VectorizedArray Warning

[peterrum]

Using GNU 9.4.0, we get the following warning:

usr/local/include/deal.II/base/vectorization.h:3267:7: note: ‘class dealii::VectorizedArray<double, 2>’ has no user-provided default constructor
 3267 | class VectorizedArray<double, 2>
      |       ^~~~~~~~~~~~~~~~~~~~~~~~~~
/usr/local/include/deal.II/base/vectorization.h:3280:3: note: constructor is not user-provided because it is explicitly defaulted in the class body
 3280 |   VectorizedArray() = default;
      |   ^~~~~~~~~~~~~~~
/usr/local/include/deal.II/base/vectorization.h:3474:11: note: and the implicitly-defined constructor does not initialize ‘__m128d dealii::VectorizedArray<double, 2>::data’

@kronbichler I guess we are setting the values not to zero intentionally?

FYI @mschreter

[kronbichler]

Yes, this is intentional, and it would indeed be very bad from a performance point of view if we had to initialize this field. What is the context in which a compiler would raise the warning, as I can't imagine (or don't hope) that it would always come up with the warning in regular code as we use it right now.

[peterrum]

OK, it helped to initialize the object by 0.0, i.e., VectorizedArray<double> a = 0.0;, in user code. The warning is very cryptic but I don't know how to improve the situation.

@mschreter You had this problem also recently? Was that when you compiled deal.II or MeltPoolDG?

[mschreter]

@peterrum I don't remember that this problem popped up at my configuration. Sorry :(.

[kronbichler]

One should get a similar warning also for plain double variables, albeit in a different form (not complaining about the constructor, but about being potentially uninitialized).

[kronbichler]

Do we still want to keep this open? Is there anything we can do?