[cfe-commits] r47850 - /cfe/trunk/Headers/mmintrin.h

[Chris Lattner]

On Mar 3, 2008, at 7:41 PM, Eli Friedman wrote:
>>> Actually, I just realized an issue with the current implementation:
>>> shouldn't the functions in mmintrin.h actually be static inline?
>>
>> No, that won't work for C99+, as that then prevents these routines
>> from being used in inline functions.  :-(  For non-c99 languages,  
>> yes,
>> I'd argue that static inline is the right choice.
>
> Ah, I forgot... yeah, that sucks.  Then I guess we should mark them
> "extern inline __attribute__((weak))"?  Just plain inline means that
> we can't in general generate the right thing.  Or we could cheat and
> add a "ignore_c99_6_7_4_3_restrictions" attribute.

Wouldn't something like this be fine:

__inline __m64 __attribute__((__always_inline__, __nodebug__))  
_mm_cvtsi32_si64 (int __i) { ... }

?  Using __inline covers C90, this has both attributes, and isn't  
static.

>>> We don't ever want to depend on an external definition.
>>
>> :-)  If they are always inline, it wouldn't be much of a
>> problem.  :-)  As I recall, the docs give us the latitude to reject
>> when someone wants the address of one of them.
>
> Where does that put us here?  Should we attach some sort of
> "address_not_allowed" attribute to these functions?

We don't need to enforce this.

-Chris