Roman

I had some unfinished work that I didn't submit/backed out. Wiki is a little outdated. So no, no support for identity tensors at the moment. Use the 'eye' method instead

What other small things did you find? Please report so we can track and fix them.

Moving things to a namespace requires the definition of assign to be available while declaring the tensor which requires pre-including a lot of header files in Tensor.h. There is got...

The above method does work for instance for tensor assignment ~~~c++ template struct assignment { template static FASTOR_INLINE void assign(AbstractTensor &dst, const Tensor &src) { if (dst.self().data()==src.data()) return; trivial_assign(dst.self(),src); }...

What were you trying to compile and with which compiler/flags?

901f471 fixes the bug part of this issue.

Yes, all these features get requested quite a lot. I will implement them as soon as I get some free time. Unfortunately these days my hands are pretty tied with...

Yes. I did not write this one intentionally. Horizontal operations are very expensive on AVX512 vector lanes. If you're stuck at compiling we can add a scalar variant for now....

That is a memory alignment issue. Define `FASTOR_DONT_ALIGN` and rerun your example. The macro does not impact performance and was designed specifically for this purpose.

There is a better solution for this that should come pretty soon. Meanwhile I'll have this documented as it's essentially a bug. Thanks for reporting