Re: [eigen] [Review] Pull request 66, Huge Tensor module improvements

[ Thread Index | Date Index | More Archives ]

On 11.06.2014 19:48, Christian Seiler wrote:
I agree with the end-result. If you think doing the merge before moving
the headers really simplifies things I won't object. But IMO for now we
should remove things affecting the Core from the pull request, anyways.

Define "Core". [...]

I meant the pblend things Benoit added to the packet logic (i.e. Eigen/src/Core/{GenericPacketMath.h,arch/*}

Hmmm.... It's been a while since I've had this issue with constexpr, so
maybe I'll have to recheck what I had in mind... The thing is that I
experimented quite a lot (especially with the involved TensorSymmetry
code) before I came to the final version I posted here, so maybe I
actually don't need it any more. I'll get back to you on that.


Btw. note that constexpr implies const in C++11 but not in C++14. And in
C++14 you may want to add constexpr also to non-const methods.

Yes, that will be a problem when solving bug 820. I think the solution is to replace the EIGEN_DEVICE_FUNC macro by different macros depending on the what attributes can be set and define those macros depending on the C++-version and CUDA availability (and hope that we don't need to introduce new variants too often).


Dipl.-Inf., Dipl.-Math. Christoph Hertzberg
Cartesium 0.049
Universität Bremen
Enrique-Schmidt-Straße 5
28359 Bremen

Tel: +49 (421) 218-64252

Mail converted by MHonArc 2.6.19+