Eduard Valeyev

Do you mean x100 slower than nested loops with explicit offset computation or than DGEMM? A code snippet would be useful.

It is most probably due to dynamic memory allocation in index construction. When you know the number of dimensions parametrize tensor by std::array. I do this when using btas with...

we definitely need an alias for constexpr number of dimensions ... perhaps something like this: /// Tensor with const number of dimensions template using TensorNd = Tensor;

@shengg Most contract-like and permute should be assumed to NOT work with TensorView. Although TensorView meets the TWG.Tensor concept spec (checked by btas::is_boxtensor), and hence gets accepted by genv() and...

I assume you don't mean the error due to the lack of comma between c and {0,3}. Validating contraction indices is doable at compile time. Checking matching dimension ranks is...

Do you mean you want Tensor/TensorView to be able to report this at compile time? rank() is easy, you need to use RangeNd whose Index is std::array (or similar), it's...

Do you mean to make TWG spec say that it's possible for rank() to be constexpr? On Mon, Jun 23, 2014 at 8:54 AM, Toru Shiozaki [email protected] wrote: > Yes....

Yes, we are using 4.8 gcc, same errors here. To me this looks like a different interpretation of how constexpr works by clang and gcc. Not sure which is correct,...

I need to read the standard, unfortunately, to understand which compiler is correct. It is possible to declare nonstatic members as constexpr, however ... and rank() must be a nonstatic...

Note that slices can also be contiguous ... I'm not sure if Toru or Miles use that in their use cases. I think to be able to properly detect the...