std::max: if versus ?:

[Marc Glisse]

On Sun, 23 Sep 2012, Gabriel Dos Reis wrote:

> On Sun, Sep 23, 2012 at 12:06 PM, Marc Glisse <marc.glisse@inria.fr> wrote:
>> On Sun, 23 Sep 2012, Gabriel Dos Reis wrote:
>>
>>> On Sun, Sep 23, 2012 at 10:09 AM, Marc Glisse <marc.glisse@inria.fr>
>>> wrote:
>>>
>>>> Then that may mean no std::max for vectors, at least until someone makes
>>>> the
>>>> inner type and size of vectors deducible (that would make the vector
>>>> version
>>>> of max more specialized than the generic one and allow overloads without
>>>> touching the generic function). Which isn't that bad, people are big
>>>> enough
>>>> to write (x<y)?y:x themselves...
>>>
>>>
>>> I do not understand why you can't have a syntax such as  __vector<T,N>
>>> those extensions.   If that syntax isn't there, you should consider adding
>>> it.  It is a natural thing to have.
>>
>>
>> The syntax exists (though it is ugly): T __attribute__((vector_size(N))).
>
> that is why I suggested __vector<T,N>.

So you were only suggesting it as a syntax and not as an internal 
representation. I guess I can look at a nicer syntax once the feature 
actually works.

>> What doesn't work currently is that T and N are not deducible, in a template
>> context. One possibility would be to make them deducible
>
> Yes.  That is a more scalable solution, IMO.
>
>>  (sounds like duplicating a bit of template machinery).
>
> Why?

For array<T,N>, deduction works. For attributes, reading PR35758, there is 
going to be a lot of work to make it work.

-- 
Marc Glisse