but it has subtly different semantics wrt duplicate
>>> symbol definitions.
>>>
>>> I'll try to dig up a concrete example as soon as I can get to my laptop.
>>>
>>
>> I think Rafael's right here: external symbols in a comdat are meaningful
>> (and presumably they mean that you can't have that symbol defined in any
>> other object file you link in, so you either get your definition or no
>> definition), and I can imagine (rare) cases when they might be what you
>> want. (And if not, don't do that...)
>>
>
> I think it means something different from what you are suggesting.
> STB_WEAK is orthogonal to COMDAT-ness.  Clang when targeting MSVC uses
> COMDATs with external linkage to mean:
> - This symbol is in some COMDAT group
> - It is not weak, don't replace my symbol with some non-weak definition.
> - This symbol cannot be discarded.
>
> I don't think there is another combination of linkage and comdat-ness
> which gets the same effects.
>

... in that case, all our interprocedural optimizations *do* need to know
about comdats, and being in a comdat needs to make an external symbol
effectively weak for optimization purposes (you can't propagate information
out of it, because it might be replaced by a different definition). This
violates Reid's desire that optimizations can be mostly ignorant of comdats.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20141218/ad72ca60/attachment.html>